Elongation of centriolar microtubule triplets contributes to the formation of the mitotic spindle in γ-tubulin-depleted cells

The assembly of the mitotic spindle after depletion of the major γ-tubulin isotype by RNA-mediated interference was assessed in the Drosophila S2 cell line. Depletion of γ-tubulin had no significant effect on the cytoskeletal microtubules during interphase. However, it promoted an increase in the mitotic index, resulting mainly in monopolar and, to a lesser extent, asymmetrical bipolar prometaphases lacking astral microtubules. This mitotic accumulation coincided with the activation of the mitotic checkpoint. Immunostaining with an anti-Asp antibody revealed that the spindle poles, which were always devoid of γ-tubulin, were unfocused and organized into sub-spindles. Despite the marked depletion of γ-tubulin, the pericentriolar proteins CP190 and centrosomin were recruited to the spindle pole(s), where they formed three or four dots, suggesting the presence of several centrioles. Electron microscopic reconstructions demonstrated that most of the monopolar spindles exhibited three or four centrioles, indicating centriole duplication with a failure in the separation process. Most of the centrioles were shortened, suggesting a role for γ-tubulin in centriole morphogenesis. Moreover, in contrast to metaphases observed in control cells, in which the spindle microtubules radiated from the pericentriolar material, in γ-tubulin-depleted cells, microtubule assembly still occurred at the poles but involved the elongation of centriolar microtubule triplets. Our results demonstrate that, after depletion of γ-tubulin, the pericentriolar material is unable to promote efficient microtubule nucleation. They point to an alternative mechanism of centrosomal microtubule assembly that contributes to the formation of abnormal, albeit partially functional, mitotic spindles.

regarding the in vivo function of γ-tubulin in microtubule nucleation. In both fungal and animal cells, attempts to deplete γ-tubulin by gene knockouts have not allowed the complete disappearance of the interphase and mitotic microtubules. In Caenorabditis elegans embryos depleted of γ-tubulin by either RNA-mediated interference (RNAi) or mutation, the centrosomal asters failed to form during interphase but assembled as cells entered into mitosis, giving rise to collapsed spindles (Strome et al., 2001;Hannak et al., 2002). In other cases, such as Drosophila 23C γ-tubulin mutants or Saccharomyces γ-tubulin conditional mutants, the spindle microtubules assembled but exhibited low microtubule numbers and/or abnormal structures (Sunkel et al., 1995;Marshall et al., 1996;Spang et al., 1996). These microtubule assemblies could result either from the presence of residual amounts of γ-tubulin or from alternative cryptic microtubule assembly mechanisms that are unfavourable under normal conditions. Microtubule assembly via a chromosomedependent pathway could account for these observations (for review, see Wittmann et al., 2001). However, no centrosomal mechanism independent of the pericentriolar matrix has yet been demonstrated.
In cultured Drosophila cells, two γ-tubulin isotypes are coexpressed, 23C and 37CD (Rubin et al., 2000;Sunkel et al., 1995;Tavosanis et al., 1997;Raynaud-Messina et al., 2001). These isoforms are 83% identical and differ mainly in their Cterminal regions (Rubin et al., 2000;Sunkel et al., 1995;Tavosanis et al., 1997). The 37CD isoform appears to be present at a low level exclusively in the cytoskeletal fraction. By contrast, the 23C isoform, the major γ-tubulin isotype in these cells, can be detected in the cytosolic fraction as both γ-TuSCs and γ-TuRCs. It is also the only isoform extensively recruited to the mitotic poles, probably accounting for the increase of nucleation activity during mitosis.
In order to investigate the mechanisms involved in spindle microtubule assembly under γ-tubulin-deficient conditions, we depleted Drosophila cells of the 23C γ-tubulin isotype by RNAi. We observed an increase in the mitotic index associated with an accumulation of cells exhibiting abnormal prometaphase stages, largely monopolar and asymmetrical bipolar figures. Most of the monopolar figures exhibited two or more short centrioles at their poles, suggesting that γ-tubulin plays an essential role in centriole morphogenesis and separation. Moreover, in contrast to the metaphases observed in control cells, in which the spindle microtubules emanated from the pericentriolar material, in the γ-tubulin-depleted cells electronic microscopy revealed a centrosomal microtubule assembly pathway involving the elongation of centriolar microtubule triplets. These results indicate that this pathway, which is associated with chromosome-dependent microtubule assembly, could account for some of the spindle organization that occurs under γ-tubulin-deficient conditions.

RNAi
The RNAi treatments were performed in S2 cells (Schneider, 1972) according to Clemens et al. (Clemens et al., 2000). The 23C-γ-tubulinand Cdk8-encoding double-strand RNAs (dsRNAs), corresponding to nucleotides 18-722 and 1-706 relative to their start codons, respectively, were used. For the depletion of 37CD γ-tubulin, two specific dsRNAs were used. First, a dsRNA targeting positions 4-834 (a highly conserved region between the two Drosophila γ-tubulins) depleted the 23C γ-tubulin but not the 37CD isotype, despite a marked reduction in its mRNA level (data not shown). Second, another dsRNA designed to target a divergent region between the two isotypes was directed against the C-terminus of the 37CD γ-tubulin (positions 1062-1276). It did not affect the level of the corresponding protein even after repeated 7-day treatments over a 30-day period.
Briefly, DNA templates were generated by the polymerase chain reaction (PCR). Both PCR primers contained the T7 RNA polymerase minimal promoter sequence followed by a 21 nucleotide sequence complementary to the cDNA. PCR reactions were performed on linearized plasmid templates corresponding to the cDNA clones LD 40196 (23C γ-tubulin), SD 06894 (37CD γ-tubulin) and PBSCDK8 (cdk8) (Leclerc et al., 1996). Purified PCR products (Qiagen, Chatsworth, CA) were transcribed in vitro with the RiboMAX largescale RNA production system T7 (Promega, Madison, WI). RNAs were treated with RQ1 DNase (Promega), sequentially extracted with phenol-chloroform and chloroform, ethanol precipitated, resuspended in diethyl pyrocarbonate (DEPC)-treated water, and then annealed. In all experiments, the depletion efficiency was checked by immunoblotting for γ-tubulin and the expression of the 37CD-γtubulin-encoding mRNA was assessed by reverse-transcription PCR (RT-PCR).

Western-blot analysis
Cell protein extracts were prepared and subjected to western-blot analysis as previously described (Raynaud-Messina et al., 2001). The mouse monoclonal antibody GTU-88 was raised against the amino acid region encompassing residues 38-53 of human γ-tubulin (Sigma-Aldrich, Saint Quentin Fallavier, France). The rabbit polyclonal antibody R70 was raised against the conserved region spanning residues 323-338 of Aspergillus γ-tubulin (Julian et al., 1993) and then affinity purified on recombinant human γ-tubulin. The mouse monoclonal antibody 1501 reacted against actin (Chemicon International, Temecula, CA). An Alpha Imager 1200 (Alpha Innotech, San Leandro, CA) was used for the quantifications.

Immunofluorescence staining
The mouse monoclonal antibody T-5168 (Sigma-Aldrich) was used to stain for α-tubulin. The rabbit antibodies R62 and R46 were used for the detection of the 13 C-terminal amino acids of 23C γ-tubulin and the pre-C-terminal amino acid sequence of 37CD γ-tubulin, respectively (Raynaud-Messina et al., 2001). Antibody Rb188 was used to detect CP190 (Whitfield et al., 1988). The rabbit antibody Rb3133, directed against Asp, was a gift from D. M. Glover (University of Cambridge, UK) (Saunders et al., 1997), the rabbit antibody Rb666 (against Bub1) was obtained from C. E. Sunkel (Bausbaa, Lopes and Sunkel, Universitade do Porto, Portugal; unpublished), the rabbit polyclonal antibody R19 (against centrosomin) was from T. C. Kaufman (Howard Hughes Medical Institute, Indiana University, Bloomington, Indiana, USA) and the rabbit antibody against Cid (for identification of the centromere) was from S. Henikoff (Howard Hughes Medical Institute, Fred Hutchinson Cancer Research Center, Seattle, WA) (Henikoff et al., 2000). The cytoskeletal microtubules were observed in cells permeabilized for 1 minute in PEM (100 mM PIPES, 1 mM EGTA, 2 mM MgCl 2 , pH 6.8) containing 0.5%v/v Triton X-100 and then fixed for 1 hour in PEM containing 1% dimethyl sulfoxide and 3.7%v/v paraformaldehyde, and immunolabelled with the monoclonal antibody T-5168 against αtubulin. For double immunolabelling, cells were fixed for 20 minutes in DES culture medium (Invitrogen, Cergy Pontoise, France) containing 3.7% v/v paraformaldehyde, permeabilized for 2 minutes in methanol (-20°C) and then incubated with the appropriate antibodies. For double labelling of γ-tubulin or Asp with α-tubulin, antibodies against Asp or γ-tubulin were added to the cells, which were incubated for 1 hour at 37°C before the antibodies against αtubulin were added and the cells were incubated for another 1 hour at 37°C, followed by overnight at 4°C. Alternatively, for double labelling of α-tubulin and CP190 (or centrosomin), the antibodies against αtubulin and CP190 (or centrosomin) were added sequentially to the cells, which then were processed with fluorescein isothiocyante (FITC) and tetramethylrhodamine isothiocyanate (TRITC)conjugated secondary antibodies (Santa Cruz Biotechnology, Santa Cruz, CA).
For the renucleation experiments, cells were incubated at 4°C for 2 hours and then allowed to recover by incubating in DES medium at 22°C. After 2 minutes, the cells were rapidly fixed in cold methanol (-20°C, 30 minutes) and double stained with the antibodies against α-tubulin and centrosomin.
Electron microscopy Cells were fixed by successive addition of 0.25 volumes of PHEM (60 mM PIPES, 25 mM HEPES, 10 mM EGTA, 2 mM MgCl 2 , pH 7.0) containing 2.5% glutaraldehyde to 1 volume of culture medium at 0 hours, 3 hours, 6 hours and 9 hours after commencing the procedure. Cells were recovered at 12 hours and processed for electron microscopy. Tannic-acid counterstaining was performed to determine the number of protofilaments (Tilney et al., 1973).

Results
RNAi strongly reduced the level of 23C γ-tubulin in cultured Drosophila cells The levels of 23C and 37CD γ-tubulins were characterized in S2 cell extracts by labelling with the two anti-γ-tubulin antibodies GTU (two exposures were performed) and R70 (Fig. 1A, see controls, lane c). As revealed by these antibodies, 23C γ-tubulin was the major isotype expressed in S2 cells. RNAi directed against 23C γ-tubulin strongly downregulated its expression (Fig. 1A, lanes γ23C). This treatment altered neither the low level of expression of 37CD γ-tubulin (Fig. 1A, second and third panels) nor the level of actin (Fig. 1A, fourth panel) used as a negative control. We reproducibly observed a reduction of at least 90-95% in the 23C γ-tubulin levels (Fig. 1B). This inhibitory effect was stable for at least 9 days. A partial reversal of inhibition was observed at day 14 after treatment (Fig.  1C, lane R14). To control for the specificity of RNAi, cells were subjected to RNAi directed against Cdk8, a transcription factor unrelated to the microtubule cytoskeleton (Barette et al., 2000;Leclerc et al., 1996). Depletion of Cdk8 did not affect the levels of γ-tubulin and actin (Fig. 1D).
Depletion of 23C γ-tubulin led to a mitotic accumulation Downregulation of 23C γ-tubulin did not seem significantly to affect the interphase microtubule cytoskeleton. However, whatever the schedule of treatment, the mitotic indices were significantly higher in 23C-γ-tubulin-depleted cells than in control cells (Table 1). When a single RNAi treatment was given on day 1, we noticed a decrease of the mitotic indices for both the control and the treated cells 7 days after treatment. To check whether this reduction in the mitotic index was related to exhaustion of the medium, we applied two successive RNAi treatments (days 1 and 5), which allowed a refeeding on day 5. The mitotic indices observed on day 7 were significantly increased (Table 1). Using this protocol, we reproducibly observed mitotic indices of 1.5-2.0% in control cells versus 9-13% in treated cells (five independent experiments).

Depletion of 23C γ-tubulin induced abnormal mitotic figures
In order to determine whether the increase in mitotic index was accompanied by an accumulation of cells at a specific stage of mitosis, we characterized the microtubule cytoskeleton by immunofluorescence. We observed mitotic figures without a

(A) Depletion kinetics. Control cells (C) and cells subjected to 23C
γ-tubulin RNAi treatment (γ23C) were harvested after 5 days, 7 days and 9 days of culture. Total-protein extracts (50 µg) were analysed by immunoblotting. From day 5 to day 9, a strong and specific depletion of 23C γ-tubulin was observed. (First and second rows) Monoclonal GTU γtubulin antibody staining. The 37CD γ-tubulin band was not visible when the signal from 23C γ-tubulin was not overexposed (first row) and could be visualized only by reacting blots with more antibody and then overexposing the membranes (second row). (Third row) Purified polyclonal R70 antibody staining. Here, 37CD γ-tubulin also appeared to be less abundant than the 23C γ-tubulin. (Fourth row) The monoclonal antibody mab1501 was used to detect actin as an internal loading control. (B) Quantification of the depletion of γ-tubulin. The γ-tubulin content of 40 µg of an extract from cells submitted to a 7-day RNAi (γ23C) was compared by immunoblotting with that of serial loadings (20-1.25 µg) of protein extracts from control cells (C) and quantified by densitometry analysis. 23C γ-tubulin RNAi reduced the γ-tubulin level by at least 95%. (C) Reversibility of γ-tubulin depletion. Extracts (50 µg), prepared from control cells (C), 7-day-treated cells (γ23C) or from cells that had been treated with the γ23C RNA for 7 days, then washed and maintained in culture for a further 7 days (R7) and 14 days (R14) without RNAi were analysed by immunoblotting with the GTU (against 23C γ-tubulin) and mab1501 (against actin) antibodies. The 23C γ-tubulin signal reappeared 14 days after the removal of the γ23C RNA by washing. (D) Depletion specificity. Extracts (50 µg) from cells submitted to a 7-day RNAi treatment against either 23C γ-tubulin (γ23C) or the transcriptional factor Cdk8 (cdk8) were immunoblotted with the antibodies GTU (against 23C γ-tubulin) and mab1501 (against actin). Depletion of Cdk8 did not affect the levels of 23C γ-tubulin and actin. distinct microtubule polarity, monopolar prometaphases, bipolar prometaphases, elongated mitoses and rare postmetaphase stages (Table 2). In all cases, the mitotic figures were devoid of 23C γ-tubulin signal (Table 3, Fig. 2B-C,F-G, Fig. 3B-E), whereas, in control prometaphase/metaphases, 23C γ-tubulin signal was present at the two poles and in the spindle (98%, n=116, 95% C.I.=96-100; Fig. 2A,A′). Under our experimental conditions, the polar localization of 37CD γtubulin was maintained after 23C γ-tubulin RNAi. In monopolar figures, only the spot corresponding to microtubule focalization was stained, whereas, in bipolar spindles either a single pole or both poles were labelled (Fig. 2L,M). Surprisingly, after 23C γ-tubulin depletion, all mitotic figures lacked astral microtubules (Figs 2, 4).

Non-polar figures
Typical prophases accounted for about 6% of the mitotic stages observed in control cells. These were mostly absent after 23C γ-tubulin depletion. By contrast, we observed figures with no evident microtubule polarity (13%), which corresponded either to late prophase stages or to a highly unstructured spindle (Table 2).

Monopolar spindles
The most common mitotic figures observed after 23C γ-tubulin depletion were monopolar spindles (Table 2, 60%). Some 10-22% of the γ-tubulin-depleted monopolar figures consisted of an aster of microtubules radiating from a central pole towards the chromosomes (Fig. 2C), which formed a rosette or an indistinct mass. The remaining 78-90% constituted a single microtubule conical half-spindle (Fig. 2B,D,E). Both of these monopolar figures corresponded to a prometaphase stage as demonstrated by the presence of fully condensed chromosomes that were either organized as a curved plate or scattered on the spindle ( Fig. 2B-E). The chromatids were unseparated, as revealed by Cid staining (data not shown) and the mitotic checkpoint was activated, as evidenced by kinetochore staining with antibodies against Bub1 (Fig. 2D,E) (Basu et al., 1999).

Bipolar spindles
Another consequence of RNAi was a decrease in the frequency of bipolar mitoses (Table 2,  Elongated spindles Atypical elongated spindles, which were rarely observed in control cells, were significantly increased in cells after γ23Ctubulin depletion (Table 2, 12%). The kinetochores were always stained by Bub1 antibodies, suggesting that the cells failed to enter metaphase ( Fig. 2J-K). The chromosomes were either scattered (Fig. 2J) or tended to accumulate at the two extremities of the spindle (Fig. 2K). Some of these figures could result from either the overall modification of the shape of a bipolar spindle (Fig. 2J) or the elongation of the microtubules of a monopolar spindle (Fig. 2K), as suggested by asymmetrical or symmetrical polar staining (Table 3). In both cases, despite the Bub1 kinetochore labelling being indicative of an active mitotic checkpoint, a process similar to anaphase B occurred, even though anaphase A was not fully completed (Cid staining, not shown).

Late mitotic stages
The proportion of cells able to proceed towards the late stages of mitosis was markedly reduced after RNAi (Table 2, 2% versus 58% in control). These figures could be clearly assigned to anaphase, telophase and cytokinesis (Fig. 3), even if individual lagging chromosomes (Cid staining, data not Journal of Cell Science 117 (23) The mitotic indices (M.I.) [i.e. the proportion of cells undergoing mitosis divided by total cell number (n)] were determined after chromosome staining (with DAPI) and immunolabelling of the microtubule cytoskeleton (with antibodies against α-tubulin). These quantifications were performed in control cells and in cells treated by RNAi against 23C γ-tubulin either once (day 1; M.I. quantifications after 5 days, 7 days and 9 days) or twice (days 1 and 5; M.I. quantification after 7 days). All confidence intervals (in parentheses) were calculated for a probability of 95%. Data from a representative experiment are presented. Four additional experiments giving comparable results were performed.
The ratio of prometaphase figures to anaphases/telophase figures, which ranged from 0.2 to 0.65 in control cells, was increased to between 7 and 46 in treated cells, depending on the experiment (n=5) ( Table 2). Therefore, we conclude that the 6-fold to 6.5-fold increase in the mitotic index (Table 1) resulted both from the higher frequency of unpolarized, monopolar and unfocused bipolar prometaphases, concomitant with a marked decrease in the proportion of cells exiting from mitosis.

Spindle abnormalities in γ-tubulin-depleted cells are associated with polar defects
The absence of a 23C γ-tubulin signal at the spindle poles suggested that the formation of monopolar and bipolar spindles was not directly under the control of the major cellular γtubulin (Table 3; Fig. 2). However, many lines of evidence based on independent approaches suggested that at least some spindle microtubules nucleated from the centrosomes towards the chromosomes. First, as revealed by immunofluorescence and electron microscopy, there were, in addition to the microtubules that interacted with the kinetochores, others that extended away from chromosomes (Fig. 4A, Fig. 5A). Second, in γ-tubulin-depleted cells, the proteins CP190 and centrosomin retained their polar localization (Fig. 4A). Third, renucleation proceeded in RNAi-treated mitotic figures from one or two focalization points (Fig. 4B).
In the prometaphase figures of the control cells, Asp (Saunders et al., 1997), a protein that accumulates at the minus ends of the spindle microtubules (Riparbelli et al., 2002;Wakefield et al., 2001), was localized to the face of the centrosomes that made contact with the spindle microtubules. In the treated cells, Asp was usually clustered around the poles in both mono-and bipolar prometaphase figures, but formed punctate dots (Table 3, Fig. 4Aa-b,g-h). This staining pattern showed that the microtubules retained their normal orientation but were organized as several distinct arrays assembled from the polar regions. This was confirmed using immunostaining for CP190 (Fig. 4Ac-d,i-j) and centrosomin ( Fig. 4Ae-f,k-l) as markers of the pericentriolar material. CP190 relocalizes to the pericentriolar material during mitosis (Whitfield et al., 1988) and is absent from the poles of acentriolar metaphase cells (Debec et al., 1995). A CP190 signal was detected at the pole in nearly all the monopolar figures (Table 3; Fig. 4Ac-d). A variable number of CP190 dots was observed, with a maximum of four dots (Fig. 4Ad). In the bipolar figures, about one-half of the spindles showed CP190 signals at both poles (Fig. 4Ai), whereas, in the other half, CP190 dots were observed in one spindle pole only (Table 3, Fig. 4Aj). Centrosomin is an essential component for centrosome assembly and function (Li and Kaufman, 1996;Megraw et al., 1999). It associates with the pericentriolar matrix during mitosis and is dispersed at interphase (Megraw et al., 2001). As for CP190, centrosomin was recruited as variable number of dots localized to the pole in nearly all monopolar figures (Table 3, Fig. 4Ae-f), whereas, in bipolar prometaphases, it was present at one pole or both poles (Table 3, Fig. 4Ak-l). Because CP190 and centrosomin were recruited to the polar structures, this indicated that mitotic centrosomal maturation had occurred to some extent, even in the absence of a detectable amount of 23C γ-tubulin. To determine whether these centrosomin-stained structures were functional, cold disassembly/reassembly experiments were performed. In control cells, renucleation always occurred from two opposed centrosomin-stained spots, giving rise to a bipolar structure (not shown). In RNAi-treated cells, renucleation proceeded from one or two focalization points (Fig. 4B) distinct from the chromosomes. In all cases, these microtubule- Centrosomin 100% (90-100%) n=415 27% at one pole (19-34%) n=98 55% at one pole (25-57%) n=20 73% at both poles (65-81%) n=98 44% at both poles (43-75%) n=20 Cells were treated twice with RNAi against 23C-γ-tubulin and observed at day 7. The spindles were labelled with antibodies against α-tubulin, whereas the poles were challenged with antibodies against γ-tubulin, Asp, CP190 or centrosomin. All confidence intervals (in parentheses) were calculated for a probability of 95%. organizing centres were decorated by the centrosomin antibody. The morphology of the renucleated microtubule bundles indicated that they could emerge from either initial monopolar or bipolar figures. Hence, it is likely that the mitotic figures generated after RNAi involved polar microtubule assembly, although other nucleation and organization mechanisms that were dependent on the chromosomes and motor proteins could not be excluded. Moreover, the pattern of CP190 and centrosomin staining suggested that centriole duplication had occurred but that the centrioles had failed to separate in the monopolar and at least some of the bipolar figures.
γ-Tubulin is required for efficient centriole separation and morphogenesis The hypotheses we raised after observing the presence of multiple CP190 and centrosomin dots at the poles of most spindles were supported by our electron microscopic observations. In most reconstructions (70%; n=27, 95% C.I.=54-88) obtained from serial thin sections, three or four centrioles were observed at the monopolar spindle poles (Fig.  5B1-4), whereas two centrioles were detected at both poles in bipolar metaphases in control cells. Hence, depletion of 23C γ-tubulin did not prevent the triggering of centriole duplication but did impair their separation in the monopolar prometaphases, perhaps owing to modifications in the dynamics of the microtubules involved in this process. Consistent with the presence of the CP190 signal at both poles in half of the bipolar prometaphases, electron-microscopic observations of serial thin sections showed that centriole separation was possible in this case, although it did not always occur (not shown).
Centrioles exhibit a typical radial structure composed of microtubule triplets or doublets in cross sections, and are unusually short in Drosophila cells (Gonzales et al., 1998). After γ-tubulin depletion, the centrioles were of normal width, excluding the possibility that they corresponded to procentrioles but, surprisingly, most of them were markedly shortened (Fig.  6), suggesting that centriole assembly was incomplete.
Journal of Cell Science 117 (23) Fig. 2. Prometaphase stages observed after 23C γ-tubulin depletion. Three characteristic mitotic figures were observed after 23C-γtubulin depletion: monopolar, bipolar and elongated spindles. Chromosomes were stained with DAPI (blue), microtubules with antibody against α-tubulin (red), whereas 23C-γ-tubulin, Bub1 or 37CD-γ-tubulin immunostaining appeared in green. Scale bar, 5 µm. (A,A′) Control metaphase, showing α-tubulin and 23C γ-tubulin merged staining (A) and γ-tubulin staining only, showing the presence of 23C γ-tubulin at the two poles and in the two halfspindles (A′). All chromosomes were localized to the chromosomal plaque. The two poles exhibited astral microtubules (arrowheads). (B-M) γ-Tubulin-depleted cells. (B-E) Monopolar figures. No γtubulin signal could be detected (B,C). Generally the microtubules radiated preferentially from a single region, giving an overall conical shape. The fully condensed chromosomes constituted a plaque (B) or were dispersed in the spindle (C-E). The shape of the single pole was variable (compare B and C) but always devoid of astral microtubules. In all cases, the kinetochores were labelled by Bub1 antibodies (D,E). (F-I) Bipolar figures. The two opposite poles defined by the microtubules were devoid of astral microtubules and of a 23C-γtubulin signal (F,G). Although the condensed chromosomes could constitute a typical chromosomal plaque (F,I), more often, some chromosomes were not aligned on the plaque (H) or were even scattered over the spindle. Their kinetochores were always labelled by Bub1 antibodies (H,I). (J,K) Elongated figures. The chromosomes were either scattered (J) or unequally distributed at each pole (K). In all cases, the kinetochores were labelled by Bub1 antibodies. (L,M) 37CD-γ-Tubulin staining, which was present at the pole(s) of formed spindles.

Elongation of centriolar microtubules contributes to spindle formation
In the monopolar metaphases, the mitotic microtubules were organized around a pole characterized by the presence of supernumerary centrioles. However, in contrast to control cells, the microtubules did not seem to be nucleated from the pericentriolar material. Electron microscopic observations showed that most of them appeared to be associated with one extremity of the centriole cylinder and in apparent continuity with the centriolar microtubules ( Fig. 5A-C, arrowheads). The orientation of the centriolar microtubules in three-dimensional reconstructions showed that the spindle microtubules were assembled from the distal end of the centriole. Most of the Fig. 3. Post-metaphase figures observed after 23C-γ-tubulin depletion. Chromosomes appear in blue, microtubules in red, and the staining in green corresponds to γ-tubulin. (A,A′) Control cytokinesis showing α-tubulin and γ-tubulin merged staining (A) and γ-tubulin staining only, showing the presence of 23C γ-tubulin at the two poles and in the mid-body (A′). (B-E) Anaphase (B), telophase (C) and cytokinesis (D,E) after γ-tubulin depletion. None of these figures exhibited γ-tubulin signals (absence of green staining). Cytokinesis was generally symmetrical (D) and exceptionally asymmetrical (E). Scale bar, 5 µm.  Asp (a,b,g,h), CP190 (c,d,i,j) or centrosomin (e,f,k,l). (Aa-f) Monopolar figures. Asp labelling was always exclusively localized to the single pole in several dots corresponding in most cases to sub-spindles (a,b). CP190 labelling was concentrated at the pole as between one and four dots (d). Centrosomin (CNN) was also concentrated at the pole, usually as a large and irregular structure (e,f). (Ag-l) Bipolar figures. Asp was always localized to the two poles, where, in most cases, it constituted a large area composed of two to five dots associated with each subspindle (g,h). CP190 and centrosomin labelling were present as one (i) or several (k) dots located at both poles or at one pole (j,l). (B) Renucleation in γ-tubulin-depleted cells. Microtubule arrays (red) renucleated from one or two centrosomin-labelled spots (green) after cold depolymerization. Scale bar, 5 µm.  mitotic microtubules exhibited 13 protofilament structure, with sections revealing 1, 14 and 2 microtubules with 12, 13 and 14 protofilaments, respectively (data not shown). Moreover, this unexpected assembly pathway and the inability of the pericentriolar material to nucleate microtubules could account for the absence of astral microtubules (Figs 2, 4).

Discussion
We have used RNAi to deplete the major γ-tubulin isotype significantly in cultured Drosophila S2 cells. Only a few cells escaped this treatment, because less than 0.7% of all mitotic cells presented a bipolar morphology with a positive 23C γtubulin signal at both poles. Depletion of 23C γ-tubulin caused S2 cells to accumulate in mitosis, with an active mitotic checkpoint. They exhibited abnormal prometaphase stages, characterized by unseparated condensed chromatids and spindle poles with supernumerary short centrioles. The emergence of a characteristic phenotype (i.e. monopolar prometaphase figures) prompted us to investigate the mechanism of spindle organization in 23C-γ-tubulin-depleted cells. Our investigations revealed that, under γ-tubulindeficient conditions, the pericentriolar material was unable efficiently to promote microtubule nucleation. Our observations point to the involvement of a previously undescribed centrosomal mechanism of microtubule assembly that operates via elongation of the distal centriolar extremities.
After depletion of 23C γ-tubulin, the duplicated centrioles were present at the one pole of the monopolar prometaphase figures and at only one of the two poles in at least half of the bipolar figures, indicating a segregation failure. This phenomenon has also been suggested to occur in Drosophila Dgrip91 mutants (Barbosa et al., 2000). Observations made on mammalian cells have demonstrated that the relative positions of the two centriole pairs depend partly on the dynamics of the microtubule cytoskeleton . Hence, after γtubulin depletion, it is likely that the microtubule subnetwork involved in the separation of the centriole pairs underwent modifications in either its dynamics or in its density. Despite the lack of ultrastructural studies excluding morphogenetic and/or separation defects, basal-body duplication has been reported to be impaired after silencing of the γ-tubulinencoding genes in Paramecium (Ruiz et al., 1999). This discrepancy with our observations could result from differences in the timing of centriole and basal-body duplication, and the many rounds of basal-body duplication in Paramecium leading to many basal bodies. Thus, it is possible that the critical concentration of γ-tubulin required for centriole duplication is lower than the one necessary for microtubule assembly by the pericentriolar material. In this regard, it is possible that the small amount of the 37CD γ-tubulin isotype constitutively associated with the centrosomal fraction plays a role in centriole duplication. However, it is unlikely that this low-abundance γ-tubulin isotype plays an important role in the nucleation of mitotic microtubules, as suggested by the absence of microtubules issuing directly from the pericentriolar material. We showed that, after 23C γ-tubulin depletion, the 37CD isotype was not affected; its expression level was maintained and it retained its polar localization. Moreover, under the experimental conditions used, the function of this minor γ-tubulin isotype could not be determined by RNAi, despite the use of two distinct dsRNAs. The poor susceptibility of the 37CD γ-tubulin isoform to the RNAi strategy might stem from its specific properties in Drosophila cells, such as its low concentration (Raynaud-Messina et al., 2001), its absence from the cytosolic fraction (Raynaud-Messina et al., 2001) and/or a possible slow turnover.
In 23C γ-tubulin-depleted S2 cells, the centrioles were often shorter than in the control cells. The presence of a short centriole among normal sized centrioles has been noticed in abnormal metaphases of neuroblasts from the Drosophila Dgrip91 mutants (Barbosa et al., 2000). Centrioles are generally believed to be very stable organelles. However, in Physarum amoebae, a transient decrease in the length of the parental centrioles occurs in mitosis, during the assembly of the daughter centrioles (Gely and Wright, 1985). If the γtubulin present in the centriole and basal-body lumen (Fuller et al., 1995;Klotz et al., 2003;Moudjou et al., 1996) interacts with the centriolar microtubules, its depletion might well alter centriole morphogenesis, possibly by an action on centriolar microtubule dynamics. Indeed, a role of γ-tubulin in microtubule dynamics and stability has been suggested in Aspergillus (Oakley and Oakley, 1989), Schizosaccharomyces (Paluh et al., 2000), Caenorhabditis (Strome et al., 2001), as well as in mammalian cells, where γ-tubulin interacts with the stable kinetochore microtubules (Lajoie-Mazenc et al., 1994).
γ-Tubulin-depleted Drosophila S2 cells mostly assembled half-spindles, resulting in monopolar metaphases with a normal overall microtubule orientation and a 13-protofilament structure. The presence of pericentriolar material was confirmed by electron microscopy and by the recruitment of the centrosomal proteins CP190 and centrosomin, which appeared to be independent of the presence of γ-tubulin (Fig.  4A). Microtubule reassembly after cold treatment suggested that the pole retained a nucleation activity (Fig. 4B). However, consistent with the severe depletion of the main centrosomal γ-tubulin, most of the spindle microtubules were not assembled from the pericentriolar material, which accounted for the absence of astral microtubules. Nucleation resulted at least partially from elongation of the distal extremity of the microtubules of the four centrioles. This process differs from the formation of primary cilia observed in several interphase animal cells (Wheatley, 1982). In mitotic 23C-γ-tubulindepleted S2 cells, the microtubules produced by the centrioles failed to form the characteristic axonemal structure. They were not surrounded by a membrane and were not associated in microtubule doublets as in typical primary cilia (Wheatley, 1982) but rather diverged in the cytoplasm, giving rise to a few sub-spindles. Moreover, they did not arise from the dispersion of the ensheathing membrane of a primary cilium, as has been reported in cells undergoing early mitosis (Rieder et al., 1979), because primary cilia were not observed in either control or treated interphase cells. It is likely that the distal extremity of the centriolar microtubules was not blocked by regulatory proteins and could assemble α/β-tubulin heterodimers, as observed in vitro with isolated centrioles (Gould and Borisy, 1977). Unless triggered by the specific axonemal program, during normal mitosis, this elongation process remains exceptional (Krishan and Buck, 1965). Our observations suggest that this atypical elongation pathway is normally overridden by the efficiency of nucleation from the high number of γ-tubulin nucleation sites present in the pericentriolar material. The failure of centriole separation and elongation of the centriolar microtubules we observed is consistent with the formation of monopolar spindles. However, this assembly pathway might only give rise to a limited number of spindle microtubules, unless microtubules detach from the distal ends of the centriole. Alternatively, the presence of additional microtubule bundles on the chromosome face opposite the main pole (Fig. 4G) suggests that microtubules might also be nucleated at the kinetochores. In contrast to the microtubule assembly pathway predominantly mediated by the pericentriolar γ-tubulin (Hannak et al., 2002), our results suggest that, below a critical concentration of γ-tubulin, other assembly mechanisms could contribute to spindle formation. These alternative mechanisms were unable to generate a fully functional mitotic apparatus in most cases.
Taken together, our results demonstrate the inefficiency of the pericentriolar material in promoting microtubule nucleation after depletion of γ-tubulin. They point to an alternative centrosome-driven mechanism of microtubule assembly that might contribute to the formation of abnormal, albeit partially functional, mitotic spindles.