Ubiquitin-proteasome system controls ciliogenesis at the initial step of axoneme extension

Primary cilia are microtubule-based sensory organelles that organize numerous key signals during developments and tissue homeostasis. Ciliary microtubule doublet, named axoneme, is grown directly from the distal end of mother centrioles through a multistep process upon cell cycle exit; however, the instructive signals that initiate these events are poorly understood. Here we show that ubiquitin-proteasome machinery removes trichoplein, a negative regulator of ciliogenesis, from mother centrioles and thereby causes Aurora-A inactivation, leading to ciliogenesis. Ciliogenesis is blocked if centriolar trichoplein is stabilized by treatment with proteasome inhibitors or by expression of non-ubiquitylatable trichoplein mutant (K50/57R). Started from two-stepped global E3 screening, we have identified KCTD17 as a substrate-adaptor for Cul3-RING E3 ligases (CRL3s) that polyubiquitylates trichoplein. Depletion of KCTD17 specifically arrests ciliogenesis at the initial step of axoneme extension through aberrant trichoplein-Aurora-A activity. Thus, CRL3-KCTD17 targets trichoplein to proteolysis to initiate the axoneme extension during ciliogenesis.

T he primary cilium is a membrane-bound, microtubulebased sensory organelle that is composed of nine doublet microtubules, also called ciliary axoneme, elongated directly from the distal end of mother centriole or basal body. Defects in formation, maintenance and function of cilia often results in numerous diseases and developmental disorders, commonly known as ciliopathies [1][2][3] . Ciliogenesis is evoked upon cell cycle exit and follows a series of ordered steps that have been characterized by detailed ultrastructural analysis of ciliated cells although there are some differences depending on cell type [4][5][6] . In the intracellular pathway, the recruitment of Golgi-derived ciliary vesicles (CVs) to the distal end of mother centrioles marks the first morphological event during ciliogenesis, followed by the extension of ciliary axoneme and its associated ciliary membrane, and finally, the docking of this complex to the plasma membrane.
Ciliogenesis and cell division are mutually exclusive events as the centrioles must be released from the plasma membrane to function as a mitotic apparatus [7][8][9][10][11] . It is therefore conceivable that a set of robust regulatory mechanism is required to suppress the inappropriate ciliogenesis in proliferating cells, and a growing number of centrosomal and ciliary components are actually reported to serve these functions 8,[12][13][14][15][16][17] . On the other hand, these proteins must be eliminated when cells exit from cell cycle and form cilia. It has been shown that some protein kinases, such as TTBK2 and MARK4, act to initiate ciliogenesis by excluding CP110 from the mother centrioles 15,18,19 . Moreover, autophagymediated protein degradation was recently reported to remove OFD1 from centriolar satellites to promote ciliogenesis 20 . However, the involvement of ubiquitin-proteasome system (UPS), one of the most important protein degradation system 21,22 , appears to be controversial and/or indirect, nevertheless a subset of ubiquitin E3 ligases, including pVHL and MIB-1, has been reported to promote ciliogenesis [23][24][25][26][27] .
We have previously shown that trichoplein, originally identified as a keratin-binding protein 28 , is concentrated at the subdistal/medial zone of both mother and daughter centrioles and activates centriolar Aurora-A kinase in growing cells 29 . During ciliogenesis, trichoplein disappears from the mother centrioles, and depletion of this protein induces the aberrant ciliogenesis, whereas overexpression blocks ciliogenesis, indicating that trichoplein negatively regulates ciliogenesis at the mother centrioles.
Trichoplein also controls the recruitment of microtubules to centrioles thorough interaction with Odf2 and ninein in nonciliated HeLa cells 30 . Other groups have reported that in some tumour cells, trichoplein (also called mitostatin) exists at mitochondria and its overexpression causes the mitochondria fragmentation, thereby inhibiting tumour growth 31,32 . A mitochondrial protein VDAC3 is also shown to negatively regulate ciliogenesis at the mother centrioles 17 .
Here we provide definitive evidence that UPS functions to initiate ciliogenesis by removing trichoplein from the mother centrioles. Our global E3 screening has identified KCTD17 as a substrate-adaptor for the Cul3-RING ubiquitin ligases (CRL3s) that polyubiquitylates trichoplein at Lys-50 and Lys-57. The CRL3-KCTD17-meidated trichoplein polyubiquitylation and degradation plays a pivotal role in the initial step of axonemal extension during ciliogenesis through the inactivation of centriolar Aurora-A.

Results
UPS targets trichoplein to proteolysis during ciliogenesis. When human RPE1 (telomerase reverse transcriptaseimmortalized retinal pigment epithelia) cells were exposed to cell cycle signals that induce ciliogenesis by serum starvation 16,33 , trichoplein prominently disappeared from the mother centrioles and partly from the daughter centrioles 29 (mother centriole was judged by the nucleating cilia ( Fig. 1a; insets) or the presence of Odf2 (ref. 34; Fig. 1b)). We further found that its protein level was notably decreased (Fig. 1c). However, these reductions were completely blocked in the presence of proteasome inhibitors (MG132, Epoxomicin, ALLN and Lactacystin; Fig. 1a-c). CP110 also disappears from the mother centrioles during ciliogenesis 8,15 , but its protein level was not regulated by proteasomal degradation after serum starvation (Fig. 1c). Considering that trichoplein was strikingly polyubiquitylated upon serum starvation (Fig. 1d), the trichoplein removal from mother centriole depends upon the UPS.
Trichoplein degradation is essential for ciliogenesis. Proteasome inhibition also blocked the ciliogenesis when the respective inhibitors were added at the timing of serum starvation in RPE1 cells (Fig. 1a,c). In contrast, if cells were treated with these inhibitors from the point at which trichoplein was considerably degraded but cilia were not formed yet, for example, 9 h after serum starvation, cilia were newly grown thereafter (Fig. 1c 0 ). We also added these inhibitors 24 h after serum starvation (trichoplein was more degraded at 24 h than at 9 h; see Fig. 1c), and found a marginal effect on ciliogenesis ( Supplementary  Fig. 1). The inverse correlation between the trichoplein level and ciliogenesis suggests that proteasomal degradation of trichoplein is a critical event for ciliogenesis.
In addition to these observations, we previously found that overexpression of trichoplein at centrioles blocked the serum starvation-induced ciliogenesis 29 . We therefore investigated whether the decrease in overexpressed trichoplein level permits the ciliogenesis using the non-degradable trichoplein mutant. We generated a series of myc-tagged trichoplein (myc-trichoplein) constructs in which Lys residues were substituted to nonubiquitylatable Arg, and found that mutation at Lys-50 and Lys-57 (K50/57R) diminished its polyubiquitylation in HEK293T cells ( Supplementary Fig. 2a). Polyubiquitylation of myctrichoplein wild type (WT) was dramatically accelerated by serum starvation in RPE1 cells; however, K50/57R mutation, but not K50R or K57R, diminished this polyubiquitylation ( Fig. 2a and Supplementary Fig. 2b). Blockade of new protein synthesis of myc-trichoplein by treatment with cycloheximide rapidly decreased the WT levels in serum-starved but not serum-grown conditions; in contrast, the K50/57R levels were stable in both conditions (Fig. 2b). These results suggest that trichoplein polyubiquitylation at Lys-50 and Lys-57 triggers its degradation. As this mutation had no effect on its ability to activate Aurora-A (Fig. 2c), we decided to use a K50/57R mutant in the following studies. However, we were unable to evaluate the effect on ciliogenesis using cycloheximide because transcriptional control of ciliary gene expression plays a critical role in ciliogenesis [35][36][37] .
We therefore established the Tet-On RPE1 cell lines that expressed MBP (maltose-binding protein)-tagged trichoplein (MBP-trichoplein) in a doxycycline (Dox)-dependent manner. In these cells, Dox withdrawal promptly decreased the protein level of MBP-trichoplein WT, but not K50/57R, in a polyubiquitylation-dependent manner upon serum starvation (Fig. 2d,e), as was the case with myc-trichoplein in cycloheximide-treated RPE1 cells (Fig. 2a,b), eliminating the concern that the differences in regent (cycloheximide versus Dox) and protein tag affected the trichoplein degradation in RPE1 cells.
In the presence of Dox, both overexpressed WT and K50/57R comparably activated Aurora-A and suppressed the serum starvation-induced ciliogenesis (Fig. 2f,g). Twenty-four hours after Dox withdrawal, WT level was drastically decreased to 10% and thereby resulted in Aurora-A inactivation ( Fig. 2h; lanes 3 and 4), whereas 70% of K50/57R remained present and kept on activating Aurora-A ( Fig. 2h; lanes 1 and 2). Under such conditions, cilia were newly formed only in WT cells ( Fig. 2i; compare WT and K50/57R). These results collectively indicate that UPS-mediated proteolysis of trichoplein is essential for ciliogenesis.   ARTICLE trichoplein degradation, we carried out the two-stepped global E3 screen (Fig. 3a). In the primary screen, 1,172 E3 ligase proteins (including putative E3s; listed in Supplementary Table 1) were purified from the human proteome expression resource library (HuPEX) using wheat germ cell-free expression system 38,39 . We evaluated their binding to bacterially purified MBP-trichoplein by protein array, and identified ten potential E3 ligases ( Supplementary Fig. 3). We then performed the secondary screen using four distinct small interfering RNAs (siRNAs) per potential E3 ligase; depletion of KCTD17 (K þ channel tetramerization domain-containing 17) protein by four independent siRNAs considerably interfered with the serum  starvation-induced ciliogenesis both in RPE1 cells and IMR-90 fibroblasts ( Fig. 3b and Supplementary Fig. 4). Immunofluoresence analyses of cell cycle markers, including cyclin A and BrdU incorporation, exclude the possibility that the defective ciliogenesis is due to the inappropriate cell cycle re-entry ( Supplementary Fig. 5). In addition, KCTD17 silencing attenuated the serum starvation-induced polyubiquitylation of trichoplein (Fig. 3c). Thus, KCTD17 is a novel regulator of ciliogenesis that involves in trichoplein polyubiquitylation.
KCTD17 controls ciliogenesis via trichoplein and Aurora-A. As noted above, serum starvation caused the trichoplein removal from mother centriole through its proteolysis in control RPE1 cells, but these were prevented in KCTD17-depleted cells (Fig. 4a,b). KCTD17 silencing also disrupted the serum starvation-induced inactivation of centriolar Aurora-A (Fig. 4a,b), as trichoplein is a centriolar activator of Aurora-A 29   caused by KCTD17 depletion, namely the defective ciliogenesis (Fig. 4c) and the deregulated trichoplein-Aurora-A pathway ( Supplementary Fig. 9). Similar results were also obtained in MBP-trichoplein-expressing Tet-On RPE1 cells (Fig. 2h,i; compare WT and WT þ KCTD17 siRNA). HEF-1 is also reported to regulate ciliary dynamics by activating Aurora-A 24,33 , but its protein levels were unchanged by KCTD17 depletion (Fig. 4b).
These results indicate that KCTD17 contributes to trichoplein degradation and thereby inactivates Aurora-A to induce ciliogenesis. It has been shown that trichoplein has numerous binding proteins other than Aurora-A 28,30 and its overexpression causes the mitochondrial fragmentation in some tumour cells 31,32 (see also Introduction). We found little changes in their protein levels ( Supplementary Fig. 10) and the mitochondrial morphology ( Supplementary Fig.11) by overexpression, depletion or stabilization of trichoplein in RPE1 cells under our experimental conditions, suggesting that trichoplein may have the cell type-specific functions. Furthermore, the defective ciliogenesis caused by KCTD17 depletion was completely reversed by co-silencing of trichoplein or Aurora-A (Fig. 4d). Thus, we propose that the KCTD17-trichoplein-Aurora-A pathway controls ciliogenesis in RPE1 cells.
KCTD17 depletion blocks extension of ciliary axoneme. To get more insight how KCTD17 regulates ciliogenesis, we analyzed mother centrioles and ciliary structures of serum-starved RPE1 cells lacking KCTD17 by transmission electron microscopy. Ciliogenesis follows a series of stereotyped steps that begin with the docking of CVs to the distal appendages of mother centrioles, followed by the formation and extension of ciliary axoneme 5,6 ( Supplementary Fig. 12a). After 24 h serum starvation, control cells often demonstrated the mother centrioles that were associated with the ciliary pockets with extended axonemal shafts (Fig. 5a). In contrast, as approximately 90% of KCTD17depleted cells lacked the elongated ciliary axoneme (see Fig. 3b; KCTD17 siRNA #1), in most of these cells (12 out of 16), the CVs efficiently docked to the mother centrioles, judged by the appearance of appendages, and became invaginated by the accumulated electron-dense materials at the distal end of centrioles (named the ciliary buds; Fig. 5b and Supplementary  Fig. 12b). No gross alteration was detected in the ultrastructure of mother centriole. We therefore conclude that KCTD17 is not required for the maturation of mother centriole and the centrioleto-membrane docking, but instead, plays a crucial role in the initial step of axoneme extension during ciliogenesis.

Discussion
Trichoplein localizes to centrioles and suppresses the aberrant ciliogenesis by activating centriolar Aurora-A in growing cells, but disappears from the mother centrioles when cells are exposed to cell cycle signals that induce ciliogenesis by serum starvation 29 . In this study, we show that the trichoplein removal depends on UPS and that CRL3-KCTD17 targets trichoplein to proteolysis through polyubiquitylation at Lys-50 and Lys-57 during ciliogenesis. If trichoplein is stabilized at mother centrioles by treatment with proteasome inhibitors, expression of nonubiquitylatable trichoplein mutant (K50/57R) or KCTD17 depletion, in any of these cases, ciliogenesis is blocked. Thus, the trichoplein degradation is an essential event for ciliogenesis. We also demonstrate that Aurora-A silencing completely reverse the defective ciliogenesis caused by trichoplein stabilization, indicating that the KCTD17-trichoplein-Aurora-A cascade controls ciliogenesis at the initial step of axoneme extension (summarized in Fig. 5c).
CP110 is also thought to inhibit the axonemal extension from the mother cenrioles 8,15,18,19 . Similar to trichoplein, CP110 disappears from mother centrioles during ciliogenesis and this removal is essential. However, there are numerous differences between trichoplein and CP110. In contrast to trichoplein localization at the subdistal/medial zone of centrioles 30 , CP110 localizes at the distal end of centrioles and structurally caps the centriole microtubles 14 . Therefore, siRNA-mediated depletion of CP110, but not trichoplein, induces the elongation of centriolar microtubules in non-ciliated cells 12,30,45 . In addition, the CP110 loss is accompanied by a recruitment of the protein kinase ARTICLE TTBK2 to the distal end of mother centriles 19 , but do not depends on proteasomal degradation (Fig. 1c), during ciliogenesis. Whether or not trichoplein-mediated and CP110-mediated regulations are mutually interrelated in axoneme extension is an interesting area for further experimentation. CRL3-KCTD17 targets trichoplein to proteolysis in response to serum starvation, but the CRL3-KCTD17 protein levels were unchanged ( Supplementary Fig. 13a). Serum starvation had less effect on the myc-KCTD17 localization (Supplementary Fig. 13b) and the trichoplein-KCTD17 interaction ( Supplementary  Fig. 6b). Thus, serum starvation-induced trichoplein degradation appears to require additional mechanisms. One possibility is that CRL3-KCTD17 activity may be modulated through posttranslational modification like phosphorylation by MARK4 or TTBK2 (refs 18,19). Alternatively, an unidentified deubiquitylating enzyme that counteracts the trichoplein polyubiquitylation may be active in growing cells and become inactive upon serum starvation. It will be future work to investigate whether the trichoplein degradation is regulated by these protein kinases or deubiquitylating enzyme and its physiological contribution to axoneme extension during ciliogenesis.
In vivo ubiquitylation assay. One day after transfection, cells were lysed in the denaturing condition with hot ubiquitin buffer (95°C) containing 25 mM Tris-HCl (pH 8.0), 1.5% SDS, 0.15% sodium deoxycholate, 0.15% NP-40, 1 mM EDTA, 1 mM okadaic acid and 5 mM N-ethylmaleimide. The lysates were diluted (1:10) with cell lysis buffer and used for immunoprecipitation with anti-trichoplein (for endogenous trichoplein), anti-Myc (for myc-trichoplein) and anti-Flag (for MBP-trichoplein-flag). Immunoprecipitates were washed three times with cell lysis buffer containing the denaturing detergent SDS (0.1%), and then analyzed by SDS-polyacrylamide gel electrophoresis. To prevent the proteasomal degradation of polyubiquitylated trichoplein constructs, we treated cells with 10 mM MG132 in the presence or absence of serum for 6 h before cell lysis.
E3 ubiquitin ligase array. cDNA clones used to screen for E3 ubiquitin ligase to trichoplein were selected from the HuPEX library 39,48 . We selected 744 genes to be human E3 ubiquitin ligase by a keyword search of the HGPD (http:// www.HGPD.jp/) 38 , Swiss-Prot database and few papers [49][50][51] . We have 622 genes to these 744 genes, which were corresponding to 1,172 HuPEX clones containing variant clones (Supplementary Table 1). These HuPEX clones were transferred to the pEW-FG expression vector 39 harbouring FLAG and GST tags by using the Gateway LR reaction. The protein synthesis was performed by the method of wheat germ protein expression system according to the manufacturer instructions 39 (CellFree Sciences). Protein array system by using magnetic beads is indicated in Supplementary Fig. 4a. The magnetic beads with glutathione ligand (Promega) were added to reaction mixture containing synthesized protein having FLAG-GST tag, and the synthesized protein was absorbed on the surface of magnetic beads. Beads adsorbed synthesized protein dispensed on the magnet plate (MEIHO Co. Ltd.), which have magnet at the bottom of each well. Each E3 ubiquitin ligase was immobilized at the bottom of magnetic plate via magnetic beads in solution. By using this E3 ubiquitin ligase array, we screened the specific E3 ubiquitin ligase for trichoplein. The array was blocked with PBST and 1% BSA (PBST þ BSA). The array was incubated for 1 h at 25°C with MBP-trichoplein (1.0 mg ml À 1 in PBST þ BSA), then washed in PBST. The array was incubated with HRPconjugated anti-MBP tag sheep antibody (GE Healthcare; diluted 1:1,000 in PBST þ BSA), washed in PBST. Reactive spots were detected with ECL plus (GE Healthcare). The quantity of protein on each spot was assayed by using FLAG M2-Peroxidase (HRP)-conjugated monoclonal antibody (Sigma-Aldrich). Reactive spots were normalized by each quantity of protein on each spot.
Immunoblotting. We performed the immunoblotting as described 46,47 . For immunoblotting, we produced a rabbit polyclonal anti-KCTD17 antibody by immunizing bacterially purified KCTD17 (193-297aa) protein (MBL) and a rabbit polyclonal antibody for trichoplein 29 . The list of antibodies with source and conditions of immunoblotting is shown in Supplementary Table 2. In some immunoblotting experiments, we used Can Get Signal immunoreaction enhancer solutions (TOYOBO) for dilution of primary and secondary antibodies. Band intensities were analyzed by densitometry (ImageJ 1.43r, for Macintosh OS X; National Institute of Health, Bethesda, MD, USA). Uncropped versions of the most important blots are shown in Supplementary Figs 14 and 15.
Microscopy. Immunofluoresence microscopy was performed using confocal microscopy (LSM510 META; Carl Zeiss) equipped with a microscope (Axiovert 200 M; Carl Zeiss), a plan Apochromat Â 100/1.4 NA oil immersion lens and LSM image Browser software (Carl Zeiss) as described 29 with slight modifications. The list of antibodies with source and conditions of indirect immunofluorescence is shown in Supplementary Table 2. In some immunofluorescence experiments, we used Can Get Signal immunostain Solution A (TOYOBO). For induction of primary cilia, RPE1 and IMR-90 cells were seeded on sterile coverslips, grown for 1 day and then subjected to serum starvation. For detection of primary cilia, cells were placed on ice for 20-30 min before fixation with cold methanol, and stained with anti-acetylated tubulin antibody 29 . BrdU incorporation was evaluated using DNA Replication Assay Kit (Millipore). Quantification of fluorescence intensity was performed using ImageJ 1.45r software.
Transmission electron microscopy. RPE1 cells cultured on coverslips were fixed with 2% fresh formaldehyde and 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) for 2 h at room temperature. After washing with 0.1 M cacodylate buffer (pH 7.4), they were postfixed with ice-cold 1% OsO 4 in the same buffer for 2 h. The samples were rinsed with distilled water, stained with 0.5% aqueous uranyl acetate for 2 h or overnight at room temperature, dehydrated with ethanol and propylene oxide, and embedded in an embedding kit (Poly/Bed 812; PolySciences, Inc.). After removal of coverslips using ice-cold hydrofluoric acid, ultra-thin sections were cut, doubly stained with uranyl acetate and Reynolds's lead citrate, and viewed with a transmission electron microscope (JEM-1010; JEOL) with a charge-coupled device camera (BioScan model 792; Gatan, Inc.) at an accelerating voltage of 100 kV.
SiRNAs. Transfection of siRNA duplexes (final concentration, 20 nM) was performed with Lipofectamine RNAiMAX reagent according to the reverse transfection protocol (Invitrogen). All siRNA duplexes were purchased from Qiagen. Target sequences are listed in Supplementary Table 3.