MeCP2 and Chromatin Compartmentalization.

Methyl-CpG binding protein 2 (MeCP2) is a multifunctional epigenetic reader playing a role in transcriptional regulation and chromatin structure, which was linked to Rett syndrome in humans. Here, we focus on its isoforms and functional domains, interactions, modifications and mutations found in Rett patients. Finally, we address how these properties regulate and mediate the ability of MeCP2 to orchestrate chromatin compartmentalization and higher order genome architecture.


Introduction
In humans, the two meter long genomic DNA is hierarchically folded to fit inside the membrane-bound micrometer-scale cell nucleus. Individual chromosomes occupy distinct subnuclear territories. The chromosome territories have been proposed to be further subdivided into two mutually excluded compartments called 'A' (active) and 'B' (inactive) with distinct accessibilities. Each compartment was reported to consist of multiple topologically associating domains (TADs) (reviewed in [1]). Within TADs, DNA/chromatin looping was predicted to promote higher DNA interaction frequencies among DNA sites located far apart within the linear DNA molecule (reviewed in [1]).
Epigenetic chromatin modifications, including DNA and histone modifications, were shown to control genome accessibility [2] and, thus, the spatial-temporal gene expression without changing the nucleotide sequence. DNA methylation, established by DNA methyltransferases, blocks the access of multiple factors to DNA, thus creating repressive regions. This DNA modification is read by methyl-CpG binding domain (MBD) protein family, which in addition recruit specific chromatin modifiers (reviewed in [3]). Methyl-CpG binding protein 2 (MeCP2) was the first member of the MBD family to be identified [4] and the most extensively studied one. Hereafter, we will focus on MeCP2 isoforms, domains, interactions, modifications and mutations before moving to its role in higher order chromatin organization.

MeCP2 Isoforms and Domains
The MeCP2 gene is highly conserved in Euteleostomi (bony vertebrates) and in humans is located on the X chromosome. Mutations in the MeCP2 gene were linked to the human neurological disorder Rett syndrome (RTT) [5]. The MeCP2 protein has two isoforms (MeCP2 e1 (exon 1) and MeCP2 e2 (exon 2)) with different amino termini due to alternative splicing and different translational start sites. The two isoforms of MeCP2 are abundantly expressed in the central nervous system, but with different expression levels and distributions in developing and post-natal mouse brains. MeCP2 e1 is Nevertheless, MeCP2 was also described to bind to actively transcribed unmethylated DNA in vivo [17,32] with only a minor portion of MeCP2-bound promoters being highly methylated [32]. A possible explanation would be that MeCP2 folds upon binding to DNA and scans the DNA for suitable binding sites making use for this of its non-specific DNA binding sites [23,33]. Thus, it would only bind non-specifically to active genes to scan the DNA for mCpG binding sites.
Recently, MeCP2 was reported to bind not only mCpG but also mCpApC [34]. The patterns of mCpApC differ between neuronal cell types and may, thus, contribute to cell type specific effects of MeCP2 [35,36].
In addition to binding DNA and methylated cytosines, MeCP2 was proposed to bind to 5-hydroxymethylcytosine (5hmC) in mouse brain [37] and embryonic stem cells [38]. 5hmC is an oxidation product of 5mC and can be further oxidized to 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC) by TET (ten-eleven-translocation) proteins, which might enable active DNA demethylation by different pathways (reviewed in [39]). In addition, 5hmC levels were reported to be differentially distributed between different tissues, much lower than 5mC levels and associated to actively expressed and developmentally regulated genes [40]. Nevertheless, these findings are highly debated, as the results are tissue and cell type dependent [37,38], the recognition mechanism of 5hmC by MeCP2 is unclear and other studies hint to a binding affinity similar to binding unmethylated DNA [41][42][43].
A more indirect way of MeCP2 to repress transcription by DNA binding is the protection of MeCP2 bound 5mC against oxidation to 5hmC by TET enzymes by restricting their access to the methylated cytosine [44]. This was proposed to contribute to restricting transcriptional noise [31] and, in particular, repressing tandem repeat DNA expression [44] and L1 retrotransposition [45][46][47]. TET-mediated L1 activation was shown to be prevented by binding of MeCP2 to 5mC [47].
Summarizing, methylation-specific and unspecific MeCP2 DNA binding are both essential for its function in transcriptional repression and chromatin organization, and its multifunctional domain structure allows the protein to simultaneously bind to DNA and interact with other proteins, which will be described next.

MeCP2 Protein-Protein Interactions
Interactions of MeCP2 with several proteins mediate and regulate its multiple functions in transcriptional regulation, chromatin organization and RNA splicing. An overview of interacting proteins, the interacting MeCP2 regions and the function of these interactions is presented in Figure 1 and Table 1.
One major mechanism by which MeCP2 represses transcription is by recruiting corepressor complexes to methylated DNA. One such complex contains mSin3A and histone deacetylases (HDACs), suggesting that transcriptional repression may in part rely on histone deacetylation [11,12], e.g., by removing active chromatin marks. mSin3A was shown to be the direct MeCP2 binding partner, whereas HDACs showed a weaker binding affinity to MeCP2 and, thus, might bind via mSin3A [11]. Another corepressor complex reported to interact with MeCP2 is the NCoR/SMRT interacting with a small region within the TRD domain, which was thus called NID. The data suggested that MeCP2 recruited NCoR/SMRT to methylated DNA and that this MeCP2 bridge function is disturbed in RTT [20]. Interestingly, binding of Sin3A was not disrupted by NID mutations [20].
In addition to transcriptional repression, MeCP2 might also work as an activator, as it was found associated with the transcriptional activator CREB1 (cyclic AMP-responsive element-binding protein 1) at the promoter of an activated gene [17]. In gene expression analysis from mouse hypothalami, the gain of MeCP2 was shown to result in more transcriptional activation than repression, whereas MeCP2 loss lead to reverse effects [17]. These results are in line with a previous study, where only a minor portion of MeCP2 was found bound to methylated CpGs, but 63% of MeCP2 were bound to actively expressed promoters [32]. In other studies though, MeCP2 was found to track methylated CpGs genome wide [31], as described above.

MeCP2 Post-Translational Modifications
Recently, several MeCP2 post-translational modifications (PTMs) were reported, mostly in large scale proteomic studies focusing on mapping one specific PTM in the whole proteome. In Table 2, experimentally determined MeCP2 modifications are summarized, together with the species in which they were identified, the methods used for identification along with references. A more detailed list can be found on PhosphoSitePlus.org [73], including additional sites only available as curated datasets.   Modifications identified by mass spectrometry (MS) might have unclear localization. x means the method as listed above was used, -means it was not used. * modification numbering according to mouse MeCP2 isoform starting in exon 2 (mouse: 484 aa, human: 486 aa, rat: 492 aa) ** references only exemplary (for more information see PhosphositePlus.org) *** residue numbering according to species mentioned as it differs from mouse.
Most of the modifications were identified in large scale studies and not further validated by any other assay. Furthermore, in most cases no additional information is available regarding their influence on MeCP2 function (e.g., [74][75][76][77]87,88,92]). Many of these PTMs were mapped using a single cell line (e.g., [74,77,88]), and their existence in vivo has not been demonstrated. For these reasons, we will focus here on the more detailed studies providing validation and functional relevance of MeCP2 PTMs, in particular, within the context of chromatin.
The first phosphorylation (phos) site identified on MeCP2 was mapped to the CTD on serine 421. S421phos was found as an upshifted band on Western blot analysis upon membrane depolarization [83,120,121] and occurring exclusively in brain, although MeCP2 was detected in many other tissues [122]. S421A/S424A double mutant mice showed better performance in hippocampal memory tests, enhanced longterm potentiation [114] and increased locomotor activity [84]. Analysis of Mecp2 S421A mice revealed an increased dendritic complexity, and defects in the response to novel experiences [123]. As global S421phos was observed upon membrane depolarization, this modification might not regulate expression of specific genes, but rather be involved in modulating global response to membrane depolarization [123].
Together with S421phos, S80phos within the NTD is one of the most studied MeCP2 phosphorylation sites with functional characterization. In contrast to S421 phosphorylation, serine 80 was reported to be dephosphorylated upon membrane depolarization and S80A mutant mice show decreased locomotor activity [84]. The modification is highly enriched in the brain and ubiquitously distributed similar to total MeCP2 [84]. S80A mutation decreased MeCP2 chromatin binding affinity, although the MeCP2 S80A protein levels and subcellular distribution did not differ relative to the wildtype MeCP2. Thus, it was suggested that the phosphorylation possibly fine-tunes chromatin association [84]. The homeodomain-interacting protein kinases 1 (HIPK1) and 2 (HIPK2) were proposed to be responsible for MeCP2 phosphorylation at serine 80 [68,69].
Another MeCP2 phosphorylation site influencing chromatin binding affinity was identified on tyrosine 120 within the MBD domain of MeCP2. This tyrosine residue is substituted in a RTT patient by aspartic acid [124], which could mimic the phosphorylated state. MeCP2 Y120D mutation was found to cause a decrease in binding affinity of MeCP2 to heterochromatin [28]. This could be explained at the structural level by computational modeling indicating that MeCP2 Y120D drastically reduces MeCP2 affinity for DNA as compared to wildtype MeCP2 [91].
A conserved serine (S164) located at the beginning of ID just after the MBD, was shown to be abundantly phosphorylated in the brain in a developmentally regulated manner [97]. While the phospho-mimicking version S164D showed minor binding to chromatin in live-cell kinetic studies, the phospho-defective mutation S164A had the opposite effect [97]. These results could be explained by in silico modeling of the 3D structure of this phosphorylation site, revealing the addition of negative charge to the protein surface as a consequence of S164 phosphorylation, hence, decreasing DNA binding. Immunofluorescence analysis of wildtype neurons versus MeCP2 S164 mutants revealed that temporal regulation of S164 phosphorylation is required for proper nuclear size and neuronal dendritic branching [97].
In addition to phosphorylation, poly(ADP-ribosyl)ation (PAR) of MeCP2 at the ID and TRD domains was reported to occur in vivo in the mouse brain and to influence heterochromatin structure. The addition of this anionic modification within the two highly cationic MeCP2 protein domains responsible to bind DNA was proposed to lead to a general decrease in DNA binding affinity [70]. Concomitantly, poly(ADP-ribosyl)ation of MeCP2 was shown to reduce binding and clustering of pericentric heterochromatin in cell-based assays, suggesting a role of this PTM in MeCP2 chromatin architecture regulation [70].
Altogether, MeCP2 modifications have been shown to regulate its ability to bind and organize DNA/chromatin, as they change the molecular properties of the respective amino acids, which can be critical depending on the position of the residue within the MeCP2 domains. Yet, as mentioned above, most of the modifications identified in MeCP2 have not been functionally characterized and their role Cells 2020, 9, 878 13 of 31 in RTT is unclear. The next section will address the consequences of MeCP2 mutations occurring in the context of RTT.

MeCP2 RTT Mutations
MeCP2 was shown to be associated with the neurological disorder Rett syndrome (RTT), as mutations in this gene were found in about 80% of RTT patients [5]. RTT affects mostly young girls and is characterized by normal development until 7-18 months of age, followed by a developmental stagnation and decline of higher brain functions [125]. Mutations causing RTT and related neurological disorders have been identified along the entire MeCP2 locus, but effects vary depending on the mutation type and location. Missense and nonsense mutations are the most commonly found and relatively well studied. A collection of all RTT related mutations can be found in the online RettBASE: RettSyndrome.org (http://mecp2.chw.edu.au/cgi-bin/mecp2/search/printGraph.cgi). Figure 2 graphically summarizes the high frequency mutations causing RTT ( Figure 2) and Table 3 describes their phenotypes.
affinity [70]. Concomitantly, poly(ADP-ribosyl)ation of MeCP2 was shown to reduce binding and clustering of pericentric heterochromatin in cell-based assays, suggesting a role of this PTM in MeCP2 chromatin architecture regulation [70].
Altogether, MeCP2 modifications have been shown to regulate its ability to bind and organize DNA/chromatin, as they change the molecular properties of the respective amino acids, which can be critical depending on the position of the residue within the MeCP2 domains. Yet, as mentioned above, most of the modifications identified in MeCP2 have not been functionally characterized and their role in RTT is unclear. The next section will address the consequences of MeCP2 mutations occurring in the context of RTT.

MeCP2 RTT Mutations
MeCP2 was shown to be associated with the neurological disorder Rett syndrome (RTT), as mutations in this gene were found in about 80% of RTT patients [5]. RTT affects mostly young girls and is characterized by normal development until 7-18 months of age, followed by a developmental stagnation and decline of higher brain functions [125]. Mutations causing RTT and related neurological disorders have been identified along the entire MeCP2 locus, but effects vary depending on the mutation type and location. Missense and nonsense mutations are the most commonly found and relatively well studied. A collection of all RTT related mutations can be found in the online RettBASE: RettSyndrome.org (http://mecp2.chw.edu.au/cgibin/mecp2/search/printGraph.cgi). Figure 2 graphically summarizes the high frequency mutations causing RTT ( Figure 2) and Table 3 describes their phenotypes. In the following, we will concentrate on RTT mutations impacting MeCP2 DNA binding and chromatin organization function.
MeCP2 RTT related missense mutations are largely found in the MBD, and a large proportion of these mutations reduce the 5mC binding affinity and, consequently, lead to impaired heterochromatin organization and function in cells [28].
MeCP2 R133 and R111 residues located within the MDB directly contact 5mC, and mutations at either site decrease MeCP2 localization at heterochromatin in vivo albeit to different extent. MeCP2 R111G is a rare RTT mutation found only in one patient, which abolishes MeCP2 localization to heterochromatin [28]. MeCP2 R133 mutation influences the pericentric heterochromatin localization depending on the amino acid substitution. MeCP2 R133C and R133L decrease the enrichment at heterochromatin, whereas R133H promotes it [28,29]. Furthermore, artificially targeting MeCP2 R111G and R133L mutants to pericentric heterochromatin rescued their ability to cluster heterochromatin [29].  T158 is the most frequently found MeCP2 MBD mutation site in RTT patients and two substitutions have been reported, T158M (frequency 419) and T158A (frequency 2). Neurons expressing MeCP2 T158M showed reduced neurite outgrowth and dendritic complexity by down regulating the expression and phosphorylation of transcriptional activator CREB1 [135,136]. Both MeCP2 T158M and T158A proteins show decreased stability, methyl-DNA binding ability and heterochromatin clustering function [28,128,139,145].
TRD is a second mutational hotspot domain in MeCP2. Considering its function in direct interaction with multiple transcriptional repressor complexes (see Figure 1 and Table 1), mutations within this region are considered to influence the recently proposed MeCP2 'bridge' function between repressors and chromatin [20,146].
R306C is the most frequent missense mutation found within the MeCP2 TRD. Mutant mice expressing MeCP2 R306C showed typical RTT phenotype: hind limb clasping, impaired mobility and motor coordination, reduced brain weight and size [147]. This mutation did not influence the MeCP2 methyl-DNA binding ability in vitro [128], but showed decreased MeCP2 DNA occupancy in vivo [147], and lack of interaction with NCoR/SMRT [85]. R306C also abolished (neuronal activity-dependent) phosphorylation at the nearby T308 residue. The effect of losing T308 phosphorylation was tested by creating a MeCP2 T308A knock-in mouse model and the analysis of these mutant mice indicated that it contributes to some of the neurological deficits in RTT [85]. Yet, it is still unclear whether the mutation of residue R306 has an influence on chromatin structure.
In addition to missense mutations, several nonsense RTT mutations have been described within the ID or the TRD. In general, these truncations showed decreased protein stability in vivo and DNA binding affinity in vitro [141].
MeCP2 R168X generates a truncated protein with a deletion of the complete TRD and C-terminal region. Male and female mice with R168X expression showed typical RTT phenotype, but little is known about the underlying mechanism. Although the entire MBD is retained, MeCP2 R168X has impaired ability to form higher order structures as tested by in vitro nucleosomal array (NA) assays [140].
The functional importance of the MeCP2 AT-hooks is highlighted by a comparative study in mice expressing either MeCP2 R270X or MeCP2 G273X (a truncation found in only one male RTT patient), which yielded a different developmental rate and phenotypic progression [148]. MeCP2 R270X mutant mice survived less time than MeCP2 G273X (85 days and 201 days, respectively) due to a disrupted AT-hook 2 (aa 264-273) in the MeCP2 R270 truncation. AT-hook 2 disruption decreased the ability of MeCP2 to promote oligomerization of NA in vitro and mislocalization of chromatin-remodeling protein ATRX in vivo [144].
In summary, the severe phenotypes of RTT patient mutations described above emphasize how essential protein stability, DNA/methyl cytosine binding, interactions with other proteins and ultimately chromatin organization are for proper MeCP2 function in vivo.

MeCP2 in Higher Order Chromatin Compartmentalization
MeCP2 is a multifunctional epigenetic reader regulated at multiple levels including, as reviewed above, specific isoforms, interacting factors, post-translational modifications and their interplay within the chromatin context. Yet, it is not well understood how MeCP2 orchestrates genome architecture. In this section, we will summarize findings related to the role of MeCP2 on higher order chromatin organization and propose a unifying model.

MeCP2 and Chromatin Looping
MeCP2 was described to compact nucleosomal arrays (NAs) [140] and to form loops involving undersaturated (DNA partially occupied by nucleosomes) nucleosomal arrays in vitro [24]. While wildtype MeCP2 was shown to form nucleosome-MeCP2-nucleosome 'sandwich' structures bringing two nucleosomes closely together, the RTT truncation mutant R294X was shown to form DNA-MeCP2-DNA 'stem' motifs, bringing nucleosome entry and exit site in close proximity [24].
Interestingly, the RTT mutation R106W, which does not bind to methylated DNA (see Table 3), did not induce any chromatin conformations. Thus, MeCP2 loop formation was proposed to proceed in a two step process involving methylation-dependent DNA binding followed by methylation-independent interactions between MeCP2 CTD and nucleosomes [24]. Of note, MeCP2 was also shown to bind to four-way junction DNA, which has a similar conformation as the 'stem' motif [24,149]. Importantly, MeCP2 was proposed to be involved in the formation of a silent chromatin loop at the imprinted Dlx5-Dlx6 locus, and this loop is lost in RTT [150].
Current models though propose that the chromatin loops are promoted by 'loop extrusion', where cohesin extrudes chromatin until it encounters boundaries created by CTCF (CCCTC-binding factor) binding [151,152], albeit the underlying mechanism is unclear. MeCP2 has been reported to interact with ATRX and cohesin subunits SMC1 (structural maintenance of chromosomes protein 1) and SMC3 using coimmunoprecipitation experiments in mouse forebrain [64]. ATRX was proposed to create an extended DNA linker region for CTCF binding [153], and CTCF was reported to promote loop formation together with the cohesin complex [154,155]. Of note, the interaction of MeCP2 with cohesin subunit SMC3 was found to be induced by S229 phosphorylation and inhibited by the S80 phosphorylation of MeCP2 [19], indicating a role of MeCP2 and its modifications on chromatin looping.
Contrary to MeCP2, it was frequently described that CTCF shows a decreased binding affinity to methylated DNA [156,157]. Wang et al. found based on DNA methylome data from 13 cell types that immortalized cells displaying DNA hypermethylation had elevated CTCF level [158]. This might constitute a compensatory mechanism for lower CTCF binding due to hypermethylation [158] and may rescue CTCF mediated insulation of known tumor suppressor genes against methylation dependent silencing [159,160]. Furthermore, the 5mC oxidation product 5caC was found to enhance CTCF association to DNA and facilitate binding to low affinity CTCF binding motifs [161,162]. As 5mC oxidation to 5hmC followed by further oxidation to 5fC and 5caC was proposed to enable cytosine demethylation ( [163], see above), CTCF association to 5caC hints to a CTCF-based mechanism reinforcing its own binding [162]. As MeCP2 has been shown to protect 5mC from TET mediated oxidation [44], MeCP2 might, thus, influence CTCF binding and DNA loop formation.
As a conclusion, the potential structural and functional interactions between MeCP2 and CTCF are still poorly understood and need to be clarified in further studies, especially considering the importance of both proteins in regulation of chromatin structure and gene expression. Mechanistically, loop formation has been proposed to give rise to TADs, whose boundaries are at least in part defined by CTCF and cohesin [164]. Although there is no evidence directly showing any effects of MeCP2 on TADs, it is still noteworthy to explore if and how MeCP2 organizes TADs, considering the role of MeCP2 on chromatin looping and counteracting CTCF binding.

MeCP2 and Heterochromatin Compartmentalization
Quantification of MeCP2 in neurons showed it to be nearly as abundant as histone octamers [31]. In MeCP2 deficient neurons, the level of histone H1 doubled [31], whereas in wild type neurons, the H1 level was half of the amount of H1 in other cells [165], indicating that MeCP2 acts as a histone H1-like chromatin linker. Accordingly, MeCP2 was shown: to accelerate H1 exchange in vivo, hence decreasing dwell time of histone H1 in chromatin [166]; to have a similar mobility to H1 in vivo; and to share with H1 an overlapping binding site on nucleosomes in vitro [31,[166][167][168]. In fact, by in vitro fluorescence anisotropy assays, it was observed that MeCP2 could replace histone H1 from chromatin [166,169] and globally alter the chromatin state. MeCP2 deficiency was also reported to affect global chromatin composition and state by increasing H3 acetylation [31]. Hence, MeCP2 was proposed to dampen transcriptional noise from repetitive DNA elements including satellite DNA in a DNA methylation-dependent manner [31]. MeCP2 was also shown to increase H3K9me2 at the promoter of the SIRT1 gene [170] and MeCP2 inhibition was shown to decrease H3K27me3 levels on silenced gene promoters [171], indicating a role of MeCP2 in facultative heterochromatin regulation.
On the other hand, MeCP2 was also reported to activate gene expression by binding the transcription activator CREB1 in euchromatin as mentioned above [17].
Based on the cytological analysis of DNA condensation level, eukaryotic chromatin can be broadly divided into the actively transcribed, open euchromatin and the densely packed, repressed heterochromatin. Heterochromatin is rich in methylated cytosines, which can be specifically recognized by multiple epigenetic readers including MeCP2.
In vivo MeCP2 was shown to be enriched at pericentric heterochromatin [4]. Pericentric heterochromatin is localized in proximity to the centromere and enriched in AT-rich major satellite DNA repeats occupying about 10% of the mouse genome [172]. In the interphase nucleus, different pericentric heterochromatin regions were shown to fuse and form locally extremely condensed regions called chromocenters [173], a distinct, supra-chromosomal, membraneless heterochromatin domain also enriched in HP1 and H3K9me3. As MeCP2 was shown to interact with HP1 and to colocalize with HP1 in heterochromatin [51] (Table 1), this enables a cross talk between histone methylation and DNA methylation pathways strengthening heterochromatin formation. The influence of MeCP2 in chromocenter organization was demonstrated by Brero et al. [174], showing that, during myogenic differentiation, the number of chromocenters decreased, i.e., heterochromatin clustered into larger compartments, concomitantly with increased MeCP2 level and genome methylation. Of note, ectopic MeCP2-YFP could promote pericentric heterochromatin clustering even in the absence of cellular differentiation.
Expanding from this initial study, the role of MeCP2 during neuronal differentiation was analyzed comparing wild type and MeCP2 deficient mouse embryonic stem cells [175]. An increased MeCP2 level and enrichment at chromocenters was measured during neuronal differentiation, together with significant chromocenter clustering. Accordingly, the chromocenter clustering function was impaired in the MeCP2 deficient mouse embryonic stem cells. Furthermore, ectopic expression of MeCP2 with RTT mutations showed impaired heterochromatin accumulation and decreased chromatin clustering function [28], suggesting a role of heterochromatin organization in RTT.
At the molecular level, using in vitro nucleosomal arrays, Georgel et al. in 2003 observed by electron microscopy that NAs formed both extensively condensed ellipsoidal particles and oligomeric suprastructures upon addition of MeCP2. This was independent of DNA methylation and relying upon regions downstream of MBD, as R168X truncation mutant failed to assemble oligomeric suprastructures [140,176]. This was further confirmed by the observation that the ID, TRD and CTD alpha could bind and compact NAs and that R270X and R273X, truncated within the TRD and missing the whole CTD, could not compact and oligomerize NAs [23,144]. These facts could in part explain how nonsense mutations of MeCP2 lead to severe symptoms of RTT.
It is still far from clear how MeCP2 organizes heterochromatin structure, but emerging evidence suggests a role of phase separation in heterochromatin condensation.

Phase Separation and Heterochromatin Condensation
Compartmentalization of heterochromatin within the cell nucleus is evolutionarily conserved. Recent evidence indicates that in eukaryotic cells, non-membrane bound compartments are present in both the cytoplasm (e.g., stress granules [177]) and the nucleus (e.g., nucleoli [178]) and chromocenters [174]. Although described decades ago, how such membraneless compartments dynamically form and function has been unclear. In 2009, Brangwynne et al. [179] proposed that germline P granules are liquid droplets with fast exchange dynamics, fusion and fission properties and round appearance formed by liquid-liquid phase separation [180], suggesting a possible mechanism for chromatin organization.
Proteins that could undergo phase separation often contain intrinsically disordered regions (IDRs) or low complexity regions (LCRs) [181]. Chemically, the process is based on weak forces (mostly hydrophobic interactions) and multiple electrostatic interactions including charge-charge, charge-π, π-π stacking interactions and hydrogen bonds [182][183][184]. Recent work implicates liquid-liquid phase separation in the nuclear organization, leading to the formation of various subdomains with distinct properties.
An earlier in vitro cryo-electron microscopy study of purified simian virus 40 minichromosome showed that the purified viral minichromosome was condensed into 10 nm globules. In high-salt buffer, these globules showed the ability to fuse, whereas at low salt conditions, they opened into filaments and nucleosome strings [185]. Maeshima et al. observed that NAs self-associate into globular oligomers in a cation-induced manner, which can be modulated by histone H1 and linker DNA [186]. Altogether, these studies suggest a 'liquid drop' model of chromosome organization.
More recently, NAs were shown to undergo histone tail dependent liquid-like phase separation in physiologic salt conditions, a phenomenon promoted by histone H1, controlled by linker DNA length and disrupted by histone acetylation [187]. Furthermore, NAs with acetylated histones could form a new liquid phase with multi-bromodomain proteins, and these droplets had distinct properties compared to droplets formed by unmodified histones.
Two recent studies found that HP1alpha protein could drive chromocenter formation via phase separation [188,189], linking phase separation to chromocenter structure and dynamics via multivalent interactions. Interestingly, MeCP2 was shown to have a highly unstructured nature [33] and to induce the formation of very large heterochromatin clusters when compared with HP1 [174]. Altogether, these studies suggest a framework to understand chromatin compartmentalization based on liquid-liquid phase separation.

Model for MeCP2 Function in Chromocenter Clustering
In summary, as described in the sections above, MeCP2 interacts with DNA, methyl cytosines and nucleosomes via separate domains, and interacts with several chromatin proteins. Furthermore, MeCP2 can replace linker histone H1 and has a highly unstructured nature. Firstly, like most proteins that could form liquid phase separation, MeCP2 intrinsically disordered regions consist of mainly positively charged residues (arginines, histidines and lysines). These residues form electrostatic interactions with the negatively charged amino acids in other proteins and phosphates in DNA or RNA, thus, building multivalent protein-protein/DNA/RNA interactions. Such interactions locally enrich or deplete factors in a dynamic manner, while being sensitive to post-translational modifications (a described above). Secondly, MeCP2 foci exhibit liquid-like properties in vivo. Brero et al. showed that MeCP2 forms round-shaped foci within the cell nucleus and foci in close proximity tend to fuse over time. Furthermore, during mitosis, these chromatin clusters undergo fission and reform again after cells have divided. MeCP2 was also shown to promote chromocenter clustering in a dose dependent manner [174]. In addition, purified MeCP2 protein alone showed oblate ellipsoid appearance in electron microscopy analysis [24]. Hence, and as depicted graphically in Figure 3, we propose that the multivalent interactions with proteins and DNA/nucleosomes, together with its ability to oligomerize and possibly create by itself phase separated compartments, altogether contribute to the in vivo ability of MeCP2 to dynamically and efficiently cluster and compartmentalize heterochromatin. chromatin clusters undergo fission and reform again after cells have divided. MeCP2 was also shown to promote chromocenter clustering in a dose dependent manner [174]. In addition, purified MeCP2 protein alone showed oblate ellipsoid appearance in electron microscopy analysis [24]. Hence, and as depicted graphically in Figure 3, we propose that the multivalent interactions with proteins and DNA/nucleosomes, together with its ability to oligomerize and possibly create by itself phase separated compartments, altogether contribute to the in vivo ability of MeCP2 to dynamically and efficiently cluster and compartmentalize heterochromatin.

Acknowledgments:
We thank all the past and present members of our laboratory for their many contributions along the years and our collaborators.

Conflicts of Interest:
The authors declare no conflict of interest.