Ciliary membrane proteins traffic through the Golgi via a Rabep1/GGA1/Arl3-dependent mechanism

Primary cilia contain specific receptors and channel proteins that sense the extracellular milieu. Defective ciliary function causes ciliopathies such as autosomal dominant polycystic kidney disease (ADPKD). However, little is known about how large ciliary transmembrane proteins traffic to the cilia. Polycystin-1 (PC1) and -2 (PC2), the two ADPKD gene products, are large transmembrane proteins that co-localize to cilia where they act to control proper tubular diameter. Here we describe that PC1 and PC2 must interact and form a complex to reach the trans-Golgi network (TGN) for subsequent ciliary targeting. PC1 must also be proteolytically cleaved at a GPS site for this to occur. Using yeast two-hybrid screening coupled with a candidate approach, we identify a Rabep1/GGA1/Arl3-dependent ciliary targeting mechanism, whereby Rabep1 couples the polycystin complex to a GGA1/Arl3-based ciliary trafficking module at the TGN. This study provides novel insights into the ciliary trafficking mechanism of membrane proteins.

P rimary cilia are microtubule-based non-motile projections on the apical surface of cells, which organize important signalling pathways mediated by specific receptors and channel proteins at the ciliary membrane in response to mechanical and chemical stimuli 1 . They are the key organelle for controlling correct tubular diameter, and defective ciliary function causes a variety of ciliopathies 1 , including autosomal dominant polycystic kidney disease (ADPKD) 2,3 . The ciliary membrane is separated from the plasma membrane by a periciliary diffusion barrier 4 , and ciliary membrane proteins must be transported to the cilia from their site of synthesis in the rough endoplasmic reticulum (ER) for proper ciliary function 5 . Experimental evidence favours a targeted delivery model, whereby ciliary membrane proteins are sorted in the Golgi and are targeted to the cilium by the vesicular pathway 5 . Numerous protein complexes have been implicated in polarized trafficking of post-Golgi vesicles to cilia, including the BBSome 5 and intraflagellar transport complexes 6 . However, the mechanism by which transmembrane proteins are sorted for ciliary trafficking is poorly understood.
Polycystin-1 (PC1) and polycystin-2 (PC2), the two ADPKD gene products 7,8 , are large transmembrane proteins that colocalize to cilia 9,10 where they play an important role in calciumbased signalling 2,11 . Loss of polycystin function in cilia is implicated in cyst formation 2 . PC1 is a 4,302-amino-acid atypical adhesion G-protein-coupled receptor (aGPCR) with 11transmemrbane domains 12,13 . A fundamental property of PC1 is cis-autoproteolytic cleavage at a juxtamembrane G-proteincoupled receptor cleavage site (GPS) [14][15][16] . GPS cleavage of PC1 is developmentally regulated in the kidney, whereby PC1 is largely uncleaved in early embryonic kidneys but becomes extensively cleaved after birth 17,18 . Mice with a 'knock-in' missense amino-acid substitution that disrupts GPS cleavage (Pkd1 V/V mice) express non-cleavable PC1 (PC1 V ) and develop cystic kidney disease during the postnatal period 15 . As shown for other aGPCRs, cleavage results in a heterodimeric PC1 form, in which the N-terminal fragment (PC1 NTF ) remains stably and non-covalently associated with the transmembrane C-terminal fragment (PC1 CTF ) 14,19 . More recently, we have shown that GPS cleavage generates a complex pattern of endogenous PC1 forms but is not per se required for PC1 to move to the Golgi 18 . How GPS cleavage affects trafficking and function of PC1 or other aGPCRs in vivo remains unclear.
PC2 is a 968-amino-acid-long 6-transmembrane-spanning transient receptor potential channel family member that acts as a calcium release channel and is most abundantly distributed to the ER 20,21 . PC1 and PC2 form a receptor/channel complex by direct interaction via coiled-coil domains in their cytoplasmic C termini [22][23][24] . Several studies have reported that this interaction is required for surface membrane localization of the complex in certain, but not in all, cell types [25][26][27] . However, there is disagreement as to whether the interaction is necessary for ciliary localization, as each protein has its own ciliary targeting signal 25,28,29 . In some studies, PC2 was able to localize to cilia independently of PC1 28,30 , while other studies show that this requires PC1 2,25,31,32 . In addition, PC1 and PC2 may take different routes to reach the cilium. PC1 is described to traffic to cilia from the trans-Golgi network (TGN) via post-Golgi vesicles in an Arf4-dependent process 29 . However, PC2 was reported to move to the cilium directly from the cis-Golgi compartment without traversing the Golgi apparatus 30 .
In this report, we use a comprehensive approach to define the ciliary trafficking mechanism of endogenous polycystins in polarized ciliated renal epithelial cells. We demonstrate that PC1-PC2 interaction and GPS cleavage of PC1 are both required for the polycystin complex to reach the TGN and for subsequent ciliary targeting. We use yeast two-hybrid screening coupled with a candidate approach to identify a novel protein complex composed of Rabep1, GGA1 and Arl3, which is responsible for the sorting and targeting of the polycystin complex to the cilium. Our study provides novel insights into the ciliary trafficking mechanism of transmembrane proteins, with implications for ADPKD, the most common human ciliopathy.

Results
Polycystin complex is required for ciliary localization. We used several strategies to determine whether native PC1 and PC2 might regulate each other's ciliary localization in vivo. We found that PC1 localizes to cilia in wild-type mouse embryonic fibroblasts (MEFs), but not in Pkd2 À / À MEFs (Fig. 1a,c). PC2 also localizes to cilia in wild-type MEFs, but not in Pkd1 À / À MEFs (Fig. 1b,c). Furthermore, exogenous mouse PC1 expressed in Pkd1 À / À MEF cells trafficked to the cilia and concomitantly restored ciliary localization of endogenous PC2 (Fig. 1d). The rescue of ciliary PC2 localization was not seen when GFP was expressed ( Supplementary Fig. 1). We then used shRNA to knock down endogenous Pkd2 expression in DBA-positive collecting duct (CD)-derived cells 15 , and we found that the ciliary localization of PC1 was abolished (Fig. 1e). Likewise, the ciliary localization of PC2 was not detectable when Pkd1 expression was knocked down (Fig. 1f). The interdependence of PC1 and PC2 ciliary localization was confirmed in IMCD cells with stable expression of full-length epitope-tagged mouse PC1 (IMCD PC1WT , Supplementary Fig. 2a-e). In mice with a floxed Pkd1 allele, PC2 was absent in the cilia of cystic kidney tubules  Supplementary Fig. 5) and analysed by confocal microscopy (a), or stained with anti-PC2 and analysed by epifluorescence microscopy (b). Arl13b or acetylated (Ac)-tubulin was used as a ciliary marker. Arrows indicate cilia. (c) The bar diagram shows the quantification of PC1 or PC2 ciliary localization in MEFs with various genotypes with numbers of cells analysed (n) indicated; ***Po0.001. (d) Exogenously expressed PC1 traffics to cilia (arrow) and rescues ciliary localization of PC2 in Pkd1 À / À MEFs. Exogenous expression of GFP served as a negative control ( Supplementary Fig. 1). (e) Confocal images of CD cells (upper panel) and CD cells expressing Pkd2 shRNA (lower panel). Cells were stained with antibodies against Arl13b (blue) and PC1(E4) (red). The bar diagram below shows the quantification of PC1 ciliary localization in CD and cells with Pkd2 knockdown performed from four independent experiments with numbers of cells analysed (n) indicated; ***Po0.001. Western blot analysis shows that Pkd2 shRNA reduced endogenous PC2 level by B90%. (f) Confocal images of CD cells (upper panel) and cells expressing Pkd1 shRNA, stained with antibodies as indicated. Four different Pkd1 shRNAs had similar results. The bar diagram shows the quantification of PC2 ciliary localization between CD and cells with Pkd1 knockdown from four independent experiments; ***Po0.001. Western blot analysis shows that Pkd1 shRNA reduced endogenous PC1 level by B90%. The white line in the XY scan in (e,f) indicates the path of the XZ scan. (g) Confocal images of PC2 ciliary localization in collecting duct at P28 in Pkd1 cko/cko :tamoxifen-Cre mouse kidney 33 , in which Pkd1 was not inactivated (left panel, normal tubule) or inactivated by tamoxifen injection on P10 (right panel, cystic tubule). PC2 was detected in 21 of 21 cilia (red) in normal kidneys. Cilia were marked by acetylated tubulin (green). Note that PC2 was not detectable in 18 of 18 cilia of cysts analysed. Scale bar, 10 mm. after postnatal Pkd1 inactivation 33 , while the protein was detected in the cilia of the normal tubular epithelial cells (Fig. 1g).
We then tested whether direct interaction of PC1 and PC2 was required for ciliary localization of the polycystin complex. In MDCK cells with stable and inducible expression of epitopetagged wild-type mouse PC1 (MDCK PC1WT , Table 1), we confirmed that recombinant PC1 formed a complex with endogenous PC2 (Fig. 2a,b). We found that, on induction of its expression, PC1 trafficked to cilia and concomitantly induced ER-resident endogenous PC2 to translocate to cilia as well ( Fig. 2c-e). We disrupted the interaction of PC1 and PC2 by introducing amino-acid substitutions at two strategic positions (L4219P and A4222P) within the coiled-coil domain 22,24 (PC1 2M ) (Fig. 2a,b). In contrast to wild-type PC1 (Fig. 2f), we found that PC1 2M did not traffic to the cilia and did not induce endogenous PC2 to translocate to the cilia (Fig. 2g,i). Similarly, PC1 R4218X (R/X) mutant 34 , which corresponds to an ADPKD patient-derived truncating mutant R4227X lacking the intact  coiled-coil domain 22,27 , did not interact with endogenous PC2 (Fig. 2a,b) and failed to cause PC2 to translocate to cilia (Fig. 2h,i). Taken together, our results show that PC1 and PC2 must interact and form a complex for ciliary localization to occur in renal tubular epithelial cells.
Polycystin complex is required to reach the Golgi apparatus. We next analysed the N-glycosylation pattern of the polycystin complex to investigate how it travels through the cell. EndoH cleaves high-mannose N-glycans attached in the ER, but not the medial/trans-Golgi-modified complex N-glycans 35 . EndoH resistance is thus indicative that glycoproteins have reached the medial/trans-Golgi compartment. Endogenous PC1 was immunoprecipitated with anti-CC (Fig. 2a) from MEFs, and the resulting precipitate was treated with EndoH followed by analysis with SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and western blotting. In wild-type MEFs, PC1 was predominantly present as GPS-cleaved forms, which are recognized as the EndoH-resistant ('Cleaved-R') and -sensitive ('Cleaved-S') PC1 NTF bands (Fig. 3a, upper panel, lanes 2-4/2 0 -4 0 ) as previously described 18 . Under these conditions, we observed very low levels of uncleaved PC1. In the whole-cell lysate, PC2 was detected as a single 120-kDa band (lower panel, lane 1/1 0 ) as previously shown 20,21 . However, the fraction of PC2 that is coimmunoprecipitated with PC1 appeared as two distinct bands of similar intensity (lane 2/2 0 ). Treatment of this PC2 fraction with EndoH revealed that the upper band (130-kDa, PC2 130 ) was EndoH resistant, whereas the lower band (120-kDa, PC2 120 ) was EndoH sensitive (lane 4/4 0 ). A similar pattern was found in both kidney tissues and CD cells (Fig. 3b). These results identify a significant amount of EndoH-resistant pool of polycystin complex in vivo. In Pkd2 À / À MEFs, GPS-cleaved PC1 remained entirely EndoH-sensitive (Fig. 3a, upper panel, lanes 5-7), implying that PC1 cannot reach the trans-Golgi without PC2. Likewise, in CD cells with Pkd2 knockdown, cleaved PC1 remains EndoH sensitive (Fig. 3c). Together, these data indicate that PC1 and PC2 form a complex in the ER and that direct interaction is required for the complex to reach the Golgi apparatus.
Polycystin complex traffics to cilia through the Golgi. To determine intracellular trafficking of ciliary PC1 and PC2, we isolated intact cilia from MDCK cells similarly as previously described 36 (Fig. 4a) and analysed their N-glycosylation patterns. The preparation contained intact cilia as visualized by IF microscopy with a ciliary marker (Fig. 4b). Moreover, western blot analysis of this ciliary preparation detected acetylated tubulin and Polaris (known ciliary proteins), but not b-actin (an abundant cytoskeletal protein), Alix (an abundant component of exosome proteomes) 37 , c-Met (an apical plasma membrane protein) 38 or IGFBP-2 (a secreted protein) 39 (Fig. 4b). These results thus validated that the cilium preparation did not contain a detectable amount of contamination from intracellular and secreted proteins, exosomes or plasma membrane. We detected a single B450-kDa PC1 band in the cilium preparation from induced MDCK PC1WT cells (Fig. 4c, upper panel, lane 3), but not from non-induced cells (lane 2) or MDCK pcDNA5 control cells (data not shown). The ciliary PC1 migrated between the uncleaved PC1 and cleaved PC1 detected in the same cells stripped of cilia (cell body) (lane 1). Ciliary PC1 was resistant to EndoH (lane 3), but migrated at the expected molecular weight (MW) of cleaved PC1 NTF (B370 kDa) 14 after removal of N-glycan by PNGaseF treatment (Fig. 4d, upper panel, lanes 2 and 5). This result identified the ciliary PC1 as the cleaved EndoH-resistent form. The uncleaved PC1 was absent in the cilia, but it was present at similar levels as cleaved PC1 in the cell body.
These results indicate that only cleaved PC1 reaches the cilia and that it traffics through the Golgi. Remarkably, we detected a PC2 band of a MW of B130 kDa in the cilium preparation from induced MDCK PC1WT cells (Fig. 4c, lane 3), which migrated slower than the B120-kDa PC2 band detected in the whole-cell body lysate (lane 1). The ciliary PC2 was also resistant to EndoH (Fig. 4d, lane 6) and migrated at the expected MW of PC2 after PNGaseF treatment (lane 5), as the PC2 130 seen in endogenous polycystin complex from native tissues and cells (Fig. 3). Taken together, the presence of exclusively EndoH-resistant PC1 and PC2 in the cilia provides direct biochemical evidence that the polycystin complex traffics to cilia through the Golgi apparatus.
Ciliary trafficking of polycystins requires GPS cleavage. The absence of uncleaved PC1 in the cilia suggests that GPS cleavage is required for ciliary trafficking of the polycystin complex. To test this we asked whether the non-cleavable PC1 V could be found in cilia. In fact we were unable to detect either ciliary PC1 V or PC2 in the cilia of CD cells derived from Pkd1 V/V kidneys (CD V/V ) (Fig. 5a). In addition, recombinant PC1 V overexpressed in MDCK (MDCK PC1V , Table 1) was unable to reach the cilia and did not induce endogenous PC2 to translocate to cilia (Fig. 5b). We obtained similar results in IMCD PC1V cells ( Supplementary Fig. 2f). Consistent with this result, we were unable to detect either PC1 V or PC2 in the cilium preparation from induced MDCK PC1V cells by western blot analysis (data not shown). Importantly, PC2 was not detectable in the cilia of cystic tubules of the Pkd1 V/V kidneys (Fig. 5c).
To investigate the basis of the observed requirement of GPS cleavage, we analysed the N-glycosylation pattern of the polycystin complex in Pkd1 V/V MEFs versus wild-type MEFs (Fig. 5d). The non-cleavable PC1 V was immunoprecipitated from Pkd1 V/V MEFs as an B600-kDa EndoH-resistant band (PC1 V-600 ) and an B520-kDa EndoH-sensitive band (PC1 V-520 ) (upper panel, lanes 4-6). We found that PC2 was still coprecipitated by PC1 V , but the co-precipitated PC2 appeared only as an B120-kDa band and was completely sensitive to EndoH (that is, PC2 120 , lanes 4-6), lacking the EndoH-resistant PC2 130 seen in WT MEFs (lanes 1-3). The same N-glycosylation pattern was found for the polycystin complex in Pkd1 V/V kidney tissues (Fig. 5e). These results imply that only the EndoH-sensitive PC1 V-520 , but not the EndoH-resistant PC1 V-600 , is associated with PC2 in the Pkd1 V/V samples. Our data collectively show that GPS cleavage is required for the ciliary localization of the polycystin complex by enabling its trafficking to the trans-Golgi compartment. In the absence of cleavage, the PC1 V /PC2 complex cannot reach the trans-Golgi.
Rabep1 binds PC1 C-terminus for ciliary trafficking. To identify the molecules that mediate polycystin complex ciliary trafficking, we performed a yeast two-hybrid screen for PC1interacting proteins using the entire 215-amino-acid C-terminal tail as bait. Thirteen unique cDNA clones were identified that specifically interact with the PC1 C-terminus. One of these clones corresponded to amino acids 465 to 799 of Rabep1 (Fig. 6a), an effector of multiple Rab GTPases involved in various steps of intracellular vesicular trafficking 40 . This protein was chosen for further analysis. We used mutational analysis to map the Rabep1 interaction region (Fig. 6b) and showed that the C-terminal 89 amino acids containing the intact coiled-coil domain of PC1 (PC1-M1/M2) was sufficient for interaction with Rabep1. The binding sites for Rabep1 and PC2 overlap but are not identical, as the C-terminal 41 amino acids of PC1 (PC1-M3) was required for Rabep1 interaction, but not for PC2.
We confirmed that the PC1-Rabep1 interaction occurs in vivo by demonstrating co-immunoprecipitation from CD cells (Fig. 6c). We were unable to co-immunoprecipitate these two proteins from CD cells with Pkd1 knockdown (Fig. 7a) nor from Pkd1 DCMYC/DCMYC knockout embryos expressing Myc-tagged truncated PC1 (ref. 41) (Fig. 7b). We found that the PC1-Rabep1 interaction occurred in Pkd2-null MEFs (Fig. 7c). This suggests that PC1 binds Rabep1 in a pre-Golgi compartment, as PC1 does not exit the ER in the absence of PC2 (Fig. 3a).
Next we tested whether Rabep1 was required for ciliary localization of the polycystin complex. We used shRNA to knock down Rabep1 in CD cells, and this abolished ciliary localization of PC1 and PC2 (Fig. 7d,e). Rabep1 knockdown in the CD cells, however, did not affect formation of the polycystin complex nor its acquisition of EndoH resistance (Fig. 7f, lane 2), indicating that Rabep1 is not required for the polycystin complex to reach the Golgi. Collectively, these results show that Rabep1 binds to the distal portion of PC1 CTF 's C-terminal tail and suggest that Rabep1 likely plays a critical role for ciliary targeting of the polycystin complex at the TGN by recruiting ciliary targeting machinery.
Rabep1 recruits polycystin complex to GGA1 and Arl3 at TGN. Rabep1 has been shown to interact with GGA1 (refs 42-44) (Golgi-localized, gamma adaptin ear-containing, ARF-binding) in the context of fusion of TGN-derived vesicles with endosomes 45 . GGA1 is an effector of Arf GTPases at the TGN and mediates protein sorting and vesicle budding from the TGN 43,46 . We therefore hypothesized that polycystin complex-associated Rabep1 might bind GGA1, which in turn might mediate the ciliary sorting of the complex. Indeed, we found that PC1 can coimmunoprecipitate GGA1, along with PC2 and Rabep1, from CD cell lysates (Fig. 7a,f) and Pkd1 MYC/MYC embryos (Fig. 7b). GGA1 knockdown abolished the ciliary localization of both PC1 and PC2 in CD cells (Fig. 7d,e), without affecting the acquisition of EndoH resistance of the polycystin complex (Fig. 7f, lane 3). PC1-GGA1 interaction was abolished by Rabep1 knockdown (lane 2), whereas PC1-Rabep1 interaction still occurred when GGA1 was knocked down (lane 3), indicating that Rabep1 bridges an interaction between PC1 and GGA1. Our data provide evidence that GGA1 is recruited to the polycystin complex by Rabep1 and plays a critical role for subsequent ciliary targeting from the TGN.
GGA proteins are recruited to the TGN membrane via their direct interaction with specific Arf members 43,46 . As GGA1 affected by the known Arfs such as Arf1 (ref. 47) or Arf3 (ref. 42) has not been shown to be involved in ciliary trafficking, we reasoned that additional small GTPases would likely be required to confer specificity of polycystin complex ciliary targeting. A previous study using a small CD16.7-PC1 chimeric construct encoding the last 112 amino acids of the PC1 cytoplasmic C-terminus 29 has implicated Arf4 (ref. 48) in the ciliary trafficking of PC1. We have asked whether Arf4 may cooperate with GGA1 at the polycystin complex. However, we were unable to detect Arf4 in the PC1 complex from the CD cell lysates (Fig. 6c). Arl3, a closely related member of the Arf family, localizes to the Golgi membrane and other microtubule-related structures including cilia 49,50 . Remarkably, Arl3 inactivation in mice results in cystic kidney disease 51 , reminiscent to what is found in Pkd1 V/V mice lacking PC1 GPS cleavage 15 . We thus hypothesized that Arl3 binds GGA1 on polycystin complex to regulate its ciliary targeting. Indeed, PC1 co-immunoprecipitated both Arl3 and GGA1, along with Rabep1 and PC2 (Fig. 7a,b). Importantly, Arl3 knockdown in CD cells abolished the ciliary localization of PC1 and PC2 (Fig. 7d,e), indicating a critical role of Arl3 for the ciliary targeting of the polycystin complex.
We then analysed mutual requirements of Arl3, GGA1 and Rabep1 for PC1 binding. Rabep1 knockdown disrupted the interaction of PC1 with both GGA1 and Arl3 (Fig. 7f, lane 2), whereas the PC1-Rabep1 interaction still occurred when GGA1 or Arl3 was knocked down (lane 3 or 4). Arl3 knockdown, however, did not disrupt the interaction of PC1 with Rabep1 or GGA1 (lane 4). These results raised the possibility that Rabep1 may recruit a GGA1/Arl3 complex to the polycystin complex by interacting with GGA1. In support of this notion, we found that endogenous GGA1 and Arl3 can be co-immunoprecipitated from CD cells depleted of Rabep1 (Fig. 7g) or in Pkd1 À / À MEFs The schematic diagram at right provides an identification guide for various PC1 and PC2 forms, with colour code that is maintained throughout the figures. Note that PC2 is co-immunoprecipitated from WT MEFs as EndoH-resistant PC2 130 and EndoH-sensitive PC2 120 ; and cleaved PC1 cannot acquire EndoH resistance in Pkd2 À / À MEFs. (b) N-glycosylation analysis of endogenous polycystin complex in kidneys (left) and CD cells (right) as in (a). Note that EndoH-resistant PC2 130 is detected in the polycystin complex in the kidney and CD lysates. (c) N-glycosylation analysis of polycystin complex in CD cells with Pkd2 knockdown. Note that Pkd2 knockdown prevented cleaved PC1 to acquire EndoH resistance. EndoH-sensitive uncleaved PC1 (B520 kDa) is visible as in Pkd2 À / À MEFs. (Fig. 7h). Furthermore, when retained in a pre-Golgi compartment in Pkd2 À / À MEFs, PC1-Rabep1 complex did not bind GGA1 or Arl3 (Fig. 7c). Collectively, our data support a model in which Rabep1 recruits the polycystin complex to GGA1/Arl3 at TGN for ciliary trafficking.

Discussion
To study ciliary membrane protein trafficking, we focused on two large transmembrane proteins, PC1 and PC2, the products of the genes mutated in human ADPKD. Our study shows that polycystin trafficking to the cilium occurs in a two-step process. First, PC1 and PC2 must interact and form a complex to traffic from the ER to the Golgi. In addition, PC1 must be cleaved at GPS for this to occur. Rabep1 binds the polycystin complex at PC1 C-terminal tail and rides on the complex to the Golgi, but is not itself required for this step in trafficking. Second, once the polycystin complex-bound Rabep1 reaches the TGN, it binds GGA1 that is associated with Arl3 for subsequent targeting to the cilium.
Our finding that a direct PC1-PC2 interaction is required for ciliary trafficking is consistent with some 2,25 , but differs from other published studies 28,30 . Some of the discrepant findings may be due to the use of overexpression systems by several authors, while we have focused on analysing trafficking of endogenous proteins. It has been previously proposed that PC2 is continuously released from the ER to the Golgi in a COPII-dependent manner but immediately returned via an ER retention/retrieval signal in the C-terminus 20,30,52 . A more recent study reported that a small portion of PC2 escapes retrograde transport and traffics to the cilium directly from the cis-Golgi compartment without traversing the Golgi apparatus 30 . This was inferred in part from the complete EndoH sensitivity of PC2 at the whole-cell lysate level. We used co-immunoprecipitation to isolate the fraction of PC2 that is complexed with PC1 from native tissue and cell lysates, and found that a significant portion was EndoH resistant (PC2 130 ). PC2 130 is present in a rather small fraction of the total cellular PC2 and has not been recognized previously when analysed at whole-cell lysate levels, probably because it is obscured by the much more abundant EndoH-sensitive ER-resident PC2 120 that migrates at a very close position on western blots. Our co-immunoprecipitation strategy simultaneously enriches the native polycystin complex and removes the predominant interfering EndoH-sensitive PC2 120 form that is not complexed with PC1, thereby allowing the visualization of PC2 130 of the polycystin complex. Our finding of exclusive EndoH resistance of both PC1 and PC2 in isolated cilia provided direct biochemical evidence that the polycystin complex traffics to cilia in fact through the Golgi apparatus. The ciliary polycystins may represent some of the EndoH-resistant polycystin complex population detected in native kidney tissues and cells. However, given the relatively small volume occupied by the cilium, most of this polycystin population probably resides on non-ciliary compartments, such as the Golgi or post-Golgi vesicles, or at the plasma membrane 18 . Further studies are required to resolve this issue.
PC1 is generally far less abundant than PC2 in kidneys and is the central determinant of cyst formation in polycystic kidney and liver diseases 53,54 . We find that ectopic expression of PC1 can induce the ER-resident endogenous PC2 to translocate to the cilium. These data support the idea that PC1 is the rate-limiting factor for ciliary trafficking of the polycystin complex in renal epithelial cells. We propose that the PC1-PC2 interaction is required to counteract and overcome the retrograde transport of the polycystin complex to the ER by masking the ER retention/ retrieval signals or alternatively by allowing the complex to pass quality-control checkpoints in the ER as found for many GPS cleavage of PC1 is also required for the polycystin complex to move to the TGN. Nonetheless, the non-cleavable PC1 V can still move to the Golgi apparatus as evidenced by its   Fig. 3, with PC1 and PC2 bands indicated with colour code. Note that the co-precipitated PC2 from Pkd1 V/V MEFs (lanes 4-6) was entirely EndoH sensitive, lacking EndoH-resistant PC2 130 seen in wild-type MEFs (lanes 1-3). (e) N-glycosylation pattern of polycystin complex in Pkd1 V/V kidney versus wild-type kidney. Polycystin complex was immunoprecipitated and analysed as in (d), expect that an anti-CC antibody was used to detect PC1. Note the presence of cleaved PC1 CTF in wild-type kidney (lanes 1-3) but not in Pkd1 V/V kidneys (lanes 4-6). ability of acquiring EndoH resistance (PC1 V-600 , Fig. 5d,e). One possible explanation is that PC1 V cannot remain associated with PC2 at the Golgi apparatus and consequently traffics to a nonciliary location yet to be determined. Cleavage may promote the trafficking of the polycystin complex to the Golgi apparatus by increasing PC1/PC2 binding affinity, thereby inhibiting PC2 dissociation. Alternatively, cleavage could increase the ER-to-Golgi transition rate of the polycystin complex, thereby outpacing PC2 dissociation. GPS cleavage could be significant for regulating polycystin trafficking in the kidney in a development-specific manner. In early embryonic stages, PC1 exists largely in its uncleaved form, and therefore it might mainly traffic, perhaps apart from PC2, to a non-ciliary location such as at the cell-cell junction where it could regulate convergent extension and elongation of developing renal tubules 17,56 . After birth, however, PC1 is mostly cleaved 15,17,18 and probably traffics to the cilia in the form of the polycystin complex to control proper tubular diameter 56 .
Previous efforts have focused on the identification of the ciliary-targeting motifs and binding proteins that may mediate ciliary trafficking 57 . Both PC1 and PC2 have been reported to contain ciliary-targeting signals. For PC1 the sequence is at its extreme C-terminus 29 , while for PC2 the ciliary targeting sequence is within the first 15 amino acids at the N-terminus 28 .
Our findings indicate that the ciliary-targeting signal in each polycystin is not sufficient for ciliary trafficking in renal epithelial cells. We have identified a novel ciliary sorting module that is composed of Rabep1, GGA1 and Arl3, which is recruited to the polycystin complex by PC1, and is required for the ciliary targeting of the complex. Rabep1-PC1 binding requires the intact coiled-coil domain of PC1, which is outside of the previously reported ciliary targeting sequence 29 . The same study using a small CD16.7-PC1 chimeric construct encoding the last 112 amino acids of the PC1 cytoplasmic C-terminus 29 has also implicated Arf4 in the ciliary trafficking of PC1. However, we were unable to detect Arf4 binding to full-length endogenous PC1 in CD cells, which makes it unlikely that Arf4 plays an important role in targeting of the polycystin complex.
GGA1 is a member of the monomeric GGA family that is known to select proteins at the TGN into clathrin-and GGAcoated carriers, which are then transported to endosomes. GGA1 exerts its function by the different domains that interact with various other proteins 44 . The GAT domain binds activated Arf-family GTP-binding proteins and thereby recruits GGA1 to the TGN. The hinge domain then binds clathrin for vesicle budding, while the GAE domain binds accessory factors that can regulate the function of GGA-coated carriers. Arl3 plays a critical role in ciliary function and is involved in intraflagellar transport in C. elegans 58 . In mice, Arl3 inactivation results in various ciliopathy-related phenotypes including cystic kidney disease and retinal degeneration 51 . In addition to the cilium, Arl3 also has been shown to localize to the Golgi, but the function of this Golgiassociated pool is unknown 50 . We show for the first time that there is a distinct pool of Arl3 that is bound to GGA1. This GGA1/Arl3 module likely binds cargo and other components necessary for clathrin binding, forming vesicle carriers destined for the cilium. As Arl3 is known to have microtubule-binding activity 50 , one possibility is that Arl3 may direct the cargo-bearing vesicles to cytoplasmic microtubules for dynein-driven transport to the cilium. The dynein-dependent system has been reported to translocate rhodopsin-bearing vesicles along microtubules towards the cilium in polarized epithelia 59 .
A central region of Rabep1 has been shown to interact with the GAE domain of GGA1 (ref. 45). The Rabep1-GGA1 interaction is bipartite, as the C-terminal coiled-coil region of Rabep1 also binds the GAT domain of GGA1. This bivalent interaction is thought to mediate fusion of post-Golgi GGA1-coated vesicles to Rabep1-bearing endosomes. PC1 binds the C-terminal coiled-coil region of Rabep1 that usually binds to GAT, thus likely leaving its central region accessible for interacting with the GAE domain of GGA1/Arl3. Our current model is that polycystin complex-bound Rabep1 serves as an accessory protein for GGA1 via its GAE domain, thereby coupling the polycystin complex to the GGA1/ Arl3 module (Fig. 8). It remains to be determined whether this module may be involved in the later stages of ciliary trafficking as recently described for Rabep1 (ref. 60) and Arl3 (refs 61,62). Further studies are also required to investigate how the Rabep1/ GGA1/Arl3 complex is related to the previously identified trafficking complexes including the exocyst 63 and BBSome 64 .
Our model has several important implications. First, this general mechanism could conceivably be used to transport the polycystin complex to other cellular locations. For example, the Rabep1-GGA1-polycystin complex could bind other members of the ARF family. This may result in trafficking to non-ciliary locations such as the plasma membrane where polycystins have also been detected [25][26][27] . Second, as the interactions between the GAE domain for ciliary targeting. In summary, this work provides novel insights into the mechanism of the ciliary trafficking of membrane proteins.

Methods
Cell lines. CD cells were affinity purified from mouse postnatal kidneys by using DBA-conjugated Dynabeads (Invitrogen). The kidneys were minced into small pieces and digested in MEM/F12 containing 0.2% collagenase, 0.2% hyaluronidase and 0.001% DNase I at 37°C for 2 h with gentle agitation. The digested tissue was incubated with Dynabeads at 4°C for 30 min. The cells bound to the beads were suspended in culture medium for propagation 15 . Tetracycline-inducible and stable MDCK and IMCD cells were generated using Flp-In System (Invitrogen) 65,66 by transfecting pcDNA5-based wild-type or mutant Pkd1 expression plasmids according to the manufacturer's protocol. MDCK cell lines were cultured in DMEM growth media with 100 mg ml À 1 hygromycin B and 5 mg ml À 1 blasticidin. Stable IMCD cell lines were cultured in DMEM/F12 growth media with 50 mg ml À 1 hygromycin B.
Cilia preparation. MDCK cells were cultured on Transwell filter for B10 days for cilium growth using DMEM with 10% FBS at the outside membrane and DMEM without FBS at the inside membrane of transwell plate. The cells were washed with PBS and treated with 30 mM ammonium sulfate for 3 h to shed cilia into the medium 36 . Intact shed cilia were collected from the medium by sequential centrifugation (at 2,000 g for 30 min, 10,000 g for 30 min and at 16,000 g for 30 min). After centrifugation, the cilia pellet was resuspended in PBS for immunofluorescence or western blot analysis.
Immunofluorescence. Cultured cells were fixed in 4% formaldehyde for 15 min and then permeabilized by 0.1% Triton X-100 for 5 min, and incubated with primary antibodies in blocking buffer (5% FBS/ 0.05% Triton X-100 in PBS) for overnight. Cells were then incubated with fluorescence-conjugated secondary antibodies (Invitrogen, CA) for 1 h and mounted by VectaShield Mounting Medium (Vector Laboratories, CA). Confocal microscope LSM 510 (Carl Zeiss Microscopy) was used to acquire images. For paraffin-embedded kidney sections, sections were first deparaffinized by xylene for 10 min and rehydrated using a graded series of alcohol solutions, followed by antigen retrieval using Target Retrieval Solution, Citrate Buffer pH 6 (Dako, CA), and then followed the procedure as previously described 15  Immunoprecipitation and immunoblot analysis. Immunoprecipitation (IP) studies of the endogenous and recombinant PC1 were accomplished on mouse tissue samples and cells (MEFs, CD cells, MDCK) homogenized in lysis buffer (20 mM sodium phosphate, pH 7.2/150 mM NaCl/1 mM EDTA/10% (vol/vol) glycerol/1% Triton X-100/protease inhibitors) 15 . In brief, the homogenate was incubated with the primary antibody or antibody-conjugated bead for overnight, followed by incubation with G Sepharose 4 Fast Flow (GE Healthcare) or goat antichicken IgY agarose beads (PrecipHen, Aves) for 3 h. Bead pellets were washed with lysis buffer three times and eluded with SDS sample buffer. The primary antibodies and antibody-conjugated bead used for IP included mouse anti-FLAG M2 affinity gel from Sigma-Aldrich (A2220), rabbit anti-FLAG from Cell Signaling Technology (#2368, 1:50), chicken anti-PC1 (a-CC, 1-2 mg ml À 1 ) 15 , rabbit anti-Myc from Cell Signaling Technology (#2278, 1:200) and rabbit anti-GGA1 from Thermo Scientific (PA5-12130, 1-2 mg ml À 1 ). The protein samples were loaded on 3-8% Tris-Acetate SDS-polyacrylamide precast gels or 4-12% Tris-Glycine SDSpolyacrylamide precast gels (Invitrogen) and transferred to PVDF membrane (Bio-Rad). The membranes were incubated with primary antibodies for overnight and then washed with TBS-Tween 20 buffer. An HRP-conjugated secondary antibody from GE Healthcare (NA934V; 1:10,000) was incubated for 1 h, and then ECL Figure 7 | Polycystin complex containing cleaved PC1 traffics to cilia through a Rabep1/GGA1/Arl3-dependent mechanism. (a) Coimmunoprecipitation of endogenous PC1 with Rabep1, GGA1 and Arl3 from CD cells. CD cells with shRNA-mediated Pkd1 knockdown were used as a negative control. The larger areas of the blots are shown in Supplementary Fig. 3d. (b) Co-immunoprecipitation of C-terminal Myc-tagged PC1 with Rabep1, GGA1 and Arl3 from Pkd1 MYC/MYC embryo lysates (MYC). Pkd1 DCMYC/DCMYC knockout embryos that expresses truncated PC1 lacking the C-terminal 257 amino acids (DCMYC) and untagged wild-type (WT) embryos were used as negative controls. (c) Wild-type (WT) or Pkd2 À / À MEF cell lysates were immunoprecipitated with anti-PC1 (a-CC), and the IP products were analysed with the antibodies as indicated. (d) Knockdown of Rabep1, GGA1 or Arl3 in CD cells abolishes ciliary localization of endogenous PC1 (left panels) and PC2 (right panels). The white line in the XY scan indicates the path of the XZ scan. For knockdown of each target protein, three shRNAs for Rabep1, four shRNAs for GGA1 and two shRNAs for Arl3 were used, and each lentiviral shRNAs similarly inhibit the expression of the target proteins by B90% (Supplementary Fig. 3a-c). Scale bar, 10 mm. (e) Quantification of PC1 and PC2 ciliary localization in CD cells expressing various shRNAs in (d) was performed from four independent experiments and presented as the mean±s.e.m.; ***Po0.001. (f) Endogenous PC1 were immunoprecipitated with anti-PC1 (a-CC) from CD cells expressing the shRNA as indicated and analysed with various antibodies as indicated. Note that acquisitions of EndoH resistance of PC1 and PC2 were not affected by any of the shRNAs. The larger areas of the blots are shown in Supplementary Fig. 3e. (g) Co-immunoprecipitation of native GGA1 and Arl3 in CD cells with Rabep1 knockdown. The specificity of the GGA1-Arl3 interaction was validated using CD cells with GGA1 knockdown (Supplementary Fig. 4d). (h) Co-immunoprecipitation of GGA1 with Rabep1 and Arl3 in both wild-type (WT) and Pkd1 À / À MEFs.

GPS cleavage
Cytoplasmic microtubule Figure 8 | Model for ciliary trafficking of the polycystin complex. PC1 is cleaved at GPS and forms a complex with PC2 in the ER (1). Rabep1 binds PC1 cytoplasmic C-terminal tail at a pre-Golgi compartment (2), and this complex traffics to the trans-Golgi network (TGN). At the TGN, Rabep1 couples the polycystin complex to a GGA1/Arl3 module (3), which is formed by Arl3-GTP binding to GGA1 via the GAT domain. GGA1 then assembles the clathrin coat (4) to form the vesicle carrier (5). The resulting polycystin complex-bearing vesicle traffics along the cytoplasmic microtubules to the base of cilia (6) and enters the cilium by an unknown mechanism.
N-glycosylation analysis. Immunoprecipitation products or lysates were denatured using glycoprotein denature buffer (New England Biolabs) for 1 min at 95°C and then quickly chilled on ice. The denatured glycoprotein was incubated with PNGaseF or EndoH (New England Biolabs) for 1 h at 37°C.
Yeast two-hybrid screening. The human PKD1 cDNA (Accession L33243) fragment corresponding to bp 12,473-13,120 encoding the C-terminal 215 amino acids was isolated by PCR and subcloned into the SalI and NotI sites of the pPC97 vector in-frame with the GAL4 DNA-binding domain 68 . YRG-2 yeast cells (Stratagene) were transformed with bait and subsequently with an adult rat hippocampus random primed cDNA library cloned into pPC86. The protocols used in this yeast two-hybrid system were performed as described previously 22 . In brief, positive yeast clones were selected on triple-deficient plates (Leu-, Trp-, and His-) and confirmed by positive b-gal activity. The pPC86-based prey plasmids were recovered and co-transformed into YRG-2 with either the bait vector or the original pPC97 vector to confirm the interaction.
Statistical analysis. Statistical analysis was performed by ANOVA to calculate statistical significance between more than two experimental groups. Student's t-test was used for the statistical significance between two experimental groups. All data are presented as mean ± s.e.m.