On the relations of phase separation and Hi-C maps to epigenetics

1. Introduction

Cursory inspection of eukaryotic nuclei using a simple light microscope shows that the optical density of chromatin is not uniform. On this basis, Heitz [1] defined heterochromatin as the dense compartment that is opaque to transmitted light and stains deeply with simple dyes, while euchromatin was the other compartment that stained lightly and through which light passed readily. Beyond this strictly empirical definition, quantitative techniques have shown that DNA in mammalian interphase nuclei is indeed more densely packed in constitutive heterochromatin compared with euchromatin. There is two to sixfold higher density of DNA in heterochromatin as measured by fluorescence intensity of DNA-binding fluorophores [2,3], which can be confirmed by measuring nucleosome density using fluorescently tagged histones [3,4]. The increased nucleosome density reflects how the 11 nm ‘beads-on-a-string’ nucleosome fibre is packaged in heterochromatin, and recent work provides a pathway that might explain the increased density observed. In H3K9me3-marked heterochromatin, the preferred contact geometry of the nucleosome fibre is a two-start helical fibre with stacked alternating nucleosomes, making the closest neighbour the second-nearest nucleosome rather than next nearest nucleosome [5]. Super-resolution imaging has revealed another level of organization that is characterized by the assembly of irregularly folded ‘clutches’ of nucleosomes where the density of larger ‘clutches’ is greater in heterochromatin compared to euchromatin [6]. The molecular crowding observed in the heterochromatic environment [2,4] has been modelled and predicted to enhance the interaction between the ‘clutches’ owing to osmotic depletion attraction [7]. The calculated entropy-driven attraction is small, approximately 0.5k_BT, but could favour the merging of ‘clutches’ to form ‘dense domains’ or ‘globules’ that have been detected by super-resolution imaging [8] and chromosome conformation capture [9].

Soon after Heitz's definition of heterochromatin, Müller [10] discovered the phenomenon of ever-sporting displacements in Drosophila, later called position-effect variegation (PEV). PEV continues to be an important experimental paradigm for interrogating the relationship of constitutive heterochromatin to euchromatin in a living animal by disrupting the natural boundary that separates the two cytologically distinguishable states of chromatin (reviewed in [11–14]). PEV led to key conceptual advances and generated invaluable molecular tools that have done much to provide an outline of the natural history of constitutive heterochromatin by unveiling conserved mechanisms that operate in species ranging from fission yeast to man [15,16]. Outstanding among the contributions of PEV were, first, the demonstration that the effect of constitutive heterochromatin on gene repression is pervasive and heritable. Pervasive because, in most cases of PEV, repression results from ‘spreading’ of the dense packaging from within constitutive heterochromatin across the variegating breakpoint into euchromatin [17–19]. Once established, repression is heritable from one cellular generation to the next [20,21]. Thus, the discovery of PEV [10] gave birth to the discipline of epigenetics more than a decade before the term itself was coined [22]. Second was the identification of second-site modifiers of variegation that encode structural and enzymatic components of constitutive heterochromatin (reviewed in [14,23]). Two of these modifiers encode proteins that are highly conserved in organisms from fission yeast to man. One is heterochromatin protein 1 (HP1) and the other H3K9 HMTases that generate the H3K9me2/3 determinant of the histone code to which HP1 binds [24]. HP1 and H3K9me2/3 are hallmarks of constitutive heterochromatin and key epigenetic regulators in eukaryotes [15,16]. They represent a potential link between compartmentalization and epigenetics that will be explored in this paper. We now turn to these hallmarks with a focus on mammalian HP1 proteins because recent in vitro work has indicated that they form liquid–liquid condensates and gel-like states [25–27] that could drive compartmentalization of cytologically visible constitutive heterochromatin in interphase nuclei.

2. Mammalian heterochromatin protein 1 proteins and polymer–polymer phase separation

In mammals, there are three HP1 isotypes, termed HP1α, HP1β and HP1γ, which are encoded by distinct genes, chromobox homologue 5 (Cbx5), Cbx1 and Cbx3, respectively [28]. Immuno-localization studies have shown that HP1α and HP1β are usually enriched within constitutive heterochromatin [29], where their concentration is around 10 µM [3]. HP1γ has a more euchromatic distribution [29]. They are small approximately 25 kDa molecules that consist of two globular domains, an N-terminal chromo domain (CD) and a sequence-related C-terminal chromo shadow domain (CSD), linked by an unstructured, flexible, hinge region (HR) [30]. Depending on the species and isoforms, there are less well-conserved N- and C-terminal extensions (NTE and CTE, respectively) (figure 1a). The CD specifically binds to the N-terminal tail of histone H3, when methylated at the lysine 9 residue (H3K9me) [36–38], with the highest (µM) affinity for the tri-methylated form (H3K9me3; [39]; figure 1b,c). From the crystallographic data, the methyl-ammonium group in K9H3 is caged by three aromatic side chains in the CD, where the binding energy is driven largely by cation–π interactions [32] (figure 1d). The CSD dimerizes and forms a ‘nonpolar’ pit that can accommodate penta-peptides with the consensus sequence motif PxVxL, found in many HP1-interacting proteins [33,34,40] (figure 1e–g). There are likely to be other modes of interaction with the nucleosome including, for example, that of the HP1 CD or CSD with the H3 histone ‘core’ [41–43], binding of the HR region to DNA and RNA [44–46] and a non-specific electrostatic interaction of the NTE with the H3 tail [31].

• Download figure
• Open in new tab
• Download PowerPoint

Mutational analysis in mice has shown that mammalian HP1 isotypes have different mutant phenotypes despite sharing extensive sequence identity [28,47]. HP1α function is essentially redundant. Cbx5^−/− mice are viable and fertile (cited in [48]) [47] albeit they exhibit a very specific defect where T_H1-specific gene expression is not silenced in T_H2 cells [49]. HP1β function cannot be compensated by HP1α and γ. The Cbx1^−/− mutation is fully penetrant with mice dying around birth possessing a variety of lesions including a severe genomic instability [47]. Disruption of the Cbx3 gene (encoding HP1γ) results in infertility and an increased postnatal mortality [50–52]. Consistent with the mutational analysis, unbiased exome data predict HP1β to have the highest probability of loss of function intolerance (pLI) out of the HP1 proteins¹ [53].

Mammalian HP1 proteins were some of the first proteins used in non-invasive fluorescence recovery after photobleaching (FRAP) studies to probe chromatin protein interactions in living cells [54]. Numerous FRAP studies, in conjunction with kinetic modelling, have shown that at steady-state equilibrium, the nuclear HP1 pool can be separated into three kinetic fractions: a highly mobile ‘fast’ fraction that freely diffuses through the nucleoplasm, a less mobile ‘slow’ fraction that binds to the HP1 ligand, H3K9me3 in heterochromatin, and a small immobile HP1 fraction whose ligand(s) is not known [55–57], although it has been suggested that this tightly bound fraction may involve the interaction of HP1 proteins with the histone H3 ‘core’ [35,58]. These data have been interpreted as heterochromatin being a stable, membrane-less, nuclear compartment whose structural integrity is mediated by protein–protein and protein–RNA interactions, where the bulk of the constituents exchange freely with the surrounding nucleoplasm [59]. This view presages thermodynamic models of intracellular phase separation, which have led to physiochemical explanations for the biogenesis and maintenance of different nuclear compartments found in living cells (reviewed in [60–63]).

An exemplar of a nuclear compartment that is formed by liquid–liquid phase separation (LLPS) is the nucleolus, which is assembled at transcriptionally active ribosomal DNA loci [64,65]. Detailed examination of nucleoli has shown they are not composed of a single condensed phase surrounded by a dilute phase but consist of subcompartments where a secondary condensed phase is contained within the primary condensed phase; the subcompartments have distinct viscosities, surface tensions and protein compositions [66,67]. In vitro studies on mammalian HP1α have shown that HP1α undergoes LLPS and led to the suggestion that HP1α-dependent constitutive heterochromatin might also consist of phase-separated subcompartments; specifically a soluble phase, a liquid droplet phase and a gel-like phase [26,27]. The propensity to form liquid droplets in vitro is peculiar to HP1α because, under the same conditions, HP1β and γ do not form liquid droplets [68]; there seems not to be a relationship between the ability to form liquid droplets and pLI. On the face of it, the three phase-separated HP1α subcompartments appear to correspond with the three kinetic fractions observed in FRAP studies. However, a notable difference is that mammalian HP1α liquid droplets form in vitro independently of chromatin [25,27] while the ‘slow’ fraction is dependent upon the in vivo interaction of HP1 with its ligand, H3K9me3 [55,57]. Moreover, the CasDrop system has shown that mammalian HP1α is unlikely to form liquid droplets in vivo [69]. Instead, HP1α-rich foci co-localize with constitutive heterochromatin [29] rather than forming droplets that surround heterochromatin [69] indicating that if HP1α does drive phase separation, it does so in a manner different from multivalent intrinsically disordered proteins [62], which are known to form endogenous liquid–liquid phase-separated condensates in the nucleus [69]. These data indicate that phase separation of mammalian constitutive heterochromatin is unlikely to be mediated by HP1-driven LLPS, albeit work in Drosophila [70] and fission yeast [71] shows that HP1 proteins can drive LLPS. Rather, we suggest that the major mechanism by which mammalian HP1 proteins drive phase separation is polymer–polymer phase separation (PPPS; [72]) as opposed to LLPS.

Evidence that LLPS is unlikely to be the mechanism by which HP1 proteins form phase-separated compartments also comes from studies on their ability to act as bridging molecules between distant chromosomal loci. Expression of a lacI–HP1βCD fusion in a mouse cell line harbouring an approximately 10 kB lac operator (lacO; [73]) integrated into the telomeric end of chromosome 11 increased the frequency of contacts between the lacI–HP1βCD fusion bound to the lacO allele and H3K9me3-marked peri-centromeric heterochromatin [74]. The bridging effect requires the CD-H3K9me3 interaction because a point mutation (T51A in HP1βCD) that disrupts the interaction of the CD with H3K9me3 decreased the frequency of contacts to that observed for the wild-type (wt) allele. This ability of HP1 proteins to promote contact between distant chromosomal loci is conserved. Using the same lacO system, a transgenic fly was generated where lacO was integrated into the telomeric end of the X-chromosome [75]. Expression of a lacI–HP1a fusion led to a discrete lacI–HP1a signal at lacO on polytene chromosomes [75]. Assembly of lacI–HP1a at the lacO promoted chromosome folding and association of the lacI–HP1a–lacO complex with loci at distant chromosomal sites. Bridging between two loci was dependent upon HP1a CSD dimerization. The bridging effect in transgenic flies was all the more impressive given it was observed in polytene chromosomes, which do not fold easily because of their stiffness that is a consequence of the approximately 1000 sister chromonemata that align in complete register, resulting in the giant cross-banded chromosomes [76]; reduced flexibility (increased stiffness) is the reason given for the absence of compartmental domains in Hi-C maps generated from polytene nuclei [77]. It is unlikely that HP1 liquid droplets could mediate this long-range bridging effect. For one, a liquid droplet would surround chromatin, whereas the lacI–HP1a forms a tightly localized domain at lacO [75]. Further, the bridging effects observed in vitro and in vivo are mediated by direct stereospecific interactions requiring the modular domains of HP1 [74,75,78], whereas liquid droplets pull genomic regions together by proxy via surface tension driven coalescence [69].

The 11 nm nucleosome fibre is a polymer [79,80]. The physics of polymers is well understood and predicts that very small interactions between monomers can strongly influence the whole structure because many small interactions can add up to stabilize different structures [81,82]. For example, when a homo-polymer is placed in a solvent, the interaction of monomers with themselves and the solvent can lead to an incompatibility that results in phase separation of the polymer (i.e. the polymer de-mixes because the energetic cost of mixing the polymer in the solvent is prohibitive), whereupon the polymer collapses into structures such as ordered globules surrounded by a solvent-rich phase [81–85]. Notably, the formation of ordered globules is likely to be one of the consequences of folding of the nucleosome fibre ‘polymer’ in the nucleus. This was concluded from the first Hi-C study, which showed that the average contact probability P_c(s) between two loci at a distance s is a power law, P_c(s)∼1/s^α, having a specific exponent α = 1.08 for genomic distances up to the size of several megabasepairs [86]. This observation led to the notion that the nucleosome fibre adopts a specific state found in ideal polymer chains models, namely the ordered (fractal) globule, which has an exponent α = 1 [85]. Scaling of s⁻¹ is best understood in terms of the nucleosome fibre folding into small ‘globular’ regions which condense into larger globules that then form even larger globules [85] (figure 6). The resulting structure is self-similar (i.e. fractal) over two orders of magnitude from around the level of whole chromosomes down to a scale of a few hundred kilobasepairs [86]. Subsequent Hi-C studies have shown that the exponent can vary depending on chromosome and cell type, indicating that the nucleus contains a complex mixture of differently folded regions, including ordered (fractal) globules, controlled by basic mechanisms of polymer physics that are strongly influenced by chromatin binding proteins and epigenetic modifications [87].

When homo-polymers in solution are cross-linked, they can form condensed structures that are incompatible with the surrounding solvent, resulting in de-mixing and phase separation [88,89]. Highly cross-linked chromatin, such as that found in mitotic chromosomes, forms phase-separated polymer gels [90]. Similarly, phase separation could arise when H3K9me2/3-marked nucleosome ‘polymers’ are ‘cross-linked’ by HP1-mediated bridging within and between the ‘polymer’ fibres. Bridging by HP1 proteins has been definitively demonstrated through elucidation of the three-dimensional structure of H3K9me3-containing di-nucleosomes complexed with human HP1α, β or γ [78] (figure 2). Two H3K9me3 nucleosomes are bridged by a symmetric HP1 dimer via the H3K9me3-CD interaction; the linker DNA between the nucleosomes does not directly interact with HP1 [78]. In vivo FRAP studies indicate that the binding of HP1 to H3K9me3-marked nucleosomes is dynamic and transient, but the rapid and constant exchange with unbound HP1 in the nucleoplasm [55–57] would maintain the bridging interactions. Given the rapid turnover of H3K9me3-bound HP1, HP1-mediated bridging may stabilize or promote pre-existing (condensed) chromatin compaction states rather than inducing these states de novo. Notably, the interaction of HP1α and β with H3K9me3-marked nucleosomal fibres revealed that HP1 dimers can bridge different segments of the same fibre [31,75]. HP1 dimers can also bridge between H3K9me3-marked nucleosomal fibres, i.e. drive inter-fibre interactions [31,75]. Given the flexible HR, HP1-bridging could promote or stabilize the conformation of H3K9me3-marked nucleosome fibres, within and between larger ‘clutches’ of nucleosomes that are known to be enriched in heterochromatin [5,6] (figure 3). Together with osmotic depletion attraction [7], the HP1-mediated bridging between ‘clutches’ would promote their coalescence with the potential to form/stabilize (fractal) globules (figure 3) that could contribute to PPPS of constitutive heterochromatin.

• Download figure
• Open in new tab
• Download PowerPoint

• Download figure
• Open in new tab
• Download PowerPoint

3. Constitutive heterochromatin, heterochromatin-like domains and complexes

Constitutive heterochromatin is found at distinct chromosomal territories—around the centromeres (peri-centromeric), at the telomeres and flanking the peri-nucleolar regions (for reviews, see [91–94]). They are huge, ranging up to 20 Mb in size in mouse and man (table 1). Their size makes them cytologically visible, which enabled Heitz [1] to demonstrate graphically compartmentalization of the eukaryotic genome. In mammals, the bulk surrounds the centromeres (peri-centromeric) and, surprisingly, there is no conserved sequence one can point to that is known to cause nucleation at this site. Instead, it is thought the generally repetitious nature of sequences promotes nucleation of peri-centromeric constitutive heterochromatin [91]. In addition to repetitious DNA, there are proteins, RNAs and epigenetic modifications enriched within constitutive heterochromatin [163,164] (table 1) that are thought to be involved in its nucleation, assembly and propagation [91]. Several of these constituents may contribute to PPPS of constitutive heterochromatin the most likely, but not exclusively, being affinity of homotypic DNA repetitive elements for each other [165], mutual affinity of nucleosomes that share the same modified histones and proteins that bridge between DNA and nucleosome fibres. Along with HP1 proteins, bridging is a property shared by many other proteins that possess two (or more) chromatin/DNA-binding motifs [166,167], any of which could contribute to PPPS of constitutive heterochromatin. For example, methyl binding proteins might contribute by bridging methylated nucleosomal DNA fibres [3]. Putting it short, there are several factors that are likely to contribute to PPPS of cytologically visible constitutive heterochromatin, with the hallmarks HP1 and H3K9me2/3 being the most conserved.

The demonstration that genes encoding HP1 proteins are highly conserved was accompanied by the prediction that HP1-containing heterochromatin-like domains and complexes would exist outside canonical constitutively heterochromatic territories and regulate the cell-to-cell (epigenetic) inheritance of chromatin states [168] (for reviews, see [28,35,95,169]). Many such domains and complexes that share structural components (e.g. HP1) and epigenetic modifications (e.g. H3K9me3) have now been identified (table 1). An attempt at a rough classification as a domain or complex has been made on the basis of size [95], with domains being in the region of 0.1–5Mb in size (table 1). Included in the domains are the odorant receptors [133,134,170], the Krüppel-associated box (KRAB) domain zinc finger (KRAB-ZNF) gene clusters [138] contained within the B4 subcompartment [140]; the protocadherin topologically associated domain (TAD; [142]), somatic cell nuclear transfer (SCNT) reprogramming resistant regions (RRRs; [141]) and the Zscan4 gene cluster [171]. There is evidence that these domains assemble regions of chromosomal DNA involved in regulating cell fate [143,144]. Moreover, perturbation in their assembly and propagation is likely to affect cellular identity [172,173] and SCNT reprogramming efficiency [141]. Heterochromatin-like complexes are small, less than 100 kb and usually only a few kb in size (table 1). They include the 3′ end of the KRAB-ZNF genes [138,139], imprinted gDMRs [153] and the SETDB1-regulated iPS reprogramming resistant regions [150]. The targeted assembly of these domains/complexes contributes to the coarse-grained chromatin-state pattern that characterizes mammalian genomes [174,175].

There are four different but related questions that need to be addressed in order to understand how heterochromatin-like domains/complexes regulate chromatin-templated processes and genome organization. First, how is the assembly of a heterochromatin-like domain/complex nucleated at a particular site in the genome? Second, how does a larger domain form by spreading along the nucleosome fibre from that site? Third, how is the domain/complex epigenetically inherited from one cellular generation to the next? Finally, how do such domains/complexes contribute to compartmentalization of the genome in terms of heterochromatin and euchromatin? Clues to what the answers might look like have come from studies on the heterochromatin-like domains that encompass the KRAB-ZNF gene clusters (table 1) [138]. In humans, the majority of clusters reside on chromosome 19. The KRAB-ZNF genes encode the largest family of transcriptional regulators in higher vertebrates [176] and the general features of the heterochromatin-like domains that encompass the KRAB-ZNF genes clusters are well described. The mammalian HP1 protein, HP1β, and the K9 HMTase SUV39H1 are enriched at the KRAB-ZNF gene clusters and the 20 domains on human chromosome 19 range from 0.1 to 4 Mb in size [138]. HP1β binding is elevated throughout the clusters compared to regions outside the clusters and high-resolution analysis of a specific cluster on chromosome 19, encompassing the ZNF77 and ZNF57 genes, has shown that HP1β binding is co-extensive with H3K9me3 [139]. The formation of large heterochromatin-like domains that encompass the clusters is thought to ‘protect’ the KRAB-ZNF gene repeats as they have expanded during evolution by preventing illegitimate recombination [138], rather than to repress and silence the KRAB-ZNF genes [139,145]. Notably, there are significant variations along a cluster, with enrichment of HP1β at the 3′ end of KRAB-ZNF genes and depletion in the 5′ promoter regions [138,145]. The enrichment observed at the 3′ end has focused attention on the molecular mechanism by which a heterochromatin-like domain can be nucleated, which brings us to the first of four questions that need to be addressed.

3.1. A localized heterochromatin-like complex nucleates formation of the larger domain

Nucleation of the domain probably involves the assembly of a specific, localized, heterochromatin-like complex at the 3′ end of the KRAB-ZNF genes (table 1 and figure 4a). Nucleation is necessarily sequence-specific and intriguingly enough requires binding of a sequence-specific KRAB domain-zinc finger protein (KRAB-ZFP) to the 3′ end of the KRAB-ZNF genes [146,148]. Once bound to its cognate recognition sequence, the KRAB-ZNP recruits the universal co-repressor of KRAB-ZFPs, KRAB-associated protein 1 (KAP1; also known as Tif1β, TRIM28 or KRIP1) [181,184,185]; the KRAB motif of the DNA-bound KRAB-ZFP binds to the RBCC domain of KAP1 [148]. KAP1 acts as a ‘scaffold’ for different enzymatic and structural components that are essential for the nucleation process (figure 4a). A key interaction is that of the SETDB1 HMTase with KAP1 that leads to the generation of the H3K9me3 modification [145,146]. SETDB1 binds the sumoylated form of the KAP1 bromodomain; the sumoylated version is the active, most repressive, form of KAP1 [178]. Sumoylation is mediated intra-molecularly—the KAP1 PHD domain is an E3 ligase that cooperates with UBE2i (also known as UBC9) to transfer SUMO2 [179] to the KAP1 bromodomain. KAP1 also recruits a dimer of HP1 molecules through the PxVxL motif in KAP-1 called the HP1-box [186,187]; KAP1 binds equally well to all three HP1α/β/γ isotypes in biochemical assays [186,187]. Nucleation is reinforced by a specific mechanism that continually replenishes the repressive H3K9me3 modification. Specifically, KAP1 binds to DAXX [188], which is an H3.3-specific chaperone [113–115] that incorporates H3.3 at the 3′ end of the KRAB-ZNF genes [147], whereupon H3.3 is tri-methylated at K9 by SETDB1 [146]. The binding of the ATRX–DAAX complex is enhanced by the known interaction of ATRX with both H3K9me3 and HP1, the former through the ADD domain and the latter through an LxVxL motif, and both interfaces combine to localize ATRX to heterochromatin [180]. The SETDB1–HP1–ATRX complex is stable in vivo: when HP1 is artificially repositioned within the nucleus, both SETDB1 and ATRX are relocated along with it [189].

• Download figure
• Open in new tab
• Download PowerPoint

Several additional enzymatic activities likely to be part of the nucleation process have been revealed using artificially reconstituted systems where a regulatable KRAB domain is targeted to a synthetic sequence that drives a reporter gene [149,190–192]. The small (approx. 1.5 kB) heterochromatin-like complexes generated by the targeted KRAB domains possess elevated levels of both H4K20me3 and H3K9me3 [191,192], indicating the recruitment of a H4K20me3 HMTase and operation of the H3K9me3:HP1:H4K20me3 pathway [182]. The KAP1 recruited by the KRAB domain also binds the NuRD complex that deacetylates histones [190]. There are also increased levels of DNA methylation [191,193,194], which is consistent with biochemical assays showing that KAP1 binds to all three DNA methyltransferases and the DNMT1 cofactor Np95 [157,158]; HP1 also interacts with all three DNA methyltransferases [195,196].

3.3. Genomic bookmarking and epigenetic inheritance

During mitosis, HP1 proteins are removed from chromatin, only to re-associate in the following interphase [202]. Consequently, in order for a heterochromatin-like domain to be epigenetically inherited from one cell generation to the next, the site at which it is nucleated in the genome must be ‘bookmarked’ so that it is re-nucleated at that specific site after mitosis. Nucleation complexes (figure 4a) are excellent candidates for genomic ‘bookmarks’. They are assembled at specific sites through the binding of KRAB-ZFPs to their cognate recognition sequences (figure 4a). The heterochromatin-like complex so targeted nucleates the subsequent ‘spreading’ and formation of a larger domain (figure 4b). A coarse-grained polymer model of genomic bookmarking predicts, as one of three parameters required for robust epigenetic inheritance of chromatin states, a critical density of bookmarks along a chromatin fibre to be 1 or 10 nucleosomes per 400 nucleosomes (ϕ_c ∼ 0.04; [203]). We have made a rough estimate of the density of nucleation sites within the B4 subcompartment using the distribution of KAP1 peaks (figure 5b–d). In the B4 sub-compartment (14 642 kb), there are conservatively 353 KAP1 peaks, which corresponds to one nucleation site per 40 kb, i.e. 200 nucleosomes, where 1 nucleosome is 200 bp. The base of the peak of enrichment for H3K9me3 and HP1β around the KAP1sites gives the size of the nucleation site at approximately 6 kb (figure 5d), which is in agreement with previous studies [138,153]. This indicates a density of bookmarks of approximately 30 nucleosomes per 200 nucleosomes in the B4 subcompartment, well within the critical density defined by the coarse-grained polymer model. The second parameter is that bookmarking involves the sequence-specific recruitment of the machinery that epigenetically modifies chromatin. The KRAB-ZFP that targets the nucleation complex (figure 4a) satisfies that requirement. The third parameter is the operation of a positive feedback mechanism that can spread and establish the domain. This too is met because of the self-reinforcing SUV39H1-mediated spreading (figure 4b) that generates the domain from the nucleation site (figure 4a). The organization of the KRAB-ZNF heterochromatin-like domains therefore satisfies the theoretical requirements for ‘epigenetic domains’ [203].

• Download figure
• Open in new tab
• Download PowerPoint

Replication of the heterochromatin-like domains/complexes during S-phase has been treated in detail elsewhere [95]. Briefly, it involves two complexes called the CAF-1 and SMARCAD1 complexes; both complexes contain KAP1 and HP1 and K9 HTMases that are key components involved in the replication of heterochromatin [100,101,204,205].

3.4. Heterochromatin-like domains/complexes and compartmentalization

Contacts of heterochromatin-like domains/complexes with constitutive heterochromatin are likely to lead to coalescence and promote macroscopic PPPS. When heterochromatin-like domains/complexes are brought into contact with constitutive heterochromatin, where the latter is enriched in high concentrations in H3K9me3, many HP1–H3K9me3 inter-fibre contacts will be formed [31] (figure 6a). The domains/complexes will merge with the large blocks of constitutive heterochromatin because their constituents are, by definition, essentially the same (table 1) and bridging molecules, such as HP1, are unable to distinguish between the translocated domain/complex and the large block of heterochromatin. Bridging and merging of the domain/complex with constitutive heterochromatin would contribute to de-mixing and macroscopic phase separation [88,89] (figures 5a and 6a), with domains/complexes becoming seamlessly part of micrometre-sized cytologically visible constitutive heterochromatin. The ZNF91 KRAB-ZNF cluster that lies within peri-centric heterochromatin [206] most probably undergoes macroscopic phase separation (asterisk in figure 5a) and its interaction with more distal sites will also lead to macro-phase separation (ovals with solid lines in figure 5a). Macro-phase separation may also take place with odorant receptor genes that form very large heterochromatin-like domains up to 5 Mb in size (table 1), 45–50% of which exhibit a preferential localization to constitutive heterochromatin in post-mitotic olfactory sensory neurons [207]. Other KRAB-ZNF clusters spread out along the arm of chromosome 19 [138] are assembled into heterochromatin-like domains where cis- contacts between them (ovals with dotted lines in figure 5a) contribute to the B4 heterochromatic sub-compartment in Hi-C experiments [140].

• Download figure
• Open in new tab
• Download PowerPoint

Heterochromatin-like domains along the arms of chromosome 19 that are flanked by stretches of ‘euchromatic’ nucleosomes (figure 5a) will, we suggest, behave like blocks in a block copolymer (BCP). Polymers that contain blocks of at least two (or more) different types of monomer are called BCPs [208,209], where a block is made up of identical monomers. BCPs share many of the properties of homo-polymers excepting that the covalent bond connecting the different types of block prevents macroscopic phase separation. Instead, BCPs undergo microscopic phase separation when one block becomes highly enriched and incompatible resulting in phase separation to form nanostructures in the range of 0.1–100 nm [209]. For a heterochromatin-like ‘block’, enrichment (leading to incompatibility) would result from HP1-mediated bridging of H3K9me2/3-marked nucleosomes within the ‘block’ (figure 6c) and this could drive micro-phase separation (asterisks in figure 6). cis- contacts mediated by HP1 ‘bridging’ between micro-phase-separated heterochromatin-like ‘blocks’ (figure 6d; ovals with dotted lines in figure 5a) could explain the emergence of the B4 compartmental domain identified in Hi-C maps [140] (figure 5a). Specifically, given that: (i) KRAB–ZNF clusters are assembled into heterochromatin-like domain (blocks) enriched in H3K9me3 and HP1 [138,139], (ii) HP1 can act as a ‘bridge’ between H3K9me3-marked nucleosomes [31,75,78], and (iii) HP1 can also act as a bridge between distantly located loci [74,75], it is unsurprising that the KRAB-ZNF heterochromatin-like domains (blocks) make far cis- contacts that emerge as the B4 subcompartment in Hi-C maps [140]. Importantly, treating chromatin fibres as BCPs has been used with considerable success to accurately simulate contact maps derived from Hi-C experiments [203,210–214].

4. Heterochromatin-like domains/complexes and Hi-C maps

Heterochromatin-like domains/complexes (table 1) are contiguous with the chromatin fibre that is tightly folded within the confines of the nucleus. As part of the fibre, the domains/complexes will fold and experience a myriad of cis- and trans-chromosomal contacts. Many will be transient and of low frequency. Others will occur more frequently and endure, as will be the case for contacts mediated by HP1–H3K9me2/3 interactions (figure 6d). A measure of folding can be assessed by Hi-C, a high-throughput technique that generates contact frequency (Hi-C) maps [215]. Hi-C maps derived from bulk populations of cells revealed the first folding paradigm—the well-known checkerboard (or plaid) pattern of contact enrichment [86] (figure 7a). The pattern is cell-type-specific [217,218] and identifies a set of loci that interact both in cis and trans between megabasepair-sized genomic intervals that can be classified on the basis of computational correlation and principal component analysis as either an A or B compartment, where there is higher contact frequency between genomic loci of the same type (A–A and B–B type contacts) and reduced contact frequency between loci of different types (A–B type contacts) [86]. Characterization showed that A-type compartments carry euchromatic marks, are gene rich, transcriptionally active and early replicating [219]. By contrast, B-type compartments were found to carry heterochromatic marks, are gene poor, late replicating and often associated with the nuclear lamina [219]. On this basis, the checkerboard contact pattern is thought to represent the folding of chromatin into euchromatin (A-type compartments) and heterochromatin (B-type compartments) [86].

• Download figure
• Open in new tab
• Download PowerPoint

Contacts between heterochromatin-like domains/complexes should segregate with the B-type heterochromatic compartments. Large domains do. As explained, the B4 heterochromatic subcompartment contains the KRAB–ZNF genes from the clusters on human chromosome 19 that are assembled into heterochromatin-like domains [138,140] (table 1). However, the checkerboard pattern in Hi-C maps [86,216] (figure 7a) does not show signs of the smaller domains/complexes (table 1). They may now have been observed in Hi-C maps generated from Nipbl^−/− liver cells where chromatin-associated cohesin was depleted [216] (figure 7e). This revealed a finer heterochromatic B-type compartmentalization that emerged from the A-type compartments; canonical B-type compartments did not exhibit fragmentation. Notably, loci in the A-type compartments retained their euchromatic epigenetic marks, i.e. A-type loci do not turn into B-type loci in Nipbl^−/− cells. The absence of the finer B-like compartments in wt Hi-C maps is because B–B contacts involving loci within the finer B-type compartments are continuously disrupted by ATP-dependent loop extrusion, whereupon the loci segregate with the A-type compartments. It is only when chromatin-associated cohesin is depleted that those contacts are re-instated and the finer, innate, B-type compartments emerge from the A-type compartments [216].

Cohesin is a loop-extruding factor (LEF) [212,214]. LEFs attach to the chromatin fibre at random positions and reel it in from both sides, thereby extruding a progressively growing chromatin loop (figure 7, top row) until they either fall off, bump into each other, or bump into extrusion barriers such as CTCF, which define TAD boundaries [212,214,216]. Loop extrusion is an energy-driven, ATP-dependent, process [220]. Based on the known activity of LEFs, a simple explanation for the ‘masking’ of the finer, innate, B-type compartments in wt cells and their emergence from A-type compartments in cohesin-depleted cells can be posited drawing upon polymer physics, where heterochromatin-like domains/complexes are treated as one of the blocks in a BCP [209] (figure 6). The general principles can be illustrated by analytical treatment of a bulk BCP made up of two blocks A and B, where electrostatic interactions are negligible. Micro-phase separation is dependent on three parameters: (i) the volume fraction of the blocks A and B (f_A + f_B = 1), (ii) the total degree of polymerization (N = N_A + N_B), and (iii) the Flory–Huggins parameter, χ_AB [221]. The χ_AB-parameter is key, because it specifies the degree of incompatibility between the A and B blocks and this is what ultimately drives micro-phase separation; a positive χ_AB-parameter means the blocks are incompatible and the larger χ_AB is, the more incompatible they are. The degree of micro-phase separation of the BCP is determined by the segregation product, χ_ABN. Given that the incompatibility is significant (i.e. the energy cost of mixing A and B is high; χ_AB is positive) and χ_ABN > 10.5, the BCP will de-mix and micro-phase separate. Because the χ-parameter varies inversely with thermal energy, an increase in temperature decreases the incompatibility between the constituent blocks and mixing takes place; de-mixing and micro-phase separation resumes upon cooling. In a similar way, LEF-dependent (ATP-consuming) extrusion of a chromatin loop containing a heterochromatin-like domain/complex (block) embedded within euchromatin results in mixing of the domain/complex with euchromatin. Mixing takes place because the energy-driven extrusion disrupts the HP1-mediated ‘bridging’ of H3K9me3-marked nucleosomes (figure 7c); incompatibility is reduced as the ‘block’ becomes less ‘heterochromatic’ and more ‘euchromatic’. Moreover, because phase separation of BCPs is dependent upon volume fraction, mixing has a greater effect on smaller complexes than larger domains (cf. figure 7b with c). Put simply, mixing of smaller heterochromatin-like complexes with larger regions of surrounding euchromatin during loop extrusion will effectively make loops ‘homogeneous’ with respect to euchromatin. This will promote incorporation of smaller complexes into A-type compartments (see blue arrows in the top row of figure 7). When energy-driven loop extrusion is inhibited by cohesin depletion (analogous to cooling a BCP), HP1 ‘bridging’ of H3K9me2/3-marked nucleosomes within the complexes is reconstituted resulting in de-mixing and micro-phase separation (figure 7f). Even the smallest heterochromatin-like complexes are likely to phase separate; polymer simulations and chromatin fragmentation experiments indicate that the minimal size of a chromatin ‘block’ required for phase separation is around 6–20 kb [211,222]. The newly micro-phase-separated complexes (blocks) can then engage in cis- and trans- contacts mediated by HP1 ‘bridging’ of H3K9me2/3-marked nucleosomes (figure 7f and red arrows in the bottom row of figure 7). It is these contacts that emerge from the A-type compartments as the finer B-type heterochromatic compartments [216]. If heterochromatin-like domain/complexes throughout the genome are likewise subject to mixing by LEF activity, compartmentalization detected by Hi-C experiments would be coarser and less well defined in wt cells. This is indeed what is observed when wt and cohesin-depleted Hi-C maps are compared (cf. figure 7a with e) [216].

5. Conclusion and perspectives

HP1-mediated bridging of H3K9me2/3-marked nucleosomes provides a mechanism that connects epigenetics, PPPS and compartmentalization. The bridging of H3K9me2/3-marked nucleosomes by HP1 is involved in nucleation and assembly of heterochromatin-like domains/complexes that epigenetically regulate chromatin-templated processes (table 1 and figures 2–5). Bridging promotes PPPS, where macro-phase separation takes place in constitutive heterochromatin and micro-phase separation with heterochromatin-like domains/complexes (figures 5a and 6). Contacts that result from HP1-mediated bridging of H3K9me2/3-marked nucleosomes are probably detected in Hi-C maps as loci that fall within B-type compartmental domains (figure 7). Based on these data, testable predictions and explanations can be posited that could provide insight into chromatin-templated processes and genome compartmentalization. We also redefine B-type heterochromatic compartmental domains as epigenetic compartmental domains (ECDs) that represent the ‘epigenetic’ component of cellular identity.

Availability of data and materials

Bioinformatic methods: Coordinates for genes within the B4 subcompartment annotation was obtained using the UCSC table browser, which was then reduced to unique positions using simple shell commands. Reads from ChIPseq were trimmed for quality using Trimmomatic SE and aligned to GRCh37/hg19 using Bowtie2 [249]. KAP1 peaks were called using MACS. Read coverage was normalized to reads per genomic context (1×), input-subtracted and plotted using deeptools [250]. The entire pipeline was deployed using Snakemake [251]. Genomic coverage was visualized using the IGV genome browser. The reprogramming-resistant regions (RRRs) annotation was graciously provided by Dr Yi Zhang, which was converted to human coordinates using the UCSC's liftover tool. Hi-C contact frequency map for chromosome 19 in H1-ESCs was obtained by using the 3D genome browser [252].

Data accessibility

See the Availability of data and materials section.

Authors' contributions

P.B.S. conceived of the synthesis presented and wrote the first draft. A.G.N. undertook the bio-informatic analyses and drew the figures. Both authors revised and approved the final manuscript.

Competing interests

The authors declare that they have no competing interests.

Funding

This work was supported by a grant from the Ministry of Education and Science of Russian Federation (grant no. 14.Y26.31.0024); P.B.S. was also supported by Nazarbayev University Grant 090118FD5311. A.G.N. was supported by Deutsche Forschungsgemeinschaft (DFG)—Projekt number 410579311.

Glossary