Requirements for NuMA in maintenance and establishment of mammalian spindle poles

The mitotic spindle is a bipolar array of microtubules required for the alignment and segregation of chromosomes during mitosis. The poles of the spindle are major focal points for the minus ends of spindle microtubules and serve as the final destination for chromosomes segregated in anaphase (Compton, 1998). In animal somatic cell mitosis, spindle poles are coincident with centrosomes, the major cellular microtubule-organizing structures. However, centrosomes are not required for the formation of the spindle or the pole. In vitro, DNA-coated beads catalyze the formation of bipolar microtubule spindles in acentrosomal Xenopus laevis egg extracts (Heald et al., 1996). Additionally, bipolar spindles also form in somatic mitotic cells in which centrosomes have been destroyed by laser ablation or prevented from forming normally by depletion of essential centrosomal components (Khodjakov et al., 2000; Mahoney et al., 2006). Previous studies in Xenopus extracts in vitro have established a role for the abundant spindle pole–localized nuclear mitotic apparatus (NuMA) protein in centrosome-independent spindle pole formation (Heald et al., 1996; Merdes et al., 2000). However, the relative contributions of NuMA and centrosomes in the establishment and maintenance of spindle pole integrity remain to be determined.

The attachment and movement of chromosomes on the mitotic spindle is mediated by kinetochore fibers, bundles of microtubules that link spindle poles to the kinetochores of each chromosome. Microtubules are nucleated in both a centrosome-dependent and -independent manner. Many are initiated directly at the pole by γ-tubulin–dependent nucleation, whereas others are nucleated adjacent to kinetochores through the local activation of components such as Tpx2 (Tulu et al., 2006). In the mitotic cytoplasm, these components are inactive as the result of binding to the importins but can be released by a chromosome-generated gradient of Ran-GTP (Gruss and Vernos, 2004). Under these conditions, microtubules nucleated in the vicinity of kinetochores are elongated and captured by centrosomal microtubules (Dasso, 2001; Khodjakov et al., 2003; Maiato et al., 2004). How these kinetochore-associated bundles are captured and focused toward centrosomes has not been determined. In addition, the means for sustained anchoring of kinetochore fibers at spindle poles and the role of centrosomes in both active focusing and the maintenance of spindle pole integrity are not established.

NuMA, along with cytoplasmic dynein, has been proposed to participate in focusing microtubules toward the poles of the mitotic spindle and physically tethering centrosomes to spindle microtubules (Merdes et al., 1996; Khodjakov et al., 2003). The foundation for our current mechanistic understanding of the role for NuMA during mitosis was established by work in cell-free extracts. Compared with typical mammalian spindles, those formed in Xenopus egg extracts are much larger, undergo substantially greater microtubule flux, and have limited astral microtubules and no cortical attachments (Ganem and Compton, 2006). Furthermore, spindles formed in egg extracts rely significantly more on a gradient of Ran-GTP emanating from chromosomes to contribute to spindle assembly and have relatively few bundled microtubules comprising kinetochore fibers (Kalab et al., 2002, 2006).

Investigations of NuMA function in mammalian mitosis have previously used antibody microinjection approaches, which have yielded contradictory outcomes. An early study reported spindle collapse to monopolarity (Yang and Snyder, 1992), and subsequent efforts demonstrated unfocused spindle poles and an extended mitotic delay (Gaglio et al., 1995). The difficulty in using siRNA to remove the abundant and long-lived NuMA protein has confounded strategies to determine mitotic NuMA function in the mammalian context (Elbashir et al., 2001; Chang et al., 2005). In addition, during interphase, NuMA accumulates in the nucleus, where it has been proposed to participate in aspects of nuclear structure and/or function, which might also be disrupted during extended siRNA treatment (Merdes and Cleveland, 1998; Harborth et al., 1999). To specifically test the principles of mammalian spindle assembly, particularly the mechanisms of spindle pole focusing and the maintenance of pole integrity, we now use gene replacement to engineer mice and cells in which NuMA's mitotic function can be selectively disrupted by administration of the small molecule 4-hydroxytamoxifen (4-OHT). Using this system, we demonstrate that NuMA is essential for early embryogenesis and cellular proliferation. During the first mitosis after inactivation of NuMA, spindles initially form with microtubules focused at centrosomes. However, subsequent to initial spindle assembly and upon generation of spindles forces, centrosome–spindle attachment is uncoupled. As a consequence, kinetochore fibers defocus, and centrosomes fail to maintain and reestablish connection with the spindle. Surprisingly, chromosome segregation is largely intact even without NuMA anchoring of kinetochore fibers to centrosomes. From these findings, we propose that NuMA is essential to maintain centrosome attachments to kinetochore fibers in mammalian mitosis and suggest that NuMA functions redundantly with centrosomes for initial focusing of microtubules at spindle poles.

Immunoblotting of extracts from a panel of mouse tissues indicated that NuMA was expressed at similar levels in most tissues examined (Fig. S1 A, available). In contrast, Mad2, BubR1, and centromere protein E (CENP-E), which are thought to be primarily or exclusively functional during mitosis, were only detected in the testes and spleen, which contain large numbers of dividing cells. Expression of the nuclear proteins lamin A and C was detected in all tissues examined and closely mirrored the expression of NuMA (Fig. S1 A). Examination of NuMA localization by immunostaining of terminally differentiated motor or Purkinje neurons demonstrated that NuMA was exclusively nuclear but excluded from nucleoli (Fig. S1 B). The near-ubiquitous expression pattern and accumulation of NuMA in nuclei of long-lived postmitotic cells supports the possibility of a nonmitotic interphase function of NuMA.

To test a mitotic role for NuMA without disrupting putative nuclear functions, we used gene targeting in mouse embryonic stem (ES) cells to create a conditional loss of mitotic function allele of NuMA. To do this, we identified an in-frame deletion within the NuMA gene that eliminated its microtubule binding without affecting intranuclear accumulation in interphase. The tubulin-binding domain of NuMA is comprised of an ∼100 amino acid region in the C-terminal globular tail of NuMA, which interacts directly with tubulin in vitro and mediates its localization to spindle poles in vivo (Haren and Merdes, 2002). Mouse and human NuMA share 88% of amino acid identity in this domain, which is partially encoded by nucleotides comprising the 22nd exon of the mouse NuMA gene. We tested the localization of GFP-tagged versions of the 410 amino acid C-terminal globular tail of the mouse NuMA protein that were either wild-type or deleted for residues encoded by exon 22 (Fig. S2, A and B, available). The C-terminal tail was used to avoid association of tagged NuMA with endogenous NuMA through the central coiled-coil dimerization domain (Harborth et al., 1995). As expected, a wild-type NuMA tail fragment tagged with GFP localized appropriately to nuclei in interphase and was present in the crescent-shaped pattern typical of endogenous NuMA at spindle poles during mitosis (Fig. S2 C). The same GFP tail fragment, but lacking the 59 amino acids encoded by exon 22, was also nuclear during interphase but was severely compromised in localization to spindle poles during mitosis (Fig. S2, C and D).

Therefore, we generated a targeting construct designed to produce an allele of NuMA in which exon 22 could be conditionally deleted through the action of the Cre recombinase. Homologous recombination in mouse ES cells was used to flank exon 22 with loxP sites to create a NuMA^Neo allele in which the Neo gene was itself deleteable by the action of the flipase (Flp) recombinase (Fig. 1 A). ES clones were screened by PCR and DNA blotting for homologous recombination at 5′ and 3′ ends of the exon 22 targeting construct, and multiple targeted clones were identified (Fig. 1, B–D). Two independently targeted ES cell lines were used to generate mice. These animals were crossed to an Flp-enhanced deleter strain (Rodriguez et al., 2000) to remove the Flp recombinase target (FRT)–flanked neomycin gene from intron 22, thereby generating a conditional allele, NuMA^flox. Subsequent mating to the ZP3-Cre mouse line expressing the Cre recombinase in the female germ line (Lewandoski et al., 1997) produced animals heterozygous for the NuMA deletion allele, NuMA^Δ22 (Fig. 1 A). Genotypes were confirmed using a three-primer PCR reaction capable of distinguishing wild type and each of the engineered NuMA alleles (Fig. 1 E).

Assessment of NuMA levels in ES cells and multiple tissues from mice heterozygous for the NuMA^Neo allele demonstrated a reduction in NuMA protein levels relative to wild-type cells and animals (Fig. 1 F), most likely the result of reduction in correctly spliced NuMA mRNA caused by the presence of the intronic neomycin cassette (Fig. S3, A–D, available). Taking advantage of this hypomorphic expression from NuMA^Neo, we tested the requirement for NuMA in embryonic development and viability and found that the NuMA^Neo allele could not be bred to homozygosity. No NuMA^Neo/Neo live pups or early embryos were ever recovered (0/88 pups and 0/50 embryos; Table I). As expected, the conditional NuMA^flox allele, produced by Flp-mediated removal of the neomycin marker, was readily bred to homozygosity (Table I). Like NuMA^Neo, NuMA^Δ22 (a 59–amino acid internal deletion) was embryonic lethal when homozygous (Table I). There was no evidence for NuMA haploinsufficiency or dominant effects that might be produced from either NuMA^Neo or NuMA^Δ22; in crosses of mice heterozygous for either of these alleles, heterozygous and wild-type animals were produced in expected Mendelian ratios (Table I). Additionally, we aged cohorts of NuMA^Neo/+, NuMA^Δ22/+, and NuMA^flox/flox mice for up to 18 mo, and in no case were any overt phenotypes observed. These observations demonstrate that NuMA is essential for one or more aspects of embryonic development and viability, NuMA^flox encodes a functional protein, and NuMA^Neo and NuMA^Δ22 are loss of function alleles.

To examine cellular phenotypes resulting from NuMA loss of function and that might explain the requirement for NuMA in embryonic viability, cell lines were generated in which NuMA deletion could be induced by the addition of the small molecule 4-OHT. This was accomplished by crossing the NuMA^flox allele into mice carrying a Cre–estrogen receptor tamoxifen mutant (ER^TM) transgene, which ubiquitously expresses the Cre recombinase fused to a mutated form of the estrogen receptor (Hayashi and McMahon, 2002). This receptor is insensitive to estrogen but binds with high affinity to the synthetic ligand 4-OHT. Binding of 4-OHT allows translocation of the receptor-Cre fusion from the cytoplasm to the nucleus, where Cre causes recombination between loxP sites.

Mating NuMA^flox/+,Cre-ER^TM to NuMA^flox/+ mice was used to obtain embryonic day 14.5 embryos that were wild type, heterozygous, or homozygous for the conditional NuMA allele and that also carried the Cre-ER^TM transgene. Mouse embryo fibroblasts (MEFs) were prepared from embryos of each genotype. To confirm that addition of 4-OHT would drive efficient recognition and recombination of loxP sites within the NuMA gene, contact-inhibited cells were treated with 4-OHT for 48 h in low serum. 4-OHT was subsequently washed out, and cells were maintained under growth arrest conditions for an additional 48 h to allow turnover of endogenous NuMA (Fig. 2 A). Multiple independent NuMA^flox/flox, Cre-ER^TM cell lines showed highly efficient deletion of the loxP-flanked region after 48 h of 4-OHT treatment (Fig. 2 B).

Quantification of recombination by quantitative PCR (qPCR) showed time and 4-OHT dose-dependent efficiency to exon 22 excision with a maximum effect observed after 48 h, resulting in a fibroblast population with 90% conversion of NuMA^flox alleles to NuMA^Δ22 by day 4 (Fig. 2 C). Immunoblotting revealed that NuMA^Δ22 accumulated to a lower than wild-type level (Fig. 2 D). Remaining NuMA protein in NuMA^Δ22/Δ22 cells could not be eliminated by doubling the duration of cell cycle arrest after 4-OHT washout, suggesting that all wild-type NuMA had been completely turned over. As expected, the amino acids encoded by exon 22 were dispensable for nuclear accumulation of NuMA but were necessary for efficient spindle pole localization. We conclude that wild-type NuMA protein is absent from NuMA^Δ22/Δ22 MEFs 4 d after 4-OHT treatment and that the mutant protein accumulates to ∼30% of the level of the wild-type NuMA polypeptide.

To directly assess the consequences of preventing NuMA accumulation at spindle poles, population growth curves were generated for NuMA^+/+, NuMA^flox/+, and NuMA^flox/flox MEFs, each carrying the Cre-ER^TM transgene. After treatment with 4-OHT to convert the NuMA^flox allele to NuMA^Δ22, NuMA^+/+ and NuMA^+/Δ22 cells grew normally after release from low serum arrest, demonstrating that there are no dominant effects on cell cycle progression of mutant protein produced from the NuMA^Δ22 allele. Strikingly, NuMA^Δ22/Δ22 MEFs failed to increase in number after release from growth arrest (Fig. 2 E), demonstrating that removal of exon 22 results in cell-autonomous growth defects. These conclusions are consistent with cell-intrinsic defects, causing early embryonic lethality in NuMA^Δ22/Δ22 mice and the lack of dominant effects or haploinsufficiency observed in NuMA^Δ22/+ animals, which develop and age normally.

To determine whether loss of NuMA from spindle poles provoked extended delays in mitosis, as has been previously reported from antibody microinjection-mediated disruption of the protein, the mitotic index of NuMA^Δ22/Δ22, NuMA^Δ22/+, and NuMA^+/+ MEFs was determined (Gaglio et al., 1995). There was no increase in the percentage of cells in mitosis after NuMA disruption (Fig. 2 F). Mitotic timing was reexamined even more directly in NuMA^+/+,Cre-ER^TM and NuMA^flox/flox,Cre-ER^TM immortalized fibroblasts, which were treated with 4-OHT, and the duration of mitosis was determined by phase-contrast time-lapse microscopy. After 4-OHT treatment, NuMA^flox was converted to NuMA^Δ22 with the same efficiency as seen in primary cells. Mitosis was defined as the period between the first stage of cell rounding and the point at which cells had completely flattened back onto the substrate. Consistent with the failure of NuMA disruption to alter mitotic index, no significant difference in the duration of mitosis was observed between NuMA^Δ22/Δ22 and control cells (Fig. 2 G).

Despite normal timing of progression through mitosis, NuMA^Δ22/Δ22 primary cells displayed obvious spindle defects, most striking of which was the detachment of microtubule-nucleating structures from the ends of mitotic spindles. Costaining metaphase spindles with a γ-tubulin antibody provided strong evidence that the dissociated structures were bona fide centrosomes (Fig. S4 A, available). In some cells, spindles retained focused arrays of microtubule ends despite having no closely associated centrosome, whereas in others, even poles with associated centrosomes lacked a discernable microtubule focus (Fig. 3 A). Greater than 50% of metaphase-like cells in the NuMA^Δ22/Δ22 population had at least one centrosome that was clearly not associated with a spindle pole (apolar centrosome), and a similar frequency had at least one pole in which microtubules did not converge to a single focal point (unfocused kinetochore fibers; Fig. 3 C). These phenotypes contrasted sharply with metaphases observed in wild-type and NuMA^+/Δ22 cells, which showed typical spindle architecture with focused microtubule arrays terminating near centrosomes (Fig. 3, A and C). In addition to centrosome–spindle coupling and microtubule-focusing defects, metaphase spindle lengths in MG132-arrested NuMA^Δ22/Δ22 cells were on average 30% longer than either wild-type or NuMA^+/Δ22 controls (Fig. 3 D).

The spindle defects observed in NuMA^Δ22/Δ22 cells were also present in anaphase (Fig. 3 B). However, immunostaining of unperturbed fibroblasts with a cyclin B1 antibody confirmed that the apolar centrosome and pole-defocusing phenotypes in NuMA^Δ22/Δ22 fibroblasts occurred before anaphase onset (Fig. S4 B). In addition, cells arrested in metaphase by blocking cyclin B and securin degradation by treatment with the proteasome inhibitor MG132 showed similar spindle defects at high frequency, corroborating this interpretation (Fig. S4 A). Tension between sister kinetochores in pole-defocused spindles was reduced by ∼33% compared with normal cells (Fig. 4, A and B), and chromosome alignment defects were observed more frequently in metaphase NuMA^Δ22/Δ22 than control NuMA^+/Δ22 fibroblasts (Fig. 4, C and D). Therefore, spindles in NuMA^Δ22/Δ22 cells are deficient in their ability to apply or retain tension across sister kinetochores and align chromosomes in a metaphase alignment.

Spindle abnormalities were likely not caused by a failure of the dynein–dynactin complex to localize appropriately. A component of dynactin, p150, localized to centrosomes independent of spindle attachment, suggesting that failure of centrosome–spindle coupling and loss of kinetochore fiber focusing did not result from mislocalization of dynein–dynactin, the major mitotic binding partner of NuMA (Fig. S4 C). Cold treatment to preferentially depolymerize dynamic microtubules confirmed that kinetochore fibers remained stably attached to kinetochores in NuMA-disrupted cells even when completely defocused at poles and detached from centrosomes (Fig. S4 D).

From these results, we conclude that spindle pole integrity is defective in the absence of NuMA and more specifically that NuMA is required at spindle poles to couple centrosomes to kinetochore fibers. We also note that although kinetochore–microtubule attachments are apparently intact, tension across sister kinetochores is reduced, and chromosome alignment is defective in NuMA^Δ22/Δ22 cells.

The displacement of centrosomes from kinetochore fibers in metaphase could arise either from an initial failure of NuMA^Δ22/Δ22 cells to focus microtubules at the poles of the spindle or from an inability to maintain centrosome attachment to a fully formed spindle. To distinguish these possibilities, we followed spindle formation and mitotic progression in live NuMA^Δ22/Δ22 MEFs. Primary cells were transduced with a retrovirus encoding tubulin-YFP 2 d before the initiation of Cre recombinase activity with 4-OHT and arrest in G0/G1 with low serum. Cells were subsequently released from arrest and filmed by time-lapse fluorescence microscopy. NuMA^+/Δ22 cells always formed bipolar spindles and maintained centrosomes at spindle poles throughout mitosis (6/6 morphologically normal spindles; Fig. 5 A; and Video 1, available). In contrast, NuMA^Δ22/Δ22 fibroblasts exhibited a high frequency of pole defocusing and centrosome detachment phenotypes (9/13 spindles aberrant) (Fig. 5 B). Live cell imaging showed clearly that in NuMA^Δ22/Δ22 cells, microtubules can establish focusing at centrosomes during the initial stages of spindle assembly but that as cells progress through mitosis, maintenance of centrosome–spindle coupling is lost, and poles subsequently defocus.

Despite these defects, completion of mitosis appeared normal, daughter cells formed morphologically normal nuclei, and cytokinesis, including abscission, also occurred (Video 2, available). We also measured the duration of mitosis in tubulin-YFP–expressing MEFs, which we defined as the length of time between nuclear envelope breakdown and the first observable stage of midbody formation. Consistent with our previous observations, mitotic timing was similar for control (33 ± 9 min, n = 6) and NuMA^Δ22/Δ22 cells that experienced spindle defects (27 ± 5 min, n = 5). Additionally, the kinetics of spindle bipolarization were similar in control (12 ± 4 min, n = 6) and phenotypic NuMA^Δ22/Δ22 cells (13 ± 3 min, n = 5). These observations indicate that loss of NuMA function during mitosis does not prevent the initial timely establishment of focused bipolar spindles but results in centrosome loss from the spindle poles before anaphase followed by microtubule defocusing. Surprisingly, despite these defects, bulk chromosome segregation proceeds largely as normal in NuMA^Δ22/Δ22 cells (Fig. 3 B and Video 2).

The ability of NuMA^Δ22/Δ22 cells to enter mitosis and the spindle phenotypes observed were consistent with specific defects in spindle pole integrity. To confirm that a potential nuclear defect in NuMA^Δ22/Δ22 cells was not responsible for the subsequent mitotic deficits, we performed an in vivo test for retention of nuclear function of the NuMA^Δ22 protein. To do this, wild-type NuMA was replaced with NuMA^Δ22 protein in spinal cord motor neurons, a cell type in which NuMA is known to be present in nuclei in adult neurons (Fig. S1 B). These cells were chosen for two reasons: first, defects in motor neuron function give rise to characteristic and easily scorable motor deficits, including gait alterations and paralysis. Second, the lifetimes of individual neurons extend for the full life span of the organism, allowing access to any age-dependent features of loss of interphase NuMA function.

NuMA was deleted specifically from motor neurons by crossing the NuMA^flox allele with animals carrying the vesicular acetylcholine transporter (VAchT)–Cre transgene. This transgene has previously been demonstrated to direct Cre recombinase expression exclusively in motor neurons after the final division of motor neuron progenitors, ensuring that any phenotypes would arise exclusively from postmitotic loss of NuMA (Misawa et al., 2003). NuMA^flox/flox,VAchT-Cre, NuMA^+/flox,VAchT-Cre, and NuMA^flox/flox littermate controls were obtained, and animals of all genotypes and genders were produced in expected Mendelian ratios. Activity of the Cre recombinase and restriction of expression to motor neurons in VAchT-Cre mice was confirmed by mating to the ROSA26R reporter strain (Soriano, 1999). Examination of spinal cord and sagittal brain sections from VAchT-Cre,ROSA26R mice for β-galactosidase activity confirmed that Cre expression was restricted to motor neurons (Fig. S5 A, available).

Extensive visual observation demonstrated that NuMA^flox/flox,VAchT-Cre animals showed no behavioral defects indicative of motor neuron dysfunction, even at up to 2 yr of age. Weight gain was also similar in control and NuMA^flox/flox,VAchT-Cre mice, providing no evidence for motor neuron dysfunction or muscle denervation (Fig. S5 B). Finally, counting of cresyl violet–stained motor neuron cell bodies in spinal cord sections of 200-d-old animals revealed no effect on motor neuron numbers (Fig. S5 C). Thus, the exon 22–deleted mutant complements all potential nuclear functions of NuMA and does so for extended periods in vivo. Alternatively, NuMA may perform limited or redundant interphase functions in these cells.

Live imaging of tubulin-YFP–marked spindles in unperturbed fibroblasts demonstrated that in the absence of NuMA, initial microtubule focusing at the pole appeared normal. This strongly supports a maintenance rather than establishment role for NuMA in the attachment of centrosomes to kinetochore fibers. To further test this, we arrested NuMA wild-type, NuMA^+/Δ22, and NuMA^Δ22/Δ22 cells in mitosis with the Eg5 inhibitor S-trityl-l-cysteine (STLC; Skoufias et al., 2006). Eg5 is a bipolar, homotetrameric, kinesin-related motor protein that is required during mitosis to both establish and maintain separation of centrosomes. It is thought to function by cross-linking and sliding apart antiparallel microtubules from opposing centrosomes, thereby pushing centrosomes apart (Kapitein et al., 2005). STLC inhibits activity of Eg5, causing collapse of bipolar spindles. Regardless of genotype, in the absence of STLC, monopolar spindles were observed with equally low frequency in all cell lines, but upon STLC treatment, all MEF cell lines produced a high frequency of monopolar spindles that appeared almost exclusively as chromosome rosettes surrounding a central γ-tubulin–positive pole (Fig. 6, A and B). This indicates that in monopolar spindles, centrosomes continue to dictate the location of spindle pole formation when NuMA function is disrupted. Furthermore, under these conditions, kinetochore fibers do not require NuMA for active focusing toward centrosomes.

Throughout the extended mitotic delay caused by treatment with STLC, centrosomes persisted at the center of monopolar spindles in NuMA^Δ22/Δ22 cells. How can this be reconciled with a maintenance role of NuMA in centrosome–spindle coupling? Relative to bipolar metaphase spindles, monopolar spindles have reduced tension forces acting between centrosomes and kinetochores. Thus, the spindle defects produced in cycling NuMA^Δ22/Δ22 cells likely reflect a requirement for NuMA in maintaining centrosomes at poles of spindles experiencing typical metaphase forces. To directly test this possibility, STLC was washed out from monopolar-arrested cells to reintroduce tension into spindles with kinetochore fibers focused toward a monopole. 1 h after release, cells were fixed for immunofluorescence. In control cells, anaphases always had one centrosome at each pole (Fig. 6 C). In contrast, NuMA^Δ22/Δ22 cells displayed anaphase figures in which chromosomes had segregated normally, but spindle poles were defocused, and centrosomes (detected as discrete foci of strong γ-tubulin staining) had failed to separate.

These results were corroborated by following the recovery of spindle bipolarity in retrovirally transduced α-tubulin-YFP–expressing fibroblasts. After release from STLC, control cells were uniformly observed to recover from monopolarity by separating centrosomes and forming bipolar spindles with well-focused poles (7/7 cells; Fig. 6 D; and Video 3, available). However, after the reestablishment of tension, NuMA^Δ22/Δ22 fibroblasts often formed spindles with one well-focused pole and one splayed spindle pole lacking a centrosome (8/13 cells; Fig. 6 D and Video 4). We conclude that mitotic NuMA function is required to maintain centrosome–spindle attachments under conditions in which tension is applied across the spindle.

Adjacent centrosomes that failed to separate in NuMA^Δ22/Δ22 cells were always initially observed at the spindle end with focused microtubules. As NuMA^Δ22/Δ22 cells progressed through mitosis, the paired centrosomes often dissociated from the pole. These cone-shaped spindles were not arrested monopolar spindles observed in profile because they uniformly proceeded into anaphase, indicating that mitotic checkpoint signaling was silenced. These data demonstrate that normal NuMA function is required for centrosome separation after recovery from a monopolar state. In addition, because the centrosome distal half of each NuMA-deficient spindle forms a splayed pole after release from STLC, we conclude that in the absence of centrosomes, NuMA is required for the establishment of microtubule focusing.

The aforementioned evidence demonstrates a requirement for NuMA in the physical tethering of centrosomes to kinetochore fibers and reveals the surprising finding that centrosome–spindle connections are required for centrosome separation after recovery from monopolarity. This finding has important implications for the mechanism of centrosome separation during reestablishment of spindle bipolarity in prometaphase. Therefore, we specifically tested whether stable microtubule–kinetochore attachments are also required for centrosome separation in recovering monopolar spindles. The Ndc80 complex, a heterotetramer composed of the Ndc80, Nuf2, Spc24, and Spc25 proteins, is a key component involved in the stable attachment of microtubule bundles at kinetochores (Cheeseman and Desai, 2005). To directly test the requirement for kinetochore–microtubule attachment in centrosome separation, we used siRNA to deplete Nuf2 in HeLa cells stably expressing fluorescently tagged tubulin and histone H2B (Fig. 7 A). These cells were arrested with monopolar spindles using STLC, the drug was removed, and time-lapse fluorescence microscopy was used to observe recovery of spindle bipolarity. 90 min after STLC washout, all but 3% of control siRNA-treated cells had separated their centrosomes and formed bipolar spindles. However, in the same period of time, 35% of Nuf2-depleted spindles failed to resolve their centrosomes and form bipolar spindles (Fig. 7, B and C). We conclude that efficient centrosome separation after recovery from monopolar spindles relies on the two-point attachment of kinetochore fibers at centrosomes and kinetochores.

In this study, we describe the creation of a separation of function allele of NuMA by specific deletion of exon 22 of the mouse NuMA gene. Removal of exon 22 produces a mutant protein that is unable to efficiently localize to spindle poles during mitosis but is able to perform all essential interphase functions of NuMA. By combining inducible inactivation of this allele with a prolonged G0/G1 arrest, we were able to achieve nearly quantitative turnover of accumulated wild-type NuMA protein before the first mitotic division. Strikingly, we show that in the absence of functional NuMA, and despite apparently normal initial bipolar spindle formation, centrosomes are unable to maintain attachment to the ends of kinetochore fibers. The subsequent defocusing of microtubules at spindle poles reveals a distinct requirement for NuMA in the maintenance of mammalian centrosome–spindle attachments.

A previous study has implicated a role for NuMA-mediated capture of kinetochore fibers in pole focusing (Khodjakov et al., 2003). This has been proposed to occur through the capture of kinetochore fibers by NuMA (Khodjakov et al., 2003), which itself is transported toward spindle poles, likely on centrosomal/astral microtubules by dynein (Merdes et al., 2000). However, centrosomes are not required for microtubule focusing or NuMA localization to spindle poles (Khodjakov et al., 2000). Our efforts in this study have extended this to demonstrate that in the absence of functional NuMA, initial bipolar spindle formation proceeds normally. However, after the establishment of tension across kinetochores, centrosomes subsequently detach from the spindle, and microtubules defocus. This demonstrates that NuMA is required to physically tether kinetochore fiber minus ends at the poles of bipolar spindles under tension and provides the first evidence that an essential function of NuMA during mitosis is in the maintenance and not establishment of mammalian centrosomal–spindle connections.

The normal initiation of bipolar spindle formation in the absence of NuMA suggests the existence of alternative processes involved in establishing focused spindle poles. These are likely to be dependent on the presence of centrosomes, which might function in cooperation with components such as HSET in mammals and Ncd in Drosophila melanogaster (Gordon et al., 2001; Goshima et al., 2005). This interpretation is corroborated by our observation that after recovery from monopolarity, microtubules in NuMA^Δ22/Δ22 fibroblasts are severely defocused in the acentrosomal spindle half. The formation of these monoastral, bipolar cone-shaped spindles provides clear evidence that NuMA is required for spindle pole focusing in the absence of centrosomes. Thus, NuMA and centrosomes function redundantly in establishing spindle poles by contributing independently to initial microtubule focusing. It is likely that mammalian acentrosomal spindle systems, including mouse oocytes, rely heavily on NuMA for the establishment of spindle poles, which is a question now testable with the NuMA^Δ22 allele.

Disruption of dynein function gives rise to defects similar to those we have observed in NuMA^Δ22/Δ22 MEFs, including centrosome–spindle uncoupling and, to a lesser extent, a mild degree of kinetochore fiber defocusing during metaphase (Heald et al., 1997; Goshima et al., 2005). In multiple contexts, dynein inhibition causes NuMA to redistribute from spindle poles to along the length of spindle microtubules. This suggests that a major function of spindle-localized dynein in mitosis may be to transport NuMA as a cargo to microtubule minus ends (Gaglio et al., 1997; Merdes et al., 2000). The rapid kinetics of NuMA localization to spindle poles and the kinetochore fiber–independent nature of NuMA accumulation at poles (unpublished data) is consistent with this interpretation. Therefore, much of the phenotype caused by inhibition of dynein function may be a direct result of a failure to properly localize NuMA.

In mitotic prophase before nuclear envelope breakdown, two duplicated centrosomes move across the nuclear envelope in a dynein-dependent manner to diametrically oppose each other in anticipation of bipolar spindle formation (Gonczy et al., 1999). Maintenance of centrosome separation after nuclear envelope breakdown requires activity of the plus end–directed motor Eg5, which acts to slide apart antiparallel microtubules that emanate directly from centrosomes. Inhibition of Eg5 activity causes centrosome collapse after nuclear envelope breakdown, triggering an extended mitotic delay (Kapitein et al., 2005; Skoufias et al., 2006). Surprisingly, we demonstrate that centrosome separation fails after restoration of Eg5 activity in monopolar NuMA^Δ22/Δ22 cells, indicating a novel function for NuMA in mitosis. Anchoring of kinetochore fibers at the spindle pole may be critical to allow Eg5 to drive initial separation of juxtaposed centrosomes. Once this occurs, overlapping antiparallel arrays of microtubules are formed between centrosomes, providing forces to drive further centrosome separation. Consistent with this interpretation, in the absence of stable kinetochore–microtubule interactions, centrosome separation after release from monopolarity is also inhibited. Thus, we provide a demonstration that Eg5-dependent centrosome separation requires the two-point attachment of microtubules at kinetochores and centrosomes.

The striking spindle defects observed in NuMA^Δ22/Δ22 fibroblasts did not delay progression through mitosis. Because a single unattached kinetochore is capable of delaying anaphase onset (Rieder et al., 1994), this indicates that microtubule capture by kinetochores occurs efficiently in the absence of functional NuMA. Furthermore, bulk chromosome segregation was observed to occur normally on spindles with detached centrosomes and defocused poles. These observations indicate that pole focusing is not an absolute requirement for (a) all kinetochores to be attached to spindle microtubules and (b) the bulk of chromosomes to be aligned and segregated in anaphase. Indeed, plant cells carry out faithful mitoses without centrosomes, using spindles without discernibly focused poles (Franklin and Cande, 1999). Additionally, mechanical severing of kinetochore fibers between spindle poles and kinetochores has been shown to not prevent continued poleward chromosome movements (Nicklas, 1989).

Although bulk chromosome segregation occurs normally, most NuMA^Δ22/Δ22 cells exhibited modest chromosome alignment defects, such as an elongated spindle and reduced tension across sister kinetochores. These defects are likely to result in chromosome segregation errors that explain the requirement of NuMA-dependent pole focusing for mouse embryonic viability and normal cellular proliferation. This is consistent with chromosome missegregation driving embryonic lethality (e.g., from loss of the centromere motor/tether CENP-E; Putkey et al., 2002). Alternatively, the early lethality of NuMA^Δ22/Δ22 embryos may result from missegregation of centrosomes or centrioles as spindle cargo (Pickett-Heaps 1969), which are not required for mitotic spindle function but are essential for normal embryonic development (Chatzimeletiou et al., 2008). Therefore, our evidence collectively demonstrates that NuMA and its role in focusing spindle poles is a required component of normal mammalian mitotic progression and embryonic development.

A mouse 129S6/SvEvTac bacterial artificial chromosome genomic library (RPCI-22; Children's Hospital Oakland Research Institute, Oakland, CA) was screened for the NuMA gene with a probe to the largest exon of NuMA. A single clone was identified and confirmed by PCR to contain the entire NuMA gene. An 8.9-kb EcoRV–ScaI fragment containing exons 16–25 was subcloned into pBluescript (Agilent Technologies). A loxP sequence and 3′ SacII restriction site were introduced upstream of exon 22 by cloning an oligonucleotide linker into a BstEII site. A loxP-FRT-PGKNeo-FRT fragment from pK-11 (Meyers et al., 1998) was inserted downstream of exon 22. Finally, the diphtheria toxin negative selection marker (Yanagawa et al., 1999) was introduced downstream of the short homology arm. The targeting vector was linearized with NotI before electroporation into ES cells. ES cell culture and transfection was performed as described previously (Putkey et al., 2002). DNA from G418-resistant ES clones was screened using PCR and genomic DNA blotting. The probe for targeting of exon 22 was amplified from R1 ES cell DNA using primers pr13 forward (5′-TTTGGTGGTGGTTTGGTC-3′) and pr13 reverse (5′-TGATAAGCAATGCCACGG-3′). Mice, ES cells, and embryos were genotyped as shown in Fig. 1 A with primers i (5′-AACCGCATCGCAGAGTTGCAG-3′), ii (5′-ATGCTCCAGACTGCCTTGGG-3′), iii (5′-GAGGAGTGGTGGCAACAGTAG-3′), and iv (5′-GGAGGTCATTCTACTGGAAG-3′). ES clones carrying the appropriate genetic modification were injected into C57/BL6 embryos at the Transgenic Mouse and Gene Targeting Core facility (University of California, San Diego, La Jolla, CA). Pups from chimeric mice were screened by coat color, and PCR was used to identify germline transmission.

Mouse embryonic fibroblasts were prepared as previously described (Putkey et al., 2002). Primary fibroblasts were maintained and passaged in an incubator set at 10% CO₂ at 37°C and maintained at 3% O₂ by a continuous flow of nitrogen gas to increase replicative lifespan as described previously (Parrinello et al., 2003). In all experiments using primary cells, fibroblasts were grown for no more than two cumulative weeks in culture after derivation.

To induce growth arrest, MEFs were cultured to confluence and shifted to media containing 2% serum. In fibroblasts carrying the Cre-ER^TM transgene, 4-OHT (10 mg/ml stock in ethanol; Sigma-Aldrich) was added to a final concentration of 0.1 µM (unless otherwise indicated) to cause nuclear translocation of Cre.

Retroviral plasmids (pBABE variants) were prepared as described previously (Shah et al., 2004). For retroviral infection of primary MEFs, cells grown in 12-well plates were washed with PBS and incubated for 15 min in a humidified incubator with retroviral supernatant mixed with 8 µg/ml polybrene (Sigma-Aldrich). Plates of cells were wrapped in parafilm and spun at 1,100 g for 30 min at room temperature, after which the supernatant was replaced with primary MEF media, and cells were returned to the incubator.

For analysis of splicing defects in NuMA^Neo/+ cells and tissues, RNA was prepared from mouse fibroblasts using the RNeasy Mini kit (QIAGEN) and from cells and tissues using TRIZOL reagent (Invitrogen). cDNA was prepared using the SuperScript III First-Strand Synthesis kit (Invitrogen). Splicing events were detected in NuMA^Neo/+ cells and tissues by PCR amplification from cDNA with primers pr4 forward (5′-GTTTCAGAGAACTCGCGGCAGG-3′) and Neo31 (5′-GGATTCATCGACTGTGGCCGGCTGGGT-3′). The C-terminal half of the NuMA cDNA was amplified from the NuMA⁺ allele as a control, using primers pr4 forward and ex24 reverse (5′-GAGGAGTGGTGGCAACAGTAG-3′). For use in qPCR, RNA from cells and tissues was treated with the RNase-Free DNase set (QIAGEN) before reverse transcription. The loss of full-length NuMA mRNA from NuMA^Neo/+ cells and tissues was measured with primers 22f.2 (5′-AACCGCATCGCAGAGTTGCAG-3′) and 23r.2 (5′-TTACGTCCTTCATGCCGGTCC-3′). Cyclophilin A was amplified as a normalizer using primers cycloA forward (5′-TTCACCTTCCCAAAGACCAC-3′) and cycloA reverse (5′-AGCACTGGAGAGAAAGGATT-3′). The degree of mRNA loss was calculated using the comparative cycle time method by normalizing the signal produced from NuMA to cyclophilin A reactions and comparing NuMA^Neo/+ and NuMA^+/+ samples.

For measurements of Cre-mediated NuMA exon 22 deletion in fibroblasts, DNA was prepared using the DNeasy Blood and Tissue kit (QIAGEN). The loss of NuMA^flox from genomic DNA was measured using primers 22f.4 (5′-CTTGCTCTATACAGTTGGGCC-3′) and 22 reverse (5′-GGTGGGTCTCAGAGGAAACTCG-3′). The mouse γ-actin gene was amplified as a normalizer using primers 6590 (5′-TGGATCAGCAAGCAGGAGTATG-3′) and 6591 (5′-CCTGCTCAGTCCATCTAGAAGCA-3′). Reactions were performed using 40 ng genomic DNA and the SYBR Green Supermix qPCR reagent (Bio-Rad Laboratories). The degree of excision was calculated by normalizing the signal produced from an experimental sample to the unexcised gene γ-actin and comparing experimental samples to DNA extracted from NuMA^+/+ MEFs as described previously (Pfaffl, 2001). All qPCR reactions were run using the iCycler (Bio-Rad Laboratories).

HeLa and MEF cells were seeded onto 35-mm glass-bottom dishes (MatTek) and incubated in CO₂-independent medium (Invitrogen) supplemented as described previously (Weaver et al., 2007). Dishes were placed in a heat-controlled stage set at 37°C. Fluorescence imaging was conducted using a spinning disk confocal (McBain Instruments) attached to an inverted microscope (TE2000e; Nikon) equipped with a 60×/1.4 NA objective lens. Fluorescence excitation was controlled by Metamorph software (MDS Analytical Technologies). Z-series images were acquired using a camera (Orca-ER; Hamamatsu Photonics) at 3- or 5-min intervals. Z stacks were compiled by maximum intensity projection for presentation.

For indirect immunofluorescence analysis of frozen tissue sections, brain and spinal cord were prepared and processed as described previously (Lobsiger et al., 2005). Antibody staining was performed overnight at room temperature. Sections were mounted on slides (Superfrost Plus; Thermo Fisher Scientific) in a solution of 0.2% gelatin in PBS, dried overnight, and covered with antifade reagent (ProLong; Invitrogen) and glass coverslips before imaging.

For immunofluorescence analysis of cultured cells, cells were grown on hydrochloric acid–washed, poly-l-lysine (Sigma-Aldrich)–coated 18-mm glass coverslips and fixed in either ice-cold methanol for 10 min (γ-tubulin and cyclin B1 staining) or 4% formaldehyde in PBS for 10 min and processed using standard conditions. DNA was visualized by staining with DAPI. Images of fixed cells were acquired using a 100× oil objective on a DeltaVision-modified inverted microscope (IX70; Olympus) using SoftWorx software (Applied Precision, LLC) and were deconvolved unless otherwise indicated. Rabbit polyclonal antibodies used in this study include anti–Xenopus NuMA tail (1:250; Merdes et al., 1996), anti–human NuMA tail (1:500; Gaglio et al., 1995), anti–CENP-E HpX (1:200; Yao et al., 1997), anti–lamin A/C (1:500; provided by L. Gerace, Scripps Research Institute, San Diego, CA), and anti-Mad2 (1:250).

Monoclonal antibodies used include SMI-32 (1:1,000; Sternberger Monoclonals), GTU88 (1:1,000; Sigma-Aldrich), DM1A (1:1,000; Abcam), GNS1 (1:100; Santa Cruz Biotechnology, Inc.), p150^Glued (1:100; BD), and 5F9 (1:500; Taylor et al., 2001). These antibodies identify neurofilaments, γ-tubulin, α-tubulin, cyclin B1, dynein complex component p150^Glued, and BubR1, respectively. Additionally, antibodies isolated from human autoantiserum available as anticentromeric antibodies (1:100; Antibodies Inc.) were used to identify kinetochores. Secondary antibodies (Jackson ImmunoResearch Laboratories) labeled with FITC, Texas red, or Cy3 were used at final dilutions of 1:200.

Fig. S1 shows the expression pattern of NuMA in mouse tissues and the localization of NuMA in adult mouse neurons. Fig. S2 provides data characterizing the localization of GFP-tagged NuMA tail fragments. Fig. S3 provides data demonstrating splicing defects underlying hypomorphic expression from the NuMA^Neo allele. Fig. S4 provides additional data characterizing the phenotype of NuMA^Δ22/Δ22 cells, including timing of centrosome detachment, dynactin localization, and kinetochore fiber stability. Fig. S5 includes data that characterize VAchT-Cre transgenic mice and show the phenotype of animals with deletion of NuMA exon 22 in motor neurons. Videos 1 and 2 illustrate spindle formation and integrity throughout mitosis in cycling NuMA^+/Δ22 and NuMA^Δ22/Δ22 primary fibroblasts. Videos 3 and 4 show mitotic spindle structure and centrosome positioning during recovery from STLC-induced monopolarity in NuMA^+/Δ22 and NuMA^Δ22/Δ22 primary fibroblasts.

We thank A. Desai and K. Oegema for helpful comments and access to imaging equipment. We also acknowledge members of the Cleveland laboratory for invaluable support and helpful discussion throughout the duration of this work.

This work was supported by a grant (GM29513) from the National Institutes of Health to D.W. Cleveland, who receives salary support from the Ludwig Institute for Cancer Research. A.J. Holland is supported by a European Molecular Biology Organization Long-Term Fellowship.

Note added in proof. Since the final submission of this manuscript, an independent study also identified a role of kinetochore-dependent forces in prometaphase centrosome separation (Toso et al., 2009).