A comprehensive analysis of Rab GTPases reveals a role for Rab34 in serum starvation-induced primary ciliogenesis

Primary cilia are sensors of chemical and mechanical signals in the extracellular environment. The formation of primary cilia ( i.e. ciliogenesis) requires dynamic membrane trafficking events, and several Rab small GTPases, key regulators of membrane trafficking, have recently been reported to participate in ciliogenesis. However, the precise mechanisms of Rab-mediated membrane trafficking during ciliogenesis remain largely unknown. In the present study, we used a collection of siRNAs against 62 human Rabs to perform a comprehensive knockdown screening for Rabs that regulate serum starvation-induced ciliogenesis in human telomerase reverse transcriptase retinal pigment epithelium 1 (hTERT-RPE1) cells and succeeded in identifying Rab34 as an essential Rab. Knockout (KO) of Rab34, but not of Rabs previously reported to regulate ciliogenesis ( e.g. Rab8 and Rab10) in hTERT-RPE1 cells, drastically impaired serum starvation-induced ciliogenesis. Rab34 was also required for serum starvation-induced ciliogenesis in NIH/3T3 cells and MCF10A cells but not for ciliogenesis in Madin-Darby canine kidney (MDCK)-II cysts. We then attempted to identify a specific region(s) of Rab34 that is essential for ciliogenesis by performing deletion and mutation analyses of Rab34. Unexpect-edly, instead of a specific sequence in the switch II region, which is generally important

Primary cilia are membrane projections from the cell surface and are thought to function as sensors of chemical and mechanical signals in the extracellular environment (1). Defects in the formation or function of primary cilia cause various human diseases called ciliopathies (2). The assembly and disassembly of primary cilia are tightly coupled to cell cycle progression, and primary cilia form in cells in the resting stage (3). Primary cilium formation (so-called ciliogenesis) is known to occur in a series of membrane-trafficking steps (4)(5)(6). First, small vesicles, called preciliary vesicles, accumulate on the mother centriole, and then the preciliary vesicles fuse with each other to form a large vesicle called a ciliary vesicle. The resulting ciliary vesicle extends together with the axoneme and finally fuses with the plasma membrane. It is generally thought that lipids and ciliary proteins must be transported from other organelles, such as the Golgi apparatus and recycling endosomes, to the mother centriole via membrane trafficking mechanisms during the ciliary vesicle formation and elongation (7,8).
Rab small GTPases, which belong to the Ras superfamily, are key regulators of membrane trafficking (9)(10)(11)(12). Rabs function as switch proteins that cycle between an active state and an inactive state. In their active state, Rabs localize to specific vesicles or organelles and recruit a specific binding partner (called a Rab effector), via which they regulate a specific membrane trafficking pathway. Recent studies have reported that several Rabs, including Rab8, Rab10, Rab11, Rab23, Rab29, and Rab34, participate in ciliogenesis (8,(13)(14)(15)(16)(17)(18). However, not all of the membrane trafficking mechanisms in ciliogenesis are fully understood, and knockout mice in which each of several cilia-regulating Rabs had been knocked out did not exhibit any ciliopathy phenotypes (15,19). Moreover, no attempts have been made to perform a comprehensive analysis of all mammalian Rabs (Rab1A-43) during ciliogenesis.
In this study, we performed a comprehensive knockdown screening for Rabs that regulate serum starvation-induced ciliogenesis in hTERT-RPE1 (human telomerase reverse transcriptase retinal pigment epithelium 1) cells and succeeded in identifying Rab34 as an essential Rab in serum starvation-induced ciliogenesis. Intriguingly, however, the known cilia-regulating Rabs, including Rab8, Rab10, Rab11B, and Rab12, were found to be dispensable for ciliogenesis, because their knockout (KO) cells formed primary cilia. Moreover, the requirement of Rab34 for serum starvation-induced ciliogenesis was confirmed in NIH/3T3 cells and MCF10A cells. On the other hand, Rab34 was not essential for ciliogenesis in cysts of Madin-Darby canine kidney (MDCK)-II cells. We then performed KO-rescue experiments on several Rab34 mutants, including a switch II swapping mutant (S1A) and an N-terminal deletion mutant (DN), to identify the region of Rab34 that is responsible for ciliogenesis. The results showed that Rab34(S1A) rescued the Rab34-KO phenotype but that Rab34 (DN) did not. These findings indicate that the unique long N-terminal region of Rab34, and not a specific sequence in the switch II region, is necessary for serum starvation-induced ciliogenesis.

Results
Comprehensive screening for Rabs whose knockdown inhibits ciliogenesis in hTERT-RPE1 cells To identify novel Rabs that participate in serum starvationinduced primary ciliogenesis, by using effective and specific This article contains supporting information. * For correspondence: Mitsunori Fukuda, nori@tohoku.ac.jp. siRNAs against 62 human Rabs that we have developed previously (20), we performed a comprehensive knockdown screening in hTERT-RPE1 cells. We used acetylated tubulin as a cilium marker and then counted the number of nonciliated cells. When we knocked down CEP164 protein that is essential for ciliogenesis (21), the number of nonciliated cells increased to more than 80% (Fig. 1A, black bar), whereas control siRNA had almost no effect (;10% of the cells were nonciliated), thereby validating our experimental setup. The results of the screening showed that knockdown of ten Rabs, i.e. Rab6C, Rab9A, Rab10, Rab11B, Rab12, Rab34, Rab40A, Rab40B, Rab42 (43), and Rab43 (41), increased the number of nonciliated cells to more than 50% (Fig. 1A, blue bars, and B), and we considered these Rabs to be primary candidates.
To confirm the knockdown effects on ciliogenesis that had been observed in the first screening and to avoid possible off-target effects by single siRNAs, we proceeded to perform knockdown experiments by using other independent siRNAs against the primary Rab candidates. The results showed that knockdown of only four Rabs, i.e. Rab6C, Rab11B, Rab12, and Rab34, by each of two independent siRNAs increased the number of nonciliated cells to more than 50% ( Fig. 2A). We also checked the protein expression levels of several candidate Rabs by immunoblotting, and the results confirmed that the Rab11B, Rab12, and Rab34 bands almost completely disappeared when the two independent siRNAs were used (Fig. 2B). siRab9A#2 and siRab10#2 seemed to decrease the protein expression level more efficiently than siRab9A#1 and siRab10#1, respectively ( Fig. 2B and data not shown), but they had a lesser effect on ciliogenesis, suggesting that siRab9A#1 and siRab10#1 have certain off-target effects on ciliogenesis. On the other hand, no protein expression of other candidate Rabs, including Rab6C, Figure 1. Screening for Rabs whose knockdown inhibited ciliogenesis in hTERT-RPE1 cells. A, the number of nonciliated hTERT-RPE1 cells (%) that have been transfected with siRNA against each Rab isoform (0.2 nM) was counted after a 24-h serum starvation (n . 50 cells). The broken red line indicates that 50% of the cells have no cilia. The black bar represents Cep164 (positive control), and the blue bars represent candidate Rabs whose knockdown increased the number of nonciliated cells to more than 50% compared with the control siRNA (negative control). Because of the high sequence identity between Rab40A and Rab40AL, Rab40A siRNA was also able to knock down Rab40AL (20). B, typical images of cells transfected with control siRNA (siControl), Cep164 siRNA (siCep164) (positive control), or siRNA against candidate Rabs (siRab). The cells were fixed after a 24-h serum starvation and then stained with anti-acetylated tubulin antibody (green; cilia) and DAPI (blue; nuclei). The broken white lines indicate the boundaries of each cell. *, nonciliated cells. Scale bars, 20 mm. was detected by immunoblotting (data not shown), and we did not pursue Rab6C in the subsequent analysis. Based on the results of the two-step screenings, we selected Rab11B, Rab12, and Rab34 as secondary candidate Rabs.
A possible function of Rab11 in ciliogenesis had already been reported (8), but involvement of Rab12 and Rab34 in ciliogenesis had not been investigated by the time we completed the comprehensive screening. During investigating Rab12 and Rab34 in the subsequent analysis, their possible involvement in ciliogenesis was reported by other groups (17,(22)(23)(24), although the detailed molecular mechanisms remained unknown.

Rab34 is required for ciliogenesis in hTERT-RPE1 cells
To determine whether the secondary candidates identified above, i.e. Rab11B, Rab12, and Rab34, are indeed essential for ciliogenesis in hTERT-RPE1 cells, we generated their respective KO cells by using the CRISPR/Cas9 system. We also established Rab8A/B-double KO (Rab8A/B-KO) cells and Rab8A/B/10-triple KO (Rab8A/B/10-KO) cells as controls, because Rab8 and Rab10 are widely thought to be involved in cilium formation in mammalian cultured cells (15). Loss of the target Rabs in each KO cell was verified by both immunoblotting (Fig. 3B) and sequencing of genomic PCR products (Fig. S1). As shown in To further confirm that the impaired ciliogenesis was directly related to Rab34-KO, we performed rescue experiments by stably expressing EGFP (enhanced GFP)-P2A-Rab34, which contains a P2A self-cleavage site (25) between EGFP and Rab34, in Rab34-KO cells (Fig. S2A). The results showed that expression of EGFP-P2A-Rab34 completely rescued the impaired ciliogenesis phenotype of Rab34-KO cells (Fig. 3, C and D). We used EGFP-P2A-Rab34 instead of EGFP-Rab34 for the KO-rescue experiment, because EGFP-P2A-Rab34 was able to rescue the phenotype of Rab34-knockdown (KD) cells more efficiently than EGFP-Rab34 was (Fig. S2B), suggesting that N-terminal EGFP tagging partly distorts the function of Rab34. A similar observation was previously reported with respect to Rab10: untagged Rab10, but not EGFP-Rab10, was found to promote neurite outgrowth of Rab10-depleted cells (26). These findings suggest that Rab34 is the most crucial Rab for serum starvationinduced ciliogenesis in hTERT-RPE1 cells.

Requirement of Rab34 for ciliogenesis depends on the cell type
To determine whether Rab34 is a general positive regulator of ciliogenesis, we investigated the effect of Rab34-KO or -KD on ciliogenesis in other cell types. As shown in Fig. 4  Error bars indicate the S.E. of data from three independent experiments. The broken red line indicates that 50% of the cells have no cilia. The black bar represents Cep164 (positive control), and the blue bars represent candidate Rabs whose knockdown with two independent siRNAs increased the number of nonciliated cells to more than 50% compared with the control siRNA (negative control). *, p , 0.05; **, p , 0.01; ***, p , 0.001; NS, not significant compared with the control siRNA (Tukey's test). B, the knockdown efficiency of candidate Rabs was evaluated by immunoblotting with the antibodies indicated on the right of each panel. Cells were harvested 48 h after transfection with control siRNA or siRNA against candidate Rabs (0.2 nM). Lane 1 and lanes 2 and 3 show the results for the control siRNA and for two independent siRNAs against each Rab, respectively. The positions of the molecular mass markers (in kDa) are shown on the left. revealed that all of the Rab-KO cysts, including the Rab34-KO cysts, had formed primary cilia in their luminal domain ( Fig. 4C and Fig. S3). Although Rab6 and Rab11 have been shown to be required for normal epithelial morphogenesis (27), their KO cysts still had primary cilia in their luminal domain (Fig. 4D). Thus, Rab34 is unlikely to be a general regulator of ciliogenesis, and it is dispensable for ciliogenesis at least in MDCK-II cysts.

Rab34 is required for early steps in serum starvation-induced ciliogenesis in hTERT-RPE1 cells
To investigate the defective ciliogenesis in the absence of Rab34 in greater detail, we then investigated the subcellular localization of ciliary membrane proteins and proteins that are required for ciliogenesis, i.e. Rabin8, CP110, IFT20, and Arl13B (7,14,28,29), in Rab34-KO cells. As shown in Fig. 5A, Rab34-KO had no effect on the localization of Rabin8 (3 h after serum starvation) or of CP110, IFT20, and Arl13B (24 h after serum starvation) in hTERT-RPE1 cells compared with that of the control parental cells for removal of CP110 from mother centrioles (Fig. 5A, arrows in the second row) (28), and the centriole localization of Rabin8, IFT20, and Arl13B (Fig. 5A, arrowheads in the top, third, and bottom rows, respectively). Essentially the same results were reported for CP110, Rabin8, and IFT20 in Rab34-KO NIH/3T3 cells (23), indicating that Rab34 is not involved in the recruitment of Rabin8 and IFT20 to centrioles or CP110 removal. Because depletion of Rab34 inhibited the axoneme elongation and extension of ciliary vesicles, Rab34 is likely to be required for early steps in serum starvation-induced ciliogenesis in hTERT-RPE1 cells.
Because our anti-Rab34 antibody was unable to detect the subcellular localization of endogenous Rab34 in hTERT-RPE1 cells by immunofluorescence analysis, we tried overexpressing FLAG-tagged Rab34, which can restore the Rab34-KO Rab34 is required for primary ciliogenesis phenotype in Rab34-KO cells (Fig. 6E), and visualized it by using an anti-FLAG tag antibody. The results showed that FLAG-Rab34 was mainly localized at the Golgi apparatus, consistent with the findings in a previous report (30), and that it rarely localized at primary cilia (;20%) ( Fig. 5B). Taken together, these results imply that Rab34, an early-step regulator of ciliogenesis, is released from ciliary membranes before the maturation of primary cilia. The unique long N-terminal region of Rab34 is required for ciliogenesis in hTERT-RPE1 cells Next, we attempted to identify a specific region(s) of Rab34 that is essential for ciliogenesis by performing mutation and deletion analyses of Rab34-KO cells. In general, Rabs are thought to recognize specific effectors through their switch II region, because mutations of specific amino acids in this region of some Rabs have been shown to abrogate their effector-binding ability (31)(32)(33)(34)(35). Sequence comparisons of the switch II region of mammalian Rabs have revealed that Lys-115 and Cys-116 are specific to Rab34 (and its closest paralog, Rab36) (Fig. 6A, asterisks) (34). Actually, we had previously shown that K120A/ C121A mutations of Rab36 impaired RILP (Rab interacting lysosomal protein)-binding activity, and that the Rab36 (K120A/C121A) mutant did not support retrograde melanosome transport in melanocytes (34). Therefore, we generated a Rab34(K115A/C116A) [here referred to as Rab34 (KC/AA)] mutant and evaluated its effect on ciliogenesis in Rab34-KD hTERT-RPE1 cells.
Consistent with the previous finding regarding Rab36 (34), Rab34(KC/AA) failed to interact with RILP in yeast two-hybrid assays (Fig. S4A) and was shown to hardly interact with it by coimmunoprecipitation assays (Fig. S4B). Unexpectedly, however, Rab34(KC/AA) completely rescued the Rab34-KD phenotype (Fig. S4C). Because a weak interaction between Rab34(KC/ AA) and RILP was still observed in the coimmunoprecipitation assays (Fig. S4B), it was possible that such a weak interaction could rescue the KD phenotype. To rule out this possibility, we generated an additional Rab34 mutant that completely lacks RILP-binding ability by swapping the switch II region of Rab34 for that of Rab1A, a Rab isoform distantly related to Rab34 [named Rab34(S1A)] (Fig. 6A), and complete loss of RILP-binding activity by Rab34(S1A) was confirmed by both yeast twohybrid assays and coimmunoprecipitation assays (Fig. 6, B and C). Nevertheless, Rab34(S1A) did rescue the Rab34-KO phenotype, the same as Rab34(WT) did (Fig. 6D). Although no specific sequence in the switch II region of Rab34 was essential for ciliogenesis, GTP-binding activity of Rab34 itself was necessary, because a constitutively active Rab34 mutant, Rab34(Q111L), but not its constitutively inactive mutant, Rab34(T66N), rescued the Rab34-KD effect on ciliogenesis, as did Rab34(WT) (Fig. S5).
To further identify the crucial region of Rab34 for ciliogenesis, we compared the entire sequences of mammalian Rab family members in greater detail and discovered that Rab34 contains a long N-terminal region (gray box in Fig. 6A) that was not found in other Rab isoforms except Rab36. Rab36, the closest paralog of Rab34, also contains a long N-terminal sequence, but the N-terminal sequences are not well conserved between them. Therefore, we produced an N-terminal deletion mutant of Rab34 [named Rab34(DN)], which lacks the 49 N-terminal amino acids of Rab34, and performed a rescue experiment. The results showed that Rab34(DN) did not rescue the Rab34-KO phenotype at all (Fig. 6E). The lack of a rescue effect by Rab34 (DN) is unlikely to be attributable to its protein expression level, because similar amounts of Rab34(DN) and endogenous Rab34 were observed in hTERT-RPE1 cells (Fig. 6F). Moreover, FLAG-tagged Rab34(S1A) and Rab34(DN) mainly localized at the Golgi apparatus, the same as Rab34(WT) did ( Fig. 6G and 5B). It should be noted that Rab34(S1A) localized at primary cilia to an extent similar to that of Rab34(WT) (;20%), whereas we did not observe any ciliary localization of Rab34(DN) under our experimental conditions. Thus, the unique N-terminal region of Rab34 is not involved in its Golgi localization, but it may be required for its ciliary targeting or localization. However, because the protein expression level of Rab34(DN) was often lower than that of Rab34(WT) and Rab34(S1A) (Figs. 3B and 6F), we could not completely rule out the possibility that its low protein expression masked the ciliary localization of Rab34 (DN). Top row, for Rabin8 staining, cells stably expressing EGFP-Rabin8 (green) were fixed after a 1-h serum starvation and then stained with anti-g-tubulin antibody (g-Tub, in magenta; centrioles). For CP110, IFT20, and Arl13B staining, cells were fixed after a 24-h serum starvation and then stained with antibodies against CP110 (green) and g-tubulin (magenta) (second row), IFT20 (green) and g-tubulin (magenta) (third row), or Arl13B (green) and g-tubulin (magenta) (bottom row). The arrows and arrowheads indicate CP110-negative centrioles and centriolelocalized Rabin8 (IFT20 or Arl13B), respectively. Scale bars, 2 mm. B, typical images of FLAG-Rab34 and GALNT2 or Arl13B in hTERT-RPE1 cells expressing FLAG-Rab34(WT). Top, for GALNT2 staining, cells were fixed without serum starvation and stained with antibodies against GALNT2 (magenta) and FLAG (green). Bottom, for Arl13B staining, cells were fixed after a 24-h serum starvation and then stained with antibodies against Arl13B (magenta) and FLAG (green). The arrows indicate FLAG-Rab34-positive primary cilia. Scale bars, 20 mm. Figure 6. The unique long N-terminal region of Rab34, not its specific sequence in the switch II region, is required for ciliogenesis in hTERT-RPE1 cells. A, schematic representation of mouse Rab1A, Rab5A, Rab7, and Rab34 (WT and DN) and sequence alignment of the switch II region (black boxes) of mouse Rab1A, Rab5A, Rab7, and Rab34. Identical and similar residues in the switch II region are shown against a black background and a gray background, respectively. Lys-115 and Cys-116 of Rab34 (asterisks) are not conserved in other Rabs except Rab36 (34) (see also Fig. S4). The unique long N-terminal region of Rab34 is indicated by a gray box. B, interaction between Rab34(WT), Rab34(S1A), or Rab34(DN) and RILP was analyzed by yeast two-hybrid assays. Yeast cells expressing pGBD-C1-Rab34 (WT), Rab34(S1A), or Rab34(DN) and pAct2-RILP were streaked on an SC-LW plate (upper) and on an SC-AHLW plate (lower, selection medium) and incubated at 30°C. C, interaction between Rab34(WT) or Rab34(S1A) and RILP was analyzed by coimmunoprecipitation assays of COS7 cell lysates. COS7 cell lysates expressing FLAG-tagged Rab34(WT) or Rab34(S1A) and T7-tagged EGFP or RILP were incubated with anti-T7-tag antibody-conjugated agarose beads. Proteins bound to the beads (IP) and a 1% volume of total cell lysates (input) were analyzed by immunoblotting with the HRP-conjugated anti-FLAG and anti-T7 tag antibodies indicated on the right of each panel. The known Rab34-interacting proteins are not required for ciliogenesis in hTERT-RPE1 cells In the final set of experiments, we investigated the possible involvement of the known Rab34-interacting proteins, i.e. RILP, RILP-like 1 (RILP-L1), RILP-like 2 (RILP-L2) (30,36,37), Folliculin (FLCN) (38), and Munc13-2 (39), in ciliogenesis in hTERT-RPE1 cells. To narrow down the candidates, we first took advantage of the Rab34(S1A) mutant described above. Because Rab34(S1A) was able to support ciliogenesis (Fig. 6D), a Rab34 effector(s) that functions in ciliogenesis was expected to bind to Rab34(S1A). However, RILP and RILP-L1/L2 failed to interact with Rab34(S1A) (Fig. 6, B and C, and data not shown), indicating that the RILP-binding activity of Rab34 is not essential for ciliogenesis. Actually, knockdown of endogenous RILP in hTERT-RPE1 cells had no or little effect on ciliogenesis (Fig. 7A), and RILP-L1 and RILP-L2 have previously been shown to inhibit rather than promote ciliogenesis in hTERT-RPE1 cells (40). Therefore, we focused on FLCN, which had previously been shown to be associated with ciliopathy (41), and Munc13-2 in our subsequent analyses. An immunoblot analysis using specific antibodies showed that FLCN was endogenously expressed in hTERT-RPE1 cells (Fig. 7B), but no endogenous Munc13-2 expression was detected under our experimental conditions (Fig. S6A). Furthermore, transfection of a specific siRNA for Munc13-2 into hTERT-RPE1 cells had no effect on ciliogenesis (Fig. S6B), indicating that Munc13-2 is not a relevant effector in ciliogenesis. Finally, we knocked down endogenous FLCN in hTERT-RPE1 cells, and the results showed that FLCN-KD had no effect on ciliogenesis, even though the FLCN immunoreactive bands almost completely disappeared after treatment with two independent siRNAs (Fig. 7B). These findings taken together indicate that the known Rab34-interacting proteins, including RILP and FLCN, are not required for Rab34-regulated serum starvation-induced ciliogenesis in hTERT-RPE1 cells.

Discussion
In the present study, we performed a comprehensive knockdown screening for human Rabs that regulate ciliogenesis (Fig.  1), and, based on the results together with the results of KO analyses, we succeeded in identifying Rab34 as an essential Rab in serum starvation-induced ciliogenesis in hTERT-RPE1 cells ( Fig. 1-3). Moreover, Rab34-KO had no effect on recruitment of IFT20, Rabin8, and Arl13B to centrioles or removal of CP110 from the mother centriole, but they inhibited axoneme elongation and extension of ciliary vesicles (Fig. 5A). These results suggest that Rab34 regulates the ciliary vesicle fusion step and/ or the subsequent expansion step of serum starvation-induced ciliogenesis.
Although ;10 Rabs have been proposed to participate in ciliogenesis in previous studies (8,(13)(14)(15)(16)(17)(18)(42)(43)(44), the results of knockdown or knockout of these Rabs in the present study demonstrated that none of the reported Rabs except Rab34 are essential for serum starvation-induced ciliogenesis, at least in hTERT-RPE1 cells under our experimental conditions (Fig. 1-3). Among the Rab-KO cells established in this study, only the Rab34-KO cells showed marked inhibition of ciliogenesis, and although Rab8/10-KO seemed to weakly inhibit ciliogenesis, the reduction was not statistically significant (Fig. 3). Nevertheless, we cannot rule out the possibility that Rabs other than Rab34 participate in ciliogenesis in other cell types or under different conditions (e.g. different upstream signals).
Intriguingly, when we checked the primary cilia in the cysts of recently established Rab-KO MDCK-II cells (27), we found that all of the Rab-KO cysts, including the Rab34-KO cysts, formed primary cilia in their luminal domain (Fig. 4, C and D,  and Fig. S3). Thus, the mechanism of ciliogenesis in the two cell types is likely to be different. Actually, it has been proposed that ciliogenesis in polarized cells and in nonpolarized cells occurs in different pathways (45,46). For example, the cilia in hTERT-RPE1 cells, NIH/3T3 cells, and MCF10A cells are induced by serum starvation, whereas the cilia in MDCK-II cells are induced during cyst growth in the absence of starvation. Thus, the upstream signals of ciliogenesis in hTERT-RPE1 and MDCK-II cells must be different, and Rab34 is specifically required for ciliogenesis in only certain cell types, such as hTERT-RPE1 cells, NIH/3T3 cells, and MCF10A cells. Consistent with this conclusion, Rab34-KO mice exhibited polydactyly, cleft lip, and cleft palate, but they did not have a polycystic kidney disease phenotype (23,47), indicating that the primary cilia in their kidney cells are functionally normal.
Several questions regarding the function of Rab34 remain unanswered in this study. One of the important questions that must be answered in a future study concerns the function of the unique long N-terminal region of Rab34. Because no specific sequence in the switch II region of Rab34 is essential for serum starvation-induced ciliogenesis ( Fig. 6 and Fig. S4), we hypothesized that the long N-terminal region contributes to recognition of an unidentified Rab34 effector that functions in ciliogenesis. Consistent with our hypothesis, N-terminal EGFP tagging of Rab34 significantly reduced its rescue efficiency (Fig.  S2), presumably because fusion of a relatively large molecule, e.g. EGFP, to the N terminus of Rab34 partially impairs its interaction with an unidentified Rab34 effector. We used N-terminally GST (GSH S-transferase)-tagged or GBD (Gal4-binding domain)-tagged Rab34 as bait in our previous comprehensive screening for Rab effectors (34,37,48), but the screening methods we used are presumably inappropriate for identifying a Rab34 effector that functions in ciliogenesis. C-terminally tagged or untagged Rab34(S1A), which lacks binding activity toward RILP family members, would be useful in future screening for a novel Rab34 effector(s) in ciliogenesis. Another unanswered question concerns the cargo(s) Rab34 transports. One possible cargo is a component of SNARE complexes, because recruitment of an EHD1-binding protein, SNAP-29, to preciliary membranes is important for ciliogenesis, and a similar early-step ciliogenesis defect phenotype has been observed in EHD1-depleted cells (49). It is also unknown which membrane trafficking steps(s), e.g. budding, transport, docking, and fusion, Rab34 regulates. It had previously been shown that the main localization of Rab34 is the Golgi apparatus (30), and we confirmed its Golgi localization in the present study (Fig. 5B). The unique N-terminal region of Rab34 is not required for its Golgi localization, but no ciliary localization of Rab34(DN) was observed, at least under our experimental conditions. Therefore, we assume that Rab34 regulates the budding/secretion step from the Golgi, the transport step of preciliary vesicles, and/or the fusion step of preciliary vesicles to form ciliary vesicles. Further extensive research will be necessary to answer all of the questions.
In summary, we found that Rab34 regulates early steps of serum starvation-induced ciliogenesis through its unique, previously uncharacterized N-terminal region and that its KO causes a loss of primary cilia even after serum starvation. In the future, it will be necessary to identify all of the membrane trafficking mechanisms responsible for Rab34-dependent ciliogenesis and determine the functional relationships between Rab34 and the previously reported Rabs, including Rab8 and Rab11, during ciliogenesis.

Cell culture and transfections
hTERT-RPE1 cells were cultured at 37°C under 5% CO 2 in Dulbecco's modified Eagle medium-nutrient mixture F-12 (DMEM/F12) medium (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum, 100 mg/ml streptomycin, and 100 unit/ml penicillin G. One day after plating, siR-NAs and plasmids were transfected into hTERT-RPE1 cells by using Lipofectamine RNAi MAX for the siRNAs (concentrations are indicated in each figure legend) and Lipofectamine 2000 (Thermo Fisher Scientific) or Fugene6 (Promega) for the plasmids, each according to the manufacturer's instructions.
MCF10A cells were cultured at 37°C under 5% CO 2 in DMEM/F12 medium supplemented with 5% horse serum, 50 ng/ml cholera toxin (Fujifilm Wako Pure Chemical, Osaka, Japan), 20 ng/ml epidermal growth factor (Fujifilm Wako Pure Chemical), 10 mg/ml insulin (Fujifilm Wako Pure Chemical), 500 ng/ml hydrocortisone (Fujifilm Wako Pure Chemical), 100 mg/ml streptomycin, and 100 unit/ml penicillin G. One day after plating, siRNAs (0.5 nM) were transfected into MCF10A cells by using Lipofectamine RNAiMAX according to the manufacturer's instructions. The same culture medium without the addition of horse serum and cholera toxin was used for serum starvation.
MDCK-II cells and NIH/3T3 cells were cultured at 37°C under 5% CO 2 in DMEM supplemented with 10% fetal bovine serum, 100 mg/ml streptomycin, and 100 unit/ml penicillin G. Establishment of Rab-KO MDCK-II cells (listed in Fig. S3) and culture of three-dimensional cysts were performed as described previously (27). Rab34-KO NIH/3T3 cells were established in a similar manner (27). Plat-E cells were donated by Dr. Toshio Kitamura (The University of Tokyo, Tokyo, Japan). The plat-E cell culture and retrovirus infection were performed as described previously (57).

Establishment of CRISPR/Cas9 KO cell lines
hTERT-RPE1 cells that had been transfected with pSpCas9bsr-Rab8A/B or -Rab11B were selected by exposure to 15 mg/ ml blasticidin for ;48 h. hTERT-RPE1 cells that had been transfected with pSpCas9-puro-Rab10, -Rab12, or -Rab34 and pDonor-tBFP-NLS-Neo-Rab10, -Rab12, or -Rab34 were selected as described previously (58). Single clones were then selected by the limiting dilution method. KO cells were first checked for loss of target proteins by immunoblotting and then checked for genomic mutations by genomic PCR and sequencing of the PCR products as described previously (27).

Immunofluorescence analysis
hTERT-RPE1 cells, NIH/3T3 cells, and MCF10A cells that had been transfected with siRNAs or in which a Rab had been knocked out were fixed with 4% paraformaldehyde for 15 min at room temperature after a 24-h or 48-h serum starvation. After fixation, the cells were permeabilized with 0.3% Triton X-100 in PBS for 1 min and then stained with specific primary antibodies and appropriate Alexa Fluor 488/555/633-conjugated secondary antibodies. After a 7-day culture in collagen gel MDCK-II, cysts were fixed with 10% TCA for 15 min at room temperature and then stained with specific primary antibodies and appropriate Alexa Fluor 555/633-conjugated secondary antibodies. The immunostained cells were examined with a confocal fluorescence microscope (FV1000, FV1000-D; Olympus, Tokyo, Japan; and Dragonfly spinning disk scanning unit [Dragonfly200], Andor, Belfast, Northern Ireland).

Binding experiments
Coimmunoprecipitation assays of COS-7 cell lysates and yeast two-hybrid assays were performed as described previously (53,59).

Data availability
All presented data are contained within the article.