Detrimental incorporation of excess Cenp-A/Cid and Cenp-C into Drosophila centromeres is prevented by limiting amounts of the bridging factor Cal1

The centromeric regions of chromosomes direct formation of kinetochores, which allow chromosome attachment to spindle microtubules. Centromeres and kinetochores are therefore of paramount importance for faithful propagation of genetic information (Santaguida and Musacchio, 2009). However, centromeric DNA sequences are not conserved (Vagnarelli et al., 2008). Most eukaryotes (including Drosophila melanogaster and humans) have regional centromeres with up to several megabases of repetitive DNA. Importantly, these repetitive sequences are neither necessary nor sufficient for centromere function, indicating that there is an epigenetic centromere specification (Vagnarelli et al., 2008).

A centromere-specific histone H3 variant (CenH3) is thought to be crucial for epigenetic centromere marking (Allshire and Karpen, 2008). CenH3 proteins are present in all eukaryotes (e.g. CENP-A in humans and Cid in Drosophila). They replace histone H3 in canonical nucleosomes or possibly variant complexes (Dalal et al., 2007; Mizuguchi et al., 2007; Camahort et al., 2009; Furuyama and Henikoff, 2009). Depletion of CenH3 results in a failure to localize most or all other centromere and mitosis-specific kinetochore proteins. Strong overexpression of Drosophila Cenp-A/Cid results in incorporation at ectopic chromosomal sites, which in part also assemble ectopic kinetochores during mitosis (Ahmad and Henikoff, 2002; Heun et al., 2006).

Ectopic kinetochores result in chromosome segregation errors and genetic instability. Ectopic CenH3 incorporation therefore must be prevented. Although still fragmentary, our understanding of the molecular mechanisms that regulate CenH3 incorporation is progressing rapidly (Allshire and Karpen, 2008; Torras-Llort et al., 2009). In proliferating cells, an additional complement of CenH3 needs to be incorporated during each cell cycle. In syncytial Drosophila embryos, this occurs during exit from mitosis (Schuh et al., 2007). Similar findings were made in human cells, where Cenp-A deposition occurs during late telophase and early G1 phase (Jansen et al., 2007). The number of factors shown to be required for normal CenH3 deposition is increasing rapidly, which suggests that there is an intricate control mechanism. Various and often dedicated chaperones (Hayashi et al., 2004; Furuyama et al., 2006; Dunleavy et al., 2009; Foltz et al., 2009), chromatin modifying and remodelling factors (Fujita et al., 2007; Maddox et al., 2007; Perpelescu et al., 2009), as well as other centromere components (Takahashi et al., 2000; Okada et al., 2006; Pidoux et al., 2009; Williams et al., 2009) are involved.

In Drosophila, Cenp-C is incorporated into centromeres concomitantly with Cenp-A/Cid (Schuh et al., 2007). High-resolution mapping with native Drosophila chromosomes has indicated that these two proteins do not have an identical localization within the kinetochore (Blower et al., 2002; Schittenhelm et al., 2007). Although these localization studies cannot exclude an association between subfractions of Cenp-A and Cenp-C, direct molecular interactions between these centromere proteins have not yet been reported. Recently, however, Cal1 has been identified in Drosophila and shown to be required for normal centromeric localization of Cenp-A/Cid and Cenp-C (Goshima et al., 2007; Erhardt et al., 2008). Moreover, these three Drosophila centromere proteins can be co-immunoprecipitated from soluble chromatin preparations (Erhardt et al., 2008). Cal1 might therefore provide a physical link between Cenp-A/Cid and Cenp-C.

Here, we report that Cal1 has distinct binding sites for Cenp-A/Cid and Cenp-C. It can link these proteins together according to yeast three-hybrid experiments. However, the level of centromeric Cal1 is far lower than that of Cenp-A/Cid and Cenp-C. Cal1 therefore cannot function as a stoichiometric linker connecting each monomer or dimer of Cenp-C to Cenp-A within the centromere. But the low levels of Cal1 effectively protect cells against mitotic defects resulting from increased centromeric incorporation of excess Cenp-A/Cid and Cenp-C.

RNAi-mediated knockdown of cal1 has been shown to result in substantially diminished Cenp-A and Cenp-C levels at centromeres in Drosophila tissue culture cells (Goshima et al., 2007; Erhardt et al., 2008). For a genetic analysis of cal1 function, we first characterized cal1 alleles (Fig. 1A). The allele cal1^c03646 was confirmed to carry a pBAC{PB} insertion 69 bp upstream of the start codon within the predicted 5′ untranslated region. The Mi{ET1} insertion in cal1^MB04866 is within the second exon and disrupts the coding sequence after 361 of a total of 979 amino acids. Both insertions are associated with recessive lethality. They failed to complement each other, as well as the deficiency Df(3R)Exel6176, which deletes the cal1 gene. The gcal1-EGFP II.2 transgene, a genomic cal1 fragment with the EGFP coding sequence inserted immediately before the stop codon (Fig. 1A), completely prevented the lethality of homo-, hemi- and transheterozygous cal1^MB04866 flies. Moreover, the rescued flies were found to be fertile. These findings demonstrate that cal1 is an essential gene and that the Cal1-EGFP fusion provides all essential Cal1 functions.

To characterize the expression pattern of cal1, we used gcal1-EGFP II.2 embryos. Microscopic analyses as well as immunoblotting experiments (supplementary material Fig. S1) indicated the presence of a maternal cal1 contribution at the onset of embryogenesis, as well as a correlation of cal1 expression with mitotic proliferation.

The maternal cal1 contribution is expected to delay the onset of phenotypic abnormalities in cal1 mutants. First abnormalities became apparent during stage 12. At this and later stages, abnormalities were largely restricted to the developing CNS (Fig. 1B). DNA staining revealed a lower number and a more irregular distribution of nuclei in the CNS of cal1 mutants compared with sibling embryos. In addition, pyknotic nuclei as well as enlarged over-replicated nuclei were more frequently observed in the mutant CNS. Phospho-histone-H3-positive mitotic cells were also more frequent and often enlarged in the mutant CNS. The great majority of mutant progeny did not reach the larval stages (97%, n=100). Comparable observations were made with homo-, hemi- and transheterozygous embryos, suggesting that both alleles (cal1^MB04866 and cal1^c03646) result in a complete loss of gene function. The observed abnormalities in cal1 mutants are consistent with the proposal that after exhaustion of the maternal cal1 contribution, proliferating cells progress through aberrant mitoses with chromosome segregation errors resulting in aneuploidy and apoptosis.

Even before the onset of mitotic abnormalities, cal1 homozygous mutant embryos displayed weaker anti-Cenp-A/Cid signals than sibling embryos (Fig. 1B). Later, when the abnormalities in the CNS became evident, Cenp-A/Cid could no longer be detected in cal1 mutants. Moreover, the same results were also obtained with anti-Cenp-C, as well as with transgenes expressing EGFP fusions of Cenp-C, Spc105, Mis12, Nsl1, Spc25, Ndc80 and Nuf2 (supplementary material Fig. S2). The localization of all these centromere and kinetochore proteins requires Cal1. However, centromere localization of Cal1 was found to depend on Cenp-A/Cid and Cenp-C, but not on Spc105, Mis12 and Ndc80 complex components (supplementary material Fig. S2). Moreover, in contrast to initial descriptions (Heeger et al., 2005; Przewloka et al., 2007), quantification of anti-Cenp-A/Cid signals in Cenp-C mutant embryos confirmed (data not shown) that normal levels of centromeric Cenp-A/Cid depend on Cenp-C (Erhardt et al., 2008). All our findings in mutant embryos confirm and extend previous observations made after RNAi in Drosophila tissue cultures (Goshima et al., 2007; Erhardt et al., 2008). Cal1 clearly functions together with Cenp-A/Cid and Cenp-C in kinetochore assembly.

Cal1, Cenp-A/Cid and Cenp-C can be co-immunoprecipitated (Erhardt et al., 2008). We analyzed whether Cal1 can interact simultaneously with Cenp-A/Cid and Cenp-C. Yeast two-hybrid (Y2H) assays clearly revealed an interaction of Cal1 with Cenp-A/Cid (Fig. 2A), but not with the kinetochore proteins Spc105, Mis12, Nsl1, Nnf1a, Bub1 or BubR1 (data not shown). The N-terminal region of Cal1 (residues 1–407) but not its middle (residues 392–722) and C-terminal (residues 699–979) regions were observed to interact with full-length Cenp-A/Cid (Fig. 2A). When the N-terminal tail or the histone fold domain of Cenp-A/Cid was assayed separately, no interactions with the N-terminal Cal1 region could be detected (data not shown). Y2H experiments also revealed an interaction between Cal1 and Cenp-C. The C-terminal regions of Cal1 (residues 699–979) and Cenp-C (residues 1009–1411) were found to interact (Fig. 2B). The interacting region within Cenp-C could be narrowed down to a smaller C-terminal subfragment (residues 1201–1411), which no longer included the Cenp-C box, a motif that is characteristic of all Cenp-C homologs (Fig. 2B). The C-terminal Cenp-C domain, which is sufficient for the Y2H interaction with Cal1, is similar in fungal and animal Cenp-C homologs (Talbert et al., 2004). It adopts a cupin fold and can mediate homodimerization (Cohen et al., 2008).

As Cenp-A/Cid and Cenp-C were observed to interact with distinct regions of Cal1, we evaluated whether Cal1 can bind to Cenp-A/Cid and Cenp-C simultaneously to form a trimeric complex (Fig. 2C). We generated a yeast strain constitutively expressing Cenp-A/Cid and Cenp-C(C) fused to the transcriptional activation and DNA-binding domain of Gal4, respectively. In addition, the strain allowed for regulated cal1 expression. In the absence of cal1 expression, we did not observe an interaction between Cenp-A/Cid and Cenp-C(C) (Fig. 2C). However, in the presence of cal1 expression, we clearly observed a Cenp-A/Cid–Cenp-C(C) interaction (Fig. 2C). Control experiments demonstrated that the inducing growth conditions were unable to promote a Cenp-A/Cid–Cenp-C(C) interaction when the inducible cal1 gene was absent (data not shown). The results of our yeast three-hybrid (Y3H) experiments therefore indicate that Cal1 can bridge Cenp-A/Cid and Cenp-C (Fig. 2D).

A perfect colocalization of Cenp-A/Cid, Cenp-C and Cal1 would be expected, if these proteins were present exclusively in a trimeric complex. Therefore, we carefully compared the localization of Cal1-EGFP with that of Cenp-A/Cid and Cenp-C in embryos and S2R+ cells expressing the gcal1-EGFP construct (supplementary material Fig. S3). Cal1-EGFP was observed at centromeres throughout the cell cycle. Importantly, during interphase, Cal1-EGFP signals, but not anti-Cenp-A/Cid and anti-Cenp-C signals, were also clearly enriched in and around the nucleolus. Our results therefore correspond to those described earlier (Erhardt et al., 2008) where antibodies against Cal1 were used. These results indicate that at least a fraction of Cal1 is not associated with Cenp-A/Cid and Cenp-C during interphase.

To evaluate which Cal1 domains contribute to localization, we generated constructs allowing expression of either the N-terminal, middle or C-terminal region fused to EGFP (Fig. 3A). The N- and C-terminal domains, which are sufficient for the interaction with either Cenp-A/Cid or Cenp-C, have been conserved more extensively during Drosophilid evolution than the middle region (Erhardt et al., 2008). None of the three Cal1 subregions was able to localize to the centromere in S2R+ cells (Fig. 3A). The middle, but not the terminal domains, became enriched in the nucleolus. To further define the requirements for Cal1 centromere localization, we generated constructs that allowed expression of different combinations of Cal1 domains (Fig. 3A). After expression of Cal1(N-M) or Cal1(M-C) we did not observe centromeric signals. However, these Cal1 fragments became enriched in the nucleolus (Fig. 3A), as expected because they contain the middle domain, which is sufficient for nucleolar localization. In agreement, Cal1(N-C), a Cal1 version lacking the middle domain, was not enriched in the nucleolus. Interestingly, however, this variant was found at the centromere throughout the cell cycle (Fig. 3A). Moreover, Y3H experiments indicated that Cal1(N-C) was still able to forge an interaction between Cenp-A/Cid and Cenp-C (data not shown). These findings suggest that Cal1 centromere localization depends on an interaction with Cenp-A/Cid and Cenp-C. Cal1 might be sequestered in the nucleolus when not in a complex with these centromeric proteins.

By expressing EGFP-tagged Cal1 variants in cal1 mutant embryos, we evaluated to what extent the different Cal1 domains contribute to its function. Expression of the regions N, M or C (Fig. 3A) from UAS transgenes could be confirmed by the resulting EGFP signals (data not shown), but did not restore centromeric Cenp-A/Cid localization and normal cell proliferation in the CNS of cal1 mutant embryos (Fig. 3B). However, expression of UAS-cal1(N-C)-EGFP prevented expression of the characteristic abnormalities in cal1 mutant embryos (Fig. 3B). This rescue was just as effective as that with full-length Cal1 (UAS-cal1-EGFP, data not shown) and resulted in an apparently wild-type CNS. Moreover, ubiquitously expressed UAS-cal1(N-C)-EGFP allowed development of cal1 mutants to the adult stage (data not shown). Importantly, simultaneous expression of UAS-cal1(N)-EGFP and UAS-cal1(C)-EGFP did not prevent the cal1 mutant phenotype (Fig. 3B). In addition, EGFP signals were not centromeric, in contrast to those obtained with UAS-cal1(N-C)-EGFP (data not shown). Therefore, we conclude that centromere localization and function of Cal1 require the presence of its Cenp-A/Cid- and Cenp-C-interacting N- and C-terminal regions, which have to be linked, but not necessarily by its M region.

The molecular interactions that are responsible for Cenp-C localization within the mitotic kinetochore are unknown. Human Cenp-C binds to DNA in vitro, although with very limited sequence preference (Yang et al., 1996; Sugimoto et al., 1997). Co-immunoprecipitation of human Cenp-C and Cenp-A has been reported (Erhardt et al., 2008; Trazzi et al., 2009), but others have failed to detect Cenp-C in Cenp-A nucleosomes (Hori et al., 2008). The co-immunoprecipitation data of Erhardt and colleagues (Erhardt et al., 2008) and our Y3H experiments are consistent with the notion that in Drosophila, Cal1 might function as a centromere component that stoichiometrically links Cenp-C to Cenp-A/Cid. To evaluate this possibility, we carefully quantified the centromeric amounts of these proteins. Wing imaginal discs of Cenp-A/cid-, Cenp-C- or cal1-null mutant larvae rescued by transgenes expressing functional EGFP fusions of these proteins were mounted next to CSE4::EGFP yeast cells (Fig. 4A). CSE4 encodes the yeast Cenp-A homolog, which is thought to be present in two copies per centromere (Meluh et al., 1998; Collins et al., 2004). Accordingly, we used the clusters of the 16 centromeres in CSE4::EGFP anaphase or telophase cells (Fig. 4A; Joglekar et al., 2006) as an internal calibration standard for the quantification of Cenp-A/Cid-EGFP, Cenp-C-EGFP and Cal1-EGFP signal intensities in Drosophila centromeres. The measured Cse4-EGFP signal intensities were found to decrease with increasing distance of the centromere cluster from the coverslip (Fig. 4B), as described previously (Joglekar et al., 2006). In groups of cells with centromere clusters at similar focal positions, the s.d. values of the Cse4-EGFP signals were found to be lower than 36% of the average signal intensity. The EGFP signal intensities measured for the Drosophila centromere protein fusions that were expressed under the control of their own regulatory regions were also plotted against their average focal z-axis positions (Fig. 4C). The y-axis intercepts of linear regressions were used for comparison of the average amounts of different centromere proteins (Table 1). Moreover, the comparison of the EGFP signal intensities obtained in Drosophilawith those of Cse4-EGFP in yeast resulted in an estimate of the absolute protein copy numbers per centromere (Table 1). The accuracy of our quantifications was confirmed in competition experiments, where we observed the expected decrease in EGFP signal intensities when a given EGFP fusion protein was analyzed in a background that also expressed the untagged version of this protein from endogenous wild-type gene copies, rather than in a null-mutant background (supplementary material Fig. S4).

In case of Cal1-EGFP, specific signals were not only detected at the centromere as for Cenp-A/Cid-EGFP and Cenp-C-EGFP, but also in the nucleolus and weakly throughout the nucleus (supplementary material Fig. S4). Based on EGFP signal quantification, the centromeric, nucleolar and residual nuclear pools were estimated to comprise on average about 3.3%, 21% and 76%, respectively, of the total nuclear Cal1. Importantly, the amount of centromeric Cal1 was clearly far lower than that of Cenp-A/Cid and Cenp-C (Table 1). Our results therefore exclude models for centromeric Cenp-C localization where every Cenp-C monomer (or dimer) is stably linked via a single Cal1 protein to one or two copies of Cenp-A/Cid. The results of a comparison of the expression levels of the different EGFP fusion proteins (Cenp-A/Cid, Cenp-C, Cal1) by immunoblotting (supplementary material Fig. S5) was entirely consistent with this conclusion when taking into account the differential distribution of Cenp-A/Cid, Cenp-C and Cal1 into subnuclear regions (centromere, nucleolus, and elsewhere in the nucleus) as suggested by the quantitative in vivo imaging.

Quantitative imaging with imaginal discs was also used for a comparison of the amounts of the centromere proteins Cal1, Cenp-A/Cid and Cenp-C with those of the kinetochore proteins Spc105, Mis12, Spc25 and Nuf2 (Fig. 4D; Table 1). The levels of these kinetochore proteins were found to be all very similar and somewhat lower than the amounts of Cenp-A/Cid. The similar abundance measured for Spc25 and Nuf2 agrees with the established fact that they are stoichiometric components of the stable heterotetrameric Ndc80 complex (Santaguida and Musacchio, 2009). The comparison of the estimated numbers of protein copies per Drosophila kinetochore with those determined in yeast (Joglekar et al., 2006; Joglekar et al., 2008) indicated that the amounts of centromere and kinetochore proteins correlate rather with the number of kinetochore microtubules (1 in budding yeast, about 11 in Drosophila) (Winey et al., 1995; Maiato et al., 2006) than with the amount of centromeric DNA (125 bp in budding yeast, 420 kb in Drosophila) (Fitzgerald-Hayes et al., 1982; Sun et al., 1997).

Cenp-A/Cid deposition needs to be carefully controlled because the CenH3 variants of the CENP-A family have a crucial role in defining the epigenetic mark that specifies centromere identity in regional centromeres (Allshire and Karpen, 2008). In principle, the interdependence of centromeric Cenp-A/Cid, Cenp-C and the low levels of Cal1 might provide robust control of centromeric Cenp-A/Cid amounts and could effectively protect cells against the consequences of accidental unbalanced Cenp-A/Cid excess. However, previous studies have demonstrated that overexpression of Cenp-A/Cid is sufficient to cause ectopic incorporation all along the chromosome and consequential mitotic defects (Van Hooser et al., 2001; Heun et al., 2006; Moreno-Moreno et al., 2006). The massive overexpression applied in these studies (70-fold) (Heun et al., 2006), which is rather unlikely to occur in physiological conditions, even accidentally, might have over-run negative regulation. Therefore, we applied more limited overexpression in Drosophila embryos (up to fourfold; supplementary material Fig. S6), to evaluate the role of Cenp-A/Cid, Cenp-C and Cal1 interdependency in centromere confinement of these proteins.

Overexpression was achieved with the prd-GAL4 driver, which directs UAS transgene expression in alternating segmental stripes within the embryonic epidermis (Fig. 5A). Overexpression starts during embryonic cell cycle 14. Embryos were fixed and analyzed 3 hours later, when the majority of the epidermal cells are in G2 of cycle 16. Intervening stripes that do not express prd-GAL4 were used as internal controls.

Interestingly, when UAS-Cenp-A/cid was expressed, we could detect at most a marginal increase in the intensity of centromeric anti-Cenp-A/Cid signals in the prd-GAL4 expressing stripes (Fig. 5B,C). However, when UAS-Cenp-A/cid and UAS-cal1-EGFP were simultaneously overexpressed, we observed a highly significant increase in centromeric Cenp-A/Cid (Fig. 5B,C; P<0.001, Student's t-test). UAS-cal1-EGFP without concomitant UAS-Cenp-A/cid expression did not result in increased centromeric anti-Cenp-A/Cid signals (Fig. 5B,C). Quantification of centromeric Cal1-EGFP fluorescence indicated that coexpression of UAS-Cenp-A/cid and UAS-cal1-EGFP resulted in slightly higher levels than when UAS-cal1-EGFP was expressed alone (Fig. 5B,C). We conclude that moderate overexpression of Cenp-A/Cid does not lead to increased centromeric Cenp-A/Cid levels because Cal1 levels are limiting. Similarly, Cenp-A levels limit centromeric Cal1 levels. These findings are consistent with the proposal that deposition of Cenp-A/Cid at the centromere requires complex formation with Cal1, probably by direct interaction as suggested by our Y2H experiments.

To analyze the interplay of Cenp-A/Cid and Cal1 with Cenp-C, we quantified centromeric anti-Cenp-C signals. As our Y3H experiments had indicated that Cal1 can form a bridge between Cenp-A and Cenp-C, these three proteins might be incorporated into the centromere as a stoichiometric stable complex. Accordingly, the increased Cenp-A/Cid and Cal1-EGFP incorporation observed after simultaneous overexpression is expected to be accompanied by a parallel increase in centromeric Cenp-C. However, we did not detect such an increase (Fig. 5C). This finding confirms that centromeres are not assembled by multimerization of stable persisting complexes of Cal1, Cenp-A/Cid and Cenp-C.

Expression of UAS-Cenp-C provided additional confirmation for the notion that centromeric accumulation of Cenp-A/Cid and Cenp-C are not necessarily coupled. Although UAS-Cenp-C expression clearly resulted in an increase of centromeric Cenp-C (Fig. 5C; P<0.001, Student's t-test), it was not paralleled with a comparable increase in centromeric Cenp-A (Fig. 5C). The increased centromeric anti-Cenp-C signals observed after UAS-Cenp-C expression suggest that the Cenp-C binding sites within the centromere are not saturated at the endogenous Cenp-C expression level. However, because UAS-Cenp-C expression caused increased anti-Cenp-C signals not only at the centromere, but also throughout the cell (data not shown), the centromeric Cenp-C binding sites appear to become limiting when Cenp-C is overexpressed.

Although our findings indicated that centromeric accumulation of Cenp-C is not necessarily coupled to that of Cal1-Cenp-A/Cid, simultaneous overexpression of all three centromere proteins clearly revealed synergism. In this case, maximal centromeric signals were obtained. Signals were significantly higher than after overexpression of UAS-Cenp-A/cid, UAS-Cenp-C and UAS-cal1-EGFP individually or in pairs (Fig. 5B; P<0.01 for all comparisons, Student's t-test). These findings are consistent with the suggestion that Cal1-mediated transient interactions between Cenp-A/Cid and Cenp-C support their centromeric deposition.

The observed increase in the centromeric levels of Cenp-A/Cid, Cenp-C and Cal1-EGFP after simultaneous prd-GAL4-driven overexpression did not appear to result in severe mitotic defects. Abnormal mitotic figures at the stage of mitosis 16 were rarely observed and chromosomal incorporation of Cenp-A/Cid, Cenp-C and Cal1-EGFP outside the centromere was not detected. However, clear mitotic defects resulted (Fig. 5D) when we used maternal α4tub-GAL4-VP16, which drives almost twofold higher expression than prd-GAL4 (data not shown). The strongest defects were caused by simultaneous expression of all three centromere proteins. Milder defects were already apparent after combined expression of UAS-Cenp-A/cid and UAS-cal1-EGFP (Fig. 5D). By contrast, all other combinations or individual expression of the UAS transgenes did not result in a distinct enrichment of abnormal mitotic figures (Fig. 5D). Expression of UAS transgenes during eye and wing development further confirmed that the combined overexpression of the three centromere proteins is far more deleterious than individual overexpression (supplementary material Fig. S7).

To address how α4tub-GAL4-VP16-driven simultaneous expression of UAS-Cenp-A/cid, UAS-Cenp-C and UAS-cal1-EGFP affects progression through mitosis, we characterized the mitotic abnormalities in further detail. The most prominent defects observed in fixed embryos were abnormal anaphase and telophase figures with chromatin bridges containing lagging centromeres (Fig. 5E). Ectopic Cenp-A/Cid incorporation throughout the chromosome arm regions was rarely detectable in these abnormal mitotic figures. Focal ectopic Cenp-A/Cid incorporation within a chromosome arm might in principle lead to multicentric chromosomes and thereby explain the observed chromosome bridges with lagging centromeres after simultaneous overexpression of Cenp-A/Cid, Cenp-C and Cal1-EGFP. Therefore, we counted the number of kinetochores in mitotic cells. Even in wild-type controls, we were unable to detect all of the 16 centromeres as distinct Cenp-A/Cid or Cenp-C foci in every mitotic cell. Apart from the occasional immediate proximity of kinetochores, accessibility problems resulting in low anti-Cenp-A/Cid or anti-Cenp-C signals specifically during prometaphase and metaphase further impaired kinetochore identification. Therefore, we used EGFP-Nuf2-expressing embryos for kinetochore counting. Moreover, we determined kinetochore counts after expression of prd-GAL4-directed UAS transgenes in adjacent control and overexpressing regions to eliminate effects of fixation variability. Analyses after prd-GAL4-driven overexpression were possible because mitotic abnormalities were frequent at the stage of mitosis 16 when UAS-cal1 was used instead of the UAS-cal1-EGFP transgene insertion selected for the initial experiments (Fig. 5A–C). The stronger effect of UAS-cal1 presumably reflects transgene position effects on expression levels or absence of the EGFP tag which might be slightly deleterious. Despite the occurrence of late-mitotic figures with lagging centromeres within the overexpressing regions, the number of discrete kinetochore spots was not significantly increased within these regions (10.3±1.6 EGFP-Nuf2 spots compared with 10.7±1.6 spots in the intervening stripes; n>50 cells from more than 10 different embryos). These results suggest that ectopic kinetochores are not the primary cause for the mitotic abnormalities resulting from co-overexpression of Cal1, Cenp-A/Cid and Cenp-C.

As ectopic kinetochores could not be observed, we determined whether increased centromeric Cal1, Cenp-A/Cid and Cenp-C was accompanied with increased levels of kinetochore proteins. Transgenes expressing EGFP fusions of a given kinetochore protein (Nuf2, Spc25, Mis12 or Spc105) under the control of their normal cis-regulatory regions were used in combination with prd-GAL4-driven simultaneous overexpression of the three centromere proteins Cal1, Cenp-A/Cid and Cenp-C in stripes. The EGFP signals in the kinetochores of mitotic cells were found to be slightly but consistently enhanced within the overexpressing stripes (Fig. 5F). This enhancement was less extensive than that of Cal1, Cenp-A/Cid and Cenp-C (compare Fig. 5C and F). However, in these experiments the kinetochore proteins were not overexpressed from UAS transgenes in contrast to Cal1, Cenp-A/Cid and Cenp-C. The comparatively mild increase of kinetochore proteins observed after co-overexpression of Cal1, Cenp-A/Cid and Cenp-C might therefore reflect limiting kinetochore protein expression levels. Interestingly, when only Cal1 and Cenp-A/Cid but not Cenp-C was overexpressed, we were unable to detect a statistically significant increase in kinetochore protein levels (Fig. 5F), suggesting that the observed mitotic defects (Fig. 5D) do not depend on increased kinetochore protein levels.

To analyze the consequences of simultaneous overexpression of Cal1, Cenp-A/Cid and Cenp-C on the dynamics of progression through mitosis, we performed in vivo imaging with embryos expressing histone H2Av-mRFP and Cenp-A/Cid-EGFP in addition to UAS-cal1, UAS-Cenp-A/cid and UAS-Cenp-C (Fig. 6; supplementary material Movies 1 and 2). First abnormalities were already apparent during mitosis 15. Compared with controls, which did not overexpress the centromeric proteins, chromosome congression into a metaphase plate was always slower (2- to 8-fold; mean 3.7-fold; n=8 cells from two embryos) and metaphase prolonged (3- to 18-fold; mean 10-fold; n=10 cells from two embryos) except in one cell. Although chromosome segregation during anaphase appeared to be normal in about half of the cases (n=11 cells from two embryos), the other half displayed subtle to strong abnormalities. Characteristically, these abnormalities consisted in lagging centromeres (Fig. 6A, data not shown). During mitosis 16, these same mitotic defects were even more pronounced (Fig. 6A). The distances between sister kinetochores in metaphase plates were found to be scattered over a wider range after co-overexpression of Cal1, Cenp-A/Cid and Cenp-C (Fig. 6B). Collectively, our analyses of the observed mitotic abnormalities suggest that increased levels of centromeric Cal1, Cenp/Cid and Cenp-C compromise kinetochore function during mitosis.

Normally, centromere loading of Cenp-A/Cid and Cenp-C occurs during and depends on exit from mitosis (Jansen et al., 2007; Schuh et al., 2007). To evaluate whether the observed increase in centromere protein levels that results from simultaneous overexpression of Cal1, Cenp-A/Cid and Cenp-C also depends on progression through mitosis, we performed experiments in string(stg)/cdc25 mutant embryos where cells remain arrested in G2 phase of cycle 14 (Edgar and O'Farrell, 1989). After overexpression of Cal1, Cenp-A/Cid and Cenp-C in these G2-arrested cells, we did not observe increased anti-Cenp-A/Cid and Cenp-C signals at centromeres (supplementary material Fig. S8). However, some ectopic accumulation throughout the cells was apparent. By contrast, an increased centromeric signal, at the expense of distributed signals was clearly obtained after progression through a successful mitosis, which was triggered with the help of a heat-inducible hs-stg transgene in stg mutant embryos overexpressing Cal1, Cenp-A/Cid and Cenp-C. We conclude that increased incorporation of Cal1, Cenp-A/Cid and Cenp-C into centromeres requires both overexpression and progression through mitosis. Moreover, the excess levels of Cal1, Cenp-A/Cid and Cenp-C that were not yet incorporated into the centromere did not lead to mitotic defects during the hs-stg-induced mitosis.

Drosophila Cal1 has been identified recently because its knockdown in cultured cells results in a loss of Cenp-A/Cid and Cenp-C from centromeres and a failure of chromosome alignment and segregation during mitosis (Goshima et al., 2007; Erhardt et al., 2008). Here, we demonstrate that Cal1 is a crucial component of the important regulatory mechanisms that prevent an excessive incorporation of Cenp-A/Cid and Cenp-C into centromeres and consequential chromosome mis-segregation.

cal1 is an essential gene that is expressed specifically in mitotically proliferating cells. To provide its function, the protein product needs its N-terminal domain, which interacts with Cenp-A/Cid, as well as its C-terminal domain, which interacts with Cenp-C. By contrast, the most rapidly diverging middle region of Cal1 seems to be of lesser importance because expression of the N-C version, which lacks the M domain, is sufficient to prevent the characteristic defects in cal1 mutant embryos. The obvious functionality of the N-C version also emphasizes the importance of the centromeric localization of Cal1. The complete Cal1 protein is observed not only at the centromere, but also in the nucleolus. The M region is both sufficient and required for nucleolar localization. However, because this M domain is not required for cal1 mutant rescue, the significance of the nucleolar Cal1 localization remains unclear.

Rescue of cal1 mutants is not observed when the N- and C-terminal domains of Cal1 are expressed without a covalent linkage. The ability to recruit Cenp-A/Cid and Cenp-C into a complex, as clearly evidenced by our yeast three-hybrid experiments, is therefore likely to be crucial for Cal1 function. Co-immunoprecipitation of Cal1, Cenp-A/Cid and Cenp-C has previously indicated that these components can associate in vivo (Erhardt et al., 2008). However, our quantification of protein levels, which is largely dependent on the accuracy of our EGFP signal quantifications, demonstrates that Cenp-C is not exclusively anchored to centromeric chromatin via persistent and stoichiometric Cal1-mediated links to Cenp-A/Cid. Centromeric Cal1 levels are more than 40-times lower than those of Cenp-A/Cid and Cenp-C.

The centromeric amount of Cal1 is also far lower than that of the other kinetochore components that we have quantified (Spc105, Spc25, Nuf2). Interestingly, per kinetochore, the copy numbers of these components appear to be scaling well with the number of kinetochore microtubules (kMTs) when comparing our results from Drosophila with those described for budding and fission yeast (Joglekar et al., 2006; Joglekar et al., 2008). Spc25 and Nuf2 are constituents of the heterotetrameric Ndc80 complex, which binds directly to kMTs (Santaguida and Musacchio, 2009). Eight copies of the Ndc80 complex are thought to bind a single kMT to the budding yeast kinetochore (Joglekar et al., 2006). In Drosophila, where the number of kMTs per kinetochore appears to be around 11 (Maiato et al., 2006), about seven copies appear to be present per kMT according to our quantification. Our quantification of kinetochore proteins fits very well with the notion that the kinetochores of higher eukaryotes might be composed of several copies of a module that is present in one copy in budding yeast. By contrast, the centromere proteins Cenp-A and Cenp-C are scaling less well with the number of kMTs. The increased complexity of lateral co-ordination within animal kinetochores and of epigenetic specification of centromere identity might explain the higher relative amount of centromere proteins apparent in Drosophila. Despite this relative increase, centromeric Cenp-A/Cid allows packaging of only about 5% of the centromeric DNA in Drosophila under the assumption that Cenp-A/Cid nucleosomes wrap about 200 bp of a 200 kb centromere (Sun et al., 1997; Allshire and Karpen, 2008).

Although our quantifications exclude the notion that Cal1 functions as a stable stoichiometric linker of Cenp-A/Cid and Cenp-C in mitotic kinetochores, our overexpression experiments provide further support for a role as a centromere protein-loading factor (Erhardt et al., 2008). Moreover, our experiments reveal additional layers of regulation that prevent excess incorporation of centromere proteins within the centromeric region. They also indicate that such excess incorporation is highly detrimental to kinetochore function. Previous work in Drosophila has demonstrated that strong overexpression of Cenp-A/Cid (about 70-fold) can lead to ectopic kinetochore formation (Heun et al., 2006). However, almost all Cenp-A/Cid that is incorporated ectopically within the chromosome arm regions is degraded rapidly (Moreno-Moreno et al., 2006), which is also observed in yeast (Collins et al., 2004). Here, we show that the limiting amounts of Cal1 provide additional, highly efficient protection against excessive chromosomal incorporation of Cenp-A/Cid. After bypassing this protection by Cal1 overexpression, even low levels of Cenp-A/Cid overexpression (about 2.5-fold) result in increased incorporation into centromeres (about 1.6-fold). When, in addition to Cal1 and Cenp-A/Cid, Cenp-C is also mildly co-overexpressed (about 3.5-fold), the levels of centromeric Cenp-A/Cid are further increased (about 2-fold) along with those of Cal1 and Cenp-C. Importantly, co-overexpression of these centromere proteins resulted not only in increased centromeric levels, but also in severe mitotic defects.

Although other interpretations are not excluded, our findings strongly suggest that the mitotic defects observed after overexpression of Cal1 and Cenp-A/Cid, and even more strongly when Cenp-C was also overexpressed, reflect the consequence of the increase in the centromeric levels of these proteins. The increase in centromeric levels of centromere proteins was accompanied by a significant increase in kinetochore proteins (Spc105 and the Mis12 and Ndc80 complex) but only to a very limited extent and only when all three centromere proteins were co-expressed. The increased amounts of centromeric Cenp-A/Cid observed after co-expression of Cal1 and Cenp-A/Cid, which were not accompanied by a statistically significant increase in kinetochore protein levels, might therefore be sufficient to disturb the spatial organization of the kinetochore, leading to inefficient chromosome congression, spindle checkpoint hyperactivation and chromosome segregation defects in anaphase.

Our experiments in stg mutant embryos, demonstrate that co-overexpression of centromeric proteins during interphase is not sufficient to cause excess centromeric incorporation, consistent with the previously demonstrated dependence of centromeric deposition of Cenp-A/Cid and Cenp-C on exit from mitosis (Schuh et al., 2007). Indeed, forcing progression through mitosis (by hs-stg induction) was observed to be sufficient to cause centromeric deposition of the overexpressed proteins. Moreover, the fact that the excess centromere proteins that were not yet incorporated into the centromere did not disturb the hs-stg induced mitosis, further supports our suggestion that the mitotic defects observed after co-expression of centromeric proteins depend on excessive incorporation into the centromere.

The severe mitotic defects observed after co-overexpression of Cal1, Cenp-A/Cid and Cenp-C emphasize the importance of careful control of centromere protein deposition. Several levels of control are effective. The interdependence of Cal1, Cenp-A/Cid and Cenp-C functions in conjunction with cell cycle control to prevent detrimental excessive centromeric incorporation. The cell cycle regulators cyclin A, Rca1/Emi1 and Fzr/Cdh1 have recently been implicated in the control of deposition of Cenp-A/Cid and Cenp-C at the centromere (Erhardt et al., 2008). How these and possibly additional cell cycle regulators control centromere protein deposition has yet to be clarified.

A possible scenario for centromere protein deposition in Drosophila might include a release of nucleolar Cal1 at the onset of mitosis, followed by conversion into a form that associates with non-centromeric soluble Cenp-A/Cid during exit from mitosis. After binding of soluble Cenp-A/Cid to the N-terminal domain of Cal1, its C-terminal domain might become exposed so that it can bind to centromeric Cenp-C and promote Cenp-A/Cid transfer onto the neighboring centromeric chromatin and thereby indirectly also additional Cenp-C deposition.

The mechanisms and the extent of control of centromeric Cenp-A deposition appear to have evolved. In fission yeast, overexpression of Cenp-A/Cid alone is sufficient to obtain excess centromeric Cenp-A/Cnp1, and this excess does not result in increased kinetochore protein levels (Joglekar et al., 2008). Spreading of Cenp-A within centromeric chromatin has also been clearly demonstrated in human cells after mild overexpression of Cenp-A (Lam et al., 2006). Mitotic defects were not detected in this case, perhaps because of the very limited increase in centromeric Cenp-A. Cal1 homologs from non-Drosophilid genomes have not yet been identified so far. Conversely, with the exception of Cenp-C, homologs of the 15 components of the vertebrate centromere chromatin-associated network (CCAN), which is related to the yeast Ctf19 and Sim4 complexes, have not been revealed in Drosophilid genomes, neither by thorough bioinformatic analyses (Meraldi et al., 2006) nor by genome-wide RNAi screens (Goshima et al., 2007; Erhardt et al., 2008). The CCAN seems also to be absent in C. elegans (Cheeseman et al., 2004; Sonnichsen et al., 2005; Gassmann et al., 2008). It is conceivable therefore that Cal1 is a functional analog of the CCAN, which has also been implicated in Cenp-A loading (Okada et al., 2006). However, because the evolutionary sequence conservation of centromere and kinetochore components is generally very low, it remains a possibility that Cal1 homologs also exist and function in centromere loading of human Cenp-A and Cenp-C.

cal1^c03646 (Thibault et al., 2004), cal1^MB04866 (Metaxakis et al., 2005) and Df(3R)Exel6176 (Parks et al., 2004) were obtained from the Bloomington Drosophila Stock Center. gEGFP-cal1 lines were generated using ΦC31-mediated germline transformation (Bischof et al., 2007) and UAS-cal1, UAS-cal1-EGFP, UAS-cal1(N)-EGFP, UAS-cal1(M)-EGFP, UAS-cal1(C)-EGFP, UAS-cal1(N-C)-EGFP and gcal1-EGFP with pP{CaSpeR-4} constructs (details provided upon request).

The wing imaginal discs analyzed for the quantification of centromere and kinetochore proteins fused to EGFP (Schuh et al., 2007; Schittenhelm et al., 2007; Schittenhelm et al., 2009) were from larvae with the following genotypes:

For the analyses in stg mutant embryos, we crossed w*; P{w⁺, UAS-cal1-EGFP} II.1/CyO, P{ry⁺, ftz-lacZ}; P{w⁺, UAS-Cenp-A/cid} III.5, P{w⁺, UAS-Cenp-C} III.1, stg^7B/TM3, Sb, P{w⁺, Ubx-lacZ} males with either w*; stg^7B, e, P{w⁺, da-GAL4} G32/TM3, Sb, P{w⁺, Ubx-lacZ} or w*; stg^7B, e, P{w⁺, hs-stg} 3.1, P{w⁺, da-GAL4} G32/TM3, Sb, P{w⁺, Ubx-lacZ} females.

Protein–protein interactions were analyzed essentially as described (Jäger et al., 2004). For Y3H analyses, the yeast strain MaV203 (Invitrogen) was cotransformed with a pBridge-cal1 and a pGADT7 construct and plated on SD-Leu-Trp selective drop-out medium. Colonies were transferred to appropriate selective drop-out medium plates (SD-Leu-Trp-Ura and SD-Leu-Trp-His) with or without methionine. cal1 expression from the pBridge construct is controlled by the P_Met25 promoter and occurs only in the absence of methionine.

Transfection of S2R+ cells was conducted with the FuGeneHD Transfection Reagent (Roche) essentially as described (Schittenhelm et al., 2007). Immunofluorescence and DNA labeling of S2R+ cells and fixed embryos was also done essentially as described (Pandey et al., 2005; Schittenhelm et al., 2007). Rabbit antibodies against EGFP (1:3000), Cenp-A/Cid (Jäger et al., 2005), Cenp-C (Heeger et al., 2005) and Spc105 (Schittenhelm et al., 2009), as well as mouse anti-α-tubulin (DM1A, 1:50,000, Sigma) and anti-lamin Dm0 (ADL67.10, 1:200) were used for immunoblotting.

For quantification of centromeric anti-Cenp-A/Cid and anti-Cenp-C signal intensities (Fig. 5B,C), we crossed prd-GAL4/TM3, Sb, P{w⁺, Ubx-lacZ} females to males carrying various UAS transgenes individually or in combination (UAS-Cenp-A/cid III.5, UAS-Cenp-C III.1 and UAS-cal1-EGFP II.1). Embryos were collected for 2 hours and aged for 5 hours at 25°C before fixation and immunolabeling with rabbit anti-Cenp-A/Cid or anti-Cenp-C, mouse anti-β-galactosidase (for genotype determination) and Hoechst 33258 (DNA stain).

For quantification of kinetochore protein levels (Fig. 5F), we crossed females carrying prd-GAL4 recombined with a transgene driving expression of a kinetochore protein fused to EGFP under the control of the endogenous regulatory region (gMis12-EGFP III.1, gSpc105-EGFP III.1, gEGFP-Nuf2 III.1 or gSpc25-EGFP III.1) over TM3, Sb, P{w⁺, Ubx-lacZ} to males carrying various UAS-transgenes individually or in combination (UAS-Cenp-A/cid III.5, UAS-Cenp-C III.1 and UAS-cal1 II.1).

Quantification of signal intensities in embryos with prd-GAL4 expressing and non-expressing regions was performed after acquisition of stacks with a 63×/1.4 oil-immersion objective and 250 nm spacing from the epidermal region of the second and third thoracic and the first abdominal segment. Within this imaged region, prd-GAL4 drives expression in the outer but not in the middle segment. A Colibri light source (Zeiss) with a 470 nm light emitting diode was used for EGFP excitation with reproducible and temporally stable intensity. The stacks were deconvolved (Huygens Remote Manager v1.0 beta 2; Montpellier RIO Imaging) and subsequently converted into maximum projections using ImageJ.

For quantification of anti-Cenp-A/Cid and anti-Cenp-C signals, stacks with 12 sections were acquired from six different embryos for each genotype. A rectangle from the middle region that does not express prd-GAL4 was first selected. Subsequently, the average intensity of the centromeric pixels within the selected rectangle was determined after applying a threshold to eliminate non-centromeric signals. Moreover, the average pixel intensity of non-centromeric pixels was determined and defined as background within the selected rectangle. Subtraction of this background from the intensity of centromeric pixels resulted in our measure of centromeric signal intensity within the middle internal control region that does not express prd-GAL4. Thereafter, rectangles from the flanking regions that express prd-GAL4 were selected, followed again by thresholding to select centromeric pixels. To arrive at our measure of centromeric signal intensities within the prd-GAL4-expressing regions, we subtracted the background determined in the middle internal control region. By subtracting the background determined in the middle internal control region from the intensity of the non-centromeric pixels within the prd-GAL4-expressing regions, we arrived at a measure for the non-centromeric excess of the overexpressed centromere protein.

For Cal1-EGFP signal quantification, we were unable to use the intervening middle region as an internal control because UAS-cal1-EGFP was only expressed within the prd-GAL4-expressing regions, Therefore, we determined centromeric GFP signal intensities by applying a threshold to select the centromeric pixels within the prd-GAL4-expressing regions followed by subtraction of the background, which was obtained by averaging signal intensities of the non-centromeric pixels. The values obtained for all six embryos of a given genotype were averaged. The average obtained with embryos expressing only UAS-cal1-EGFP was set to 100% to arrive at the bars presented in Fig. 5C.

Quantification with or without prior deconvolution resulted in identical ratios of centromeric signal intensities between prd-GAL4-expressing and non-expressing regions in case of the anti-Cenp-A/Cid staining. In case of anti-Cenp-C and Cal1-EGFP, higher non-centromeric signals precluded a reliable, exclusive segmentation of centromeric signals without prior deconvolution. However, in these cases, quantification of individually selected centromeres using the two-square method with local background correction (see below) also resulted in very similar results, irrespective of prior deconvolution.

For quantification of EGFP-tagged kinetochore components, we acquired stacks with 12 sections from at least seven different embryos. Individual prometaphase or metaphase cells in the maximum projections were selected consecutively by two concentric squares (side length, 50 pixel and 55 pixels, respectively). The total pixel intensity of each square was determined and the average pixel intensity within the region encircled by the larger, but not by the smaller square was determined as local background. The average background pixel intensity integrated over the smaller square was subtracted from the total pixel intensity within the smaller square to yield the kinetochore signal intensity of a cell.

Embryos obtained from a cross of α4tub-GAL4-VP16, gHis2AvD-mRFP II.2, gcid-EGFP-cid II.1 / CyO, P{ry⁺, ftz-lacZ} females with UAS-cal1 II.1; UAS-Cenp-A/cid III.5, UAS-Cenp-C III.1 males were analyzed by in vivo imaging essentially as described (Pandey et al., 2005) at the stage when epidermal cells progress through the fifteenth (4–5 hours) or sixteenth (6.5–7.5 hours) round of mitosis. Time-lapse imaging was performed with an Olympus FV1000 system. Stacks (four sections, 250 nm spacing) were acquired at intervals of 20 seconds using a 60× oil-immersion objective and converted to maximum projections. Embryos from α4tub-GAL4-VP16, gHis2AvD-mRFP II.2, gcid-EGFP-cid II.1 / CyO, P{ry⁺, ftz-lacZ} females crossed against w¹ males were analyzed for control.

For the comparison of the levels of Drosophila centromere and kinetochore proteins fused to EGFP with those of Cse4-EGFP in yeast, we dissected wing imaginal discs from third instar wandering stage larvae in Schneider's Drosophila medium (Invitrogen). The imaginal discs were mounted in phosphate-buffered saline (PBS) on a coverslip previously coated with yeast cells of strain KBY7006 (S. cerevisiae 473a CSE4-GFP:KAN; (Joglekar et al., 2006) kindly provided by Kerry Bloom (University of North Carolina, Chapel Hill, NC). Yeast cells from a fresh overnight culture grown in YPD at 25°C were resuspended in PBS after sedimentation and a wash in H₂O. The suspension was spread on a coverslip coated with concanavalin A for about 5 minutes. Immediately before mounting freshly dissected wing imaginal discs, a region in the center of the coverslip was wiped dry. Imaginal discs were mounted in this region with their peripodial membranes facing the cover slip. Stacks (20–27 sections, 250 nm spacing) were acquired using a 63×/1.4 oil immersion objective and a Zeiss Cell Observer HS. The stacks were converted into maximum projections using ImageJ. For Cse4-EGFP signal quantification, individual centromere clusters of anaphase or telophase cells were selected by two concentric squares (side length, 20 and 22 pixels, respectively) and centromeric signal intensity was determined after background subtraction as described above for kinetochore EGFP fusion proteins. For quantification of EGFP-tagged centromere and kinetochore components in imaginal wing discs, at least 40 individual cells from more than three wing discs per genotype were also selected by two concentric squares (side length, 50 and 55 pixels, respectively) followed by determination of centromeric signal intensity after background subtraction as described above.

We would like to thank Romanas Chaleckis for assistance during the Y2H experiments, Sebastian Heeger for initial characterization of Cenp-A/Cid and Cenp-C overexpression, Alf Herzig for the stg^7B, hs-stg, da-GAL4 stock, Johannes Bischof and Konrad Basler for sharing the unpublished cloning vector pattB, as well as Brigitte Jaunich and Sina Niebur for technical help. This work was supported by grants of the Swiss National Science Foundation (SNF 3100A0-120276/1) and the Deutsche Forschungsgemeinschaft (DFG He 2354/2-4).