Centromeres Are Specialized Replication Domains in Heterochromatin

Centromeres are the attachment points of chromosomes to the mitotic and meiotic spindles. Centromere activity in the budding yeast Saccharomyces is conferred by a specific DNA sequence (Clarke 1990). However, the centromeres of chromosomes in complex eukaryotes are not defined by any distinct sequence: they are usually composed of highly repetitive satellite sequences that are indistinguishable from surrounding noncentromeric satellite blocks (Csink and Henikoff 1998; Murphy and Karpen 1998; Willard 1998). In Drosophila, a centromere was mapped to a 420-kb region that carries full centromeric function, and which is composed of repetitive sequences that are found at noncentromeric regions of chromosomes (Murphy and Karpen 1995; Sun et al. 1997). Minimal centromeres in mammalian cells are also hundreds of kilobases in size, and entirely composed of repetitive alphoid sequences (Lo et al. 1999; Yang et al. 2000). Because centromeres cannot be delineated from their flanking sequences by any sequence criteria, and analyzing highly repetitive DNA is difficult, the size and location of centromeres in most organisms remains unknown. For example, molecular mapping of functional centromeres in Arabidopsis could only map the centromeres to within the extensive repetitive regions of chromosomes that includes megabases of canonical 180-bp repeats, clusters of transposons, and even functional genes (Copenhaver et al. 1999).

Highly repetitive regions of chromosomes adopt a heterochromatic chromatin structure, with distinctive properties and chromatin components (Elgin 1996). Centromeres have a different set of chromatin components, of which centromeric nucleosomes containing special histone H3-like proteins are most conspicuous. Nucleosomes are normally assembled during replication, and it has been suggested that the centromeric histone CENP-A is incorporated at the centromere at the time of its replication (Shelby et al. 1997). Heterochromatin replicates late (Lima-de-Faria and Jaworska 1968), and so centromeres are also thought to replicate late. Very late replication of centromeres has been proposed to play a role in centromere function (Dupraw 1968; Csink and Henikoff 1998). In contrast to expectation, we show that centromeres replicate as isolated domains early in S phase. At this time, they are surrounded by heterochromatin that has not yet replicated. Therefore, a fundamental feature of DNA, replication timing, distinguishes centromeres from surrounding heterochromatin. We further show that nucleosome assembly using histone H3 is inhibited as centromeres replicate. We suggest that pericentromeric heterochromatin sequesters centromeres away from histone H3 and thereby participates in centromere maintenance.

All experiments were conducted using a Drosophila tetraploid Kc cell line. Culturing, transfection methods, and constructs were previously described (Henikoff et al. 2000), except that the colcemid treatment in cytological preparations was omitted. To label replicating DNA with nucleotide triphosphate analogues, we administered a 15-min hypotonic treatment using KHB buffer containing 0.1 mM digoxigenin (dig)-dUTP (Boehringer; Koberna et al. 1999), and then returning cells to insect media for 15 min. For dual-pulse labeling, cells were treated with 0.1 mM dig-dUTP/KHB, resuspended in insect cell culture media for the chase interval, and then treated with 0.1 mM biotin (Bio)-dUTP/KHB (Boehringer).

Labeling of replicating DNA with 5-bromodeoxyuridine was performed by incubating cells in culture medium supplemented with 10 μg/ml BrdU and 10 μg/ml deoxycytidine for 1 h. Immunological detection of Cid was performed on immobilized cells, which were then fixed with methanol/acetic acid/H₂O (11:11:1). Digestion with exonuclease III (30 U in 100 μl buffer) was used to expose the BrdU epitope. BrdU was detected using a FITC-conjugated monoclonal anti–BrdU antibody (Boehringer), and dig-dUTP was detected using FITC- or Texas red–conjugated anti–dig antibodies (Jackson ImmunoResearch Laboratories). In dual-pulse experiments, the two nucleotide analogues were detected using FITC-conjugated anti–dig antibodies and Texas red–conjugated streptavidin (Pierce Chemical Co.).

Transfection efficiencies with histone-green fluorescent protein (GFP) constructs were typically 35–70%. To assay the localization of histone-GFP fusion proteins when replication was blocked, we split one transfected culture into a control and an experimental sample. Aphidicolin was added to the experimental culture at a final concentration of 0.1 mg/ml 5 min before heat shock induction. Cells were allowed to recover after induction for 2 h at 25°C.

HP1 or Cid proteins were detected using anti–HP1 rabbit antibodies (Smothers and Henikoff 2001) or by using anti–Cid rabbit antibodies (Henikoff et al. 2000) followed by anti–rabbit IgG Cy5-conjugated goat antibodies (Jackson ImmunoResearch Laboratories). All images were collected as previously described (Henikoff et al. 2000).

Images were analyzed using DeltaVision software (Applied Precision). Cells were categorized according to their overall pattern of replication (euchromatic, scattered, or heterochromatic). Centromeres and nuclei were identified by thresholding anti–Cid signals and DAPI fluorescence, respectively.

To measure the amount of nucleotide incorporated in centromeres, we examined all optical sections that included an anti–Cid signal. We selected nuclei in which euchromatin was labeled, summed the intensity of the nucleotide analogue signal in an area defined by Cid signals, and divided this by the summed nucleotide signal intensity within the euchromatin of the same section. Both sums were corrected for background. The summed signal intensity indicates the amount of replication in a cell during the labeling period, and the average ratio from randomly selected early S phase cells (n = 22) is an estimate of the fraction of replication at centromeric foci relative to euchromatin. These integrated intensities allow an estimate of the amount of DNA replicated early in S phase at a centromeric replication focus. The amount of nucleotide analogue incorporated at centromeric replication foci above background relative to that incorporated in euchromatin throughout the early S phase period was 0.005 (SEM = 0.002). Since euchromatin contains ∼120,000 kb of DNA (Adams et al. 2000), the average centromeric replication focus contains ∼500 kb of DNA.

The Hsp70 promoter has very low-level constitutive activity, so in all cases we determined the mean GFP signal from uninduced control samples and removed from the analysis any nuclei in which the GFP signal was within two standard deviations of this mean. We counted the number of nuclei with appropriate GFP signals (Cid-GFP localized to centromeres or H3GFP labeling replicating DNA), and scaled these numbers to the transfection efficiency (estimated as the number of nuclei showing localization after induction and with no aphidicolin treatment). The amount of histone-GFP fusion proteins incorporated at centromeres when produced from cid promoter constructs was estimated by comparing the integrated intensities from GFP in euchromatin to that in centromeres (n = 23 for cid-H2BGFP, n = 10 for cid-H3GFP).

Immortal cell lines are traditionally used to examine replication timing in higher eukaryotes, because cell types are uniform and they can be easily manipulated. We used Drosophila Kc cells, which have been used in previous studies of replication timing, and have been documented as having a stable karyotype (Echalier 1997). In Kc nuclei, there is a clear separation between the early replication of euchromatin and the late replication of heterochromatin during the ∼10-h-long S phase (Barigozzi et al. 1966; Dolfini et al. 1970). Furthermore, the heterochromatic regions of chromosomes coalesce into a large irregular chromocenter that occupies ∼25% of the nuclear volume and can be visualized by detection of the HP-1 protein (van Steensel and Henikoff 2000; Fig. 1A and Fig. B). Centromeres are embedded within the chromocenter (van Steensel and Henikoff 2000). A chromocenter is not unique to the Kc nucleus, as we have observed consistent chromocenter formation in other dividing cells as well, including S2 and S3 cell lines (Ahmad, K., unpublished results). This chromocentric organization of the nucleus and the regularity of the cell cycle allowed us to precisely follow replication timing of centromeres in asynchronous Kc cell populations.

Pulse-labeling of Kc cells with the nucleotide analogue dig-dUTP confirms earlier studies of mitotic replication banding (Barigozzi et al., 1966; Dolfini et al. 1970,) that heterochromatin and euchromatin replicate in distinct portions of S phase (Fig. 1A and Fig. B). A typical experiment showed that 51% of nuclei were in S phase at the time of the pulse: 23% showed euchromatic replication patterns, 25% showed heterochromatic replication patterns, and 3% showed discrete foci scattered throughout both regions. This last type indicates that there is only minor overlap of the euchromatic and heterochromatic replication periods during the pulse. Euchromatin replicates in the first portion of S phase (early S), and heterochromatin in the last (late S) (Barigozzi et al. 1966; Dolfini et al. 1970). This spatial and temporal separation of replication in Kc cells facilitates a cytological analysis of centromere replication.

We have previously described the centromere identifier (cid) gene in Drosophila, which encodes a centromeric histone H3-like protein (Henikoff et al. 2000). This protein is analogous to CENP-A in mammals (Palmer et al. 1991), Cse4p in Saccharomyces (Stoler et al. 1995), HCP-3 in nematodes (Buchwitz et al. 1999), and SpCenpA in Schizosaccharomyces (Takahashi et al. 2000). All of these proteins are thought to be assembled into specialized nucleosomes, although direct evidence has only been obtained for CENP-A (Palmer et al. 1987; Yoda et al. 2000). Because Cid appears to be a constitutive centromeric component, antibodies to Cid can be used to locate centromeres in interphase nuclei, and typically detect four to six centromeric spots (5.6 ± 2.1) in both diploid and tetraploid Kc cells (Henikoff et al. 2000). Chromosomes in Drosophila somatic cells are paired; therefore, each centromeric spot probably represents a cluster of centromeres from homologous chromosomes, regardless of their total number. These spots divide into individual centromeres only in late-G2 phase of the cell cycle (data not shown).

We used anti–Cid antibodies on nuclei that had been pulse labeled with nucleotide analogues to determine when centromeric DNA replicates. The Cid epitope and nucleotide analogues colocalize when euchromatin is also replicating, revealing that centromeres replicate in early S phase (Fig. 1 C). Centromeres appear to be the only sites in the chromocenter to replicate at this time. Of 42 nuclei in which euchromatin was labeled, 20 also showed labeling at centromeres, indicating that centromeres are replicating through about one half of early S phase. However, within that period, centromeres must replicate asynchronously, since we observed incorporation at only a subset of centromeric spots in any one nucleus. Asynchronous replication of centromeric satellite has been observed in human cells (O'Keefe et al. 1992).

The nucleotide triphosphate pulse labeling procedure that we used allows for detection of incorporation without requiring DNA denaturation, thus providing excellent cytology. The brief hypotonic shock required to deliver nucleotide triphosphate analogues was found to have little or no ill effects on a wide variety of cell processes (Koberna et al. 1999). To rule out the possibility that the hypotonic treatment used to deliver nucleotide triphosphate analogues had affected DNA replication, we performed the incorporation using BrdU, which penetrates cells without hypotonic treatment. Identical results were obtained (Fig. 1 D). We conclude that replication at centromeres initiates substantially before that of the rest of the DNA in the heterochromatic compartment of the nucleus.

We determined the relative timing of replication in different regions in a nucleus using two sequential pulses of nucleotide analogues. A pulse of dig-dUTP nucleotide analogue was administered, and this was followed by a chase in regular media and a second pulse of Bio-dUTP. Nuclei that had been labeled with a chase interval of 1, 2, or 3 h generally showed clear distinction between sites of dig- and Bio-dUTP incorporation, consistent with the completion of euchromatic replication in individual foci within these time intervals (Fig. 2A and Fig. B; Manders et al. 1992). We categorized cells according to their overall pattern of nucleotide incorporation from these two pulses (Fig. 2 E: E/E, euchromatin/euchromatin; E/H, euchromatin/heterochromatin; or H/H, heterochromatin/heterochromatin), and then assessed incorporation at centromeres in each category. Some cells with an E/E pattern showed no incorporation of dig-dUTP at centromeres, but did show centromeric incorporation of the second nucleotide Bio-dUTP (Fig. 2A and Fig. E). Incorporation of dig-dUTP but not Bio-dUTP at the centromeres of E/E pattern nuclei was never observed (n = 125). These results demonstrate that replication begins at euchromatic sites before centromeric origins fire. We also observed nuclei with an E/E labeling pattern in which both nucleotide analogues were incorporated into the same centromeric site, suggesting that this site has been replicating for the entire 1-h interval between the two pulses (Fig. 2 E, right-most column). The frequency of centromeres labeled with both nucleotide analogues suggests that replication continues for 2–3 h during the early S-phase period. Experiments in which the two nucleotide analogues were delivered 2 and 3 h apart confirmed this calculation, because occasional nuclei were seen in which both dig- and Bio-dUTP were incorporated at a single centromeric spot (Fig. 2 B). This is an unusually prolonged period of replication since most sites complete replication in 40–60 min (Fig. 2; Manders et al. 1992; Ma et al. 1998). Based on the integrated intensity of nucleotide incorporation at centromeric replication foci (see Materials and Methods), we estimate that ∼500 kb of DNA at each centromere within a centromeric replication focus is replicated in the early S-phase period.

The detection of early replication at centromeres is aided by their clear spatial separation from simultaneously replicating euchromatin. In nuclei with heterochromatic labeling patterns, many replication foci are near centromeres (Fig. 2C and Fig. D), and the high density of foci might obscure centromere replication. To examine whether any replication occurs at centromeres in the late S period, we quantitated the amount of incorporated nucleotide analogue in centromeric spots in the early and late periods. These measurements show that ∼90% of all detectable nucleotide incorporation at centromeres occurs in the early period (316 U mean intensity/pixel per centromere in the early period versus 34 U in the late period). Thus, the upper limit on the amount of replication at centromeres that occurs in the late S period is ∼10%, but even this may be due to nearby replication foci that could not be distinguished from centromeres. We conclude that centromeric replication is largely complete before the rest of the heterochromatic chromocenter replicates.

Models for centromere specification by unique DNA sequences are ruled out by the absence of centromere-specific sequences in natural centromeres (Karpen and Allshire 1997), and the establishment of neocentromeres at previously noncentromeric regions (Warburton et al. 2000). An attractive alternative to DNA sequence–based specification of centromeres is that centromeric nucleosomes define the centromere. In all eukaryotes, centromeres are likely to be contained in specialized nucleosomes that include an H3-like protein (Tyler-Smith and Floridia 2000). We have previously shown that Cid-GFP fusion protein produced from the cid promoter localizes to centromeres, and that this promoter is active early in S phase (Henikoff et al. 2000). Thus, the normal localization of Cid takes place as centromeric DNA is replicating. However, the deposition of specialized nucleosomes at replicating centromeres must be difficult because histone H3 is in vast excess throughout S phase (Osley 1991).

We wondered whether the deposition of Cid in centromeric replication domains might be facilitated by preventing the deposition of histone H3. Conventional nucleosomes are normally assembled during replication; therefore, we examined the deposition of conventional histones fused to GFP when centromeres replicate. Previous work has shown that histone-GFP fusion proteins can localize to chromatin (Kanda et al. 1998; Henikoff et al. 2000). We had previously characterized the deposition of H2B-GFP and H3-GFP fusion proteins by examining mitotic chromosomes. When expressed from the cid promoter, which is active early in S phase, these fusion proteins localize to the euchromatic arms of chromosomes (Henikoff et al. 2000). These experiments did not address whether small quantities of the fusion proteins were also incorporated at centromeres. Therefore, we examined the deposition of histone-GFP fusion proteins after production from transfected cid-promoter constructs in interphase nuclei to determine whether these histones were incorporated in centromeric replication domains. While H2B-GFP was readily deposited at centromeres, H3-GFP was not (P = 0.0002; Fig. 3A and Fig. B). Both histone-GFP proteins give similar intense labeling in euchromatin, implying that the lack of H3-GFP at centromeres is not due to a general reduced deposition of H3-GFP. Labeling in heterochromatin by H3-GFP and H2B-GFP produced from heat-shock promoter constructs was also indistinguishable in intensity and pattern (Fig. 3C and Fig. D). That the centromere may be deficient of histone H3 has been previously suggested, based on the absence of phosphorylated H3 antibody labeling (van Hooser et al. 1999), and our results with H3-GFP fusion protein support this idea. We conclude that the deposition of H3-containing nucleosomes is inhibited at the time that the cid promoter is active in early S phase.

The centromeric histone CENP-A can form nucleosomal particles in vitro (Yoda et al. 2000), and cofractionates with other histones in vivo (Palmer et al. 1987; Shelby et al. 1997). The similarity of Drosophila Cid to histone H3 suggests that Cid is also incorporated into nucleosomes. Since the deposition of histone H3 is inhibited as centromeres replicate, Cid-containing nucleosomes might be formed using an alternate pathway. To investigate the nature of a Cid deposition pathway, we tested the dependence of histone-GFP protein localization on DNA replication. A pulse of histone-GFP fusion proteins can be produced in cells by transfecting with a heat-shock promoter construct, and then inducing the promoter. Newly produced H2B-GFP and H3-GFP proteins localize in subnuclear patterns similar to patterns produced by pulses of nucleotide analogues (Fig. 3; Henikoff et al. 2000), and fail to localize in cells that are blocked by the replication-inhibiting drug aphidicolin (Fig. 4 A). Thus, the deposition of these fusion proteins is coupled to the replication of the underlying DNA, just as is the formation of conventional nucleosomes (Adams and Kamakaka 1999; Krude 1999).

To determine whether the localization of Cid to centromeres is also coupled to replication, we transfected a heat shock promoter-Cid-GFP fusion construct (HS-CidGFP) into cells. Cells were then induced to express the fusion protein, or treated with aphidicolin to block DNA replication, and then induced. Centromeric localization of Cid-GFP occurred both in treated and untreated cells (Fig. 4 A), demonstrating that the deposition of Cid need not be coupled to DNA replication. The replication independence of Cid-GFP localization distinguishes it from that of the bulk of conventional histones.

We further examined the localization of Cid-GFP to centromeres when produced at different times in the cell cycle. The deposition of histones onto DNA is predominantly limited to cells that are in S phase of the cell cycle (Wu et al. 1986). We transfected cells with heat shock promoter-histone-GFP fusion constructs (HS-H3GFP or HS-CidGFP). We then induced the construct and pulsed these cells with nucleotide analogues to identify cells in S phase. A pulse of H3-GFP protein localized in early and late replication patterns similar to the nucleotide analogue in the same cells, and no detectable localization occurred in gap-phase cells (Fig. 4 B, left). However, we found that newly produced Cid-GFP protein localized to centromeres in both S-phase and gap-phase cells with similar efficiencies (Fig. 4 B, right). We confirmed that localization of Cid-GFP protein can occur in post-replicative cells, whereas that of H3-GFP is strictly replication dependent, by examining metaphase chromosomes at various times after induction of transfected constructs. In cells transfected with HS-H3GFP, occasional labeled metaphase figures first appear 4 h after induction and are labeled in heterochromatic segments (Fig. 5 A). The pattern of labeling and timing of these figures reflects the deposition of H3-GFP in late S-phase cells and the shortest time required to traverse the G2 phase of the cell cycle, and is consistent with previous observations in Kc cell populations (Barigozzi et al. 1966; Dolfini et al. 1970). However, Cid-GFP first appears at the centromeres in 48% of mitotic figures within 2 h of induction: these must be from cells that were induced during the G2 phase (Fig. 5 B). By 4 h s after induction, 100% of the mitotic figures from transfected cells show centromeric labeling. We conclude that the Cid deposition pathway is present and active throughout the cell cycle.

Analysis of centromeres in complex eukaryotes has been hampered by the lack of sequence differences between the centromere and flanking heterochromatin, and the repetitive nature of these regions (Sun et al. 1997; Willard 1998). These sequence commonalities have led to the attribution of heterochromatic features to the centromere, including late replication. Our analysis demonstrates that the replication of centromeres in Drosophila cells actually precedes that of pericentromeric heterochromatin. We estimate that, on average, ∼500 kb of centromeric DNA is replicated in the early S phase period. This size is in agreement with a genetically defined fully functional centromere in Drosophila (Murphy and Karpen 1995), suggesting that the early replication domain corresponds to the complete centromere. The early timing of its replication distinguishes the centromere from other repetitive sequences and rules out models for defining centromeres that have invoked their very late replication (Dupraw 1968; Csink and Henikoff 1998). Early replication appears to be a general feature of centromeres, as Saccharomyces centromeres are known to replicate early in S phase (McCarroll and Fangman 1988).

Our observations on the controlled assembly of conventional (H3-containing) and specialized (Cid-containing) nucleosomes at replicating centromeres suggests that chromatin assembly is a critical step in centromere maintenance. The cid promoter drives expression early in S phase (Henikoff et al. 2000), and centromeres are replicating during this time. Therefore, Cid synthesis and centromere replication appear to be tightly coordinated. In Schizosaccharomyces yeast, the Cnp1 gene (encoding the centromeric SpCenpA histone) is also expressed early in S phase (Takahashi et al. 2000), and we expect that a similar coordination with centromeric replication will be found.

At the time that Cid is being deposited, H3 deposition is inhibited. It is striking that early replicating centromeres are typically surrounded by late-replicating heterochromatin, and we suggest that inhibiting histone H3 incorporation at centromeres when they replicate is one function of this juxtaposition. Inhibiting histone H3 incorporation at centromeres requires the uncoupling of conventional chromatin assembly and DNA replication. These two processes are thought to be linked by interactions between replication machinery and the CAF1 chromatin assembly factor (Krude 1995; Verreault et al. 1996). Uncoupling may be accomplished if histone H3 or some component of its assembly machinery is excluded from the heterochromatic chromocenter early in S phase. Regions deficient in histone H3 would then be incorporated into Cid-containing nucleosomes by a replication-independent pathway.

The observation that centromeric histone H3-like proteins from worms and yeast preferentially localize to fly or human heterochromatin (Henikoff et al. 2000) suggests that heterochromatin sequesters centromeric H3-like proteins in general. Sequestering Cid in the heterochromatic chromocenter would increase the local concentration of Cid around centromeres and thereby promote Cid deposition. Centromeres in many organisms are typically surrounded by heterochromatin, and genetic evidence suggests that heterochromatin is important for centromere function (Allshire et al. 1995). The centromeres of Saccharomyces chromosomes are the only known exception to this rule, but in this organism centromeric activity is conferred by a specific DNA sequence and associated DNA-binding proteins (Meluh and Koshland 1997). The importance of heterochromatin for the function of complex centromeres is reinforced by the finding that a human neocentromere shows M31 staining (a marker for mammalian heterochromatin), whereas the parental chromosomal region does not (Saffery et al. 2000). Perhaps the exclusion of histone H3 during replication is one of the prerequisites for “centromerization” (Choo 2000), thus necessitating that neocentromeres acquire heterochromatic proteins.

We expect that centromeres must be protected from conventional nucleosome assembly pathways in all dividing cells, but heterochromatin may not always perform this function. For example, distinct heterochromatin does not form in the rapidly dividing nuclei of Drosophila syncytial embryos (Hiraoka et al. 1993), and replication initiates throughout the chromosomes simultaneously (Kriegstein and Hogness 1974). In these unusual nuclei, conventional nucleosome assembly might be prevented by excluding histone H3 from the apical edge of interphase nuclei, where centromeres lie (Foe et al. 1993). Similarly, varying local concentrations of proteins around nuclei have been proposed to explain progression of the syncytial cell cycle even though bulk cyclin levels are always high (Foe et al. 1993). In later cycles, it appears to be most efficient to produce Cid when centromeres replicate.

It has been of great interest to understand how the location of the centromere is stably maintained in successive cell divisions, because it does not appear that DNA sequence is responsible (Brown and Tyler-Smith 1995; Karpen and Allshire 1997). Nucleosome particles form the fundamental unit of chromatin, and so an attractive alternative to DNA sequence-based inheritance of centromere identity is that centromeric nucleosomes participate in centromere maintenance (Palmer et al. 1990). Replication initiation appears to depend on chromatin structure (Ofir et al. 1999; Stevenson and Gottschling 1999) and we suggest that Cid-containing nucleosomes predispose DNA to replicate early. This early replication and the exclusion of histone H3 in heterochromatin would preclude conventional chromatin assembly, thus allowing the assembly of Cid-containing nucleosomes and ensuring early replication again in the next cycle. This process would maintain centromeres.

This work is dedicated to the memory of Doug Palmer. We thank members of the Henikoff lab for many valuable discussions, Jim Smothers for HP1 antibody, and Adrian Quintanilla for advice on image analysis.

K. Ahmad is an American Cancer Society postdoctoral fellow.