Epigenetic inheritance and boundary maintenance at human centromeres

Abstract

Centromeres are chromosomal regions that provide the foundation for microtubule attachment during chromosome segregation. Centromeres are epigenetically defined by nucleosomes containing the histone H3 variant CENP-A and, in many organisms, are surrounded by transcriptionally repressed pericentromeric chromatin marked by trimethylation of histone H3 lysine 9 (H3K9me3). Pericentromeric regions facilitate sister chromatid cohesion during mitosis, thereby supporting centromere function. Heterochromatin has a known propensity to spread into adjacent euchromatic domains unless it is properly bounded. Heterochromatin spread into the centromere can disrupt kinetochore function, perturbing chromosome segregation and genome stability. In the fission yeast Schizosaccharomyces pombe, tRNA genes provide barriers to heterochromatin spread at the centromere, the absence of which results in abnormal meiotic chromosome segregation. How heterochromatin-centromere boundaries are established in humans is not understood. We propose models for stable epigenetic inheritance of centromeric domains in humans and discuss advances that will enable discovery of novel regulators of this process.

Introduction

As cells divide, they must precisely distribute their genetic material to daughter cells to stably maintain the genome. This process of chromosome segregation is facilitated by two specialized regions of the chromosome, the centromere, which serves as the assembly site for the mitotic kinetochore, and the pericentromere, which facilitates cohesion and proper sister chromatid alignment in metaphase [1–3]. In humans, centromeric regions are assembled on highly repetitive alpha-satellite DNA sequences that are neither necessary nor sufficient for centromere function [4]. Instead, the centromere is epigenetically defined by the presence of nucleosomes containing the histone H3 variant CENP-A, which are interspersed with histone H3 containing nucleosomes [5]. In contrast, the surrounding pericentromeric regions are repressed by the formation of heterochromatin characterized by di/trimethylation of histone H3 on lysine 9 (H3K9me2/3) [6]. H3K9me3 is bound by Heterochromatin Protein 1 (HP1) and two H3K9 methyltransferases, SUV39H1 and SUV39H2, which methylate additional H3K9 residues and promote chromatin condensation [6]. Careful maintenance of the centromere and pericentromeric heterochromatin is essential for accurate chromosome segregation. Defects in either chromosomal domain can give rise to chromosome segregation errors leading to aneuploidies that promote diseases including Down syndrome and cancer [7].

While heterochromatin plays important roles in chromosome segregation, it has a tendency to spread into nearby euchromatic regions if not properly bounded [8]. This potential for spreading poses an important challenge for centromere stability and maintenance. Loss of heterochromatin boundaries in the fission yeast Schizosaccharomyces pombe leads to heterochromatin spread into the core centromeric domain, causing abnormal chromosomal segregation [9]. Similarly, CENP-A containing domains of the centromere are also spatially fixed over multiple cellular generations, the expansion of which is observed in cancer cells [10, 11]. Despite its importance, our understanding of how human cells establish boundaries between centromeres and pericentric heterochromatin have been limited by a lack of genomic information at repetitive centromeric sequences. In this review, we focus on our current understanding of centromeric boundaries and propose models for epigenetic boundary maintenance at human centromeres. We refer the reader to other recent reviews for details on centromere organization [12, 13], heterochromatin organization [6, 14] and heterochromatin boundary regulation [8].

Epigenetic organization at centromeres

Multiple organisms, including fission yeast, Drosophila, and humans, have regional centromeres, where CENP-A spans multiple kilobases of repetitive sequences flanked by pericentromeric heterochromatin [12]. Studies in humans and Drosophila used immunofluorescence on stretched chromatin fibers to show that the CENP-A containing domain is interspersed with nucleosomes modified by dimethylation of histone H3 at lysine 4 (H3K4me2), a modification normally found in euchromatin [15]. Interestingly, these studies also detected the heterochromatin modification of dimethylated histone H3 lysine 9 (H3K9me2) at the edges of, and occasionally overlapping with, CENP-A containing chromatin (Figure 1) [15]. Notably, H3K9me3 was absent from regions within and immediately flanking the CENP-A regions and, in many fibers, was only observed on one side of centromeric chromatin 5–30 μm away [15]. These studies suggested that in humans, CENP-A chromatin is bordered by H3K9me2 domains, whereas H3K9me3 domains are further away. Using long-read sequencing, recent studies resolved the epigenetic landscape of centromeres further to indicate that CENP-A containing higher order repeat arrays (HORs) are in fact enriched in H3K9me3 by around three-fold compared to non-centromeric regions and demonstrate a notable dip in H3K9me3 at the border (Figure 1) [16, 17]. Interestingly, these CENP-A containing HORs contain zones of low and high CENP-A density. Specifically, in HOR regions of the centromere where DNA cytosine methylation is low, called “centromeric dip regions” (CDR), CENP-A containing nucleosomes are present at a density of one CENP-A containing nucleosome for every four H3-containing nucleosomes (Figure 1) [16, 18]. Outside of these regions but within the same HOR array, the concentration of CENP-A-containing nucleosomes rapidly declines by around six-fold. It remains unclear, however, whether this decline in CENP-A is accompanied by a rise in levels of H3K9me2/3, and whether the H3K9me3 dip at the HOR border corresponds to high levels of H3K9me2. We also do not yet have high resolution genome-wide profiling data of H3K9me2/3 outside of the CENP-A occupying HOR, although multiple studies have shown using immunofluorescence that H3K9me3 signal colocalizes with centromeric regions in human cells [19].

Figure 1:

Epigenetic organization at centromeric regions. At centromeres, CENP-A peaks correspond with regions of reduced CpG methylation and recently expanded repeats and transposable elements (TEs). While H3K9me2 immediately abuts CENP-A, H3K9me3 is proposed to be high within the centromere, with a noticeable dip in levels towards the border of the CENP-A occupying higher-order-repeat. During mitosis, H3.3S31P is also high at regions bordering CENP-A, although the relative levels and patterns of these epigenetic modifications remain unknown (Created with BioRender.com).

Epigenetic maintenance at the centromere

The movement of the replication fork through chromatin requires nucleosome disassembly, following which nucleosomes must be reassembled and new histones incorporated into chromatin. Through this process, core centromeres and the surrounding pericentromeric regions are heritably maintained. Multiple studies suggest that histones containing repressive modifications, as in the pericentromere, can be redeposited locally to the parental and daughter strands during replication [20, 21]. Following replication, histone methyltransferases like SUV39H1/2, which mediate H3K9me3 deposition in pericentric regions, can specifically bind the H3K9me3 modification on nucleosomes distributed during replication and modify nearby nucleosomes by trimethylating H3K9 [6], regenerating pre-replicative levels of heterochromatin by the following G1 phase [22]. In the fission yeast S. pombe, pericentric transcripts processed by the RNAi machinery recruit SUV39H1^Clr4 to maintain H3K9me3 in a positive feedback loop [23]. In humans and mice, pericentric transcripts stabilize SUV39H1 at pericentric regions [19, 24, 25]; however, whether they play a role in its targeting remains unknown. Interestingly, human centromeric regions are replete with transposable elements including endogenous retroviruses (ERVs) that are known to be repressed by a different methyltransferase, SETDB1 [26–29] (Figure 1). SETDB1 localizes to pericentromeric regions in the S phase of the cell cycle in mouse embryonic fibroblasts, where it could potentially monomethylate H3K9 for subsequent methylation by SUV39H enzymes [30]. It remains unclear, however, whether SUV39H1 and SETDB1 act on the same repeat elements within human pericentric regions.

Similar to H3K9me3, CENP-A levels are also diluted in the S phase of the cell cycle, when CENP-A is thought to be stochastically distributed to the parental and daughter strands [31]. Studies in human cells proposed that “placeholder” H3.3 nucleosomes preserve positional information of preexisting CENP-A nucleosomes, which are subsequently replaced in the G1 phase by new CENP-A delivered to the centromere by the histone chaperone HJURP [32–34]. HJURP is recruited to centromeric regions by preexisting CENP-A via its interaction with CENP-C and the Mis18 complex [12, 35]. In this way, both the CENP-A and H3K9me3 occupying domains at centromeres and pericentromeres respectively appear to have a “memory” of their spatial positioning, which could have implications in boundary maintenance, as detailed below.

Models for centromere boundary maintenance

Templated model

Given that both H3K9me3 and CENP-A information is retained after DNA replication, centromeres may use this information as a template for reestablishing epigenetic modifications [20, 35–37]. Consequently, controlled redeposition of both CENP-A and H3K9me3 could be sufficient to maintain a boundary between centromeric domains (Figure 2A). Such “inherently bounded” domains have been simulated for heterochromatin [38, 39]. Consider that if a target site is modified by preexisting H3K9me3, a reader-writer histone methyltransferase can propagate this modification at a rate assigned k+. Random turnover, such as that from replicative dilution, would occur at a rate of k−. One model predicts that if k+/k− ≤ 1.5, heterochromatin domains can be stably maintained in their one-dimensional spatial extent [40]. Interestingly, heterochromatin domains have been proposed to cluster by multiple mechanisms, including association with the nuclear lamina, nucleolus or via phase separation [41]. Given this spatial organization, it is intriguing to also consider a three-dimensional bounded model of heterochromatin maintenance, where H3K9me3-marked domains come together and spread in 3D [42]. In such a model, “k+” would be driven by both self-attraction of the H3K9me3 modification into a compact “self-propagating” domain as well as the rate of “read-write” by histone methyltransferases, whereas “k−” is driven by replicative dilution and histone turnover, as in the previous model. Intuitively, if the rate of deposition of an epigenetic modification is much higher than the rate of replicative dilution, which occurs once per cycle, the epigenetic mark would tend to spread. Therefore, for such “inherently bounded” models to work in a biological context, deposition of the epigenetic modification should be restricted by, for example, limiting the amount of the modifying enzyme or its spatiotemporal regulation. Indeed, HP1 needs to be overexpressed for ectopic heterochromatin induction in fission yeast [43], and the presence of an extra heterochromatin-rich Y chromosome disrupts heterochromatin structure genome-wide in Drosophila [44]. It is interesting to note that these observations could suggest a role for mass action in templated maintenance of heterochromatin at pericentromeres, akin to what has been proposed for CENP-A maintenance in a templated model [12, 45].

Figure 2:

Models for spatial maintenance of epigenetic information at human centromeres. A) Two models are represented here, 1) a templated model where after replicative dilution, templated but controlled addition of CENP-A and H3K9me3 nucleosomes results in stable epigenetic inheritance of core- and peri-centromeric regions and 2) an exclusion model, where the core- and peri-centromeric domains preclude addition of the H3K9me3 and CENP-A, respectively, thereby maintaining the epigenetic status at the regions. B) A genetic boundary model where a DNA element (gray) protects the core centromere containing CENP-A nucleosomes (orange) and H3 containing nucleosomes (blue) from heterochromatin invasion C) Epigenetic boundary model where an epigenetic transition zone (gray) between core- and peri-centromeric domains maintains heterochromatin boundaries (Created with BioRender.com)

Like models for heterochromatin assembly, a predominant model for CENP-A assembly also predicts that preexisting CENP-A directs the assembly of new CENP-A [32, 35−37]. Studies that use fluorescent protein labeling strategies to track new CENP-A deposition relative to preexisting CENP-A showed that new CENP-A is assembled near old CENP-A nucleosomes, thus maintaining the spatial positioning of centromeric regions [10]. Importantly, like pericentromeres, centromeres also exhibit three-dimensional clustering at the nuclear periphery in mouse and human cells [46], and centromeric assembly is restricted to the G1 phase by regulation of the Mis18 complex. Further, the CENP-A chaperone CAL1 is limiting in Drosophila [47] and centromeric assembly is temporally restricted to the G1 phase by regulation of the Mis18 complex in humans [48, 49]. These observations suggest that both centromeres and pericentromeres could self-sustain given specific conditions. However, the templated model has certain drawbacks that argue against this possibility. For example, if CENP-A deposition could occur on either side of the existing CENP-A domains, centromeric drift is likely to occur over a small number of generations [12, 50]. It is possible however, that other mechanisms safeguard templated deposition of CENP-A at the centromere to prevent centromere drift. For example, studies in S. pombe have shown that centromeric DNA sequences promote eviction of histone H3 in the G2 phase of the cell cycle, when CENP-A levels are replenished [51]. Overall, while templated addition of CENP-A and H3K9me3 appears to be important to maintain the epigenetic marks at centromeres, it is unlikely to be sufficient for governing positional information at centromeres.

Exclusion model

Another potential model to maintain distinct domains of CENP-A and H3K9me3 could be an “exclusion model”, where core CENP-A containing domains are not permissive for H3K9me3 modification and pericentric H3K9me3 domains are not permissive for CENP-A deposition (Fig. 2A). One such mechanism could be that centromeric regions contain histones that cannot be methylated by SUV39H1. In fact, CENP-A has very low lysine content and is generally refractory to histone posttranslational modifications [52], which already limits the available H3K9-containing nucleosomes available for methylation at centromeric regions. A recent study where formation of a new centromere was induced by deleting the canonical centromere on chr4 showed that CENP-A seeding is accompanied by H3K9me3 eviction, which also supports an exclusion model for CENP-A maintenance [53]. Additionally, histone H3.3, which is thought to be a placeholder for CENP-A at centromeres [32], is also associated with active chromatin in Drosophila and humans [54]. H3.3 often contains acetylated H3K9 residues and is more likely to be associated with H4 containing acetylation marks, which may render it relatively resistant to H3K9me3 modification [55]. In this way, the presence of CENP-A and H3.3 at centromeres could preclude H3K9me3 formation.

Paradoxically, H3.3 is also found at pericentromeres and telomeres in mouse embryonic stem cells, where it can promote heterochromatin establishment and transcriptional repression, respectively [56–58]. However, pericentromeric H3.3 may be functionally distinct from centromeric H3.3. H3.3-containing nucleosomes immediately flanking centromeres are phosphorylated at serine 31 in mitosis and depleting serine 31 phosphorylation causes genomic instability [59, 60]. While the functional relevance of this modification at pericentromeric regions remains unknown, it is interesting to note that regions flanking centromeres are also enriched in H3K9me2 in chromatin fibers derived from interphase cells [15]. Given that H3K9me2 is a preferred substrate for SETDB1 but not SUV39H1, and a proportion of non-nucleosomal H3.3 is dimethylated at lysine 9 [55], it is tempting to speculate a model where H3.3 is dimethylated by SETDB1 prior to its incorporation into centromeres, which restricts further modification by SUV39H1. Since heterochromatin spreading seems to require a critical density of H3K9me3 [61], such a mechanism could curtail heterochromatinization of centromeric regions.

Although CENP-A domains could be refractory to H3K9me3 deposition, it is unclear whether pericentromeres are similarly refractory to CENP-A deposition. Work in both yeast and human? neocentromeres, which are evolutionarily new centromeres, suggests that pericentromeric domains provide a favorable environment for CENP-A deposition [53, 62]. In contrast, when CENP-A is overexpressed in Drosophila, it is misincorporated into chromosomal arms but excluded from pericentromeric heterochromatin [63]. Studies have also shown that CENP-A can be erroneously deposited into euchromatin at the G1 phase of the cell cycle, after which DNA replication acts as an error-correction mechanism to remove CENP-A in euchromatin and restrict it to centromeres [64]. Therefore, in addition to an exclusion mechanism, CENP-A maintenance may require additional boundary mechanisms. For example, templated addition of CENP-A at centromeric domains could reinforce its exclusion from pericentromeric domains, by limiting the availability of centromere maintenance factors. In this way, multiple boundary mechanisms could work in conjunction to maintain the epigenetic landscape at centromeres and pericentromeres.

Genetic and epigenetic boundaries

In general, two types of heterochromatin borders have been proposed: a “fixed” or static border that is encoded in the underlying DNA sequence and is correlated with a sharp transition from active to repressive histone modifications, and a “negotiable” or dynamic/diffused border, formed by a gradient of active and repressive histone modifications in boundary regions [8, 65] (Figure 2B–C). The centromere on chr1 in the fission yeast S. pombe appears to have a “fixed” border for pericentric heterochromatin, where the CENP-A containing inner cnt repeats and the H3K9me3-rich outer otr repeats are separated by a RNA polymerase III-transcribed tRNA^Ala gene necessary to protect the core from heterochromatin invasion [9]. Accordingly, deletion of the tRNA^Ala gene results in H3K9me3 spreading into the centromeric region in a SUV39H1^Clr4 and HP1^Swi6 dependent manner, causing abnormal meiotic segregation [4]. Centromeres in chr2 and chr3 are also bordered by tRNA gene clusters, which suggests a similar mechanism of boundary maintenance in these regions [66]. Additionally, tRNA genes also protect euchromatin surrounding centromeres from pericentromeric invasion, except for on the q arm of chr1, where inverted repeats (IRC3) coincide with a sharp drop in H3K9me3 [67]. Interestingly, deletion of the tRNA boundary does not induce ectopic CENP-A deposition. Instead, CENP-A appears to have an independent mechanism of boundary maintenance, where the chromatin remodeler RSC causes decompaction of boundary regions to prevent CENP-A from misloading in pericentromeric regions [62].

Studies have also shown the involvement of the JmJC protein Epe1 in regulating heterochromatin at S. pombe centromeres [68]. Reporter genes inserted outside the otr/euchromatin border are repressed when epe1 is knocked out, which suggests that Epe1 protects the genome from pericentromeric heterochromatin invasion [69]. Curiously, reporter genes inserted between cnt/otr regions exhibit variegation in epe1-deficient cells, which indicates an additional role for Epe1 in heterochromatin stability [69]. Indeed, Epe1 is recruited to heterochromatin through its interaction with HP1^Swi6 [70, 71]. However, recent studies suggest that it is degraded by a conserved ubiquitin ligase Cul4-Ddb1 in heterochromatin, thereby restricting its localization to heterochromatin boundaries [72].

Due to the technical challenges associated with studying repetitive sequences at endogenous human centromeres, many studies to date have focused instead on human artificial chromosomes and neocentromeres, both of which can assemble a functional kinetochore onto non-repetitive sequences [73]. One such study found that a balance of the histone acetyltransferase KAT7 and the histone methyltransferase SUV39H1 is required to prevent heterochromatin spreading on synthetic centromeres [74], which is consistent with a “negotiable” border of heterochromatin maintenance. However, SUV39H1 overexpression does not seem to affect CENP-A levels at endogenous centromeres as visualized using immunofluorescence [75], suggesting that heterochromatin may not have negotiable borders at endogenous centromeres.

At neocentromeres, formation of the centromere can remodel the surrounding epigenetic domains to form two distinct domains comprising a core domain that contains modifications associated with active transcription, flanked by heterochromatic pericentromeric domains [76]. While it remains unclear how a boundary is established between the open CENP-A occupying region and the compact heterochromatin domain, this study did reveal that the pericentromeric heterochromatin continued to spread outwardly till it encountered convergent genes that bind the transcription factor CTCF. This is akin to a recent study in budding yeast centromeres where pericentromeric domains, which do not contain heterochromatin and are instead defined by cohesin loading, were found to be delimited by convergent genes [2]. Interestingly, CTCF sites are often found neighboring facultative heterochromatin in mammals, where they could function as heterochromatin insulators [77]. CTCF has also been implicated in centromere clustering and pericentromeric repression in Drosophila [78], making it an appealing candidate for centromeric boundary maintenance.

The T2T genome provides novel insights into human centromeric sequences

Harnessing advances in long read sequencing technologies, the Telomere-to-telomere (T2T) consortium has completely characterized human centromeric sequences providing new avenues for studying boundary maintenance. First, as described in previous sections, the T2T revealed that the majority of CENP-A is assembled at regions of hypomethylated DNA, which could point to a potential role of DNA methyltransferases in centromere boundary maintenance [18]. Consistent with this, prior studies have shown that the DNA methyltransferase DNMT3b associates with CENP-A assembly factors Mis18 and CENP-C, in the absence of which histone modifications are altered in both centromeric and pericentromeric regions [79, 80]. The sequence of chromosomes 3 and 4 centromeres showed that the CENP-A containing higher order repeats on these chromosomes were split by a human satellite 1 (HSAT1) array insertion, suggesting the existence of more than two core/pericentromeric boundaries in these centromeres [18]. Characterizing transcription through centromeres showed that much of centromeric transcription arises from transposable elements that are frequently embedded in centromeric regions outside of the homogenous HOR arrays [81]. Transposable elements have been implicated as heterochromatin barriers in non-centromeric loci in the vertebrate genome raising the question of whether they may play the same role in centromeres [82]. Additionally, the complete genome sequence revealed 676 genes and pseudogenes that lie in human pericentromeric regions, which provokes the question of how these genes remain transcriptionally active in a heterochromatic environment, and whether they share mechanisms of boundary maintenance with CENP-A occupying regions [18]. Perhaps earlier work in Drosophila where functional genes exist in constitutive heterochromatin, and even require heterochromatin localization for their function, will inform future studies into these questions in humans [83]. Finally, assembly of long-read protein occupancy data on the T2T genome showed that CENP-A containing HORs contain H3K9me3[Reference]. If H3K9me3 can freely spread into centromeres as suggested by this observation, perhaps boundaries only exist for CENP-A.

Discussion and Future Directions

In this review, we discuss three mechanisms for boundary maintenance at centromeres, which are not necessarily mutually exclusive. It is interesting to note, however, that recent data suggest that H3K9me3 is enriched in CENP-A containing HORs, which begets the question of whether a heterochromatin boundary even exists at human centromeres. To determine this, we will need to employ DiMeLo-sequencing, a long-read sequencing-based methodology, to resolve genomic occupancy of H3K9me3 and CENP-A distribution on single fibers [16]. With this and other approaches that can span long centromeric repeat regions, we may find that CDRs within CENP-A containing HORs exclude H3K9me3 completely or alternatively, that CENP-A regions are interspersed with nucleosomes modified with heterochromatin marks.

Studies of centromere dependent meiotic drive have shown that genetically stronger centromeres are preferentially incorporated into the oocyte during female meiosis [84]. In this context of centromere drive, pericentromeric heterochromatin equalizes centromeres to offset the cost of strong centromeres [85]. For example, H3K9me3 from pericentromeres may invade a genetically stronger centromere to equalize it with its genetically inferior counterpart, while templated addition of CENP-A safeguards the centromere from unmitigated heterochromatinization. Given that constitutive heterochromatin is established de novo in the paternal pronucleus immediately following fertilization [86], perhaps it is at this stage that centromeric boundaries in paternal and maternal chromosomes are calibrated to equalize genetically distinct centromeres.

While we do not yet completely understand the establishment and maintenance of centromere boundaries, the recent T2T assembly and the advent of long-read sequencing allows us to begin asking important questions. For example, we can use DimeLo-sequencing to map CENP-A and H3K9me3 over development and intersect their maps with those of candidate factors such as DNMT3b and CTCF. Using genome engineering methods, we can perturb putative boundaries and regions replete with transposable elements, to decipher their function, if any, in centromere boundary maintenance. Altogether, recent technological advancements are paving the way to a better understanding of centromere organization and maintenance.