|Home | About | Journals | Submit | Contact Us | Français|
New methods for the genome-wide analysis of chromatin are providing insight into its roles in development and their underlying mechanisms. Current studies indicate that chromatin is dynamic, with its structure and its histone modifications undergoing global changes during transitions in development and in response to extracellular cues. In addition to DNA methylation and histone modification, ATP-dependent enzymes that remodel chromatin are important controllers of chromatin structure and assembly, and are major contributors to the dynamic nature of chromatin. Evidence is emerging that these chromatin-remodelling enzymes have instructive and programmatic roles during development. Particularly intriguing are the findings that specialized assemblies of ATP-dependent remodellers are essential for establishing and maintaining pluripotent and multipotent states in cells.
An essential aspect of building a mammalian cell is packing 1.7 metres of DNA into a 5-micrometre nucleus in a form that allows it to be replicated and transcribed in stable, tissue-specific patterns. The basic unit of chromatin assembly is the nucleosome1, which compacts DNA about sevenfold. However, because the overall level of compaction of the vertebrate genome is several thousand fold, relatively little of the DNA in vertebrates is present on simple nucleosomal templates in vivo. Instead, most chromatin is present in undefined, highly compacted structures that remain available for the induction of developmental programs that specify cell fate and morphogenesis.
At least three processes control the assembly and regulation of chromatin: DNA methylation (see ref. 2 for a review); histone modifications (see ref. 3 for a review); and ATP-dependent chromatin remodelling, which is the focus of this Review. ATP-dependent remodelling seems to be crucial for both the assembly of chromatin structures and their dissolution. About 30 genes encode the ATPase subunits of these complexes in mammals. With few exceptions, these ATPases seem to be genetically non-redundant, with mutation of the encoding genes often having severe effects on the early embryo or giving rise to maternal-effect phenotypes (in which the phenotype of the embryo reflects the genotype of the mother). Indeed, in many cases, the genes encoding the ATPases or their subunits are haploinsufficient (that is, one copy is insufficient for development), indicating that their role in specific processes is rate limiting. Despite their genetic non-redundancy, the various ATPases seem to have similar activities when studied in vitro: they all increase nucleosome mobility4. Therefore, it is clear that better in vitro assays are needed to tease apart their biological functions.
With the advent of genome-wide analysis techniques such as combining chromatin immunoprecipitation with serial analysis of gene expression (ChIP–SAGE) or with massively parallel sequencing (ChIP–Seq), our understanding of chromatin regulation has improved markedly5. These approaches, combined with rapid RNA interference (RNAi) screening and simpler genetic methods, are allowing a new appreciation of the role of the ATP-dependent remodellers in development, particularly in stem cells. Here, we review the key developmental roles of the four classes of ATP-dependent chromatin-remodelling enzyme in Drosophila melanogaster and mice, and we present evidence that these remodellers have an important role in establishing and maintaining the pluripotency of embryonic stem cells, perhaps as a result of the unique configuration of the chromatin ‘landscape’ of a pluripotent cell. These studies show that chromatin remodellers consist of a large number of assembled complexes, some of which are cell-type specific and developmental-stage specific. Many of these assemblies have specialized and largely non-redundant functions during development.
ATP-dependent chromatin-remodelling complexes seem to have evolved to accommodate the major changes in chromatin regulation that occurred during the evolution of vertebrates from unicellular eukaryotes (Box 1). As an example, complexes of the SWI/SNF family, which is one of the most-studied families of chromatin-remodelling complexes, have lost, gained and shuffled subunits during evolution from yeast to vertebrates. In particular, the transition to vertebrate chromatin-remodelling complexes involved the expansion of several of the gene families encoding the subunits and the use of combinatorial assembly, which together are predicted to allow the formation of several hundred complexes. But what is the advantage of combinatorial assembly?
During the evolution of multicellularity and complex body plans, the demand for tissue-specific and developmental-stage-specific expression of genes coincides with increased complexity in chromatin organization and in strategies for chromatin regulation. In the figure, the arrows represent the timescale of evolution, and the appearance of specific strategies of chromatin regulation is indicated, together with relevant important developments in eukaryotic evolution. For instance, the chromatin of yeast (Saccharomyces cerevisiae), which is unicellular, is simpler than that of vertebrates and does not contain linker histones or methylated DNA, the latter of which is also rare in Drosophila melanogaster.
ATP-dependent chromatin-remodelling enzymes also evolved, and we take SWI/SNF complexes as an example. In the figure, the homologous subunits of these complexes in yeast, D. melanogaster and mice are shown as similar shapes of the same colour, allowing them to occupy specific positions in the illustration of the complex, as in a jigsaw puzzle. The domains that enable the subunits to interact with DNA are depicted at the surface of each protein, as explained in the key. In yeast, these complexes are monomorphic in composition and seem to contribute mainly, if not exclusively, to transcriptional activation and transcriptional elongation.
The evolutionary emergence of multicellular organisms was accompanied by the loss of some of the subunits that are present in yeast Swi/Snf complexes (Snf6, Snf11, Swp29 and Swp82) and the gain of others. Unlike in yeast, there are two D. melanogaster SWI/SNF complexes — the Brahma (BRM)-associated proteins (BAP) complex and the polybromo-containing BAP (PBAP) complex, and these can mediate transcriptional activation and transcription repression. In the figure, these are depicted collectively as BAP complexes.
In the transition to vertebrate complexes, there was a large increase in the number of possible complexes as a result of vertebrates gaining the ability to combinatorially assemble several subunits encoded by gene families. The possible subunits at each position are listed in the figure (for example, BAF60A, B, C indicates that one of these three subunits is present). Some assemblies of the vertebrate SWI/SNF complexes known as brahma-associated factor (BAF) complexes are tissue-specific and have unique developmental roles, for example the npBAF complex (which is specific to neuronal progenitors) and the nBAF complex (which is specific to neurons) (Fig. 1). Other assemblies might coexist in a specific cell type and perhaps target specific genes or function together with specific transcription factors. As in D. melanogaster, PBAF complexes are a subset of mammalian BAF complexes defined by the incorporation of polybromo (also known as BAF180) and BAF200 (also known as ARID2), although these were purified from HeLa extracts and may represent partly assembled complexes. Thus, although the fundamental activity of promoting nucleosome mobility is highly conserved from yeast to humans, additional mechanisms that have not yet been discovered could account for the evolution of functionally different complexes. It is not known whether these ideas can be generalized to other ATP-dependent remodelling enzymes.
BRD, bromodomain-containing protein; BRG1, brahma-related gene 1; MOR, Moira; PHD, plant homeodomain; PtdIns(4,5)P2, phosphatidylinositol-4,5-bisphosphate; SAYP, supporter of activation of yellow protein; SNR1, Snf5-related protein 1.
One of the surprises of the genomic era is the relatively small number of genes that are present in vertebrates but not in flies (D. melanogaster). Hence, the greater complexity of vertebrates cannot be attributed to an increase in gene number. Instead, the vertebrate genome, which is about 30-fold larger than the fly genome, contains more genetic regulatory information outside protein-coding genes. Perhaps in response to this expansion of the genome, another strategy was used to regulate chromatin: combinatorial diversity. Current evidence indicates that many vertebrate chromatin-regulatory complexes are assembled combinatorially (see ref. 6 for a review), thereby greatly expanding the potential for diverse gene-expression patterns compared with unicellular eukaryotes. Arguably, the greatest need for diverse patterns of gene expression occurs in the development and function of the brain, and it may be no accident that an extraordinary diversity of neural phenotypes is emerging from genetic studies of the subunits of chromatin remodellers in the nervous system (see ref. 7 for a review).
The evolutionarily conserved SWI-like ATP-dependent chromatin-remodelling complexes can be broadly divided into four main families on the basis of the sequence and structure of the ATPase subunit: SWI/SNF, ISWI, CHD and INO80 complexes. However, many of the predicted SWI/SNF-like ATPases do not fit any of these classes and await characterization. Why does the regulation of a genome require so many functionally non-redundant ATP-dependent chromatin remodellers if they all act to increase nucleosome mobility? Emerging evidence supports at least two possible explanations. First, new roles and molecular functions of chromatin remodellers have been discovered recently. For example, ISWI complexes have been shown to be required for maintaining the higher-order structure of the D. melanogaster male X chromosome8, and INO80 complexes are involved in telomere regulation, chromosome segregation, and checkpoint control and DNA replication during cell division (see ref. 9 for a review). Hence, it is becoming clear that SWI-like remodellers are intricately involved in many aspects of cell biology beyond transcription. Second, even within their traditional role of transcriptional regulation, ATP-dependent chromatin remodellers do not function in a consistent manner. Brahma-associated factor (BAF) complexes, which belong to the SWI/SNF family, can function as both transcriptional activators and repressors and can even switch between these two modes of action at the same gene10. In addition, tissue-specific BAF complexes have been reported to interact with a variety of transcription factors in different cell types (see ref. 11 for a review), allowing the complexes to take on context-dependent functions arising from their different interaction partners. For these reasons, the roles of ATP-dependent chromatin remodelling may be wider, yet more precise and programmatic, than was previously thought. Indeed, modulating the expression of a single target gene can partly suppress the phenotypes of mutations in the BAF complex in the heart12 and in post-mitotic neurons13. This focus on a single target is also seen for polycomb group (PcG) proteins. These proteins mediate transcriptional repression and often oppose the function of trithorax group (TrxG) genes such as those encoding BRG1 and MLL (discussed in the next section), by regulating chromatin structure. Early developmental effects of mutations in the mouse PcG gene Ring1b (also known as Rnf2) can be partly repressed by a mutation in Ink4a (also known as Cdkn2a or Arf), which is a BMI1-target gene and cell-cycle inhibitor14. In addition, late developmental effects of Bmi1 mutation can be partly repressed by null mutations of Chk2 (also known as Chek2), which normally induces a checkpoint evoked by the mitochondrial dysfunction in Bmi1-mutant mice15. Furthermore, the neural developmental phenotypes of mice lacking the TrxG protein MLL, can be reversed by rescuing expression of just one of its targets, Dlx2 (ref. 16). The surprising dedication of chromatin regulators to a single gene suggests that in vitro studies of mechanism will need to focus on appropriate biological targets in the correct cell type.
The diverse developmental roles of each family of chromatin remodeller in mammals are summarized in Tables 1 and and2.2. What emerges from this distillation is the large number of phenotypes associated with mutations in these complexes. It seems that most, if not all, developmental transitions require chromatin regulation and that such regulation is more specific than was initially thought. This is consistent with the findings mentioned above that alterations to single genes can often rescue at least part of the null phenotype. In retrospect, this conclusion is perhaps not surprising given that several hundred proteins seem to be involved in a non-redundant manner in chromatin regulation during development.
SWI/SNF complexes are crucial for the proper development of all organisms in which they have been studied. In this section, we highlight their widespread developmental functions, emphasizing the importance of combinatorial diversity to their specialized roles throughout development.
In D. melanogaster, correct body segmental identity is determined by the proper expression patterns of homeotic genes of the Antennapedia complex and the Bithorax complex. Misexpression of homeotic genes leads to segmental transformations and other patterning defects. The expression patterns of genes in the Antennapedia and Bithorax complexes are first established by the actions of the gap and pair-rule groups of genes and are then maintained by the opposing actions of PcG proteins (which are repressive) and TrxG proteins (which are activating). The genes encoding the core homologues of yeast (Saccharomyces cerevisiae) Swi/Snf in D. melanogaster — brahma (brm), osa and moira (mor) — were first identified in screens for suppressors of homeotic transformations caused by mutations in the Polycomb gene17 and were hence classified as TrxG genes. D. melanogaster SWI/SNF proteins are present and function in a multisubunit complex known as Brahma-associated proteins (BAP). Subsequently, analysis of several other subunits of the BAP complex — SNR1 (Snf5-related protein 1; a homologue of mammalian BAF47 and yeast Snf5) and SAYP (supporter of activation of yellow protein; also known as E(Y)3; a homologue of mammalian BAF45-family members) — showed that they are also required for the antagonism of PcG proteins18,19. By light microscopy, PcG proteins and BAP complex proteins are mutually exclusive on salivary-gland polytene chromosomes20 and PcG proteins might directly counteract the chromatin-remodelling activity and recruitment of the BAP complex to chromatin21.
Although maternal BAP complex proteins are required for the early stages of specifying segmental identity in D. melanogaster, when BRM or other components of the BAP complex are depleted from the zygote, this leads to multiple defects in organ and gamete formation and is lethal in embryos at late stages of development (see ref. 22 for a review), revealing that the roles of the BAP complex extend beyond antagonizing PcG proteins.
The mammalian homologues of the BAP complex have similarly widespread roles in development, although at present there is no evidence that mammalian complexes have TrxG-protein-like functions during the specification of segment identity in vertebrates. Studies of BAF complexes in mammals indicate that these complexes undergo progressive changes in subunit composition during the transition from a pluripotent stem cell to a multipotent neuronal progenitor cell to a committed neuron (Fig. 1a). In mammals, the ATPase subunit of the SWI/SNF complex is encoded by two homologues, Brm and Brg1 (brahma-related gene 1). The ATPase is 1 of 12 subunits that seem to be non-exchangeable in vitro. Several subunits are encoded by gene families (for example, BRG1 and BRM are encoded by one gene family, and only one of these is present in each complex), giving rise to a diversity of stable assemblies that differ between cell types and that have distinct functions23,24 (Fig. 1a). Mice deficient in either of the two ATPases have different phenotypes. Maternal Brg1 is required for zygotic genome activation in a two-cell-stage embryo25, and zygotic Brg1 is essential for both the survival and the proliferation of the cells of the inner cell mass and the trophoblast26 (Table 1 and Fig. 2). By contrast, Brm-knockout mice develop normally, albeit with a slight increase in body mass27. Consistent with the fact that lethality occurs near the time that the embryo is implanted in the uterine wall (at peri-implantation), recent studies have demonstrated that BAF complexes have a crucial role in maintaining the self-renewal and pluripotency of mouse embryonic stem cells (ESCs)28–31. Mouse ESCs produce a complex called esBAF, which is characterized by containing BRG1 but not BRM, and BAF155 but not BAF170 (refs 28, 29); this complex regulates the core pluripotency transcriptional network of mouse ESCs32. So far, complexes containing BAF155 and not BAF 170 have not been found in other cell types. In addition to its role in transcriptional regulation during early embryogenesis, a homologue of BRG1 was also identified in a screen for genes that are essential for nuclear reprogramming in the cytoplasm of frog (Xenopus laevis) oocytes33. It is not clear whether the oocyte contains a specialized complex that is responsible for the nuclear-reprogramming activity. However, given that mouse ESCs are also capable of nuclear reprogramming, the complex present in oocytes might be akin to the esBAF complex. Additional evidence that the esBAF complex is essential for pluripotency comes from the observation that deletion of BAF47 or BAF155 is lethal to the embryo before it has implanted34,35. To differentiate into cells of different lineages, pluripotent ESCs need to exit from the state of self-renewal by silencing genes that potentiate the ESC state. BAF complexes are also crucial for this exit from the ESC state: this is evident from a study showing that RNAi-mediated depletion of BAF57 or BAF155 (components of esBAF) prevents silencing of Nanog, which encodes a master regulatory transcription factor, and also hinders chromatin compaction and hetero-chromatin formation during differentiation36.
As ESCs differentiate into neuronal progenitors, the esBAF complex undergoes several subunit exchanges: it incorporates BRM and BAF60C and excludes BAF60B24,29 (Fig. 1a). BRG1 is required for the self-renewal of neuronal progenitors and for the normal differentiation of neurons from these progenitors24. BRG1-deficient neuronal progenitors misexpress key components of the NOTCH-and sonic-hedgehog-signalling pathways, which direct neurogenesis24. BAF45A is sufficient to induce the proliferation of neuronal progenitors past their normal mitotic exit point24. As neuronal progenitors leave their stem-cell niche in the subventricular zone of the brain and exit from mitosis, they express BAF45B and BAF53B, which replace BAF45A and BAF53A in the neuronal-progenitor-specific BAF (npBAF) complex to form a neuron-specific BAF (nBAF) complex24. The nBAF complex promotes activity-dependent dendritic outgrowth by interacting with the Ca2+-responsive dendritic regulator CREST (also known as SS18L1), thereby directly regulating genes essential for dendritic outgrowth13. The function of BAF53B in dendritic morphogenesis cannot be replaced by BAF53A, demonstrating the functional specialization of BAF complexes of different compositions. Genes encoding several subunits of BAF complexes were also found in an RNAi screen for factors involved in dendritic morphogenesis in D. melanogaster37, indicating that there is a conserved chromatin-regulatory program of dendritic morphogenesis. Interestingly, the downregulation of BAF53A during neuronal differentiation in mice is mediated by two microRNAs (miRNAs): miR-9* and miR-124 (ref. 38). These miRNAs are targeted by REST, a transcriptional repressor that is selectively downregulated in post-mitotic neurons. Mutating the miRNA-binding sites in the 3′-untranslated region of BAF53A gene leads to prolonged expression of BAF53A and reduced amounts of BAF53B in post-mitotic neurons. In accordance with the neuron-specific function of BAF53B, the persistent expression of BAF53A in neurons causes defects in activity-dependent dendritic outgrowth, illustrating the biological significance of subunit switching in BAF complexes during neural development. These studies suggest that the tissue-specific BAF complexes that arise from combinatorial assembly might allow matching between chromatin-remodelling complexes and ambient transcription factors, such as CREST in the case of post-mitotic neurons.
Additional evidence for this coordination between tissue-specific BAF complexes and transcription factors comes from studies of heart development. Baf45c-containing complexes are required for heart and muscle development in zebrafish (Danio rerio)39. In addition, BAF60C is selectively expressed in the regions of the mouse embryo that give rise to a heart and is required for morphogenesis of the heart, differentiation into cardiac and skeletal muscle cells40 and establishment of left–right asymmetry in the early embryo41. Remarkably, ectopic expression of BAF60C but not BAF60A, in coordination with the transcription factors GATA4 and TBX5, is sufficient to induce the development of beating cardiomyocytes from non-cardiogenic mesoderm in the developing embryo42 (Figs 1a and and2).2). Hence, BAF complexes are required for cardiac fate determination. But the closely related polybromo-containing BAF (PBAF) complexes (a subfamily of mouse SWI/SNF complexes defined by the incorporation of polybromo (also known as BAF180)) are expressed in the epicardium and have roles that are non-redundant with those of BAF60C in mediating coronary development and cardiac chamber maturation43,44. By contrast, PBAF complexes are not required for ESC formation or function: deletion of the signature subunit of these complexes, polybromo (BAF180), does not impair the formation of the inner cell mass or the production of different germ layers43.
The first evidence that BAF complexes could repress transcription through long-range interactions came from studies of mouse T cells. Normal T-cell development in the thymus depends on BAF complexes, although it is unclear whether thymocytes also have a specialized BAF complex. During development, T-cell differentiation is coupled to the differential expression of two co-receptors of the T-cell antigen receptor: CD4 and CD8. The expression of Cd4 is developmentally repressed by a distant silencer located 2 kb from the transcription start site of Cd4, and deletion of this silencer results in de-repression of Cd4 at an early stage of development45. The silencer binds to BRG1, BAF57 and presumably the entire BAF complex, which subsequently recruits a transcription factor required for Cd4 silencing, RUNX1 (which mediates repression by an unknown mechanism)46. BRG1-deficient thymocytes prematurely express CD4 and fail to activate CD8 expression, resulting in arrested thymocyte development47. This mode of action is distinct from that of the yeast Swi/Snf complex, which in all cases investigated is recruited to promoters by transcription factors in order to activate transcription48. Surprisingly, in mice, BRG1 is required again at a later stage to activate CD4 expression, indicating that BAF complexes can both activate and repress the transcription of a single gene10 depending on the developmental context. Because mammalian BAF complexes often repress transcription at a distance, whereas the yeast Swi/Snf complex always activates transcription by binding to promoters, there are likely to be significant mechanistic differences between these two complexes. These observations indicate that the initial nomenclature that we proposed, mSWI/SNF49, may be an inappropriate extrapolation.
BRG1 and BAF complexes might also be necessary for the induction of differentiation into skeletal muscle, given that the ability of the myogenic transcription factors MYOD1 and MEF2D to transactivate myogenic genes in mice was found to be inhibited by a dominant-negative allele of Brg1 (ref. 50) and also by short-hairpin-RNA-mediated depletion of BAF60C40.
The second family of SWI-like ATP-dependent chromatin-remodelling complexes is the ISWI family. These complexes were first identified in D. melanogaster, which has a single ISWI ATPase. This ATPase is the core component of three types of ISWI complex: NURF (nucleosome-remodelling factor), ACF (chromatin-assembly factor) and CHRAC (chromatin accessibility complex) complexes (see ref. 51 for a review). Loss-of-function mutations in Iswi are lethal during late pupal or larval development8, perhaps as a result of impaired expression of homeotic genes in the imaginal discs. Restricted expression of an ATPase-dead, dominant-negative allele of Iswi leads to defects in organogenesis, owing to its widespread role in cell viability and cell division8. More intriguingly, ISWI complexes are involved in regulating higher-order chromatin structure. This is evident from a study showing that dominant-negative Iswi or Iswi loss-of-function mutations results in marked global decondensation of mitotic chromosomes, owing to the apparent requirement for ISWI in incorporating the linker histone protein H1 (ref. 52), which is in turn required for normal chromosome condensation and compaction53. In females, ISWI deficiency leads to complete sterility8. This results from the misregulation of bone morphogenetic protein (BMP)-mediated gene expression in the germinal stem cells of females, which leads to a rapid loss of self-renewal of these stem cells54. Although the ISWI ATPase is present in three complexes, NURF might be the functionally predominant complex, because deletion of nurf301 (a dedicated component of the NURF complex) recapitulates almost all the phenotypes of the Iswi mutant55.
In mammals, the core ISWI ATPase is SNF2H or SNF2L. SNF2L and SNF2H have non-overlapping protein expression patterns in mice51. They are functionally distinct and are found in different complexes, as distinguished by their accessory subunits51 (Table 2). SNF2L is present in the NURF and CERF (CECR2-containing remodelling factor) complexes, whereas SNF2H is present in the NoRC (nucleolar remodelling complex), WICH (WSTF ISWI chromatin remodelling), ACF and human CHRAC complexes. NURF complexes and NoRC are involved in transcriptional activation and repression. By contrast, ACF, CHRAC and WICH complexes are required for the regulation of chromatin structure (including nucleosome assembly and spacing), the replication of DNA through heterochromatin and the segregation of chromosomes51.
There are no reports of Snf2l-null mice. However, in mice, disruption of the gene encoding the largest subunit of mammalian NURF complexes, bromodomain PHD-finger transcription factor (Bptf), is lethal to the embryo between embryonic day (E) 7.5 and E8.5 (Fig. 2). In these organisms, although the inner cell mass forms normally, the embryo does not develop a distal or anterior visceral endoderm, leading to a lack of the anteroposterior axis and primitive streak and a lack of subsequent mesodermal and endodermal differentiation56. Although BPTF-null mouse ESCs are viable, they are impaired in their ability to generate mesodermal and endodermal cell fates. BPTF also interacts with transcription factors of the SMAD family and regulates BMP-mediated signalling during the establishment of the germ layers in the embryo56. Although this is reminiscent of ISWI function in the D. melanogaster ovary8, it is unclear whether mammalian NURF complexes are similarly required for proper oogenesis. The functions of ISWI are not limited to early embryonic development in mice. Although NURF complexes function during early embryonic patterning, another SNF2L-containing complex, the CERF complex, is required later in embryogenesis: deletion of the gene encoding the CERF-specific subunit CECR2 disrupts cranium formation and causes exencephaly57.
Mice deficient in SNF2H die after embryonic implantation (between E5.5 and E7.5, a checkpoint at which impaired proliferation is often associated with lethality) and have not been fully characterized58. The role of SNF2H-containing complexes in development is less clear than that of SNF2L-containing complexes, although the severe phenotype of Snf2h-null mice suggests that such complexes have crucial roles. An accessory subunit of the WICH complex, known as WSTF (Williams–Beuren syndrome transcription factor), is encoded in the 1.6-megabase haploinsufficient region of chromosome 7. This region is responsible for the genetic disorder Williams–Beuren syndrome in humans, a developmental disorder characterized by a specific form of mental ability or retardation combined with congenital cardiovascular disease and growth deficiency59. Wstf-haploinsufficient mice recapitulate the cardiovascular defects observed in human patients60. Hence, WICH complexes might be responsible for these phenotypes, through either their functions in regulating the replication of DNA or the assembly and structure of chromatin.
The third family of SWI-like ATP-dependent chromatin-remodelling complexes is the CHD family. Theses complexes contain members of the CHD family of ATPases, which comprises nine chromodomain-containing members (see ref. 61 for a review) (Table 2). CHD proteins are broadly classified into three subfamilies based on their constituent domains: subfamily I (CHD1 and CHD2), subfamily II (CHD3 and CHD4) and subfamily III (CHD5, CH6, CHD7, CHD8 and CHD9).
The subfamily I member CHD1 was initially thought to be integral to transcriptional activity. The tandem chromodomain of human CHD was found to specifically recognize and bind to the trimethylated lysine residue at position 4 of histone H3 (H3K4me3; a hallmark of actively transcribed chromatin)62, mediating subsequent recruitment of post-transcriptional initiation and pre-messenger-RNA splicing factors63. However, this role in transcriptional activation is either not general or not conserved, because Chd1-mutant D. melanogaster zygotes are viable and display only a mild notched-wing phenotype. Instead, in D. melanogaster, CHD1 seems to have a more important role in gametogenesis and as a maternal product. Both Chd1-null male and female D. melanogaster are sterile64. In females, oogenesis depends on the presence of functional CHD1 (ref. 64) (Fig. 2). Closer examination reveals that Chd1-mutant females, when mated to wild-type males, lay fertilized eggs that die before hatching. Maternal CHD1 is required for the incorporation of H3.3 into the male pronucleus during decondensation after fertilization. Failure to incorporate H3.3 may render the paternal genome unable to participate in mitosis in the zygote, resulting in non-viable haploid embryos65. Interestingly, mutation of both genes encoding H3.3 in D. melanogaster recapitulates the Chd1 loss-of-function phenotype — that is, viable adults are completely sterile66 — suggesting that this remodeller is specifically dedicated to H3.3 incorporation during early development after fertilization.
Although there are no reports of Chd1-knockout mice, RNAi-mediated depletion of CHD1 in mouse ESCs results in a loss of pluripotency. CHD1 seems to maintain mouse ESC chromatin in a hyperdynamic and euchro-matic state, thereby preserving lineage plasticity67 (Fig. 3). The findings in mice therefore seem inconsistent with the phenotype of null mutants in D. melanogaster. Developing null mice in which Chd1 can be conditionally deleted will be essential to test these conclusions rigorously.
In mammals, the subfamily II members CHD3 and CHD4 are subunits of NURD (nucleosome-remodelling and histone deacetylase) complexes, which contain histone deacetylases (HDACs) and function as transcriptional repressors68. Like BAF complexes, mammalian NURD complexes achieve diversity in regulatory function through combinatorial assembly (Fig. 1b). The core ATPase is CHD3 or CHD4. There are three main accessory subunits, which are encoded by gene families: MTA (metastasis-associated), MBD (methyl-CpG-binding domain) and RbBP (retinoblastoma-associated-binding protein). Each complex contains one MTA protein: MTA1, MTA2 or MTA3. These are mutually exclusive (see ref. 69 for a review) and nucleate complexes with markedly different, and sometimes opposite, functions. Each complex also contains MBD2 or MBD3, which are functionally distinct and contribute to different forms of the complex70, and RbBP4 and/or RbBP7. The composition of the NURD complexes varies with cell type and in response to signals within a tissue (see ref. 71 for a review), giving rise to a diversity of complexes with distinct functions.
Genetic studies of components of the mammalian NURD complex have shed light on its functions during development. Inactivation of mouse Mbd3 results in death during mid-gestation, stemming from the failure of the inner cell mass to develop into a mature epiblast and the subsequent failure of embryonic and extra-embryonic tissues to organize properly after implantation72. Surprisingly, Mbd3-null ESCs are viable and can initiate differentiation in culture, but they fail to commit to developmental lineages, as a result of impaired silencing of pluripotency genes73. Loss of Mbd3 results in the failure to assemble NURD complexes and probably reflects a loss of function for these complexes. Hence, the NURD complex is crucial for the correct silencing of genes during early development to allow proper patterning and lineage commitment. NURD complexes are also required in later development. Conditional inactivation of Chd4 in the haematopoietic cells of mice leads to impaired haematopoietic stem-cell homeostasis and impaired differentiation into myeloid cells, and to defective thymocyte development and defective activation of the Cd4 locus74,75. Thus it seems that NURD and BAF complexes have specific, opposing, roles at the Cd4 locus.
The most extensively studied member of CHD subfamily III is CHD7. Mutations in CHD7 result in CHARGE syndrome in humans (which is characterized by coloboma of the eyes, heart defects, choanal atresia, severe retardation of growth and development, and genital and ear abnormalities)76 and more than 40 alleles have been defined. Chd7-heterozygous (Chd7+/−) mice recapitulate several aspects of the human disease, such as inner-ear vestibular dysfunction (resulting from defective sensory epithelial innervation)77. Molecular studies suggest that CHD7 is involved in transcriptional activation of tissue-specific genes during differentiation78. Indeed, the D. melanogaster counterpart of Chd7, kismet, may be globally required for RNA-polymerase-II-driven elongation and for counteracting PcG-protein-mediated repression, by recruiting the histone methyltransferases ASH1 and TRX to chromatin during development79. The widespread defects observed in patients with CHARGE syndrome suggests that, as is the case in D. melanogaster, CHD7 in humans has a general and non-redundant function in gene transcription and cellular development.
The last family of SWI-like ATP-dependent chromatin-remodelling complexes is the INO80 family. These complexes contain INO80 ATPases, which in mammals include INO80 and SWR1 and are characterized by the presence of a conserved split ATPase domain. The INO80 and SWR1 complexes are large multisubunit machines with in vitro nucleosome-remodelling activity, which might contribute to their in vivo roles in transcriptional regulation (see ref. 80 for a review). In the light of recent discoveries, we focus on the developmental functions of the SWR1 complex and a related complex known as the TIP60-p400 complex. In yeast, the Swr1 complex incorporates the variant histone Htz1 (known as H2AZ in mammals) into chromatin by replacing H2A81. In mammals, the genes encoding p400 and SRCAP are closely related homologues of yeast SWR1 and the gene products are found in distinct complexes that have common subunits but potentially different functions82. SRCAP-containing complexes are required for the deposition of H2AZ in vivo83, although the role of these complexes in H2AZ incorporation during development has not been studied. H2AZ is conserved from yeast to mammals: how it functions in different organisms has been controversial, but it is essential for the development of both mice and D. melanogaster. Moreover, a recent genome-wide study by Boyer and colleagues showed that the incorporation of H2AZ during mouse ESC differentiation is essential for proper lineage commitment84. Hence, mammalian SWR1 complexes might be required for H2AZ incorporation during differentiation, a possibility that requires further exploration.
In contrast to SRCAP, p400 is present in a complex together with the mammalian histone acetyltransferase TIP60 and about 16 other subunits, which together carry out functions in transcriptional regulation (particularly in transcriptional activation) and DNA damage repair (see ref. 85 for a review). There are no reports of mice that are null for components of the TIP60-p400 complex, except for TRRAP (homozygous Trrap-knockout embryos die at the peri-implantation stage)86. But in a recent RNAi screen for chromatin proteins that are required for mouse ESC self-renewal and pluripotency, Panning and colleagues found that reducing the protein levels of several components of the TIP60-p400 complex, including p400 and TIP60 themselves, resulted in premature differentiation and arrest of mouse ESCs87.
Thus, four ATP-dependent chromatin remodellers — CHD1, NURD complexes, the TIP60-p400 complex and the esBAF complex — seem to have non-redundant roles in pluripotency (Fig. 3), raising the question of whether each has a programmatic, specialized, non-overlapping role in maintaining the ‘landscape’ of pluripotent chromatin.
With the advent of genome-wide approaches for probing gene expression, protein occupancy of DNA sites and nucleosome positioning, a molecular framework for understanding the mechanism that underlies differentiation is emerging. ESCs have become the model for generating a unified map of the network of mechanisms that controls pluripotency, self-renewal and differentiation. Master regulatory transcription factors such as OCT4, SOX2 and Nanog work together with miRNAs and chromatin-regulatory proteins to maintain a transcription circuitry that allows both self-renewal and pluripotent lineage commitment (see ref. 88 for a review). Early genetic studies indicated that BAF complexes have an essential role in pluripotency. More recent screens for chromatin-related proteins that are required for ESC morphology identified components of NURD complexes, the TIP60-p400 complex, CHD1 and, as expected, BAF complexes67,87. BAF complexes are the only ATP-dependent remodellers that have been studied by genome-wide ChIP–Seq analysis, but the functions of TIP60 and CHD1 have been studied by using promoter microarrays, providing global mechanistic insights into the functions of chromatin remodellers32,87 (Fig. 3).
As described earlier, the SWI/SNF-like complex in ESCs, esBAF, has a functionally and biochemically distinct composition that is required for self-renewal and pluripotency28,29. High-resolution genome-wide ChIP–Seq studies showed that the esBAF complex is present at about one-quarter of all promoters in mouse ESCs, with the intensity of binding correlating positively with the expression level of a gene. The targets of the esBAF complex are enriched for genes that are expressed highly and selectively by ESCs, and these overlap extensively with the targets of the transcription factors OCT4, SOX2, Nanog, STAT3 and SMAD1, suggesting functional interplay between the esBAF complex and these pluripotency factors in regulating genes involved in maintaining ‘stemness’30,32. A small number of genes have been studied in ESCs cultured in the prolonged absence of BRG1 or BAF components, and the findings suggest that the esBAF complex maintains ESC fate simply by activating ‘ESC genes’ and repressing genes involved in differentiation29–31. However, a genome-wide analysis carried out after acute depletion of BRG1 points to the esBAF complex having additional, more complex, modes of action32. Reduction of esBAF levels by RNAi-mediated knockdown of the core ATPase Brg1 causes a large number of esBAF targets to be both upregulated and downregulated32. Surprisingly, the upregulated genes include ESC-enriched genes that were already being actively transcribed, suggesting that the esBAF complex refines the expression levels of some ESC genes to keep them within the correct boundaries32. The de-repression of several pluripotency genes, including Oct4 and Nanog, in blastocysts treated with Brg1-directed short interfering RNAs30 suggests that this refinement might also occur in vivo. Nonetheless, it is unclear whether the activating or repressing functions of the esBAF complex are more crucial, and it is similarly unclear how these different outcomes of remodelling (activation versus repression) are achieved.
In addition to esBAF, the TIP60-p400 complex is also required for maintaining the self-renewal potential and pluripotency of ESCs87. Using promoter microarrays, p400 was found at more than one-half of all promoters across the genome in mouse ESCs, and the intensity of binding was similarly correlated with the activity of the gene. The complex seems to be recruited to its targets in two ways, directly by the H3K4me3 mark and indirectly by Nanog. Although the TIP60-p400 complex has histone-acetyltransferase activity, which is generally associated with gene activation, this complex functions mainly to repress developmental genes. Hence, TIP60-p400 might deposit H4 acetylation marks that function in an unconventional manner to mediate gene repression.
CHD1 was also recently implicated in pluripotency, through its ability to maintain the open chromatin configuration that is characteristic of mouse ESCs67. Mouse ESCs in which Chd1 has been knocked down using RNAi maintain many of the characteristics of self-renewing ESCs, but they are defective in multilineage differentiation. CHD1 associates with the promoters of active genes and prevents the accumulation of heterochromatin through an unknown mechanism. The conversion of euchromatin to heterochromatin is presumably the cause of the impaired lineage commitment of CHD1-deficient ESCs, because the induction of differentiation transcription programs potentially requires all genes to be generally accessible, including transcriptionally silent developmental genes in ESCs. Although it is not known how CHD1 maintains open chromatin structure, one intriguing possibility is that it is involved in incorporating H3.3 into the chromatin. Because the phenotypes observed in RNAi studies are sometimes unreliable, the implication that CHD1 is involved in pluripotency will need to be confirmed by analysing embryos with null mutations of Chd1.
These studies illustrate the non-overlapping roles and different modes of action of the various ATP-dependent chromatin remodellers in a single cell type, and they show that these remodellers have genetically non-redundant and programmatic roles in pluripotency, perhaps as a result of their coordinated action with the master regulatory transcription factors. Given the crucial roles of chromatin remodellers in pluripotency, it is curious that they were not identified in the search for factors capable of generating induced pluripotent stem (iPS) cells. The reason for this might be the strict stoichiometry of the complexes, the requirement for which is shown by the observation that overexpression of just one subunit often results in a dominant-negative phenotype.
Before mammalian genomes were sequenced and genome-wide analyses of chromatin function became possible, ATP-dependent chromatin remodelling was thought to be largely a permissive mechanism that operates to allow the binding of general transcription factors. However, the discovery that a large number of non-redundant genes are involved in chromatin remodelling and the ability to carry out more rigorous genetic analyses is enabling the specialized and instructive functions of these complexes to be defined. These functions arise partly from the combinatorial assembly of the complexes. The assembly of complexes from products of gene families suggest that biological specificity is produced in much the same way that letters produce meaning by being assembled into words. But the mechanisms by which these chromatin-remodelling ‘words’ are ‘translated’ into specific biological functions are still unclear, and new ways to probe complex chromatin structure might be needed before we can improve our mechanistic understanding. This is particularly true of BAF complexes, which can regulate transcription from significant distances from promoters and therefore cannot be accurately studied using the available in vitro assays.
The predominant view that chromatin remodellers work to unwind DNA from nucleosomes is derived from elegant but non-physiological in vitro studies. These experiments used cell-free extracts and artificial nucleosome templates that do not recapitulate the three-dimensional complexity of chromatin structure in vivo, such as its organization into heterochromatin or euchromatin and into chromosome territories. The evidence presented in this Review and elsewhere indicates that chromatin remodellers have an intricate role in determining the overall structure and long-range interactions of chromatin. It is also becoming clear that nuclear architecture and gene positioning contribute to gene regulation (see ref. 89 for a review), so new assays must be developed to examine the contribution of chromatin remodellers to the establishment and regulation of higher-order chromatin structure. Techniques such as chromosome conformation capture (3C)90 will be crucial, but other assays, for example using artificially reconstructed three-dimensional chromatin in vitro, will be needed to answer such questions.
A long-standing controversy in the field is the mode by which chromatin-remodelling complexes are targeted to their site of action, which is key to achieving biological specificity. Because most remodellers lack sequence-specific DNA-binding motifs, the predominant view is that they are recruited by way of transient interactions with transcription factors and with DNA-binding proteins that recognize specific DNA sequences. From this viewpoint, in the context of development, remodellers would have an accessory role rather than an instructive role in shaping the transcriptional program and therefore the lineage outcome of a cell. However, there is also ample evidence that, in some cases, remodellers are pretargeted to their sites of action, either by histone modifications or by unknown mechanisms, and that they prepare the target site for binding by transcription factors91. Such examples suggest that remodellers have an instructive role in determining the ability of a particular genome to respond transcriptionally to signals that lead to recruitment of a transcription factor or DNA-binding factor to chromatin. If this is true, the obvious questions are how remodellers recognize their targets and whether this targeting is required for a cell to respond specifically to developmental cues during lineage commitment.
At present it is not known whether chromatin remodelling can transmit the memory of cell fate from one generation to the next. With mounting evidence of the transience and reversibility of chromatin modifications (such as the presence of histone demethylases), the view that chromatin configuration is fixed after being established is giving way to the view that the chromatin landscape can be altered in response to both extrinsic signals and intrinsic signals, such that de-differentiation through nuclear reprogramming is possible. So are remodellers similarly dynamic in their mode of action? That is, are they highly responsive to signalling events? If their program of action is transmitted from one generation to another, then uncovering the mechanisms that direct remodellers back to their appropriate sites of action after each cell division will be crucial for understanding how the specificity and the memory of chromatin-remodelling action are achieved during development.
We thank A. Yoo, A. Shalizi and J. Ronan for comment and critique throughout the preparation of this article. L.H. is funded by the Agency for Science, Technology and Research (Singapore). G.R.C. is funded by grants from the National Institutes of Health and the Howard Hughes Medical Institute.
Author Information Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests.