Structure of the human inner kinetochore bound to a centromeric CENP-A nucleosome

Abstract

Kinetochores assemble onto specialized centromeric CENP-A nucleosomes (CENP-A^Nuc) to mediate attachments between chromosomes and the mitotic spindle. We describe cryo-EM structures of the human inner kinetochore CCAN (Constitutive Centromere Associated Network) complex bound to CENP-A^Nuc reconstituted onto α-satellite DNA. CCAN forms edge-on contacts with CENP-A^Nuc, while a linker DNA segment of the α-satellite repeat emerges from the fully-wrapped end of the nucleosome to thread through the central CENP-LN channel that tightly grips the DNA. The CENP-TWSX histone-fold module further augments DNA binding and partially wraps the linker DNA in a manner reminiscent of canonical nucleosomes. Our study suggests that the topological entrapment of the linker DNA by CCAN provides a robust mechanism by which kinetochores withstand both pushing and pulling forces exerted by the mitotic spindle.

The centromere is a specialized genetic locus that interacts with the mitotic spindle to facilitate chromosome segregation. Centromeres of humans and other primates comprise multiple copies of the conserved ~171 bp α-satellite repeat that contains the histone H3 variant CENP-A (1, 2). Kinetochores assemble specifically onto centromeres to form large macromolecular machines that directly mediate chromosome segregation (3). A critical function of the kinetochore is to generate a load-bearing attachment between chromosomes and the spindle apparatus. How this function is achieved at the molecular level is unclear.

Throughout the cell cycle, CENP-A nucleosomes specify the constitutive association of the 16-subunit inner kinetochore CCAN complex with centromeric chromatin, whereas the ten-subunit KMN network of the outer kinetochore assembles onto CCAN during mitosis to couple CCAN to microtubules for spindle-mediated chromosome movement. Two human CCAN proteins, CENP-C and CENP-N, directly interact with CENP-A^Nuc (4, 5). CENP-C interacts with the C-terminus of CENP-A, and histones H2A and H2B (6), whereas CENP-N binds the L1 loop and adjacent DNA gyre of CENP-A^Nuc (7–11). The 17 bp B-box motif that interacts with CENP-B is the only DNA sequence within the α-satellite repeat that is specifically recognized by a centromere-associated protein.

In budding yeast, a single kinetochore complex associates with a 'point' centromere accommodating a sole CENP-A nucleosome. In contrast, the regional kinetochores of humans and other metazoans are generated from multiple copies of the CCAN and KMN network complexes to form large disk-like structures. Here, we determined the cryo-EM reconstruction of a CCAN-CENP-A^Nuc complex using a native 171 bp α-satellite repeat DNA, representing the repeating modular unit of a human inner kinetochore.

Results

CCAN is assembled from a network of interdependent modules

Human CCAN comprises four modules: CENP-LN, CENP-HIKM, CENP-OPQUR and CENP-TWS, that are additionally linked by the largely disordered CENP-C through its interactions with CENP-LN and CENP-HIKM (7, 12, 13). We expressed and purified CCAN sub-complexes, and used the N-terminal half of CENP-C (CENP-C^N) that contains all known CCAN binding sites and the central domain essential for centromere localization (14, 15) (fig. S1A). Phosphoproteomic analysis indicated an interphase phosphorylation state of CCAN, as judged by the absence of CDK1-phosphorylation sites on CENP-T (3). Initially, we focused on the 11-subunit CCAN core complex (CENP-OPQUR-LN-HIKM) which could assemble without CENP-C^N (termed CCAN^ΔCT) (fig. S1B). We determined the cryo-EM structure of CCAN^ΔCT at 3.2 Å resolution (Fig. 1, fig. S2A-D and Table S1). The cross-linked structure was identical to a non-cross-linked reconstruction obtained at lower resolution (fig. S2E). We also determined crystal structures of the CENP-OPQUR and CENP-HIK^Head modules (fig. S2F and Table S2).

Figure 1. CCAN is assembled from a network of interdependent modules.

(A, B) Cryo-EM density map (A) and molecular architecture (B) of the apo-CCAN^ΔCT fitted into the cryo-EM map with resolved OPQUR^Foot and HIK^Head. (C) Atomic model of the apo-CCAN^ΔCT with details of the CENP-N-CENP-M interface as well as CENP-O binding to the CENP-HI pocket shown as insets. (D) Cartoon schematic of the CCAN^ΔCT modules.

The architecture of CCAN^ΔCT resembles budding yeast CCAN (ScCCAN) (16, 17) with CENP-OPQUR and CENP-HIKM 'lobes' assembled on either flank of a central CENP-LN module (Fig. 1). The arc-like CENP-LN module is the structural keystone of CCAN^ΔCT, and generates a deep channel that is extended on one side by the N-terminal coiled-coil of CENP-QU. The RWD domains of CENP-O and CENP-P, with the C-terminal regions of CENP-QU and CENP-R, organize into a 'cap' domain above CENP-N. In human CCAN^ΔCT, CENP-R partially substitutes for yeast Nkp1 and Nkp2. CENP-M, that is also unique to vertebrate CCAN, adopts a pseudo-GTPase-fold and forms a stable complex with CENP-HIK (18). CENP-M makes extensive contacts with the HEAT domain of CENP-I and the N-terminal helical bundle of CENP-H and CENP-K, and in the context of CCAN, with CENP-L (Fig. 1). CENP-M has a small deletion that removes the switch I region and an adjacent β-strand (18). A conserved β-strand of CENP-N occupies this position to augment the central β-sheet of CENP-M (Fig. 1C). An N-terminal α-helix of CENP-O, connected through a flexible linker to CENP-OPQUR, interacts with a conserved pocket formed by CENP-I and CENP-K (Fig. 1C). Disrupting this interaction by deleting the N-terminal 35 residues of CENP-O destabilizes CCAN^ΔCT assembly, as does mutating residues on CENP-N that bind CENP-M (fig. S3A, B). Further supporting the human CCAN architecture described here is its consistency with the interdependent localization of CCAN subunits to centromeres (12, 13, 18, 19).

Two peripheral and conformationally flexible regions of CCAN are poorly resolved in the cryo-EM reconstruction: the CENP-HIK^Head and N-terminal region of CENP-QU (CENP-QU^Foot) (Fig. 1). Focused 3D classification of CENP-HIK^Head identified multiple poses, two of which readily accommodate the CENP-HIK^Head crystal structure (figs S2F, S3C).

Reconstitution of CCAN-CENP-A^Nuc onto a 171 bp α-satellite repeat

To assemble a complete CCAN^Δ_T-α-satellite nucleosome complex, we reconstituted CCAN^ΔT complexes with CENP-C^N and CENP-A^Nuc using 171 bp α-satellite repeat DNA (CENP-A^Nuc-171). The α-satellite DNA matches an α-satellite repeat from human chromosome 2, and also shares 93% sequence identity with at least 14 other human centromeres (20) (fig. S4A). We used two separate 171 bp α-satellite sequences, differing in register by 22 bp (termed AS1 and AS2) (fig. S4A). To define the exact nucleosome positioning of the AS2 sequence, we determined a cryo-EM structure of a CENP-A^Nuc-CENP-C^N complex at 2.4 A resolution (fig. S4B-D and Table S1). The dyad axis of the nucleosome is equivalent to NCPs comprising X-chromosome α-satellite DNA, with which it shares 71% sequence identity (10, 21) (fig. S4A). This also matches nucleosome positions defined by in vivo ChIP micrococcal nuclease (MNase)-seq analyses of human centromeric chromatin (21, 22). Relative to histone H3 nucleosomes (H3^Nuc), CENP-A^Nuc exhibits continuous conformational flexibility of its terminal DNA segments as shown by a 3D variability analysis. The DNA ends fluctuate between fully wrapped at both ends, to being unwrapped by ~22 bp at one end, with the majority of particles in intermediate states (fig. S4E and Movie S1). Unwrapped and flexible terminal DNA segments are seen in yeast and vertebrate CENP-A^Nuc structures (17, 23), and this correlates with a 110-120 bp nuclease-resistant DNA core (17, 23). These in vitro studies are also in concordance with CENP-A^Nuc protecting 110-150 bp of α-satellite DNA based on in vivo ChIP-MNase-seq studies of natural human centromeres and neocentromeres, as well as observations that CENP-A^Nuc is a canonical octameric nucleosome in cells (21, 24, 25). A structure of CENP-A^Nuc reconstituted with AS1 (from the CCAN^ΔT-CENP-A^Nuc-AS1 cryo-EM data set) was positioned identically to CENP-A^Nuc-AS2, except the 3' unwrapped end is absent due to the 22 bp 5' register shift relative to AS2 (fig. S4A). In addition to resolving the CENP-A^Nuc, we were able to build the CENP-C central domain interacting with the C-terminus of CENP-A, and histones H2A/H2B, a structural feature of CENP-A^Nuc recognition conserved from yeast to human (6, 10, 11, 17).

The positively charged CCAN channel grips linker α-satellite DNA of CENP-A^Nuc

We then determined cryo-EM structures of CCAN^ΔT-CENP-A^Nuc complexes using both the AS1 and AS2 DNA (figs S5, S6). These two structures were identical except for their DNA boundaries (figs S4A, S6B). Because the AS1 sequence produced a more stable complex, as judged by SEC, a larger cryo-EM data set was collected with this DNA (Table S1).

Cryo-EM micrographs of the CCAN^ΔT-CENP-A^Nuc data sets and 2D-class averages showed particles corresponding to three species: (i) CENP-A^Nuc-CENP-C^N, (ii) CCAN^ΔT bound to free DNA and (iii) CCAN^ΔT bound to CENP-A^Nuc (fig. S5A-C). The observation of one CCAN^ΔT associated with CENP-A^Nuc is in agreement with an assessment of the oligomeric state of CCAN^ΔT-CENP-A^Nuc in solution, indicating a single CCAN^ΔT assembled onto CENP-A^Nuc (fig. S5D).

In the CCAN^ΔT-DNA complex, determined at 4.5 Å resolution, EM density is visible for a linear ~24 bp DNA duplex threading through the CENP-LN channel (Figs 2A, 3A and fig. S5E). The CCAN^ΔT-CENP-A^Nuc complexes exhibited conformational heterogeneity, limiting the resolution of a consensus cryo-EM reconstruction to 8.9 Å (Fig. 2A and figs S5, S6). Focused 3D classification and refinement of the CCAN^ΔT-DNA component alone produced a 7.3 Å resolution reconstruction with a protein and DNA structure identical to that of the CCAN^ΔT-DNA complex (fig. S5C, E). We therefore generated a composite CCAN^ΔT-CENP-A^Nuc structure from the higher resolution CENP-A^Nuc-CENP-C^N and CCAN^ΔT-DNA reconstructions (Fig. 2A). In the CCAN^ΔT-CENP-A^Nuc complex, the CENP-A^Nuc DNA is wrapped similarly to the free CENP-A^Nuc-CENP-C^N (fig. S6).

Figure 2. Structure of the CCAN^ΔT-CENP-A^Nuc complex.

(A) Right panel shows the consensus CCAN^ΔT-5 CENP-A^Nuc cryo-EM map (transparent white) overlaid onto the composite CCAN^ΔT-CENP-A^Nuc cryo-EM density map based on individual cryo-EM maps for the CENP-A^Nuc-CENP-C^N and CCAN^ΔT-DNA reconstructions. Left panel: composite map alone. (B) Two orthogonal views of the CCAN^ΔT-CENP-A^Nuc complex depicted in cartoon representation for protein and space filling for DNA. CENP-A^Nuc has a diameter of 11 nm.

Figure 3. The extranucleosomal linker DNA is gripped by the CENP-LN channel.

(A) Cartoon representation of CCAN^ΔT-CENP-A^Nuc with cryo-EM density shown for extranucleosomal linker DNA (from CCAN^ΔT-DNA cryo-EM density map). (B) The CENP-LN module features a positively-charged channel that complements the DNA duplex. Electrostatic surface charge shown for CENP-LN. (C and D). Details of the CENP-LN channel-DNA interaction. R169 of CENP-N inserts into the DNA minor groove. CENP-LN surface charge shown in C. (E) Cartoon schematic of the CCAN^ΔT modules showing DNA path. (F and G) CENP-C binds to two sites on CCAN. The DEFxIDE motif binds to CENP-LN (E), whereas the FxxLFL motif binds to CENP-HIKM (G).

A striking feature of the CCAN^ΔT-CENP-A^Nuc complex is that CCAN^ΔT forms few direct contacts with CENP-A^Nuc, but instead its primary contact with CENP-A^Nuc is through the extended extranucleosomal linker DNA. CENP-A^Nuc forms a small end-on contact, mediated through its DNA gyre, with the back-face of CCAN^ΔT (Fig. 3B, C). This positions the extranucleosomal DNA to thread through the CENP-LN channel. Clearly resolved cryo-EM density for the DNA phosphate backbone shows it tightly gripped by conserved basic residues of the CENP-LN channel (fig. S5E). These create a dense, continuous positively-charged surface complementing the shape and charge of the 24 bp linker DNA duplex (Figs 2, 3). Interactions of the side chains of CENP-N residues K148, M167 and R169 with the DNA minor groove contribute to defining a fixed DNA register (Fig. 3C, D). In the DNA-bound state, the CENP-LN channel contracts, tightening its grip on the DNA duplex (fig. S5F), and the channel is extended by the apposition of CENP-HIK^Head in the raised conformation (Fig. 3A, E).

CENP-C interaction sites on CENP-LN and CENP-HIKM modules

CENP-C is required for the centromere localization of CENP-LN, CENP-T, CENP-O and the CENP-HIKM module (5, 7, 12, 13, 18). The CCAN-interacting region of CENP-C maps to the PEST sequence (CENP-C^PEST), a 200-residue intrinsically disordered region N-terminal of the CENP-C central domain that features two highly conserved short linear sequence motifs (7, 13, 14). Cryo-EM maps of the CCAN^ΔT-DNA and CCAN^ΔT-CENP-A^Nuc complexes indicated two volumes of unassigned map density not present in apo-CCAN^ΔCT and CCAN^ΔC-DNA complexes (discussed below), associated with the CENP-HIKM and CENP-LN modules (fig. S6D), previously shown to interact with CENP-C (7, 12–14). To guide the interpretation of these densities, we used AlphaFold2 (26) to predict potential models of how CENP-C^PEST would interact individually with CENP-LN, and CENP-HIKM. The first run predicted the conserved DEFxIDE motif (residues 301-307) (7, 12–14) of CENP-C^PEST would form an edge β-strand with the CENP-N β-sheet (Fig. 3F). This position corresponded well to additional cryo-EM density associated with CENP-N (fig. S6D). Phe and Ile residues of the DEFxIDE motif dock into a hydrophobic pocket at the CENP-LN interface, whereas the flanking acidic residues form electrostatic interactions with conserved Arg and Lys residues of CENP-N, and CENP-L. In support of this model, a prior study mutating the EFxID residues of the CENP-C DEFxIDE motif abolished its interaction with CENP-LN in vitro and disrupted CENP-N recruitment to centromeres in vivo without perturbing CENP-C localization (7). For CENP-HIKM, AlphaFold2 also predicted that the CENP-C^PEST FxxLFL motif (residues 262-267) (13) would bind as an α-helix to a hydrophobic site at the combined CENP-H-CENP-K-CENP-M interface (Fig. 3G). This prediction was validated by cryo-EM density at this location (fig. S6D), and is consistent with a previous report that mutating the LFL residues of this motif disrupts CENP-C interactions with CENP-HIKM, in vitro and in vivo (13). Both the LN and HIKM interaction sites are located on the back-face of CCAN, facing the CENP-A^Nuc, suggesting a mechanism for how CENP-C tethers kinetochores to CENP-A^Nuc (5) (Fig. 3E).

α-Satellite linker DNA is a crucial determinant of stable CCAN association with CENP-A^Nuc

Our CCAN^ΔT-CENP-A^Nuc structure suggests that the interaction of the extranucleosomal DNA duplex with the CENP-LN channel is a major determinant of CCAN assembly onto a centromeric α-satellite-CENP-A^Nuc. To test this, we performed reconstitutions assessing the role of the extranucleosomal DNA, CENP-C^N, basic residues lining the CENP-LN channel, CENP-HIK^Head and CENP-QU^Foot. We first tested the requirement of the extranucleosomal DNA. Using SEC, we found that CENP-A^Nuc wrapped with 147 bp of α-satellite DNA (10) did not form stable complexes with CCAN^ΔT without CENP-C^N, whereas a 171 bp α-satellite that included the extranucleosomal DNA formed stable complexes with CCAN^ΔT independently of CENP-C^N (fig. S7A, B). Interactions with CENP-A^Nuc reconstituted with a 171 bp α-satellite were equally stable in the presence or absence of the B-box, consistent with sequence-independent interactions with the DNA phosphate backbone (fig. S7C).

Having established a requirement for linker DNA in the assembly of CCAN^ΔCT-CENP-A^Nuc complexes, we next assessed how modifications to the CENP-LN channel affected CCAN^ΔT interactions with CENP-A^Nuc. In the absence of CENP-C^N, mutating three positive patches of CENP-L (CENP-Lcm), that interact with linker DNA and the nucleosome gyre (Fig. 3B-D and fig. S5F), abolished binding of CCAN^ΔCT to CENP-A^Nuc-171 without perturbing CCAN assembly (fig. S7D). Identical results were obtained by deleting either CENP-HIK^Head, or CENP-QU^Foot (fig. S8A, B). These deletions also moderately reduced CCAN binding to isolated DNA (fig. S8C). To assess the validity of the CCAN-CENP-A^Nuc model in a cellular context, we tested the effect of the CENP-Lcm mutant that disrupted CCAN^ΔCT binding to CENP-A^Nuc-171 in vitro, on centromere localization of CENP-L in vivo. In HEK293 cells, localization of the CENP-Lcm mutant to kinetochores was substantially impaired compared to wild-type CENP-L (fig. S8D).

In previous structures of the CENP-N N-terminal domain (CENP-N^NT) associated with CENP-A^Nuc (7, 8, 10), CENP-N^NT interacts with the L1 loop of CENP-A (CENP-A^{L1 loop}) and adjacent DNA gyre at SHL2-3. In agreement with these structures, we found that mutating CENP-A^{L1 loop} disrupts the interaction of the isolated CENP-LN dimer with CENP-A^Nuc wrapped with 171bp α-satellite DNA (fig. S9A). By contrast, our structure of CCAN-CENP-A^Nuc-171 does not involve an interaction of CENP-N^NT with CENP-A^{L1 loop}, and consistent with this, mutating CENP-A^{L1 loop} did not disrupt interactions of CENP-A^Nuc-171 with CCAN^ΔCT, as assessed by both pulldown assays and during reconstitution on SEC (fig. S9A, B). These results, as well as previous structural information (7–10), suggest that the mode of interaction between CENP-LN and CENP-A^Nuc is different when CENP-LN is part of the full CCAN complex compared to CENP-LN alone. To further test this hypothesis, we performed competition assays where we added CENP-N^NT, which binds specifically to CENP-A^{L1 loop}, at a four-fold excess over the CCAN-CENP-A^Nuc complex (fig. S9C). The CCAN-CENP-A^Nuc complex readily accommodated the additional CENP-N^NT, without evident diminution of CCAN binding, and this additional CENP-N^NT incorporation was dependent on CENP-A^{L1 loop}. This suggests that in a fully assembled CCAN-CENP-A^Nuc complex, CENP-A^{L1 loop} is not occupied and therefore available for CENP-N^NT binding (fig. S9D), consistent with our cryo-EM reconstruction. Modelling a CENP-N^NT-based interaction of CCAN with CENP-A^{L1 loop} shows that CENP-A^Nuc would clash extensively with the CENP-L, CENP-HIK and CENP-OPQUR modules (as well as CENP-TWSX module as discussed below) (fig. S9E). In both models, CENP-N^NT forms the same interactions with the DNA backbone, although at SHL7-8 for CENP-N^NT in CCAN (fig. S9F). Thus, previously described mutants at the CENP-N^NT-DNA interface that disrupt CENP-N-centromere localization in vivo (4, 7, 8), are consistent with, but do not distinguish between, both modes of CENP-N-CENP-A^Nuc interaction.

Overall, the interaction of free CENP-LN with CENP-A^Nuc is dependent on CENP-A^{L1 loop}, but independent of the linker DNA, whereas CCAN interactions with CENP-A^Nuc are dependent on the linker DNA but independent of CENP-A^{L1 loop}. These results are consistent with different modes of interaction between CENP-LN and CENP-A^Nuc when CENP-LN is incorporated into CCAN. This agrees with a previous observation that the L1 loop is not required for the in vitro assembly of a functional kinetochore (27).

CENP-C confers selectivity of assembled CCAN for CENP-A mono-nucleosomes

The only direct interaction between CCAN^ΔT and the CENP-A histone in our structure is mediated by CENP-CN. To validate this finding, we produced human H3 nucleosomes reconstituted either with a 147 bp 601 sequence or a 601 sequence extended at the 5' end by 30 bp (177 bp) (fig. S4A). As expected, CCAN^ΔT failed to bind H3^Nuc reconstituted with the minimal 147 bp 601 sequence, whereas CCAN^ΔT bound CENP-A^Nuc with the same DNA only in the presence of CENP-C^N, similar to 147 bp α-satellite DNA (figs S7A and S10A). Consistent with our structure, binding was observed with the extended 177 bp sequence for both H3^Nuc-177 and CENP-A^Nuc-177 (fig. S10B). The binding of CCAN^ΔCT to CENP-A^Nuc-177 but not to H3^Nuc-177, was enhanced four to five-fold by CENP-C^N, consistent with selective recognition of CENP-A^Nuc by CENP-C (fig. S10C).

CENP-TWSX binds and partially wraps linker DNA

The CCAN^ΔT-CENP-A^Nuc structure was reconstituted without the CENP-TWSX module due to our initial difficulties incorporating CENP-TWSX into CCAN. Removing all affinity tags from CENP-WSX enabled CENP-TWSX incorporation into CCAN (fig. S10D), and as judged by SEC, this 16-subunit holocomplex interacts with the 171 bp α-satellite-CENP-A^Nuc similarly to CCAN^ΔT (fig. S10D), consistent with CENP-A, CENP-C and CENP-T comprising a single 20-subunit inner kinetochore complex (28). The full CCAN complex also requires extranucleosomal linker DNA to bind CENP-A^Nuc without CENP-C, suggesting that CENP-TWSX does not change the fundamental mechanism of CCAN assembly onto CENP-A^Nuc (fig. S10E). EMSA experiments revealed that CENP-TWSX increases the affinity of CCAN^ΔCT for both CENP-A^Nuc and H3^Nuc nucleosomes containing 177 bp 601 DNA approximately 3-fold, suggesting that it does not confer additional selectivity for CENP-A^Nuc over canonical H3^Nuc (fig. S10 C, F).

To understand how CENP-TWSX contributes to CCAN recognition of DNA and CENP-A^Nuc, we determined a cryo-EM structure of CCAN^ΔC with a 53 bp DNA linker segment (fig. S4A). We obtained a well-resolved cryo-EM map at 2.8 Å resolution with all regions of CCAN and the DNA clearly defined (Fig. 4A, B, fig. S11, Table S1 and Movie S2).

Figure 4. Architecture of the full CCAN^ΔC-DNA complex.

(A) Cryo-EM density map, and (B) molecular model of the CCAN^ΔC-DNA complex with DNA map density shown in yellow. (C) CCAN assembles an enclosed chamber that topologically entraps linker DNA. (D) Electrostatic representation of the DNA-binding chamber formed by CENP-LN, CENP-TW and CENP-HIK^Head modules shows a positively charged chamber that tightly grips DNA. (E) Schematic of CCAN showing DNA path through the CENP-LN, CENP-HIKM and CENP-TWSX modules. (F) The CENP-TWSX module resembles the canonical H3/H4 nucleosome tetramer, and partially wraps DNA.

The four histone fold domains of CENP-TWSX resemble the histone H3-H4 tetramer, and are similar to crystal structures of isolated chicken CENP-TWSX (19). CENP-TWSX forms multiple interactions with neighboring CCAN subunits and DNA (Fig. 4B, C). Contacts with a slightly repositioned CENP-HIK^Head are mediated by both CENP-T and the extended N terminus of CENP-W. CENP-W also directly contacts the CENP-N pyrin domain. A CENP-TWSX-mediated linkage of CENP-HIK^Head to CENP-N and CENP-QU^Foot creates an enclosed chamber that topologically entraps the extranucleosomal DNA (Fig. 4A-D). The chamber comprises CENP-LN, CENP-HIK and CENP-TW modules and is augmented and rigidified by all other CCAN subunits (Fig. 4E). As in the CCAN^ΔT-DNA complex, the DNA duplex inserts into the deep CENP-LN channel, however CENP-TWSX and the repositioned CENP-HIK^Head, induce a marked curvature of the DNA. Strikingly, the DNA starts to wrap around the CENP-TW histone fold domains as it emerges from the CENP-LN channel in a manner comparable to how a canonical H3-H4 tetramer wraps DNA (Fig. 4F). This DNA binding is augmented by CENP-HIK^Head. Consequently, the DNA duplex engages a 50 A-long positively-charged groove at the interface of CENP-IHead and CENP-TW (Fig. 4D). Similar to nucleosomes, the N-terminus of the CENP-T (HFD) inserts into the minor groove of the extranucleosomal DNA positioning the highly conserved Arg450 to contact the DNA backbone (Fig. 4F). The CENP-LN channel and CENP-TW-CENP-IHead groove perfectly complement the DNA duplex, with the DNA phosphate backbone forming a continuous interface with CCAN, burying some 976 A2 of DNA surface area.

To complete our structural analysis, we determined cryo-EM structures of CCAN in complex with CENP-A^Nuc reconstituted with 171 bp α-satellite DNA (Fig. 5A, B and fig. S11C-E). CCAN assembles onto CENP-A^Nuc similarly to CCAN^ΔT. The nucleosome associates with the back-face of CCAN, and the extranucleosomal DNA threads through the CENP-LN channel. The linker DNA of the α-satellite repeat is gently curved and interacts with CENP-TW-CENP-IHead reminiscent of the CCAN^ΔC-DNA complex. We note that the AS2 α-satellite DNA of 171 bp in length is approximately 15 bp too short to wrap all the way around the CENP-TWSX module.

Figure 5. Structure of the complete CCAN-CENP-A inner kinetochore module.

(A) Atomic model of the CCAN-CENP-A^Nuc complex. (B) Cartoon schematic of the CCAN-CENP-A^Nuc complex. (C) Mutations of CENP-N and CENP-TW previously shown to impair centromere function contribute to the DNA-binding channel of the CCAN (4, 7–9). Residues implicated in DNA binding by CENP-L, and mutated in the current study, are also shown.

CENP-TW contributes to DNA binding via conserved basic residues that were previously implicated in mediating DNA binding in vitro, and for CENP-W, are required for mitotic progression, and its recruitment to kinetochores in vivo. These functions of CENP-TW, and CENP-TWSX's ability to supercoil DNA, led to proposals that CENP-TWSX is a DNA-binding nucleosome-like particle, consistent with the curvature of the DNA in our structure (19, 29). These residues of CENP-TW, together with basic residues of CENP-N, whose mutation disrupt CENP-N association with CENP-A^Nuc in vitro, and centromere localization in vivo (4, 7–9), form a continuous spine lining the conserved CCAN-DNA interface (Fig. 5C).

Many vertebrate cell lines are viable and exhibit normal chromosome segregation without the CENP-OPQUR module (12, 30), contrasting with our finding that deletion of the CENP-QU^ΔFoot abrogates CCAN^ΔCT binding to CENP-A^Nuc (fig. S8B, C). A likely explanation for this is the absence of the CENP-TWSX module in CCAN^ΔCT, that contributes to DNA binding (Figs 4 and 5A, B). Consistent with this, we found that in the presence of CENP-TWSX, removing the entire CENP-OPQUR module (CCAN^ΔCO), caused only a small decrease in the affinity for both DNA and CENP-A^Nuc (fig. S12A, B). Additionally, CCAN^ΔCO readily formed a complex with CENP-A^Nuc (fig. S12C). These in vitro results support the idea that CENP-OPQUR contributes relatively little to the attachment of vertebrate kinetochores to chromatin.

Discussion

The absence of both mitotic phosphorylation and the outer kinetochore suggests the structure we describe likely reflects an interphase kinetochore. A cryo-EM structure of human apo-CCAN (31) agrees with our CCAN reconstructions, and the proposed model of DNA engagement by the CENP-LN channel is shown from our CCAN-CENP-A^Nuc complex to represent extranucleosomal linker DNA. Defining the structure of the holo-kinetochore at mitosis remains an outstanding challenge, and it will be important to validate the cryo-EM structures in cells.

The remarkably similar features of DNA duplex recognition by the basic CENP-LN channel of HsCCAN and ScCCAN (17), suggests a common evolutionary origin of centromere-kinetochore assembly. In humans, however, the DNA duplex engaged by CCAN is the α-satellite linker DNA rather than the unwrapped CENP-A^Nuc DNA of budding yeast. CCAN engages ~40 bp of extranucleosomal DNA which could be facilitated either by unwrapping of ~16 bp of the upstream CENP-A^Nuc, consistent with our structures showing DNA unwrapping in CENP-A^Nuc, and/or by the sparse occupancy of CENP-A^Nuc on centromeric DNA (32). Such organization is also consistent with evidence that CENP-C and CENP-T bridge adjacent nucleosomes (28). The topological entrapment of linker DNA provides a molecular explanation for how the inner kinetochore can withstand strong pushing and pulling forces applied by the mitotic spindle, a central function of this large macromolecular assembly. Our work also confirms CENP-C as a key determinant of CCAN specificity for CENP-A^Nuc, providing a molecular explanation for how this essential CCAN component directly links the inner kinetochore to centromeric chromatin.

Our finding that CCAN can bind to either canonical or CENP-A nucleosomes bearing adequate linker DNA could provide an explanation for how the inner kinetochore maintains attachments at centromeres where H3^Nuc vastly exceeds CENP-A^Nuc across the entire centromeric α-satellite repeats (33). Furthermore, it could explain why synthetic depletion of CENP-A once CCAN is assembled onto centromeres, nevertheless permits mitotic centromere function in human cells (34). Salt extraction of CENP-A does reduce CENP-N and CENP-C levels in centromeric chromatin (35), suggesting that once CCAN is assembled at centromeres in a CENP-A dependent process (36, 37), CENP-A is dispensable for localization. Lastly, assembly of CCAN onto linker DNA provides a possible explanation for how some organisms that lack CENP-A yet still depend on CCAN for chromosome segregation might function (38).

An emerging question from this study is how human CCAN localizes specifically to the centromere given its modest preference for CENP-A^Nuc mono-nucleosomes compared to H3^Nuc (33). One explanation is that the CENP-LN module is selectively recruited to the centromere, through recognition of CENP-A^{L1 loop}, during early stages of the cell cycle, and is then displaced to form a fully assembled CCAN, thus confining kinetochore assembly to a specific chromatin locus. CENP-LN levels at the centromere decline upon mitotic entry (10, 39), supporting this sequential and specific localization mechanism. CDK-mediated CENP-C phosphorylation (11, 40) and/or chromatin compaction (39) might regulate this process. CENP-B recognizes the B-box motifs of the α-satellite repeat and interacts with CENP-C, providing another possible mechanism for specific CCAN recruitment (41). Different degrees of wrapping of CENP-A^Nuc in centromeric arrays compared to H3^Nuc might also provide selectivity for CCAN assembly, as could the H1 histone that would block access of CCAN to the linker DNA at H3^Nuc. Our structure permits a model for the arrangement of CCAN within the centromeric arrays of regional kinetochores (fig. S12D).

This study also highlights the common mechanisms of topological entrapment used by biological systems responsible for accurately mediating chromosome segregation, such as the cohesin complex that, likewise resisting vast forces exerted by the spindle, maintains cohesion between sister chromatids.

Supplementary Material

Movie 1

3D variability analysis demonstrates a continuous conformational heterogeneity of the DNA termini of the CENP-A^Nuc.

Movie 2

Overview of the cryo-EM reconstruction and molecular model of the CCAN^ΔC-DNA complex.

One-Sentence Summary.

The human inner kinetochore CCAN complex tightly grips the linker DNA of the α-satellite CENP-A nucleosome.

Data and materials availability

PDB and cryo-EM maps have been deposited with RCSB and EMDB, respectively. Accession numbers are listed in Tables S1 and S2.