Ciliary accumulation of human INPP5E requires the conserved FDRxLYL motif (aa 609–615) (Humbert et al., 2012). This motif lies downstream of INPP5E’s highly conserved phosphatase domain, as defined by the InterPro protein signature database (InterPro domain IPR000300, aa 297–599) (Humbert et al., 2012; Blum et al., 2021). However, INPP5E crystal structure reveals a more extensive globular phosphatase domain, spanning residues 282–623, on whose surface the FDRxLYL motif folds (PDB ID: 2xsw) (Figure 1a; Tresaugues et al., 2010). Indeed, the crystallographic data show that the FDRxLYL motif, and the alpha-helix it is nested in, interact with several other catalytic domain residues (such interactions include D610-L362, Y614-P358, R620-E347, and R621-E354). Recently, a 3D model of full-length INPP5E was generated using AlphaFold, a remarkably accurate machine learning algorithm for protein structure prediction (Jumper et al., 2021). The AlphaFold model closely matches the crystal structure, including the FDRxLYL motif’s structure and location. Thus, based on the structural data, we conclude that INPP5E’s catalytic domain spans residues 282–623, and therefore encompasses the FDRxLYL motif, which lies on the domain’s surface. Accordingly, from now on, when we speak of the catalytic or phosphatase domain, we will be referring to the globular domain spanning residues 282–623, unless we specify otherwise.

To clarify how INPP5E’s phosphatase domain controls ciliary targeting, we generated a series of INPP5E deletion mutants lacking different portions of the N or C-terminal regions, or lacking only the InterPro-defined catalytic domain (Figure 1b). Interestingly, INPP5E ciliary accumulation closely correlated with the presence of an intact structurally-defined catalytic domain (aa 282–623) (Figure 1b–c). Deletion of the InterPro-defined catalytic domain (Δ297–599) completely abolished ciliary targeting in EGFP-INPP5E-transfected hTERT-RPE1 cells (Figure 1c). N-terminal deletions not affecting the phosphatase domain did not affect ciliary localization, as was the case for constructs 100–644, 200–644, 251–644, and 274–644 (Figure 1c). In contrast, 288–644 displayed strongly reduced ciliary targeting, and 351–644, 451–644 and 551–644 completely failed to localize to cilia (Figure 1c). C-terminal deletions showed a similar pattern. As previously reported, the MORM mutant (1-626) was found inside cilia and at the ciliary base, and the same was true for 1–623 (Figure 1c; Jacoby et al., 2009). The 1–621 mutant was only occasionally ciliary, at low levels near the ciliary base (Figure 1c). Further deletions from the C-terminus resulted in complete loss of ciliary targeting, as was the case for 1–618, 1–616, 1–608, and 1–283 (Figure 1c). Hence, integrity of the structurally defined catalytic domain is essential for INPP5E ciliary accumulation.

At least partly, this lack of ciliary targeting could be due to protein instability of these mutants. To assess this, we used western blot to measure the expression levels of many of these mutants in transfected HEK293T cells. Given the nature of this expression system, we expect any changes in protein levels between constructs to be likely due to changes in protein stability, at least when comparing proteins of similar sizes (if not, then slower translation rates of longer mRNAs need to be accounted for). Other factors affecting protein levels in other contexts are unlikely to play important roles in this system, including changes in transfection efficiency (always very high in HEK293T), transcriptional regulation (expression of all constructs driven by constitutive and strong cytomegalovirus promoter), or post-transcriptional mRNA regulation (no anticipated differences between constructs).

When we did these experiments, we observed a clear effect of the catalytic domain (aa 282–623) on EGFP-INPP5E protein levels (Figure 1—figure supplement 1). This was very clear, for instance, when comparing 1–626 to 1–618, and also 274–644 to 288–644. In both cases, as catalytic domain residues where removed in the second mutant of each pair, protein levels decreased sharply, by at least 4-fold. On the other hand, virtually complete removal of the catalytic domain in Δ297–599 only reduced expression moderately, about 2-fold. Yet ciliary targeting of this mutant was completely abolished, not just reduced by half (Figure 1c). Altogether, these data suggest that the catalytic domain, in addition to affecting protein stability, has a specific role in promoting INPP5E ciliary accumulation.

We first confirmed that, as previously reported, the FDRxLYL motif is essential for INPP5E ciliary targeting (Figure 2a; Humbert et al., 2012). Mutation to alanines of both the FDR (aa 609–611) and LYL (aa 613–615) triplets completely abolished ciliary localization (Figure 2a). To assess the relative importance of each residue within the FDRxLYL motif, we also made the individual alanine mutants (F609A, D610A, R611A, E612A, L613A, Y614A, L615A). Surprisingly, all of them still localized to cilia, indicating redundancy within the FDR and LYL triplets (Figure 2—figure supplement 1).

Given that the FDRxLYL motif lies on the catalytic domain surface (Figure 1a), and that catalytic domain integrity is essential for ciliary targeting (Figure 1b–c), we hypothesized that INPP5E ciliary targeting relies on a catalytic domain surface including not only the FDRxLYL residues, but also other residues whose proximity to FDRxLYL is dependent upon a folded domain. To test this, we examined the 3D structure of INPP5E’s catalytic domain (PDB ID: 2XSW) in order to identify candidate residues for this putative surface. We did this by selecting residues meeting all or most of the following criteria: (i) located near FDRxLYL motif, on same side of domain; (ii) high exposure to solvent; and (iii) highly conserved in vertebrate INPP5E orthologs. Out of this analysis, we selected sixteen candidate residues: R345, R346, E347, W348, E349, Q353, E354, Y360, V361, R378, R379, D380, I382, W383, F384, E387. Alanine mutation of most of these residues did not affect ciliary targeting of EGFP-INPP5E in hTERT-RPE1 cells (Figure 2b–c). Exceptions were W348A, Y360A, W383A, and R378A+R379 A, all of which completely failed to accumulate in cilia (Figure 2c–d). The R378 and R379 arginines are redundant, as the double but not single mutations impeded cilia localization (Figure 2c–d).

Mistargeting of these four mutants could be due to loss of catalytic domain integrity, which is critical for ciliary targeting (Figure 1). If so, then phosphatase activity would also be disrupted in these mutants. To test this, we immunoprecipitated these EGFP-INPP5E mutants from transfected HEK293T lysates and measured their PI(4,5)P₂ 5-phosphatase activity in the immunoprecipitates (IPs). EGFP-INPP5E wild type was used as positive control, whereas negative controls included: (i) a reaction with substrate but no enzyme, to assess the rate of basal PI(4,5)P₂ dephosphorylation; and (ii) the catalytically inactive D477N INPP5E mutant, lacking a critical active site aspartate but normally localizing to cilia (Bielas et al., 2009; Kong et al., 2000; Garcia-Gonzalo et al., 2015; Kong et al., 2006; Whisstock et al., 2000).

Compared to negative controls, phosphate release by WT was about 12-fold faster, a very significant difference (Figure 2e). W348A and Y360A were completely inactive, and R378A+R379 A nearly so (Figure 2e). Hence, lack of ciliary targeting in these mutants may be non-specific. Unlike R378A+R379 A, the cilia-localized R378A and R379A single mutants retained activity, either fully (R378A) or partially (R379A). Like R379A, W383A was half as active as WT, with a very significant ≈6-fold increase over negative controls (Figure 2e). The same was true for the LYL triplet mutant, whereas the FDR mutant was somewhat less active, with a ≈3-fold increase over controls (Figure 2a–e). Quantitation by anti-EGFP immunoblot of the protein levels of each mutant in the IPs used to measure activities revealed no significant differences for any of the mutants (Figure 2f). We also analyzed expression levels of these mutants at 48 hr post-transfection by Western blot of lysates in a separate HEK293T cell experiment, in which protein levels of the W383A, LYL and FDR mutants were ≈75%, 60%, and 50% of WT, respectively. The lowest expression was seen for W348A, Y360A, R378A+R379 A, and R379A, all between 20 and40% of WT (Figure 2—figure supplement 2).

All of this suggests that W383 and FDRxLYL function as bona fide CLSs, as their strict requirement for ciliary targeting cannot fully be accounted for by their moderate effects on enzyme activity or protein levels. Accordingly, R379A, a mutation immediately adjacent to FDRxLYL and W383, displayed similar moderate effects on activity and protein levels, but had no effect on ciliary localization (Figure 2b–f).

Still, the partial activity loss in W383A and FDRxLYL mutants suggests that catalytic domain integrity may be partially lost as well. To examine this further, we assessed the half-lives of these mutants in HEK293T cells using the protein translation inhibitor cycloheximide (Figure 2—figure supplement 3). After 5, 10 or 24 hr in cycloheximide, EGFP-INPP5E(WT) levels were virtually unaffected, indicating a long half-life of several days. For W383A and the triplet FDR and LYL mutants, initial levels, at 24 h post-transfection, were not significantly different from WT (Figure 2—figure supplement 3a-b). From those initial levels (100%), W383A fell to 60% in the first 10 hr, but then remained at 60% by 24 hr, rather than diminishing further (Figure 2—figure supplement 3c). Similar results were seen for the FDR mutant, whereas LYL’s curve ran closer to WT’s (Figure 2—figure supplement 3c).

Thus, W383A showed a biphasic kinetics, whose most parsimonious explanation appears to be the existence of two protein populations: ≈40% of W383A would be unstable with a half-life of ≈5 hr, whereas the remaining ≈60% would be stable, with a half-life of days, like WT (Figure 2—figure supplement 3c). Presumably, the unstable form would not be properly folded and would be inactive, whereas the stable form would be folded and active, which would explain why W383A reduces activity ≈2-fold (Figure 2e). Although this model remains speculative, the fact that ≈50–60% of W383A is stable and enzymatically active supports the idea that W383 is a bona fide CLS, since loss of ciliary targeting in W383A is virtually complete, and much stronger than a twofold reduction (Figure 2c–d). Similar points support the specificity of FDRxLYL as CLS.

We also explored how W383 substitution to residues other than alanine affects ciliary targeting and enzyme activity (Figure 2—figure supplement 4). To do this, we mutated W383 to another aromatic residue (W383F), to several aliphatic residues (W383I, W383L, W383M, W383V), or to acid (W383E) or basic (W383R) residues. Interestingly, W383F was fully active and ciliary, indicating that an aromatic ring at this position suffices for both functions (Figure 2—figure supplement 4). In contrast, all other mutations fully suppressed ciliary targeting, like W383A (Figure 2—figure supplement 4a). Interestingly, enzyme activity of W383I, W383L, W383M, and W383V was the same as for W383A, suggesting that an aromatic ring in this position is important for both targeting and activity (Figure 2—figure supplement 4b-c). Overall, the data in this section show that W383 and FDRxLYL, despite their moderate effects on enzyme activity, function as specific CLSs to target INPP5E to cilia.

Besides W383 and FDRxLYL, the C-terminal CaaX box (641-CSVS-644) also modulates INPP5E ciliary targeting. Although CaaX box deletion causes mistargeting of some INPP5E molecules to the ciliary base or other cellular membranes, CaaX box mutants still accumulate in cilia (Figure 1c; Jacoby et al., 2009; Thomas et al., 2014; Kösling et al., 2018). This is somewhat puzzling, as INPP5E ciliary targeting strongly depends on the farnesyl receptor PDE6D (Thomas et al., 2014).

In the course of our experiments, we made a serendipitous observation that led us to novel insights into how non-farnesylated INPP5E manages to still accumulate in cilia. Initially, we found that INPP5E’s N-terminus (aa 1–273), while being dispensable for ciliary targeting of INPP5E (Figure 1b–c), is required for ciliary targeting of INPP5E’s MORM mutant (Δ627–644) (Figure 3e). We subsequently saw that this effect of the N-terminus is mediated by residues 251–273, and that these residues are strongly required for ciliary targeting of the C641S mutant, in which the farnesylated CaaX box cysteine is replaced by non-farnesylatable serine (Figure 3a–c). Thus, while both Δ251–273 and C641S single mutants localized inside cilia, the double (Δ251–273)+C641 S mutant completely failed to do so, despite accumulating at the ciliary base (Figure 3a–c). Interestingly, upon careful observation and quantitation, both single mutants already showed a partial loss of intraciliary targeting, aside from their ciliary base accumulation (Figure 3a–c). Thus, the single Δ251–273 mutant behaves as previously reported for CaaX mutants (and as shown here for C641S), except that Δ251–273 reduces ciliary targeting even more than C641S (Figure 3c). These data indicate that the CaaX box and residues 251–273 cooperate to target INPP5E inside cilia: not only are both sequences required for optimal ciliary targeting of the wild type protein, but their functions are partially redundant, each becoming essential when the other one is missing (Figure 3a–c).

Even though both these mutations are outside the phosphatase domain, it remains possible that their effects on ciliary targeting are due to disruption of catalytic domain integrity. To test this, we measured the PI(4,5)P₂ 5-phosphatase activity of the aforementioned 274–626 mutant, lacking both the N-terminus (Δ1–273) and the MORM region (Δ627–644). This mutant was fully active, with activity and protein levels indistinguishable from WT (Figure 3—figure supplement 1). In contrast, the 288–626 mutant, additionally missing residues 275–287 at the beginning of the catalytic domain, had much less activity and reduced levels (Figure 3—figure supplement 1). Therefore, residues 251–273 and 627–644 are completely dispensable for enzyme activity, indicating that their effects on ciliary targeting are not caused by disruptions in the phosphatase domain. This was confirmed by Western blot of HEK293T lysates transfected with (Δ251–273)+C641 S or the single Δ251–273 and C641S mutants: in all cases protein levels were indistinguishable from WT (Figure 3—figure supplement 2). Thus, residues 251–273 and the CaaX box also behave as bona fide CLSs.

We then mapped which residues within 251–273 are responsible for CLS function. To do this, we started with the 274–626 mutant, which fails to accumulate in cilia as mentioned above. To this mutant, we gradually added residues to the N-terminus, thus creating four more mutants: 269–626, 264–626, 257–626 and 251–626 (Figure 3d). Of these, 269–626 and 264–626 failed to accumulate in cilia, whereas 257–626 and 251–626 readily did so (Figure 3e). Next, starting from 257 to 626, we generated five alanine-substitution mutants within the 257-FSLLAPIRSKDV-268 region, removing residues FS, LL, PIR, SK, and DV, respectively (Figure 3f). Of these, residues LL and PIR were essential for ciliary targeting, whereas FS, SK and DV were fully dispensable (Figure 3f–g). We then created the single residue mutants spanning the 259-LLAPIR-264 motif. All these mutants (L259A, L260A, P262A, I263A, and R264A) abolished cilia localization of 257–626, although weak residual targeting was still observed only for L259A (Figure 3f–g). Protein levels of 257–626 and the individual LLxPIR motif point mutants were all similar to WT (Figure 3—figure supplement 2). Therefore, the LLxPIR motif is a novel CLS that cooperates with the CaaX box to mediate optimal ciliary targeting of INPP5E.

The above data indicate that INPP5E ciliary targeting depends on four CLSs, which we will heretofore refer to as CLS1 (the LLxPIR motif, aa 259–264), CLS2 (W383), CLS3 (the FDRxLYL motif, aa 609–615), and CLS4 (the CaaX box, aa 641–644). Our data actually show that CLS3 goes beyond the FDRxLYL motif, as deletion of residues 616–621 also interferes with ciliary targeting (Figure 1b–c). This is consistent with the original report on CLS3, where residues 619–621 were also shown to modulate ciliary targeting (Humbert et al., 2012). Thus, CLS3 spans residues 609–621 in human INPP5E.

Since INPP5E ciliary targeting has been reported in different vertebrate species, we examined whether CLS1-4 are conserved in vertebrate evolution. To do this, we aligned the human INPP5E sequence with those of another mammal (mouse), a bird (crow), a reptile (python), an amphibian (toad), and a fish (zebrafish; Figure 4a). From this analysis, it is clear that all four CLSs are highly conserved in vertebrates. For CLS1, the consensus sequence is [VL]LxPIR, with only the first leucine admitting a conservative change (Figure 4a; Figure 3g). CLS2’s tryptophan is fully conserved, and so is CLS3, as previously shown (consensus: FDRxLYLxGI[KR]RR) (Figure 4a; Humbert et al., 2012). For CLS4 the consensus is C[ST][IV]S, which in all cases encodes a farnesyl transferase-specific CaaX box (Wright and Philips, 2006). Thus, the four CLSs are highly conserved. Moreover, they are all found on the same side of INPP5E according to the AlphaFold model, even though CLS1 location in this model has a low confidence score (Figure 4b; Jumper et al., 2021).

By affecting INPP5E ciliary targeting, these CLSs are expected to affect INPP5E ciliary function. To test this, we assessed the ability of the CLS mutants to rescue the well-documented abnormal ciliary accumulation of TULP3 in INPP5E-KO cells (Garcia-Gonzalo et al., 2015; Chávez et al., 2015). We performed this assay using INPP5E-KO hTERT-RPE1 cells, which we generated in two steps. First, by using CRISPR/Cas9 technology, we created puromycin-sensitive hTERT-RPE1 cells to knock out the puromycin acetyltransferase (PAC) gene, stably inserted into these cells upon their immortalization (Figure 4—figure supplement 1a). As expected, the PAC-KO clones thus obtained were highly sensitive to puromycin, in contrast to the original hTERT-RPE1 cells (Figure 4—figure supplement 1b-c). In the second step, using puromycin selection-based CRISPR on these cells, we generated INPP5E-KO cells, as confirmed by genomic analysis (Figure 4—figure supplement 2a). Immunofluorescence with anti-INPP5E antibodies confirmed the absence of INPP5E from these cells, despite some non-specific non-ciliary staining (Figure 4—figure supplement 2b). As expected, anti-TULP3 antibodies revealed a massive accumulation of this protein in the INPP5E-KO cilia (Figure 4—figure supplement 2c).

In non-transfected or EGFP-transfected INPP5E-KO cells, the percentage of TULP3-negative cilia was very low (≈10%) (Figure 4c–d). As shown previously in a similar assay using Inpp5e-null mouse embryonic fibroblasts (MEFs; Garcia-Gonzalo et al., 2015), transfection of EGFP-INPP5E WT into our INPP5E-KO cells effectively rescued the phenotype, driving the percentage of TULP3-negative cilia to ≈90%. Also as previously reported, the catalytically inactive D477N mutant completely failed to rescue (Figure 4c–d; Garcia-Gonzalo et al., 2015). Having validated the assay, we assessed rescue by the following CLS mutants: Δ251–273 (ΔCLS1), W383A (ΔCLS2), F609A+D610A+R611 A (ΔCLS3), C641S (ΔCLS4), the double ΔCLS1+ΔCLS4 mutant (ΔCLS1 +4), and the catalytic domain deletion Δ297–599. Interestingly, ΔCLS4 was fully functional in this assay, ΔCLS1 a little less so, and ΔCLS1 +4 had no rescue activity whatsoever (Figure 4c–d). Thus, the relationship between CLS1 and CLS4 in this assay largely mirrors the redundancy observed in their ciliary targeting (Figure 3c). As expected, ΔCLS3 and Δ297–599 were also completely inactive (Figure 4c–d). To our surprise, ΔCLS2 showed partial rescue: less than ΔCLS1, but more than ΔCLS3 (Figure 4c–d). The reasons for this partial rescue are still unclear to us and will require further investigation. In any case, it remains true that all non-ciliary CLS mutants showed strongly reduced functionality in the TULP3 rescue assay.

We then examined how CLS1-4 affect INPP5E cilia localization as determined by methanol fixation of hTERT-RPE1 cells (rather than fixing first with paraformaldehyde (PFA), as done so far). We did this on the basis of a recent study showing that, in these cells, live imaging of GFP-INPP5E shows a strong accumulation at the ciliary base (in addition to some intraciliary staining), and that this ciliary base accumulation is best recapitulated with methanol-only fixation (whereas PFA is better suited to visualize the intraciliary staining) (Conduit et al., 2021). Indeed, with methanol fixation, EGFP-INPP5E WT was clearly accumulated at the base of most cilia (≈60%), with the remaining cilia showing a more uniform distribution along the axoneme (≈40%) (Figure 4—figure supplement 3a-b). Regardless of cilia base-enrichment, some intraciliary axonemal staining was always seen with EGFP-INPP5E WT under these conditions. In contrast, no axonemal staining was ever observed for any of the CLS mutants (ΔCLS1, ΔCLS2, ΔCLS3, ΔCLS4, or ΔCLS1+4) (Figure 4—figure supplement 3a-b). This confirms the importance of all these CLSs for INPP5E to reach inside cilia. It also shows that the axonemal staining seen with ΔCLS1 and ΔCLS4 in PFA-fixed cells (Figure 3b–c) is lost with the methanol-only fixation (Figure 4—figure supplement 3a-b). Interestingly, with the latter fixation, ΔCLS1, ΔCLS4, and ΔCLS1+4 were all strongly accumulated at the ciliary base (in 100% of cilia). These accumulation occurred at the ciliary transition zone, distal from basal body and proximal from axoneme, as revealed by signal intensity scans along these cilia (Figure 4—figure supplement 3c). In contrast, ΔCLS3, was never seen at the ciliary base, whereas ΔCLS2 was occasionally seen there (30% of cilia) (Figure 4—figure supplement 3a-b). Altogether, the data in this section confirm the importance of INPP5E’s four CLSs.

INPP5E ciliary targeting is dependent on CLS1, CLS2, CLS3, and CLS4 (Figures 2—4), and on the integrity of its catalytic domain (Figure 1). The MORM mutation (Δ627–644), which deletes CLS4 and moderately reduces cilia localization, as mentioned above, is the only known INPP5E ciliopathy mutation affecting its ciliary targeting. However, whether and how other ciliopathy mutations in INPP5E also affect ciliary targeting is largely unknown. To address this, we sought ciliopathy mutations locating near a CLS, or that were likely to compromise catalytic domain integrity. In this manner, we identified twelve mutations, eleven from JBTS and one from LCA-like disease (Figure 5a–b; Bachmann-Gagescu et al., 2015; Bielas et al., 2009; Travaglini et al., 2013; de Goede et al., 2016; Xu et al., 2015; Wang et al., 2013).

One of the mutations, C641R, replaces the farnesylatable cysteine in CLS4 by an arginine. Not surprisingly, this mutation affected ciliary targeting in the same way as C641S or MORM, and did not affect protein levels (Figure 5b–d). Several other mutations affected residues located near the CLS2-CLS3 region. These included R345S, T355M, R378C, C385Y, V388L, R621Q, and R621W. None of these affected ciliary localization, despite causing moderate reductions in protein levels (Figure 5b–d and g). Since almost all INPP5E ciliopathy mutations affect the catalytic domain, none were found near CLS1. The closest one to CLS1 was G286R, at the beginning of the catalytic domain. Interestingly, G286R largely abolished ciliary targeting, although this is more likely due to its affecting catalytic domain integrity (Figure 5b–e). We also tested mutations, like W474R and V303M, that we thought likely to disrupt the catalytic domain, given their location at the core of the beta-sandwich. Indeed, W474R abolished ciliary targeting, and V303M reduced it considerably, with levels of both reduced (Figure 5b–d and f). Since W474R and G286R were found in homozygosis in JBTS patients, this suggests that some JBTS patients cannot target INPP5E to cilia (Figure 5a). Finally, we also tested the D490Y mutation, located near a beta-hairpin close to the active site, but far from any CLS. D490Y did not affect ciliary targeting, despite a strong reduction in protein levels (Figure 5b–e). Altogether, these data show that INPP5E ciliary targeting is sometimes affected in Joubert syndrome, which could contribute to pathogenesis in these cases.

After identifying novel INPP5E CLSs and exploring their role in ciliopathies, we focused on the mechanisms of action of these CLSs. Presumably, these CLSs act by binding to other proteins implicated in INPP5E ciliary targeting. One such protein is PDE6D, a ciliary cargo receptor for prenylated proteins. Although PDE6D binding to INPP5E is CLS4-dependent and CLS3-independent, whether its interaction requires our newly identified CLSs (CLS1 and CLS2) is unknown (Thomas et al., 2014; Humbert et al., 2012; Qiu et al., 2021; Rao et al., 2016). To test this, we cotransfected HEK293T cells with plasmids encoding Flag-PDE6D and EGFP-INPP5E in order to perform coimmunoprecipitation (co-IP) experiments (Figure 6a). As expected, Flag-PDE6D robustly co-immunoprecipitated (co-IPed) with EGFP-INPP5E(WT), but not EGFP control (Figure 6a). This interaction was completely dependent on CLS4, and completely independent of CLS1, CLS2, and CLS3 (Figure 6a). Accordingly, ΔCLS1 +4 behaved the same as ΔCLS4, with virtually no co-IP observed, whereas ΔCLS2 and ΔCLS3 co-IPed the same as WT (Figure 6a). We also tested whether the co-IP involved INPP5E’s N-terminal (aa 1–283) or C-terminal (251-644) regions, with the latter being the case for PDE6D (Figure 6a).

In addition to co-IP, we also studied the INPP5E-PDE6D interaction in vivo by means of co-recruitment assays (Figure 6e–f). To do this, we generated bait constructs expressing mCherry-FKBP-INPP5E fusion proteins (WT, ΔCLS1, ΔCLS2, ΔCLS3, ΔCLS4, or ΔCLS1+4), or mCherry-FKBP as negative control. Each of these bait constructs was separately co-expressed in HeLa cells with the prey construct (mVenus-PDE6D), and with FRB-CFP-CaaX, whose prenylated CaaX box tethers FRB to the inner leaflet of the plasma membrane (Figure 6e). Upon inducing the FKBP-FRB interaction with rapamycin, this should recruit bait constructs to the plasma membrane, a recruitment that can be monitored and quantitated by total internal reflection fluorescence (TIRF) microscopy (Figure 6e–f; Lin et al., 2013; Gallego et al., 2013). Additionally, if bait and prey interact, then prey co-recruitment to the plasma membrane will also be observed (Figure 6e–f). As expected, rapamycin-induced robust plasma membrane recruitment of all bait constructs (Figure 6g). Also as expected, mVenus-PDE6D co-recruitment was much higher with mCherry-FKBP-INPP5E(WT) than with the mCherry-FKBP control, confirming the specificity of PDE6D-INPP5E binding (Figure 6g–h). As observed in the co-IPs, mVenus-PDE6D co-recruitment was strongly reduced by both the ΔCLS4 and ΔCLS1+4 mutations, indicating a strong CLS4-dependence (Figure 6g–h). Also consistent with the co-IPs, ΔCLS1 and ΔCLS3 did not affect mVenus-PDE6D co-recruitment, whereas ΔCLS2 caused a modest reduction (Figure 6g–h). Altogether, the co-IP and in vivo co-recruitment data demonstrate that CLS4 is the key CLS controlling INPP5E-PDE6D binding.

RPGR also interacts with INPP5E and is required for its ciliary targeting (Rao et al., 2016). Moreover, RPGR ciliary targeting is also dependent on PDE6D, which binds to both its geranylgeranylated CaaX box and its RCC1-like domain (Zhang et al., 2019; Rao et al., 2016; Dutta and Seo, 2016; Lee and Seo, 2015; Fansa et al., 2015; Wätzlich et al., 2013; Vössing et al., 2021). Despite all these connections, how RPGR mediates INPP5E ciliary targeting is unclear. To address this, we performed co-IPs between Flag-RPGR and the same EGFP-INPP5E constructs used above for the PDE6D co-IPs. As with PDE6D, the INPP5E-RPGR co-IP was abolished by ΔCLS4 and ΔCLS1+4, but was untouched by ΔCLS1, ΔCLS2 or ΔCLS3 (Figure 6b). Therefore, INPP5E’s farnesylated CaaX box is also key for its association to RPGR. Consistently, RPGR strongly co-IPed with INPP5E’s C-terminal fragment (251-644) but not with the N-terminal one (1-283).

ARL13B is another key mediator of INPP5E ciliary targeting (Humbert et al., 2012; Fujisawa et al., 2021; Qiu et al., 2021). ARL13B regulates INPP5E in at least two different ways. First, ARL13B acts as a guanine nucleotide exchange factor (GEF) for ARL3, whose active GTP-bound form promotes dissociation of the PDE6D-INPP5E complex after it reaches the ciliary lumen (Gotthardt et al., 2015; Fansa et al., 2016; Ivanova et al., 2017; ElMaghloob et al., 2021). Additionally, ARL13B directly interacts with INPP5E and is required for its ciliary retention (Humbert et al., 2012; Fujisawa et al., 2021; Qiu et al., 2021). Since the first mechanism is mediated by ARL3, we checked whether INPP5E binds ARL3. However, we detected no co-IP between EGFP-INPP5E and ARL3-myc, or its constitutively active form ARL3(Q71L)-myc, in accordance with previous data (data not shown; Humbert et al., 2012). Likewise, we found no interaction between INPP5E and BART, a protein that cooperates with ARL13B as a co-GEF for ARL3 (data not shown; ElMaghloob et al., 2021).

We then carried out co-IPs to assess the CLS-dependence of the ARL13B-INPP5E interaction. We readily detected co-IP of endogenous ARL13B with EGFP-INPP5E in HEK293T cells. This co-IP was much lower with ΔCLS4 and ΔCLS1+4, but not clearly affected by CLS1-3 (Figure 6c). As with PDE6D and RPGR, INPP5E’s C-terminal region (251-644), but not the N-terminal (1-283), sufficed for the INPP5E-ARL13B interaction (Figure 6c). Since CLS3 was previously reported to affect INPP5E-ARL13B co-IP, we examined this question further (Humbert et al., 2012). In total, we performed 11 co-IP experiments between these proteins. In three of them (including the one in Figure 6c), we measured co-IP of endogenous ARL13B. In the other eight, we measured co-IP of co-transfected ARL13B-Flag. Quantitation of all these experiments showed no discernible effect of CLS1-3 on ARL13B co-IP, while largely confirming the effect of CLS4 (Figure 6d). Since the previously reported effect had been observed using deletions (ΔFDR, ΔLYL) rather than alanine-substitutions (FDR >AAA, LYL >AAA) of the CLS3 residues, we also performed an experiment comparing ARL13B-Flag co-IP for all these four mutants. This experiment showed no differences among any of these mutants, all of which co-IPed ARL13B like WT control (data not shown). Thus, we conclude that CLS4, but not CLS1-3, affect ARL13B co-IP by INPP5E in our experiments.

We also studied INPP5E-ARL13B association by means of co-recruitment assays like those in Figure 6e–h. In this case, ARL13B-EYFP was used as prey instead of mVenus-PDE6D (Figure 6i). ARL13B-EYFP co-recruitment strongly increased with mCherry-FKBP-INPP5E(WT), as compared to mCherry-FKBP alone, demonstrating a specific interaction. This interaction was unaffected by ΔCLS1 but was strongly reduced with ΔCLS2, ΔCLS3, ΔCLS4 and ΔCLS1+4 (Figure 6i). Therefore, in addition to confirming the key role of CLS4, the in vivo co-recruitment data showed that CLS2-3 do affect INPP5E-ARL13B association after all. The reason why the effects of CLS2-3 are readily detectable in our co-recruitment but not out co-IP assays remains to be determined. Taken together, our data indicate that CLS2, CLS3 and CLS4 all promote INPP5E association with ARL13B.

TULP3 is a ciliary trafficking adapter needed for ciliary targeting of membrane proteins such as G-protein-coupled receptors, polycystins, ARL13B and INPP5E (Badgandi et al., 2017; Han et al., 2019; Mukhopadhyay et al., 2010; Mukhopadhyay et al., 2013; Legué and Liem, 2019; Hilgendorf et al., 2019; Hwang et al., 2019). However, whether INPP5E and TULP3 interact is not known. We therefore tested this. Indeed, EGFP-INPP5E specifically co-IPed TULP3-myc (Figure 7a). Such co-IP was unaffected by ΔCLS1, ΔCLS4, and ΔCLS1+4, but was clearly reduced by ΔCLS2 and ΔCLS3 (Figure 7a–b). Consistently, TULP3 interacted strongly with EGFP-INPP5E(251-644), and much less so with EGFP-INPP5E(1-283) (Figure 7a).

TULP3 functions as an adapter by connecting the IFT trafficking machinery (which it binds via its N-terminal domain, NTD: aa 1–183) to membrane proteins (which it binds via its phosphoinositide-binding C-terminal Tubby domain, CTD: aa 184–442) (Mukhopadhyay et al., 2010). To test how TULP3 binds INPP5E, we assessed co-IP of Flag-INPP5E by different EGFP-TULP3 constructs (Figure 7c). Flag-INPP5E specifically co-IPed with full length TULP3, an interaction that was largely dependent on TULP3’s CTD, even though weak binding to NTD was also observed (Figure 7c). In addition, we tested Flag-INPP5E binding to two TULP3 mutants, namely K268A+R270 A and K389A. The former cannot bind phosphoinositides, and may also be hypoacetylated (Mukhopadhyay et al., 2010; Kerek et al., 2021), whereas the latter removes a key lysine needed for TULP3 to interact with ARL13B and target it to cilia, according to a recent preprint (Palicharla et al., 2021). Interestingly, both mutations clearly reduced INPP5E binding, though none abolished it completely (Figure 7c). Therefore, our data show that INPP5E interacts with TULP3, and that this interaction is dependent on: (i) CLS2 and CLS3 in INPP5E’s catalytic domain, and (ii) TULP3’s Tubby domain and its ability to bind ARL13B and phosphoinositides (Figure 7d).

INPP5E also interacts with CEP164, a ciliary base protein essential for ciliogenesis. In CEP164-silenced non-ciliated cells, INPP5E fails to accumulate at the centrosome (Humbert et al., 2012). This suggests that CEP164, by recruiting INPP5E to the ciliary base, may contribute to its ciliary targeting. Because of this, we also examined the INPP5E-CEP164 interaction. To do this, we looked at how EGFP-INPP5E and its mutants co-IP endogenous CEP164 in HEK293T cells. Interestingly, while WT, ΔCLS1, ΔCLS4 and ΔCLS1+4 all co-IPed similar amounts of CEP164, the ΔCLS2 and ΔCLS3 mutants displayed a stronger interaction, suggesting that CLS2 and CLS3 downregulate CEP164 binding (Figure 7e). This effect of CLS2-3 was confirmed after quantitation of five similar experiments (Figure 7f). Moreover, CEP164 strongly interacted with INPP5E’s N-terminal fragment (1-283), but not with the CLS2/3-containing C-terminal one (251-644; Figure 7e). Since much more CEP164 was pulled down by EGFP-INPP5E(1-283) than EGFP-INPP5E(WT), despite both fusion proteins being expressed similarly, this further indicates that CLS2-3 antagonize CEP164 binding, which is mediated by INPP5E’s N-terminal region (Figure 7e).

CEP164 contains a WW domain near its N-terminus and several coiled coils in the rest of its long sequence (1460 aa) (Cajanek and Nigg, 2014; Schmidt et al., 2012; Graser et al., 2007). Since WW domains, like SH3 domains, interact with proline-rich ligands, we hypothesized that CEP164’s WW might interact with INPP5E’s proline-rich N-terminus (Macias et al., 2002). To test this, we first checked whether Flag-INPP5E was co-IPed by three CEP164 fragments spanning its N-terminal (NT, aa 1–467), middle (Mid, aa 468–1135) and C-terminal (CT, aa 1136–1460) regions (Cajanek and Nigg, 2014). Consistent with our hypothesis, CEP164(NT)-EGFP strongly interacted with Flag-INPP5E, as did full-length CEP164-EGFP (Figure 7g). Instead, no interaction was seen with CEP164(Mid)-EGFP and only a very weak one with CEP164(CT)-EGFP. Moreover, a mutation disrupting CEP164’s WW domain (WWmut: Y73A+Y74 A) abolished INPP5E binding to CEP164(NT)-EGFP (Figure 7g; Cajanek and Nigg, 2014). Hence, CEP164’s NT is sufficient for INPP5E binding, and CEP164’s WW domain is required for it.

Altogether, these data suggest a model of CLS2-3 action: presumably, excessively strong binding to CEP164 would retain INPP5E at the ciliary base and prevent its translocation into the ciliary compartment. CLS2-3 might overcome this by loosening the CEP164-INPP5E interaction. If this is the main reason why CLS2-3 are required for INPP5E ciliary targeting, then deletion of the CEP164-interacting N-terminal region should rescue ciliary targeting of INPP5E-ΔCLS2 and INPP5E-ΔCLS3, as CLS2-3 would no longer be needed for CEP164 dissociation. To test this, we combined the ΔCLS2 and ΔCLS3 mutations with the Δ1–273 deletion, thereby generating the 274-644(ΔCLS2) and 274-644(ΔCLS3) mutants. Unlike the 274–644 control, which readily accumulated in cilia, both 274-644(ΔCLS2) and 274-644(ΔCLS3) completely failed to accumulate in cilia, just as the single ΔCLS2 and ΔCLS3 mutants (Figure 7h). Hence, even though CLS2-3 promote CEP164 dissociation, this is not sufficient for INPP5E ciliary targeting (Figure 7i). This might be due to CLS2-3 being required for binding to other ciliary trafficking proteins, such as TULP3 (Figure 7a–d).

The INPP5E-CEP164 interaction is also known to regulate ciliogenesis. Specifically, INPP5E prevents ectopic ciliogenesis by regulating the interaction between CEP164 and TTBK2, a key ciliogenic kinase that is in turn regulated by casein kinase 2 (CSNK2A1) (Xu et al., 2016; Loukil et al., 2021). Since both TTBK2-CEP164 and INPP5E-CEP164 interactions involve CEP164’s WW domain, we reasoned that the CSNK2A1-TTBK2 pathway might also regulate INPP5E-CEP164 binding and hence INPP5E ciliary targeting (Xu et al., 2016). Indeed, we found INPP5E ciliary levels significantly reduced in CSNK2A1-null fibroblasts (Figure 7—figure supplement 1a-b). Also consistent with the above hypothesis, CSNK2A1 did not co-IP with INPP5E, suggesting that CSNK2A1’s effect is indirect (Figure 7—figure supplement 1c). Whether it is mediated by TTBK2 awaits further study.

Recent work shows that INPP5E ciliary targeting requires ATG16L1, an autophagy protein that forms a complex with IFT-B complex component IFT20 (Boukhalfa et al., 2021; Finetti et al., 2021). Moreover, ATG16L1 was shown to interact with INPP5E and its product PI4P (Boukhalfa et al., 2021). We therefore assessed the CLS-dependence of the ATG16L1-INPP5E interaction. Interestingly, although the single ΔCLS1, ΔCLS2, ΔCLS3, and ΔCLS4 mutants did not noticeably alter binding between EGFP-INPP5E and Flag-ATG16L1, a reduction was observed with the ΔCLS1 +4 mutant, suggesting that CLS1 and CLS4 may cooperate in ATG16L1 binding, thus mirroring their cooperation in INPP5E ciliary targeting (Figure 8a). Consistent with CLS1 and CLS4 partaking in the interaction, the C-terminal INPP5E fragment (251–644, containing both CLS1 and CLS4) was sufficient for binding, whereas the N-terminal fragment (1–283, containing only CLS1) interacted only weakly (Figure 8a). In three independent experiments, the ΔCLS1 +4 mutant was always found to interact less than the corresponding single mutants or WT (Figure 8b). However, despite the strong tendency, one-way ANOVA did not show a significant reduction with ΔCLS1 +4 (Figure 8b). Therefore, more independent experiments will be needed to clarify this point. In any case, these data suggest that CLS1 and CLS4 may jointly be implicated in how ATG16L1 targets INPP5E to cilia (Figure 8c). On the other hand, CLS2 and CLS3 would act via CEP164, TULP3, and ARL13B, with the latter also acting via CLS4, like PDE6D and RPGR (Figure 8c).

Most cell types form primary cilia when their centrosomes are not engaged in cell division. The main exception to this is the hematopoietic lineage, where the centrosome is often engaged in other specialized structures, such as the immune synapse (IS) in lymphocytes (Stinchcombe and Griffiths, 2014). Interestingly, numerous parallels exist between primary cilia and the IS, such as the use of IFT trafficking machinery (Stinchcombe and Griffiths, 2014; Finetti et al., 2015). ARL13B and ARL3 were shown to localize to the IS, and a recent preprint has shown that INPP5E does as well (Powell et al., 2021; Stephen et al., 2018; Chiu et al., 2021). With advice from the preprint authors, we confirmed that endogenous INPP5E accumulates at the IS between Jurkat T-cells and Raji antigen-presenting cells, a well-established IS model (Figure 8—figure supplement 1a-b; Chiu et al., 2021; Herranz et al., 2019; Montoya et al., 2002; Calvo and Izquierdo, 2021). Given the parallels between cilia and IS, we wondered whether IS targeting of INPP5E shares the same mechanisms that INPP5E uses for ciliary targeting in other cell types. To test this, we assessed the CLS-dependence of INPP5E IS targeting. As also reported in the preprint, EGFP-INPP5E WT is also detected at the Jurkat-Raji IS (Figure 8—figure supplement 1c; Chiu et al., 2021). This localization was not noticeably perturbed in the ΔCLS1, ΔCLS2, ΔCLS3, ΔCLS4, or ΔCLS1+4 mutants (Figure 8—figure supplement 1c). Therefore, INPP5E IS targeting does not follow the same mechanisms as INPP5E ciliary targeting.

For information on all the plasmids used in this study, see the Key Resources Table. Site-directed mutagenesis was performed by overlap extension PCR. Amplifications for all cloning were performed with Platinum SuperFi DNA polymerase (Thermofisher), and all finished constructs were validated by DNA sequencing (Eurofins Genomics).

All cell lines were grown at 37 °C and 5% CO₂ in a humidified atmosphere and were regularly tested to ensure they were mycoplasma-free. hTERT-RPE1 cells were cultured in DMEM/F12 basal medium supplemented with 10% fetal bovine serum (FBS) and were reverse transfected using JetPrime (Polyplus-transfection), and their cilia analyzed 48 hr later, after 24 hr of serum starvation. HEK293T cells were maintained in DMEM +10% FBS, transfected using the calcium phosphate method, and lysed 40–48 hr later. HeLa cells were cultured in DMEM +10% FBS +Penicillin/Streptomycin and transfected using FuGENE 6 (Promega). Co-recruitment assays were performed 24 hr after transfection. CRISPR-engineered control and Csnk2a1-null mouse embryonic fibroblasts have been described elsewhere (Loukil et al., 2021). Raji B and Jurkat T (clone JE6.1) cell lines from ATCC were cultured in RPMI-1640 medium containing L-glutamine, penicillin/streptomycin, and 10% heat-inactivated FBS. All cell lines used in this study have been validated and used extensively in previous publications, with the exception of CRISPR cell lines, whose validation is described below.

For information on all the antibodies used in this study, see the Key Resources Table. hTERT-RPE1 cells grown to confluence on coverslips were fixed 5 min in PBS +4% paraformaldehyde (PFA) at room temperature (RT), followed by freezer-cold methanol for 3 min at −20 °C. Cells were then blocked and permeabilized for 30–60 min at RT in PBS +0.1% Triton X100 +2% donkey serum +0.02% sodium azide (blocking solution). Coverslips were then incubated in a humidified chamber for 2 hr at RT (or overnight, 4 °C) with blocking solution-diluted primary antibodies. After three PBS washes, PBS-diluted secondary antibodies and DAPI (1 µg/ml, Thermofisher) were added for 1 hr at RT in the dark. After three more PBS washes, coverslips were mounted on slides using Prolong Diamond (Thermofisher), incubated overnight at 4 °C, and imaged with a Nikon Ti fluorescence microscope. Brightness and contrast of microscopic images were adjusted for optimal visualization using Adobe Photoshop or Fiji (Image J). For methanol-only fixation (Figure 4—figure supplement 3), cells were fixed 5 min in freezer-cold methanol at –20 °C, washed thrice in PBS, blocked as above but without Triton X100, and stained with antibodies as above.

Puromycin-sensitive PAC-KO hTERT-RPE1 cells, and from them INPP5E-KO cells, were generated using CRISPR/Cas9 methods essentially as we described previously (Barbeito et al., 2021). The sequences of guide RNAs and information on the plasmids encoding them are in the Key Resources Table. PAC-KO cells were not analyzed genomically, but their puromycin sensitivity was confirmed in dose-response assays as shown in Figure 4—figure supplement 1b-c. The genomic analysis of INPP5E-KO cells was performed as previously reported (Figure 4—figure supplement 2a; Barbeito et al., 2021).

For the rescue assays, INPP5E-KO hTERT-RPE1 cells were reverse transfected on coverslips using JetPrime (500 ng DNA and 1 µl JetPrime reagent, in 50 µl JetPrime buffer, were prepared following manufacturer’s instructions and added to 2·10⁵ suspended cells in 1 ml serum-containing medium, per well). Transfection medium was replaced 4 hr post-transfection (hpt). At 24 hpt, cells were starved with DMEM:F12 +0.2% FBS. At 48 hpt, cells were fixed (PFA +methanol), stained and visualized as described above. For quantitation, transfected-cell cilia without visible TULP3 staining were expressed as percentage relative to total transfected-cell cilia.

HEK293T cells were lysed 40–48 hr post-transfection in buffer containing 50 mM Tris-HCl pH = 7.5, 150 mM NaCl, 1% Igepal CA-630 (Sigma-Aldrich) and 1 X Halt protease inhibitor cocktail (Thermofisher, #78429). Lysates were then rotated (15 min, 4 °C) and centrifuged (10 min, 20,000 g, 4 °C), and protein levels in the postnuclear supernatants measured with Pierce BCA Protein Assay Kit (Thermofisher). After equalizing protein amount and concentration in all samples, EGFP fusion proteins were immunoprecipitated with GFP-Trap magnetic agarose (GFP-Trap_MA, Chromotek) beads for 2 hr or overnight at 4 °C with rotation. Beads were then washed thrice in lysis buffer without protease inhibitors, eluted with 2 X Laemmli buffer containing 200 mM DTT, and boiled 5 min at 95 °C. SDS-PAGE and Western blots were performed as previously described (Barbeito et al., 2021; Martin-Hurtado et al., 2019). For information on all antibodies used for western blot, see the Key Resources Table.

Activity assays were performed essentially as described (Bielas et al., 2009). Briefly, HEK293T cell lysates were obtained as above, and their protein levels measured with Pierce BCA Protein Assay kit (Thermofisher). EGFP-INPP5E or its mutants were then immunoprecipitated from 1 mg of cell lysate by overnight rotation at 4 °C with GFP-Trap_MA beads (Chromotek). Beads were then washed thrice in buffer containing 50 mM Tris-HCl pH = 7.5 and 150 mM NaCl buffer (no protease inhibitors or detergent), and twice more in activity buffer (50 mM Tris-HCl pH = 7.5, 150 mM NaCl, 3 mM MgCl₂ and 0.1% octyl-β-D-glucopyranoside (Alfa Aesar)). For the activity assays, beads were incubated in activity buffer supplemented with 120 µM diC8-PtdIns(4,5)P2, from Echelon Biosciences. After incubating the enzyme reactions for 20 min at 37 °C, the supernatant was retrieved from the beads and its phosphate concentration measured at 620 nm using the Malachite Green Assay Kit (Echelon Biosciences). Beads were then processed for western blot as above.

Co-recruitment assays were performed using Eclipse Ti microscope (Nikon, Japan) with a 100 X TIRF objective (1.0 X zoom and 4X4 binning) in TIRF mode and PCO-Edge 4.2 BI sCMOS camera (PCO, Germany), driven by NIS Elements software (Nikon) and equipped with 440 nm, 514 nm, and 561 nm laser lines. Time lapse imaging was performed at 2 min intervals for 20 min, with 100 nM rapamycin addition after the fifth time point. All live cell imaging was conducted at 37 °C, 5% CO₂ and 90% humidity with a stage top incubation system (Tokai Hit). Vitamin and phenol red-free media (US Biological) supplemented with 2% FBS were used in imaging to reduce background and photobleaching. Adequate co-expression of all relevant plasmids was confirmed by fluorescent imaging at appropriate wavelengths. To minimize variability due to relative expression levels, only cells showing at least 30% increase in mCherry intensity after addition of rapamycin were considered for quantification. All image processing and analyses for co-recruitment assays were performed using Metamorph (Molecular Devices, Sunnyvale, CA, USA) and FIJI software (NIH, Bethesda, MD, USA). Co-recruitment assay graphs show means ±95% confidence interval with n≥40 different cells pooled from at least three independent experiments. For information on plasmids used in these assays, see the Key Resources Table.

Raji cells were attached to glass-bottom microwell culture dishes (IBIDI) using poly-L-lysine (20 µg/mL). Raji cells were then labeled with 10 µM CMAC (7-amino-4-chloromethylcoumarin, Molecular Probes), pulsed with 1 µg/ml SEE (Staphylococcus enterotoxin E, Toxin Technologies), and mixed with Jurkat T cells (clone JE6.1) (ATCC). To promote synaptic conjugate formation, cell-containing dishes were centrifuged at low speed (200xg, 30 seg) and incubated 5 min at 37 °C. For recombinant protein expression, exponentially growing Jurkat T cells were electroporated with 20–30 µg of plasmid as previously reported (Herranz et al., 2019), and Raji cells were added as above 40–48 hr post-electroporation. Cell conjugates were then fixed, first with PBS +2% PFA (10 min, RT), then with cold acetone (10 min at –20 °C). Immunofluorescence staining was done as previously described (Herranz et al., 2019). Imaging was performed using a Nikon Eclipse TiE microscope equipped with a DS-Qi1MC digital camera, a PlanApo VC 60 x NA 1.4 objective, and NIS-AR software (all from Nikon). Epifluorescence images were then deconvolved with Huygens Deconvolution Software from Scientific Volume Image (SVI), using the ‘widefield’ optical option.

Swiss PDB Viewer was used to visualize the crystallographic structure of INPP5E aa 282–623 (Protein Data Bank (PDB) ID: 2xsw) in order to identify candidate CLS or ciliopathy residues near the FDRxLYL motif (Tresaugues et al., 2010). Final figures were rendered using the PyMOL Molecular Graphics System (Version 2.0 Schrödinger), from either the PDB structure or the full length AlphaFold model (AF-Q9NRR6-F1) (Jumper et al., 2021).

For quantitation of Western blot levels, including co-IP experiments, Fiji/Image J was used. Using 8-bit blot images, average pixel intensity (API) and surface area of a rectangular region of interest (ROI) containing the protein band or background region was obtained using the Measure function. Data were then transferred to an Excel file, where API values were inverted (so that blacker in blot means higher intensity: API’=255 API), background-subtracted, and the resulting specific intensity multiplied by the corresponding ROI’s surface area to get the band’s total specific intensity. Ratios were then calculated for each sample as appropriate (e.g. EGFP-INPP5E construct levels normalized to tubulin, or co-IPed protein normalized to IPed protein, and to its own levels in lysates). Specific details for each experiment are described in figure legends. Quantitation of intensities along cilia length in the methanol immunofluorescence stainings were obtained using Fiji/Image J’s Plot Profile function. For graphs and statistical analyses, GraphPad Prism 9 software was used. Details of statistical analyses are found in the corresponding figure legends.