We investigated the requirements for targeting the centromeric histone H3 homologue CENP-A for assembly at centromeres in human cells by transfection of epitope-tagged CENP-A derivatives into HeLa cells. Centromeric targeting is driven solely by the conserved histone fold domain of CENP-A. Using the crystal structure of histone H3 as a guide, a series of CENPA/histone H3 chimeras was constructed to test the role of discrete structural elements of the histone fold domain. Three elements were identified that are necessary for efficient targeting to centromeres. Two correspond to contact sites between histone H3 and nucleosomal DNA. The third maps to a homotypic H3–H3 interaction site important for assembly of the (H3/H4)2 heterotetramer. Immunoprecipitation confirms that CENP-A self-associates in vivo. In addition, targeting requires that CENP-A expression is uncoupled from histone H3 synthesis during S phase. CENP-A mRNA accumulates later in the cell cycle than histone H3, peaking in G2. Isolation of the gene for human CENP-A revealed a regulatory motif in the promoter region that directs the late S/G2 expression of other cell cycle–dependent transcripts such as cdc2, cdc25C, and cyclin A. Our data suggest a mechanism for molecular recognition of centromeric DNA at the nucleosomal level mediated by a cooperative series of differentiated CENP-A–DNA contact sites arrayed across the surface of a CENP-A nucleosome and a distinctive assembly pathway occurring late in the cell cycle.
The accurate transmission of replicated eukaryotic chromosomes is mediated by centromeres. Structurally distinct loci present once per chromosome, centromeres provide the essential functions of chromosome segregation. These include specifying the assembly of the kinetochore, a microtubule-dependent motor complex at the surface of the chromosome, and the maintenance of sister chromatid cohesion until their separation at the onset of anaphase (Bloom, 1993; Miyazaki and Orr-Weaver, 1994; Pluta et al., 1995). In addition to these primarily mechanical functions, centromeres act as important regulators of mitosis and meiosis through a mechanism that monitors attachment of chromosomes to the spindle and reports to a spindle assembly checkpoint that regulates progression into anaphase (McIntosh, 1991; Li and Nicklas, 1995; Nicklas et al., 1995; Rieder et al., 1995). Understanding how these functions are specified at a molecular level begins with identification of the molecular recognition events that initiate centromere assembly on the chromosome.
By elegant molecular genetic approaches, it has been possible to identify discrete cis-acting DNA sequences from Saccharomyces cerevisiae (Clarke and Carbon, 1980; Hieter et al., 1985) and Saccharomyces pombe (Hahnenberger et al., 1989; Niwa et al., 1989) that are sufficient to establish centromere function on artificial chromosomes. Dissection of these sequences has revealed that centromere function is established at both the primary structural level of DNA sequence as well as at higher levels of DNA structure within chromatin. DNA sequence recognition is driven by sequence-specific DNA–protein interactions, exemplified by the essential CDE III element of the S. cerevisiae centromere; single point mutations in this 25-bp DNA sequence can completely abolish centromere activity (McGrew et al., 1986; Hegemann et al., 1988). CDE III plays a primary role in kinetochore assembly on the S. cerevisiae centromere by binding to a 240-kD multiprotein complex, CBF3, that mediates the association of a microtubule- dependent motor activity with the chromosome (Lechner and Carbon, 1991; Hyman et al., 1992; Middleton and Carbon, 1994). Cbf3b, a 60-kD subunit of CBF3, is an essential zinc finger protein that is thought to provide the DNA binding function of CBF3 (Lechner, 1994). Other examples of centromere proteins that directly recognize DNA sequence are the yeast helix-loop-helix protein CBF1 (Cai and Davis, 1990) and the mammalian protein CENP-B, which recognizes a discrete sequence element found in centromeric satellite DNA (Earnshaw et al., 1987; Masumoto et al., 1989; Sullivan and Glass, 1991). Thus, molecular recognition of centromeric loci occurs, at least in part, through direct DNA sequence recognition by proteins, interactions similar to the familiar DNA binding activities observed for transcription factors (Mitchell and Tjian, 1989; Harrison, 1991).
Centromere function is also established through essential interactions that take place at the level of DNA structure within chromatin. From the earliest cytological observations of centromeres as the primary constriction of mitotic chromosomes, it has been understood that centromeres are packaged distinctly as constitutive heterochromatin. In the point centromeres of budding yeast, 150–200 bp of cen DNA sequences are packaged in a core particle flanked on both sides by arrays of highly phased nucleosomes (Bloom and Carbon, 1982), and this specialized chromatin structure is necessary for centromere function (Saunders et al., 1988; Bloom et al., 1989). Sequence element CDE II, which comprises a 78–86-bp AT-rich segment conserved in composition but not in sequence among yeast centromeres, appears to adopt a uniquely folded conformation that plays an important role in providing complete centromere function (Sorger et al., 1994; Tal et al., 1994; Sears et al., 1995). The more complex centromeres of fission yeast exhibit a different type of chromatin structure, with several kilobases of DNA in the central core domain packaged in a highly irregular nucleosomal array that is assembled only in conjunction with functional centromere sequences in S. pombe (Polizzi and Clarke, 1991). The dependence of this structure on sequences distal to the central core DNA suggests that large scale folding of the centromere locus is required for the segregation function (Polizzi and Clarke, 1991; Marschall and Clarke, 1995).
Understanding what constitutes a functional centromere sequence in animal cells has been confounded by their large size, ranging from 500–5,000 kb in human chromosomes (Tyler-Smith and Willard, 1993). Nevertheless, it has been possible to map a Drosophila centromere to a 420-kb segment, revealing that both simple satellite sequences as well as islands of complex sequence are required for complete centromere function (Le et al., 1995; Murphy and Karpen, 1995). In mammalian cells, centromere function is also associated with large blocks of heterochromatin comprised of highly repetitive satellite DNA typified by the α satellite of primate chromosomes: extensive tandemly repeated arrays of a 171-bp monomer sequence (Willard, 1991). One of the abiding mysteries of animal centromeres, however, is the lack of sequence conservation of centromeric satellite DNA (Beridze, 1982). With the exception of a small sequence that functions as the binding site for CENP-B, the CENP-B box (Masumoto et al., 1989), no homology is seen in satellite DNA across different classes of vertebrates, and, indeed, satellite DNA is one of the most rapidly evolving compartments of the genome in vertebrates. An important role for CENP-B and the CENP-B box in centromere function is in doubt, however, since its presence is not correlated with centromere function (Earnshaw et al., 1989). Two hypotheses have been suggested to explain this lack of conservation in centromere sequences: either animal cell centromere DNA contains small, as yet unidentified sequence elements similar to yeast centromeres that possess kinetochore-nucleating capabilities, or centromere function is not specified directly by DNA sequence, but rather by higher order DNA or chromatin structure.
One protein situated to play a role in specifying the properties of centromeric chromatin is CENP-A, a centromere-specific homologue of the core nucleosomal protein histone H3 (Palmer et al., 1991; Sullivan et al., 1994). CENP-A was originally identified as a centromere-specific autoantigen that copurified with nucleosomal core particles (Earnshaw and Rothfield, 1985; Palmer and Margolis, 1985; Palmer et al., 1987). A potentially homologous protein has recently been identified in yeast as the product of a gene, CSE4, that is essential for mitotic chromosome segregation (Stoler et al., 1995). Together, CENP-A, CSE4, and histone H3 form a roughly equidistant triangle of homologous proteins linked at the level of ∼60% sequence identity, limited to a COOH-terminal domain of ∼90 amino acids (Sullivan et al., 1994; Stoler et al., 1995). This region corresponds to the domain in histone H3 that is sufficient for nucleosome assembly in vitro (for review see van Holde, 1989) and in vivo (Mann and Grunstein, 1992), and that is part of the highly ordered core of the histone octamer (Arents et al., 1991). Surprisingly, this conserved histone fold domain of CENP-A is required for targeting to human centromeres, rather than the unique sequences of the NH2 terminus (Sullivan et al., 1994).
In this work we have dissected the molecular features of CENP-A that are required for its assembly at human centromeres. By systematic replacement of structural elements of the CENP-A histone fold domain with the corresponding sequences of histone H3, we have identified three regions of the molecule that are required for targeting CENP-A to centromeres. These correspond to two nucleosomal DNA contact sites of histone H3 and a region that mediates self-association between the two copies of histone H3 within the nucleosome. In addition to these structural features, we show that CENP-A expression is uncoupled from normal histone H3 expression, occurring later in the cell cycle, and that this synthetic timing is important for appropriate targeting of CENP-A. Taken together, these data suggest a mechanism for specific molecular recognition of centromeric DNA at the level of the nucleosome.
Materials And Methods
Cell Culture and Transfection
HeLa (ATCC CCL3) and tTA-HeLa cells (Gossen and Bujard, 1992) were maintained in DME with 10% FCS (GIBCO BRL, Gaithersberg, MD) at 37°C in a 5% CO2 atmosphere. tTA-HeLa cultures were supplemented with 400 μg/ml G418. Stably transformed tTA cell lines (see below) were cultured in the presence of 400 μg/ml G418, 330 ng/ml puromycin, and 1 μg/ml tetracycline. For immunofluorescence experiments, cells were plated on glass coverslips at a density of 2–2.5 × 104 cells per cm2 the night before transfection. Transfection was performed in serum-free medium using Lipofectamine (GIBCO BRL) as previously described (Sullivan et al., 1994).
To establish a stable, inducible cell line expressing CENP-A–HA1, the CENP-A insert from pcDL-CAepi was subcloned into plasmid pUHD10.3 (Gossen and Bujard, 1992), forming plasmid pUHD10.3CAepi. tTA HeLa cells expressing the tetracycline transactivator were cotransfected with pUHD10.3-CAepi and pBS-PAC, a puromycin resistance marker (de la Luna et al., 1988) at a 10:1 ratio using Lipofectamine. Transformants were selected using 330 ng/ml puromycin in the presence of 1 μg/ml tetracycline, and individual clones were assayed by induction in tetracycline-free medium followed by Western blot analysis.
Construction of Segmental Mutants.
General methods were essentially as described by Ausubel et al. (1995) unless specified. Mutations were constructed in plasmid pcDL-CAepi, which is identical to pcDL-CAHA1 (Sullivan et al., 1994) except that three copies of the hemagglutinin (HA)1 1 epitope are present at the COOH terminus of the coding region. Histone H3 sequences were obtained from plasmid pMH3.2-614 (Taylor et al., 1986). Segments of CENP-A were replaced with the corresponding H3 sequence using a bimolecular recombinant PCR strategy. A pair of standard 5′ and 3′ oligonucleotide primers (GIBCO BRL) flanking the CENP-A coding region of pcDL-CAepi was prepared and used in all experiments. For each mutant, two divergent overlapping primers were constructed, each containing at least 12–15 bp of CENP-A sequence at their 3′ ends and a segment encoding the desired mutations at their 5′ ends. Each mutagenic oligonucleotide pair was designed with a 15–17-bp overlap. PCR reactions were performed using each mutagenic primer in conjunction with the appropriate flanking primers, synthesizing two DNA fragments that overlapped by 15–17 bp within the mutated region. PCR reactions (95°C × 90 s; 20 × [95°C × 30 s, 55°C × 60 s, 72°C x 90 s]; 72°C × 10 min) were performed in 50 μl using 5 μg/ml pcDL-CAepi, 1.5 mM Mg, 1 μM each primer, 100 mM dNTPs (Pharmacia Fine Chemicals, Piscataway, NJ), 10% DMSO, and 1.25 U of an 8:1 unit ratio mixture of Taq DNA polymerase (Promega, Madison, WI) and Pfu DNA polymerase (Stratagene, La Jolla, CA) made fresh for each experiment. PCR products were purified via QX Matrix (Qiagen, Chatsworth, CA), combined, and used as template for a second round of PCR using only the standard primers, performed as above except with 10 amplification cycles. Full-length recombinant PCR products were cloned into plasmid pCRII (Invitrogen, San Diego, CA). The sequence of the entire coding region was verified (Sequenase 2.0; United States Biochemical Corp., Cleveland, OH), and inserts from correct clones were isolated as NarI–SacI fragments and cloned into NarI- and SacI-digested pcDL-CAepi. Plasmids for transfections were prepared with Qiagen DNA purification columns.
Codon 86 was randomized by the same bimolecular recombinant PCR strategy described above, using primers possessing the sequence NN(C/T) on the coding strand. Approximately 50 pCRII transformants were picked and colony sequenced with the CircumVent thermal cycle sequencing kit (New England Biolabs Inc., Beverly, MA) using a 32P–end labeled primer that spanned nucleotides 275–293 in the CENP-A cDNA to determine the sequence of codon 86. 13 different mutants were recovered (Y, F, I, L, V, C, N, D, H, R, S , P, and G) and cloned into pcDL-CAepi as above. Constructs that failed to localize were sequenced completely to verify that loss of function was due to mutation at codon 86 (Sequenase 2.0; United States Biochemical Corp.).
Transfer of CENP-A Helix II to Histone H3.
A trimolecular recombinant PCR strategy was used to replace helix II codons 85–112 of histone H3 with the corresponding codons 87–114 of CENP-A. Three fragments were generated in the first round of PCR. Fragment 1, the 5′ fragment, containing codons 1–84 of histone H3, was amplified from plasmid pcDLH3HA1 using the standard 5′ primer as above and a 30-mer primer at the 3′ end corresponding to codons 80–84 of histone H3 plus codons 87–91 of CENP-A (the insertion of two codons in CENP-A relative to histone H3 accounts for the difference in residue numbering). Fragment 2, the central fragment corresponding to helix II of CENP-A, was amplified from pcDLCAHA1 using a 5′ primer complementary to the 3′ primer of fragment 1 and a primer at the 3′ end corresponding to the last five codons of CENP-A helix II and codons 113–117 of histone H3. Fragment 3, the 3′ fragment encoding the COOH-terminal portion of histone H3 and the HA-1 epitope, was amplified from pcDL-H3HA1 using a 5′ primer corresponding to the last five codons of CENP-A helix II and codons 113–117 of histone H3 and the standard 3′ primer an oligo in the 3′ untranslated region of CENP-A. The three fragments were purified, and then combined in equimolar amounts to provide the template for a second round of PCR using the standard 5′ and 3′ primers; the product was isolated and subcloned for expression as described above. The complete coding region sequence of the resulting plasmid was verified by sequencing.
Plasmid pMH3-CAHA1 was constructed by replacing the histone H3 coding region of pMH3.2-614 with an epitope-tagged CENP-A fragment. PCR primers for CENP-A were constructed incorporating an NcoI site at the ATG initiator codon at the 5′ end and a single copy of the HA-1 epitope followed by an AflIII site at the 3′ end. The amplified product was cloned into NcoI–AflIII-digested pMH3.2-614 and verified by DNA sequencing.
For analysis of protein localization in transfected cells, immunofluorescence microscopy was performed 18–72 h after transfection, as described previously (Sullivan et al., 1994). Endogenous centromere antigens were visualized with a human anticentromere antiserum, hACA-M detected with a rhodamine–coupled secondary antibody, while HA epitope–tagged proteins were visualized with mAb 12CA5 (a kind gift from Dr. Ian Wilson, The Scripps Research Institute, La Jolla, CA) and fluorescein-coupled secondary antibody.
Immunoblots were performed as described previously (Sullivan et al., 1994) using human anticentromere serum hACA-M at a dilution of 1:2,000 and mAb 12CA5 at a concentration of 5 μg/ml. Blots were developed using HRP-coupled secondary antibodies (Amersham Corp., Arlington Heights, IL) and a chemiluminescence detection reagent (Pierce Chemical Co., Rockford, IL).
For immunoprecipitation analysis, protein expression in a stable pUHD10.3-CAepi transformant was induced for 3 d. Nuclei from 3–5 × 107 cells were isolated according to Masumoto et al. (1989), washed in buffer A ( 5 mM Hepes, pH 7.5, 10 μM leupeptin, 1.5 μM aprotinin, 1 mM DTT), and centrifuged at 3,000 g. The nuclear pellet was resuspended in 500 μl digestion buffer at a concentration of 0.5–1 × 108/ml (buffer A containing 200 U/ml micrococcal nuclease, 1 mM CaCl2) and incubated at 37°C for 5 min. Digestion was stopped by addition of EDTA to a final concentration of 10 mM. After centrifugation at 8,000 g, the supernatant was collected, and the pellet was resuspended in buffer A and subjected to two additional rounds of extraction by sonication for 10 s followed by centrifugation and collection of the supernatants. Supernatants were pooled in a siliconized Eppendorf tube, supplemented with 0.1% NP-40 and 25 μg of mAb 12CA5, and mixed end over end for 2 h at 4°C. A 100-μl aliquot of protein A–Sepharose (Pharmacia Fine Chemicals) previously equilibrated with buffer A was added and incubated for an additional 2 h at 4°C. The beads were collected by centrifugation and the supernatant was saved. Immunoprecipitates were washed five times with buffer A, and then resuspended in SDS-PAGE sample buffer. Equivalent amounts of all soluble fractions and one half of the immunoprecipitated proteins were analyzed by Western blotting.
Isolation of a Human CENP-A Genomic Sequence
A human Caucasian male placental genomic DNA library prepared in Lambda Fix II (Stratagene; a kind gift from Edward Chan, The Scripps Research Institute) was screened by PCR (Israel, 1993) using CENP-A primers that span a small intron. Two phage with overlapping inserts spanning 20 kb of genomic DNA were isolated and characterized by restriction mapping using a series of probes derived from the CENP-A cDNA (to be described in detail elsewhere). A 2,878-bp EcoRI fragment containing a 5′ flanking genomic sequence was isolated and sequenced by a combination of manual and automated methods (GenBank accession number U82609). The 2.9-kb fragment was found to contain 1,101 bp upstream of the start of our CENP-A cDNA clone, the first 250 bp of the cDNA and 1,527 bp of the first intron in CENP-A.
Cell Cycle Analysis
HeLa cells were grown in 10-cm dishes to ∼60% confluence. The first block was initiated by replacing medium with complete DME containing 2 mM thymidine. After 15 h, cells were released by washing twice with dPBS and adding normal complete DME, and were allowed to grow for 9 h. Cells were blocked a second time for 15 h as above. After release as above, samples were collected at 2 h intervals for 16 h by trypsinization and washed twice with PBS, and pellets were kept at −70°C until preparation of RNA. For time points exhibiting an increased mitotic index (8–12 h after release), cells were also recovered from the media and the washes before trypsinization.
RNA was isolated by acidic guanidinium thiocyanate/phenol-chloroform extraction (Xie and Rothblum, 1991). CENP-A mRNA was assayed by RNase protection using a probe constructed by cloning a 155-bp EcoR1– ApaI fragment containing the 5′ end of the CENP-A cDNA into pBSSK(+) (Stratagene). Plasmid was linearized with XbaI for transcription by T7 RNA polymerase (Maxiscript kit; Ambion, Austin, TX) and α-[32P]UTP (Amersham Corp.) according to the manufacturer's instructions. The probe length was 203 bp with a protected fragment length of 153 bp. RNase protection asays (HybSpeed RPA kit; Ambion) were performed using 10 μg of total HeLa RNA isolated from synchronized cells. Hybridization of probe (350K cpm/rxn) and RNA was carried out for 1 h at 68°C in siliconized tubes followed by digestion with an RNase A/T1 mixture used at a dilution of 1:100 from the supplied concentration. End-labeled size markers were prepared from an HaeIII digest of pBS-SK(+). Reactions were electrophoresed on 6% sequencing gels and exposed for 2 h on a phosphor imaging screen from Molecular Dynamics (Sunnyvale, CA). ImageQuant software (Molecular Dynamics) was used to quantitate signal intensities. Histone H3 mRNA abundance was determined in the same samples by Northern blot analysis, using the coding region of plasmid pMH3.2-614 as a probe, and similarly quantitated.
Structural Determinants of Centromeric Targeting
The histone fold domain consists of a set of three α helices (H I, H II, H III) separated by two turn/β sheet structures (strand A, strand B); histone H3 and, by homology, CENP-A contain an additional α helix at the NH2 terminus of the fold domain (N helix; Fig. 1,A) (Arents et al., 1991). To evaluate CENP-A targeting within the context of this structure, we prepared a set of substitution derivatives by replacing CENP-A sequences within the fold with the homologous histone H3 sequences (Fig. 1,A). Mutations were constructed using an epitope-tagged version of CENP-A carrying three copies of the influenza hemagglutinin HA-1 epitope (Wilson et al., 1984) at the COOH terminus, allowing us to monitor the expression (Fig. 1,B) and localization of CENP-A derivatives in transfected cells (Fig. 2; WT).
We first asked whether the histone fold domain is sufficient to direct centromeric targeting. In previous experiments, the NH2-terminal tail of CENP-A was replaced with that of histone H3, which, although lacking sequence homology, shares its highly basic character with CENP-A (Sullivan et al., 1994). To determine if a basic NH2-terminal tail is dispensable for targeting, codons 4–31 of CENP-A were excised. The resulting protein showed no impairment of targeting to centromeres, demonstrating that a basic NH2-terminal tail is dispensable for this function (Fig. 2; NΔ). Thus, the COOH-terminal portion of CENP-A, corresponding to the histone fold homology domain, is both necessary and sufficient for assembly of CENP-A at centromeres.
Within the histone fold domain, we initially examined four regions corresponding to secondary structure segments of the domain based on the data of Arents et al. (1991) and Richmond et al. (1993): helices I and II, strand A, and strand B (Fig. 1,A). We also tested the COOH terminus, which is longer in CENP-A and divergent from histone H3. Helix III was not tested since only a single conservative (Ile-Val) substitution is found in this segment of CENP-A. The two most conserved regions, helix I (Fig. 2; hI) and strand B (Fig. 2; sB), could be substituted without affecting targeting, as could the COOH terminus (Fig. 2; C). The strand A substitution, residues 75–86, was profoundly deficient in targeting ability (Fig. 2; sA). It retained a small degree of targeting specificity that was observed as a slight increase in centromere staining over an essentially uniform nuclear incorporation in a minority of cells (see also Fig. 3,D below). Substitution of helix II of the histone fold domain resulted in a complete loss of targeting to centromeres (Fig. 2; hII). These data demonstrate that sequences in the central portion of the histone fold domain are primarily responsible for targeting CENP-A to centromeres.
The two segments identified in this experiment comprise a large contiguous region at the center of the domain and contain most of the divergent amino acids that distinguish CENP-A from histone H3. To further refine identification of targeting sequences, strand A and helix II were each divided into NH2-terminal, central, and COOH-terminal portions, each containing three to five CENP-A specific residues that were substituted with histone H3 sequences as above (Fig. 1,A). This analysis showed the NH2 and COOH termini of the long central helix II were necessary for targeting CENP-A, but replacement of residues in the central portion of the helix had no effect (Fig. 3,B). In contrast, none of the strand A subregion mutations, including the deletion of two amino acids in the center of this region, showed any significant impairment of targeting (Fig. 3 A). Thus, the CENP-A–specific sequences at the two ends of helix II, residues 88–92 and 109–114, are necessary for assembly at centromeres, while strand A, residues 75–86, can accommodate significant change in amino acid sequence and length without disruption of targeting activity.
One residue in this region, Trp86, was selected for specific mutagenesis. This residue is notable because Trp is absent in the core histones, but is present at this same position in CSE4 (Stoler et al., 1995). This codon was randomized with a PCR procedure, and 13 mutants encoding different amino acids were recovered and tested for targeting (data not shown). Replacement of Trp86 with the aromatic residues Tyr or Phe (the amino acid normally found in this position in histone H3) had no effect on targeting. Aliphatic residues showed intermediate levels of targeting roughly correlated with hydrophobicity, while hydrophilic and charged residues failed to target at all. This experiment rules out a specific role for this Trp residue in centromeric targeting, but demonstrates a requirement for an aromatic amino acid.
Histone H3 contains an additional α helix at the amino terminus of the histone fold domain, the N-helix. Secondary structure prediction reveals a putative α helix in this segment of CENP-A, spanning residues 43–55. When this region was replaced along with the entire NH2-terminal tail of histone H3, the resulting protein, H3-CA, retained targeting activity but was less efficient at localizing to centromeres (Sullivan et al., 1994). Two additional replacement mutants were constructed, HN1 and HN2, to ask whether a secondary targeting element could be identified in this region (Fig. 1,A). Mutant HN1, replacing the NH2terminal portion of this helix, had a distribution similar to H3-CA, with localization at centromeres detected over variable levels of nonspecific nuclear staining (Fig. 3, C and D). Mutant HN2, spanning the COOH-terminal portion of the N-helix, targeted normally.
Quantitative assessment of the relative roles of the different targeting elements of CENP-A is complicated by the fact that levels of gene expression vary considerably within the transiently transfected cell population. Even for wild-type CENP-A–HA1, a substantial fraction of cells was observed in which overexpression results in uniform nuclear staining. To compare the targeting defects of the strand A and HN1 mutations, we assayed populations of transfected cells, judging the distribution of epitopetagged CENP-A as being primarily localized at centromeres (e.g., Fig. 2; WT and C), detectably localized (ranging from Fig. 2; sA, to Fig. 3,C), or unlocalized (e.g., Fig. 2, hII). Data are presented in histogram form in Fig. 3,D. For two control constructs assayed simultaneously, the majority of cells had primarily localized epitope with the remaining cells approximately evenly distributed in the other two classes (Fig. 3,D; WT and HC). For the N-helix mutant, only a small proportion of cells (6%) exhibited primarily localized mutant protein, while most cells (49%) showed detectably localized centromeric CENP-A over varying levels of general nuclear staining (Fig 3,D; HN1). For the strand A mutant, no cells were observed with staining primarily at centromeres, and only 18% showed any detectable targeting above the general nuclear staining (Fig. 3 D; HSA). These results suggest that the predicted N-helix of CENP-A contains sequences that are required for efficient targeting to centromeres but to a lesser extent than sequences of strand A or helix II.
Since the long central helix II was the only region that was absolutely required for targeting to centromeres, we sought to determine whether it could act by itself to direct histone H3 preferentially to centromeres. A derivative of histone H3 was constructed by replacing residues 85–112 of histone H3 with the corresponding residues of CENP-A (87–114). The resulting protein showed no ability to localize to centromeres, not even at the level of the strand A mutant of CENP-A (data not shown). Thus, while helix II sequences are required for targeting CENP-A to centromeres, they function only in conjunction with other components of CENP-A.
Self-association of CENP-A Predicts Formation of Homotypic Nucleosomes
The COOH-terminal segment of histone H3 helix II provides a unique function within the nucleosome, mediating protein–protein association at the dyad axis that links the two symmetric halves of the nucleosome (Camerini-Otero and Felsenfeld, 1977; Xie et al., 1996). A requirement of this sequence for targeting CENP-A was revealed by mutant HH2.3 (Fig 3,B), suggesting that protein–protein interactions within the nucleosome are important for CENP-A function. To ask whether CENP-A exhibits selfassociation properties, we constructed a stable HeLa cell line that inducibly expresses the epitope-tagged CENP-A derivative, CENP-A–HA1. Upon induction, cells accumulate CENP-A–HA1 (Fig. 4,A) at their centromeres (data not shown), allowing us to assay protein interactions under conditions in which CENP-A was primarily localized at centromeres. Chromatin was solubilized from isolated nuclei by micrococcal nuclease digestion followed by brief sonication to release centromeric chromatin. After immunoprecipitation from this soluble chromatin extract using mAb 12CA5, fractions were analyzed by SDS-PAGE and immunoblot analysis using human anti–centromere antibodies, allowing detection of both epitope-tagged and endogenous CENP-A (Fig. 4, B and C). Under conditions in which CENP-A–HA1 was present at a lower abundance than endogenous CENP-A, the immunoprecipitated fraction always contained equimolar quantities of endogenous CENP-A recovered with CENP-A–HA1, as judged by Western blot signal intensity (Fig. 4,B). When CENP-A–HA1 was overexpressed relative to endogenous CENP-A, the endogenous protein was still recovered in immunoprecipitates, but in diminished quantities (Fig. 4,C). These data provide strong evidence for CENP-A self-association in vivo. A nucleosome containing CENP-A could in principle be either heterotypic, containing one copy each of CENP-A and histone H3, or homotypic with two copies of CENP-A. Recovery of endogenous CENP-A in the presence of a vast amount of the potential competitor histone H3 demonstrates a preference for self-association. The presence of equimolar amounts of endogenous and epitope-tagged proteins under the conditions of Fig. 4,B, where the quantitatively minor CENP-A–HA1 is essentially doping the CENP-A pool, indicates that this association is highly efficient—essentially all CENP-A–HA1 is present in an equimolar complex. Competition by CENP-A–HA1 when it is quantitatively overexpressed, as in Fig. 4 C, is further evidence for efficient CENP-A/CENP-A self-association. We conclude that CENP-A nucleosomes are homotypic for CENP-A.
Regulatory Determinants of Centromeric Targeting
The assembly of normal histone H3 into chromatin takes place concurrently with DNA replication, as histone H3/H4 tetramers are deposited on newly synthesized DNA within minutes (Worcel et al., 1978). For our initial experiments, we reasoned that expression of CENP-A during S phase would be appropriate, and we prepared a construct in which the coding region of CENP-A was placed under the regulatory signals of a mouse S-phase–dependent histone H3 gene (Taylor et al., 1986; Harris et al., 1991) (Fig. 5). Surprisingly, even at low levels of expression, CENP-A synthesized from this plasmid failed to accumulate at centromeres but was distributed throughout the nucleus (Fig. 5). Since we observe targeting in cells that express CENP-A–HA1 constitutively, we interpret these results to show that uncoupling CENP-A expression from normal histone expression in S phase is an important component of the CENP-A targeting mechanism.
The cell cycle–dependent expression of CENP-A was examined directly in cells synchronized at the G1/S boundary using a double thymidine block procedure. CENP-A mRNA was detected using an RNase protection assay (Fig. 6,A) while histone H3 transcripts were detected by Northern blot analysis (Fig. 6,B). HeLa cells released from a block at the G1/S boundary take 7 h to complete S phase, and spend 3.5 h in G2 and ∼1 h in mitosis (Rao and Johnson, 1970). A plot of the relative abundance of each transcript as a function of time after release (Fig. 6 C) showed that accumulation of histone H3 mRNA paralleled previously published analyses, peaking in mid S phase 4–5 h after release from the thymidine block, followed by a rapid decline to baseline levels by 8–10 h (Harris et al., 1991). In contrast, CENP-A mRNA accumulation did not begin until mid S phase and reached maximal levels 8–10 h after release. CENP-A mRNA levels also rapidly declined between 10 and 12 h after release. CENP-A protein was assayed in parallel by Western blot analysis using a human autoantiserum, showing a gradual increase in abundance starting 4–6 h after release, consistent with an approximate doubling of the CENP-A pool (data not shown).
The pattern of mRNA accumulation observed for CENP-A is similar to that of several cell cycle related gene products, including cdc2, cdc25C, and cyclin A (Dalton, 1992; Zwicker et al., 1995). A common repressor-mediated transcriptional control mechanism has recently been identified among these three cell cycle–regulated genes, conferred by a conserved DNA sequence motif that spans 15 bp located within 20 nucleotides 5′ of the transcription start site (Lucibello et al., 1995; Zwicker et al., 1995). This element contains two conserved segments, seven and five nucleotides in length, separated by a 3-bp linker of unconserved sequence (Fig. 6,D). A genomic clone for human CENP-A was isolated, and a 2.9-kb fragment containing the first exon and 1.1 kb of 5′ flanking genomic DNA was subjected to DNA sequence analysis. A sequence nearly identical to the cell cycle repressor motif was found 11 bp upstream of the 5′ end of the CENP-A cDNA (Fig. 6 D). In CENP-A, the two conserved elements of the motif shared 100% identity with the cell cycle repressor motif. Curiously, these were separated by 8 bp rather than 3 bp, precisely an additional half helical turn of the DNA, as compared with cdc2, cdc25C, and cyclin A. Nevertheless, coupled with the observation that CENP-A mRNA accumulates with a similar kinetic pattern during the cell cycle, it is reasonable to propose that this motif is involved in linking CENP-A gene activity to the cell cycle. Taken together, these results strongly suggest that expression late in the cell cycle is necessary for proper assembly of CENP-A at centromeres.
Three pieces of evidence suggest that CENP-A acts as a core histone, replacing histone H3 within the histone octamer. The first is the biochemical demonstration that CENP-A copurifies with nucleosomes and with the histone H3/H4 tetramer during fractionation of chromatin (Palmer and Margolis, 1985; Palmer et al., 1987). The second is the high degree of amino acid sequence homology shared by CENP-A and histone H3, specifically within the histone fold domain (Palmer et al., 1991; Sullivan et al., 1994). Finally, association with chromatin and a genetic interaction with normal histone H4 suggest that CSE4, the putative S. cerevisiae homologue of CENP-A, is a nucleosomal protein (Stoler et al., 1995; Smith et al., 1996). From these considerations, it is logical to consider the overall organization of CENP-A as similar to that of histone H3 within the histone octamer (Arents et al., 1991; Arents and Moudrianakis, 1993; Richmond et al., 1993).
Structural Basis for CENP-A Assembly at Centromeres
At a structural level, the ability of CENP-A to assemble into centromeric chromatin is specified solely by the histone fold domain. As with the other core histones, histone H3 makes several contacts with nucleosomal DNA as it winds over the surface of the histone octamer (Mirzabekov et al., 1978; Shick et al., 1980; Richmond et al., 1984, 1993; Hill and Thomas, 1990; Arents and Moudrianakis, 1993). Two of the three targeting elements of CENP-A correspond to histone H3–DNA contact sites. The first of these is near the site where DNA enters and exits the octamer, corresponding to the position of the N-helix (Fig. 7,A, peach), which acts as a weak targeting element in our experiments (Richmond et al., 1984, 1993; Hill and Thomas, 1990). A second major H3–DNA contact takes place at the position of strand A and the NH2 terminus of helix II, one of the most concentrated sites of divergence between CENP-A and histone H3 (Mirzabekov et al., 1978; Shick et al., 1980; Richmond et al., 1984, 1993). These sequences form a fairly broad strip on the surface of the nucleosome lying directly across the DNA path (Fig. 7 A, yellow and tan). Strand A is a part of a parallel β sheet structure that has been proposed to act as a specific DNA binding element of the histone octamer (Arents and Moudrianakis, 1993), while the NH2 terminus of helix II is directly adjacent to this region and exposed on the surface. Since small substitutions in strand A had no significant effect on centromeric targeting, it is unlikely that specific side-chain interactions with DNA are required for this region's contribution to the targeting function. Rather, it may act by imparting some general structural features to this portion of the core particle, perhaps influencing the structure of the NH2 terminus of helix II.
A third region of CENP-A that is necessary for targeting is the COOH-terminal portion of the long central helix II (Fig. 7, orange). This region, largely buried in the interior of the H3/H4 tetramer, forms an important protein– protein interaction between the two copies of histone H3, directly on the dyad axis of the nucleosome (Camerini- Otero and Felsenfeld, 1977). This is the only homotypic interchain interaction that can be detected by contact site cross-linking experiments with nucleosome core particles, indicating that the H3/H4 tetramer is held together primarily by this H3–H3 interaction (for review see van Holde, 1989). The role of this region in mediating protein–protein interaction is quite apparent in the structure of the dTAFII62/dTAFII42 heterotetramer, a component of TFIID whose structure is strikingly similar to the heterotetrameric histone H3/H4 core of the nucleosome (Xie et al., 1996). In this structure, two molecules of dTAFII42, the histone H3 homologue, make an extensive contact at the COOH terminus of helix II, which links the two symmetric halves of the heterotetramer. In CENP-A, this region, 109-AYLLTL114, presents more hydrophobic and bulky side chains than the corresponding region of histone H3, 107-TNLCAI112. Thus, this element is situated to affect the protein–protein interactions across the dyad axis of a CENP-A nucleosome, differentiating it from histone H3.
DNA Recognition by Specialized Nucleosomes: A Model
Taken together, these structural considerations suggest a model for the selective recognition of centromeric DNA by CENP-A driven by specialized DNA contact surfaces and self-association. We propose that the specific function of the COOH terminus of helix II is to promote CENP-A– CENP-A self-association, presumably in the context of a (CENP-A/H4)2 heterotetramer, to form a homotypic CENP-A nucleosome. A homotypic CENP-A nucleosome will possess a duplicated set of differentiated DNA contact sites arrayed across the nucleosome surface. The repetition and geometry of these sites provides the possibility for cooperative interaction of the specialized DNA binding surfaces of CENP-A nucleosome, allowing what individually may be only weakly selective binding sites to sum to a significant affinity for centromeric DNA sequence or structure (Fig. 7 B).
Two predictions of this model are that (a) CENP-A should form homotypic nucleosomes, and (b) target DNA should have a repeating substructure that matches the specialized surfaces of CENP-A. Experimental support for the self-association of CENP-A has been obtained by coimmunoprecipitation of endogenous with transfected CENP-A (Fig. 4). While we have not yet determined experimentally the DNA sequences or structures to which CENP-A is bound, it is very likely that they include the satellite DNA component of mammalian chromosomes. Satellite DNA is unconserved at the level of primary sequence (Beridze, 1982). Theoretical analysis of satellite DNAs, however, reveals a substructure comprised of two 50–60-bp bending elements that are separated by 20–30 bp of low bending potential that is conserved among satellites from numerous species (Fitzgerald et al., 1994). Recognition of such conserved structural features of DNA by CENP-A might explain how centromere structure and function are conserved without apparent DNA sequence conservation. The strong strand A–helix II targeting site corresponds to a region where nucleosomal DNA is deformed, bending more sharply across the protein surface than flanking regions (Fig. 7 B, arrows) (Richmond et al., 1984; Wolffe, 1995). DNA bending or curvature is known to be an important determinant of histone octamer positioning on DNA (Shrader and Crothers, 1989; Sivolob and Khrapunov, 1995). Furthermore, analysis of the nucleosomal positioning signal in the ribosomal 5S RNA gene suggests that the regions 2–3 helical turns on either side of the dyad axis, very close to the predicted site of interaction with strand A, play a dominant role in specifying octamer position on the DNA (FitzGerald and Simpson, 1985). DNA recognition by CENP-A may therefore be a specialized implementation of general nucleosomal positioning features. These considerations support the notion that some of the molecular recognition events that specify centromere formation in higher eukaryotes take place at the level of DNA structure rather than DNA sequence, per se, and that these occur in the context of a specialized nucleosome.
Chromatin Assembly and the Specification of Centromeres
Structural recognition alone is not sufficient to explain the specific localization of CENP-A to centromeres, since overexpression of CENP-A results in a distribution throughout the nucleus. Thus, there does not appear to be an efficient mechanism to degrade ectopically localized CENP-A as is observed for CENP-C (Lanini and McKeon, 1995). Rather, our evidence points to regulation of the timing of CENP-A synthesis as an important feature of the targeting mechanism. Restricting expression of CENP-A to S phase abolished targeting, and analysis of steady state CENP-A mRNA abundance revealed that, indeed, it is uncoupled from normal histone expression, beginning late in S phase and extending through G2. Replication of centromeric chromatin therefore occurs through a process that is at least partially independent of normal chromatin replication. One reason for this may be simply to couple CENP-A synthesis with centromere DNA replication, which occurs in mid to late S phase (O'Keefe et al., 1992). A second possibility is that the temporal offset is required to promote the assembly of homotypic CENP-A nucleosomes, by expression at a time when concentrations of potentially competitive histone H3 are diminished. A third explanation for the role of temporal segregated from bulk histone synthesis is that a unique replication pathway for centromeric chromatin is part of the process by which cells recognize and propagate centromeres as distinct functional compartments of the chromosomes. Epigenetic features of centromere structure and function have been identified through analysis of position effect variegation in Drosophila (Spradling and Karpen, 1990; Henikoff, 1992) and of activation of deficient centromere sequences in S. pombe (Steiner and Clarke, 1994). Understanding how CENP-A chromatin replication is linked to the maintenance and function of centromeres on human chromosomes will provide new insight into the question of what constitutes an animal cell centromere.
Why Histone H3?
The heart of the nucleosome is the histone (H3-H4)2 heterotetramer. As discussed above, the heterotetramer possesses most of the DNA binding properties of the nucleosome as well as the information required for positioning (FitzGerald and Simpson, 1985; Dong and van Holde, 1991; Wolffe, 1995) and is deposited first onto DNA after replication, followed by the slower addition of histone H2AH2B dimers (Worcel et al., 1978). Histones H3 and H4 are thus uniquely situated to play a primary role in nucleosomal DNA recognition. Of all the four core histones, only histone H3 has the opportunity to direct its own selfassembly through homotypic interactions (Camerini-Otero and Felsenfeld, 1977; Arents et al., 1991; Xie et al., 1996). Homotypic H3–H3 interactions are therefore a key to harnessing the cooperative binding potential inherent in the dyad symmetry of the nucleosome. Additional histone H3 variants have been identified at the sequence level in Caenorhabditis elegans (Gown et al., 1996) and as a mouse cDNA (GenBank accession number AA008158). The mouse sequence contains a histone fold domain that is only 80% identical to that of mammalian CENP-A and may correspond to mouse CENP-A or represent yet another histone H3 homologue. Taken together, these observations reveal that histone H3 occupies a unique niche in the structure of the nucleosome, one that may provide an important element of adaptability for the structural differentiation of the chromatin fiber.
In summary, analysis of the histone fold domain structures of CENP-A that are required for its localization into centromeres reveals that this process depends upon the specialization of key elements of the histone H3 molecule: DNA binding surfaces and the unique H3–H3 homotypic dimer interface. Examining these features in the context of a histone octamer reveals how these elements can combine to provide modified DNA binding sites distributed in a cooperative array spanning ∼120 bp of nucleosomal DNA. In addition to providing a framework for understanding how centromeric chromatin may be built upon a nucleosomal DNA recognition mechanism, these experiments focus on the unique aspects of histone H3 within the nucleosome. Thus, understanding the relationships between structure and function for the specialized centromeric CENP-A nucleosome may provide new insight into the functions that histone H3 provides for chromatin in general.
We thank H. Damke, S. Schmid, and H. Bujard for their gifts of tTA vectors and HeLa cell line, and E. Chan for the human genomic library.
This work was supported by National Institutes of Health grant GM39068 to K.F. Sullivan, and in part by a grant from the Markey Charitable Trust to the Department of Cell Biology.
1. Abbreviation used in this paper: HA, hemagglutinin.
Address all correspondence to Kevin F. Sullivan, Department of Cell Biology, The Scripps Research Institute, 10550 N. Torrey Pines Rd., La Jolla, CA 92037. Tel.: (619) 784-2350. Fax: (619) 784-2345. e-mail: ksulli firstname.lastname@example.org