Ninein is released from the centrosome and moves bi-directionally along microtubules.

Cell-to-cell contact and polarisation of epithelial cells involve a major reorganisation of the microtubules and centrosomal components. The radial microtubule organisation is lost and an apico-basal array develops that is no longer anchored at the centrosome. This involves not only the relocation of microtubules but also of centrosomal anchoring proteins to apical non-centrosomal sites. The relocation of microtubule minus-end-anchoring proteins such as ninein to the apical sites is likely to be essential for the assembly and stabilisation of the apico-basal arrays in polarised epithelial cells. In this study, we establish that ninein is highly dynamic and that, in epithelial cells, it is present not only at the centrosome but also in the cytoplasm as distinct speckles. Live-cell imaging reveals that GFP-ninein speckles are released from the centrosome and move in a microtubule-dependent manner within the cytoplasm and thus establishes that epithelial cells possess the mechanical means for relocation of ninein to non-centrosomal anchoring sites. We also provide evidence for the deployment of ninein speckles to apical anchoring sites during epithelial differentiation in both an in situ tissue and an in vitro culture system. In addition, the findings suggest that the non-centrosomal microtubule anchoring sites associate with adherens junctions in polarised epithelial cells.


Introduction
Differentiation and polarisation of epithelial cells involve a dramatic reorganisation of the microtubule cytoskeleton and centrosomal components. The centrioles migrate to the apical cell surface, the radial microtubule array is largely lost and an apico-basal array develops that is no longer anchored at the centrosome (Bacallao et al., 1989;Bre et al., 1987;Bre et al., 1990;Meads and Schroer, 1995;Tucker et al., 1992). We have previously proposed that a release-and-capture mechanism is responsible for the generation of non-centrosomal apico-basal arrays in most centrosome-containing polarised epithelial cells. This not only involves centrosomal nucleation, release and repositioning of microtubules but also the relocation of centrosomal anchoring proteins such as ninein to apical noncentrosomal anchoring sites (Mogensen, 2004;Mogensen, 1999;Mogensen et al., 2000).
Ninein is a coiled-coil centrosomal protein that has a major role in microtubule minus-end anchorage and also acts as a docking site for ␥-tubulin-containing complexes (Delgehyr et al., 2005;Stillwell et al., 2004). The affinity of ninein for the minus-end of microtubules is highlighted by its localisation revealed by immuno-electron microscopy to the tips of the subdistal appendages of the mother centriole, which support the minus-ends of the core radial microtubules, and to the minus-ends of both centrioles, where it colocalises with C-Nap1 (Mogensen et al., 2000;Ou et al., 2002). The role of ninein in anchorage is strongly supported by overexpression and depletion studies (Abal et al., 2002;Dammermann and Merdes, 2002). In addition, both centrioles nucleate similar numbers of microtubules by means of their ␥-tubulinassociated complexes following nocodazole removal, but only the mother centriole, which possesses a higher concentration of ninein than the daughter, is able to maintain a radial array (Mogensen et al., 2000;Piel et al., 2000).

Cell-to-cell contact and polarisation of epithelial cells involve a major reorganisation of the microtubules and centrosomal
components. The radial microtubule organisation is lost and an apico-basal array develops that is no longer anchored at the centrosome. This involves not only the relocation of microtubules but also of centrosomal anchoring proteins to apical non-centrosomal sites. The relocation of microtubule minus-end-anchoring proteins such as ninein to the apical sites is likely to be essential for the assembly and stabilisation of the apico-basal arrays in polarised epithelial cells. In this study, we establish that ninein is highly dynamic and that, in epithelial cells, it is present not only at the centrosome but also in the cytoplasm as distinct speckles. Live-cell imaging reveals that GFP-ninein speckles are released from the centrosome and move in a microtubule-dependent manner within the cytoplasm and thus establishes that epithelial cells possess the mechanical means for relocation of ninein to noncentrosomal anchoring sites. We also provide evidence for the deployment of ninein speckles to apical anchoring sites during epithelial differentiation in both an in situ tissue and an in vitro culture system. In addition, the findings suggest that the non-centrosomal microtubule anchoring sites associate with adherens junctions in polarised epithelial cells.
Release and relocation of microtubule minus-end anchoring proteins such as ninein are likely to be central to the assembly and stabilisation of the apico-basal arrays in polarised epithelial cells. Highly polarised epithelial cells in the mammalian inner ear have proved ideal for investigating microtubule minus-end anchoring proteins and their deployment during assembly of the apico-basal arrays as nucleation and anchorage are effected at separate sites (Mogensen et al., 1997). This is particularly the case for the supporting pillar cells of the organ of Corti that assemble large non-centrosomal apico-basal arrays of several thousand microtubules during tissue morphogenesis. In these cells, ␥tubulin is concentrated at the centrosome where nucleation occurs, whereas the vast majority of the microtubule minusends are anchored at ninein-containing non-centrosomal apical sites (Mogensen et al., 2000;Mogensen et al., 2002). Ninein redistribution during development of the apico-basal arrays in the pillar cells suggests that ninein relocates to apical anchoring sites in a microtubule-dependent manner (Mogensen et al., 2000).
The aim of this study was to analyse the dynamics of ninein movement in epithelial cells in order to establish a model for its relocation to apical non-centrosomal anchoring sites during epithelial polarisation. Using high-resolution confocal microscopy, we show that the cytoplasmic ninein pool in cochlear pillar cells contains distinct ninein speckles, which are redeployed to apical non-centrosomal anchorage sites. Furthermore, live-cell imaging of ninein tagged with green fluorescent protein (GFP) and immuno-electron microscopy studies provide evidence for the release of ninein from the centrosome and for its subsequent microtubule-dependent motile behaviour within the cytoplasm. In addition, fluorescence recovery after photobleaching (FRAP) experiments suggest that a dynamic exchange exists between the centrosomal and cytoplasmic pools of ninein and that this movement is dependent on microtubules and not actin filaments.

Results
Cytoplasmic ninein speckles relocate to apical non-centrosomal anchoring sites during inner ear epithelial development High-resolution confocal microscopy was used to further analyse cytoplasmic ninein at different stages during assembly of the apico-basal array in inner pillar cells of the cochlea. Inner pillar cells assemble apico-basal arrays consisting of some 3000 microtubules anchored at apical noncentrosomal peripheral sites during a 6-day postbirth period in the mouse (Fig. 1A). ␥-Tubulin is concentrated at the apical centrosome throughout development and has not been detected at the apical anchoring sites, where most of the microtubule minus-ends are concentrated (Fig.  1F). In this study, we have identified a transit population of distinct non-centrosomal ninein speckles in the inner pillar cells, whose location varied with the stage of development. Ninein was concentrated at the centrosome, with only a few cytoplasmic speckles evident in the apical half of the cells, during the initial stages of assembly of the apicobasal array, when most of the microtubules are still concentrated at the centrosome ( Fig. 1B) (Mogensen et al., 2000 and Fig. 4 within). However, as microtubule assembly progressed, ninein speckles became widespread throughout the apical region, and speckles started to accumulate in an apical peripheral ring by day 2-3 after birth (Fig. 1C). A distinct peripheral ring of ninein aggregates was evident by day 4-6 when the apico-basal microtubule array was close to completion and the vast majority of the minus-ends were anchored at the apical peripheral sites (Fig. 1D,E). This suggested that ninein speckles are released from the centrosome and translocated to the non-centrosomal anchoring sites. We therefore decided to investigate the dynamics of ninein and the role of microtubules in ninein speckle motility in cultured epithelial cells.
Cytoplasmic ninein speckles associate with microtubules in cultured epithelial cells Two affinity-purified rabbit polyclonal antibodies were used to localise ninein in various cultured epithelial cells. In agreement with previous reports, the ninein antibodies were found to label the centrosome in all cell types studied, including UE-1 (Lawlor et al., 1999), PtK2, MDCKII, Caco2, FHL124 (Reddan et al., 1999), hTERT RPE (Rambhatla et al., 2002) and ARPE-19 (Dunn et al., 1996;Bouckson-Castaing et al., 1996;Mogensen et al., 2000;Piel et al., 2000) (Fig. 2A). Interestingly, ninein was also evident as distinct speckles throughout the cytoplasm in all these cells ( Fig. 2A,B,E and data not shown). The authenticity of these cytoplasmic ninein speckles was verified by using siRNA ninein depletion in ARPE-19 cells; this resulted in loss of speckles as well as loss of centrosomal labelling ( Fig. 2C,D). In all cells studied, widefield fluorescent and confocal immuno-localisation of ninein and microtubules suggested that the cytoplasmic ninein speckles associated with microtubules. Distinct ninein speckles were evident both along the length, and at the end, of microtubules, including at the ends of free microtubules (Fig. 2B,. In order to verify that ninein speckles did colocalise with microtubules, the percentage of speckles observed on microtubules was calculated before and after a random shift of Journal of Cell Science 120 (17) the microtubule image with respect to the speckle image in cells double labelled for ninein and microtubules. The percentage of speckles observed on microtubules was 81.14% (±2.66%, n=37) before image shift and 44.36% (±4.3%, n=37) after image shift. This difference is highly significant (P<0.001), and it is therefore unlikely that the ninein speckle colocalisation with microtubules was coincidental. However, some ninein speckles were evidently not associated with microtubules (Fig. 2E).
Immuno-gold labelling confirmed the centrosomal localisation of ninein and its specificity for the centriolar appendages (Fig. 6B), as previously documented (Mogensen et al., 2000). In addition, gold particles were also observed along microtubules in groups of two to four or as single particles (Fig.  2I,J). Control samples, where the primary antibody had been omitted, showed no specific labelling (data not shown). This further supported the idea that ninein speckles associate with microtubules.
Ninein speckles depend on microtubules for their distribution throughout the cytoplasm A nocodazole assay was used to determine whether the distribution of ninein speckles in the cytoplasm was dependent on microtubules. Nocodazole caused a marked centralisation of the ninein speckles in most cells, with very few speckles remaining in the peripheral region (Fig. 3A). Some speckles were associated with nocodazole-resistant microtubules, but most were not (Fig. 3B). In order to determine whether the speckles were free in the cytosol or attached to other cytoskeletal filaments, nocodazole-treated PtK2 cells were preextracted with a detergent buffer (see Materials and Methods) prior to fixation and staining for ninein. This resulted in a similar centralised distribution of ninein speckles as observed for cells treated with nocodazole without detergent extraction (compare Fig. 3A with 3B). By contrast, cells treated with 2 M latrunculin B to depolymerise the actin cytoskeleton revealed no apparent change in ninein distribution (data not shown). Nocodazole removal and microtubule regrowth led to the ninein speckles regaining their full distribution similar to that observed in untreated cells (Fig. 3D,E).
The intermediate filament network is linked to both microtubules and actin filaments and is noted for collapsing around the nucleus and withdrawing from the periphery in the absence of microtubules (Gard et al., 1997;Klymkowsky, 1982). Nocodazole treatment of PtK2 cells did cause a collapse of the keratin network in many cells. Double labelling for keratin 18 and ninein demonstrated a correlation between a central concentration of ninein and a collapsed keratin cytoskeleton in nocodazole-treated cells (Fig. 3C). In addition, ninein speckles appeared to decorate the keratin filaments ( Fig.  3H). Double labelling for ninein and keratin 18 in untreated cells also suggested that some ninein speckles associated with intermediate filaments in PtK2 cells (Fig. 3F). Furthermore, immuno-gold labelling for ninein revealed gold particles along filaments of ~15 nm diameter in the electron microscope, suggestive of there being an association between ninein and intermediate filaments (Fig. 3G).
These results suggest that ninein speckle distribution throughout the cytoplasm is dependent on microtubules rather than actin filaments and that some speckles associate with the intermediate filament cytoskeleton. Cytoplasmic ninein speckles are highly dynamic and show short-and long-range movements to and from the centrosome Time-lapse recordings of PtK2, ARPE-19 and MDCKII cells expressing GFP-tagged ninein were used to visualise and investigate the dynamics of ninein. The GFP-ninein construct has been previously characterised (Abal et al., 2002). GFPninein was concentrated at the centrosome, and distinct GFPninein speckles were also evident in the cytoplasm, as observed for endogenous ninein in untransfected epithelial cells. Timelapse imaging revealed that the cytoplasmic ninein speckles were highly dynamic, moving bi-directionally to and from the centrosome (Figs 4, 5; supplementary material Movies 1 and 2). Analyses of time-lapse recordings and image frames strongly suggested that some ninein speckles were released from the centrosome (Fig. 4D,E; supplementary material Movies 2-4), whereas others moved towards the centrosome ( Fig.  5; supplementary material Movie 2). Two types of movement were also apparent: short and long range. The short-range movements were confined to a region relatively close to the centrosome (2-3 m radius). This region was packed with highly dynamic ninein moving to and from the centrosome ( Fig. 4A; supplementary material Movie 1). The GFP-ninein in this region appeared more string-like than that further from the centrosome (Fig. 4B,D; supplementary material Movie 4) and this was also confirmed for endogenous ninein by immunolabelling of non-transfected cells (Fig. 4C). It is therefore unlikely to be an artefact of the GFPninein. The long-range speckle movements were also bi-directional, and speckles could be observed throughout the cytoplasm (Fig. 4E,F, Fig. 5A,B, supplementary material Movie 2). Average speckle velocities of 0.53 m/second (±0.06 m/second, n=27) away from the centrosome and 0.37 m/second (±0.05 m/second, n=23) towards the centrosome were apparent in PtK2 cells expressing GFP-ninein (see Materials and Methods). Tracking several ninein speckles revealed that they often used the same path ( Fig. 6A,AЈ). Overall, analyses of speckle movements and their velocities suggested that ninein speckles were released from the centrosome and transported along microtubules by motor proteins. This was further supported by doubletransfection studies using GFP-ninein and YFP-␣tubulin, which enabled visualisation of ninein speckle release from the centrosome and subsequent movement along microtubules ( Fig. 6C; supplementary material Movie 5). In addition, immuno-gold localisation of ninein in the vicinity of the centrosome revealed gold particles aligned along centrosome-anchored microtubules (Fig. 6B).
Ninein speckle dynamics is inhibited by nocodazole but not latrunculin B or inhibitors of GSK3␤ The data presented above strongly suggested that ninein speckle dynamics was dependent on microtubules. In order to confirm this, a nocodazole assay was used to depolymerise the microtubules in GFP-ninein-transfected PtK2 cells. All but a few stable microtubules remained following treatment with 10 g/ml nocodazole, but full recovery occurred within ~10 minutes following nocodazole We also tested whether the actin cytoskeleton had any significant influence. Time-lapse imaging of cells treated with 2 M latrunculin B showed no significant effect on GFP-ninein speckle dynamics (Fig. S1A in supplementary material). Movements away from the centrosome produced an average Journal of Cell Science 120 (17) velocity of 0.58 m/second (±0.06 m/second, n=36) and towards the centrosome of 0.43 m/second (±0.04 m/second, n=35) in cells treated with latrunculin B (Fig. 7B). Fixation of cells post imaging and labelling for microtubules and actin filaments confirmed a lack of actin filaments, whereas the radial microtubule organisation was maintained ( Fig. S1B in  supplementary material).
The possible involvement of glycogen synthase kinase 3␤ (GSK3␤) in ninein speckle release and dynamics was also investigated. Phosphorylation by polo-like kinase 1 (Plk1) has been shown to be responsible for the release of nineinlike protein (Nlp) from the centrosome but this is not the case for ninein (Casenghi et al., 2005). However, GSK3␤, which is present at the centrosome, has been reported to phosphorylate ninein (Hong et al., 2000a;Hong et al., 2000b;Howng et al., 2004) and could potentially influence its association with the centrosome. We therefore investigated the affect of GSK3␤ inhibition on ninein release from the centrosome. GFP-ninein-expressing cells were treated with one of three different inhibitors of GSK3␤ (LiCl, SB415286, SB216763; see Materials and Methods) prior to time-lapse imaging. None of the three inhibitors of GSK3␤ appeared to affect ninein release or cause any significant reduction in its dynamics, suggesting that GSK3␤ is not responsible for ninein release from the centrosome (Fig. S2 in supplementary  material).

FRAP analyses suggest microtubule-dependent centrosomal GFP-ninein recovery
The time-lapse studies indicated that ninein speckles were actively transported to the centrosome along microtubules ( Fig.  5; see supplementary material Movie 2). We decided to explore further the recruitment of ninein to the centrosome by using FRAP. FRAP analyses were carried out in PtK2 cells expressing GFP-ninein by using a Zeiss LSM 510 META confocal scanning microscope, and the centrosome was selected for photobleaching (see Materials and Methods). In the presence of an intact microtubule cytoskeleton, almost complete (average 98.48±1.52%, n=5) fluorescent recovery of centrosomal ninein occurred within 2.5 minutes following photobleaching (Fig. 8A,C,D). However, FRAP analysis of centrosomes in nocodazole-treated cells showed little or no recovery (average 23.83±3.68%, n=6) after 4 minutes or even after extended time periods of up to 10 minutes (Fig. 8B-D). These FRAP studies suggest that there is a dynamic exchange between the cytoplasmic and centrosomal pools of ninein and that microtubules are required for efficient ninein transport to the centrosome.
Ninein speckles localise to apical anchoring sites in polarised MDCKII cells Ninein deployment was also investigated in polarised MDCKII cells; ninein was present at the centrosome and as speckles in the cytoplasm, as also observed in non-polarised epithelial cells. However, in single optical sections through the apical region of polarised cells, a distinct punctate pattern of ninein was evident at the cell periphery (Fig. 9A). A marked accumulation of ninein within the apical cell peripheral region became apparent in lateral and apical views of 3D reconstructions of polarised MDCKII cells (Fig. 9B,G,H). Polarised MDCKII cells, like the pillar cells, contain an apico- basal array of microtubules, with most of the minus-ends anchored at noncentrosomal apical sites (Bacallao et al., 1989;Meads and Schroer, 1995;Musch, 2004). 3D reconstructions of the microtubule organisation in polarised MDCKII cells revealed that most of the apico-basal microtubules form a tube, with the minus-ends organised in a peripheral ring at the apex (data not shown). So, even though there are significantly (~tenfold) fewer microtubules in polarised MDCKII cells compared with inner pillar cells, a peripheral ninein ring was again evident at the non-centrosomal microtubule minus-end anchoring sites, as observed in the pillar cells (Fig. 9H).
The peripheral location of the ninein ring suggested that it was within the domain of adherens junctions. E-cadherin is responsible for cell-cell adhesion, and its cytoplasmic tail interacts with ␤-catenin. Double labelling for ninein and either E-cadherin or ␤-catenin revealed no significant colocalisation with E-cadherin but considerable colocalisation at the cell periphery with ␤-catenin ( Fig. 9C; see also Fig. S3 in supplementary material). Colocalisation of ninein and ␤-catenin was confirmed using both the Zeiss and Volocity classification programs (Fig. 9D-F). Similar colocalisation was also observed in supporting Kollikers cells of the cochlea (data not shown). However, a direct physical interaction between ninein and ␤catenin could not be confirmed by coimmunoprecipitation.

Discussion
The centrosomal protein ninein acts as a microtubule minus-end anchoring protein at the centrosome in cells deploying a radial microtubule array and at apical sites in polarised epithelial cells, which assemble non-centrosomal apicobasal arrays. Functional inhibition studies including microinjection of antibodies against ninein (Ou et al., 2002) and siRNA depletion (Dammermann and Merdes, 2002), which result in loss of centrosomal focus, and ninein overexpression, which enhances centrosomal anchorage (Abal et al., 2002), all provide strong evidence for a role for ninein in anchorage. However, little is known about the mechanisms responsible for ninein relocation during epithelial differentiation, and yet ninein is clearly central to microtubule reorganisation in polarised epithelial cells.
Ninein speckles associate with microtubules and intermediate filaments and require microtubules for their distribution throughout the cytoplasm Ninein was observed not only at the centrosome but also within the cytoplasm as distinct speckles in several epithelial cell types, suggesting that these speckles are a general feature of epithelial cells. Cytoplasmic diffuse ninein has previously been observed in developing cochlear epithelial cells (Mogensen et  al., 2000), and ninein speckles have been reported in neurons (Baird et al., 2004). These are both systems that assemble large non-centrosomal microtubule arrays. However, what role the ninein speckles serve in undifferentiated epithelial cells and neurons is currently not known. Interestingly, ninein was observed at the ends of free microtubules in some epithelial cells in this study, and it is tempting to speculate that it could have a role in minus-end capping and stabilisation of noncentrosomal microtubules. This is supported by microinjection of antibodies against ninein, which not only caused the release of microtubules but also their subsequent depolymerisation (D.K.M., J.M.C. and M.M.M., unpublished). It is also of note that ninein speckles have been reported to be highly abundant in neurons, which contain predominantly non-centrosomal microtubules (Baird et al., 2004). Alternatively or additionally, these speckles might represent a cytoplasmic store in dynamic exchange with the centrosome, as found for other core centrosomal components including ␥-tubulin and centrin Paoletti et al., 1996). This is supported Journal of Cell Science 120 (17) by our FRAP data. The cytoplasmic ninein speckles might also represent an intermediate stage in the redistribution process, which requires final signals from developing cell junctions in order to progress to the anchoring sites. For example, in suprabasal cells of the epidermis, desmoplakin is required for the recruitment of ninein to desmosomes, with ninein remaining centrosomal and cytoplasmic in desmoplakin knockouts (Lechler and Fuchs, 2007).
Immuno-fluorescence and immuno-electron microscopy as well as drug (nocodazole and latrunculin B) treatments strongly suggest that ninein speckles associate with and depend on microtubules for their distribution throughout the cytoplasm. Ninein speckles were observed to decorate microtubules along their length in all the cells studied. An association with keratin filaments was also observed in PtK2 cells, whereas there was no apparent association with actin filaments. So far, no microtubulebinding domain has been identified for ninein, but an interaction with the p150 glued , p50 and Arp1 subunits of the dynactin complex has been reported (Casenghi et al., 2005). Ninein speckles could therefore bind to microtubules via the dyneindynactin complex. Similarly, ninein speckles could bind indirectly to intermediate filaments by means of dynein, which has also been shown to associate with intermediate filaments (Helfand et al., 2002).
Nocodazole induced a centripetal localisation of the ninein speckles in many cells, but microtubule regrowth reestablished their distribution throughout the cytoplasm. Latrunculin B, by contrast, caused no apparent change in ninein distribution. Interestingly, the few nocodazole-treated cells that still showed speckles throughout the cytoplasm retained a fully extended keratin network. The centralisation of the speckles is thus likely to be due to ninein speckle association with a peripherally withdrawn and collapsed keratin filament network. Intermediate filaments are linked to both microtubules and actin filaments and are particularly dependent on microtubules for their distribution (Gard et al., 1997;Klymkowsky, 1982). Microtubules are clearly required for ninein speckle distribution throughout the cytoplasm, but intermediate filaments might also influence their deployment.
Ninein speckles are released from the centrosome and move bi-directionally along microtubules These studies provide visual evidence for the release of ninein speckles from the centrosome and for their subsequent bidirectional movements within the cytoplasm. Two types of movements were observed, one short range, pericentrosomal with string-like ninein, and the other long range, involving extensive movements to and from the centrosome by distinct speckles. The string-like or short filaments of ninein observed close to the centrosome might represent an intermediate stage in the ninein release process. Interestingly, ultrastructural analyses of centrosomal ninein aggregates have revealed a network of filaments and nodes (Abal et al., 2002). How ninein speckles are released from the centrosome is not known, but it is likely that phosphorylation is involved. For example, phosphorylation of Nlp by Plk1 is responsible for its displacement from the centrosome at G 2 -M phase (Casenghi et al., 2005). Plk1 does not phosphorylate ninein (Casenghi et al., 2005), but GSK3␤, which is present at the centrosome, interacts with the C-terminus of ninein and phosphorylates ninein (Hong et   2004). However, three different inhibitors of GSK3␤ used in this study did not cause any significant reduction in ninein dynamics or release from the centrosome, suggesting that GSK3␤ is not responsible for ninein release.
Several lines of evidence provide strong support for ninein speckle dynamics being dependent on microtubules. Analyses of GFP-ninein speckle release and movements from the centrosome showed several speckles following the same tracks, and double-transfection studies with GFP-ninein and YFP-␣tubulin suggested that the speckles moved along microtubules. In addition, immuno-electron microscopy localised ninein to centrosomally anchored microtubules. Furthermore, nocodazole greatly reduced all speckle movements, whereas latrunculin B had no significant affect. The bi-directional speckle movements along microtubules are most likely mediated by motor proteins. The estimated average ninein speckle velocity away from the centrosome was 0.53 m/second in PtK2 cells, which is within the range of kinesin motors (0.1-1.5 m/second). However, which kinesin motor is responsible for ninein speckle movement towards the cell periphery is unknown but warrants future investigation. However, the estimated average speckle velocity for movements towards the centrosome was 0.37 m/second in PtK2 cells, which is less than that expected for dynein (0.5-2.0 m/second). Analyses of speckle dynamics in the PtK2 cells did include speckles displaying brief periods of pausing, and it is likely that these pauses account for the relatively low average speckle velocity. Furthermore, velocity calculations based only on moving speckles (i.e. excluding pauses) in MDCKII cells were 0.81 m/second away from and 0.83 m/second towards the centrosome (Fig. S2), which falls within the expected ranges for kinesin and dynein.
The observation of bi-directional movements of ninein speckles to and from the centrosome suggested an exchange between centrosomal and cytoplasmic ninein. This was supported by the FRAP studies, which revealed that 60-90% of ninein was exchanged within 2.5 minutes. Pre-treatment with nocodazole resulted in no significant recovery following photobleaching, suggesting that ninein transport to the centrosome is dependent on microtubules. Dynein has previously been shown to be required for centrosomal targeting of ninein, with dynein inhibition by excess of either p50 (dynamitin/dynactin subunit) or p150 Glued CC1 causing centrosomal depletion of ninein (Casenghi et al., 2005;Dammermann and Merdes, 2002). We have also shown that overexpression of p150 Glued CC1 causes ninein depletion at the centrosome as well as disorganised microtubules lacking centrosomal focus (data not shown). PCM-1 and bicaudal-D (BICD) have both been reported to be involved in the transport of ninein to the centrosome in a dynein-dependent manner (Dammermann and Merdes, 2002;Fumoto et al., 2007). However, the reported interaction between ninein and the dynein-dynactin complex suggests that at least a fraction of ninein moves independently of PCM-1 and BICD.
Most importantly, these studies of dynamics provide evidence for epithelial cells possessing the mechanical means for the release and translocation of ninein from the centrosome to non-centrosomal anchoring sites.
Centrosomal ninein is released and relocated to junctionassociated apical sites during epithelial polarisation These studies have identified distinct ninein speckles in supporting cochlear epithelial cells and provided further evidence for the redeployment of ninein to the apical anchoring sites during normal tissue morphogenesis in situ. In addition, we show for the first time that ninein is similarly redeployed to apical peripheral anchoring sites following in vitro polarisation of MDCKII cells. We also show that cytoplasmic ninein speckles are a general feature of epithelial cells and that they are highly dynamic, displaying microtubule-dependent movements within the cytoplasm.
Interestingly, we report extensive colocalisation of ninein and ␤-catenin at cell junctions in polarised MDCKII and supporting cochlear cells. This suggests that ninein accumulates at and the noncentrosomal anchoring sites form in close proximity to adherens junctions. Expression of cadherin junction components has been reported to increase microtubule minus-end stability, and ␤-catenin has been suggested to mediate this link (Chausovsky et al., 2000). The data presented in this study thus suggest a possible link between ninein and ␤-catenin and support a role for adherens junctions and ␤catenin in the establishment of non-centrosomal apico-basal microtubule arrays in polarised epithelial cells. We therefore propose that ninein speckles are released from the centrosome and transported along microtubules by plus-end-directed kinesin motors to apical non-centrosomal anchoring sites. In addition, we suggest that the anchoring sites form in association with adherens junctions during assembly of the apico-basal microtubule arrays. These studies clearly also highlight the vital role that ninein dynamics and redistribution play in microtubule reorganisation during epithelial polarisation and differentiation.

Antibodies
Two rabbit anti-ninein antibodies generated against the same GST fusion peptide [Pep3, sequence 952-1227(Bouckson-Castaing et al., 1996; gift from M. Bornens, Institute Curie, Paris, France] were used in these studies. One was a gift from M. Bornens and the other was newly generated, affinity purified and tested for specificity. The ninein antibodies were used at 1:1000 (0.1 mg/ml). Control experiments involving pre-incubation of the antibody with the Pep3 peptide resulted in no detectable labelling. Furthermore, siRNA depletion of ninein in ARPE-19 cells (human cell line, using two sequences: sense CGGUACAAUGAGUGUAGAATT and antisense: UUCUACACUCAUUGUACC -GTC from Ambion and sense: GGAAGACCUAAGAAAUGU and antisense: UACAUUUCUUAGGUCUUCC from Qiagen) was used to verify the specificity of the ninein antibody. Cells were transfected with the siRNAs twice, with an interval of 48 hours between transfections and for a total of 96 hours and the cells were fixed and labelled with the ninein antibody. A combination of the two sequences produced the best depletion results. A scramble sequence was used as control.
Cells were transfected at 60-70% confluence using FuGene VI (Roche) and incubated for 12-16 hours at 37°C. In double transfection experiments, both plasmid DNAs were mixed at a 1:1 ratio. Cells for live imaging were grown on 40 mm diameter cover slips for use in a POC-R chamber. The GFP-ninein construct was a gift from M. Bornens [Institute Curie, France (Abal et al., 2002)]. The EYFP-tubulin construct was obtained by cloning full-length cDNA sequence of ␣-tubulin into the XhoI and BamHI sites of the pEYFP-C1 vector (Clontech).

Immuno-fluorescence labelling of cultured and isolated pillar cells
Cultured cells and isolated cochlear pillar cells [dissected as described previously; ] were fixed in cold (-20°C) 90% methanol/10% MES (100 mM MES, 1 mM EGTA, 1 mM MgSO 4 at pH 6.9, Sigma) for 4 minutes, rehydrated in 1% goat serum in PBS and detergent treated with 1% NP40 in PBS for 4 minutes, blocked in 10% goat serum in PBS and incubated in primary antibodies for 30-90 minutes followed by secondary antibodies for 30 minutes. Cells labelled with phalloidin were fixed with 4% paraformaldehyde in PBS and extracted in 1% NP40 in PBS, and cells labelled for both actin and microtubules were fixed in 3.7% paraformaldehyde in PHEMO buffer (68 mM PIPES, 25 mM HEPES, 15 mM EGTA and 3 mM MgCl 2 ) with 0.5% glutaraldehyde and 0.5% Triton X-100 (Sigma) for 10 minutes and quenched with 50 nM NH 4 Cl in PBS for 10 minutes and labelled as above.
Fluorescent images were recorded using either a Zeiss Axiovert 200M with a Zeiss AxioCam camera and Axiovision software or a Zeiss LSM 510 META confocal microscope with associated software. Z-stacks were generated from optical sections taken at 0.2 m intervals, and Volocity (Improvision) software program was used for 3D reconstruction. The Volocity classification and Zeiss LSM510 META software programs were used to analyse colocalisation and determine signal overlap. Digital image files were transferred to Photoshop CS2 for image handling.

Immuno-gold labelling and TEM
PtK2 cells were fixed in 4% paraformaldehyde and 0.1% glutaraldehyde in cytoskeletal buffer (10 mM MES, 150 mM NaCl, 5 mM EGTA, 5 mM MgCl 2 , 5 mM glucose at pH 6.9) with 0.3% TritonX-100, blocked in 10% goat serum in PBS, incubated with ninein antibody overnight at 4°C, washed and incubated in protein-A conjugated to 8 nm gold particles, post-fixed in 2.5% glutaraldehyde in cacodylate buffer (0.1 M sodium cacodylate, 2.0 M calcium chloride) for 40 minutes followed by 1% osmium tetroxide for 40 minutes, processed for electron microscopy and imaged using a JOEL 200 TEM.

Drug treatments
Cells were treated with 10 g/ml nocodazole (Sigma) for 60 minutes on ice followed by 2 hours at 37°C for microtubule depolymerisation and fixed or washed several times in cold medium prior to regrowth at 37°C. Latrunculin B (Sigma) at 2 M was used for actin depolymerisation, and cells were incubated at 37°C for 30 minutes. Cell extraction was carried out with 0.5% Triton X-100 in PHEMO for 30 seconds after nocodazole treatment but before fixation.

Live cell imaging and analyses
Cells expressing GFP-ninein or GFP-ninein and YFP-␣-tubulin were maintained in a POC-R chamber on a heated stage at 37°C in phenol-red free medium (Invitrogen) for live recording. Images were captured using a 63ϫ 1.4 NA objective on a fully automated Zeiss Axiovert 200M using a Zeiss AxioCam and Axiovision software. For double transfections of GFP-ninein and YFP-␣-tubulin, filter sets JP3 (Chroma) were used in conjunction with Delta Vision software (Applied Precision Inc.) to distinguish between the two fluorochromes. Images were captured on a Micromax CCD camera (Roper Scientific, Trenton, NJ) camera and deconvolved using softWoRx (Applied Precision Inc.). Cells destined for nocodazole treatment were treated as described above and then imaged with 10 g/ml nocodazole added to the POC-R chamber.
Ninein speckle velocities were calculated by tracking individual speckles over time (1-2 second time intervals). One Way Analysis of Variance (ANOVA) followed by Tukey's post hoc analysis was used to determine significance at the 5% level. An overall average speckle velocity was calculated based on all trackable speckles, including speckles displaying brief periods of pausing in GFP-ninein-expressing control-, nocodazole-and latrunculin-B-treated PtK2 cells. However, for analyses of the affect of GSK3␤ inhibitors on ninein dynamics in MDCKII cells, only moving speckles were used for velocity estimations.
The colocalisation of ninein speckles with microtubules was verified by calculating the percentage of speckles observed on microtubules before and after a random shift of the microtubule image with respect to the speckle image in two cells double-labelled for ninein and microtubules. The percentage of speckles located either on or off a microtubule was calculated within 37 areas in two cells and the significance was determined using a t test.

FRAP analyses
FRAP data were acquired using the Zeiss LSM 510 META confocal. Images were collected prior to photobleaching as a measure of the original intensity. The Argon/Ion laser (488) was used at 4% intensity for image acquisitions. The centrosome in GFP-ninein transfected PtK2 cells was selected using the LSM imaging software and photobleached using the Argon/Iron laser at 30-40 iterations at 100% intensity. Images were recorded post-bleaching during centrosomal recovery. The centrosomal GFP recovery was determined by measuring the signal intensity of the photobleached region over time, using normal imaging parameters set prior to bleaching.