PDE6δ-mediated sorting of INPP5E into the cilium is determined by cargo-carrier affinity

The phosphodiesterase 6 delta subunit (PDE6δ) shuttles several farnesylated cargos between membranes. The cargo sorting mechanism between cilia and other compartments is not understood. Here we show using the inositol polyphosphate 5′-phosphatase E (INPP5E) and the GTP-binding protein (Rheb) that cargo sorting depends on the affinity towards PDE6δ and the specificity of cargo release. High-affinity cargo is exclusively released by the ciliary transport regulator Arl3, while low-affinity cargo is released by Arl3 and its non-ciliary homologue Arl2. Structures of PDE6δ/cargo complexes reveal the molecular basis of the sorting signal which depends on the residues at the −1 and −3 positions relative to farnesylated cysteine. Structure-guided mutation allows the generation of a low-affinity INPP5E mutant which loses exclusive ciliary localization. We postulate that the affinity to PDE6δ and the release by Arl2/3 in addition to a retention signal are the determinants for cargo sorting and enrichment at its destination.

P rimary cilia are antenna-like microtubule-based cell surface protrusions which can be found on eukaryotic cells and serve as sensory organelles. Genetic disorders affecting structure or function of cilia result in a large number of diseases collectively termed ciliopathies 1,2 . While the cilium appears as a protrusion in the plasma membrane that is open to the cell body, the ciliary content and membrane composition are different than that of the cell body and plasma membrane 3,4 . This is in part achieved by the presence of a diffusion and transport barrier, where entry and exit decisions of ciliary components have to be taken 5,6 .
PDE6d is a prenyl-binding protein that was originally discovered as the delta subunit of rod photoreceptor-specific phosphodiesterase PDE6 (ref. 7). It was found as a solubilizing factor for the prenylated subunits of this enzyme and was later shown to be a general prenyl-binding protein (hence also called PrBP/PDE6d) [8][9][10][11] . PDE6d was shown to bind prenylated peptides or proteins of the Ras subfamily with approximately micromolar affinity 12,13 and to play a critical role in their cellular distribution [14][15][16] . Since it is believed to be crucial for the localization and thus the activity of the oncoprotein Ras, inhibitors of the Ras-PDE6d complex were actually considered as promising Ras drug candidates 17 .
INPP5E belongs to the inositol polyphosphate 5 0 -phosphatase family that hydrolyzes the 5 0 -phosphate of phosphatidylinositols and localizes to primary cilia 18,19 . The importance of the 5 0 -phosphatase activity for ciliary function is underscored by the finding that INPP5E is mutated in Joubert syndrome, a ciliopathy characterized by motor and intellectual disabilities [18][19][20] , and that the gene mutated in the OCRL (Oculocerebrorenal) or Lowe syndrome also encodes an inositol polyphosphate 5 0 -phosphatase 21,22 . INPP5E contains a C-terminal CaaX motif where the C-terminal residue Cys644 is farnesylated 23 . A mutation encoding a stop codon near to the CaaX motif (Q627) of INPP5E was identified in a family with MORM syndrome 18 , a ciliopathy characterized by intellectual disability, obesity, retinal dystrophy and micropenis 24 . This mutation was shown to affect INPP5E ciliary localization, which in combination with other reports 25 indicates the importance of the C-terminus and its farnesylation for the ciliary localization of INPP5E (ref. 18).
Recently, PDE6d was co-purified with INPP5E and siRNAmediated knockdown of PDE6d resulted in impaired ciliary localization of INPP5E (ref. 26). Moreover, a PDE6d deletion mutation, which was identified in Joubert syndrome, was shown to impair the targeting of farnesylated INPP5E protein to the primary cilium 25 . Knockdown of PDE6d also impeded the transport of GRK1 and PDE6 catalytic subunits to photoreceptor outer segments, which are considered specialized forms of cilia 27,28 .
The homologous small Arf-like GTP-binding proteins Arl2 and Arl3 have been shown to act as nucleotide-dependent-specific release factors of farnesylated cargo from PDE6d in vitro and in vivo. Structural and kinetic analyses have shown that Arl2/3 act allosterically to increase the dissociation rate constants for cargocarrier complexes 13,15,29,30 . In contrast, it was shown recently by pull-down experiments with cellular extracts that Arl3 but not Arl2 can efficiently release INPP5E from its complex with PDE6d (ref. 25).
In analogy to nuclear localization signals a number of different ciliary localization signals have been identified for different transmembrane proteins [31][32][33] . However, not much is known about the molecular mechanism of how these signals are recognized and how decisions on ciliary entry based on these signals are made. For certain membrane-associated, posttranslationally modified proteins carrying an N-terminal myristoyl or a C-terminal prenyl motif, it has been shown that the import into cilia is dependent on the carrier proteins PDE6d, UNC119a and UNC119b and on Arl3 as displacement factor 13,25,28,30,34 . However, it has been extensively documented that Ras proteins as well as Rheb require PDE6d for their proper localization at the plasma membrane or internal membranes, but do not appear to be localized in cilia 15,16 .
This begs the question about the mechanism of PDE6dmediated sorting of farnesylated cargo between the cilium and other cellular compartments. Thus, we set out to investigate the molecular basis of farnesylated cargo sorting using ciliary INPP5E and non-ciliary Rheb as an example. Here, we show that a 100-fold difference in the binding affinity of farnesylated cargo with PDE6d and the specific release of high-affinity cargo by activated Arl3GTP determines cargo sorting into cilia, while low-affinity cargo can be released by both Arl3GTP and Arl2GTP and stays outside the cilium. Moreover, we show by structural, biochemical and cell biological approaches, how and why the binding affinity is dependent on the residues at the À 1 and À 3 positions preceding the farnesylated cysteine and that sorting of farnesylated cargo can be manipulated by changing the affinity to PDE6d.

INPP5E and Rheb localization and binding affinity to PDE6d.
Using IMCD3 cells stably expressing either INPP5E or Rheb fused to a localization and tandem affinity purification (LAP) tag 35 , we can show that INPP5E localizes almost exclusively to the primary cilium with very small fraction in the cell body ( Fig. 1a; upper), which is consistent with previous reports 18,19,26 . In contrast, Rheb mainly localizes to endomembranes ( Fig. 1a; lower), this observation is consistent with previous reports 13,36 . Given that the prenyl-binding protein PDE6d is the shuttle factor mediating the localization of INPP5E and Rheb 13,16,18,25,26 , we set out to characterize the interaction of PDE6d with INPP5E and Rheb. Previously we have shown that farnesylated C-terminal peptides derived from Rheb or KRas bind to PDE6d in exactly the same way and with similar affinities as the full-length farnesylated proteins 12,13 . Hence, we used a fluorescently labelled C-terminal farnesylated and carboxy-methylated peptide of INPP5E (residues 637-644) and Rheb (residues 175-181) to measure the affinity to PDE6d by fluorescence polarization. Figure 1b (left) shows that PDE6d binds to INPP5E peptide with low nanomolar affinity (K d ¼ 3.7 nM ± 0.2, ± indicates s.d., n ¼ 9). In contrast, the affinity between PDE6d and the farnesylated C-terminal peptide of Rheb falls into the submicromolar range (K d ¼ 445 ± 83 nM, ± indicates s.d., n ¼ 10) (Fig. 1b; right), which is in the same range with the previously described values 12,13 . These data raised the question, whether the almost 100-fold higher affinity of INPP5E towards PDE6d as compared to Rheb is involved in the sorting mechanism of these two proteins to different destinations.
High-affinity cargo is specifically released by Arl3GTP. Towards an explanation for the possible sorting mechanism that leaves some PDE6d-cargo in the cell body but allows others to be enriched in the cilia we turned to the release activities of Arl2 and Arl3. Both GTP-binding proteins in their active conformation have been shown to be responsible for releasing cargo from PDE6d. While Arl2 is a non-ciliary protein, Arl3 localizes along the length of the cilium 37 . Using fluorescence polarization, we measured the release of INPP5E and Rheb peptides from PDE6d by the addition of Arl2 or Arl3 bound to the non-hydrolysable GTP analogue GppNHp. The data show that Rheb peptide can be released by both Arl2GppNHp and Arl3GppNHp (Fig. 2a), supporting earlier observations 13 . In contrast, INPP5E peptide can only be released by Arl3GppNHp under the same conditions ( Fig. 2b). To compare the cargo release kinetics of Arl2GppNHp and Arl3GppNHp, we measured the dissociation rate constants of INPP5E and Rheb peptides from PDE6d in the presence and absence of Arl3GppNHp or Arl2GppNHp, by adding a large excess of unlabelled peptide to silence the back reaction. In the absence of Arl2/3, Rheb showed an intrinsic dissociation rate (k off ¼ 0.95 ± 0.004 s À 1 , ± indicates s.d., n ¼ 4), while no measurable dissociation rate could be observed for INPP5E in a reasonable time window. This observation is in line with the almost 100-fold difference in the binding affinity between both peptides determined from the steady state equilibrium measurements. The presence of Arl3GppNHp or Arl2GppNHp has a similar acceleration effect on the dissociation rate of Rheb peptide from PDE6d (k off ¼ 27.2±0.7 and 15.3±0.3 s À 1 , respectively, ± indicates s.d., n ¼ 4) (Fig. 2c,d). However, the release of INPP5E peptide in the presence of Arl3GppNHp shows an estimated 10,000-fold acceleration (k off ¼ 10.7±0.2 s À 1 ,±indicates s.d., n ¼ 4), while release by Arl2GppNHp (k off ¼ 0.018 ± 0.0005 s À 1 , ± indicates s.d., n ¼ 4) is almost 600-fold slower (Fig. 2e,f). Taken together, our data suggest that high-affinity farnesylated cargo can be specifically released by Arl3, while low-affinity cargo can be released similarly by both Arl2 and Arl3.
Role of Arl3 N-terminal helix in the release mechanism. Previously we have shown that the N-terminal helix of Arl3 is important to release myristoylated cargo from a complex with the shuttle factor UNC119 (ref. 30). To find out whether the N-terminus of Arl3 and/or Arl2 has a similar if any role in the interaction with PDE6d, fluorescence polarization measurements using full-length Arl3 (Arl3 fl ) or an N-terminal truncated form (Arl3 DN ) were performed. Supplementary Fig. 1 shows that Arl3 DN is unable to release the INPP5E peptide from PDE6d as compared with Arl3 fl . To investigate the role of the N-terminal helix of Arl3 in the release mechanism, we measured association and dissociation rate constants to determine the affinity of PDE6d towards Arl2 and Arl3 in both full-length and N-terminal truncated forms. Association rate constants between the four proteins Arl3 fl , Arl3 DN , Arl2 fl and Arl2 DN are rather similar although association is almost twice as fast for full-length Arl3 as compared with Arl2 (Fig. 3a,b). In contrast, determination of the dissociation rate constants shows large differences. While the difference in k off between full-length protein Arl2 fl and N-terminal deleted Arl2 DN is only threefold, Arl3 fl shows a 26-fold higher residence time with PDE6d, as compared with Arl3 DN (Fig. 3c-e). By calculating the equilibrium dissociation constants (   316 ± 6.3 nM, respectively, ± indicates s.d., n ¼ 4), whereas Arl3 fl has an affinity in the low nanomolar range (K d ¼ 5.8 ± 0.5 nM, ± indicates s.d., n ¼ 4) (Fig. 3f). The K d values for Arl2 fl and Arl3 fl differ from previously determined values 38 , likely because of the different techniques used.
Our data suggest that the N-terminal helix of Arl3 makes a significant contribution to the interaction with PDE6d and increases the affinity between the proteins by 37-fold. This additional input of Arl3 compared with Arl2 is probably a major factor in the ability of Arl3 to release high-affinity farnesylated cargo from PDE6d. A similar effect was shown for the Arl3/UNC119 complex where in contrast to Arl2 (and any other Arf protein), the N-terminal helix of Arl3 did not detach from the surface of the protein after the GDP-GTP conformational change and actively participates in the release mechanism in the closed position 30 .
The sorting signal of PDE6d-related farnesylated cargo. To investigate the nature of the affinity difference between INPP5E and Rheb peptides towards PDE6d in more details, we solved the crystal structure of the INPP5E peptide in complex with PDE6d at 1.85 Å resolution (data collection and refinement statistics summarized in Supplementary Table 1). Superimposition of the INPP5E peptide/PDE6d complex with the structure of PDE6d in complex with Rheb (PDB code: 3T5G) shows that the immunoglobulin-like b-sandwich folds of PDE6d overlay well with an r.m.s. deviation of 0.5731 Å. The proteins show a hydrophobic cavity, where the farnesyl moieties of INPP5E and Rheb are inserted ( Fig. 4a; upper). The prenyl groups overlay well and make an identical interaction pattern with the surrounding hydrophobic residues of PDE6d ( Fig. 4a; lower). However, the side chains of the residues on the À 1 and À 3 positions upstream of the farnesylated cysteine (the 0 position) in INPP5E and Rheb show different contacts with PDE6d. As shown in Fig. 4b (upper), the serine side chain of Rheb on the À 1 position makes a hydrogen bond with the side chain of glutamic acid (Glu88) from PDE6d, whereas the hydrophobic side chain of the isoleucine of INPP5E at the equivalent position is situated in a highly hydrophobic environment mediated by five hydrophobic residues of PDE6d (Val80, Trp90, Met118, Leu123 and Ile128). On the other hand, the lysine side chain of Rheb at the À 3 position is pointing away from the binding pocket of PDE6d, while the serine side chain of INPP5E at the equivalent position makes a hydrogen bond with the side chain of glutamic acid (Glu88) ( Fig. 4b; lower).
Thus, we reasoned that the different contact patterns of INPP5E and Rheb peptides with PDE6d are responsible for the difference in affinities. To prove this, we generated two peptides, where the amino acids on the À 1 and À 3 positions were swapped between INPP5E and Rheb, creating INPP5E(KS) (S641K/I643S) and Rheb(SI) (K178S/S180I) peptides. Affinities of the swapped peptides to PDE6d were determined by titrating increasing amounts of unlabelled INPP5E(KS) and Rheb(SI) into a preformed complex of fluorescent Rheb peptide with PDE6d and monitoring the displacement by the decrease in fluorescence polarization. Analysis of the data with a competition model derived from the law of mass action as described 17,39 shows that the affinities to PDE6d can be reversed, with a K d values of (697 ± 54 nM, ± indicates s.d., n ¼ 14) for INPP5E(KS) and (12 ± 2.7 nM, ± indicates s.d., n ¼ 12) for Rheb(SI) (Fig. 4c).
To confirm the conclusion relating to the À 1 and À 3 positions, we measured the affinities of farnesylated peptides derived from rhodopsin kinase GRK1 and the g-subunit of transducin GNGT1 (Tg) with PDE6d. It is important to note that, GRK1 carries Met and Ser at À 1 and À 3 positions similarly with INPP5E, whereas GNGT1 (Tg) carries Gly and Lys at À 1 and À 3 positions similarly with Rheb ( Supplementary Fig. 2a). The results showed high binding affinity (7.2 ± 1.3 nM, ±indicates s.d., n ¼ 12) of GRK1 and low binding affinity (6,573 ± 477 nM, ± indicates s.d., n ¼ 9) of Tg for PDE6d ( Supplementary Fig. 2b). These data suggest that the binding affinity between PDE6d and farnesylated cargo is dependent on the sequence of the farnesylated C-terminus, in particular on the À 1 and À 3 positions relative to the farnesylated cysteine.

Dependency of INPP5E ciliary localization on PDE6d and Arl3.
To test whether reducing the affinity of INPP5E to PDE6d is affecting its ciliary localization, we stably transfected the INPP5E(KS) mutant into IMCD3 cells and compared its localization with INPP5E(WT). Figure 5a shows that INPP5E(KS) mutant is not enriched in cilia anymore but is localized all over the cell including the cilium, while INPP5E(WT) is highly enriched in cilia with only a minor fraction in the cell body (Fig. 1a). Evaluation of mean fluorescence intensity ratio between cilia and whole cell shows that INPP5E(WT) has a 5.3-fold enrichment in  the cilia, while the INPP5E(KS) mutant loses its ciliary enrichment and is more evenly distributed over the entire cell ( Fig. 5b and Supplementary Fig. 3). We propose that the mislocalization of INPP5E(KS) mutant could result from its weak affinity to PDE6d, which enables its release by Arl2 outside the cilium, resulting in its retention at the endomembranes. To support this assumption, we used the stably transfected IMCD3 cells expressing INPP5E(WT) or mutant INPP5E(KS) and performed a GST pull-down experiment with PDE6d in the presence and absence of Arl3GppNHp or Arl2GppNHp. The results show that the INPP5E(KS) mutant can indeed be released by both Arl2GppNHp and Arl3GppNHp, while INPP5E(WT) is specifically released only by Arl3GppNHp (Fig. 5c). Confirming with this, siRNAmediated knockdown of Arl3 shows loss of dominant ciliary localization of INPP5E and its redistribution between cilia and cellular endomembranes ( Fig. 6 and Supplementary Fig. 4).
In line with these experiments, we tested whether increasing the affinity of Rheb to PDE6d permits its ciliary entry. For this we stably transfected the Rheb(SI) mutant into IMCD3 cells and compared its localization to that of Rheb(WT). Rheb(SI) showed a more than fourfold increase in ciliary localization as compared with Rheb(WT) (Fig. 7). This result indicates that increasing the affinity of Rheb towards PDE6d shifts the equilibrium of Rheb distribution towards the cilium as compared to the entire cell. The non-exclusive ciliary localization of Rheb(SI) mutant could be explained by the absence of a Rheb specific retention signal inside the cilia.
Taken together, our data suggest that the high binding affinity between INPP5E and PDE6d and the specific release by Arl3GTP are essential determinants for the ciliary localization of INPP5E.

Discussion
Consistent with our previous reports 12, 13 , here we show that nonciliary farnesylated cargo such as Rheb binds to PDE6d with submicromolar affinity. Interestingly, the binding affinity between PDE6d and the ciliary farnesylated protein INPP5E is in the low nanomolar range. Structural analysis revealed that the residues at the À 1 and À 3 positions relative to the farnesylated cysteine are the determinants for the binding affinity to PDE6d. This finding was confirmed by mutational analysis and by the binding affinity measurements of farnesylated peptides derived from rhodopsin kinase (GRK1) and the g-subunit of transducin (Tg). The high binding affinity of GRK1 to PDE6d could explain its mislocalization in the outer segment of photoreceptor in the absence of PDE6d, while Tg, which has a low-affinity to PDE6d, is only minimally affected 28 . The latter suggests that another farnesyl binding protein might exist to take over the role as a shuttle factor for Tg or that the ciliary entry of the heterotrimeric transducin does not rely solely on the farnesylated g-subunit. Our findings suggest that the affinity of farnesylated cargo is an essential determinant of its PDE6d-mediated sorting into the ciliary compartment.
It has been reported that Arl3 is localized in the cytoplasm and inside cilia 37 , while no ciliary localization for Arl2 has been reported so far. Considering that the complex between highaffinity cargo such as INPP5E or GRK1 with PDE6d can be released specifically by Arl3 and that both proteins are highly enriched in cilia, one would have to predict that the active GTP-bound form of Arl3 is only localized inside the cilium and thus is able to release cargo exclusively in this compartment. This assumption is supported by our recent study which showed that the ciliary protein Arl13B is the specific guanine nucleotide exchange factor for Arl3 (ref. 40) as well as by studies showing that retinitis pigmentosa 2 (RP2), the GTPase activating protein of Arl3, localizes at the basal body of the cilium or the preciliary region 41,42 , so that Arl3GTP should reside exclusively inside the cilium and would get hydrolyzed to Arl3GDP while exiting the cilium. Confirming with this, Arl3 does not seem to take over the role of Arl2 in releasing low-affinity farnesylated cytosolic cargo, as siRNA-mediated knockdown of Arl2 was shown to be sufficient to mislocalize KRas (ref. 15). Thus, our data suggest that high-affinity farnesylated cargo is specifically released by Arl3 inside cilia and Arl2 is specific for the release of low-affinity cargo outside cilia.
Our results are apparently not in agreement with previous results 25,26 , who showed that the transport of INPP5E is independent of Arl3. In these reports, data were analysed in terms of ciliary localization (INPP5E-positive cilia), not taking the distribution of INPP5E between cilia and the entire cell into account. Such analysis has enabled us to determine the fold enrichment of INPP5E inside cilia and how it is affected by either changing the affinity to PDE6d or by Arl3 knockdown. The redistribution of INPP5E in the cells, which were treated with siRNA against Arl3, showed similar but generally weaker effect as compared with the redistribution of the low-affinity mutant  (Figs 5b and 6b). The effect of Arl3 knockdown might be limited by the incomplete knockdown and by the fact that staining of INPP5E inside cilia does not differentiate between free or PDE6d-bound phosphatase. Both ciliary cargo and Arl3 seem to bind to PDE6d with high affinities, non-ciliary cargo and Arl2 on the other hand bind to PDE6d with low affinities. Thus we assume that the cargo release by Arl3 inside cilia or Arl2 in the cytosol might not be complete at comparable concentrations of all components. As a consequence an additional signal would be required to drive the equilibrium to completion and to retain cargo at its destination. A retention signal could be achieved by the interaction with membrane or other interacting partners. The endomembrane system offers a large surface area and could play the role as retention signal for cytosolic farnesylated cargo such as Rheb. A possible ciliary retention signal for INPP5E could be Arl13B. The specific ciliary protein Arl13B has been shown to directly interact with INPP5E and its knockdown results in INPP5E mislocalization 26 .

INPP5E(KS)
In this report, we propose a three step model for PDE6dmediated sorting of farnesylated cargo into different cellular compartments. The binding affinity of farnesylated cargo to PDE6d is the first fundamental step in the sorting mechanism, followed by the specific release of high-affinity cargo by Arl3 inside cilia or the release of low-affinity cargo by Arl2 in the entire cell. Finally, a retention signal keeps the farnesylated cargo at its destination (Fig. 8). Interfering with any of these steps can provide valuable insights in studying the role of INPP5E in ciliopathies especially that a mutation which influences its localization to cilia is associated with MORM syndrome. Furthermore INPP5E localization studies for Arl13B patient mutations associated with Joubert syndrome will deepen our understanding of the molecular basis of ciliopathies. Finally it would be interesting to exploit available small molecules that inhibit the interaction of PDE6d with farnesylated cargo in studying the role of INPP5E in cilia and ciliopathies. Proteins. All proteins were expressed in Escherichia coli strain BL21-Codon-Plus(DE3)-RIL. Cells were induced at OD B0.6 with 100 mM IPTG and incubated at 20°C overnight. Cells were harvested and lysed in lyses buffer (25 mM Tris-HCl, pH 7.5, 150 mM NaCl and 1 mM b-mercaptoethanol, 1 mM PMSF) using French press. Supernatants of C-terminal histidine-tagged full-length Arl3, Arl2 and N-terminal histidine-tagged PDE6d were loaded onto a Ni-NTA column (QIAGEN). Proteins were eluted with elution buffer (25 mM Tris-HCl, pH 7.5, 150 mM NaCl and 1 mM b-mercaptoethanol, 250 mM imidazole), followed by gel filtration on a Superdex 75 S26/60 column using elution buffer without imidazole. Supernatants of N-terminal GST-tagged truncated Arl3 and Arl2 (aa 18-177, 17-178, respectively) were expressed, harvested and lysed similar to the histidinetagged proteins. The supernatants were loaded onto GSH-column (Amersham Biosciences). Proteins were eluted with elution buffer (25 mM Tris-HCl, pH 7.5, 150 mM NaCl and 1 mM b-mercaptoethanol, 20 mM glutathione). The GST-fusion proteins were separated from the tag by proteolytic cleavage followed by gel filtration on a Superdex 75 S26/60 column using elution buffer without glutathione. Nucleotide exchange of the GDP bound Arl3 and Arl2 proteins was achieved by overnight incubation at 4°C with 4 U mg À 1 alkaline phosphatase (Roche Diagnostics) and 1.5-fold excess of the non-hydrlysable GTP analogue (GppNHp) or the fluorescently labelled GppNHp (mantGppNHp) and followed by gel filtration. Crystallization and structure determination. The INPP5E-peptide (SQNSSTIC(Far)-OMe) was dissolved in DMSO and mixed with 500 mM solution of PDE6d at 1:1 molar ratio in a buffer containing 25 mM Tris-HCl (pH 7.5), 150 mM NaCl and 3 mM DTE. The crystals appeared in Protein Complex suite from Qiagen, 1.4 M sodium malonate (at 20°C) and were flash frozen in a cryoprotectant solution that contains the mother liquor in addition to 16% (v/v) glycerol. Diffraction data set was collected at the X10SA beamline of the Suisse Light Source, Villigen. XDS program was used for data processing. The structure was solved by molecular replacement using Molrep from CCP4 (suite) and PDE6d from the PDE6d-farnesylated Rheb complex (PDB code: 3T5G) as a search model. The farnesylated INPP5E peptide was built using WinCoot and refinement was done with REFMAC5. Refinement and data collection statistics are summarized in Supplementary Table 1. Structure coordinates were deposited in the Protein Data Bank (PDB code 5F2U). A stereo image of a portion of the electron density map is displayed in Supplementary Fig. 5. Cell culture and stable cell line generation. Mouse renal epithelial cells from the inner medullary collecting duct containing a stably integrated FRT cassette (IMCD3 Flp-In, kind gift from M.V. Nachury lab; Flp-In cell line technology by Life technologies) were cultured at 37°C and 5% CO 2 in DMEM/F-12, HEPES (Life technologies) complemented with 10% fetal bovine serum and 2 mM L-Glutamine. Stable cell lines were generated as previously described 43,44 . Briefly, IMCD3 cells were seeded in six-well plates at a density of 100,000 cells per well. On the following day the cells with a confluence of 40-60% were cotransfected with the pG-LAP3 vector (Addgene) containing the gene of interest and pOG44 vector (Life technologies) encoding the FLP recombinase using Lipofectamine 2,000 (Life technologies). Transfected cells were selected with hygromycine in a concentration of 100-200 mg ml À 1 complemented culture medium. Expression of the respective proteins was proven by immunoblotting with an anti-GFP antibody (1:500; Santa Cruz Biotechnology sc-9996).
Immunostaining and microscopy. IMCD3 cells stably expressing GFP-tagged protein were plated on poly-L-lysine coated coverslips in six-well plates, each well containing 100,000 cells. Twenty-four hours later, cilia were induced by 48 h serum starvation. Cells were washed in PBS and fixed with 4% formaldehyde in cytoskeletal buffer (2,75 M NaCl, 100 mM KCl, 25 mM Na 2 HPO 4 , 8 mM KH 2 PO 4 , 40 mM MgCl 2 , 40 mM EGTA, 100 mM PIPES, 100 mM Glucose, pH 6.0) for 20 min. After two washes with PBS cells were permeabilized with 0.3% Triton X100 in cytoskeletal buffer for 10 min. Cells were rinsed in 0.1% Tween20 in PBS and blocked in 10% FBS in PBS for 30 min. For immunostaining of primary cilia, mouse 6-11B-1 anti-acetylated tubulin antibody (1:5,000; Sigma T6793) in 10% FBS in PBS was incubated overnight at 4°C. Alexa Fluor 647 anti-mouse secondary antibody (1:800; Life technologies A-31571) was added for 45 min at room temperature after washing four times with 0.1% Tween20 in PBS. Coverslips were rinsed three times in 0.1% Tween20 in PBS and afterwards in PBS. Nuclei were stained with DAPI (Serva), diluted 1:10,000 in PBS for 1 min. After three washes with PBS, coverslips were fixed on glass slides with Mowiol (Merck). Images were taken using an Olympus IX81 microscope with a CCD camera and a 60x NA 1.35 oil immersion objective.
Knockdown experiment. The INPP5E(WT) stable cell line was plated on poly-Llysine coated coverslips in six-well plates at a density of 100,000 cells per well. After 24 h cells were transiently transfected with Lipofectamine 2,000 with siRNAs directed against mouse Arl3 and a negative control siRNA, following the manufactureŕs recommendations. The siRNAs against Arl3 and for a negative control were provided from Qiagen with the following sequences: for Arl3 (sense: 5 0 -GGGUCAGGAACUAACGGAATT-3 0 , antisense: 5 0 -UUCCGUUAGU UCCUGACCCGT-3 0 ); for negative control (sense: 5 0 -UUCUCCGAACGUGUC ACGUdTdT-3 0 , antisense: 5 0 -ACGUGACACGUUCGGAGAAdTdT-3 0 ). Eightyfour hours later, cells were serum-starved for 24 h and subsequently treated for immunofluorescence microscopy as described before. Image collection was performed utilizing identical settings for every sample.
GST pull-down assay. IMCD3 cells stably expressing GFP-INPP5E(WT) or GFP-INPP5E(KS) were lysed in lysis buffer containing 75 mM Hepes pH 7.5, 150 mM KCl, 1.5 mM EGTA, 1.5 mM MgCl 2 , 15% glycerol, 0.2% NP-40 and one protease inhibitor cocktail tablet (Roche). Cell lysates were cleared and supernatants were incubated for 1 h at 4°C with 100 ml GSH-beads conjugated with 20 mM GST-PDE6d. For the release assay, 20 mM of either Arl2 or Arl3 were added to the previous mixture and incubated for further 1 h at 4°C. After 5 times washing with the lysis buffer, the complexes were analysed by western blotting using anti-GFP antibody (1:500; Santa Cruz Biotechnology sc-9996) and anti GST (1:5,000; home source). Full scans of western blots are provided in Supplementary  Fig. 6.