A role for Rab23 in the trafficking of Kif17 to the primary cilium

Rab23 belongs to the Rab subfamily, the largest branch of the Ras superfamily of small GTPases. To date, more than 60 Rabs and Rab-like proteins have been identified (Elias et al., 2012; Grosshans et al., 2006; Pfeffer, 2005) but the functions of only about 36 Rabs are known (Schwartz et al., 2007). Rab GTPases are known to localize to distinct membrane-bound compartments, conferring vesicle and organelle membrane identities (Sonnichsen et al., 2000). More importantly, they orchestrate multiple steps of vesicular transport between different organelles of the endocytic and secretory pathways in a spatial and temporal manner (Schwartz et al., 2007; Zerial and McBride, 2001).

The first insight into the function of Rab23 came from cloning of the mouse open brain (opb) allele (Eggenschwiler and Anderson, 2000; Gunther et al., 1994; Sporle et al., 1996) of Rab23 (Eggenschwiler et al., 2001). In vertebrate embryonic patterning, Rab23 is required for dorsalization and acts antagonistically to Shh, which is essential for ventral cell type specification (Chiang et al., 1996; Sporle et al., 1996). Successful sequential activation of both genes is crucial in Shh-dependent patterning of the neural tube, and normal development of the brain and spinal cord. Rab23 mutant mouse embryos exhibit exencephalic malformation of the forebrain, midbrain and hindbrain regions, and do not survive beyond the second half of gestation (Eggenschwiler et al., 2001; Gunther et al., 1994). How exactly Rab23 silences the Shh pathway in dorsal neural cells and how Rab23 controls Shh target genes, is unknown. In light of the general function of Rabs, Rab23 could potentially regulate trafficking of Shh signaling components (Lim et al., 2011; Wang et al., 2006), but evidence for this is still lacking. Rab23 has been shown to work distally and downstream of both patched and smoothened, but upstream of the Gli transcription factors (Eggenschwiler et al., 2006; Evans et al., 2003).

The widespread distribution of Rab23 across metazoans suggests that there is a fundamental physiological role of Rab23 beyond Shh signaling (Guo et al., 2006; Lumb and Field, 2011). In addition, high Rab23 expression levels in adult mouse brains hinted at a postnatal function beyond embryogenesis. Elevated Rab23 has been found in hepatocellular carcinomas (HCC) (Liu et al., 2007; Sun et al., 2012), lung cancer tissues (Huang et al., 2011) and atrophic gastritis with intestinal metaplasia (Kim et al., 2007), and it is also associated with diffuse-type gastric cancer (dGC) (Hou et al., 2008) and the follicular adenoma type of thyroid tumors (Denning et al., 2007). Besides tumorigenesis, Rab23 is also upregulated in kidney mesangial cells of focal segmental glomerulosclerosis (FSGS) mice (Huang et al., 2009; Shui et al., 2008). In humans, homozygous nonsense mutations in Rab23 result in a congenital pleiotrophic disorder – Carpenter's syndrome – where patients suffer from anatomical and physiological deformities (Alessandri et al., 2010; Jenkins et al., 2007). Although these Rab23-associated diseases could be linked to dysregulated Shh signaling in postnatal life (Ruiz i Altaba et al., 2002), Rab23 could also have physiological functions beyond embryogenesis and Shh signaling (Chia and Tang, 2009).

Rab23 is a candidate ciliary protein because it has been identified in the ciliary proteome of isolated mouse kidney primary cilia (Ishikawa et al., 2012) and the Ciliary Proteome Database version 3 (Gherman et al., 2006). It is also present at the flagellum of Trypanosoma brucei throughout its life cycle (Lumb and Field, 2011). In fact some features of the human Carpenter's syndrome, for instance cranial malformations, obesity and polysyndactyly, do resemble those frequently observed in ciliopathies (Mykytyn et al., 2002). Intriguingly, studies have shown that overexpression of Evi5L, a Rab23 cognate GTPase-activating protein (GAP), leads to reductions in ciliation (Yoshimura et al., 2007), and that Rab23 modulates recycling of the Shh signaling activator smoothened at the primary cilium (Boehlke et al., 2010). Although this implies that Rab23 is associated with the primary cilium, the exact function of Rab23 and its exact mode of action remain obscure.

We report here a previously unknown cilia-associated function of Rab23 and that Rab23 could be regulating the ciliary transport of Kif17. Our results suggest that one role of Rab23 could be to facilitate binding of Kif17 with its cognate ciliary import carrier importin β2 [also known as transportin 1 (TNPO1) or karyopherin β2]. Formation of such a tripartite complex is likely to be essential for targeting Kif17 to the primary cilium.

Despite several investigations into functional associations between Rab23 and the primary cilium, it remains uncertain whether Rab23 is indeed present at the primary cilium. Immunostaining with anti-Rab23 antibodies (Guo et al., 2006) has shown that endogenous Rab23 is generally cytosolic and associated with subcellular membranous structures, but that its localization at the primary cilia cannot be easily discerned. To test whether overexpression of Rab23 could demonstrate clearer cilia localization, cells expressing wild-type Rab23, GTPase-activity-deficient Rab23 Q68L (constitutively active), GTP-binding-defective Rab23 S23N (dominant negative) and the non-phosphorylatable Rab23 T150A mutants (hereafter denoted Rab23WT, Rab23QL, Rab23SN and Rab23TA, respectively) were compared. Overexpressed mCherry-tagged Rab23WT and Rab23QL displayed distinct colocalization with the primary cilia marker acetylated α-tubulin in NIH3T3 (Fig. 1A), human (h)TERT-immortalized retinal pigment epithelial (hTERT-RPE1) (supplementary material Fig. S2A) and CRL1927 mouse mesangial cells (data not shown). Although a majority of mCherry–Rab23WT and –Rab23QL cells localized to the primary cilia of NIH3T3 cells, only ∼18% of Rab23TA cells exhibited Rab23 ciliary localization, whereas overexpressed Rab23SN could not be detected at the cilia (Fig. 1B). This differential enrichment in the cilia was observed despite comparable expression levels of Rab23 constructs after transient transfection in NIH3T3 cells (Fig. 1C). Similarly, overexpressed untagged Rab23WT and Rab23QL mutant, when immunostained with anti-Rab23 antibodies, were also enriched at the primary cilia (Fig. 1D). These results suggest that the degree of Rab23 ciliary localization is dependent on the GTP-binding status of Rab23. In addition, disruption of phosphorylation site at T150 of Rab23 also appeared to affect the targeting of Rab23 to the primary cilium.

Before cell fixation, a brief incubation with 0.1% Triton X-100 can be used to extract ciliary membrane proteins and release other membrane and cilioplasmic proteins, but leave the ciliary axoneme intact. Incubation with Triton X-100 resulted in a loss of mCherry–Rab23QL staining along the primary cilium, whereas acetylated α-tubulin staining of the axoneme remained (Fig. 1E, top two panels), suggesting that Rab23 is not tightly associated with the ciliary axoneme but is more likely to be mostly bound to the ciliary membrane or existing in a soluble form within the cilioplasm. This loss was also observed with a known ciliary membrane protein, ADP-ribosylation factor-like 13b (Arl13b) (Fig. 1E, lower two panels) (Larkins et al., 2011).

The ciliary enrichment of GTP-Rab23 hinted at a possible role for Rab23 in the morphological development of primary cilia. We therefore checked the effect of Rab23 manipulation on ciliogenesis in NIH3T3 fibroblasts. Compared to mCherry-tagged Rab23WT, Rab23QL, Rab23TA and mock-transfected (‘MT’) control NIH3T3 cells, only Rab23SN-expressing cells exhibited a significant decrease in ciliation and shorter cilia lengths. By contrast, mCherry–Rab23WT and –Rab23QL cells grew moderately longer cilia than vector control cells (Fig. 1F). Although the results imply that higher levels of ciliary GTP–Rab23 present can be correlated with increased ciliary growth, Rab23 silencing did not seem to have an overall adverse effect on cilia morphology in NIH3T3 fibroblasts (Fig. 1G,H). Expression of the dominant-negative Rab23SN mutant thus possibly had indirect or non-Rab23-specific consequences, which led to the markedly more severe ciliogenesis phenotype. Based on these results, however, loss of Rab23 does not appear to have a crucial effect on ciliogenesis in NIH3T3 fibroblasts.

We next asked whether Rab23 could be involved in the trafficking of known ciliary proteins, such as the Shh signaling transducer smoothened (Wang et al., 2009), Arl13b, a ciliary membrane-associated GTPase mutated in Joubert syndrome (Cantagrel et al., 2008), and Kif17, a soluble kinesin-2 family motor protein responsible for transport processes along axonemal microtubules (Hirokawa et al., 2009; Verhey et al., 2011). Although there were no significant differences observed in the ciliary localization of Arl13b after Rab23 silencing in NIH3T3 fibroblasts (supplementary material Fig. S1D–F), Rab23-silenced cells exhibited a moderate increase in the intensity of ciliary SmoothenedA1 (SmoothenedA1 is a constitutively active W549L mutant) (supplementary material Fig. S1A,B). This corroborated the previously defined role of Rab23 as a negative regulator of the Shh pathway (Eggenschwiler et al., 2006) and the finding that Rab23 regulates the recycling of ciliary smoothened (Boehlke et al., 2010), explaining how loss of Rab23 could lead to an accumulation of SmoothenedA1 at the primary cilium. However, in our hands, immunoprecipitation of Rab23 with anti-Rab23 antibodies did not show an interaction with SmoothenedA1 (data not shown). At this point, it is unclear how exactly Rab23 could be involved in the ciliary trafficking of smoothened or in the modulation of Shh signaling.

In investigating Kif17 as a potential ciliary cargo of Rab23, however, we found that the ciliary tip localization of Kif17 was significantly enhanced after exogenous expression of mCherry–Rab23WT and –Rab23QL. This was in contrast to a drastic reduction in Kif17 signals in Rab23SN-expressing cells and a more moderate reduction in the phosphorylation mutant Rab23TA-expressing cells (Fig. 2A,B). For the populations expressing Rab23SN and Rab23TA mutant proteins, there were also significantly fewer Kif17-positive cilia (Fig. 2C). To overcome any possible effects of Rab23 manipulation on ciliogenesis that might indirectly impinge on ciliary transport, cells were serum-starved to induce cilium formation before carrying out the Rab23-knockdown treatments. Similar to what was observed after mCherry–Rab23SN overexpression, Rab23 silencing resulted in significant decreases in the ciliary intensities of Kif17 and the Kif17-positive cilia number in NIH3T3 fibroblasts (Fig. 2D–F). As overexpression of dominant-negative Rab23SN in NIH3T3 cells led to stunted cilial growth that could adversely affect ciliary transport, the phenotype elicited by Rab23SN overexpression should be interpreted with caution. The Kif17 phenotype elicited by the Rab23 dominant-negative mutant is nonetheless similar to the Rab23-silencing phenotype. To further validate that perturbation of the localization of Kif17 was more likely due to defective ciliary entry of Kif17 rather than an inhibition of Kif17 transcription, we showed that Kif17–mCitrine expression levels were relatively unaffected after Rab23 silencing (Fig. 2G).

When the Rab23 manipulations were repeated in hTERT-RPE1 epithelial cells, the ciliary intensities of Kif17 were also enhanced in cells overexpressing Rab23WT and Rab23QL, and significantly reduced in cells overexpressing Rab23SN compared to vector control (supplementary material Fig. S2A,B). This was observed despite no significant differences in Kif17-positive cilia numbers among the differently transfected cells (supplementary material Fig. S2C). By contrast, the ciliary localization of Kif17 was significantly disrupted in serum-starved Rab23-silenced cells (supplementary material Fig. S2D–F), and this was again observed despite comparable expression levels of Kif17–mCitrine between scrambled control and Rab23-knockdown cells (supplementary material Fig. S2G).

To show that ciliary mislocalization of Kif17 was due to a specific effect of Rab23 depletion, the presence of Kif17 at the primary cilia was assessed after expression of a non-degradable form of Rab23WT in Rab23 short hairpin RNA (shRNA)-silenced cells. Transfection of mCherry–Rab23WT in Rab23-shRNA-treated cells led to increases in the ciliary localization of Kif17 compared to Rab23-depleted and even scrambled control cells, with significantly higher Kif17 ciliary intensities and numbers of Kif17-positive cilia (Fig. 3).

Interestingly and importantly, we found that the ciliary localization of Kif17 was rather specifically dependent on Rab23, but not another known ciliary Rab GTPase, Rab8a, which has been previously established as being essential for cilium formation and ciliary cargo transport (Yoshimura et al., 2007; Nachury et al., 2007; Follit et al., 2010). Unlike in Rab23SN-expressing cells, dominant-negative Rab8aTN did not diminish the localization of Kif17 at the ciliary tip (supplementary material Fig. S3A). Likewise, depletion of endogenous Rab8a, unlike after Rab23 knockdown (Fig. 2D–G), did not result in a significant disruption of the ciliary localization of Kif17 (supplementary material Fig. S3B–E).

Rab8a has been previously shown to have a role in the ciliary transport of a C-terminal fragment of fibrocystin (harboring the ciliary-targeting sequence, CTS) in mouse kidney inner medullary collecting duct (IMCD3) cells (Follit et al., 2010). In agreement with these results, Rab8a knockdown resulted in a significant loss of ciliary fibrocystin-CTS (supplementary material Fig. S4A–D) in NIH3T3 cells. In contrast, Rab23 was not important in modulating ciliary transport of fibrocystin-CTS, as Rab23 silencing did not affect the ciliary localization of fibrocystin-CTS (supplementary material Fig. S4F–I).

To better examine how Rab23 could be influencing the ciliary transport kinetics of ciliary Kif17, we used fluorescence recovery after photobleaching (FRAP). After laser-pulse-induced photobleaching of the Kif17 signal at the ciliary tip, we compared the recovery of the normalized, mean Kif17 fluorescence intensity at the bleached region between scrambled and Rab23-specific shRNA-treated cells. Whereas scrambled cells could recover up to ∼30% of initial ciliary Kif17 intensity, Rab23-silenced cells exhibited little or no recovery of the ciliary Kif17 signal over time (Fig. 4). Collectively, the results so far strongly indicate that Rab23 is important for the entry of Kif17 into primary cilia.

Kif17 has been shown to depend on nuclear transport adaptor importin β2 for its import into the primary cilium (Dishinger et al., 2010). To test how Rab23 could be involved with its potential ciliary cargo Kif17 and its putative transport adaptor importin β2, co-immunoprecipitation analysis was performed to check for possible interaction between the three proteins. Using anti-Rab23 antibodies for immunoprecipitation experiments in lysate from Kif17–mCitrine- and Rab23WT-transfected cells, Rab23 was found to co-precipitate with endogenous importin β2 and Kif17–mCitrine. Both interactions were enhanced upon addition of the non-hydrolyzable GTP analog GTPγs, with respect to untreated cell lysate, as well as the negative controls (cell lysate treated with GDP and cell lysate incubated with rabbit control IgG) (Fig. 5A). In the reciprocal co-immunoprecipitation using anti-importin β2 antibodies, Rab23 binding to Kif17–mCitrine and importin β2 also increased in a GTP-dependent manner (Fig. 5B). This interaction between Rab23, Kif17–mCitrine and importin β2 could also be observed in hTERT-RPE1 cells (supplementary material Fig. S2H). Notably, Rab8a, whose manipulation did not perturb Kif17 ciliary localization, did not interact with either Kif17 or importin β2 (supplementary material Fig. S3F). By contrast, whereas Rab8a interacted with fibrocystin-CTS (supplementary material Fig. S4E), there was no detectable interaction between Rab23 and fibrocystin-CTS (supplementary material Fig. S4J).

One plausible reason for Rab23 existing in a complex with importin β2 and Kif17 would be that it facilitates the binding of Kif17 to importin β2. The affinity of importin β2 for Kif17 was hence assessed after Rab23 knockdown. Whereas Kif17–mCitrine and importin β2 binding increased in the presence of GTPγs in scrambled cell lysate, there was no detectable interaction between importin β2 and Kif17 in Rab23 knockdown cell lysate, even in the presence of GTPγs (Fig. 5C). Sufficient levels of Rab23 could therefore be required for binding of Kif17 to importin β2, and this requirement could underlie the Rab23-mediated regulation of Kif17 ciliary transport.

To investigate whether Rab23 interacts directly with Kif17 in vitro, we produced recombinant glutathione-S-transferase tagged wild-type Rab23 (GST–Rab23) and hexahistidine-tagged Kif17 (His₆–Kif17) in bacteria. GST–Rab23-bound glutathione beads were used in an affinity binding (pulldown) analysis. However, GST–Rab23 could not pull down purified His₆–Kif17 (data not shown). Although in vitro binding studies appear to indicate a lack of direct interaction between both proteins, this was not unexpected as Rab23 possibly requires activation by a yet unknown guanine nucleotide exchange factor (GEF) not present in bacteria. Indeed, GST–Rab23 was able to pull down both endogenous Kif17 and importin β2 in cell lysate in a GTPγS-enhanced manner (Fig. 5D). We attempted to investigate whether importin β2 is important for the affinity of GST–Rab23 binding of Kif17 by immuno-depleting importin β2 from the cell lysate. Interestingly, but not unexpectedly, a fraction of Kif17 was co-depleted with importin β2. GST–Rab23 could still pull down an appreciable quantity of Kif17 from the importin-β2-depleted lysate (Fig. 5E). We quantified the respective signal ratio of Kif17 and importin β2 in the GST–Rab23 pulldown versus that of the lysate (input) (Fig. 5E), and it appeared that GST–Rab23 pulled down more Kif17 than importin β2 relative to their respective amounts in the cell lysate. The experiments described above suggest that Rab23 interacts with Kif17 and importin β2 in a GTP-dependent manner in cell lysates. However, the difference in the antibodies used and lack of a way to accurately determine the respective molar ratios of Kif17 and importin β2 captured by GST–Rab23 did not permit any conclusive inference on direct binding between these molecules.

Together with the potential role of importin β2 in the transport of ciliary Kif17, Verhey's group also discovered that maintenance of an asymmetric distribution of Ran–GTP across the cytoplasm and cilium is essential for the ciliary entry of Kif17 (Dishinger et al., 2010). Taking the current findings into consideration, there are now two small GTPases influencing ciliary transport of Kif17. It is unclear how the influence of Rab23 on Kif17 ciliary trafficking could be functionally connected to the Ran-mediated mechanism. We also observed that abolishment of the ciliary–cytoplasmic Ran-GTP–Ran-GDP gradient across the ciliary base (by overexpressing constitutively active mCherry–RanQ69L) led to a significant perturbation of the ciliary localization of Kif17 in NIH3T3 fibroblasts (data not shown). To ask whether Rab23 and Ran could be working cooperatively or in distinct pathways, ciliary Kif17 was quantified after combined silencing of Rab23 and overexpression of either mCherry–RanQ69L or dominant negative RanT24N mutant. Intriguingly, siRNA-mediated depletion of Rab23 in cells expressing RanQ69L exhibited the largest reduction in ciliary intensities of Kif17 (Fig. 6A,B). This additive loss of ciliary Kif17 suggests that Rab23 or Ran could have independent roles in regulating Kif17 transport to the primary cilium. Notably, overexpression of Rab23QL, but not RanQ69L, resulted in a detectable interaction between importin β2 and Kif17–mCitrine (Fig. 6D), further implying that Rab23 and Ran have different modes of ensuring that Kif17 is imported into the ciliary compartment. Hence, although activated Ran promotes dissociation of importin β2 from Kif17, activated Rab23 enhances their association.

There is controversy regarding the localization of Rab23 at the primary cilium. Although Yoshimura and co-authors did not observe Rab23 ciliary localization after overexpression of GFP–Rab23WT in hTERT-RPE1 cells (Yoshimura et al., 2007), overexpressed Rab23 has been reported to localize to the cilia of Madin–Darby canine kidney (MDCK) cells (Boehlke et al., 2010). In our hands, endogenous Rab23 was not noticeable at the primary cilia, but overexpressed wild-type Rab23 and Rab23QL displayed distinct ciliary localization in NIH3T3 (Fig. 1), hTERT-RPE1 cells (supplementary material Fig. S2A) and CRL1927 mouse mesangial cells (data not shown). By contrast, dominant-negative Rab23SN was not present at the primary cilia in these cells, whereas the phosphorylation mutant Rab23TA exhibited reduced Rab23 ciliary localization. Given that ciliary enrichment of Rab23 appeared to be mostly associated with GTP-bound forms, the ciliary targeting of Rab23 could well depend on its GTP-binding ability.

In a mass spectrometric analysis, Rab23 was identified as a protein phosphorylated by cytidine 3′,5′-cyclic monophosphate (cCMP) (Bond et al., 2007). Following this finding, no functional significance had been associated with this phosphorylation. Examining the ciliary localization of Rab23WT and Rab23TA has led us to discover that the T150 phosphorylation could serve to regulate ciliary localization of Rab23 (Fig. 1A,B,D). Pre-fixation detergent treatment of Rab23QL-expressing cells (Fig. 1E) further indicated that Rab23 is mainly affiliated with ciliary membranes and cilioplasm. This is in accordance with how the Rab-GTP–Rab-GDP molecular switch is associated with Rab proteins alternating between being attached to membrane and being soluble, cytosolic proteins (Pylypenko and Goud, 2012; Seabra and Wasmeier, 2004).

We have observed that manipulation of Rab23 levels did not significantly abrogate cilia formation, suggesting that Rab23 does not play a crucial role in the building of the ciliary architecture. However, enrichment of overexpressed Rab23 at the primary cilia could still point to a ciliary function. We discovered that ciliary localization of Kif17 depended on the GTP-binding status of Rab23 (and thus its presence at the primary cilium) as well as Rab23 expression levels in both NIH3T3 and hTERT-RPE1 cells. An increase in the ciliary localization of Kif17 following non-degradable Rab23 expression in Rab23-shRNA-treated cells further indicated that loss of ciliary Kif17 was specifically due to insufficient levels of Rab23 (Fig. 3). Notably, in assessing the ciliary localization of Kif17, Rab23 silencing in NIH3T3 and hTERT-RPE1 cells led to significant reductions in the ciliary intensities of Kif17 and the number of Kif17-positive cilia; however, the number of Kif17-positive cilia exhibiting localization of Kif17 at the distal ends were not significantly different, suggesting that Rab23 was likely to be more crucial in targeting of Kif17 towards the primary cilium than in its transport from the ciliary base to the distal tip. Taken together with ciliary FRAP experiments that showed Rab23-knockdown cells exhibiting almost no recovery of ciliary Kif17 over time compared to scrambled control cells, there is strong evidence that Rab23 could be required for the trafficking of cytoplasmic (unbleached) Kif17 towards the primary cilium.

The influence of Rab23 over Kif17 ciliary trafficking prompted us to question the specificity of the effects observed. Analysis of another Rab protein, Rab8a, which is essential in cilium formation and ciliary transport (Follit et al., 2010; Moritz et al., 2001; Nachury et al., 2007; Omori et al., 2008), could help indicate whether Kif17 ciliary targeting could be relying on general or specific ciliary trafficking regulators. In our investigations, neither overexpression of a dominant-negative Rab8a mutant or Rab8a-specific silencing resulted in any apparent perturbations in ciliary localization of Kif17 (supplementary material Fig. S3A–D). Co-immunoprecipitation with anti-Rab8a antibodies further showed no detectable interaction between Rab8a and Kif17–mCitrine or importin β2 (supplementary material Fig. S3F), suggesting that Rab8a is not likely to be crucial for the ciliary trafficking of Kif17. By contrast, although Rab8a bound to fibrocystin-CTS and influenced its ciliary localization, Rab23 did neither. The influence of Rab23 on ciliary localization of Kif17 is therefore relatively specific (supplementary material Fig. S4E,J).

Control of the ciliary entry of Kif17 is known to crucially depend on two components: the nuclear and ciliary import transport carrier importin β2, and the ciliary–cytoplasmic Ran-GTP–Ran-GDP gradient (Dishinger et al., 2010). Our co-immunoprecipitation studies showed Rab23 to be an additional player in ciliary transport of Kif17. Rab23 exists in a complex with Kif17 and importin β2, and this interaction is enhanced in a GTP-dependent manner in both NIH3T3 and hTERT-RPE1 cells (Fig. 5; supplementary material Fig. S2H). Rab23 likely facilitates binding of Kif17 to importin β2, because the interaction between importin β2 and Kif17 was mostly lost upon Rab23 depletion (Fig. 5C). Conceivably, Rab23-mediated regulation of Kif17 ciliary import could be through promoting the interaction between Kif17 and its transport carrier, importin β2. Formation of this ternary complex could in turn be important for targeting of Kif17 to the primary cilium. Our GST–Rab23 pulldown assays further suggest that Kif17 in the lysate binds Rab23 in vitro, in a manner that might even be stoichiometrically independent of importin β2. However, as we were unable to detect direct physical association between recombinant GST–Rab23 and His₆–Kif17 alone in the absence of cell lysate, Rab23–Kif17 interaction in vivo appears to require other cellular factors.

Increased loss of the Kif17 ciliary localization observed after Rab23 depletion and constitutive active RanQL overexpression, as well as the differing effect of overexpression of Ran and Rab23 on the interaction between importin β2 and Kif17, all imply that these small GTPases likely have separate roles in mediating ciliary transport of Kif17 (Fig. 6A–D). Although Rab23 is first required for binding of Kif17 to importin β2, influx of Kif17 across the ciliary barrier could be subsequently determined by the ciliary–cytosolic Ran-GTP–Ran-GDP gradient. Perhaps similar to the process of nuclear import into the nucleoplasm (Azuma and Dasso, 2000; Madrid and Weis, 2006), upon Kif17 import into the cilium, Ran–GTP would associate with importin β2 and thus release Kif17 and Rab23 into the ciliary compartment (Fig. 6E).

As mentioned, Rab GTPases are known modulators of multiple stages of exocytic and endocytic membrane trafficking in eukaryotic cells. In cycling between GDP-bound and GTP-bound states, Rab proteins are recruited to specific membrane compartments, where they come in close proximity and interact with distinct effector proteins. Rab–effector complex formation then allows for the execution of precise intracellular trafficking steps, for instance, vesicle motility. Since the 1990s, numerous reports describe that Rab proteins are involved with the regulation of the microtubule network (Horgan and McCaffrey, 2011). For example, Echard and colleagues have reported a direct, GTP-dependent interaction between Rab6A and Kif20A (kinesin-6 family) (Echard et al., 1998). A more recent paper has demonstrated how the mammalian Cos2 homologue Kif7 (kinesin-4 family), which is essential in relaying the signal transduction from smoothened to the Gli family of transcription factors, is more than likely required in formation of the ciliary structure than transport of hedgehog signaling ciliary proteins (He et al., 2014). Kif7, like Kif17, localizes to the distal ciliary tips or microtubular plus ends, but it is unclear whether Rab23 could have a role in its ciliary transport.

To date, Kif17 has been found to be involved in the regulation of NMDA receptor subunit 2B (GluN2B)-containing vesicles in cultured hippocampal dendrites and ciliary targeting of cyclic-nucleotide-gated (CNG) channels to the non-motile cilia of olfactory sensory neurons (Guillaud et al., 2003; Jenkins et al., 2006; Setou, 2000; Yin et al., 2012). Based on the strong association of Rab23 with the primary cilium, its high levels of expression in neurons (Guo et al., 2006), and the involvement of Rab23 in Kif17 ciliary trafficking, Rab23 could be required in the transport of the known cargoes of the motor protein, which are, together, essential for various neuronal functions. Beyond the neuronal context, the role of the ciliary transport of Rab23 could also extend across various cell types, such as fibroblasts and epithelial cells. The current findings could hopefully contribute towards our understanding of how Rab23-null mutations could lead to the pleiotrophic Carpenter's syndrome in humans (Alessandri et al., 2010; Jenkins et al., 2007), and partly connect some similarities observed between the symptoms of the Carpenter's syndrome and ciliopathies.

Full-length Rab23 was obtained by PCR and other standard procedures from a mouse brain cDNA library (Guo et al., 2006). Single point mutations giving rise to Rab23 Q68L, S23N and T150A mutants were generated using the GeneTailor™ site-directed mutagenesis kit (12397-014, Invitrogen Life Technologies, Carlsbad, CA). Rab23WT and mutants were subcloned into either pCIneo (E1841, Promega, Madison, WI, USA) or pmCherry-C1 vectors (632524, Clontech Laboratories Inc., Mountain View, CA, USA). Wild-type Rab23 was also cloned into a pGEX-4T-1 expression vector (28-9545-49, GE Healthcare Life Sciences, Buckinghamshire, UK) and transformed into BL21 (DE3) Escherichia coli cells (C6000-03, Invitrogen Life Technologies) for production of glutathione-S-transferase (GST)–Rab23 fusion proteins. Both pmCherry-RanQ69L (human; 30309) and pTK21-Ran T24N (human; 37396) were purchased from Addgene (Addgene, Cambridge, MA). These constructs were separately contributed by the Jay Brenman (Kazgan et al., 2010) and Iain Cheeseman (Kiyomitsu and Cheeseman, 2012) laboratories. To ensure uniformity in the comparison of Ran mutants, the Ran T24N gene fragment was isolated from pTK21 and inserted into the pmCherry-C1 vector. Kif17–mCitrine [pmCit-N1 (Dishinger et al., 2010)] was obtained from Kristen J. Verhey (University of Michigan Medical School, Ann Arbor, Michigan, MI). For ciliary FRAP analysis, Kif17 was subcloned in frame with C-terminal mCherry reporter [pmCherry-N1 (632523, Clontech Laboratories Inc.)]. Gregory J. Pazour (University of Massachusetts Medical School, Worcester, MA) and Wolfgang E. Kühn (University Medical Centre, University of Freiburg, Germany) provided the GFP–fibrocystin-CTS [pJAF99 (Follit et al., 2010)] and pLXSN-SmoothenedA1-Venus [SmoothenedA1 is a constitutive active mutant that consists of a mutation of the 549 tryptophan residue to leucine residue (Boehlke et al., 2010)] constructs, respectively. Wild-type Arl13b tagged with a C-terminal GFP reporter was acquired from an OriGene TrueORF™ cDNA clone library (MG206808, OriGene Technologies, Inc., Rockville, MD).

Rabbit antibodies against Rab23 were generated by repeated immunization with recombinant His₆-tagged Rab23 fusion proteins (Guo et al., 2006). Commercial antibodies include: anti-acetylated α-tubulin (6-11B-1 clone, T6793, Sigma-Aldrich, St Louis, MO), anti-Arl13b (17711-1-AP, Proteintech Group Inc., Chicago, IL), anti-pericentrin (ab4448, Abcam, Hong Kong), anti-GFP (ab5450, Abcam), anti-Kif17 (ab11261, Abcam), anti-transportin (558660, BD Pharmingen Inc., San Diego, CA), anti-β-actin (AC-74 clone, A2228, Sigma-Aldrich) and anti-Rab8 (610844, BD Biosciences, San Jose, CA) antibodies. Fluorescein isothiocyanate (FITC)-, Texas Red (TxR)- or Cy5-conjugated secondary antibodies used in immunofluorescence assays were purchased from Jackson ImmunoResearch Laboratories, Inc., West Grove, PA, and horseradish peroxidase (HRP)-conjugated secondary antibodies used in western immunoblotting were purchased from Thermo Fisher Scientific Inc., Waltham, MA.

NIH3T3 fibroblasts were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1 mM sodium pyruvate (11360-070, Gibco Life Technologies, Grand Island, NY) and 1× penicillin-streptomycin (15140-122, Gibco Life Technologies). hTERT-RPE1 epithelial cells were maintained in 1:1 mixture of DMEM and Ham's F-12 medium supplemented with 5% FBS and 1× penicillin-streptomycin. All cells were grown at 37°C with 5% CO₂ levels. To induce cilium formation, cells were incubated with low-serum medium (basal medium containing 0.5% FBS) for 24–48 h.

Plasmid DNA transfection was performed using Lipofectamine™ 2000 reagent (11668-019, Invitrogen Life Technologies), whereas Lipofectamine™ RNAiMAX reagent (13778-150, Invitrogen Life Technologies) was used in the siRNA-mediated knockdown of Rab23 and Rab8a. The 27-mer siRNA duplexes targeted against mouse Rab23 cDNA are: siRNA2, 5′-GACCUAACAAACAAAGGACCAAGAGAA-3′; siRNA7, 5′-GAUGCAUGAAUCAUCCAGCAGAUCGAU-3′ (in siRNA7.2 knockdown treatment, an equimolar mixture of siRNA7 and siRNA2 was prepared before silencing); and siRNA3, 5′-GCUCGUACAACCAUUGCGUAUGUUUCU-3′. hsiRNA1, 5′-AAGAAUGAGGAAGCUGAGGCACUGGCA-3′ and hsiRNA2, 5′-AACAAGAUUGAUCUUCUGGAUGAUUCT-3′ were designed to target human Rab23. The sequence that targets mouse Rab8a is based on 5′-GAAUAAGUGUGAUGUGAAUGACAAGAG-3′. The Rab23-specific shRNA sequence 5′-GGACATACTTTACAGAAAG-3′, kindly provided by Dr Heidi Liou and Dr Eyleen Goh (Duke-NUS Graduate Medical School, Singapore), targets the 5′ untranslated region (UTR) of mouse Rab23. The sequence was cloned with or without the GFP reporter into the retroviral pUEG vector (a modification of the pSIREN-RetroQ-ZsGreen1 vector) (Goh et al., 2008). Retroviruses containing scrambled and Rab23 shRNA were generated by calcium phosphate transient transfection of the retroviral plasmids into HEK293GP cells. Freshly collected 0.45-μm-filtered retroviral supernatant was then used to infect recipient NIH3T3 cells.

Cultured cells were harvested in lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% Triton X-100, 1% NP-40), containing fresh additions of protease inhibitor cocktail (11873580001, Roche Diagnostics Corporation) and 0.1 mM phenylmethanesulfonylfluoride (PMSF; P7626, Sigma-Aldrich). For GTPγs and GDP treatments (consistently carried out prior to binding), cell lysate was loaded with 40 μM of GTPγs (20-176, EMD Millipore Corporation) or GDP (20-177, EMD Millipore Corporation). In the immuno-depletion of importin β2 from cell lysate before affinity pulldown, 1 µg of anti-transportin [or mouse control IgG (sc-2025, Santa Cruz Biotechnology) in mock-depleted lysate] and protein-G–Sepharose beads (17-0618-02; GE Healthcare Life Sciences) were added to 1 mg of total cell lysate, and left to incubate at 4°C for 2 h. After centrifugation, supernatant was carefully collected for subsequent steps.

For co-immunoprecipitation experiments, cell lysate, relevant antibodies [or rabbit control IgG (sc-2027, Santa Cruz Biotechnology)] and Sepharose affinity beads [protein-A–Sepharose beads (17-0974-01, GE Healthcare Life Sciences) or cyanogen bromide (CNBr)-activated Sepharose 4B beads (17-0430-01, GE Healthcare Life Sciences)] were allowed to bind overnight. For affinity pulldown assays, GST-tagged proteins bound to glutathione–Sepharose-4B beads (GE Healthcare) were incubated with cell lysate and allowed to bind overnight. After thorough washing to ensure maximal removal of non-specific binding, bead-bound ligands were eluted with SDS sample buffer [with the addition of 100 mM dithiothreitol (DTT)]. Denatured proteins were separated by SDS-PAGE and electro-blotted onto nitrocellulose membranes. Membranes were incubated with primary antibodies, such as anti-Rab23 (1:2500), anti-β-actin (1:5000), anti-GFP (1:1000), anti-Kif17 (1:500) and anti-transportin (1:1000) antibodies, and their respective HRP-conjugated secondary antibodies (1:5000). Chemiluminescent protein bands were captured using the Chemidoc™ MP Imaging System (Bio-Rad Laboratories Inc.). Alternatively, polyacrylamide gels were stained with Coomassie Brilliant Blue R-250 solution (161-0400, Bio-Rad Laboratories Inc.) for the visualization of GST fusion proteins. ImageJ 1.45 software was used in the quantification of western blot band intensities. Densitometric data were all presented as bar charts created in Microsoft Excel.

Cells were fixed with 3.7% paraformaldehyde (P6148, Sigma-Aldrich) and permeabilized using 0.05% saponin (S7900, Sigma-Aldrich) for 15 min at room temperature. Primary (1:200) and secondary (1:400) antibodies, prepared in blocking buffer [5% FBS and 2% bovine serum albumin (BSA) diluted in 1× PBS], were directly incubated with the coverslips. Antibodies against acetylated α-tubulin and Arl13b were typically used for the marking of primary cilia, whereas anti-pericentrin antibody was used to stain the pericentriolar matrix surrounding the basal bodies. Hoechst 33342 (H1399, Molecular Probes Inc, Eugene, OR), prepared at a 1:10,000 dilution, was used as a nuclear counterstain.

Immunofluorescence images were captured with the Carl Zeiss LSM710, LSM700 (Carl Zeiss, Oberkochen, Germany) (Core facilities, Department of Biochemistry, National University of Singapore, Singapore) or Olympus FV1000 (Confocal Microscopy Unit, Yong Loo Lin School of Medicine, National University of Singapore, Singapore) confocal imaging systems, and presented as collapsed z-stacks in full resolution. ZEN 2012 edition (Carl Zeiss), FV10-ASW 4.0 Viewer (Olympus), Imaris 6.1.5 and Adobe Photoshop CS6 software were used for editing and analysis of images. Image montages were usually displayed in x-y view, unless otherwise stated. For a clearer depiction of the localization at Rab23 at the primary cilia, z-views – a z-plane projection of stacked z-slices – were also included. In particular, the ciliary intensities of ciliary cargoes were measured using the ‘histogram’ function in the ZEN 2012 software. The final, normalized ciliary intensity value was calculated by subtracting the cell background intensity from the mean fluorescence intensity. All localization data for ciliary cargoes was consolidated from at least three independent experiments and presented as bar charts, created in Microsoft Excel. For the quantification of the intensities of ciliary cargoes, bar charts were plotted so that the top (white box) and bottom (grey box) lines of each column mark the 75th percentile and 25th percentile of the data set respectively. The center line marks the median and standard errors span the mean (‘diamond’). The numbers of cells sampled for each condition are included near the bottom of all bar charts.

NIH3T3 cells were left to adhere onto 35-mm glass-bottomed dishes overnight (81151, iBidi GmbH, Martinsried, Planegg, Germany). At ∼18–24 h before imaging, cells were maintained in serum-free DMEM without Phenol Red dye (21063-029, Gibco Life Technologies). To minimize focus shifts, a 37°C temperature and 5% CO₂ level were stably maintained within an enclosed chamber. Using the Carl Zeiss LSM710 confocal microscope, the ciliary localization of Kif17 at the distal tips was photobleached using 100% laser power at 500 iterations. Images were then captured at 1-min intervals over 20 min at a lower 30% laser power. For each image, six additional z-slices were taken above and below the z-slice of the focus (the total number of z-slices taken per image was 13). Net fluorescence intensity, after normalizing the total intensity of Kif17 against the corresponding background staining (within the same z-slice), was then plotted in a phase-decay graph using GraphPad Prism version 6.00 for Windows, GraphPad Software (La Jolla, CA).

Statistical data was analyzed with SPSS statistics version 20 (IBM^®, Armonk, NY, USA). Fisher's exact tests were employed in the analysis of all categorical data. For continuous data sets, Shapiro–Wilk tests were first used to check for normal or non-normal data distribution. Once established as a normally distributed or parametric data set, Microsoft Excel's Student's t-test (unpaired, unequal variance, two-tailed) was performed for each pairwise comparison. Otherwise, for non-normal or a non-parametric data distribution, a Wilcoxon rank-sum test was applied to each pairwise comparison. In all bar charts, error bars represent s.e.m.

We sincerely thank Dr Kristen J. Verhey for Kif17–mCitrine; Dr Gregory J. Pazour for GFP–fibrocystin-CTS; Prof. Wolfgang E. Kühn for SmoothenedA1-Venus plasmids; Dr Heidi S. C. Liou and Dr Eyleen Goh for the GFP-pUEG-scrambled shRNA and Rab23 shRNA constructs. We are also grateful to Prof. Marie-Véronique Clément and Dr Gireedhar Venkatachalam for sharing the NIH3T3 and hTERT-RPE1 cell lines. We thank the reviewers for their constructive comments and suggestions, which have improved the manuscript.