Diverse mechanisms of centromere specification

Summary

The centromere performs a universally conserved function, to accurately partition genetic information upon cell division. Yet, centromeres are among the most rapidly evolving regions of the genome and are bound by varying assortments of centromere-binding factors that are themselves highly divergent at the protein sequence level. A common thread in most species is the dependence on the centromere-specific histone variant CENP-A for the specification of the centromere site. However, CENP-A is not universally required in all species or cell types, making the identification of a general mechanism for centromere specification challenging. In this review, we examine our current understanding of the mechanisms of centromere specification in CENP-A-dependent and independent systems, focusing primarily on recent work.

eTOC blurb

In this review, Mellone and Fachinetti highlight diverse mechanisms that control centromere specification in CENP-A-dependent and -independent systems, with emphasis on recent work. They discuss the possible roles of different components, from centromeric DNA to its associated factors (including transcripts), in centromere identity focusing on multicellular species.

Introduction

Cell division is an essential process in the life of all organisms: it contributes to growth and development, wound repair, response to infection, and cell turnover. When a cell divides, it needs to accurately distribute its chromosomes, the molecular basis of genetic heredity.

The centromere plays a key function in this process. It is the chromosomal docking site for the assembly of the kinetochore, a multimeric protein machinery responsible for the formation of load-bearing spindle-fiber attachments. The centromere and the kinetochore also mediate sensing and regulatory functions for microtubule attachment and chromosome separation during cell division (reviewed by Santaguida and Musacchio ¹). Centromeres can be positioned at single sites (in monocentric species) or be scattered along the entire chromosome (in holocentric species). Defects in centromere formation or kinetochore assembly lead to chromosome mis-segregation and consequently to numerical (aneuploidy) and structural chromosome alterations in the daughter cells (reviewed by Santaguida and Amon ²). The centromeres of multicellular organisms are complex DNA–protein structures frequently built on repetitive DNA sequences and typically composed of a specialized type of chromatin embedded within large flanking heterochromatic regions that form the so-called pericentromere. In addition, centromeres can form de novo at genomic locations not usually associated with centromere activity (also known as neocentromeres), with important implications for speciation (reviewed by Murillo-Pineda and Jansen ³). The existence of neocentromeres as well as functional dicentric chromosomes (chromosomes containing two centromere satellite arrays with only one functionally active; reviewed by Sullivan and Sullivan ⁴) suggests that centromeres are epigenetically regulated chromatin loci (reviewed by Allshire and Karpen ⁵). During the last two decades substantial effort has gone into understanding what defines and maintains the centromere position. Central to these efforts was the identification of CENP-A (Centromeric Protein A), a histone H3 variant^6,7 that acts as the epigenetic mark required to maintain centromere position and function (reviewed by Fukagawa and Earnshaw ⁸). CENP-A has been observed at the centromeres of species from all kingdoms of life, as well as at human neocentromeres. In these species, CENP-A is essential for centromere formation and assembly through the recruitment of additional centromeric proteins (referred to as the ‘constitutive centromere-associated network’, or CCAN, which includes CENP-C, CENP-N, CENP-T and CENP-I in human cells) and stably maintains the centromeric position throughout generations (reviewed by McKinley and Cheeseman ⁹ and briefly discussed in this review).

Although CENP-A plays a key role in centromere specification, other components upstream of CENP-A and/or uncoupled from CENP-A activity have also been implicated in regulating centromere identity. In this review, we focus on aspects of centromere specification that function in parallel, intersect or are independent of the CENP-A pathway.

CENP-A chromatin distribution and organization

CENP-A localizes to centromeres forming distinct foci, visible cytologically in both interphase and mitosis. However, CENP-A is not the only protein present at centromeric sites^10,11 and other H3 variants such as histone H3.1 and H3.3 are also associated with centromeric chromatin^12,13. Estimates of the number of CENP-A molecules found at individual centromeres are surprisingly low relative to the overall size of centromeric DNA: the fission yeast Schizosaccharomyces pombe only contains 26; Drosophila melanogaster 84; chicken DT-40 cells; and human cells ~400. These estimates raise a few questions about how the presence of CENP-A nucleosomes may specify centromere identity.

Analysis of extended chromatin fibers in human and Drosophila cells revealed that CENP-A nucleosomes exhibit discontinuous staining, visible as several ‘spots’ of CENP-A alternating with ‘spots’ of histone H3 ¹² At D. melanogaster’s centromeres, the CENP-A domain ranges in size between 101-171 kb, depending on the centromere, and extended chromatin fibers show ~18 CENP-A spots on average¹⁸. If each Drosophila centromere contains an estimated average of 42 nucleosomes/centromere¹⁵, we can infer that each ‘spot’ consists only of approximately 3 nucleosomes in the span of 6-9 kb (the estimated length of each ‘spot’)¹⁸. Similarly, the CENP-A domain for human centromere 8 is 632-kb long¹⁹. Therefore, the estimated 200 CENP-A nucleosomes¹⁰ would represent only ~6 % of all nucleosomes occupying this centromere (since approximately 3,200 nucleosomes are expected to occupy a region this size). Although more studies are needed to illuminate the precise distribution of CENP-A nucleosomes at the centromere, it appears that just few CENP-A nucleosomes are sufficient to mark the site of centromere formation in both flies and humans. Even though only approximately one fifth of human CENP-A localizes to the centromere, CENP-A is still ~50 fold enriched at the centromere relative to the rest of the genome¹⁰. Thus, on average, no other regions of the genome harbors comparable local enrichment of CENP-A than centromeres.

Two models have proposed ways in which the sparse CENP-A nucleosomes within centromeric chromatin may fold during cell division to promote inner and outer kinetochore nucleation, both implicating the involvement of highly ordered fibers^12,16,20. In human cells, rosette like-structures comprising several CENP-A nucleosomes and spanning up to 300 nm were revealed with super resolution during interphase²¹. Such compacted structures could be mediated by CENP-C through cross-array clustering²² or by CENP-N through stacking of CENP-A nucleosomes^23,24. Given that CENP-A nucleosomes are likely to be quite spaced from one another in vivo, their assembly into stacked and compacted CENP-A nucleosomes would require that they be bundled through the formation of chromatin loops. Such CCAN-mediated clustering of CENP-A nucleosomes could contribute to centromere identity by differentiating centromeric from non-centromeric CENP-A nucleosomes (Figure 1A) (reviewed by Nagpal and Fierz ²⁵). These bundled loops (observed during interphase) could then reorganize to form more ordered structures during cell division^8,12,16,20.

Figure 1. Genetic and epigenetic factors control centromere position.

(A-C) Models for maintenance of human centromeres via CENP-A nucleosomes clustering (A), via a CENP-A self-assembly template mechanism (B) and via CENP-B/CENP-C interaction (C). (D-E) Models for de novo centromere formation of human centromeres via protein/protein interactions (D), changes in the epigenetics landscape (E) and presence of secondary structures and/or topological changes (F). Dashed lines denote proposed interactions, question marks represent untested interactions.

CENP-A stabilization via multiple mechanisms (reviewed by Mitra et al. ²⁶) and, possibly, bundling of CENP-A nucleosomes at centromeric regions, could be an important mechanism to restrict centromere activity at its native position, enabling kinetochore assembly only at these stable sites. An additional model for centromere specification was suggested by a recent study that monitored CENP-A stability during DNA replication, a time when histones disassemble from the parental chromosome and are subsequently re-assembled onto sister chromatids²⁷. Nucleosomes are either maintained, ensuring the inheritance of epigenetic marks throughout replication, or assembled de novo. CENP-A is maintained by the centromeric protein ‘Holliday junction recognition protein’ (HJURP) via a direct interaction with the replication machinery^11,28 and CENP-A itself is required for proper centromere replication²⁹. Nechemia-Arbely et al. observed that the DNA replication machinery evicts the previously deposited, non-centromeric CENP-A and retains only the centromeric pool through a protective mechanism enacted by CENP-C and other CCAN members ¹¹. This data demonstrates that DNA replication can act as an error correction mechanism to restrict CENP-A accumulation to the centromere and to preserve centromere identity.

Given that CENP-A nucleosomes can be found at non-centromeric genomic regions, it is tempting to speculate that CENP-A post-translational modifications (PTMs) found exclusively at the centromere could be another way to restrict CENP-A’s function to centromeric regions. However, the function of most CENP-A PTMs remains controversial and to date no specific PTM associated with only centromeric bound CENP-A have been identified (reviewed by Srivastava et al. ³⁰). Nonetheless, a special importance for correct kinetochore formation has been attributed to mono-methylation of lysine 20 of histone H4; this modification is enriched in CENP-A nucleosomes, but not in pre-nucleosomal CENP-A–H4 complexes or canonical H3 nucleosome due to allosteric inhibition ^31,32.

New CENP-A assembly at the centromere has been proposed to be directed by the presence of preexisting CENP-A nucleosomes though an epigenetic self-assembly loop (Figure 1B) (reviewed by Musacchio and Desai ³³). Analyses of extended chromatin fibers led to the observation that CENP-A deposition at native centromeres occurs in nucleosomes that are in close proximity to preexisting CENP-A nucleosomes³⁴ with flanking pericentric heterochromatin acting as a regional boundary³⁵. These observations suggest a model where new CENP-A deposition is limited by the availability of paired CENP-A-H3 di-nucleosomes, which are connected through CENP-C (and possibly other CCAN proteins)³³. In this model, once all H3 nucleosomes of all di-nucleosomes are replaced, new CENP-A deposition is terminated in G1 phase. During this stage, it has been proposed that part of the Mis18 complex (a key component required for CENP-A deposition, see below) dissociates from the centromere, possibly via ubiquitination³⁶, halting CENP-A reloading^37,38. However, the extent to which hetero-di-nucleosomes are important for new CENP-A deposition remains to be tested.

Recent findings in human cells suggest that the number of centromeric CENP-A molecules is quantitatively determined by the total pool of CENP-A available³⁹, but is independent of both the number of CENP-A molecules previously present at centromeric sites³³ and any specific post-translational modifications on pre-deposited CENP-A⁴⁰ (Figure 1C). This model was further supported by observations in a subpopulation of temporally quiescent CD4⁺ T cells, which lack CENP-A yet are capable of rapidly loading CENP-A at native centromeres³⁹. Interestingly, a complete turnover of all preexisting CENP-A molecules was previously observed in the holocentric species C. elegans⁴¹. These findings suggest that additional factors other than CENP-A provide key marks of centromere identity necessary to maintain certain levels of CENP-A at centromeres.

Deposition of CENP-A by specific assembly factors

CENP-A deposition is controlled by Cyclin Dependent Kinases 1 and 2 (CDK1/2) and Plk1 and mediated by specific chaperones. To date, three CENP-A chaperones have been identified: HJURP in tetrapods, Scm3 in yeast, and CAL1 in Drosophila (reviewed by Westhorpe and Straight ⁴²). HJURP and CAL1 bind pre-nucleosomal CENP-A/H4 dimers preventing tetramerization of CENP-A/H4 ^43–45 and mediating new CENP-A deposition at mitotic exit (for more details see review by McKinley and Cheeseman ⁹).

HJURP is also required to retain CENP-A at the centromere during DNA replication^11,28 and could be critical for CENP-C⁴⁶ and CENP-T assembly⁴⁷. However, the molecular mechanisms underlying both events remain poorly understood. Targeting HJURP to an ectopic location is sufficient to trigger CENP-A assembly both at native^48,49 and artificial chromosomes⁵⁰. De novo centromere formation via tethering of CAL1 was also demonstrated in vivo⁵¹. These studies highlight that CENP-A assembly factors are the driving force for centromere formation.

In vertebrate cells, HJURP binding to centromeres requires the Mis18 complex, an octameric complex of two M18BP1 (Mis18 Binding Protein 1), four Mis18α, and two Mis18b subunits^52–54 (Figure 2). Both HJURP and the Mis18 complex are phospho-regulated to temporally restrict their binding to centromere to prevent premature CENP-A deposition (for more details see review by McKinley and Cheeseman ⁹). The Mis18 complex is highly conserved and essential for CENP-A assembly; notably, it has not been identified in Drosophila, suggesting the evolution of distinct CENP-A reloading pathways (reviewed by Zasadzińska and Foltz ⁵⁵).

Figure 2. Protein/protein interactions govern centromere identity.

Schematic map of reported protein-protein interactions between HJURP, CENP-B, CENP-C, CENP-A, M18BP1, Mis18α and Mis18β. Numbers indicate amino acid (aa) positions for each protein relative to their domains. Specific aa residues are also marked to highlight a particular PTM or interaction. HIKLMN are CENP proteins, CATD = CENP-A targeting domain, HCTD1/2 = HJURP C-terminal domain, SANT-A = SANT associated domain, green and red P = phosphorylation that lead to protein activation or inhibition, respectively. Top left inset: model for new CENP-A deposition based on an interplay of two CENP-C molecules to assemble a new CENP-A nucleosome at the centromere during early G1 phase.

As the Mis18 complex is upstream of CENP-A and HJURP deposition, a key question is how it is recruited to the centromeres. In human cells, this complex does not require HJURP ⁴⁸ nor CENP-A or CENP-B for its recruitment³⁹, while in chicken cells and Xenopus egg extracts, M18BP1 directly connects to CENP-A nucleosomes through residues that are, however, not conserved in humans^56,57. Converging evidence in multiple species indicates that M18BP1 recognizes centromeres via the direct binding to the constitutive centromeric components CENP-C ^39,49,58,59 and CENP-I ⁴⁹, as well as binding to other factors such as MgcRacGAP ⁶⁰ and KAT7 ⁶¹, both of which transiently localize to centromeres (Figure 2). This suggests that M18BP1 is the connecting factor between the centromeric regions and the pre-nucleosomal CENP-A/H4/HJURP complex. Mis18β also directly interacts with the C-terminus of CENP-C ⁶². Remarkably, tethering M18BP1 to an ectopic site is not sufficient to recruit CENP-A in human cells^49,63, suggesting that, in addition to the presence of the Mis18 complex, human HJURP also requires components of the CCAN such as CENP-C and CENP-I to successfully assemble CENP-A de novo ^49,58,64,65. It remains to be understood which factor is responsible for initiating CENP-A deposition at the centromere.

Is CENP-C a simple reader of centromeric chromatin or a marker for centromere formation?

CENP-C is a constitutive centromere component that acts as a “blueprint” of the CCAN network, mediating its assembly and promoting kinetochore formation via an interaction with Mis12 ^66,67 (reviewed by Hara and Fukagawa ⁶⁸). CENP-C depletion leads to a strong reduction of CCAN proteins levels, but not to their immediate and complete loss⁶⁹. CENP-C was proposed to act as a “reader” of CENP-A nucleosomes (reviewed by Ali-Ahmad and Sekulić ⁷⁰). Its replenishment follows rapidly after CENP-A reloading in G1/S phase^69,71 and CENP-C binding to CENP-A nucleosomes is regulated by PTMs^72,73. CENP-C binds to CENP-A nucleosomes through two binding sites (Figure 2), with its central region showing higher affinity (reviewed by Ali-Ahmad and Sekulić ⁷⁰) and is required for CENP-A stabilization along with CENP-N ^74,75.

When bound to an ectopic site, CENP-C is sufficient to trigger the incorporation of endogenous CENP-A at this locus, albeit inefficiently (~30%), in chicken, human and fly cells^49,64,76. At endogenous centromeres, CENP-C is a key factor to mediate de novo CENP-A assembly: following co-depletion and subsequent re-expression of both endogenous CENP-A and CENP-C, only CENP-C had the ability, although limited, to reinstate centromere memory³⁹. This observation suggests that de novo centromere formation is likely initiated by a critical level of CENP-C rather than by CENP-A (Figure 1D). Once CENP-A is deposited, centromere position is efficiently maintained indefinitely by the CENP-A-dependent self-assembly loop³⁹.

How CENP-C is capable of promoting de novo centromere formation still remains to be tested. A likely possibility is that in human cells this ability is mediated by interacting with and recruiting the Mis18 complex to the centromeres^39,49,58,59 (Figure 1 and 2). In contrast, in chicken cells and Xenopus egg extracts, CENP-C is not sufficient for the assembly of the Mis18 complex^56,57. In Drosophila, where the Mis18 complex is absent, CENP-C was shown to directly interact with CAL1^15,77. Interestingly, an interaction of HJURP with CENP-C fragments (but not the full-length protein) was also observed in human cells⁴⁶. This raises the possibility that HJURP may also be important for CENP-C recruitment (and vice versa), although a transient interaction with the Mis18 complex could play a role in new CENP-A reloading. It is important to note that CENP-C not only binds to CENP-A nucleosomes and Mis18, but it also connects to chromatin in other ways: indirectly via CENP-I/H/K/M/L/N/T complexes (reviewed by Ali-Ahmad and Sekulić ⁷⁰) and via CENP-B ^69,78,79 and, directly, through an interaction with DNA (Figure 1 and 2). Although its DNA sequence specificity remains uncertain, CENP-C has DNA binding activity^80–82, which could potentially initiate CENP-C loading during neocentromere formation.

Emerging roles of CENP-B

Many species including mammals, fission yeast, and insects have evolved sequence-specific DNA-binding proteins (CENP-B or CENP-B-related proteins) originating from the domestication of pogo-like transposases (reviewed by Gamba and Fachinetti⁸³). CENP-B is a 2-helix-turn-helix, DNA-binding protein that binds to a 17-bp DNA-sequence motif called the CENP-B box⁸⁴, which is present at all human centromeres except the Y chromosome⁸⁵. Although pogo-like transposases are found in several different species, CENP-B-box-like motifs display a certain degree of conservation, being found only in vertebrates (reviewed by Gamba and Fachinetti⁸³).

Contrary to all other components that constitute the core centromere, CENP-B is not essential, as CENP-B KO mice are viable and CENP-B-deficient functional centromeres exist (e.g. the Y chromosome and neocentromeres in humans, and centromeres of species such as certain equids and some primates) (reviewed by Fukagawa and Earnshaw ⁸, Dumont and Fachinetti ⁸⁶, and Giulotto et al ⁸⁷). Paradoxically, CENP-B evolved to bind centromeric DNA, is widely conserved in vertebrates, and plays functional roles at the centromere.

CENP-B interacts with CENP-A ^79,88 and CENP-C ^78,79 (Figure 2) and it was shown to be sufficient for preserving kinetochore assembly and for the maintenance of chromosome segregation fidelity⁶⁹. Further, the enrichment of CENP-B boxes, CENP-B and other centromeric/kinetochore components is associated with a lower rate of chromosome mis-segregation⁸⁹. However, it is important to note that while CENP-A/B/C are interacting partners, about half of CENP-B molecules do not engage in these interactions⁹⁰. This non-CENP-A/C-interacting CENP-B pool could participate in heterochromatin formation^91,92 via the recruitment of the DAXX chaperone⁹³. CENP-B binding to repetitive DNA has been proposed to be regulated by PTM on CENP-B itself^94,95 and by DNA methylation⁹⁶ and facilitated by Nap1 ⁹⁷ and INMAP ⁹⁸ (Figure 2). However, further studies are required to test if and how these PTMs regulate the interaction of CENP-B with CENP-C.

In addition to its active role in kinetochore function, CENP-B is also required for centromere formation and maintenance. ‘Bottom-up’ assembly strategies, in which large fragments of centromeric DNA of a minimum length of 30-kb are introduced into cells, demonstrated that CENP-B boxes and CENP-B facilitate centromere formation on human artificial chromosome (HACs) (reviewed by Ohzeki et al ⁹⁹). These findings indicate that CENP-B may act as a mark for the centromere identity upstream of CENP-A⁹¹. Along the same line, it was recently shown that under temporal centromere inactivation in which CENP-A level is compromised, CENP-B can specify the ‘memory’ of centromere position on native human centromeres by preserving a critical level of CENP-C and, possibly, by preventing neocentromere formation³⁹ (Figure 1C). The finding that CENP-B enables CENP-A recruitment via CENP-C ³⁹ leads to a possible model for the temporal events occurring during centromere formation, at least at centromeres normally bound by CENP-B: CENP-B-mediated HAC formation occurs first through the recruitment of CENP-C and subsequently of CENP-A (Figure 1D). Notably, a direct interaction between CENP-A and CENP-B has been observed^79,88,91, even though this interaction appears to be insufficient to load and stabilize CENP-A in the absence of CENP-C ^39,75 (Figure 2). Masumoto and colleagues also proposed that CENP-B favors CENP-A incorporation by promoting the formation of a permissive chromatin environment through the direct recruitment of ASH1L, a H3K36 methylase⁹² (Figure 1E).

An alternative and more speculative model to explain such CENP-B-dependent CENP-A deposition arises from studying the topology of centromeric chromatin. The presence of CENP-B at the centromere has been proposed to affect centromere organization by dictating, but not directly preserving, nucleosome positioning along centromere DNA^100–102 and by remodeling centromere architecture^101,103. In addition, the binding of CENP-B to centromeric DNA induces conformational changes to centromeric chromatin by generating a ~60° bending of the DNA^104–106. The ability of CENP-B to bend DNA may be important at centromeres that do not contain DNA sequences able to fold into non-B DNA structures, which have been proposed as possible marks of centromere identity¹⁰⁷ (Figure 1F). The specific structure and biological relevance of the chromatin changes induced by CENP-B still need to be determined.

Centromeric DNA composition and its potential roles

Two nearly universal characteristics of centromeric DNA are its repetitive nature and the low degree of sequence conservation. Despite the striking sequence divergence, a recurring feature of regional centromeres in fungi, plants and animals is the presence of both transposable elements and repeated arrays of satellite DNA that span from several kilobases to megabases and could provide favorable conditions for centromere formation and/or function (reviewed by Gamba and Fachinetti⁸³ and Talbert and Henikoff¹⁰⁸; Figure 3).

Figure 3. Centromere organization in model species.

Schematic showing the DNA organization ¹⁸³for three multicellular species with fully sequenced centromeres, Drosophila melanogaster centromere 3¹⁸, Zea Mays centromere 10 ¹⁸³, and Homo sapiens centromere 8 ¹⁹. A) Drosophila melanogaster centromere 3 is composed of blocks of 10-mer (Prodsat) and 12-mer (dodeca) satellites flanking an island of complex DNA, named Giglio, containing the centromere-enriched enriched retroelement G2/Jockey-3 and the Ribosomal Intergenic Spacer (IGS). B) Z. mays centromere 10 includes two CentC enriched-clusters, called C1 and C2, separated by a retroelement enriched region devoid of CentC (NC). CRs are Centromeric Retroelements 1-6. C) H. sapiens centromere 8 is composed of flanking monomeric α-sat interspersed with Long Interspersed Elements (LINEs), Short Interspersed Elements (SINEs) and other human satellites surrounding a region containing α-sat Higher Order Repeats (HORs). Dotted lines represent intervening sequences with the same organization of those shown. The brackets demarcate the region of the centromere that contains CENP-A and its corresponding length.

In addition to form on repetitive DNA, naturally-occurring centromeres can also emerge de novo, albeit rarely. Such de novo centromeres have been observed in human patients (neocentromeres; reviewed by Marshall et al. ¹⁰⁹) and during speciation in plants and mammals (evolutionary new centromeres or ENCs^110–112). However, neocentromeres display defective segregation, error correction and cohesion in cells-based assays^79,113,114 and ENCs typically accumulate centromeric satellites after their inception during evolution¹¹⁵. These findings suggest that certain intrinsic features of centromeric DNA may facilitate centromere stability or function. Yet, knowledge of what such advantageous features may be has remained elusive in part because of the difficulties associated with functionally and phylogenetically analyze such large, complex, and inaccessible regions.

In humans, all conventional centromeres are positioned on α-satellite (α-sat) DNA, an AT-rich monomer 171 bp long that is repeated head-to-tail¹¹⁶. α-sat DNA is specific to primates and has two basic organizations: 1) monomeric α-sat, which displays variable organization and sequence and 2) ‘higher-order repeats’ (HORs) consisting of divergent monomers forming a larger repeating unit. HORs are tandemly arranged to form large arrays up to several megabases and are thought to evolve through concerted evolution¹¹⁷. HORs arrays differ in sequence and organization between chromosomes and the high degree of polymorphism observed across individuals demonstrates they are rapidly evolving (reviewed by Miga ¹¹⁸ ). Each centromere spans multiple arrays, yet the kinetochore only forms on a single HOR on each chromosomes that is referred to as ‘active’ (reviewed by McNulty and Sullivan ¹¹⁹).

The recent advent of long-read sequencing and assembly technologies has allowed unprecedented access to the DNA sequences of complex centromeres in maize¹²⁰, fission yeast¹²¹, Drosophila¹⁸, and humans^19,122,123. Twenty years since the announcement of the sequencing of the human genome, full centromere assemblies were recently revealed for the sex chromosomes and for chromosome 8, and the remaining centromeres are described in a recent preprint¹²⁴. While the X and Y centromeres contain arrays composed of canonical chromosome-specific HORs^122,123, centromere 8 consists of an HOR array (D8Z2) composed of four distinct HOR types of different lengths. Interestingly, the CENP-A region of centromere 8 shows a high degree of HOR subtype mixing, suggesting that this region is undergoing optimization. While the centromere 8 core is composed of HORs, the flanking satellite blocks are composed of monomeric α-sat DNA displaying more sequence variation¹⁹. HORs make up the centromere 8 of Great Apes as well, whereas in both humans and Great Apes monomeric α-sat is typically relegated to the pericentromere. Interestingly, monomeric α-sat is retained at the centromeres of lower primates¹⁹, suggesting a very recent reorganization of centromeric α-sat through homogenization (reviewed by Balzano and Giunta ¹²⁵). A recent preprint reported the sequences associated with each of the active human centromeres, confirming that CENP-A associates with a single HOR on each chromosome that is depleted of CpG methylation and showing that CENP-A consistently associates with the most recent HOR haplotypes¹²⁶.

In addition to satellite DNA, retroelements (REs) are also associated with centromeres across widely divergent taxa such as fungi, plants, flies, and mammals. In Drosophila melanogaster, the centromeres form on complex DNA ‘islands’ enriched in REs and flanked by asymmetric large arrays of simple satellites¹⁸ (Figure 3). The centromeres of maize and rice are composed of satellite repeats and interspersed centromeric REs^120,127,128 and REs invade maize neocentromeres after their formation¹²⁹. In mammals, REs have been found associated with the centromeres of species such as gibbons^130,131 and the koala¹³², to name just two. Computational models of human centromeres have uncovered REs associated with most centromeres (reviewed by Klein and O’Neill ¹³³). Complete maps for the three centromeres available at the time of this review identified a single LINE (long interspersed nuclear element — a retroelement) present at the active X centromere¹²³, no retroelement insertions at the Y centromere¹²², and LINEs and SINEs (short interspersed nuclear element — a non-autonomous retroelement) scattered at centromere 8, but only within the outer monomeric alpha-satellite region¹⁹. Examination of the phylogeny of pericentromeric and centromeric LINE repeats at the human X centromere supports a model whereby the functional centromere has evolved more recently compared to flanking monomeric α-sat DNA, which might have preceded HORs as the functional primate centromere¹¹⁷. Ongoing sequencing and annotation efforts will further shed light on the prominence of these elements within active centromere regions.

Why REs are common at centromeres is still unexplained. One model is that their activity may spur the formation of de novo centromeres. This model is consistent with the findings that LINEs (along with endogenous retroviruses) are enriched at evolutionary breakpoints with latent centromere activity in marsupials¹³⁴, at the 10q25 neocentromere found on a chromosome 10-derived marker chromosome mardel(10)¹³⁵, and at all ENCs found in donkey¹¹⁵. Furthermore, the presence of LINE elements interspersed with monomeric α-sat at what are considered the ancestral centromere sequences of primates, now relegated to the flanking pericentromere^19,117, is consistent with the possibility that REs contribute to centromere inception, although they later become ‘extradited’ to the pericentromere during satellite homogenization and optimization (reviewed by Wong and Choo ¹³⁶).

Centromeric REs have also been implicated in satellite ontogenesis (reviewed by Klein and O’Neill ¹³³ and Presting ¹³⁷) and have been proposed to promote centromere integrity by maintaining centromeric chromatin transcriptionally active to either facilitate CENP-A maintenance or by generating non-coding RNAs with structural roles (reviewed by Klein and O’Neill ¹³³). In Drosophila embryos, the centromere-enriched LINE-like element G2/Jockey-3 is transcribed at low levels¹⁸, and transcripts emanating from a single, transcriptionally active LINE at the human 10q25 neocentromere influence CENP-A levels and neocentromere function¹³⁵. Although these findings are consistent with the proposed roles for retroelements transcription in centromere integrity, more studies are necessary to better establish these correlations across species.

The vast degree of sequence variation observed at the centromeres across species is puzzling, given the highly conserved function that centromeres perform. However, cell biological and genetic evidence support the notion that centromeres represent battlegrounds of conflict, spurring their rapid evolution, a phenomenon named “centromere drive” (reviewed by Henikoff and Malik ¹³⁸, Rosin and Mellone ¹³⁹, and Lampson and Black ¹⁴⁰). Nonetheless, satellite DNA and transposable elements are nearly ubiquitous, hinting at some type of ‘universal’ role that these elements may confer. What could such shared sequence features be, given the extraordinary divergence of centromeric DNA? Using a computational approach, a study identified the recurrence of dyad symmetries in small (<10bp) centromeric DNA sequence units that are predicted to adopt energetically stable non-B-form DNA structures such as DNA melting and cruciform extrusions, which may induce centromere formation¹⁰⁷ (Figure 1F). Interestingly, in humans these dyad-symmetries were only found at centromeres that lack CENP-B boxes, such as neocentromeres and the Y centromere, suggesting that at all other centromeres the DNA-bending properties of CENP-B may induce secondary structures that mimic non-B DNA. ‘I-motifs’, a type of non-B-form DNA, were observed in vitro for a Drosophila centromeric satellite using NMR spectroscopy ^141,142. Larger topological structures such as DNA loops forming during DNA replication were observed by electron microscopy in a heterologous Xenopus egg extract system with a BAC containing human centromeric DNA¹⁴³. A functional implication of centromeric DNA loops has been proposed during mitosis, where they may act as molecular springs to absorb the microtubule pulling force exerted on centromeric chromatin^144,145. Interestingly, in S. pombe, centromeric DNA inherently destabilizes H3 nucleosomes even when relocated to non-centromeric ectopic sites, presenting yet another mechanism for how centromere sequences could favor centromere formation/propagation: through programmed H3 eviction during the cell cycle¹⁴⁶.

In summary, whether such secondary DNA structures exist in vivo, and what their biological relevance may be remains uncertain, as non-B-form DNA is unlikely to be unique to centromeric regions. The emergence of full centromere sequences for multicellular species are paving the way for the types of functional and evolutionary studies needed to gain new insights into the contributions of centromeric repeats to centromere organization and identity.

Centromere Transcription, Transcripts or Both?

Although it is clear that a single sequence does not dictate the identity of complex centromeres, species-specific assortments of centromeric repeats may evolve to attain an optimal genomic landscape for centromere chromatin homeostasis^127,133. Several lines of evidence support the notion that centromeric repeats are transcriptionally active (see Corless et al. ¹⁴⁷ and Liu et al. ¹⁴⁸ for recent reviews). However, the questions of why centromeres are transcribed — and what the role of centromere-emanating transcripts may be — have yet to be answered.

In fission yeast, while pericentric repeats are processed into small interfering RNAs (siRNAs) via the RNA interference pathway involving Dicer¹⁴⁹, centromere-derived transcripts are immediately degraded by the exosome, suggesting that they are mere byproducts of centromeric transcription unlikely to play a structural role as non-coding RNAs¹⁵⁰. The central core region of S. pombe’s centromere 2 displays extensive low quality RNAPII (RNA polymerase II) transcription¹⁵¹ and its sequence can be extensively altered and still function as a centromere as long as it contains cryptic transcription start sites¹⁵¹. In G2 phase of the cell cycle, a sequence of centromere 2 is ‘programmed’ to lose histone H3 coinciding with the time of Cnp1/CENP-A deposition. Interestingly, this programmed loss of H3 occurs in parallel with the recruitment of elongating RNAPII ¹⁴⁶. Furthermore, a GATA-like transcription factor, Ams2, is required for proper CENP-A accumulation at centromeres in fission yeast¹⁵².

Transcription was also observed at ectopic sites upon de novo centromere formation resulting from DNA tethering of CAL1 to an ectopic location in Drosophila S2 cells. CAL1 promotes this transcription through an interaction with both RNAPII and the histone chaperone FACT ¹⁵³ (Figure 4A). Similarly, an increase in transcription was observed at the site of neocentromere formation in human cultured cells¹¹⁴. While transcription may facilitate the removal of histone H3 in exchange for CENP-A at the centromere, it also runs the risk of removing pre-deposited CENP-A; however, such transcription-coupled loss of CENP-A is prevented by the conserved elongation factor Spt6¹⁵⁴. Collectively, these studies point to a role for transcription in the chromatin reorganization necessary for stable CENP-A incorporation into nucleosomes at centromeres^147,148,153.

Figure 4. Centromere transcription.

Diagram depicting two ways that transcription has been shown to influence centromere function. A) In Drosophila, transcription facilitates histone exchange and CENP-A deposition through the destabilizing effect of transcribing RNA polymerase II (RNAPII) and the histone chaperone FACT, which are recruited by the CENP-A assembly factor CAL1. The elongation factor Spt6 prevents loss of pre-existing CENP-A. Analogous findings have been shown in S. pombe and human cells. B) In human cells, α-satellite (α-sat) transcripts localize to their site of origin in cis, promoting centromere integrity and the deposition of new CENP-A.

As mentioned previously in reference to the possible roles of centromeric retroelements, transcription has been implicated in centromere assembly and integrity also through the formation of centromere-derived non-coding RNAs (Figure 4B). Transcripts emanating from centromere-associated sequences have now been described for a large number of species spanning multiple taxa (reviewed by Grenfell et al. ¹⁵⁵ and Chan and Wong ¹⁵⁶). In human cells, α-sat transcripts are produced from both the centromere and the pericentromere – with those derived from the active HORs being more stable – and they have been observed localized to their cognate DNA sequences¹⁵⁷. α-sat RNAs do not depend on centromere activity¹⁵⁷, yet removal of some centromeric components leads to a moderate increase in centromeric α-sat transcripts¹⁵⁸, particularly at replicating centromeres²⁹. Importantly, depletion of array-specific α-sat transcripts results in defective new CENP-A centromeric assembly at the targeted centromere, reduced level of centromeric CENP-C and, consequently, mitotic defects^157,159. These findings, along with the physical interaction observed between α-sat RNAs and centromere proteins, prompted the speculation that these transcripts function in cis as non-coding RNAs, with an intrinsic sequence-specific role in centromere integrity¹⁵⁷. In contrast, another study that detected α-sat transcripts using single-molecule RNA FISH emphasized instead a role for transcription itself rather than for α-sat non-coding RNAs¹⁵⁸. More work is needed to discriminate between these models and to extend these observations to other species.

Recent genomic analysis of nascent RNAs from a hydatiform human cell line shows very low ongoing transcription coming from α-sat¹⁶⁰, suggesting that α-sat RNAs accumulate and remain stably associated with the centromere¹⁵⁷. During mitosis, α-sat transcription has been proposed to maintain centromeric cohesion¹⁶¹ and to facilitate chromosome segregation in a manner dependent on R-loops¹⁶². Elongating RNA Polymerase II (RNAPII) has been observed co-localizing with the centromeres of metaphase chromosomes in human and Drosophila cells, and in Xenopus egg extracts^{159,163–166}. However, in human cells, RNAPII retention seems to result from persistent cohesion at the centromere rather than from active recruitment of the transcriptional machinery¹⁶⁷. What is puzzling is that RNAPII is presumably lost from centromeres upon removal of cohesion in anaphase¹⁶⁷, yet alpha-satellite transcripts are present throughout the cell cycle¹⁵⁷, including during G1 when new CENP-A is deposited. This suggests that either RNAPII is re-loaded at the centromeres in G1 or that new CENP-A loading at this time is merely regulated by the presence of α-sat transcripts rather than by ongoing active transcription.

It’s unclear if the function of non-coding RNAs and of transcription per se work in synergy or are mutually exclusive mechanisms in the same context¹⁵⁶. One possibility is that transcription provides an optimal chromatin environment for de novo CENP-A deposition and maintenance during centromere establishment, but that centromere-derived transcripts acquire non-coding RNA functions during evolution through, for instance, the establishment of protein-RNA interactions that further stabilize a newly formed centromere¹³³.

Non-canonical centromere specification mechanisms

Centromere divergence reaches well beyond the variation in the primary sequences of centromeric DNA and centromere-binding proteins. The reliance on CENP-A for centromere specification is all but universal, and examples of CENP-A-independent modes of centromere specification and function continue to emerge. Such CENP-A-independent centromere function has been observed in specialized tissues of organisms that otherwise rely on CENP-A for mitotic centromere function as well as in species that display evolutionarily ‘re-invented’ centromere conformations entirely lacking CENP-A orthologs.

In C. elegans, both oocyte meiotic divisions involve a reorganization of the kinetochore in cup-like structures that assemble and mediate chromosome segregation in a CENP-A- and CENP-C-independent manner¹⁶⁸. In Arabidopsis thaliana, expression and localization studies revealed the presence of CENP-A at the centromeres of sperm cells but not at those of egg cells. Following zygote formation, paternal CENP-A is removed, and CENP-A is expressed from both paternal and maternal copies of the gene and deposited at centromeres de novo ¹⁶⁹.

Several orders of holocentric insects, including Lepidoptera (e.g. butterflies and moths), completely lack CENP-A orthologs¹⁷⁰. Many of these lineages also lack CENP-C, yet contain several members of the CCAN, including the histone-fold domain containing protein CENP-T¹⁷¹. Interestingly, chromatin profiling approaches uncovered a mutually exclusive relationship between the position of centromere sites and transcriptional activity in the silkworm Bombyx mori. Perturbations of transcriptional states and analysis of differentially expressed genes in a related species revealed that centromere occupancy can switch for a given position, suggesting a sequence-independent centromere formation mechanism¹⁷².

Holocentric insects are not the only species not relying on CENP-A for centromere formation. Mucor circinelloides, an opportunistic human pathogen, displays monocentric chromosome containing conserved centromere (CENP-T but not CENP-C) and kinetochore proteins (Dsn1 and Mis12) over a short AT-rich DNA sequence motif surrounded by a novel type of retrotransposon. Remarkably, this organism lacks CENP-A or any other centromeric-specific histone variants¹⁷³. As their centromeres share features with both point and regional yeast centromeres, the authors proposed that the centromeres of M. circinelloides are mosaic, a new type of centromere configuration that emerged in early-diverging fungi.

Even more remarkably, the kinetoplastid parasite Trypanosoma brucei encodes for entirely evolutionarily distinct kinetochore proteins (KKT1-25), that even lack the nearly universal CENP-C and Ndc80 proteins¹⁷⁴. T. brucei displays monocentric chromosomes that are large and repetitive¹⁷⁵. Although microtubule-binding proteins that might perform the role of Ndc80 have been identified¹⁷⁶, a protein that may specify centromere position similarly to CENP-A is not yet known. Histone H3 variants in T. brucei are not specific to the centromere¹⁷⁷. A recent study reports the characterization of two of the six proteins that constitutively localize to T. brucei’s centromeres called KKT2 and KKT3 ¹⁷⁸. Both proteins harbor a kinase domain and a polo-like box and have redundant functions in recruiting several other proteins to the kinetochore. The central region of KKT2/3 contains two zinc finger domains, suggesting it can bind DNA directly¹⁷⁸. Together, these examples underscore the diversity of centromere specification mechanisms behind the canonical view centered around CENP-A.

Concluding remarks

The centromere has been the focus of fascination for biologists since its discovery as the primary site of chromosome constriction by Walter Flemming in the 19^th century¹⁷⁹. Today, the centromere is recognized as a fundamental player in genome inheritance, well beyond a simple constriction on the chromosome. The centromere biology field has made tremendous progress in isolating and functionally characterizing centromere-bound proteins and, more recently, in identifying the centromeric DNA sequences of multicellular species. However, many aspects of how the centromere site is specified remain elusive, and since in nature centromeres rarely form de novo, scientists have resorted to several strategies to experimentally create de novo centromeres. Deleting centromeric DNA (reviewed by Murillo-Pineda and Jansen ³, and Hori and Fukagawa ¹⁸⁰), depleting centromeric components (reviewed by Hoffmann and Fachinetti ⁷¹), creating HACs (reviewed by Ohzeki et al. ⁹⁹), or mis-targeting centromeric proteins (reviewed by Hori and Fukagawa ¹⁸⁰) are just a few examples of how scientists are trying to understand the ‘rules’ of what defines a functional and heritable centromere. Even though CENP-A is the most recognized pillar of centromere specification, how it is faithfully directed to centromeres is turning out to be very complex, involving several protein networks and even centromeric RNAs or other epigenetic or genetic features (reviewed by Talbert and Henikoff ^108,181 and Scelfo and Fachinetti ¹⁸²). The existence of organisms that do not rely on CENP-A at all underlies the fact that no factor is universally conserved, as is true for centromeric DNA. Thus, centromeres are figuratively and literally ‘always on the move’ and still hold many secrets that will keep scientists busy for years to come.