Rab35 controls cilium length, function and membrane composition

Rab and Arl guanine nucleotide‐binding (G) proteins regulate trafficking pathways essential for the formation, function and composition of primary cilia, which are sensory devices associated with Sonic hedgehog (Shh) signalling and ciliopathies. Here, using mammalian cells and zebrafish, we uncover ciliary functions for Rab35, a multitasking G protein with endocytic recycling, actin remodelling and cytokinesis roles. Rab35 loss via siRNAs, morpholinos or knockout reduces cilium length in mammalian cells and the zebrafish left‐right organiser (Kupffer's vesicle) and causes motile cilia‐associated left‐right asymmetry defects. Consistent with these observations, GFP‐Rab35 localises to cilia, as do GEF (DENND1B) and GAP (TBC1D10A) Rab35 regulators, which also regulate ciliary length and Rab35 ciliary localisation. Mammalian Rab35 also controls the ciliary membrane levels of Shh signalling regulators, promoting ciliary targeting of Smoothened, limiting ciliary accumulation of Arl13b and the inositol polyphosphate 5‐phosphatase (INPP5E). Rab35 additionally regulates ciliary PI(4,5)P2 levels and interacts with Arl13b. Together, our findings demonstrate roles for Rab35 in regulating cilium length, function and membrane composition and implicate Rab35 in pathways controlling the ciliary levels of Shh signal regulators.

molecular composition of the ciliary membrane and cytosol and, therefore, the appropriate sensory and signalling output of the organelle.
With the aim of investigating new Rab proteins that regulate cilium formation and/or function, we focused on Rab35, a plasma membrane and endosomal protein with roles in cargo recycling, cytokinesis, actin cytoskeleton regulation, and autophagy, amongst others [58][59][60][61][62][63][64][65][66][67][68]. Initial clues towards a Rab35 ciliary role include its presence in a photoreceptor outer segment proteome, and highthroughput siRNA or CRISPR screens implicating Rab35 as either a positive or negative regulator of ciliogenesis, or a regulator of Hedgehog signalling in NIH3T3 mouse fibroblasts [33,[69][70][71][72]. Here, using mammalian cell culture and zebrafish models, we show that Rab35 localises to the ciliary membrane and regulates primary and LRO cilium length. We also reveal that Rab35 controls the ciliary levels of Shh signal regulators, Smoothened, Arl13b, and INPP5E, as well as the INPP5E target, PI(4,5)P 2 . Furthermore, we identify the Rab35 GEF and GAP, DENND1B and TBC1D10A, respectively, as the regulators of Rab35 in the ciliary context. Together, our data uncover a novel conserved role for Rab35 in controlling cilium length and the ciliary levels of mammalian Shh signalling regulators.

Rab35 localises to the ciliary membrane
Rab35 has been found in several proteomes of mammalian primary cilia [70,71,73]. To more directly investigate a possible ciliary localisation of Rab35, we employed two well-established mammalian cell models, namely mouse renal epithelial (IMCD3) and human retinal pigment epithelial (hTERT-RPE1) cells, where ciliogenesis can be induced by serum withdrawal [74,75]. First, we analysed mouse IMCD3 and human hTERT-RPE1 cells transiently expressing GFP-tagged Rab35 and found that it localises along the ciliary axoneme in~60% of transfected ciliated cells (Fig 1A and B). We also observed GFP-Rab35 at the plasma membrane and in vesicular structures, consistent with previously described localisations in other cell types and its known endosomal functions [58,[60][61][62][63][64]66]. To further analyse the ciliary association of Rab35, we established a stable hTERT-RPE1 cell line expressing GFP-RAB35 and performed super-resolution microscopy to localise RAB35 in relation to markers of the ciliary membrane (ARL13B) and axoneme (IFT88 and acetylated tubulin; Fig 1C). The GFP-RAB35 signal coincides with that of ARL13B, with the radial extent of both signals being wider than that of the axonemal markers, indicating association of RAB35 with the ciliary membrane. Notably, in the proximal-most part of the cilium, the radial extent of the GFP-RAB35 signal is wider than that of ARL13B (Figs 1C and D, and EV1A and B), and in~20% of cells, GFP-RAB35 is also more concentrated in this region ( Fig EV1C). This proximal staining is reminiscent of the ciliary pocket localisation for the membrane remodelling Eps15 homology domain (EHD) 1 protein [76] and super-resolution imaging of GFP-RAB35 expressing hTERT-RPE1 cells stained for endogenous EHD1 revealed co-localisation of both proteins in the proximal region of the cilium; importantly, these EHD1/GFP-RAB35 co-localising signals appeared broader than the radial extent of the ARL13B ciliary membrane signal, indicative of their association with the ciliary pocket membrane (Fig 1E). Taken together, these data suggest that GFP-RAB35 localises to the ciliary pocket in addition to the ciliary membrane.

Rab35 regulates cilium length
The ciliary localisation of Rab35 suggests that this G protein may regulate cilium formation, structure or function. Indeed, there are recent suggestions of a role for Rab35 in Hedgehog signalling regulation [33,72]; also, Rab35 was identified in an siRNA screen as a positive ciliogenesis modulator [69], although additional highthroughput siRNA screening studies indicate that Rab35 is either a negative regulator of cilium formation [71], or plays no role at all in this process [77]. To shed light on these seemingly contradictory results and test for a potential role of Rab35 in ciliogenesis, we employed siRNA-mediated depletion of RAB35 in hTERT-RPE1 cells and analysed cilia using acetylated tubulin staining. Two independent siRNAs targeting human RAB35 were used, both of which reduce RAB35 expression by > 90% (Fig 2A, and Appendix Fig  S1A). Although RAB35 depletion does not affect the number of ciliated hTERT-RPE1 cells observed after serum withdrawal (Fig 2B), the cilia are significantly shorter, with a median cilia length of 2.1 lm in RAB35-depleted cells, compared to~2.8 lm in nondepleted cells (Fig 2C and D). Expression of an siRNA-resistant GFP-RAB35 construct rescues the cilia length defect, thus ruling out the possibility that this phenotype is due to off-target effects (Fig 2C  and D). Similar results were observed in IMCD3 cells using a pool of siRNAs targeting mouse Rab35, where cilia length is reduced bỹ 20% (Figs 2E-H, and EV2B and C, and Appendix Fig S1B). To further validate these results and to address the possibility that residual Rab35 due to incomplete siRNA-mediated depletion may still function in ciliogenesis, we generated Rab35 knockout (KO) NIH3T3 cell lines using the CRISPR/Cas9 system [78]. We obtained two independent Rab35 KO clones (KO#1 and KO#2) using two different sgRNAs targeting exon 3 and confirmed Rab35 protein loss by immunoblot analysis and genomic locus disruption by sequencing (Fig 2I and Appendix Fig S1C). Analysis of cilia using acetylated tubulin staining revealed no differences in the number of ciliated cells between wild-type (WT) and Rab35 KO NIH3T3 cells; however, like Rab35-depleted cells, the KO cells display a shorter cilium, with a median length of~1.8 lm compared to~2.4 lm in the WT control (Fig 2J-L). Collectively, these findings suggest that Rab35 is not involved in initial cilium formation and demonstrate that Rab35 regulates the length of primary cilia.

Rab35 ciliary localisation and function depend on its nucleotidebound state
Rab GTPases are regulated by their nucleotide-bound state. To assess the requirements of GDP and GTP binding for the ciliary localisation of Rab35, we overexpressed GFP-tagged dominant-negative (GDP-bound) or constitutively active (GTP-bound) mutants in serum-starved hTERT-RPE1 cells [58]. Both WT and GTP-bound RAB35-Q67L localise to cilia in~60% of transfected ciliated cells. In contrast, ciliary signals for GDP-bound RAB35-S22N are much less frequent (~13%; Fig 3A and B). We also assessed whether overexpression of the Rab35 mutants affects hTERT-RPE1 cilium formation and/or structure. We found that RAB35-S22N overexpression exerts a dominant-negative effect on ciliogenesis, resulting in a severe reduction in ciliation (down to 25% of cells; Fig 3C), with those cilia  A Localisation of transiently expressed GFP-Rab35 in IMCD3 cells after 48 h serum starvation and staining for acetylated tubulin (acetyl. tub.) and DNA. Insets show higher magnification images of the cilia region. Scale bars, 10 lm. B Localisation of transiently expressed GFP-RAB35 in hTERT-RPE1 cells after 24 h serum starvation and staining for polyglutamylated tubulin (polyglu. tub.) and DNA.
Insets show higher magnification images of the cilia region. Scale bars, 10 lm. C Super-resolution (FV-OSR) imaging of hTERT-RPE1 stably expressing GFP-RAB35 after 24 h serum starvation and staining for GFP, ARL13B (cilia membrane), IFT88 (axoneme), acetylated tubulin (acetyl. tub.; axoneme), and c-tubulin (c-tub; centrosome). Top panels show representative images, and bottom panel shows line profile plots of fluorescence intensity (arbitrary units; a. u.) in the distal (dotted line) and proximal (dashed line) cilia regions. Data are mean AE SEM (n = 5 cilia), with the solid line in the line profile plots indicating mean and the dotted lines indicating SEM values. Scale bars, 1 lm. D Representative FV-OSR images of 24 h serum-starved hTERT-RPE1 stably expressing GFP-RAB35 and stained for GFP, ARL13B and acetylated tubulin. Right panels show a cilium with proximal enrichment of GFP-RAB35 and left panels a cilium with even GFP-RAB35 localisation along the full length. Scale bars, 1 lm. E Representative FV-OSR images of 24 h serum-starved hTERT-RPE1 stably expressing GFP-RAB35 and stained for GFP, EHD1 and ARL13B. Line profile plots of fluorescence intensity in the proximal cilia region (dashed line) are shown to the right. Data are mean AE SEM (n = 6 cilia), with the solid line in the line profile plots indicating mean and the dotted lines indicating SEM values. Scale bars, 1 lm.
Rab35 GEF DENND1B and Rab35 GAP TBC1D10A regulate cilium length and Rab35 ciliary localisation The activity of Rab GTPases is regulated by specific guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs) that control their nucleotide-bound state. Several such Rab35 regulators have been identified, including GEFs of the DENND1 family and GAPs of the TBC1D10/EPI64 family [59,64,67,79]. To test   Fig S2A). For DENND1C and TBC1D10C, we did not detect expression using qPCR in hTERT-RPE1 cells and, therefore, they were excluded from subsequent experiments. First, we analysed ciliary length and found that DENND1B depletion leads to significantly shorter cilia (~20% reduction), which phenocopies what we observed for RAB35 depletion or overexpression of GDPbound RAB35-S22N (Fig 3E and F). Conversely, TBC1D10A depletion leads to longer cilia, similar to the overexpression of GTPbound RAB35-Q67L (Fig 3E and F). These ciliary length phenotypes were not observed when DENND1A or TBC1D10B was depleted ( Fig 3F and Appendix Fig S2B). Given that Rab35 ciliary localisation is GTP-dependent and that GEFs are known to recruit their cognate Rab GTPases to target membranes [80], we analysed the effect of GEF and GAP depletion on Rab35 localisation. We found that the number of GFP-RAB35-positive cilia is considerably reduced in DENND1B-depleted cells, while TBC1D10A depletion increases the number of GFP-RAB35 positive cilia ( Fig 3G). Finally, we investigated the localisation of the Rab35 GEF and GAP regulators in ciliated hTERT-RPE1 cells using transiently expressed GFPtagged constructs. Consistent with their roles in cilium length control and Rab35 ciliary localisation, pools of both DENND1B and TBC1D10A were found at the cilium and/or ciliary base iñ 45% of transfected cells, whereas no ciliary localisation was observed for DENND1A or TBC1D10B (Fig 3H and I, and Appendix Fig S2C and D). Interestingly, DENND1B localises preferential along the axoneme, while TBC1D10A is found more often at the ciliary base ( Fig 3J). Together, these findings identify ciliary roles and localisations for the DENND1B GEF and TBCD10A GAP regulators, suggesting that they control Rab35 GTP status in cilia and copperfasten our conclusion that Rab35 serves cilia-related functions.

Rab35 ciliary function is conserved in zebrafish
We investigated whether Rab35 has a conserved ciliary localisation and function in an animal model, namely zebrafish. We focused our studies on the left-right organiser (LRO) or Kupffer's vesicle (KV), a fluid-filled vesicular organ transiently present in the early embryonic life of the fish that is essential to establish internal body laterality [6]. First, we injected mRNA encoding mCherry-tagged mRab35 at the one-cell stage and analysed its localisation in KV cells. At 8 somite stage [13 h post-fertilisation (hpf)], we observed a signal, albeit faint, corresponding to mCherry-Rab35 along KV monocilia ( Fig 4A). Thus, Rab35 also localises to the cilium in zebrafish KV cells.
Next, we depleted rab35 using a translation blocker antisense morpholino and compared the phenotypes with those obtained with a mismatch morpholino and non-injected embryos. Zebrafish has one Rab35-encoding gene, Rab35b, that leads to a transcript of 201 amino acids, which is highly homologous to the mouse and human orthologs. Rab35b was successfully silenced by MO injection (Fig 4B and C) with a dose determined by titration to cause less than 30% mortality (Appendix Fig S3A). Importantly, Rab35 morphants display aberrant left-right (LR) patterning of internal organs ( Fig 4D). Normal organ patterning is referred to as situs solitus, while defects in laterality can take many forms [81]. A full reversal in the placement of all organs is termed situs inversus, while heterotaxia describes uncoupled defects in some of the organs. At 30 hpf, we observed that in 34% of Rab35 morphants, the heart is misplaced at the centre or the right side of the body, ◀ Figure 2. Rab35 regulates cilium length.
ª 2019 The Authors EMBO reports 20: e47625 | 2019 compared with 1 and 10% of non-injected and mismatch MO control embryos, respectively. We also scored liver positioning and found that 34% of Rab35 morphants display misplacement of this organ, compared with 1 and 8% of controls (Figs 4D and EV3A and B). Also, morphological analysis at 48 hpf revealed that Rab35 morphants possess pericardial oedema, which is a typical feature of ciliary impairment (Fig EV3C-E) [82][83][84]. To investigate LR phenotypes at an earlier time point, we analysed whether the asymmetric expression of dand5, one of the earliest transcriptional responses to proper KV function, was affected in Rab35 morphants and found that 45% of embryos exhibit bilateral expression ( Fig 4E). Thus, Rab35 function is required for the establishment of LR asymmetry in early zebrafish development.
Since KV cilia produce directional fluid flow responsible for the establishment of LR asymmetry of dand5 [7], we analysed the length and number of KV cilia in Rab35 morphants. We found that cilia are abnormally short when Rab35 is depleted, compared with mismatch MO-injected or non-injected embryo controls (Fig 4F and  G). Importantly, this phenotype is rescued by co-injection of Rab35 mRNA (Fig 4F and G). Consistent with the results in mammalian  cell lines, we did not observe a significant change in the number of cilia per KV in Rab35 morphants (Appendix Fig S3E).
To assess whether the regulatory role of Rab35 in KV cilia length depends on its GTPase activity, we overexpressed mCherry-tagged dominant-negative (GDP-bound Rab35-S22N) or constitutively active (GTP-bound Rab35-Q67L) Rab35 by injecting one-cell stage embryos with the corresponding mRNA. We observed that KV cilia length is significantly reduced in embryos overexpressing Rab35-S22N, when compared with the overexpression of Rab35-WT or mCherry alone (Fig 4H and I). Together, these findings show that Rab35 regulates cilia length in zebrafish and is required for the asymmetric expression of dand5 and subsequent left-right organ patterning during embryo development.

Rab35 regulates Arl13b levels at the ciliary membrane
Having established a conserved function for Rab35 in mammalian and zebrafish cilium length control, we next wondered if Rab35 regulates the localisation of specific proteins at the ciliary membrane. First, we analysed the localisation of endogenous ARL13B in RAB35-depleted hTERT-RPE1 cells and found that the ciliary levels of ARL13B are elevated 1.5-to twofold upon RAB35 depletion (Fig 5A and B). The increase in ARL13B ciliary levels was observed with two different ARL13B antibodies (Figs 5A and B, and EV4A and B) and was rescued in the stable hTERT-RPE1 cell line expressing an siRNA-resistant form of RAB35 (Fig 5A and B). Given the effect of Rab35 depletion on both ciliary length and ARL13B, we wondered whether these two phenotypes are correlated; however, closer analysis at a single-cell level revealed no correlation between ciliary ARL13B intensity and ciliary length (Fig EV5A), suggesting that Rab35 may function in more than one cilia-related pathway. Interestingly, immunoblot analysis of whole-cell lysates showed increased total ARL13B protein levels in RAB35-depleted cells (Fig 5C), suggesting that RAB35 may regulate ARL13B protein stability and/or degradation. Similar results were also observed with the Rab35 KO NIH3T3 cell lines, where ciliary Arl13b levels are elevated 1.5-fold compared to wild-type NIH3T3 cells (Fig 5D-F).
Next, using siRNAs targeting the cilia-related RAB35 GEFs and GAPs discussed above, we assessed the requirement of GTP binding for RAB35-mediated regulation of ARL13B localisation. We found that depletion of the GEF DENND1B leads to an increase in ciliary ARL13B levels, which phenocopies what we observed for RAB35 depletion (Figs 5G and EV4C). Conversely, depletion of the GAP TBC1D10A has the opposite effect, resulting in a significant reduction of ARL13B in cilia (Figs 5H and EV4D). We also found that ARL13B localisation is unaffected in cells depleted for either DENND1A or TBC1D10B. Thus, RAB35 control of ARL13B levels at the ciliary membrane depends on GTP binding and is regulated by the same GAP and GEF regulators that control RAB35 ciliary association (see Fig 3 above). We also tested whether the RAB35mediated effect on ciliary ARL13B levels requires the ciliary localisation of RAB35. By analysing hTERT-RPE1 cells overexpressing GFP-RAB35 and comparing ARL13B ciliary levels in GFP-positive and negative cilia, we observed a significant decrease in ciliary ARL13B levels in cells with GFP-RAB35 ciliary localisation (Fig 5I and J). Collectively, these findings suggest that Rab35 negatively regulates Arl13b levels at the ciliary membrane.

Rab35 interacts with Arl13b
Given that Rab35 and Arl13b co-localise at the ciliary membrane ( Fig 1C) and that cilia-localised Rab35 regulates Arl13b ciliary levels ( Fig 5I and J), we assessed their potential interaction via co-immunoprecipitation (co-IP) assays in HEK293T cells. We found that ARL13B-FLAG specifically co-precipitates with GFP-RAB35 but not GFP alone, and that this interaction is greatly enhanced in cells where ciliogenesis is induced by serum withdrawal (Fig 6A). Reciprocal co-IP of mCherry-RAB35 with ARL13B-GFP confirmed the interaction ( Fig 6C). Next, we investigated the domains or motifs of Arl13b required for Rab35 interaction. Arl13b consists of an N-terminal guanine nucleotide-binding (G) domain and an unusually long C-terminal tail with coiled coil (CC) and proline-rich (PR) domains ( Fig 6B) [85]. All these domains were previously shown to be important for protein-protein interactions; for example, IFT-B  proteins interact with Arl13b via its PR domain, while INPP5E binds to the G-domain [86]. Arl13b also contains a C-terminal RVEP motif for ciliary targeting and N-terminal palmitoylation for ciliary membrane anchoring [45,[86][87][88]. By co-expressing various GFPtagged ARL13B mutants and truncations with mCherry-RAB35 in HEK293T cells, we found that ARL13B membrane association is crucial for RAB35 interaction because ARL13B lacking the palmitoylation motif (ARL13B-C8S, C9S) fails to co-IP with RAB35, as do ARL13B truncations lacking the N terminus ( Fig 6C). On the other hand, RAB35 interaction with ARL13B does not require the C-terminal PR domain or the RVEP motif, whereas ARL13B truncations containing only the G-and/or CC-domain show reduced levels of RAB35 co-IP. Together, these findings show that several ARL13B sequence motifs are essential for the interaction with RAB35, namely the N-terminal palmitoylation motif, as well as a binding region in the C-terminal tail mapped to amino acids 245-356. Finally, we investigated whether the nucleotide-bound state of RAB35 affects the interaction with ARL13B. For this, we immunoprecipitated GFP-RAB35 with anti-GFP beads and then exchanged this GTPase with either GTPcS, a non-hydrolysable GTP analogue, or GDP. After incubation with lysate from HEK293T cells expressing ARL13B-FLAG, we found that GFP-RAB35 loaded with GTPcS pulled down significantly more ARL13B-FLAG than GDP-loaded GFP-RAB35 ( Fig 6D). Since ARL13B is also a GTPase and its binding to   GTP could also influence the interaction, we performed a similar pull-down experiment and found that the interaction of GTPcSloaded ARL13B-GFP with mCherry-RAB35 is greatly increased compared to GDP-loaded ARL13B-GFP ( Fig 6E). Thus, a stronger RAB35-ARL13B interaction is observed when either protein is GTPbound. Taken together, these findings demonstrate an interaction between Rab35 and Arl13b, which further supports a role for Rab35 in regulating Arl13b at the ciliary membrane.
Rab35 regulates INPP5E ciliary localisation and PI(4,5)P 2 ciliary levels in hTERT-RPE1 cells Given Rab35 function in regulating Arl13b ciliary levels, we asked whether Rab35 also affects the localisation of other ciliary proteins. Arl13b was previously shown to mediate the ciliary targeting and distribution of various ciliary proteins [42,47,89] such as farnesylated INPP5E, whose ciliary targeting involves the GDP dissociation inhibitor (GDI)-like solubilising factor PDE6D and GTP-bound Arl3 that acts as a cargo-release factor [89,90]. Arl13b acts as a GEF for Arl3, activating Arl3 specifically in the cilium due to its own restricted ciliary localisation, and ensures the specific release of INPP5E from PDE6D in the ciliary compartment [51]. We analysed the endogenous localisation of INPP5E in RAB35-depleted hTERT-RPE1 cells and found that INPP5E is significantly increased at the cilium, resembling the phenotype observed for ARL13B localisation in these cells (Fig 7A and B). Interestingly, closer analysis at a single-cell level revealed a clear correlation between ciliary INPP5E and ARL13B levels in RAB35-depleted cells, although no correlation was found for ciliary INPP5E intensity and cilium length (Fig EV5B  and C). To determine whether the increased INPP5E ciliary levels in RAB35-depleted cells depend on ARL13B, we examined INPP5E localisation in cells depleted of RAB35 and ARL13B. We found that ARL13B loss reverses the RAB35 depletion phenotype, resulting in decreased INPP5E ciliary levels, similar to that of cells depleted for ARL13B alone (Fig 7C) [89]. Similar results were also observed with the Rab35 KO NIH3T3 cell lines, where ciliary Inpp5e levels are elevated 1.5-fold compared to wild-type NIH3T3 cells (Fig 7D and  E). These data show that Rab35 regulates INPP5E ciliary localisation in an Arl13b-dependent manner and indicate that the increased ciliary Arl13b levels observed in Rab35-depleted cells result in increased Arl13b activity towards its ciliary targets. The phosphoinositide 5-phosphatase INPP5E catalyses the conversion of phosphatidylinositol (4,5)-bisphosphate (PI(4,5)P 2 ) to phosphatidylinositol 4-phosphate (PI4P). Due to its ciliary localisation, INPP5E establishes a distinct localisation pattern for phosphoinositides along the ciliary axoneme with PI(4,5)P 2 being more enriched in the proximal region of the cilium [91,92]. Given the effect of Rab35 loss on ciliary INPP5E levels, we examined the PI (4,5)P 2 localisation in the Rab35 KO NIH3T3 cells using the PI(4,5) P 2 -specific probe PH-PLCd1-GFP [93]. In wild-type NIH3T3 cells, the majority of cilia are positive for PI(4,5)P 2 with~28% of cilia displaying signal only at the most-proximal region and~52% of cilia with staining extending beyond the first third of the axoneme. Importantly, in the Rab35 KO cell line the number of PI(4,5)P 2 -positive cilia is significantly reduced with 45% of cilia displaying no ciliary PI(4,5)P 2 localisation (Fig 7F and G). These results are consistent with increased INPP5E in cilia observed after Rab35 depletion and loss.

Rab35 depletion impairs SAG-induced ciliary translocation of Smoothened
Having established Arl13b as a target of Rab35, we wondered whether Rab35 serves Arl13b-related roles in ciliary signalling pathways. Specifically, we focussed on the Shh pathway because of the conserved role of Arl13b in regulating this pathway [41,42,94]. Shh A HEK293T cells were transiently co-transfected with ARL13B-FLAG and GFP or GFP-RAB35. Cells were cultured either in full growth medium (+serum) or serum-starved for the final 8 h before harvesting. Immunoprecipitations (IP) were performed using anti-GFP beads. Cell lysates (2.5% of input) and immunoprecipitates were loaded on the same gel, subjected to immunoblot (IB) analysis with indicated antibodies and detected with the same exposure. B Schematic representation of the human ARL13B structure depicting the guanine nucleotide-binding (G), coiled coil (CC) and proline-rich (PR) domains as well as the point mutations affecting N-terminal palmitoylation or C-terminal ciliary-targeting RVEP motif. C HEK293T cells were transiently co-transfected with mCherry-RAB35 together with GFP or indicated ARL13B-GFP wild-type, mutant or truncation constructs. Cells were serum-starved for the final 8 h before harvesting. IPs were performed using anti-GFP beads and interacting proteins detected by immunoblotting (IB). D HEK293T cells were transiently transfected with ARL13B-FLAG and serum-starved for the final 8 h before harvesting. HEK293T cell lysates expressing ARL13B-FLAG were subjected to pull-down with GFP-RAB35 bound to anti-GFP beads and preloaded with either no nucleotide, GTPcS or GDP. Bound proteins and cell lysates (1% of input) were analysed by immunoblotting. The graph shows the ratio of precipitated ARL13B and RAB35, normalised to the no nucleotide control. Data are mean AE SEM of three independent experiments. Statistical significance according ANOVA followed by Tukey post hoc test (*P = 0.0333). E HEK293T cells were transiently transfected with mCherry-RAB35 and serum-starved for the final 8 h before harvesting. HEK293T cell lysates expressing ARL13B-FLAG were subjected to pull-down with ARL13B-GFP bound to anti-GFP beads and preloaded with either no nucleotide, GTPcS or GDP. Bound proteins and cell lysates (1% of input) were analysed by immunoblotting. The graph shows the ratio of precipitated ARL13B and RAB35, normalised to the no nucleotide control. Data are mean AE SEM of three independent experiments. Statistical significance according ANOVA followed by Tukey post hoc test (*P = 0.0466).
Source data are available online for this figure.   requirement of RAB35 for SMO ciliary translocation, we treated RAB35-depleted hTERT-RPE1 cells with the SMO agonist SAG, which binds to SMO and activates the pathway independently of PTCH [96]. In both control and RAB35-depleted cells, SMO localisation to cilia is rarely observed under basal conditions (DMSO vehicle control). After SAG treatment, whilst SMO is present in~75% of cilia in control-depleted cells, only 40% of cilia are SMO-positive in RAB35-depleted cells (Fig 8A and B) and those depleted cilia that retain SMO localisation display significantly decreased SMO ciliary levels (see data on Smo + cilia in Fig 8C). Thus, Rab35 loss interferes with SAG-induced ciliary accumulation of SMO.
Previous studies have shown that Arl13b loss leads to aberrant SMO accumulation in cilia, suggesting that Arl13b limits the ciliary accumulation of SMO, possibly by promoting its ciliary export [42,86]. We wondered, therefore, whether elevated ciliary ARL13B levels in RAB35-depleted cells are required for the SMO localisation defect in these cells. To answer this question, we analysed SAGtreated hTERT-RPE1 cells co-depleted of RAB35 and ARL13B and observed near wild-type levels of ciliary SMO (Fig 8D-F). To test whether the ARL13B-mediated effect on SMO requires its ciliary localisation, we employed a GFP-tagged ciliary-targeting defective mutant of ARL13B (V358A) [87] and found that SAG-induced SMO ciliary translocation is greatly reduced in hTERT-RPE1 cells expressing ARL13B-WT but not ARL13B-V358A or GFP (Fig 8G and H). Thus, the reduced levels of ciliary SMO observed in SAG-induced RAB35-depleted cells depend on elevated levels of ARL13B in the cilium.

Discussion
In this study, we identify Rab35 as a new regulator of cilium structure and function. We show evolutionarily conserved localisation of Rab35 at the ciliary membrane and demonstrate a role for Rab35 in restricting the ciliary accumulation of Arl13b and INPP5E and promoting the ciliary accumulation of SMO as well as PI(4,5)P 2 . In addition, we demonstrate that loss of Rab35 causes left-right asymmetric patterning defects in zebrafish. Thus, a ciliary function is now added to the multiple cellular roles of Rab35 in diverse processes such as endocytic recycling, actin remodelling, autophagy and cytokinesis.
A number of our observations support a role for Rab35 as a positive regulator of cilium length. First, cilia are abnormally short in Rab35-depleted, knockout and dominant-negative Rab35-S22Nexpressing mammalian cells, in Rab35-depleted zebrafish KV cells, as well as in Rab35-S22N-expressing KV embryos. Second, cilia are abnormally long in mammalian cells expressing constitutively active Rab35-Q67L. Third, depletion of a Rab35 GAP (TBC10D1A) elongates cilia, whereas depletion of a Rab35 GEF (DENND1B) truncates cilia. Thus, Rab35 joins a number of other Rabs, namely Rab8, Rab11 and Rab23 (reviewed in [32]) required for normal cilium structure. In the context of cilium length regulation, it is interesting that the truncated RAB35-deficient cilia exhibit increased ARL13B levels, given reports of abnormally long cilia in cells overexpressing Arl13b [44,97]. However, it should be recognised that the level of overexpression in those studies likely reflects a much higher ciliary concentration of Arl13b compared to the 1.5-to twofold increase that was observed for endogenous Arl13b in Rab35-disrupted cells. It has also been shown that cilia lacking Arl13b can be abnormally long [98], thereby highlighting the lack of a correlation between cilium length and Arl13b levels. Additionally, Rab35 disruption may impact more than one ciliarelated pathway, and thus, the cilium length defect in cells lacking Rab35 may not arise directly from the elevated ciliary Arl13b levels. Indeed, this notion is supported by our observation of no correlation between the ciliary length and ARL13B level phenotypes in Rab35-disrupted cells (Fig EV5A). It is also noteworthy that Rab35 knockout and/or depleted mammalian or zebrafish KV cells display normal cilia number. Our finding that Rab35 positively regulates cilium length but is not required for ciliogenesis agrees with data from a genome-wide siRNA screen in mouse IMCD3 cells [77], although two additional high-throughput depletion studies in RPE1 cells report, conflictingly, that Rab35 is a positive or negative regulator of ciliary incidence [69,71].
Consistent with a function for Rab35 at the cilium, we identify ciliary roles for specific Rab35 GEF and GAP proteins, DENND1B and TBC1D10A, respectively. We found that DENND1B-and TBC1D10A-depleted cells display opposing cilium length defects (truncated/elongated), thereby phenocopying what is observed in WT cells overexpressing GDP or GTP-bound Rab35. Also, loss of DENND1B or TBC1D10A results in opposing effects on ARL13B ciliary levels (reduced/increased), consistent with the ARL13B phenotype in RAB35-depleted cells (discussed further below). Notably, ciliary defects were not found in DENND1A-or TBC1D10Bdepleted cells, and DENND1C and TBC1D10C are not expressed in our ciliated cell model. Thus, only a subset of DENND1 and TBC1D10 subfamily members of Rab35 regulators have ciliary roles, at least in those cells we investigated. Multiple DENN domaincontaining (DENND1A/B/C, Folliculin) GEFs and TBC domaincontaining (TBC1D10A/10B/10C/13/24) GAPs control Rab35 activation, most likely because of its multitasking roles and requirement for regulation at distinct intracellular sites [59,60]. Of the Rab35 GEFs and GAPs, we identify in the ciliary context, DENND1B regulates a fast endocytic recycling pathway and endocytic uptake of cell surface T-cell receptors, and TBC1D10A impairs exosome secretion [99][100][101][102]. It is important to note that some Rab35 regulators such as TBC1D10A target other Rabs (e.g. Rab27a, Rab8a) [103,104] and, therefore, it remains possible that part of the ciliary phenotype in Rab35 GAP/GEF-depleted cells could result from dysregulation of Rabs other than Rab35. Indeed, overexpression of the dominantnegative mutant of Rab35 (Rab35 S22N) leads not only to a reduction in cilia length but also in the percentage of ciliated cells. This could be caused by the sequestration of GEFs and/or effectors that are common to other Rabs.
By analysing embryonic KV cell cilia, we found that the role of Rab35 in controlling cilium length is conserved in zebrafish. We found that Rab35 faintly localises to KV cell cilia, supporting the ciliary localisations we find in mammalian cells. The low levels of Rab35 in KV cell cilia suggest that Rab35 may transiently localise to cilia, a notion supported by our results in mammalian cells where Rab35 ciliary localisation is observed in only~60% of Rab35-expressing cells. Importantly, the reduction in cilia length upon Rab35 depletion correlates with abnormal patterning of the internal organs. These results suggest that KV fluid flow is likely perturbed as a consequence of shortened cilia in Rab35-depleted embryos, as evidenced by abnormal bilateral expression of dand5. Both theoretical and experimental studies have demonstrated that short KV cilia cause KV flow defects leading to the disruption of the whole left-right cascade of gene expression, namely by changing dand5 and nodal asymmetric expression patterns [7,[105][106][107].
We also demonstrate roles for Rab35 in regulating the ciliary membrane levels of Sonic hedgehog signal regulators, Arl13b, SMO and INPP5E. Specifically, RAB35 prevents overaccumulation of ARL13B and INPP5E in cilia, whilst promoting the signalinduced accumulation of SMO to cilia (summarised in Fig 9B). Our observations are consistent with the role of Arl13b in promoting the ciliary release of lipidated INPP5E from its transport carrier [89] and provide an explanation for why more ciliary ARL13B in RAB35-depleted cells correlates with abnormally high levels of ciliary INPP5E. In addition, reduced ciliary SMO levels in RAB35depleted cells, where ciliary ARL13B levels are elevated, correlates with the opposite phenotype (i.e. increased SMO accumulation) in cells with a loss-of-function mutation in Arl13b [42]. However, in cancer cells, Arl13b promotes the ciliary localisation of SMO, although this difference may be explained by the transformed state of these cells [50,108,109]. One scenario to explain our observations is that Rab35 principally targets Arl13b, possibly as a direct effector or a more distant upstream regulator. Thus, the SMO and INPP5E localisation defects in RAB35-depleted cells would derive from misregulation of the proposed roles of Arl13b in INPP5E ciliary targeting, as well as retrograde IFT and the ciliary removal of Shh signalling molecules [86,89]. A role for Rab35 in Arl13b regulation also agrees with their interaction in co-immunoprecipitation assays. Furthermore, Rab35 regulation of mammalian Arl13b ciliary levels is consistent with the left-right defects observed in zebrafish Rab35 (this study) and Arl13b morphants, as well as in zebrafish overexpressing Arl13b [44,97,110].
As to how Rab35 might limit the ciliary membrane levels of Arl13b, one possibility is that Rab35, operating in the cilium, interacts with ciliary transport pathways that establish and maintain the organelle's composition (Fig 9A). For example, Rab35 may facilitate an IFT function that either promotes the ciliary removal, or limits the ciliary entry, of Arl13b. In support of this notion, Arl13b undergoes IFT, interacts with IFT-B, and requires IFT for its ciliary targeting in Caenorhabditis elegans [86,88,111]. Alternatively, Rab35 could regulate membrane diffusion barriers at the transition zone that keep Arl13b in the cilium [88,112,113]. Another possibility is that ciliary Rab35 controls Arl13b localisation and cilium structure by regulating ectosomes that bud from ciliary membranes to control their composition and length [20][21][22][23][24]114]; indeed, some ciliary ectosomes are Arl13b-positive [108]. However, this model is mitigated A Model of how activated (GTP-bound) Rab35 at the ciliary and/or periciliary (pocket) membrane limits Arl13b ciliary levels. In one scenario, Rab35 inhibits Arl13b entry into cilia, possibly through an IFT pathway. In a second scenario, Rab35 promotes Arl13b exit/retrieval from ciliary (and periciliary) membranes via: (i) interaction with retrograde IFT, (ii) modulation of the transition zone membrane diffusion barriers that prevent Arl13b from exiting the cilum, and/or (iii) endocytic processes at or near the ciliary pocket that target Arl13b for degradation or recycling to non-ciliary plasma membrane destinations. B Ciliary membrane and ciliary length phenotypes following Rab35 disruption. In wild-type cells, Arl13b, INPP5E and Smoothened (SMO; under conditions of Shh signal activation) all localise to the ciliary membrane. Arl13b promotes the ciliary localisation of INPP5E [89], whilst limiting (or altering) the ciliary distribution of SMO [42]. PI(4,5)P 2 (blue line) is enriched within the proximal portion of the ciliary membrane. In Rab35-depleted cells, ciliary Arl13b levels are elevated, causing a concomitant increase and decrease in the ciliary levels of INPP5E and SMO, respectively. Rab35 disruption also leads to a reduction in the length of the proximal ciliary PI(4,5)P 2 signal and a reduction in the frequency of PI(4,5)P 2 -positive cilia (thin blue line). Misregulation of ciliary membrane protein composition in Rab35-disrupted cells leads to a decrease in cilium length.
ª 2019 The Authors EMBO reports 20: e47625 | 2019 by our finding that Rab35-disrupted cells possess reduced ciliary levels of PI(4,5)P 2 , which is a positive regulator of ciliary ectosome formation [20]. A second non-mutually exclusive possibility is that Rab35 functions at the ciliary pocket and/or nearby endosomal compartments to regulate Arl13b uptake from the periciliary membrane and subsequent sorting into recycling or degradation pathways (Fig 9A). Indeed, the ciliary pocket is a site of endocytosis, and a role for Rab35 in promoting Arl13b uptake and subsequent degradation is consistent with our observation of increased total Arl13b levels in Rab35-depleted cells, as well as known roles for Rab35 in endocytic events such as endosomal sorting of proteins for recycling or degradation [17,[115][116][117][118][119]. An uptake function for Rab35 could act on Arl13b molecules that are either en route to the cilium or exiting from the cilium as part of a natural cycling process of Arl13b transport between ciliary and non-ciliary compartments. Consistent with its established roles at the endosome [58,60,100,120], Rab35 could also limit ciliary Arl13b levels by facilitating its recycling to nonciliary plasma membrane destinations. Notably, Arl13b is known to co-localise with recycling endosomal markers and regulate endocytic recycling traffic [121].
In summary, we have uncovered a new and conserved role for Rab35 at the cilium of mammalian and zebrafish cells. We show that Rab35 is required for normal cilium length and the ciliary membrane levels of vertebrate Sonic hedgehog signalling regulators SMO, Arl13b and INPP5E, as well as organ patterning. Given the strong association of Sonic hedgehog signalling defects in ciliary disease, our findings also suggest possible roles for Rab35 in ciliopathy mechanisms.

Cell culture and transfection
All cell lines were grown at 37°C under 5% CO 2 . Human hTERTimmortalised retinal pigmented epithelial cells, hTERT-RPE1, were cultured in DMEM/F12 medium supplemented with 10% foetal bovine serum (FBS), 2 mM L-glutamine and 0.348% sodium bicarbonate. Human embryonic kidney cells, HEK293T, were cultured in DMEM with 10% FBS. Murine inner medullary collecting duct cells, IMCD3, were cultured in DMEM/F12 medium supplemented with 10% FBS (Sigma), 100 U/ml penicillin G and 100 lg/ml streptomycin (Gibco). NIH3T3 cells were grown in DMEM supplemented with 10% calf serum, and ciliogenesis in these cells was induced by serum starvation with medium containing 1% calf serum for 24 h. Cilia formation in hTERT-RPE1, IMCD3 or HEK293T cells was induced by serum withdrawal for 24-48 h, 48 h or 8 h, respectively. Smoothened agonist SAG (Santa Cruz) was added at 200 nM for 24 h in starvation media to induce Sonic hedgehog pathway activation.
HEK293T cells were transiently transfected with plasmid DNA using the calcium phosphate precipitation method: Briefly, 1.5 × 10 6 cells were plated in a 10-cm dish 16-20 h before transfection. For each dish, 1-3 lg plasmid DNA was diluted in 500 ll 0.25 M CaCl 2 , mixed with 500 ll 2× HEBS (280 mM NaCl, 1.5 mM Na 2 HPO 4 , 50 mM HEPES, 10 mM KCl, 12 mM dextrose, pH 7.10) and added to the cells. After 8 h, media was exchanged and cells were harvested 2 days after transfection for cell lysis and co-immunoprecipitation assays. For transfection of hTERT-RPE1 cells, 150,000 cells/well were seeded in 12-well plates. After 16 h, cells were transiently transfected with 0.5 lg plasmid DNA using 1.5 ll TransIT-LT1 (Mirus Bio), serum-starved 6 h after transfection and fixed after 24-48 h for immunofluorescence. Stable hTERT-RPE1 cells were generated by transfecting 1 lg plasmid using 2 ll Lipofectamine 3000 and 2 ll P3000 reagent (Invitrogen) and subsequent selection with 500 lg/ml G418 (Invitrogen). For transfection of IMCD3 cells, 300,000 cells/well were seeded in 24-well plates. After 16 h, cells were transiently transfected with 1 lg plasmid DNA using 1 ll of Lipofectamine 2000 (Invitrogen), serum-starved 12 h after transfection and fixed after 48 h for immunofluorescence. For live imaging, 800,000 NIH3T3 cells were seeded in a 35-mm Ibidi glass plate. They were transiently transfected 16 h after with 2 lg of DNA mix containing equal parts of HTR6-RFP and PI(4,5)P 2 sensor PH-PLCd1-GFP with 4 ll of TurboFect (Thermo Fisher) in starvation media. After 24 h, cells were imaged.

RNA interference and qPCR
For depletion experiments, hTERT-RPE1 cells were seeded in 12well plates (150,000 cells/well) and immediately transfected with Silencer Select siRNAs at a final concentration of 20 nM (Ambion) using 1.5 ll Lipofectamine 2000 (Invitrogen). For RAB35/ARL13B co-depletion experiments, hTERT-RPE1 cells were first transfected with either Neg or RAB35 siRNA (2 wells for each siRNA). After 24 h, cells were serum-starved and transfected with either Neg or ARL13B siRNA. Regarding IMCD3 depletion experiments, cells were seeded in 24-well plates (300,000 cells/well) and immediately transfected with 1.25 ll of Dharmafect 1 (Dharmacon) with 80 nM of siGENOME siRNAs pools (Dharmacon). Efficiencies of mRNA depletion were measured by quantitative real-time PCR (qPCR). For this, total RNA was purified with a NucleoSpin RNA Minikit (Macherey-Nagel) or RNeasy kit (Qiagen) following the manufacturer's instructions. cDNA was synthesised with High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) or SuperScript II (Life Technologies) using 200-500 ng of RNA per reaction. qPCR analysis was performed using SYBR green detection in an Applied Biosystems 7500 system or a LightCycler 96 system (Roche Life Sciences). GAPDH mRNA levels were used for normalisation and results were obtained using the DDC t method [122]. Primers and siRNAs used in this study are listed in Appendix Tables S3 and S4, respectively.
Zebrafish culture and injections of morpholino oligonucleotides and/or mRNA WT zebrafish line [AB background and Tg(b-actin2:loxP-DsRed-loxP-GFP; s928Tg)] were maintained and staged as described in Tavares et al [4]. Embryos were kept at 32°C in the dark and in E3 embryo medium. Morpholinos blocking translation of Rab35 or Mismatch (Appendix Table S5) were used as previously described [4] at 140-280 lM, by injecting one-cell stage embryos. These were left to develop at 32°C or 25°C until the desired stage. While preparing this manuscript, a paralogue for Rab35 was identified, Rab35a or Rab35-like, which encodes a shorter protein (195 amino acid) with a divergent C-terminal region (not found in mouse, human or Drosophila) and lacking key basic residues characteristic of Rab35. We ensured that the sequence of the used morpholino does not match any sequence in the Rab35-like gene. Mouse Rab35 coding sequence was cloned into a pCS2 (+) vector using the primers indicated in Appendix Table S3 and a mCherry tag was added at the N terminus of Rab35. Linearised plasmid DNA was used to produce in vitro mRNA with the mMESSAGE mMACHINE T7 transcription kit (Invitrogen), which was then purified with the RNA Clean and Concentrator-5 kit (Zymo). At one-cell stage, embryos were injected with 200 pg of mRNA and left to develop at 32°C or 25°C until the desired stage. To analyse KV cilia, embryos were fixed at 8 somite stage [13 h post-fertilisation (hpf)] with 4% PFA at 4°C overnight and processed for immunofluorescence. To evaluate the left-right patterning of internal organs, embryos were observed from their ventral side at 30 hpf and their heart positioning was evaluated using a stereoscopic zoom microscope (SMZ745, Nikon Corporation, Japan). At 53 hpf, whole-mount in situ hybridisation was performed as described previously [105] for foxa3 probe and embryos were scored for liver patterning. To evaluate KV function, whole-mount in situ hybridisation was performed with the dand5 probe at 8 somite stage embryos using the same protocol, and embryos were scored for dand5 expression laterality using a stereoscopic zoom microscope (SMZ745, Nikon Corporation, Japan).

Immunofluorescence staining
Mammalian cell lines were grown on glass coverslips, fixed with 3% PFA in PBS for 15 min and permeabilised for 5 min in either 0.2% Triton X-100/PBS or ice-cold methanol. Blocking was performed for 30 min with 3% BSA (hTERT-RPE1, NIH3T3) or 1% BSA (IMCD3) in 0.1% Triton X-100/PBS, or with 3% BSA in PBS (for staining with rat anti-Arl13b antibody). Antibodies (Appendix Table S2) were diluted in blocking solution. Cells were incubated with primary antibodies at room temperature for 2 h. Samples were washed three times with PBS and incubated with appropriate secondary antibodies conjugated to fluorescent dyes for 1 h at room temperature. DAPI (4 0 ,6-diamidino-2-phenylindole, Sigma-Aldrich) was included with secondary antibodies for DNA staining. Samples were washed three times with PBS, and coverslips were mounted on glass slides in Mowiol (Sigma-Aldrich).
Zebrafish embryos were fixed with 4% PFA at 8 somite stage (13 hpf) and stored overnight at 4°C. They were then washed 5 times for 5 min with PBS with 1% Triton X-100 and dechorionated using sharp forceps. Permeabilisation was performed using Proteinase K (10 mg/ml) for 1 min, re-fixed for 20 min with 4% PFA and washed in cold acetone for 7 min at À20°C. Blocking was performed for 2 h with 1% BSA/1% DMSO/1% Triton X-100 in PBS supplemented with 1.5% FBS. Antibodies against acetylated a-tubulin (mouse) and mCherry (rabbit) were diluted at 1:300 and incubated overnight at 4°C with rotation. On the second day, embryos were washed in blocking solution for 4 h at room temperature with rotation. Then, embryos were incubated overnight with the appropriate secondary antibodies conjugated to Alexa Fluor dyes (1:500) in equal parts of blocking and DAPI solution at 4°C with rotation. On the third day, embryos were washed in blocking solution 2 times for 10 min, washed 30 min in PBS with 1% Triton X-100, fixed for 5 min in 4% PFA, re-washed in PBS with 1% Triton X-100 and mounted between a coverslip and glass slide in PBS.

Microscopy
Widefield images were acquired as z-stacks at 0.4 lm intervals using a Zeiss AxioImager M1 microscope equipped with a Zeiss Plan Neofluar 40×/NA1.3 oil-immersion objective and a QImaging Retiga R6 CCD camera or Zeiss CellObserver equipped with an Apochromat 63×/NA1.4 oil-immersion objective and a CoolSNAP HQ2 camera. Confocal images of hTERT-RPE1 and NIH3T3 cells were acquired on an Olympus Fluoview FV1000 microscope with a 60×/1.35 NA UPlanSApo oil-immersion objective. For IMCD3 cells, confocal Zseries stacks were acquired on a Leica SP5 confocal, mounted on a DM6000 inverted microscope, using a 63× 1.3NA oil-immersion objective and a Zeiss LSM 710 confocal microscope equipped with a Plan-Apochromat 63/1.40 Oil Ph3 lens. Live imaging of NIH3T3 cells to detect Pi(4,5)P 2 was performed using a spinning disc confocal microscope (Nikon TI) with full temperature atmosphere control (37°C + Humidity + CO 2 ) equipped with a Photometrics 512 EMCCD camera and a Nikon 100× Oil Apo TIRF NA1.4. For whole-mount embryos, confocal Z-series stacks were made on the same microscopes with a 40× 0.85NA objective. Super-resolution imaging was performed using an Olympus FV3000RS Advanced Laser Scanning microscope equipped with a PLAPON60XOSC2 NA 1.40 oil-immersion objective and a FluoView Olympus Super Resolution (FV-OSR) software module. FV-OSR images were captured as z-stacks at 220 nm intervals and with a pixel size of 43 nm. A noise reduction filter was applied to the images in the FV3000 acquisition software, and images were deconvolved using Olympus cellSens software.

Image analysis
Fluorescence intensity profile plots were created using the Twin Slicer tool of the Huygens Professional software (Scientific Volume Imaging B.V) by drawing lines across cilia in the proximal or distal region and exporting the values into Excel and GraphPad Prism. Cilia length measurements in hTERT-RPE1 cells were performed in Fiji/ImageJ (NIH) [125] using acetylated alpha tubulin as a cilia marker, and ≥ 50 cilia were analysed for each experimental condition. Cilia length measurements in IMCD3 and Kupffer's vesicle (KV) cells were also performed in Fiji/ImageJ (NIH) [125] in 3D, using the Segmentation plugin and the Simple Neurite Tracer tool [126], and acetylated alpha tubulin as a cilia marker. Representative maximum projections of confocal images of KV and IMCD3 cells stained with anti-acetylated alpha tubulin are shown. For IMCD3 cells, > 50 cilia were analysed for each experimental condition, whereas for Zebrafish embryos, 8-12 embryos were analysed per condition and per experiment, which corresponds to > 250 cilia per condition and experiment. Staining intensity of ciliary membrane proteins was quantified with Cellprofiler 2.1.1 software using an automated image analysis pipeline for the identification of the ciliary area in each cell [127]. Single-plane confocal images of cells stained for an axonemal marker (acetylated or polyglutamylated tubulin) and the cilia membrane protein of interest (ARL13B, INPP5E or SMO) were used. First, background subtraction was performed using a rolling ball algorithm in the RunImageJ module. To identify cilia, the area stained by the axonemal marker was segmented using the IdentifyPrimaryObjects module with the RobustBackground thresholding method. Average pixel intensity within these regions was measured for the protein of interest using the MeasureObjectIntensity module, and ≥ 25 cilia were analysed for each experimental condition. Intensity values were normalised to the mean of the control sample, which was either cells treated with non-targeting siRNA for depletion experiments or cells expressing GFP alone for overexpression experiments. For quantification of SMO ciliary intensity levels and SMO + cilia, cells were stained for acetylated tubulin and SMO. Acetylated tubulin staining was used to identify the ciliary area (acetyl. tub + cilia) of all cilia as described above and the average pixel intensity of SMO staining was measured within these regions. To identify cilia that displayed a discernible enrichment of SMO (SMO + cilia), images stained for SMO were thresholded using the IdentifyPrimaryObjects module with the RobustBackground thresholding method. This resulted in the segmentation of ciliary area stained by SMO above the threshold, which were then correlated with the acetyl. tub-positive cilia.
Cell lysis, immunoprecipitation and immunoblotting IMCD3 cells were lysed in cold lysis buffer [50 mM Tris-HCl pH 7.4, 1% IGEPAL, 150 mM NaCl, 1 mM EDTA, 1 mM ethylene glycol tetraacetic acid (EGTA), 2 mM MgCl 2 , 1 mM dithiothreitol (DTT)] in the presence of EDTA-free protease and phosphatase inhibitors (Roche) for 20 min on ice, followed by centrifugation at 12,000 × g for 20 min at 4°C. Protein concentration in total cell lysates was determined using the DC protein assay kit (Bio-Rad). Zebrafish embryos at 8 somite stage were dechorionated using sharp forceps, then disrupted in media with 0.1% EDTA to remove yolk and dissociate cells. Next, they were lysed in cold extraction buffer (0.5 M Tris-HCl pH 6.8, 3% SDS, 1% glycerol, 1 mM DTT) supplemented with complete EDTA-free protease inhibitor cocktail (Roche) with the help of a sonicator. NIH3T3 and hTERT-RPE1 cells were lysed on ice for 20 min in lysis buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 1% NP-40, 5 mM MgCl 2 , 1 mM DTT, 1 mM PMSF, 2 mM Na 3 VO 4 ) supplemented with complete EDTA-free protease inhibitor cocktail (Sigma-Aldrich) and lysates were cleared by centrifugation. For immunoprecipitation, one 10-cm dish of transfected HEK293T cells per immunoprecipitation reaction were lysed at 4°C for 20 min in co-IP buffer (50 mM Tris-HCl pH 7.5, 100 mM NaCl, 1% NP-40, 5 mM MgCl 2 , 1 mM DTT, 1 mM PMSF, 2 mM Na 3 VO 4 ) supplemented with complete EDTA-free protease inhibitor cocktail (Sigma-Aldrich) and lysates were cleared by centrifugation. Protein concentrations were determined with the Bradford assay (Sigma-Aldrich). Equal protein amounts (~0.5-1 lg in 800 ll) were subjected to immunoprecipitation with 5 ll GFP-Trap beads (Chromotek) at 4°C for 4 h. Beads were washed three times with extraction buffer, and the bound proteins were eluted in Laemmli sample buffer. For pulldown experiments, HEK293T cells expressing either GFP-RAB35 or ARL13B-GFP were lysed at 4°C for 20 min in 1.6 ml buffer (50 mM HEPES pH 7.2, 150 mM NaCl, 1 mM MgCl 2 , 1% Triton X-100) supplemented with complete EDTA-free protease inhibitor cocktail (Sigma-Aldrich). Lysates were cleared by centrifugation, divided equally into three tubes (500 ll each), and subjected to immunoprecipitation with 5 ll GFP-Trap beads (Chromotek) at 4°C for 2 h. Beads were washed once with nucleotide exchange buffer (50 mM HEPES ph 7.2, 150 mM NaCl, 1 mM MgCl 2 , 1% Triton X-100, 2.5 mM EDTA), resuspended in 50 ll nucleotide exchange buffer and incubated for 10 min on ice. A sample without nucleotide addition was used as negative control, and to other samples, either GTPcS or GDP (Sigma-Aldrich) was added at a final concentration of 1 mM and incubated for 20 min on ice. To lock the nucleotidebound state, MgCl 2 was added to a final concentration of 20 mM and incubate for 10 min on ice. Beads were washed once in IP buffer (50 mM HEPES pH 7.2, 150 mM NaCl, 3 mM MgCl 2 , 1% Triton X-100), and 500 ll of cell lysates, which was prepared from HEK293T cells expressing either ARL13B-FLAG or mCherry-RAB35 in IP buffer, was added and incubated at 4°C for 2 h. Beads were washed three times with IP buffer, and bound proteins were eluted in Laemmli sample buffer. Protein samples were separated by SDS-PAGE and transferred to nitrocellulose membranes. Membranes were blocked in 5% milk/0.1% Tween-20/PBS and probed with primary antibodies and HRP-conjugated secondary antibodies (Appendix Table S2), followed by detection with ECL (Thermo Fisher). Band intensities on immunoblots were quantified using Fiji/ ImageJ (NIH) [125]. For the analysis of ARL13B total protein levels, the ratios of ARL13B and b-tubulin (loading control) levels were calculated and normalised to the control treated with non-targeting siRNA to allow the comparison of independent experiments.

Statistical data
Graphs were drawn, and statistical analysis was performed using Prism (GraphPad). Data are presented as mean AE standard error of the mean (SEM) or box-and-whisker plots with horizontal lines showing 25, 50 and 75 th percentiles and whiskers extending to minimum and maximum values. For data in which more than two datasets were compared, ANOVA followed by Bonferroni post hoc test or Kruskal-Wallis test followed by Dunn's post hoc test were used. For data comparing two datasets, unpaired t-test with Welch's correction was used. For zebrafish data, Fisher exact test was performed. Differences were considered significant if P < 0.05 (*P < 0.05, **P < 0.01, ***P < 0.001).

Animal welfare statement
The procedures performed in zebrafish were approved by the Portuguese Veterinary General Administration (DGAV-Direção Geral de Alimentação e Veterinária) and guaranteed the welfare of the animals.