Chromatin features and the epigenetic regulation of pluripotency states in ESCs

Embryonic stem cells (ESCs) possess the remarkable abilities of self-renewal and differentiation. They have the capacity to generate differentiated cells comprising all three embryonic germ layers. This pluripotent capacity makes them an excellent in vitro system with which to study mammalian development (Gifford et al., 2013; Xie et al., 2013) and disease modeling (Merkle and Eggan, 2013).

To fulfill this pluripotent capacity, ESCs have acquired a sophisticated transcriptional circuitry, helmed by the triad of master transcription factors, OCT4 (POU domain, class 5, transcription factor 1; POU5F1), SOX2 (SRY box 2) and NANOG, which operate in an interconnected autoregulatory loop (Chambers and Tomlinson, 2009). Additional transcription factors, signaling effectors and epigenetic regulators converge on this ‘core pluripotency’ network, to stabilize the self-renewing state through positive-feedback mechanisms (Young, 2011). Yet to retain multi-lineage differentiation potential, this self-organizing gene circuitry must remain amenable to perturbation(s) elicited by extrinsic stimuli, e.g. developmental signaling cues, to bring about appropriate changes in gene expression programs. This necessitates a suitably structured chromatin configuration, one that is highly tractable and tailored for the physiological needs of pluripotent cells.

In this Review, we will present an updated view of pluripotency within the context of epigenetic regulation, placing particular emphasis on the relationship between transcription factors, chromatin regulators and signaling cascades in shaping pluripotency sub-states. We will discuss the molecular attributes that underlie the dynamic ESC chromatin landscape and how breaking the nucleosome symmetry adds an additional layer of flexibility in gene regulation that may contribute to the diversification of cellular states.

Developmental specification is accompanied by a progressive loss of cellular potential, and is marked by an increase in epigenetic restriction (Gifford et al., 2013; Skora and Spradling, 2010). As development commences, the zygote loses its totipotent capacity and undergoes successive cleavages to give rise to the blastocyst, which comprises three distinct lineages: the epiblast and primitive endoderm cells, which are derived from the inner cell mass (ICM), and the extra-embryonic trophectoderm cells (Rossant and Tam, 2009). The pluripotent epiblast cells in the ICM will give rise to all somatic lineages of the embryo and the germline. Notably, this pluripotent capacity persists only transiently for a few days in vivo; however, when explanted in vitro, further development is halted and pluripotency can be captured and propagated indefinitely in the form of ESCs.

Mouse ESCs were first isolated from the ICM compartment of blastocysts in 1981 (Evans and Kaufman, 1981), and are one of the earliest and best-studied prototypes of pluripotent cells. The derivation of human ESCs, also from the ICM of explanted blastocyst prior to implantation, came almost 20 years later (Thomson et al., 1998). Since then, different stem cells corresponding to distinct embryonic precursors and pluripotency sub-states in vivo have been described. They differ from the conventional ESC pluripotent state in their developmental origins, transcription factor and signaling requirements, as well as their epigenetic configurations (Table 1). For example, unlike mouse ESCs, conventional human ESCs exhibit a pronounced tendency for X-chromosome inactivation in female cells, and are generally less amenable to genetic manipulation (Buecker and Geijsen, 2010; Hanna et al., 2010b). In this regard, it was postulated that conventional human ESCs might bear a greater resemblance to the mouse epiblast stem cells (EpiSCs), which originate from a developmentally more advanced post-implantation epiblast that has undergone X chromosome inactivation and cannot efficiently form germline chimeras when injected into blastocysts (Brons et al., 2007; Tesar et al., 2007). This diminished potency may be due to the differences in transcriptional and chromatin constituency between mouse EpiSCs and ESCs (Song et al., 2012).

The ability to derive stem cells of different molecular and phenotypic characteristics at different developmental stages in the mouse embryo has led to the idea that the pluripotent state is not invariant, but rather a continuum of states that can be modulated by extrinsic signaling cues, both in vivo and in vitro (Pera and Tam, 2010). Numerous studies have delineated the signaling principles central to the establishment and maintenance of these different pluripotency states (Ng and Surani, 2011). For example, the LIF (leukemia inhibitory factor)/STAT3 (signal transducer and activator of transcription 3) and FGF (fibroblast growth factor)/ERK (extracellular signal-regulated kinase; MAPK1) pathways constitute opposing axes that govern the balance between self-renewal and differentiation, respectively, in conventional mouse ESCs (Niwa et al., 2009). However, FGF signaling is crucial for the maintenance of mouse EpiSCs (Lanner and Rossant, 2010). Notably, experimental manipulation of these signaling pathways can perturb the equilibrium between these two pluripotent sub-states (Bao et al., 2009; Chou et al., 2008). More recently, it has been demonstrated that blocking WNT signaling in mouse ESCs likewise leads to destabilization of the self-renewing state and promotes efficient transition to EpiSCs (ten Berge et al., 2011). Collectively, these studies highlight the surprisingly facile interconversion between mouse ESCs and EpiSCs, and the power of extrinsic signals to surmount epigenetic barriers. Of importance, pioneering work by Smith and colleagues showed that dual inhibition (‘2i’) of the MEK (mitogen activated protein kinase)-ERK and GSK3 (glycogen synthase kinase 3β) signaling pathways in conventional serum-cultured mouse ESCs reinforces their capacity for self-renewal, culminating in the so-called pluripotent ‘groundstate’ (Ying et al., 2008). This seminal work provided the foundation for subsequent studies directed at deriving germline-competent ESCs from previously recalcitrant models, such as the nonobese diabetic mouse strain, as well as from the rat (Buehr et al., 2008; Hanna et al., 2009; Li et al., 2008; Nichols et al., 2009). More recently, an equivalent human groundstate-like naïve pluripotent stem cell has been described (Chan et al., 2013; Gafni et al., 2013). Unlike previous instances (Buecker et al., 2010; Hanna et al., 2010a; Wang et al., 2011b), these naïve human ESCs were obtained independently of transgene expression, via the use of various small-molecule inhibitors and cytokines that include 2i/LIF, among others. Finally, this ‘2i’ inhibition strategy has also proven to be highly effective in promoting nuclear reprogramming of somatic cells into induced pluripotent stem cells (iPSC), in part through conditioning the chromatin for widespread remodeling (Rais et al., 2013; Silva et al., 2008).

Chromatin is the basic regulatory unit of the eukaryotic genetic material. It comprises repeating arrays of nucleosomes, each consisting of 147 bp of DNA wrapped around a histone octamer made up of two molecules of each histone: H2A, H2B, H3 and H4. Histones are classified as either canonical or variant, depending on their primary sequence and mode of deposition during the cell cycle. They are subjected to various post-translational modifications and demarcate different transcriptional domains in the genome (Fig. 1). The different combinations in which these modifications occur can significantly expand the regulatory properties of histone, beyond that of a structural scaffold upon which the DNA is wrapped (Kouzarides, 2007). This can occur in at least two ways: first, by directly altering the biophysical properties of the nucleosome particle through steric changes in histone-histone (within the same or neighboring nucleosomes) and/or histone-DNA interactions; and second, via the recruitment of effector proteins that directly recognize the modifications (Taverna et al., 2007).

Given the instructive nature of signaling pathways in governing cell state transitions, it is conceivable that signaling effectors must, directly or indirectly, converge on chromatin, orchestrating changes in chromatin structure and gene expression. Indeed, increasing evidence now shows that signaling effectors can bind directly to chromatin, leading to gene expression changes (Badeaux and Shi, 2013). For example, in mouse ESCs, the nuclear localization of tyrosine kinases, JAK1 and JAK2 (Janus kinases 1 and 2), which are involved in the LIF-STAT3 pathway, promotes phosphorylation of tyrosine 41 of histone H3 (H3Y41), and interferes with heterochromatin protein 1α (HP1α) binding. This direct signaling to chromatin contributes to the expression of pluripotency genes such as Nanog and Sox2 (Griffiths et al., 2011). Furthermore, the β-catenin-TCF1/3 complex, a core component of the WNT signaling pathway, as well as the mitogen-activated protein (MAP) kinases, can also act directly on chromatin to regulate self-renewal and differentiation (Cole et al., 2008; Goke et al., 2013; Tiwari et al., 2012). In the case of the ‘2i-LIF’ mouse ESCs (Table 1), inhibition of ERK and GSK3 signaling pathways leads to global DNA hypomethylation and a profound decrease in the repressive histone modification mark H3K27me3, which is imposed by the polycomb repressive complex 2 (PRC2) on developmental loci (Ficz et al., 2013; Habibi et al., 2013; Leitch et al., 2013; Marks et al., 2012). The loss of DNA methylation is mediated through the TET enzymes, as well as the increased expression of PRDM14, which represses the expression of both DNMT3A and DNMT3B DNA methyltransferases (Grabole et al., 2013; Leitch et al., 2013; Yamaji et al., 2013). Our own findings also indicate that in mouse ESCs, ERK directly regulates PRC2 deposition and promotes the establishment of the poised RNA polymerase II transcriptional apparatus on developmental promoters (Tee et al., 2014). In addition, the LIF/STAT3 pathway is also known to cooperate with epigenetic regulators such as the SWI/SNF chromatin remodeler complex esBAF to antagonize PRC2 deposition on active genes (Ho et al., 2011). In this case, esBAF may cooperate with Topoisomerase 2a to modulate chromatin accessibility, thereby conditioning the genome for STAT3 binding (Dykhuizen et al., 2013; Thakurela et al., 2013). Together, the parallel activation of ERK and STAT3 pathways help to promote the distinctive chromatin features inherent to the serum-primed pluripotent state.

The reduced level of DNA methylation and altered PRC2 occupancy in the ‘2i-LIF’ mouse ESC model may be indicative of the epigenetic foundation of groundstate pluripotency in vitro. These cells are also thought to parallel the early epiblast cells of the ICM in vivo, wherein the DNA methylation level is lowest following the development of the zygote to blastocyst (Smith et al., 2012), concomitant with imprinted X chromosome reactivation in female embryos (Mak et al., 2004; Okamoto et al., 2004). Interestingly, a recent finding demonstrates that a small fraction of mouse ESCs grown under this specific culture regime co-expresses markers of both embryonic and extra-embryonic lineages (termed Hex+ ESCs), and are able to generate both trophectoderm and epiblast lineages when reintroduced into embryos (Morgani et al., 2013). These cells clearly exhibit an expanded cellular potency in a chimeric setting, when compared with conventional ESCs, which do not typically contribute towards the trophectoderm lineage. The possibility that alleviation of particular restrictive epigenetic modifications by ‘2i-LIF’ may provoke expansion of cellular plasticity is consistent with the notion that lineage specification is generally accompanied by stabilizing epigenetic configurations. However, a second study provided a surprising twist. It was observed that conventional mouse ESCs grown in the presence of serum and LIF could transit through a previously undescribed stem cell state that resembles that of the totipotent 2-cell (2C) blastomeres in vivo (Macfarlan et al., 2012). The authors reported a striking upregulation of retrotransposons, such as endogenous retroviruses (ERVs), that accompanied the expression of 2C-specific protein-coding genes. Interestingly, many of the latter generated chimeric transcripts linked to the retrotransposons, suggesting that they had co-opted the regulatory elements of neighboring retroelements for coordinated expression, a property also likely exploited in the human ESC transcriptional circuitry (Kunarso et al., 2010). The expression of ERVs continues to persist in a sub-population of in vitro cultured ESCs, termed 2-cell like ESCs (2C-ESCs). Although these 2C-ESCs are rare and transitory in nature, most, if not all, of the ESCs cultured in vitro are capable of passing through this peculiar state at some point. These 2C-ESCs have a transcriptional and epigenetic profile similar to that of the totipotent embryonic cells and are lacking in OCT4, SOX2 and NANOG. In this regard, they are molecularly distinct from the Hex+ESCs. Contrary to expectation, addition of ‘2i’ is antagonistic to the emergence of these 2C-ESCs, whereas conditions of high oxygen promote entry into this privileged state. Importantly, abrogation of repressive chromatin modifiers such as KAP1 (tripartite motif-containing 28; TRIM28), LSD1 (lysine (K)-specific demethylase 1A; KDM1A) and G9a (euchromatic histone lysine N-methyltransferase 2; EHMT2) were also able to promote the derivation of these totipotent-like cells, suggesting that this innate totipotent-like capacity may be epigenetically constrained in conventional ESCs. Whether this constitutes a bona fide mechanism to allow ESCs to drift into alternative pluripotent states, or whether it is simply the result of a compromised epigenetic machinery remains to be determined. In the case of the Hex+ and 2C-ESCs, this event may provide a means to alleviate mutational loads acquired during consecutive DNA replication and cell divisions, perhaps in a manner analogous to the epigenetic reprogramming events that occur in the mammalian germline (Hackett et al., 2012; Surani and Tischler, 2012).

It is increasingly evident that environmental conditions such as oxygen tension and nutrient content can influence the metastability and fidelity of stem cell states, in part through epigenetic alterations (Lengner et al., 2010; Macfarlan et al., 2012; Mathieu et al., 2013). The regulatory importance of oxygen availability in influencing stem cell phenotypes, beyond that of cellular bioenergetics, may perhaps be better appreciated when considering the bioavailability of these elements and conditions with respect to the in vivo contexts where stem cells reside. Stem and multipotent progenitor cells are typically localized in highly specialized microenvironments in vivo, which can either be hypoxic or well oxygenated, depending on the proximity to blood vessels (Simon and Keith, 2008). Many studies have uncovered a role for hypoxia in promoting the undifferentiated progenitor state, through activation of different intracellular signaling pathways, as well as the hypoxia-inducible transcription factors (Cipolleschi et al., 1993; Ezashi et al., 2005; Yoshida et al., 2009). The latter, for example, bind the Oct4 promoter to promote an ‘open’ configuration (Covello et al., 2006). In this regard, the atypical response of the 2C-ESCs to high oxygen content may strike as an apparent oddity, highlighting a fundamental difference in the prototypic ESC pluripotent state that may relate to their distinct developmental origins.

Other studies have also established a role of vitamin C in pluripotency and nuclear reprogramming. This antioxidant promotes TET enzyme-mediated DNA demethylation and acts as a co-factor for histone demethylases JHDM1A/1B in ESCs (Blaschke et al., 2013; Chen et al., 2013b; Esteban et al., 2010; Stadtfeld et al., 2012; Wang et al., 2011a). This finding is in line with the emerging importance of metabolic regulation in ESCs (Shyh-Chang et al., 2013a). Of notable mention is threonine dehydrogenase, an enzyme that is highly expressed in pluripotent stem cells and modulates the generation of S-adenosylmethionine, a metabolite used in protein methylation reactions such as that used by histone-modifying enzymes for the catalysis of histone methylation. Inhibition of this pathway leads to a profound loss of histone H3K4 tri-methylation, and compromises ESC pluripotency (Shyh-Chang et al., 2013b). In summary, these studies show that extensive signal-to-chromatin-based mechanisms direct developmental potency, and that differential epigenetic control in stem cell populations underpins transcriptional fluctuation and promotes cellular plasticity.

Epigenetic regulators play a central role in controlling transcriptional output and variability. These regulators include the chromatin remodeling complexes, such as SWI/SNF and NuRD, that act to regulate nucleosome spacing and dynamics, as well as histone-modifying complexes such as histone deacetylases (HDACs) and the PRCs. Studies using different model organisms have shown that many of these chromatin regulators are involved in regulating transcriptional variability (Lehner et al., 2006; Raj et al., 2010). In mice, haploinsufficiencies in Dnmt3a and Trim28 have been shown to increase phenotypic variation in inbred littermates (Whitelaw et al., 2010). In the case of serum-cultured ESCs, heterogeneous expression of pluripotency factors is postulated to regulate phenotypic plasticity (Chambers et al., 2007; Hayashi et al., 2008; Toyooka et al., 2008). However, it is noteworthy that this phenomenon is less pronounced in the ‘groundstate’ ESCs (Wray et al., 2011). At the molecular level, heterogenous gene expression must be imparted by differential modulation of chromatin architecture (Hayashi et al., 2008). Although unrestrained transcriptional fluctuations are likely to be deleterious to cell viability, it is conceivable that cells may benefit from some degree of gene expression heterogeneity as a means to increase phenotypic variance during lineage specification and for survival fitness in general (Huang, 2009). In this regard, it is perhaps of no coincidence that a similar repertoire of epigenetic regulators is involved in the transition of pluripotent ESCs into the 2C totipotent-like state.

The equilibrium and frequency of sub-state switching enforced by signaling pathways must invariably feed back into events that occur at the level of the enhancer and promoter. For example, studies have shown that the disassembly of promoter nucleosomes is rate limiting for gene expression (Boeger et al., 2008), and that competition between nucleosome and promoter proximal occupation of RNA polymerase II can fine-tune gene expression (Gilchrist et al., 2010). These and other studies indicate that the assembly of the transcriptionally competent chromatin state is a highly dynamic process that can be regulated on several fronts, thus affording various opportunities for the fine-tuning of gene expression. In serum-cultured ESCs, the NuRD complex appears to be especially important for promoting transcriptional heterogeneity of pluripotency genes and lineage commitment (Reynolds et al., 2012a). NuRD harbors both chromatin remodeling and histone deacetylase activities, mediated through different constituents in the complex (Zhang et al., 1998,, 1999). Situated at the heart of this complex is methyl-CpG binding protein 3 (MBD3), which serves as a scaffold for the assembly of other components (Kaji et al., 2006,, 2007). Notably, loss of MBD3, and hence NuRD, function in mouse ESCs leads to elevated expression of select pluripotency genes that impedes lineage commitment (Reynolds et al., 2012a). Several of these NuRD-repressed pluripotency genes, e.g. Tbx3 (T-box 3), Klf4 (Kruppel-like factor 4) and Foxd3 (forkhead box D3) are also targets of the PRCs (Walker et al., 2010). Notably, in a separate study by Brookes et al., these genes are classified as ‘PRC2 active’, based on their apparent active transcription yet paradoxical presence of PRCs on their promoters in the context of a heterogenous ESC population (Brookes et al., 2012). Further analysis revealed an independent association of PRCs with the elongative form of RNA polymerase II on these genes (Fig. 1), suggesting that these two components may exist separately within a cell (i.e. on different alleles) or in distinct cell populations. It was also postulated that the heterogenous expression of key pluripotency genes, such as Nanog might impact on this dynamic (Brookes et al., 2012).

Mechanistically, NuRD-mediated deacetylation of H3K27 may facilitate the recruitment of PRC2 to repressed lineage genes, including these ‘PRC2-active’ genes (Brookes et al., 2012; Reynolds et al., 2012b). In this instance, it will also be of interest to assess the impact of NuRD on the global nucleosome occupancy in ESCs, particularly at enhancers and promoters where NuRD resides (Reynolds et al., 2012b; Whyte et al., 2012; Yildirim et al., 2011). This may reveal further insights into how the interplay between nucleosome remodeling and RNA polymerase II pausing (Gilchrist et al., 2010) may contribute to the dynamic gene expression regulation in ESCs (Henikoff, 2008).

The aforementioned group of NuRD-PRC2 co-target genes also make up part of the extended network of pluripotent transcription factors that function to stabilize the core OCT4, NANOG and SOX2 regulatory circuitry (Walker et al., 2011). Thus, an important role of NuRD in ESCs may be to modulate expression of these ‘extended’ pluripotency genes, thereby destabilizing the pluripotency network and lowering the threshold required for cells to respond to differentiation cues and exit the self-renewing state. Consistent with this, depletion of MBD3 during iPSC generation yields unprecedented reprogramming efficiency, effectively eliminating stochastic effects in this otherwise inefficient process (Rais et al., 2013); however this effect may be context-dependent (dos Santos et al., 2014). In addition, UTX [lysine (K)-specific demethylase 6A; KDM6A], a histone H3K27me3 demethylase, is required during iPSC generation to remove trimethyl marks from H3K27me3 in the case of early expressed pluripotency genes (Mansour et al., 2012). Notably, the loss of PRC2 and LIN53, a component of the NuRD complex, is an absolute pre-requisite for triggering germ to somatic cell conversion in C. elegans (Patel et al., 2012; Tursun et al., 2011), which collectively points to an important role of these transcriptional repressors in safeguarding lineage fidelity. However, it is interesting to note that the loss of PRC2 apparently impedes iPSC generation even when using the optimized homogenous ‘secondary’ reprogramming cells (Buganim et al., 2012; Onder et al., 2012; Pereira et al., 2010), suggesting that loss of PRC2 may result in additional deleterious effects during nuclear reprogramming.

Genome-wide charting of DNA and histone modifying enzymes, along with their associated modifications in stem cells and differentiating tissues have given rise to chromatin-modification ‘maps’, which may be used to identify DNA elements of regulatory importance (Gifford et al., 2013; Stergachis et al., 2013; Xie et al., 2013; Zhu et al., 2013). This wealth of epigenetic information on the chromatin is further organized spatially, in a three-dimensional manner in relation to the nuclear structure (Jin et al., 2013; Phillips-Cremins et al., 2013). Several epigenetic features have been extensively described in ESCs (Young, 2011) and some of these are highlighted in Fig. 1.

The preponderance of the bivalent H3K4me3-H3K27me3 chromatin domain is arguably one of the more distinctive epigenetic characteristics described in ESCs and has received considerable attention (Azuara et al., 2006; Bernstein et al., 2006). Bivalent domains are historically described as chromatin stretches comprising oligonucleosomes that harbor both the activating H3K4me3 and the repressive H3K27me3 marks. At the time of the discovery, no distinction was made as to whether the marks co-exist on a single nucleosome particle or on different nucleosomes. In addition, it is important to note that bivalent domains are not exclusive to stem cells as they also occur in differentiated tissues, albeit at lower frequency. While the factors involved in the establishment of this bivalent state are well defined, the exact function of this chromatin configuration remains debatable, in part due to a lack of an appropriate genetic experimental model (reviewed by Voigt et al., 2013). Towards this end, two groups have recently identified mixed-lineage leukemia 2 (MLL2) as the enzyme that directs H3K4me3 deposition on bivalent promoters (Denissov et al., 2014; Hu et al., 2013a). Whereas loss of MLL2 abolished H3K4me3 specifically on bivalent promoters, it did not significantly alter the transcriptional kinetics of these genes upon retinoic acid treatment. This calls into question the purported role of bivalency in priming gene expression (Box 1), and also highlights a redundant mechanism involving other H3K4 methyltransferases, such as MLL1. It was also proposed that MLL2 engages unmethylated CpG islands and functions primarily as a pioneer H3K4 methyltransferase to mark bivalent promoters, with the presence of PRCs somehow blocking the engagement of SET1 H3K4 methyltransferases known to deposit H3K4me3 at active genes (Denissov et al., 2014). However, it will be important to further examine the gene expression kinetics of Mll2-null as well as Mll1/2 double-knockout ESCs in other physiological settings that are representative of development in vivo, e.g. in the context of germ layer differentiation. This is pertinent given that both Mll2 and PRC2 knockout mice are embryonic lethal, with PRC2 mutants showing defects during gastrulation (reviewed by Voigt et al., 2013), and that Mll2-null embryoid bodies are defective in select HoxB gene inductions (Glaser et al., 2006).

Bivalent domains have also been detected in mammalian gametes (Ng et al., 2013; Sachs et al., 2013), although it remains unclear whether the marks co-exist on the same nucleosome. Nevertheless, the retention of residual H3K4me3-K27me3-marked nucleosomes in the sperm genome and the observation that genes marked by higher levels of H3K27me3 at their promoters are more likely to stay repressed during early embryogenesis (Brykczynska et al., 2010; Erkek et al., 2013; Hammoud et al., 2009) may argue for a potential role of bivalency in transgenerational epigenetic inheritance and for development in general. However, further in vivo functional assessment is necessary to render support for or to debunk this hypothesis (Box 1).

In ESCs, the bivalent chromatin domains are embedded in a wider context of a highly dynamic and accessible chromatin landscape, which is postulated to constrain PRC2 occupancy and H3K27me3 spreading (Zhu et al., 2013). Indeed, it is well established that pluripotent cells have a more dynamic chromatin structure compared with their differentiated counterparts (Gaspar-Maia et al., 2011). For example, undifferentiated ESCs contain a lower abundance of constitutive heterochromatin, marked by fewer DAPI-dense foci (Efroni et al., 2008). Pluripotent chromatin is also marked by strategic deposition of histone variants at key regulatory elements, as well as a dynamic binding of structural chromatin proteins such as HP1 (Meshorer et al., 2006). The regulated displacement of linker histones H1 at active regulatory regions also contributes towards chromatin decondensation (Christophorou et al., 2014). These chromatin features are likely to be important for attaining the transcriptionally pervasive pluripotent state (Efroni et al., 2008). Notably, some of these chromatin features are also observed in pluripotent cells in vivo, such as the ICM (Ahmed et al., 2010), and are recapitulated during nuclear reprogramming (Fussner et al., 2011; Mattout et al., 2011).

The plasticity of ESCs is often attributed to hyperdynamic chromatin features. However, whether permissive chromatin structure is instructive for pluripotency or merely a consequence thereof, remains an open question. Emerging evidence suggests it is likely a combination of both. For example, the extent of heterochromatin compaction in ESCs and ICM cells inversely correlates with NANOG expression (Fussner et al., 2011). More importantly, loss of OCT4 is sufficient to trigger ectopic formation of silenced heterochromatin domains in the ICM cells (Ahmed et al., 2010). Using OCT4 as a further example, an additional study showed that the differential kinetics of OCT4 binding in the early mouse embryo might contribute toward lineage allocation, and it was postulated that this differential binding might be regulated by the state of chromatin accessibility (Plachta et al., 2011). This study thus implies a role for local chromatin structure in regulating transcription factor occupancy. Interestingly, a reduced level of OCT4 in mouse ESCs apparently results in enhanced self-renewal, comparable with that observed in ‘2i-LIF’ (Karwacki-Neisius et al., 2013). Somewhat counter-intuitively, this overall reduction in OCT4 level is accompanied by an increase in OCT4 and NANOG occupancies at pluripotency-associated enhancers. This phenomenon reflects either an active redistribution of OCT4 to key regulatory sites and/or an inherent tighter, or additive association of OCT4 at these specific genomic regions. On a speculative note, it is plausible that certain chromatin features may underlie these ‘attractor’ sites, and in the case of the early embryos, may contribute in part to the differential kinetics of OCT4-chromatin binding (Plachta et al., 2011). One such chromatin feature may be the presence of architectural proteins such as cohesin. It is well documented that transcription factor binding often occurs in a clustered fashion that is orchestrated by the cohesin complex. Depletion of cohesin results in reduced DNA accessibility and transcription factor binding, indicating a causative role for cohesin in promoting the formation of these transcription factor clusters (Yan et al., 2013). Cohesin may thus help to confer robustness in gene expression by helping to stabilize highly occupied cis-regulatory modules (Faure et al., 2012) (Fig. 1). This is consistent with the findings in ESCs that pluripotency factors such as OCT4, NANOG and KLF4 interact and cooperate with cohesin complex in a highly coordinated fashion to access and regulate expression of target loci (Apostolou et al., 2013; Wei et al., 2013). As noted previously, it will be important to assess transcription factor binding occupancy in the context of local chromatin features, and how global changes in nucleosome dynamics correlate with transcription heterogeneity and developmental specification (Li et al., 2012; Teif et al., 2012).

The binary relationship between transcription factor binding and open chromatin may be overly simplistic. Recent evidence suggests that OCT4 itself can function as pioneer factor, readily accessing regions of closed chromatin structure (Soufi et al., 2012). In fact this property is not merely restricted to pluripotency factors, but also exhibited by numerous lineage-specific regulators (Zaret and Carroll, 2011). Although pioneer factors can bind to nucleosomal DNA, additional determinants are likely in place to direct target specificity in a context-dependent manner. A recent study aimed at understanding the role of ASCL1 in somatic cell-neuron transdifferentiation identified a ‘trivalent’ chromatin feature that may help guide this pioneer factor to target regions (Wapinski et al., 2013). Notably, this trivalent signature, comprising H3K9me3, H3K4me1 and H3K27ac putatively within a single nucleosome, is present only in cell types that are amenable to neuronal transdifferentiation, perhaps serving as a means to identify other recipient cells that are permissive for neural reprogramming. This work raises the exciting possibility that certain chromatin features can instruct transcription factor-target specificity, paving the way for future research to assess the universality of such a ‘chromatin-guided’ mechanism.

Given the importance of nucleosome dynamics in gene regulation, what then are the molecular attributes that affect nucleosome turnover and/or remodeling? Among many factors, histone variants may be particularly well poised to fulfill such a role. Unlike canonical histones such as H2A and H3.1/2, which are primarily expressed during S phase and deposited in a strict DNA replication-dependent manner, histone variants such as H2A.Z and H3.3 are expressed and deposited throughout the cell cycle, uncoupled from DNA synthesis (Campos and Reinberg, 2009). Histone variants are also subjected to different post-translational modifications and as such could be selectively deployed to alter local epigenetic configuration through replacement of canonical histones at any point during the cell cycle (Hake et al., 2006; Loyola et al., 2006; McKittrick et al., 2004). When incorporated, histone variants can alter the overall stability of the nucleosome particle, regulating nucleosome turnover at key regulatory elements in the genome (Bonisch and Hake, 2012; Chen et al., 2013c; Henikoff, 2008; Jin and Felsenfeld, 2007). Some histone variants have also evolved highly specialized functions, whereas others are expressed in a tissue-specific manner (Talbert and Henikoff, 2010). Interestingly, in certain cases the regulatory function of histone variants can be compensated by upregulation of other related histone isoforms, indicative of functional redundancy (Lin et al., 2000; Montellier et al., 2013). In fact, a recent report suggested that canonical histones bearing particular post-translational modification combinations could compensate for the lack of a particular histone variant, highlighting remarkable plasticity in the process (Montellier et al., 2013). In this case, depletion of a testis-specific H2B variant, TH2B (histone cluster 1, H2ba; Hist1h2ba), during spermatogenesis induced compensatory mechanisms via the upregulation of canonical H2B, as well as via the additional installation of specific post-translational modifications on histones H3 and H4, to drive the completion of histone-protamine exchange. It is likely that such crosstalk between histone genes is necessary to ensure robust completion of key biological processes and therefore is presumably employed in other developmental settings. Along this line, a similar mechanism may also operate in mouse ESCs depleted of H3.3. Notably, despite the prevalent deposition of histone H3.3 at key regulatory and genic sites, only modest transcriptional abnormalities were observed in H3.3-depleted mutant ESCs that successfully differentiated into all three germ layers during teratoma formation (Banaszynski et al., 2013). This highlights the possibility that, as in the case of TH2B, canonical H3 (or post-translational modification changes on other histones) might rescue, in part, the lack of H3.3 at affected loci (Banaszynski et al., 2013; Goldberg et al., 2010).

The engagement of histone variants significantly expands the flexibility of epigenetic control in maintaining pluripotency as well as during development. Consistent with the crucial importance of histone variants in modulating gene expression is their frequent presence at regulatory elements in the genome, coinciding with promoters and enhancers, many of which are also ‘hotspots’ of transcription factor binding. For example in mouse ESCs, H2A.Z occupies both active and inactive regions, including distal regulatory elements (Hu et al., 2013b; Ku et al., 2012). Notably, H2A.Z depletion leads to reduced chromatin accessibility, concomitant with impaired MLL, PRC2 and OCT4 targeting (Hu et al., 2013b). This has led to the view that H2A.Z may act in a general manner to modulate local chromatin accessibility, thereby facilitating efficient targeting of transcriptional activating or repressing complexes to their respective sites in the genome. As mentioned previously, H3.3 also maps to both active and inactive regulatory elements, although, unlike H2A.Z, the loss of H3.3 deposition specifically impacts targeting of the polycomb, but not that of the trithorax machinery (Banaszynski et al., 2013).

Regions enriched for H2A.Z/H3.3 often coincide with regions of nucleosome depletion (Chen et al., 2013c; He et al., 2010), indicative of active nucleosome exchange and/or remodeling (Fig. 1). The observation that H2A.Z, when in complex with H3.3, constitutes a highly unstable nucleosome in vivo is relevant here, highlighting that a key regulatory role for H2A.Z might be to amplify the intrinsic instability of H3.3-containing nucleosomes (Jin and Felsenfeld, 2007) (see Fig. 3). This ‘hybrid’ configuration, as initially referred to by the authors to draw attention to the co-existence of two distinct histone variants in the same nucleosome particle, maps to nucleosome-free regions of active promoters and other regulatory regions (Jin et al., 2009). Although these studies were performed in somatic cells, the findings very likely also pertain to ESCs, given that both cell types share common features with respect to genomic regions co-occupied by H2A.Z and H3.3.

In addition to promoting nucleosome turnover, another interesting feature of H3.3-containing nucleosomes is that a small but significant proportion of them undergo a higher incidence of (H3.3-H4)₂ tetramer splitting in vivo compared with H3.1/2-containing nucleosomes that are highly stable (Fig. 2). In yeast, where the single H3 isoform is homologous to H3.3, (H3-H4)₂ tetramer splitting can be observed on transcriptionally active loci (Katan-Khaykovich and Struhl, 2011). The same is observed in mammalian cells, although the frequency of (H3.3-H4)₂ splitting appears to correlate more with enhancer specificity and potentially during DNA replication (Huang et al., 2013; Xu et al., 2010). Considering that ESCs have a significantly larger number of epigenetically marked enhancers compared with differentiated lineages (Stergachis et al., 2013; Whyte et al., 2013), it will be interesting to examine the frequency and specificity of (H3.3-H4)₂ tetramer splitting as a function of differentiation, and whether pluripotency and/or lineage-specific transcription factors, either acting in isolation or in collaboration with histone chaperones, may play a role in orchestrating this process. Such analysis is important given the realization that it is the chromatin state of enhancers, rather than that of promoters, that better reflects cell type-specific activities (Heintzman et al., 2009; Hnisz et al., 2013; Rada-Iglesias et al., 2011; Visel et al., 2009). Additionally, many studies have now shown that epigenetic patterning of enhancers occurs prior to cell fate specification, and that this enhancer-priming event permits differential access of signaling molecules and transcription factors that drive cell type-specific gene expression programs (Bergsland et al., 2011; Liber et al., 2010; Mullen et al., 2011; Trompouki et al., 2011). The finding that a given transcription factor binds first to the enhancer before potentiating epigenetic changes at the promoter during reprogramming, further attests to the instructive nature of enhancer marking in directing cell fate changes (Taberlay et al., 2011). Therefore, understanding the dynamics of tetramer splitting and identifying the players that regulate this process will reveal novel insights into how enhancers participate in cell fate specification. Indeed, if (H3.3-H4)₂ tetramer splitting is restricted to active enhancers during DNA replication, it may provide a semi-conservative means for active propagation of these instructive regulatory elements through cell divisions, an exciting possibility that warrants further testing.

The aforementioned studies demonstrating the impact of nucleosome composition in gene expression are an appropriate reminder of the fundamental importance of nucleosome stoichiometry in gene regulation, a feature perhaps overshadowed by the overwhelming attention on histone tail post-translational modifications. A prevailing view is that the octameric nucleosome particle is symmetrical, comprising identical pairs of each of the four core histone proteins. This type of nucleosome composition is referred to as homotypic. However, the observation that the nucleosome core particle can actually accommodate a non-identical H2A-H2A.Z pair, a composition referred to as heterotypic, has challenged this view (Fig. 3). In the case of H2A-H2A.Z heterotypic nucleosomes, although initial structural studies did not support the feasibility of such a conformation (Suto et al., 2000), their presence has been detected in vivo, in organisms spanning from yeast, fly and mouse to human (Luk et al., 2010; Nekrasov et al., 2012; Viens et al., 2006; Weber et al., 2010). Importantly, as shown in mouse trophoblast stem cells, the composition of the H2A.Z-H2A nucleosome particle changes dynamically throughout the cell cycle, with heterotypic and homotypic nucleosomes localizing to different regions of the genome and assuming different functions (Nekrasov et al., 2012). Thus, heterotypic nucleosomes represent one spectrum of nucleosome asymmetry wherein the two copies of sister histones are clearly non-equivalent. It is conceivable that other histone proteins, in particular H3-H3.3, may be similarly engaged given their overall similarity in amino acid composition and identical interaction surface within the tetramer (Luger et al., 1997) (Fig. 3). Importantly, the presence of a mixture of homotypic and heterotypic H3 nucleosome populations can potentially influence the choice between two different modes of nucleosome deposition (Ahmad and Henikoff, 2002) and how active transcriptional states, e.g. primed enhancers, are epigenetically inherited. Understanding the precise mechanisms of how histone chaperones regulate the deposition and/or exchange of histones may reveal further insights into how the different permutations of histone stoichiometry are established (Fig. 3). Given our previous discussion on the regulatory roles of ‘hybrid’ H2A.Z/H3.3 nucleosomes, it will be interesting to examine the full histone stoichiometry in this case, to assess whether sister histones are homotypic or heterotypic. If different ‘hybrid-heterotypic’ permutations were possible, then this would greatly expand the number of epigenetic conformations, in addition to the epigenetic diversity made possible by various combinations of post-translational modifications (Fig. 3).

Theoretically, histone variant-imposed nucleosomal asymmetry can be further amplified by the presence of non-identical post-translational modifications carried by each sister histone (Fig. 4). Indeed, as shown by Voigt et al., asymmetrically modified nucleosomes exist in vivo, in both pluripotent and differentiated cells (Voigt et al., 2012). By using an affinity-purification-based mass spectrometry (MS) methodology, the authors were able to assess in a quantitative manner the distribution of post-translational modifications on H3 and H4 sister histones within a mononucleosome. The study revealed that a significant population of mononucleosomes harbors H3K4me3 exclusively on one sister H3 histone, and H3K27me3 on the other. This may represent the strictest embodiment of chromatin ‘bivalency’, as both marks are present not only on the same stretch of chromatin, but on the very same mononucleosome. Importantly, this observation corroborates the inability to detect H3K4me3 and H3K27me3 on the same histone tail peptide at a population level using MS approaches (Voigt et al., 2012; Young et al., 2009).

The presence of asymmetrically marked H3K4me3-H3K27me3 mononucleosomes can further inform us of the modus operandi of the trithorax and PRC2 machineries in vivo. As Voigt and colleagues further show, the activity of PRC2 is inhibited by the presence of active modifications (Voigt et al., 2012). For example, pre-installation of H3K4me3 on both sister histones impedes PRC2 methylation of H3K27, whereas asymmetric presentation of H3K4me3 is ineffectual (Voigt et al., 2012). This is consistent with previous studies in which it was found that H3K36me3, another active marker that is associated with transcription elongation, inhibits PRC2 activity (Schmitges et al., 2011; Yuan et al., 2011). The inverse relationship has also been observed whereby the presence of H3K27me3 impacts the activity of SET1 H3K4 methyltransferases (Kim et al., 2013) (Fig. 4). These findings thus provide a molecular insight into how PRC2 and trithorax complexes function in ESCs. We and others have recently proposed a regulatory logic for how PRC2 may be targeted to various sites in the genome, in a manner that is dependent on transcriptional activity (Brookes et al., 2012; Davidovich et al., 2013; Kaneko et al., 2013; Tee et al., 2014). We envisage a hierarchical series of events whereby PRC2 constantly scans the genome for transcriptional activity, directly assessing the chromatin and transcriptional configurations at promoters, prior to engagement and H3K27 methylation. In this case, even though PRC2 may readily gain access to the promoter regions of highly transcribed loci, perhaps through contact with nascent RNA, the symmetric presence of either H3K4me3 and/or H3K36me3 will curtail PRC2-mediated H3K27 methylation. Indeed, in vitro experiments suggest that the presence of symmetric active modifications does not affect PRC2 binding to the nucleosomes, but rather regulates its activity (Voigt et al., 2012). This interpretation accounts for the observed binding of PRC2 as being potentially promiscuous in the genome (Kaneko et al., 2013), whereas occupancy of H3K27me3 is largely restricted to repressed developmental genes and repetitive elements in ESCs. In summary, the study by Voigt et al. demonstrates that sister nucleosomes can be asymmetrically modified. How this may be integrated with histone variant imposed ‘asymmetry’ to regulate gene expression in vivo remains to be fully elucidated.

ESCs largely divide in a symmetrical fashion with equal partitioning of genetic and epigenetic materials, giving rise to functionally equivalent daughter cells. However, a recent report documents that altering the cell polarity in ESCs can perturb this symmetry (Habib et al., 2013). When subjecting mouse ESCs to localized WNT3A signaling, the authors observed that the ESCs underwent asymmetric division instead. Remarkably, daughter cells proximal to the source of WNT3A had higher expression levels of pluripotency genes, whereas distal cells acquired differentiation hallmarks. This is reminiscent of adult stem cells in vivo that mostly undergo asymmetric cell divisions to generate a self-renewed progenitor and a daughter cell destined for differentiation (Clevers, 2005). A central question relating to this process is how is the parental epigenetic information maintained in the progenitor stem cell, but altered in the differentiating cell? Studies using Drosophila male germline stem cells (GSCs) as a model system have revealed some fascinating insights into this process. Male GSCs undergo asymmetric cell division, giving rise to a self-renewed GSC and a daughter cell gonialblast that undergoes differentiation. By using a dual-color labeling strategy to distinguish between ‘old’ and ‘new’ histones, Tran et al. discovered that pre-existing old canonical H3, but not H3.3, histones are specifically retained in the GSCs, whereas newly synthesized H3 histones are enriched in the differentiating gonialblast cells. Notably, this asymmetric mode of H3 distribution is lost when GSCs are experimentally manipulated to divide symmetrically (Tran et al., 2012). Mechanistically, this asymmetric distribution of histones can occur if (1) pre-existing H3 and newly synthesized H3 are already differentially localized onto the two sets of sister chromatids during S phase prior to mitosis, and (2) the mitotic machinery can somehow distinguish between these epigenetically distinct sister chromatids. It is unclear at this moment how the different epigenetic states between the two sister chromatids are established and subsequently distinguished, although it is tempting to speculate that similar principles governing X chromosome inactivation and imprinting in mammals may be operating in this instance. Further insight into this elusive process may be gained from the observation that asymmetric epigenetic regulation is required to generate neuronal bilateral asymmetry in C. elegans. In a truly remarkable fashion, it has been shown that a single mutation in one of the 24 C. elegans H3 genes could abolish the left-right neuronal symmetry owing to defective H3-H4 tetramer formation and nucleosome assembly (Nakano et al., 2011). This finding led to the intriguing hypothesis that in wild-type worms, unequal loading of nucleosomes onto the leading and lagging DNA strands during DNA replication gives rise to sister chromatids marked with differing nucleosome densities at discrete loci required for neuronal specification. In this regard, the difference in nucleosome density may henceforth constitute a bona fide epigenetic feature that serves as an initial ‘mark’ to distinguish between the two sister chromatids, which is subsequently transmitted following mitosis.

Does the non-random segregation of epigenetically distinct sister chromatids also occur in mammalian stem cells? Some studies do support such an occurrence. Klar et al. reported that mouse ESCs and neuroectoderm cells, but not other cell types tested, show varied segregation patterns of chromosome 7 (Armakolas and Klar, 2006). However, a separate group showed that in a subpopulation of adult mouse skeletal muscle stem cells, apparently all sister chromatids (not just select chromosomes), could undergo biased segregation (Rocheteau et al., 2012). To what extent and significance this asymmetric inheritance of sister chromatids may regulate cell fate decisions in mammalian stem cells remains to be fully defined.

In this Review, we have discussed the emerging importance of epigenetic regulators in governing lineage fidelity and explored how the manipulation of chromatin factors can perturb the equilibrium between discrete pluripotent states. The increasing ease in genome editing and the means to engineer lineage-specific reporters in human ESCs in a high-throughput fashion have provided exciting inroads into our understanding of how epigenetic barriers are established during distinct phases of mammalian development, and how they are overcome during nuclear reprogramming (Cong et al., 2013; Poser et al., 2008; Wang et al., 2013). Mapping of histone post-translational modifications as a function of lineage specification has proven to be extremely informative, and has been the primary method of choice to assess genome-wide chromatin states. However, most studies performed to date do not provide information about the exact histone composition within the nucleosomes or distinguish between modifications on sister histones.

The view that a nucleosome is symmetrical must be revised in light of observations demonstrating the functional differences between heterotypic and homotypic nucleosomes, and how asymmetric presentation of post-translational modifications on sister histones can impact the kinetics of histone-modifying enzymes. Clearly, the extent to which these nonequivalent nucleosomal states exist in vivo remains to be investigated, but they may indeed be the norm rather than the exception. The continual discovery of new histone variants, e.g. through alternative splicing, adds to the growing list of possible nucleosome conformations (Bonisch et al., 2012; Rasmussen et al., 1999). Last but not least, an additional layer of complexity that warrants attention is the finding that histone modifications, as well as nucleosomes, can often be asymmetrically positioned around discrete transcription factor binding sites, with skewed enrichment on one side of the binding site, but not on the opposite end (Kundaje et al., 2012). It is unclear how this directionality impacts on gene regulation, but a clue to this may be gleaned from the recent description that a class of pioneer transcription factors apparently exerts a stronger effect in chromatin opening on one side of their motif than on the other (Sherwood et al., 2014). In conclusion, epigenetic asymmetry is manifested in different forms, from within the most fundamental nucleosome particle, to larger regulatory transcription factor binding units, as well as the inheritance of epigenetically distinct sister chromatids, collectively attesting to the remarkably complex realm of gene expression regulation.

We thank Drs Eric Campos, Matthias Stadtfeld, Maria Elena Torres-Padilla and Lynne Vales for critical reading of the manuscript.