Myomegalin is necessary for the formation of centrosomal and Golgi-derived microtubules

The interphase microtubule (MT) cytoskeleton regulates cell morphology and motility, organelle transport and localization, and vesicular trafficking. Nascent MTs are formed from nucleation centers containing γ-tubulin ring complexes (γ-TuRCs). Subsequently, MTs grow by tubulin dimer polymerization at their plus-end (Kollman et al., 2011). The latter process is tightly regulated and requires plus-end stabilization by interacting proteins, the best characterized being end-binding (EB) proteins, mainly EB1 (Akhmanova and Steinmetz, 2008). The major MT nucleation center in vertebrate cells is the centrosome, also known as MTOC (microtubule organizing center), but other organelles such as the nucleus and the Golgi apparatus (GA) can concentrate MT-nucleating activities at their surface (Bartolini and Gundersen, 2006; Bornens, 2002; Chabin-Brion et al., 2001). The function of MTs is strongly linked to their capacity to bridge different cellular compartments at their minus- and plus-ends, such as the centrosome and the cell cortex, to organize directional movements and cell polarity.

In many eucaryotes, proteins related to Drosophila centrosomin bind and regulate subcellular localization of γ-tubulin, the major component of γ-TuRCs (Fong et al., 2008; Sawin et al., 2004; Zekert et al., 2010; Zhang and Megraw, 2007). Centrosomin regulates the recruitment of γ-tubulin to mitotic centrosomes, the formation of astral MTs and the proper orientation of mitotic spindles (Megraw et al., 2001). In fission yeast, Pcp1 and Mto1P are related proteins with similar functions that recruit γ-tubulin to spindle pole body (the equivalent of the centrosome in yeast) and non-spindle pole body associated MTOCs, respectively (Samejima et al., 2008; Sawin et al., 2004; Venkatram et al., 2004). In this protein family, Aspergillus nidulans AspB (Zekert et al., 2010) and mammalian CDK5RAP2/CEP215 (Fong et al., 2008) also associate with γ-tubulin to promote MT nucleation from cytoplasmic sites and centrosomes, respectively. All these proteins are large coiled-coil proteins with a small (around 60 amino acids long) N-terminal conserved domain known as the centrosomin motif 1 (CM1) (Samejima et al., 2008; Zhang and Megraw, 2007). Interestingly, the CM1 domain in centrosomin is required for γ-tubulin, D-TACC and Msps recruitment to centrosomes, but not of other centrosomal proteins such as Aurora-A and Map60 (Zhang and Megraw, 2007). In addition, a recent study has demonstrated that the CM1 domain of CDK5RAP2 is able to bind γ-TuRCs and enhance its ability to nucleate MTs (Choi et al., 2010). This motif was therefore named γ-TuNA (γ-TuRC-mediated nucleation activator).

Myomegalin/PDE4DIP is a CDK5RAP2 paralog in vertebrates. Myomegalin is highly expressed in muscle tissue and its product has been described as an interactor of phosphodiesterase 4D, an enzyme controling cAMP level (Taskén et al., 2001; Verde et al., 2001). In some mammalian cells, Myomegalin localizes to both the GA and the centrosome (Verde et al., 2001), but its function is currently unknown. Other proteins, such as CAP350 and AKAP450 (also known as AKAP9 or CG-NAP) localize at the centrosome and GA. CAP350 participates in MT anchoring at the centrosome and may stabilize MTs in the GA area to maintain its pericentrosomal structure (Hoppeler-Lebel et al., 2007). AKAP450, a γ-tubulin-interacting protein, in addition to its role as a kinase-anchoring scaffold protein at the centrosome, is also important for MT nucleation from the centrosome and GA, and for GA assembly (Hurtado et al., 2011; Takahashi et al., 1999; Takahashi et al., 2002). The GA and the centrosome cooperate in different cellular processes such as cell polarity, cell migration and ciliogenesis (Bisel et al., 2008; Follit et al., 2006; Hurtado et al., 2011; Magdalena et al., 2003; Marie et al., 2009; Sütterlin and Colanzi, 2010). Interestingly, activation of CDC42 at the GA regulates centrosome organization and function, and depends on the GA matrix protein GM130 (Kodani et al., 2009; Kodani and Sütterlin, 2008). Another GA matrix protein, GRASP65, also regulate centrosomes during mitosis (Sütterlin et al., 2005). In addition to the centrosome, the GA can be a potent MT-organizing organelle (Chabin-Brion et al., 2001; Efimov et al., 2007). Centrosome and GA-derived MTs cooperate for different functions such as proper ribbon formation and polarization during GA assembly (Vinogradova et al., 2012). The molecular machinery underlying the ability of the GA to organize MTs has begun to be identified (Efimov et al., 2007; Hurtado et al., 2011; Kim et al., 2007; Rivero et al., 2009). It includes AKAP450 and CLASP, a MT plus-end binding protein. AKAP450 is recruited to cis-Golgi membranes via GM130, and CLASP is anchored to the trans-Golgi membranes via GCC185.

In this study, we show that Myomegalin is a centrosome and cis-Golgi protein required for MT growth from the centrosome and GA. We also show that Myomegalin exists as several isoforms that participate by different means in MT regulation. We describe the role of two of them. The first isoform, CM-MMG, regulates MT nucleation via an interaction with the γ-TuRCs. The second isoform, EB-MMG, regulates MT dynamics via an interaction with the plus-end tracking protein (+TIP) EB1.

Immortalized human retinal pigment epithelial cells htert-RPE1 (thereafter referred as RPE1) were grown in DMEM/F12 (Life Technologies) supplemented with 10% fetal calf serum (heat inactivated 30 minutes at 56°C).

A rabbit polyclonal antibody (Ab#1) against Myomegalin was raised against the fragment 322–939 of clone IMAGE n° 4813248, fused to GST. The fusion protein was injected in rabbits and the serum was depleted in anti-GST immunoglobulins by affinity column. The specificity of this antibody was assessed by immunofluorescence (IF) and Western blotting (Fig. 1; supplementary material Fig. S1).

Rabbit polyclonal antibody against Myomegalin (HPA008162, denoted Ab#3 in this study) was from Sigma and mouse monoclonal antibody against Myomegalin was from Abnova (M01, clone 2B5, denoted Ab#2 in this study). For Golgi compartment identification, GM130 antibody was from Abcam (rabbit monoclonal) or from BD Transduction Laboratories (mouse monoclonal), TGN46 from AbD serotec (sheep polyclonal). MTs were stained with a rat monoclonal YL1/2 antibody (Abcam). Anti-EB1 antibodies were from BD transduction Laboratories (mouse monoclonal), and Santa Cruz Biotech (rat monoclonal). Anti-γ-tubulin antibodies were from Sigma–Aldrich (mouse monoclonal GTU-88 and rabbit polyclonal T3559). Anti-GCP2 and NEDD1 were from Sigma–Aldrich and Novus Biological, respectively. Rabbit and mouse anti-AKAP450 were from Bethyl Laboratories and BD Biosciences, respectively. Mouse anti-c-Myc (9E10) was from Santa Cruz Biotechnology. Anti-GFP antibodies were purchased from Abcam (rabbit polyclonal) and Roche (mouse monoclonal).

For MT depolymerisation and/or Golgi dispersal, nocodazole (10 µM) or BFA (brefeldin-A) (5 µg/ml) was added to the culture medium for 2 hours. For dynein inhibition experiments, cells were transfected with a myc-p50 dynamitin coding expression plasmid.

Full-length Myomegalin cDNAs (KIAA0477/EB-MMG and KIAA0454/CM-MMG) were obtained from the Kazusa DNA Research Institute. The sequence coding for the full-length and truncated proteins were amplified by PCR with Gateway compatible primers using Platinum HiFi polymerase (Invitrogen) and transferred into pDONR/ZEO by Gateway® recombination. The sequences were subsequently transferred by Gateway® recombination into modified Gateway-compatible pRK5-Myc, pEGFP-C1 and pmCherry-C1 expression vectors. A pcDNA3/γ-tubulin coding vector was a gift from Alexey Khodjakov (Wadsworth Center, Albany, NY).

For plasmids, cells were transfected with Fugene6 (Roche Applied Science) according to manufacturer's protocol.

Single siRNA oligonucleotides or different combinations of mixed siRNA oligonucleotides against Myomegalin were transfected into cells using Oligofectamin (Life Technologies) according to the manufacturer's protocol. The sequences are the following: siRNA#1 (5′-CUAACGAGCUGGAGAAAUA-3′), siRNA#2 (5′-GAAGGGAAUAGUAAACUUA-3′), siRNA#3 (5′-AGAGCGAGAUCAUGACUUA-3′), siRNA#5 (5′-CAAGAAGAAUUGCAGAAUA-3′), siRNA#7 (5′-AGAGGAAGCCCUUGGAGAA-3′). In all the experiments, a second transfection was performed at 48 hours and experiments were conducted 72 hours after the first transfection. This protocol gave the best reduction of protein expression. To deplete GM130 or AKAP450 we used a mixture of two siRNA. The sequences are as described previously (Rivero et al., 2009). To deplete EB1 we used siRNA (5′-UUGCCUUGAAGAAAGUGAA-3′). A non-targeting siRNA or siRNA targeting GAPDH were used for controls. All siRNA were purchased from Dharmacon.

Cells were grown on coverslips, fixed in cold methanol (6–8 minutes at −20°C), saturated for 30 minutes in PBS 3% BSA and incubated with the appropriate mixture of primary antibodies for 60 minutes at room temperature. After washing in PBS 0.1% tween, they were incubated for another hour with the adequate secondary antibodies conjugated to cy2, cy3 or cy5 (Jackson laboratories) to which was added 1 µg/ml DAPI for DNA staining, rinsed and mounted in Prolong Gold anti-fade reagent (Life Technologies). Confocal images were acquired by capturing Z-series with 0.3 µm step size on a Zeiss LSM 510 laser scanning confocal microscope.

RPE1 cells were cultivated on coverslips and treated with appropriate siRNA for 72 hours before the addition of 1 µg/ml of nocodazole to the medium for 2 hours at 37°C followed by a 30-minute incubation on ice to depolymerize MTs. For repolymerization, coverslips were washed with medium and incubated 1, 3 or 5 minutes in warm medium. Before fixation, cells were permeabilized for 15 seconds in PHEM (60 mM PIPES, 25 mM HEPES, 10 mM EGTA, 4 mM MgSO4, pH 7.0) with 0.25% Triton and 320 mM sucrose. One volume of cold methanol was added and left for an additional 45 seconds before transferring cells to pure cold methanol for 4 minutes on ice. IF was then done as usual.

For time-lapse imaging, RPE1 EB1-GFP stable transfected cells were placed in 35 mm glass based dishes (Iwaki) and transfected as previously described with plasmid encoding mCherry-EB-MMG or mCherry-CM-MMG for 24 hours. Microscopy was carried out using a Zeiss LSM 510 laser scanning confocal microscope with a 63× oil lens. Images were taken every 2 seconds over 400-second periods. To determine EB1 velocity MT ends were tracked using the “track points” function of the MetaMorph software.

3×10⁴ RPE1 cells were cultivated in 6 well plates and treated with appropriate siRNA as usual. The day before filming, cells were split in 6 well plates coated with collagen (25 µg/ml in 0.2% acetic acid) to be able to film individual cells. Phase contrast images were acquired every 2 minutes on a Zeiss Axiovert 200 microscope equipped with a thermocontroled chamber with a 10×/0.30 Zeiss objective for 4 hours. The velocity, the distance from start to end point – as a straight line – (D) and the total track distance (T) were measured in Image J with the manual tracking plugin. Directionality was calculated as the ratio D/T. Statistical analysis of the results was done with the GraphPad Prism software.

Myc and/or GFP-tagged constructs were transfected in RPE1 cells. Twenty four hours later, cells were washed in PBS and lysed in 1 ml of buffer containing 20 mM TRIS, 2 mM EDTA, 150 mM NaCl, 1% NP40 and a complete protease inhibitor cocktail (Roche). Lysates were collected, incubated on ice for 15 minutes and centrifuged for 30 minutes at 4°C at 13,000 g. Fifty µl were saved as input material. 10 µl anti-Myc conjugated sepharose beads (Santa Cruz Biotechnology) or anti-GFP conjugated sepharose beads (Abcam) were incubated for 2 hours at 4°C. Control immunoprecipitation was performed with lysates of untransfected cells. Beads were washed 3 times in lysis buffer, suspended in Laemmli sample buffer and boiled. Samples were electrophoresed on NuPAGE Novex Bis-Tris minigels (Life Technologies) and proteins were detected by Western blot.

The human Myomegalin gene (PDE4DIP) encompasses 224 kb on chromosome 1q12. In the Ensembl database it is predicted to encode 24 isoforms, 8 of which have a strong support from transcript sequences. These 8 isoforms originate from the use of common and alternative exons, leading to proteins with differences in their N-terminus, C-terminus and internal sequences (Fig. 1A). Seven of these isoforms have a conserved CM1 domain at their N-terminus, known to bind γ-TuRC (Choi et al., 2010). The three longer isoforms have a conserved CM2 domain at their C-terminus, characterized as a centrosome/Golgi targeting motif in CDK5RAP2 (Wang et al., 2010).

We first analyzed which isoforms are expressed in RPE1 cells using three different antibodies, and three siRNAs (Fig. 1B). Of the four major protein bands detected by Ab#2, two were consistent in apparent MW with isoforms 001/002/003 above 247 kDa and 004/005 around 120 kDa. The inhibitory effects of pooled or single siRNA confirmed this prediction, as only siRNA#3 inhibited the expression of the ∼120 kDa isoforms, while all 3 siRNA inhibited the large ones. Two other major bands around 80 and 30 kDa were also observed; they do not correspond to predicted isoforms, and were not inhibited by siRNA, possibly reflecting non-specific antibody reactivity.

As expected, Ab#3 detected protein products corresponding to the same large (above 240 kDa) and ∼120 kDa isoforms, albeit with a lower sensitivity than Ab#2. This lower sensitivity may account for the absence of residual large isoform in cells treated with siRNA#2, as seen with Ab#2. Interestingly, Ab#3 recognized a unique 140 kDa isoform not detected by Ab#2, and only inhibited by siRNA#3, which likely corresponds to isoform 006. This isoform lacks a CM1-containing N-terminus and has a unique C-terminus encoded by a unique alternative exon. A minor band corresponding to a protein product of 70 kDa is likely to be nonspecific for the same reasons as mentioned above for Ab#2. Ab#1 detected high molecular weight proteins very similar to Ab#2. Bands detected below 107 kDa are likely to be nonspecific as they do not correspond to any predicted Myomegalin products and are not inhibited by siRNAs.

RPE1 therefore express a complex pattern of different isoforms, which probably exert different functions within the cell.

We determined the subcellular localization of Myomegalin in RPE1 cells by IF. Ab#2 and Ab#3 antibodies were not used in these experiments because i) Ab#2 showed weak cross-reactivity with the human Myomegalin paralog CDK5RAP2 at the centrosome and ii) Ab#3 gave a strong centrosomal and cytoplasmic staining that was not specific as it was not inhibited by any Myomegalin-specific siRNA either used alone or pooled. A rabbit polyclonal antibody (Ab#1) raised against a CM1-containing fragment was the most reliable and specific reagent for IF experiments. Immunostaining with this antibody indicated that Myomegalin localizes at the centrosome, the GA and punctually in the cytoplasm (Fig. 2A). A similar staining has been described in Cos-7 cells and in FRTL-5 thyroid cells (Verde et al., 2001). However, we noticed that the GA staining in RPE1 cells was present in only 20 to 30% of proliferating cells as compared to 70 to 80% of resting cells. Consequently, subsequent experiments were done with overnight serum-starved cells. In these conditions, we observed that siRNA#1 and siRNA#3 were the most efficient at inhibiting Myomegalin expression at the GA but did not induce complete depletion at the centrosome (data not shown). The decrease in Myomegalin-specific staining by siRNA#2 was weak, as expected from Western blot experiments (data not shown; Fig. 1B). However, pooled siRNA (#1+#3) showed the strongest inhibitory effect and were used in subsequent experiments (supplementary material Fig. S1A,B). Of note, the GA showed enlargement and partial fragmentation upon Myomegalin depletion (supplementary material Fig. S1B). The IF results obtained with Ab#1 were confirmed with Ab#2, with the exception of a stronger centrosome staining probably due to the weak cross reactivity of the antibody with CDK5RAP2 (data not shown).

Our results show that RPE1 cells express CM1-containing isoforms of Myomegalin at the GA and centrosomes. However, the localization of endogenous isoform 006 was not addressed in these experiments as it is not recognized by Ab#1.

To more precisely determine Myomegalin localization at the GA, RPE1 cells were labeled with anti-Myomegalin antibody together with antibodies directed against GM130/golgin95/GOLGA2 and TGN46/TGOLN2 to label the cis- and trans-regions of the GA, respectively. We observed an extensive colocalization of Myomegalin with GM130 at the cis-Golgi cisternae (Fig. 2B). TGN46-labeled structures were close but distinct from anti-Myomegalin staining (Fig. 2B).

We next studied the fate of Myomegalin during GA fragmentation. First, we examined the effect of a treatment with BFA or nocodazole. These drugs disrupt the structural integrity of the GA. They respectively block the traffic from the ER to the GA and depolymerize MTs, and cause Golgi fragmentation into small punctate remnants. After 30 minutes of BFA or 3 hours of nocodazole treatment, Myomegalin was found associated with GA remnants distributed throughout the cytoplasm, indicating that the localization of Myomegalin to the GA is independent of the integrity of this organelle. Like in intact GA, the cis-Golgi marker GM130 showed partial colocalization with Myomegalin (supplementary material Fig. S2A). Because nocodazole depolymerizes MTs, this result further indicated that Myomegalin is associated to the GA in a MT-independent manner. Second, we followed the fate of Myomegalin in cells overexpressing the p50 subunit of dynactin to inhibit dynein function, which causes fragmentation of the GA (Corthésy-Theulaz et al., 1992). Myomegalin association to GA ministacks persisted upon dynein inhibition (supplementary material Fig. S2A). Myomegalin thus localizes to the cis-Golgi independently of MTs and Golgi structure.

The human Myomegalin paralog CDK5RAP2 localizes to the GA in an AKAP450-dependent manner (Wang et al., 2010). Similarly, we could show that Myomegalin colocalizes with AKAP450 and depends on this protein for GA localization (supplementary material Fig. S2B,C). Interestingly, Myomegalin depletion altered localization of AKAP450 at the GA, which showed a more diffuse staining pattern in the cytoplasm and an impaired colocalization with GM130 compared to control cells. Myomegalin may thus control to some extent AKAP450 localization and/or stability (supplementary material Fig. S2D). As expected from the dependence of AKAP450 on GM130 for its presence at the GA (Rivero et al., 2009), GM130 depletion also prevented localization of Myomegalin on this organelle (supplementary material Fig. S3).

To better determine which isoforms are present at the GA and centrosome, we expressed full length and truncated GFP-tagged isoforms. We concentrated on isoforms 001 and 006 as representatives of two strikingly different forms of Myomegalin. We named isoform 001 CM-MMG because it has a CM1 and a CM2 domains and isoform 006 EB-MMG because, as detailed below, it has a unique tandem of putative EB1 binding motifs (Fig. 3A). At low expression levels, the two full length isoforms localized at the GA in close association with the cis-Golgi protein GM130 (Fig. 3B), but of the two only the CM-MMG isoform was detected at the centrosome. Both N and C-terminal CM-MMG truncated proteins localized at the centrosome but only the C-terminal part of CM-MMG localized at the GA. Strikingly, EB-MMG was addressed to the GA through its N-terminus, indicating that the two isoforms have two different, non-overlapping, Golgi-targeting regions (Fig. 3B,C).

Human CDK5RAP2 and Myomegalin orthologs in various species, like Drosophila centrosomin, are known to associate with and to anchor γ-TuRCs MT nucleating complexes (Fong et al., 2008; Sawin et al., 2004; Zhang and Megraw, 2007). More importantly, the conserved CM1 domain not only binds γ-TuRCs directly but also stimulates its MT nucleation activity (Choi et al., 2010).

When overexpressed in RPE1 cells, four out of the six MMG constructs formed dense aggregates. In IF, aggregates formed by CM-MMG or its N-terminal part (CM-MMG_N), which has the CM1 domain, strongly recruited the γ-TuRCs components γ-tubulin (Fig. 4A), GCP2 (supplementary material Fig. S4B) and the associated protein NEDD1 (supplementary material Fig. S4A). In contrast, aggregates formed by EB-MMG_N, which is devoid of the CM1 domain, did not recruit γ-TuRCs (Fig. 4A). However, it was unique in its capacity to concentrate pericentrin (supplementary material Fig. S4C). The CM-containing Myomegalin isoforms are thus likely to bind γ-TuRCs through their CM1 domain, like other members of the same family. Surprisingly, we could not demonstrate specific binding between γ-tubulin and CM-MMG or CM-MMG_N. However, we found that a CM1-containing fragment, but not a fragment lacking the C-terminus of CM1 (Fig. 3A, CM1 and CM1ΔC), specifically associated with endogenous and ectopic GFP-tagged γ-tubulin (Fig. 4B).

Next, we asked whether the aggregates of CM-MMG and CM-MMG_N associated with γ-TuRCs possess MT nucleation capacity. When we depolymerized MTs in transfected cells and subsequently allowed them to regrow, MT asters formed around nearly all CM-MMG and CM-MMG_N, but not control CM-MMG_C, aggregates (Fig. 5A). Strikingly, we observed that many CM-MMG_N associated asters also colocalized with anti-GM130 labeled GA remnants (Fig. 5B), suggesting that Myomegalin may participate in the regulation of MT production from the GA.

To test our hypothesis that Myomegalin regulates MT growth from the GA and/or the centrosome, MTs were depolymerized and allowed to regrow for various lengths of time in control and Myomegalin siRNA-treated RPE1 cells. α-tubulin staining after 1 minute of regrowth revealed numerous non-centrosomal MTs in addition to the centrosomal asters in control cells. These non-centrosomal MTs were associated with the dispersed Golgi remnants, as already described (Efimov et al., 2007; Rivero et al., 2009), to which Myomegalin was associated (Fig. 6A). Myomegalin siRNA-treated cells showed a striking difference with control cells in the formation of centrosomal and non-centrosomal MTs (Fig. 6B). After 1 minute regrowth, the aster size of Myomegalin siRNA-treated cells was diminished at the centrosome (1.35±0.39 µm versus 0.65±0.20 µm for control and Myomegalin siRNA, respectively) (Fig. 6C) and almost totally abrogated at the Golgi remnants. After 5 minutes, the difference at the centrosome was even higher with 7.85±2.60 µm and 1.61±0.71 µm for control and Myomegalin siRNA, respectively (Fig. 6C). However, a seemingly normal array of cytoplasmic MTs was recovered at later time points. Thus, MT regrowth is delayed in the absence of Myomegalin.

The unique sequence found at the N-terminus of Myomegalin isoform 006/EB-MMG is conserved in vertebrates. It is encoded by a single exon and does not show predicted domains. However, examination of the human sequence revealed three putative EB-1 binding motifs SxI/LP. These motifs are known to bind the end binding homology (EBH) domain of EB1 in +TIP binding partners such as CLASP, APC and CDK5RAP2 (Honnappa et al., 2009) (supplementary material Fig. S5). Validated binding motifs are found within basic serine/proline rich sequence regions. Site prediction using the Eukaryotic Linear Motif (ELM) resource confirmed that the second and third motifs of the 006/EB-MMG isoform are likely to constitute EB1 binding sites. The second motif is the best conserved in different species.

We found that EB-MMG associates with EB1 by co-immunoprecipitation in a cell line expressing GFP-EB1 and transfected with Myc-tagged full length or truncated constructs of EB-MMG. As expected, only EB-MMG and EB-MMG_N, which have the putative EB1 binding sites, but not EB-MMG_C, which does not, were able to bind EB1 (Fig. 6A). CM-MMG did not coprecipitate with EB1 (data not shown).

In RPE1 cells with moderate levels of overexpressed EB-MMG, the latter colocalized with EB1 at MT plus-ends (Fig. 7B), suggesting that EB-MMG may have a role in MT growth and dynamics. To test this hypothesis, we first analyzed the effect of overexpressing this isoform on the localization of EB1 in RPE1 cells. We observed a striking reduction of endogenous EB1 at MT plus ends in cells overexpressing EB-MMG but not CM-MMG (Fig. 7C,D; data not shown). We also looked at EB1 localization in RPE1 cells stably expressing ectopic GFP-EB1. In this cell line overexpressed EB1 labeled MTs along their entire length, as described previously by others (Askham et al., 2002; Ligon et al., 2003). Overexpressed EB-MMG strongly displaced EB1 from MTs to a diffuse cytoplasmic and nuclear localization identical to EB-MMG staining in these cells. This effect was not observed with CM-MMG (data not shown). These observations suggest that overexpressed EB-MMG may negatively regulate MT growth by trapping EB1 away from MT plus-ends. This was indeed the case in a MT regrowth assay where MT aster size was diminished by 50% in EB-MMG overexpressing cells compared to control cells (Fig. 6E). In addition, we used videomicroscopy to measure the velocity of GFP-EB1 comets in control and EB-MMG or CM-MMG overexpressing cells. This velocity was diminished by ∼30% in EB-MMG but not CM-MMG cells (Fig. 7F).

Finally, Myomegalin depletion by siRNA treatment did not alter EB1 localization at MTs plus-ends. In the reverse experiment, Myomegalin still localized at the GA and centrosome when EB1 was depleted (supplementary material Fig. S6).

Golgi-derived MTs are important for cell polarization and cell motility (Efimov et al., 2007; Miller et al., 2009; Nakano et al., 2010; Rivero et al., 2009). Because decreased Myomegalin expression affects MTs originating from the GA in RPE1 cells we wondered whether cell mobility was altered. We analyzed the individual movements of RPE1 cells seeded at low concentration (to prevent alteration of trajectories or speed by one cell encountering another one on its way). We tracked individual cells by time-lapse microscopy to analyze both the direction of the cell movement and the distance run by cells. We used two different siRNA pools targeting similar regions of Myomegalin mRNA, and thus likely to target the same Myomegalin isoforms (Fig. 1A). Control RPE1 cells filmed for 4 hours usually moved in a directionally persistent fashion. Cells treated with siRNA directed against Myomegalin ran shorter distances or were even immobilized, probably according to the level of the knock-down obtained with the siRNAs. The decrease in the distance run by cells was due to a reduced velocity but not to a change in migration directionality (measured by the ratio of the distance between start and finish point divided by the total track distance) (Fig. 8A,B). Values are probably underestimated as RNAi efficiency is not absolute (especially for the #1+#3 siRNA pool).

In conclusion, Myomegalin depletion affected cell motility but not migration directionality, which suggests that cell polarization was not affected. This is reminiscent of the role of AKAP450 in migration (Rivero et al., 2009). We propose that reduced MT polymerization (by reduction of MT nucleation and/or loss of EB1 from MT +ends) is responsible for the slower migration in cells with reduced Myomegalin expression.

The consensus coding sequences (CCDS) project has identified eight CCDS in the human Myomegalin/PDE4DIP gene, which are consistently annotated and of high quality. Most protein variability is found at the N and C-terminus. In both cases, a specific transcript is likely to be obtained from the initiation or termination of transcription downstream of alternative promoters and polyadenylation sites, respectively. Four putative promoters and three polyadenylation sites may thus account for part of the isoform diversity. Further complexity stems from alternative splicing events, such as those regulating the use of exons 29 and 37. Our study shows that a subset of the isoforms encoded by these CCDS is expressed in RPE1 cells. Centrosomin, the Drosophila ortholog of Myomegalin, acts during oogenesis and embryogenesis as a complex mix of isoforms to allow centrosome function (Eisman et al., 2009). We found centrosomin isoforms related to human CM-MMG (i.e. possessing a CM1 domain at their N-terminus) but none related to EB-MMG. Different isoforms can be present in a single cell or can be differentially regulated among various tissues or developmental stages to exert different functions (Bourdon et al., 2005; Pan et al., 2008; Wang et al., 2008). At least three Myomegalin isoforms are expressed in RPE1 cells, indicating that their functions are not exclusive but that they work together in independent or coordinated cellular processes, as we discuss below. There are rare cases in which two isoforms produced from one gene display alternative sequences at a similar location in the protein, conferring two different properties or function. Such exon switching is observed for example in the FGFR2 and calcitonin/CRGP genes, allowing the production of receptors binding different ligands and two different tissue-specific hormones sharing their N-terminus, respectively (Rosenfeld et al., 1983; Yan et al., 1993). The two different N-termini of Myomegalin (possessing or not a CM1 domain) confer to the protein the capacity to bind either γ-TuRCs (with CM1) or EB1 (without CM1). However, the mechanism responsible for the production of Myomegalin isoforms with or without a CM1 domain at the N-terminus does not involve alternative splicing but rather alternative promoters. Myomegalin orthologs from other species and vertebrate paralog CDK5RAP2 are related to the CM-MMG isoform. Interestingly, all vertebrate species have a Myomegalin gene that encodes the two isoforms with unique N-termini. It is therefore likely that the two different functions of Myomegalin, associated with the two types of isoforms, cannot work independently.

Myomegalin isoforms detected in RPE1 cells localize at the centrosome and the cis-Golgi. We found that, more specifically, the CM-MMG isoform (with conserved CM1 and CM2 domains) is present at both locations, whereas EB-MMG isoform is only detected at the cis-Golgi. The CM2 domain directs CDK5RAP2 to the centrosome and GA in a pericentrin and/or AKAP450-dependent manner (Wang et al., 2010). The preserved localization of the CM-MMG_C truncation mutant at the two organelles is consistent with a similar role of the CM2 domain in Myomegalin. The CM-MMG_N mutant, which is not present at the GA, is consistently localized at the centrosome (Fig. 3B; data not shown). This is possibly the result of an interaction of the Myomegalin CM1 domain with γ-TuRCs anchored at the centrosome through interactions with pericentrin, AKAP450 and/or CDK5RAP2 (Fong et al., 2008; Takahashi et al., 2002). Our study also underlines the central role of AKAP450 as a scaffold to recruit proteins important for MT nucleation at the GA, such as CDK5RAP2 and Myomegalin.

Surprisingly, EB-MMG is targeted to the GA through its N-terminus, suggesting a second, CM2-independent targeting mode. The colocalization of non-overlapping CM-MMG and EB-MMG truncated forms (CM-MMG_C and EB-MMG_N) at the GA (Fig. 3C) indicates that heterodimerization is an unlikely mechanism to explain this observation. Supporting this, CM-MMG_C, but not EB-MMG_N, localizes at the centrosome.

We have demonstrated that inhibiting Myomegalin expression leads to decreased MT growth from the centrosome and the GA during the recovery phase following nocodazole treatment. This is consistent with the localization of Myomegalin at these two organelles (this study; Verde et al., 2001), and with the recruitment of γ-TuRCs by Myomegalin in humans and related proteins in various species (Fong et al., 2008; Sawin et al., 2004; Terada et al., 2003; Veith et al., 2005). Nucleation at a specific site relies on γ-TuRC anchoring by large scaffolding proteins such as pericentrin, AKAP450 or CEP192. Recent studies suggest that it also depends on the nucleation-activating role of Myomegalin-related proteins such as CDK5RAP2 in humans and SPC110 in budding yeast (Choi et al., 2010; Kollman et al., 2011). Indeed, the two latter proteins are direct interactors of γ-TuRC and γ-TuSC (γ-tubulin small complex) respectively. They may help position γ-tubulin molecules at a distance compatible with tubulin protofilaments arrangement in the MT lattice (Kollman et al., 2011). Myomegalin does not seem to participate in γ-tubulin anchoring on interphase centrosomes (supplementary material Fig. S1B); it may instead play a nucleation-activating role at the centrosome and possibly at the GA. In contrast, CDK5RAP2 is required for both γ-TuRC anchoring and activation at the centrosome, because centrosomal γ-tubulin is greatly diminished following CDK5RAP2 depletion (Fong et al., 2008). It will be of interest to determine to what extent CDK5RAP2 and Myomegalin are redundant or complementary at the centrosome. Myomegalin may also play a role similar to other centrosomal proteins such as CEP70 and Centrobin, which regulate MT elongation and/or stabilization, but are not involved in centrosomal anchoring of γ-tubulin (Lee et al., 2010; Shi et al., 2011; Shi et al., 2012).

The molecular bases of the MT organizing activity of the GA only begin to be delineated. Clearly, it requires a GM130-AKAP450-γ-TuRC complex at the cis-Golgi and a GCC185-CLASP complex at the trans-Golgi (Efimov et al., 2007; Rivero et al., 2009). We found that aggregates of the overexpressed CM-MMG isoform promote MT nucleation at sites associated with the cis-Golgi protein GM130 and AKAP450, but not pericentrin. This suggests that a GM130-AKAP450-Myomegalin-γ-TuRC complex functions in MT nucleation at the cis-side of the GA. CDK5RAP2 was also found at the GA and may play a redundant or complementary role together with Myomegalin (Barr et al., 2010; Wang et al., 2010). Indeed, MT nucleation was delayed but not abolished following siRNA-mediated downregulation of Myomegalin. A defect in Golgi-associated MT organization impairs Golgi polarization and directional migration in RPE1 cells (Miller et al., 2009). The defect in cell migration that we observed after inhibiting Myomegalin expression suggests that the MT cytoskeleton and/or Golgi organization is not fully functional in the absence of Myomegalin.

We found that a specific Myomegalin isoform (EB-MMG) may regulate MT growth and organization through its direct interaction with the +TIP EB1 proteins. A number of +TIP protein bind MT ends indirectly through an interaction between the EBH C-terminal domain of EB1 and an SxI/LP motif in the +TIP (Honnappa et al., 2009). EB-MMG has tandem SxLP motifs, with the N-terminal one being highly conserved in vertebrate species. The sequence context of these motifs matches the basic serine/proline-rich consensus as defined in the Eukaryotic Linear Motif (ELM) resource for a functional EB1 binding motif. Such motifs are present in MT stabilizing (CLASP, APC) and destabilizing (MCAK) +TIPs. These motifs are likely to drive the binding of EB-MMG to EB1 we observed. Interestingly the interaction of overexpressed EB-MMG with EB1 resulted in a depletion of EB1 from MT plus-ends. MT dynamics was affected, at least using GFP-EB1 velocity measurements as a readout. To what extent this isoform was required for MT regulation in our MT regrowth assay is also not known, since we used siRNAs to downregulate most isoforms. Future experiments will require EB-MMG-specific antibodies and siRNA to confirm localization of the endogenous protein at the GA and possibly at plus-ends and to delineate its specific role in MT regulation. Similarly, we cannot exclude that depletion of the short CM1-containing isoforms 004 and 005 contributes to the phenotypes we observed. It is tempting to speculate that EB1, recruited at the GA in an EB-MMG-dependent manner, participates to the stabilization of MTs growing from the cis-Golgi, together with CLASP subsequently recruited at growing plus-ends at the trans-Golgi. EB-MMG at the Golgi surface may serve as a docking site for MTs plus-ends important for Golgi structure and/or positioning. In addition, it is possible that a cytoplasmic EB-MMG pool associates with EB1 to regulate EB1 availability or stability.

CDK5RAP2 exists as two isoforms resembling the CM-MMG isoform and also binds EB1 through an SxLP motif (Fong et al., 2009). The role of this interaction is still unclear because CDK5RAP2 inhibition only slightly affects MT dynamics and the SxLP motif in CDK5RAP2 is not conserved in all vertebrate species. We can hypothesize that in complex vertebrate organisms, MT nucleation and EB1-mediated stabilization activities must localize in very close proximity to ensure efficient and coordinated MT growth. This may be the case at the GA, but this will have to be substantiated by demonstrating the presence of EB1 (or its relative EB3) at the surface of this organelle. It is the case at the centrosome where γ-TuRC nucleation complexes are concentrated, EB1 is recruited by a CAP350-FOP complex (Yan et al., 2006). CDK5RAP2, but not Myomegalin, may recruit additional EB molecules at this site.

CDK5RAP2 alterations are associated with mitotic defects such as the formation of anastral spindles, impaired centrosomal attachment to spindle poles, abnormal mitotic spindle pole number and mitotic orientation (Barr et al., 2010; Barrera et al., 2010; Fong et al., 2008; Lizarraga et al., 2010). Mitotic Myomegalin-depleted RPE1 cells displayed normal levels of γ-tubulin, CDK5RAP2 and NEDD1 associated with spindle poles, and did not have an increased mitotic index suggestive of defects in mitosis (data not shown). CDK5RAP2 may compensate for the absence of Myomegalin in mitotic cells, but a detailed analysis of spindle structure/orientation and chromosome segregation in different cell type will be necessary to clarify the role of Myomegalin during mitosis.

In conclusion, our work establishes a role of Myomegalin in MT regulation. Future experiments are now required to clarify the specific functions exerted by the two Myomegalin isoforms at the cellular and molecular levels, and to identify the molecular complexes associated with these isoforms to perform their roles in a space and time-regulated manner. How these roles in MT regulation are integrated with the signaling activity of Myomegalin as an A-kinase anchoring and phosphodiesterase-associated protein will also be a matter of interest (Uys et al., 2011; Verde et al., 2001).

We thank A. Khodjakov for the kind gift of reagents and F. Gergely for comments on the manuscript. This work was supported by INSERM and Institut Paoli-Calmettes. F.S. was a recipient of a fellowship from the Ministère de la Recherche et de l'Enseignement Supérieur.