Bub1 is required for kinetochore localization of BubR1, Cenp-E, Cenp-F and Mad2, and chromosome congression

Chromosome segregation in eukaryotes is mediated by a microtubule spindle apparatus. In addition to the spindle, kinetochores are essential for successful chromosome segregation. Kinetochores are large complex protein structures that assemble at the centromeric regions of each sister chromatid and perform three key functions (Nicklas, 1997; Rieder and Salmon, 1998). First, kinetochores attach chromosomes to the spindle. Second, kinetochores co-ordinate microtubule dynamics to allow chromosomes to move along the spindle. Third, kinetochores generate the `wait' signal that prevents anaphase onset until all the chromosomes are correctly aligned on the spindle. This signal forms part of the spindle checkpoint mechanism, a highly conserved cell cycle checkpoint that ensures accurate chromosome segregation (Jallepalli and Lengauer, 2001; Musacchio and Hardwick, 2002).

Electron microscopy studies show that vertebrate kinetochores are trilaminar structures that sit back-to-back on top of the chromatin (Biggins and Walczak, 2003; Cleveland et al., 2003). The inner kinetochore plate, directly adjacent to the centromeric heterochromatin, is separated from an outer plate by a middle layer. Microtubules embed into the outer kinetochore, beyond which extends a fibrous corona. This trilaminar structure is not visible during interphase. Rather, an amorphous ball-like structure called the pre-kinetochore lies adjacent to the centromeric heterochromatin (Rieder, 1982). This suggests that kinetochores undergo an assembly process or morphogenesis upon entry into mitosis, maturing from the pre-kinetochore to a trilaminar structure.

Light microscopy studies are consistent with the notion that kinetochores undergo a maturation process upon entry into mitosis. Whereas a number of proteins, including Cenp-A, Cenp-C and Cenp-I, localize to the centromere region throughout the cell cycle (Liu et al., 2003; Palmer et al., 1987; Tomkiel et al., 1994), many other proteins only localize to kinetochores transiently during mitosis. These include motor proteins such as cytoplasmic dynein (Echeverri et al., 1996) and Cenp-E (Yen et al., 1992), and other proteins such as Cenp-F (Liao et al., 1995), ZW10, ROD (Chan et al., 2000) and Hec1 (Martin-Lluesma et al., 2002). The spindle checkpoint proteins Bub1, Bub3, Mad1, Mad2, Mps1 and a Mad3/Bub1-related protein kinase called BubR1, also only localize to kinetochores during mitosis (Musacchio and Hardwick, 2002; Shah and Cleveland, 2000). Consistent with a role in monitoring chromosome alignment, the levels of these latter proteins, including Mad2, Bub1 and BubR1, diminishes following microtubule capture and/or bi-orientation (Chan et al., 1998; Chen et al., 1996; Taylor and McKeon, 1997). Another group of proteins that localize transiently during mitosis includes INCENP, Aurora B and Survivin (Adams et al., 2001a; Bischoff and Plowman, 1999). At the onset of anaphase, these proteins relocate from the chromosomes to the spindle and are hence termed chromosome passenger proteins (Earnshaw and Bernat, 1991). Although these proteins do not localize to kinetochores – in vertebrates, they localize to the inner centromere region – they do appear to play a role in kinetochore assembly. Specifically, inhibition of Aurora B or Survivin inhibits recruitment of several kinetochore proteins including BubR1, Cenp-E and Mad2 (Carvalho et al., 2003; Ditchfield et al., 2003; Hauf et al., 2003; Lens et al., 2003).

Proteins that associate transiently with kinetochores in mitosis are not recruited simultaneously in human cells. Rather, there appears to be a defined order of assembly. Specifically, two independent co-staining studies show that Bub1 is recruited to kinetochores very early in prophase, followed by Cenp-F and then BubR1, and finally with Cenp-E being recruited in mid- to late prometaphase (Jablonski et al., 1998; Taylor et al., 2001). One model to explain this order of assembly is that recruitment of the latter proteins is dependent on the prior recruitment of the early ones. Thus, as with the assembly of bacteriophage capsids (Casjens and King, 1974), perhaps proteins recruited early create binding sites that facilitate the sequential binding of others. Consistent with this notion, Bub1 binds Cenp-F in a two-hybrid assay (Chan et al., 1998), BubR1 co-precipitates Cenp-E (Chan et al., 1998; Yao et al., 2000) and Bub1 co-precipitates BubR1 (Taylor et al., 2001). However, owing to the lack of suitable mammalian kinetochore assembly assays, whether this temporal order reflects an underlying dependency remains to be determined.

Xenopus egg extracts do however provide a tractable kinetochore assembly assay. Immunodepletion of Bub1 from egg extracts prevents kinetochore localization of BubR1, Mad2, Mad1 and Cenp-E (Sharp-Baker and Chen, 2001). In addition, immunodepletion of BubR1 prevents recruitment of Bub1, Mad2, Mad1 and Cenp-E (Chen, 2002). However, this observation seems to be at odds with the order observed in mammalian cells: because Bub1 binds kinetochores before BubR1 in mammalian cells, one might predict that recruitment of Bub1 is not dependent on BubR1. One possible explanation for this difference is that, in order to facilitate many rapid cell divisions, the Xenopus embryo stockpiles pre-assembled kinetochore complexes. Indeed, it is not clear whether there is a temporal order of recruitment upon entry into mitosis in the Xenopus system or whether all the transient kinetochore proteins are recruited simultaneously. Therefore, it is at present unclear whether the observations derived from the Xenopus in vitro system are applicable to mammalian somatic cells.

To dissect the kinetochore assembly process in mammalian somatic cells, we have used RNA interference (RNAi) (Elbashir et al., 2001) to repress individual proteins selectively in human cells and then determined the effect on the localization of other kinetochore proteins using quantitative optical sectioning microscopy (Taylor et al., 2001). In contrast to the data from Xenopus, our observations are consistent with the notion that the order of assembly does indeed reflect an underlying order of dependency, possibly mediated by direct protein-protein interactions. Surprisingly, our observations also show that, although repression of Bub1 delays chromosome congression, it does not appear to inhibit spindle checkpoint function.

DLD-1 and TA-HeLa cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% foetal calf serum, 2 mM glutamine, penicillin and streptomycin at 100 U ml^–1 and 100 μg ml^–1, respectively (all from Gibco BRL) and grown in a humidified 5% CO₂ incubator, as described previously (Taylor et al., 2001). Nocodazole [Sigma; 5 mg ml^–1 in dimethyl sulfoxide(DMSO)] was used at a final concentration of 0.2 μg ml^–1. MG132 (Calbiochem) was dissolved in DMSO and used at a final concentration of 20 μM.

To create a sheep polyclonal anti-Mad2 antibody and a DLD-1-derived cell line expressing a Myc-tagged Mad2 fusion protein, a human Mad2 cDNA was generated by reverse-transcription polymerase chain reaction (RT-PCR) amplification of HeLa cell mRNA using the SuperScript™ One-Step RT-PCR system (Invitrogen), subcloned and sequenced. For antibody production, the cDNA was cloned into pGEX-4T3 (Pharmacia). A glutathione-S-transferase/Mad2 fusion protein was then expressed, purified and used for immunization as described previously (Taylor et al., 2001). The anti-Mad2 antibody SM2.2 was then affinity purified following standard procedures.

The DLD-1 Myc-tagged Mad2 cell line was created using the Flp-In™ system (Invitrogen). Briefly, a Flp-In host cell line was created by integrating a single Flp recombination target (FRT) recombination site into the DLD-1 genome. The Mad2 cDNA was then cloned into a pcDNA5/FRT-based Myc-tagged vector and co-transfected into the DLD-1 FRT line along with a plasmid expressing the FLP recombinase (pOG44, Invitrogen) using LipofectAMINE™ Plus (Invitrogen). Cells were selected in 400 μg ml^–1 Hygromycin B (Roche) and colonies were pooled and expanded.

Immunofluorescence analysis was basically done as described (Taylor et al., 2001). With the exception of the anti-Mad2 antibody SM2.2 (see above), all other antibodies have been described previously (Ditchfield et al., 2003; Hussein and Taylor, 2002; Taylor et al., 2001; Tighe et al., 2001). Briefly, cells were fixed in 1% paraformaldehyde, washed in PBS plus 0.1% Triton X-100 (PBST), blocked in PBST plus 5% non-fat dried milk then stained with the following primary antibodies: ACA (human anti-centromere, 1:1000); SB1.3 (sheep anti-Bub1, 1:1000); 4B12 (mouse anti-Bub1, 1:10); SBR1.1 (sheep anti-BubR1, 1:1000); 5F9 (mouse anti-BubR1, 1:50); RCE.1, (rabbit anti-Cenp-E, 1:2000); SCF.1 (sheep anti-Cenp-F, 1:1000); AIM-1 (mouse anti-Aurora B, Transduction Laboratories, 1:200); TAT-1 (mouse anti-tubulin, 1:500); RAA.1 (rabbit anti-Aurora A, 1:10,000); rabbit anti-phospho-histone H3, (Upstate Biotechnology, 1:200). Following washes, cells were stained with appropriate Cy2-, Cy3- and Cy5-conjugated secondary antibodies (Jackson Immunoresearch Laboratories) all diluted 1:500, stained with Hoechst 33358 (Sigma) and then mounted. For analysis of microtubules, cells were fixed in ice-cold methanol for 20 minutes and then processed as described above.

Small interfering RNA (siRNA) duplexes (Dharmacon Research) designed to repress Bub1, BubR1, Cenp-E, Cenp-F, Aurora B and Survivin (Fig. 2A) were transfected using OligofectAMINE™ (Invitrogen) according to the manufacturer's instructions. In brief, 0.5×10⁵ cells were seeded in 24-well plates 24 hours before transfection in growth media without antibiotics. The siRNA duplexes at a final concentration of 240 nM and 3 μl OligofectAMINE were diluted in media without antibiotics, mixed and incubated for 20 minutes. The siRNA-lipid complexes were then added to cells for 4 hours followed by addition of complete medium containing 20% foetal calf serum. 24 hours later, the cells were replated and analysed 48-72 hours after transfection.

Immunoblotting was basically done as described (Taylor et al., 2001). Total cell lysates were prepared by solubilizing cell pellets in SDS sample buffer. Proteins were then resolved by SDS PAGE and electroblotted on Immobilon-P membranes (Millipore). Blots were blocked in TBST (50 mM Tris, pH 7.6, 150 mM NaCl, 0.1% Tween-20) plus 5% non-fat dried milk, then incubated overnight with SB1.1 (anti-Bub1, 1:2000), SBR1.3 (anti-BubR1, 1:2000), SAB.1 (anti-Aurora B, 1:1000), an anti-Survivin rabbit polyclonal (Novus Biologicals, 1:1000), SCF.1 (anti-Cenp-F, 1:1000), an anti-Cenp-E rabbit polyclonal (1:200) (kindly provided by D. Cleveland, University of California, San Diego, USA) or TAT-1 (1:1000) in TBST. After washing in TBST, bound primary antibodies were labelled with the appropriate horseradish-peroxidase-conjugated secondary antibodies (Zymed), all diluted at 1:2000 in TBST. After washing in TBST, bound secondary antibodies were detected using the SuperSignal chemiluminescence system (Pierce) and imaged on Biomax MR film (Kodak). The level of Bub1 protein repression was determined by densitometry analysis. AIDA ImageQuant software (Raytest) was used to quantify each band within a given blot with appropriate background subtraction. Bub1 protein levels after RNAi are shown as a percentage of control protein levels. The percentage given is representative of four independent RNAi experiments±s.e.m.

Standard fluorescence microscopy was carried out on a Zeiss Axiovert 200 equipped with epifluorescence using 100× objective and a Photometrics Cool Snap HQ CCD camera driven by Metamorph software (Universal Imaging). Deconvolution microscopy was performed using a wide-field optical sectioning microscope (Deltavision, Applied Precision) as previously described (Taylor et al., 2001). Briefly, for each cell, a z-series of 15 images at 0.2 μm intervals was captured at each wavelength and then processed using constrained iterative deconvolution. Deconvolved image stacks were projected and fluorescence signal intensities quantified using SoftWoRx (Applied Precision). To quantify the amount of kinetochore bound protein, the average pixel intensities from at least 30 kinetochore pairs from three or more nocodazole-treated cells were measured and background pixel intensities subtracted. The pixel intensities at each kinetochore pair were then normalized against ACA pixel values to account for any variations in staining or image acquisition. The values from cells that had been subjected to RNAi were then plotted as a percentage of the values obtained from cells transfected with a control siRNA duplex.

Flow cytometry was done as described (Ditchfield et al., 2003). Briefly, loosely attached cells were first collected and then pooled with attached cells removed by trypsinization. Following fixation in 70% ethanol, the cells were washed in PBS and incubated with a fluorescein-5-EX succinimidyl ester (FSE)-conjugated MPM2 monoclonal antibody (Upstate Biotechnology) diluted 1:500 in PBS for 1 hour at 4°C. The cells were then washed twice for 5 minutes in PBS, stained with propidium iodide (40 μg ml^–1 final concentration) and treated with RNase A (50 μg ml^–1) for 30 minutes at room temperature. 10,000 cells were then analysed on a Becton Dickinson FACScan and the number of mitotic MPM2 positive cells determined.

Bub1 localizes to kinetochores very early in prophase, followed by Cenp-F and then BubR1, with Cenp-E being recruited in mid- to late prometaphase (Jablonski et al., 1998; Taylor et al., 2001) (Fig. 1A). To determine when Mad2 is recruited to kinetochores in human cells, we generated two polyclonal anti-Mad2 antibodies. Although one, SM2.2, detected Mad2 by western blotting (Fig. 1B), neither detected Mad2 by immunofluorescence (not shown). We therefore created a stable cell line expressing a Myc-tagged Mad2 fusion protein under the control of a constitutive viral promoter. The Myc-Mad2 fusion was detectable on western blots using both anti-Mad2 and anti-Myc antibodies (Fig. 1B). Analysis of this cell line by immunofluorescence using anti-Myc antibodies demonstrated that Myc-Mad2 localized to the nuclear envelope during interphase (not shown), consistent with observations showing that Mad2 localizes to nuclear pores (Campbell et al., 2001). Myc-Mad2 was abundant at kinetochores during prometaphase but not metaphase (Fig. 1C-E), suggesting that Myc-Mad2 behaves like the endogenous protein (Chen et al., 1996; Li and Benezra, 1996; Waters et al., 1998). When exposed to nocodazole, the Myc-Mad2 line accumulated a similar number of mitotic cells to a control line (see Fig. S1), indicating that expression of the exogenous Myc-Mad2 fusion protein does not appear to have an adverse effect on the spindle checkpoint. In prophase and prometaphase cells, we frequently observed kinetochores that stained positive for Bub1, BubR1 and Cenp-E but not Myc-Mad2 [Fig. 1C-E, Table 1 and Fig. S2]. Furthermore, we never saw Myc-Mad2 at kinetochores that lacked Bub1, BubR1 or Cenp-E (Table 1). Thus, it appears that Mad2 is recruited to kinetochores in late prometaphase after Bub1, Cenp-F, BubR1 and Cenp-E (Fig. 1A).

The existence of a defined temporal order of association raises the possibility that there is a linear assembly pathway whereby the localization of latter components is dependent on the prior association of earlier proteins (Jablonski et al., 1998; Taylor et al., 2001). To test this, we designed siRNA duplexes (Elbashir et al., 2001) to repress Bub1, Cenp-F, BubR1 and Cenp-E (Fig. 2A) and then examined the ability of the unrepressed proteins to localize to kinetochores. Western blotting of total cell lysates following transfection of siRNA duplexes demonstrated that Bub1, Cenp-F, Cenp-E and Survivin could be efficiently repressed (Fig. 2B). Repression of BubR1 and Aurora B by RNAi has been described previously (Ditchfield et al., 2003). However, we have been unable effectively to repress Mad2 by transfection of siRNA duplexes. 48 hours following transfection, cells were treated with nocodazole for 1 hour to inhibit kinetochore-microtubule interactions, thus creating conditions under which all the checkpoint proteins should be enriched at kinetochores. After fixation, the relative abundances of Bub1, BubR1, Cenp-F, Cenp-E and Mad2 at kinetochores was determined by quantitative immunofluorescence microscopy as detailed in the Materials and Methods.

First, we examined the effect of repressing Bub1. In control cultures, BubR1, Cenp-F, Cenp-E and Mad2 all colocalized with Bub1 at kinetochores (Fig. 3A). However, in cells in which Bub1 had been repressed, the levels of BubR1, Cenp-F, Cenp-E and Mad2 present at kinetochores were greatly reduced. Quantitation of normalized pixel intensities shows that, when Bub1 was reduced to less than ∼5% of its control value, BubR1 was reduced to ∼30%, Mad2 to ∼25%, Cenp-E to ∼20% and Cenp-F to ∼10% (Fig. 3B). Thus, Bub1 appears to be required for efficient kinetochore localization of BubR1, Cenp-F, Cenp-E and Mad2.

Next, we examined the effect of repressing BubR1. In control cultures, Bub1, Cenp-F, Cenp-E and Mad2 all localized with BubR1 at prometaphase kinetochores (Fig. 4A). In cells in which BubR1 had been repressed, the levels of Bub1, Cenp-F and Mad2 present at kinetochores appeared largely unaffected. By contrast, the levels of kinetochore-bound Cenp-E appeared reduced. Quantitation of normalized pixel intensities shows that, when BubR1 was reduced to less than ∼10% of its control value, kinetochore-bound Bub1 and Cenp-F increased about 1.75 times (Fig. 4B). Whether these increased pixel counts reflects a true increase in kinetochore-bound protein rather than an increase in antigen accessibility remains to be determined. Although Mad2 levels were very similar to their control value, Cenp-E levels were reduced to ∼35%, indicating that BubR1 is required for efficient kinetochore localization of Cenp-E. However, in contrast to work in Xenopus (Chen, 2002), BubR1 is not required for kinetochore localization of Bub1, Mad2 or Cenp-F.

Next, we examined the effect of repressing Cenp-E. In control cultures, Bub1, BubR1, Mad2 and Cenp-F were readily detectable at Cenp-E-positive kinetochores (Fig. 5A). In cells in which Cenp-E had been repressed, the levels of Bub1, BubR1 and Cenp-F detectable at kinetochores appeared largely unaffected. By contrast, the levels of Mad2 appeared to be lower. Quantitation of normalized pixel intensities shows that, when Cenp-E was reduced to less than ∼10% of its control value, kinetochore-bound Bub1 and BubR1 increased by about 1.5 times (Fig. 5B). Again, whether these increased values are due to a true increase rather than to changes in antigen accessibility is unclear. Although Cenp-F levels were slightly reduced relative to their control value, Mad2 levels were reduced to less than 40%, suggesting that Cenp-E is required for efficient kinetochore localization of Mad2. In this instance, our observations are consistent with those from Xenopus because depletion of Cenp-E from egg extracts does not appear to reduce kinetochore localization of BubR1 (Mao et al., 2003). However, in mouse embryonic fibroblasts homozygous for a Cenp-E null allele, levels of kinetochore bound BubR1 and Mad2 are reduced to ∼50% and ∼30%, respectively (Weaver et al., 2003). Although the reduction in Mad2 levels is consistent with our RNAi-based observations, it is not clear why, in our system, BubR1 is not reduced following repression of Cenp-E.

Finally, we examined the effect of repressing Cenp-F. In control cultures, Bub1, BubR1, Mad2 and Cenp-E were readily detectable at Cenp-F-positive kinetochores (Fig. 6A). In cells in which Cenp-F had been repressed, the levels of Bub1, BubR1 and Mad2 present at kinetochores were largely unaffected. By contrast, the levels of Cenp-E were more variable than in controls. Quantitation of normalized pixel intensities shows that, when Cenp-F was reduced to less than ∼10% of its control value, kinetochore bound Bub1, BubR1 and Mad2 were similar to their control values (Fig. 6B). In addition, the quantitative data confirm that Cenp-E levels are reduced to ∼40% of their control values upon repression of Cenp-F.

Taken together, the observations in Figs 3, 4, 5, 6 are consistent with the notion that sequential protein-protein interactions are responsible for the temporal order of recruitment. However, because Cenp-F (which is recruited early) is not required for kinetochore localization of BubR1 and Mad2 (which are recruited later), we can effectively rule out a linear pathway as shown in Fig. 1A.

Recent evidence suggests that Aurora B is required for spindle checkpoint function (Ditchfield et al., 2003; Hauf et al., 2003; Kallio et al., 2002). Indeed, a small-molecule Aurora kinase inhibitor diminishes kinetochore localization of BubR1, Cenp-E and Mad2 (Ditchfield et al., 2003). In addition, repression of Aurora B by RNAi appears to compromise kinetochore function severely (Ditchfield et al., 2003). Consistent with this, kinetochore localization of Bub1 (Fig. 7A,B) and Cenp-F (not shown) was also diminished in nocodazole-treated cells upon repression of Aurora B. Aurora B RNAi also reduced kinetochore localization of Bub1 during prophase (Fig. 7C). Interestingly, although repression of Aurora B reduced kinetochore-bound Bub1 levels to ∼50%, the effect on BubR1 and Cenp-E was more dramatic (∼40% and ∼20%, respectively; Fig. 7B). These observations suggest that Aurora B might not promote kinetochore localization of BubR1 and Cenp-E simply by targeting Bub1 to kinetochores. Rather, it suggests that Aurora B has two effects: targeting Bub1 to kinetochores and increasing the affinity of BubR1 and Cenp-E for Bub1-positive kinetochores (Fig. 10).

Aurora B is recruited to centromeres in late G2 then relocalizes to the spindle mid-zone at anaphase (Adams et al., 2001a). By contrast, the checkpoint proteins are recruited to kinetochores in prophase and prometaphase, and then largely dissociate before anaphase (Waters et al., 1998). These observations suggest that it is unlikely that the checkpoint proteins are required to target Aurora B to centromeres. Indeed, repression of Bub1, BubR1 and Cenp-E had little apparent effect on the ability of Aurora B to localize to centromeres (Fig. 7D).

Consistent with previous observations (Ditchfield et al., 2003; Carvalho et al., 2003; Lens et al., 2003; Yao et al., 2000), repression of BubR1, Cenp-E, Aurora B and Survivin resulted in chromosome alignment defects (Fig. 8A). In addition, we observed that repression of Bub1 also affected chromosome alignment. However, the alignment defects observed in these five cultures appeared to fall into two categories. Specifically, repression of BubR1, Aurora B or Survivin resulted in chromosomes adjacent to the spindle. By contrast, when Cenp-E and Bub1 were repressed, most chromosomes did align on the metaphase plate (Fig. 8A). However, the number of cells with one or more unaligned chromosomes increased from ∼10% in controls to ∼78% and ∼48% in Bub1- and Cenp-E-repressed cultures, respectively (Fig. 8B). Thus, based on spindle morphology, it appears that repression of BubR1, Aurora B and Survivin inhibits chromosome attachment and/or movement on the spindle, whereas repression of Bub1 and Cenp-E results in a congression defect.

To confirm this difference, we analysed the ability of cells to accumulate in metaphase following exposure to the proteasome inhibitor MG132. We predicted that, if chromosomes could bi-orient but had a congression defect, inhibiting the metaphase-anaphase transition would provide more time for the cells to align their chromosomes, thus increasing the metaphase index. By contrast, if chromosomes could not attach and/or bi-orient then delaying the metaphase-anaphase transition would have little effect on the metaphase index. Following exposure to MG132, the number of prophases and anaphases reduced in all cultures, demonstrating that the proteasome inhibitor was preventing mitotic entry and exit (Fig. 8C). In addition, in control cultures, the metaphase index increased from ∼30% to ∼80% in 1 hour, indicating that the metaphase-anaphase transition was being prevented. By contrast, following repression of BubR1, Aurora B and Survivin, the metaphase index increased only marginally to ∼45%. Furthermore, rather than falling, the proportion of prometaphase cells increased from ∼30% to ∼55%, consistent with an attachment/bi-orientation defect. However, following repression of Bub1 or Cenp-E, the metaphase index reached ∼80% within 3 hours, indicating that, in these cells, the chromosomes can align on the metaphase plate if anaphase is prevented. Indeed, when the prometaphase-to-metaphase ratio was determined, control, Bub1- and Cenp-E-repressed cells yielded relatively low values (0.3-0.6), whereas the BubR1-, Aurora-B- and Survivin-repressed cells gave relatively high values (1.1-1.3) (Fig. 8D). Thus, based on spindle morphology and the MG132 assay, this comparative analysis indicates that repression of Bub1 inhibits chromosome congression.

Repression of BubR1 by RNAi compromises the spindle checkpoint (Ditchfield et al., 2003). Specifically, the number of metaphase cells is reduced and the number of anaphases increases. These anaphases are typically aberrant, with one or more lagging chromosomes. In addition, when exposed to spindle toxins, the mitotic index of BubR1 RNAi cultures does not increase as it does in controls. During the course of this work, we were struck by the observation that repression of Bub1 did not yield a similar effect. Indeed, the number of cells in metaphase and anaphase appeared normal (Fig. 9A). Furthermore, anaphases with lagging chromosomes were rarely seen (not shown). When exposed to nocodazole, Bub1-repressed cultures accumulated mitotic cells in a manner very similar to controls, typically ∼20-25% in 6 hours (Fig. 9B). By contrast, the mitotic index in the BubR1 RNAi culture reached only ∼7% after 6 hours. Together, these data suggest that repression of Bub1 by RNAi does not compromise the spindle checkpoint. To confirm this, control, Bub1 and BubR1 RNAi cultures were exposed to nocodazole for 18 hours and the mitotic index determined using two independent markers, namely MPM-2 and phospho-histone H3. Although protein lysates confirmed extensive protein repression in both the Bub1 and BubR1 RNAi cultures (Fig. 9C), the mitotic index of the Bub1 culture was almost identical to the control. Specifically, as judged by MPM-2 staining, ∼60% of cells in the control and Bub1 cultures were mitotic compared with only ∼40% in the BubR1 culture. In addition, whereas phospho-histone-H3 cells were rare in the BubR1 culture (∼35%), they were readily apparent in both the control and Bub1 RNAi culture (∼75%) (Fig. 9E).

Recently, it has been shown that cells undergo mitotic arrest in response to nocodazole but not Taxol when they are exposed to an Aurora kinase inhibitor (Ditchfield et al., 2003; Hauf et al., 2003) or Survivin is repressed by RNAi (Carvalho et al., 2003; Lens et al., 2003). Although repression of Bub1 does not appear to prevent mitotic arrest in response to nocodazole, we asked whether the same was true for Taxol. However, when exposed to Taxol, Bub1-repressed cultures accumulated mitotic cells in a manner very similar to controls, typically ∼20% in 6 hours [see supplementary data Fig. S3]. Thus, repression of Bub1 does not appear to have differential effect on nocodazole-versus Taxol-mediated arrest.

In mammalian cells, several spindle-checkpoint-associated proteins (including Bub1, Cenp-F, BubR1 and Cenp-E) assemble at the kinetochore in a defined order (Jablonski et al., 1998; Taylor et al., 2001). Here, we have shown that Mad2 localizes to kinetochores during prometaphase, after Cenp-E. In addition, we have used RNAi to determine whether these checkpoint proteins are mutually dependent on each other for kinetochore localization. Our observations are consistent with the notion that Bub1, which localizes to kinetochores very early in prophase, is required for the subsequent localization of BubR1, Cenp-E, Cenp-F and Mad2. In addition, whereas repression of Bub1 delays chromosome congression, it does not appear to compromise the spindle checkpoint.

Although RNAi has opened up new opportunities to investigate gene function in human cells (Elbashir et al., 2001), it is not without its limitations. Quantitation of pixel intensities following immunofluorescence analysis of kinetochores indicates that, within a given cell, Bub1 (and the other proteins analysed in this study) can be reduced to less than 10% of its level in control cells. Clearly, therefore, and in contrast to gene knockout approaches, RNAi-mediated gene inactivation is not 100% efficient. Consequently, when interpreting the phenotypes observed following repression, the possibility that the residual protein might be sufficient to provide significant protein function has to be taken into account. Furthermore, when interpreting the data from population-based assays (e.g. Figs 8, 9), the transfection efficiency also has to be taken into account. Although western blotting indicates that repression is extensive in most cells (∼83% reduction, results not shown), immunofluorescence indicates that some cells have wild-type protein levels. Although non-transfected cells provide useful internal controls in single cell assays, they reduce the quantitative effect when analysing a whole population.

These two points have to be taken into account when interpreting the observations following repression of Bub1 where a spindle checkpoint defect was not apparent. A trivial explanation for this is that perhaps the repression, either within any given cell or across a population, was not sufficient to expose the requirement for Bub1 in the checkpoint. Three reasons suggest that this is unlikely. First, quantitative western blotting indicates that, across a population, Bub1 levels can be reduced to ∼17±1.85%. Second, when single cells are analysed, the amount of Bub1 present at kinetochores is reduced to ∼5%. Third, repression of Bub1 is clearly not without consequences: kinetochore localization of at least four other proteins is markedly reduced, with Cenp-F down to ∼10%. In addition, the number of metaphase cells with one or more misaligned chromosomes increases to 80%. Clearly, therefore, repression of Bub1 by RNAi is sufficiently extensive to expose the role of Bub1 in kinetochore assembly and chromosome alignment. Why a checkpoint defect was not observed is unclear but will be discussed further below.

We have used RNAi followed by quantitative immunofluorescence microscopy to analyse kinetochore assembly in human cells. One limitation of this approach is that repression of any given kinetochore protein might affect kinetochore structure, which might in turn affect the accessibility of the other antigens being studied. Indeed, when BubR1 is repressed, the levels of Bub1 and Cenp-F detectable at kinetochores increases. Because BubR1 binds kinetochores after Bub1 and Cenp-F, the simplest explanation for this is that depletion of BubR1 makes the pre-bound Bub1 and Cenp-F more accessible. However, because checkpoint proteins flux through kinetochores (Howell et al., 2000; Howell et al., 2001), we cannot rule out the possibility that repression of BubR1 interferes with this mechanism, leading to the accumulation of other checkpoint proteins at kinetochores. One way to address this issue – although it would open up a separate set of technical issues – would be to perform the RNAi on cell lines ectopically expressing green-fluorescent-protein-tagged fusion proteins, which should be detectable regardless of epitope accessibility. However, despite the limitations of the approach used, our observations are consistent with a simple model, outlined below, which takes into account both the temporal assembly order and the known protein-protein interactions exhibited by these proteins.

Kinetochore localization of the checkpoint proteins occurs in a defined order (Jablonski et al., 1998; Taylor et al., 2001). Here, we have tested the various dependencies that underlie this temporal order. Our observations are consistent with a model (Fig. 10) whereby Bub1 is a master regulator in terms of assembly of a number of checkpoint proteins at kinetochores. Bub1 localizes to kinetochores as soon as chromosome condensation becomes visible (Jablonski et al., 1998) and is required for efficient kinetochore localization of Cenp-F, BubR1, Cenp-E and Mad2. Because Bub1 has been shown to bind Cenp-F (Chan et al., 1998) and BubR1 (Taylor et al., 1998), it is likely that direct protein-protein interactions are responsible for recruiting Cenp-F and BubR1. Cenp-F, which is recruited after Bub1 but before BubR1 (Jablonski et al., 1998), does not appear to be required for kinetochore localization of any of the other proteins analysed here, with the possible exception of Cenp-E. Indeed, Cenp-F can bind Cenp-E in a yeast two-hybrid assay (Chan et al., 1998), suggesting that Cenp-F might play a role in recruiting or stabilizing Cenp-E at kinetochores. Consistent with a stability rather than recruitment role, we observed that repression of Cenp-E also reduces kinetochore-bound Cenp-F, indicating that the Cenp-E/Cenp-F interaction might stabilize the binding of both proteins at kinetochores. Consistent with several previous observations demonstrating that BubR1 and Cenp-E physically interact (Chan et al., 1998; Yao et al., 2000), BubR1 is required for recruitment of Cenp-E. Furthermore, recent observations in Xenopus have shown that activation/inactivation of BubR1 is dependent on Cenp-E (Mao et al., 2003). Cenp-E is also required for efficient kinetochore localization of Mad2, which is in agreement with observations from Xenopus egg extracts (Abrieu et al., 2000) and our observation that Mad2 localizes to kinetochores after Cenp-E (Fig. 1).

Although most of our observations can explained by the model outlined above, a straightforward linear pathway appears to be too simplistic. First, Cenp-F localizes to kinetochores in prophase in a Bub1-dependent manner but, unlike Bub1, Cenp-F is not required for the subsequent localization of the other checkpoint proteins (with the possible exception of Cenp-E). Although Cenp-F might physically interact with Bub1 (Chan et al., 1998), there is at present no evidence to implicate Cenp-F in the spindle checkpoint (Hussein and Taylor, 2002). Thus, Cenp-F might represent another branch of kinetochore assembly that does not involve the spindle checkpoint. However, regardless of Cenp-F, a simple linear model is still not sufficient to explain our observations. In particular, BubR1 is required for Cenp-E recruitment and Cenp-E is required for Mad2 recruitment. Yet, when BubR1 is repressed, Mad2 is largely unaffected. One possible explanation for this is that, although the prior recruitment of proteins in the pathway might well promote the subsequent recruitment of others, additional factors might also promote the assembly and/or stability of these later binding proteins.

A good candidate for one such factor is Aurora B (Adams et al., 2001a). Aurora B accumulates and becomes active shortly before the onset of mitosis (Adams et al., 2001a; Bischoff and Plowman, 1999; Giet and Prigent, 1999). Aurora B is required for correct chromosome alignment in Drosophila (Adams et al., 2001b; Giet and Glover, 2001), Caenorhabditis elegans (Kaitna et al., 2002) and mammalian cells (Kallio et al., 2002; Murata-Hori and Wang, 2002). In addition, Aurora B is required for the localization of all the transient kinetochore proteins tested (Ditchfield et al., 2003). Interestingly however, when Aurora B is repressed to less than 10% of its control value, even though kinetochore bound Bub1 is only reduced to ∼50%, BubR1 and Cenp-E are reduced to ∼40% and ∼20%, respectively. Thus, although Aurora B appears to promote Bub1 localization, proteins that bind later in the assembly pathway appear to be more dependent on Aurora B. However, Aurora B localizes to the inner centromere and is therefore unlikely to play a structural role in terms of recruiting these proteins to kinetochores. Rather, Aurora B might be required to prime either the kinetochore scaffold or the individual proteins, which then localize to kinetochores. Consistent with the notion that multiple pathways are involved in Mad2 recruitment are observations showing that Cenp-I (Liu et al., 2003) and Hec1 (Martin-Lluesma et al., 2002) are required for the kinetochore localization of Mad2 but not Cenp-E.

Bub1-deficient yeast mutants do not undergo mitotic arrest when exposed to spindle toxins (Hoyt et al., 1991; Roberts et al., 1994). Depletion of Bub1 from Xenopus extracts allows MPF inactivation in an in vitro assay that reconstitutes the spindle checkpoint (Sharp-Baker and Chen, 2001). In addition, expression of a Bub1 dominant negative mutant in HeLa cells accelerates progression through a normal mitosis and compromises the checkpoint following spindle damage (Geley et al., 2001; Taylor and McKeon, 1997). We were therefore surprised that repression of Bub1 by RNAi did not yield an apparent checkpoint defect (Fig. 9). If, as outlined above, this is not simply due to insufficient repression of Bub1, how can we account for this observation? Recent studies have shown that repression of Hec1 by RNAi prevents kinetochore localization of Mps1, Mad1 and Mad2. However, despite the absence of these proteins, Hec1-repressed cells are still able to maintain an active spindle checkpoint (Martin-Lluemsa et al., 2002). Although Aurora B is required for checkpoint function, cells nevertheless undergo mitotic arrest when they are exposed to an Aurora kinase inhibitor and nocodazole (Ditchfield et al., 2003; Hauf et al., 2003). To explain this observation, we put forward two possibilities, both of which are applicable here, one of which is qualitative in nature and the other quantitative (Ditchfield et al., 2003). The qualitative possibility is that more than one pathway contributes to mitotic arrest following spindle damage and that Aurora B (as in Ditchfield et al., 2003) or Bub1 (as described here) is only required for one pathway. Thus, in the absence of Bub1, a Bub1-independent mechanism might be sufficient to maintain mitotic arrest. This possibility seems to be at odds with the observation that a dominant negative murine Bub1 mutant (N-mBub1) compromises the checkpoint when expressed in human cells (Geley et al., 2001; Taylor and McKeon, 1997). However, because the domains present in N-mBub1 are also present in BubR1 (Taylor et al., 1998), it is conceivable that N-mBub1 also interferes with BubR1, which does appear to be essential for checkpoint function (Chan et al., 1999; Ditchfield et al., 2003; Shannon et al., 2002). Therefore, whereas Bub1 RNAi does not compromise the checkpoint even though N-mBub1 can, this does not necessarily rule out the possibility that Bub1 is not essential for spindle checkpoint function. Rather, it highlights a limitation of dominant negative studies.

In contrast to the qualitative argument, the quantitative argument posits that the spindle checkpoint is composed of only one pathway but that inhibition of either Aurora B (Ditchfield et al., 2003) or Bub1 (as described here) only partially inhibits this pathway. Indeed, although Bub1 was repressed to low levels, kinetochore-bound BubR1 and Mad2 were only reduced to ∼25%. Thus, if the rate-limiting step in the checkpoint pathway is the recruitment of BubR1 and Mad2 to kinetochores, it is conceivable that, despite repression of Bub1, perhaps 25% of a checkpoint signal emanating from each kinetochore creates a combined signal strength that is sufficient to maintain the checkpoint. Whether levels of kinetochore-bound BubR1 and Mad2 would fall below the threshold level required to prevent mitotic exit upon complete depletion of Bub1 in somatic cells remains to be seen. Therefore, although repression of Bub1 by over 80% appears to be sufficient to dissect its role in kinetochore assembly, perhaps it is not sufficient to expose its role in the checkpoint.

We thank D. Cleveland for anti-Cenp-E antibodies and members of the Taylor lab for reagents, advice and comments. V.L.J. and D.H. are funded by the Biotechnology and Biological Sciences Research Council (BBSRC), M.I.F.S. by the Wellcome Trust, and S.V.H. by the Medical Research Council and Astra Zeneca. S.S.T. is a BBSRC David Phillips Research Fellow.