Arf-like Protein 3 (ARL3) Regulates Protein Trafficking and Ciliogenesis in Mouse Photoreceptors*

Arf-like protein 3 (ARL3) is a ubiquitous small GTPase expressed in ciliated cells of plants and animals. Germline deletion of Arl3 in mice causes multiorgan ciliopathy reminiscent of Bardet-Biedl or Joubert syndromes. As photoreceptors are elegantly compartmentalized and have cilia, we probed the function of ARL3 (ADP-ribosylation factor (Arf)-like 3 protein) by generating rod photoreceptor-specific (prefix rod) and retina-specific (prefix ret) Arl3 deletions. In predegenerate rodArl3−/− mice, lipidated phototransduction proteins showed trafficking deficiencies, consistent with the role of ARL3 as a cargo displacement factor for lipid-binding proteins. By contrast, retArl3−/− rods and cones expressing Cre recombinase during embryonic development formed neither connecting cilia nor outer segments and degenerated rapidly. Absence of cilia infers participation of ARL3 in ciliogenesis and axoneme formation. Ciliogenesis was rescued, and degeneration was reversed in part by subretinal injection of adeno-associated virus particles expressing ARL3-EGFP. The conditional knock-out phenotypes permitted identification of two ARL3 functions, both in the GTP-bound form as follows: one as a regulator of intraflagellar transport participating in photoreceptor ciliogenesis and the other as a cargo displacement factor transporting lipidated protein to the outer segment. Surprisingly, a farnesylated inositol polyphosphate phosphatase only trafficked from the endoplasmic reticulum to the Golgi, thereby excluding it from a role in photoreceptor cilia physiology.


Arf-like protein 3 (ARL3) is a ubiquitous small GTPase expressed in ciliated cells of plants and animals. Germline deletion of Arl3 in mice causes multiorgan ciliopathy reminiscent of
Bardet-Biedl or Joubert syndromes. As photoreceptors are elegantly compartmentalized and have cilia, we probed the function of ARL3 (ADP-ribosylation factor (Arf)-like 3 protein) by generating rod photoreceptor-specific (prefix rod ) and retinaspecific (prefix ret ) Arl3 deletions. In predegenerate rod Arl3 ؊/؊ mice, lipidated phototransduction proteins showed trafficking deficiencies, consistent with the role of ARL3 as a cargo displacement factor for lipid-binding proteins. By contrast, ret Arl3 ؊/؊ rods and cones expressing Cre recombinase during embryonic development formed neither connecting cilia nor outer segments and degenerated rapidly. Absence of cilia infers participation of ARL3 in ciliogenesis and axoneme formation. Ciliogenesis was rescued, and degeneration was reversed in part by subretinal injection of adeno-associated virus particles expressing ARL3-EGFP. The conditional knock-out phenotypes permitted identification of two ARL3 functions, both in the GTP-bound form as follows: one as a regulator of intraflagellar transport participating in photoreceptor ciliogenesis and the other as a cargo displacement factor transporting lipidated protein to the outer segment. Surprisingly, a farnesylated inositol polyphosphate phosphatase only trafficked from the endoplasmic reticulum to the Golgi, thereby excluding it from a role in photoreceptor cilia physiology.
Replacement of entire outer segments every 10 days in mice necessitates efficient trafficking of membrane-associated proteins (19,20). Lipidated proteins associate transiently with the endoplasmic reticulum (ER) where post-translational processing occurs (21). Trafficking to the outer segment by diffusion requires solubilization factors that interact with lipid side chains, e.g. PDE␦ (also known as PrBP/␦ or PDE6D, a prenylbinding protein originally thought to be a subunit of PDE6) (22,23) and UNC119 paralogs (UNC119a and UNC119b, where UNC119 is uncoordinated 119, a human Caenorhabditis elegans homolog) (24). PDE␦ is a prenyl-binding protein that interacts with C-terminal farnesyl and geranylgeranyl lipids, whereas UNC119 is an acyl-binding protein specific for N-terminal C-12 and C-14 fatty acids. PDE6D null mutations in human are associated with Joubert syndrome (25), caused by impaired ciliary targeting of INPP5E (inositol polyphosphate-5-phosphatase E), an enzyme thought to mediate ciliary stabi-lization (26). Deletion of Pde6d in mice produced retinal degeneration caused by trafficking defects of GRK1 and PDE6 (22). A missense mutation in UNC119 is associated with cone-rod dystrophy (27).
We generated rod-specific and retina-specific Arl3 knockouts to study defects in photoreceptor protein trafficking. In rod Arl3 knock-outs, Cre recombinase, an enzyme carrying out site-specific recombination, is expressed post-ciliogenesis allowing formation of outer segments, whereas in ret Arl3 knock-outs, Cre is expressed during embryonic development. The results show that ER to OS trafficking of lipidated OS proteins in rod Arl3 Ϫ/Ϫ retina was impaired, leading to outer segment shortening and slow retinal degeneration, consistent with loss of CDF function. The ret Arl3 Ϫ/Ϫ photoreceptors formed neither connecting cilia nor outer segment membrane, leading to protein accumulation in the inner segment and rapid degeneration. INPP5E, which localizes exclusively to the Golgi apparatus of WT photoreceptors, was significantly reduced in ret Arl3 Ϫ/Ϫ inner segments suggesting that trafficking of INPP5E from ER-to-Golgi is ARL3-dependent.

Experimental Procedures
Animals-All procedures were approved by the University of Utah Institutional Animal Care and Use Committee and were conducted in compliance with the Guide for Care and Use of Laboratory Animals from National Institutes of Health. Mice were maintained in a 12:12-h dark/light cycle. Flp mice were used to invert the gene trap at FRT sites. Six3-Cre and iCre75 transgenic mice were used to generate retina-and rod-specific Arl3 knock-outs ( Fig. 1) (47)(48)(49). A transgenic mouse expressing the EGFP-CETN2 fusion protein (JAX stock no. 008234) was used to identify centrioles with fluorescence microscopy (50).
Generation of Arl3 Gene Knock-out Mouse-A mouse embryonic stem cell line containing a gene trap cassette in intron 1 of the Arl3 gene was purchased from the European Mouse Mutant Cell Repository (EUCOMM, Helmholtz Zentrum München, Germany). The gene trap was flanked by antisense FRT and loxP sites facilitating trap inversion (51,52) by Flp-and Creinduced recombination (Fig. 1, B-D). The presence of the gene trap and the integrity of short and long arms were verified by PCR. Chimeric mice were generated by the transgenic mouse core facility, University of Utah. The WT allele was genotyped by PCR using primer pair ARL3-F1 (5Ј-AGACCACGTGCTC-TTCCATC) and ARL3-R3 (5Ј-GTGGTTATGTGTTGTCAT-GAG), yielding a 367-bp amplicon. The presence of the gene trap was verified using primer pair ARL3-F1 and RP2-R5 (5Ј-CTAGACAATCGGACAGACAC), yielding a 1080-bp amplicon. Heterozygous Arl3 germline knock-out mice (Arl3 GT/ϩ ) were outbred to Flp mice to invert the gene trap at FRT sites. Flp-FRT recombination (53) was verified using the primer pair RP2-R5 and Eu-R2 (5Ј-CTTGAACCTCCTCGTTCGAC), yielding an ϳ500-bp amplicon. Mice carrying the inverted gene trap (floxed mice) were mated with transgenic mice expressing Cre to generate conditional knock-outs. Cre-loxP recombination was checked with a Cre-specific primer set. The iCre75 transgene was identified using the primer pair iCre75-F (5Ј-GGATGCCACCTCTGATGAAG) and iCre75-R (5Ј-CACAC-CATTCTTTCTGACCCG). Six3-Cre mice were genotyped using the primer pair Six3Cre159 (5Ј-TCGATGCAACGAGT-GATGAG) and Six3Cre160 (5Ј-TTCGGCTATACGTAACA-GGG). Absence of the rd8 mutation was confirmed by PCR (54).
Electroretinography (ERG)-ERGs were measured as described (41) with minor modifications. Single-flash scotopic responses were recorded at stimulus intensities of Ϫ40 db (Ϫ3.4 log cds⅐m Ϫ2 ) to 20 db (2.39 log cds⅐m Ϫ2 ). For photopic ERGs, the mice were light-adapted under background light of 1.48 log cds⅐m Ϫ2 for 5-10 min. Single-flash responses were recorded at stimulus intensities of Ϫ10 db (Ϫ0.6 log cds⅐m Ϫ2 ) to 20 dB (2.39 log cds⅐m Ϫ2 ).
OptoMotry (OKT)-Optomotor reflex-based tests were performed using an OptoMotry system (Cerebral Mechanics). Rotation speed (12°/s) and contrast were kept constant as described (57). The mice were adapted to room light (150 -250 lux), and OptoMotry was performed under light conditions to test cone-mediated vision.
Optical Coherence Tomography (OCT)-Mice were anesthetized initially by inhalation of 3% isoflurane/O 2 mixture (flow rate of 1.0 liter/min) in a closed canister, and anesthesia was maintained with 1.5% isoflurane/O 2 mixture at the same flow rate. After each mouse was transferred to a heated water pad (37°C), pupils were dilated with a 1% tropicamide solution, and corneas were hydrated periodically with 1ϫ PBS, pH 7.4, to prevent corneal desiccation. Images were acquired from both eyes using a 30°lens and Heidelberg Retina Angiograph-optical coherence tomography (Spectralis Heidelberg, Germany) with a 488-nm argon blue laser and standard 500-nm long-pass filter. The mice were transferred to a recovery heat pad maintained at 37°C and monitored until fully conscious.
EGFP-INPP5E Construct and Neonatal Electroporation-The mouse Inpp5e gene was amplified from the cDNA prepared from mouse retina using the primer pair of 5Ј-ATGCC-ATCCAAGTCAGCTTGC (forward) and 5Ј-CATCCTTTGC-AGTGACCTTGG (reverse). Inpp5e cDNA was cloned into a pEGFP-C1 vector (Clontech) that was amplified with the primer pair of 5Ј-GCCAAGGTCACTGCAAAGGATGCAGC-CATACCACATTTGTAGAG (forward) and 5Ј-CAGGCAA-GCTGACTTGGATGGCATCTTGTACAGCTCGTCCATG (reverse). The Inpp5e cDNA was ligated to the vector DNA using a NeBuilder kit (New England Biolabs). INPP5E was fused in-frame to the C terminus of EGFP and verified by sequencing. The EGFP-INPP5E-expressing plasmid was introduced into photoreceptors of neonatal mice using in vivo electroporation (41), and postnatal day 15 (P15) mouse eyes were harvested for immunohistochemistry.
Generation of ARL3-EGFP Shuttle Vector and AAV (scAAV-ARL3-EGFP)-Recombinant self-complementary AAV2/8 (scAAV2/8) was used to express ARL3-EGFP under the control of the Grk1 promoter. Mouse Arl3 was cloned into a scAAV shuttle vector, pscAAV-CAG-EGFP (provided by W. W. Hauswirth), to generate scAAV2/8 packaging construct scAAV-ARL3-EGFP. To make a self-complementary AAV vector, the left ITR (ITR near the promoter region) was modified to eliminate the terminal resolution site and AAV D sequence. The resultant plasmid was named as scAAV-ARL3-EGFP. Triple-plasmid transfection into HEK293 cells was used to produce the AAV vector (58). The self-complementary ARL3-EGFP construct was packaged into AAV2/8 capsid. The vector was purified by polyethylene glycol precipitation followed by cesium chloride density gradient fractionation (58). The purified vector was maintained in solution containing 10 mM Tris-HCl, 180 mM NaCl, pH 7.4. The virus was titered by real time PCR using the following primers and fluorescent labeled probes: forward primer 5Ј-GCACCTTCTTGCCACTC-CTA-3Ј and reverse primer 5Ј-GACACAGCACCAGGCTAAA-TCC-3Ј; probe 5Ј-6-carboxyfluorescein-CGTCCTCCGTGA-CCCCGGC-tetramethylrhodamine-3Ј. Linearized plasmid DNA was used as standard to quantify the virus. In P15 WT and mutant mice, 0.5 l of 3.9⅐10 12 ml Ϫ1 scAAV-ARL3-EGFP was injected subretinally, and the eyes were harvested at 2 months of age.
Statistics-SigmaPlot12 was used for statistical analysis using Student's t test, and the level of statistical significance was set to p ϭ 0.05.

Generation of Conditional Arl3
Knock-out Mice-Embryonic stem cells containing a ␤-GEO gene trap (52) in intron 1 of the mouse Arl3 gene were used to generate chimeric mice (Fig. 1A). Before injection into blastocysts, position of the gene trap in intron 1 and presence of inverse FRT and loxP sites flanking the trap were confirmed by PCR and sequencing (Fig. 1B). The gene trap contains a splice-acceptor sequence upstream of the promoterless ␤-Geo reporter gene that essentially converts the gene trap into an exon. The trap is predicted to truncate ARL3 after exon 1, which ends after the start codon ATG, thereby preventing expression of ARL3. Arl3 GT/ϩ parents with verified germline transmission did not produce live Arl3 GT/GT mice confirming embryonic or early postnatal embryonic lethality of Arl3 Ϫ/Ϫ mice (46). In several Arl3 GT/ϩ matings, only 2 Arl3 GT/GT mice (of 150) were born but did not survive (results not shown).
Arl3 GT/ϩ mice were mated with transgenic mice expressing flippase recombinase (flp), causing inversion of the gene trap (Fig. 1C). Inversion of the trap renders the gene trap inactive (no effect on ARL3 expression). Cre-loxP recombination was initiated with two transgenic lines, iCre75 mice expressing Cre recombinase (Cre) in rods (49,56) or Six3-Cre mice expressing Cre in retinal progenitors (47). Cre-loxP recombination with iCre75 at postnatal day 7 (P7) and later inverted the trap only in rods to generate conditional rod knock-outs termed rod Arl3 ϩ/Ϫ , whereas Cre-loxP recombination with Six3-Cre inverted the trap in all retinal progenitors generating conditional retina knock-outs termed ret Arl3 ϩ/Ϫ (Fig. 1D). Genotyping of offspring (Fig. 1E) and Western blot results (Fig. 1F) confirmed the generation of each conditional knock-out line. The rd8 mutation of the Crb1 gene (54) was detected in the F1 mutant mouse line but was subsequently eliminated by successive matings with rd8-free C57BL6/J mice.
Immunohistochemistry with available ARL3 antibodies was unsuccessful. To show the distribution of ARL3 unambiguously, we generated a self-complementary adeno-associated virus (scAAV2/8) expressing the fusion protein, ARL3-EGFP. Subretinal injection of scAAV-ARL3-EGFP revealed localization of ARL3, presumably in the GDP-bound form, throughout the outer nuclear layer (ONL), inner segment (IS), and connecting cilium (CC) but not the outer segment (Fig. 1G). ARL3 is likely in the GDP (inactive) form in the inner segments because its activating protein (GAP), RP2, is present in the inner segment cell membrane (41) and its GEF (which activates GDP/ GTP exchange) is located at the connecting cilium and outer segment. ARL3-GTP is needed only at the ciliary ridge to unload cargo (see below). Enlargement of the IS/OS junction (Fig. 1G, right panel) reveals that ARL3 is enriched in the CC/BB area. rod Arl3 Ϫ/Ϫ Photoreceptors Degenerate with RP-like Phenotype-During retina development, the mother centriole docks to the cortex of the rod IS around P5-6 to engage in ciliogenesis and axoneme formation (59). We initiated rod deletion of ARL3 by crossing Arl3 flox/flox mice with iCre75 transgenic mice expressing Cre at P7 and later (49). Scotopic ERG recordings of P15 rod Arl3 Ϫ/Ϫ mice at various intensities ( Fig. 2A) revealed nearly normal responses at P15 but reduced rod a-and b-wave amplitudes at 1 month of age. The presence of rod responses demonstrates that postnatal development, including ciliogenesis, occurred normally. The rod Arl3 Ϫ/Ϫ photopic responses (Fig. 2B) were robust and comparable with WT at P15 but started to decline (15-20%) at 1 month of age, a phenotype resembling RP. Visual function was further tested by Opto-Motry, a technique that enables the rapid screening of functional vision using the OKT response. At 1 month of age, rod Arl3 Ϫ/Ϫ OKT was normal but extinct at 2 months (Fig. 2C). The normal response at 1 month is most likely due to survival of cones (Fig. 2D, right panel). At 2 months of age, rods and cones were degenerated (Fig. 2, D and F), and the OKT response was extinguished. By optical coherence tomography (OCT), rod Arl3 Ϫ/Ϫ retina thickness declined at P15 (Fig. 2, E and F) and approached 160 m at P50 (Fig. 2F). Heterozygous rod Arl3 Ϫ/Ϫ retina thickness (ϳ260 m) was identical to WT suggesting haplosufficiency (Fig. 2F). In mice expressing iCre75 on the WT background, retina thickness was unaffected over the first 60 postnatal days (56).
Retina-specific Deletion of ARL3-Six3-Cre-mediated recombination is expected to disable ARL3 expression as early as embryonic day 10 (47), before photoreceptor differentiation and ciliogenesis. Unexpectedly, ret Arl3 Ϫ/Ϫ scotopic and photopic ERG traces were recordable at P15 and 1 month (Fig. 4, A  and B). The scotopic a-waves were suppressed at P15 and highly reduced at 1 month using flash intensities of Ϫ1.63 log cds m Ϫ2 and higher (Fig. 4A, right panel). Photopic b-waves were significantly reduced in P15 and 1-month-old animals to about 35% of normal ERG traces (Fig. 4B, right panel). The ret Arl3 Ϫ/Ϫ OKT response was 25% reduced at 1 month and nearly extinguished at 2 months (Fig. 4C). Significant visual activity in ret Arl3 Ϫ/Ϫ animals at 1 month may be due to survival mutant cones in the peripheral retina (Fig. 4D, right panel). ML-opsinlabeled cone outer segments were clearly present in the ret Arl3 Ϫ/Ϫ retina periphery, whereas OS were absent in the central ret Arl3 Ϫ/Ϫ retina. At 2 months, the ret Arl3 Ϫ/Ϫ retina was degenerated and unable to respond to light (Fig. 4D), a phenotype resembling Leber congenital amaurosis (61).
Six3-Cre-mediated recombination is known to be delayed in the retina periphery (60,62). We therefore tested CC/OS formation by rhodopsin immunohistochemistry in the presence of EGFP-CETN2 at P10 and P15, the times when ERG and OKT showed significant visual responses. The results show normal OS in the peripheral ret Arl3 Ϫ/Ϫ retina, comparable with heterozygous controls, although degeneration in the central retina was far advanced (Fig. 5).
ret Arl3 Ϫ/Ϫ Rod and Cone Photoreceptors Do Not Form Outer Segments-OCT of ret Arl3 Ϫ/Ϫ animals showed a decrease in retina thickness to 200 m as early as P15, further decreasing to ϳ160 m at P30 (Fig. 6, A and B). ONL thickness is slightly reduced at P10, and active degeneration ensues at P15 (Fig. 6C). In control P10 and P15 ret Arl3 ϩ/Ϫ retinas, connecting cilia formed as documented with the centriole/CC marker EGFP-CETN2 (Fig. 6, D and E, left panels). By contrast, in the P10 and P15 ret Arl3 Ϫ/Ϫ retinas, only mother and daughter centrioles were present, and functional basal bodies, connecting cilia and axonemes, were not formed (Fig. 6, D and E, right panels). Rhodopsin mislocalized throughout the ONL and IS (Fig. 6, F and G) similarly as observed in P10 ret Kif3a Ϫ/Ϫ retina lacking connecting cilia and outer segments (60). Trafficking of lipidated proteins (PDE6) (Fig. 6, H and I), rod T␣ (Fig. 6, J and K), and GRK1 (data not shown) was also impaired as outer segment disks are absent. Pigments M/L-and S-opsin distributed throughout the ret Arl3 Ϫ/Ϫ cones (Fig. 6, L-O). At P15, rod and cone photoreceptors degenerate massively (Fig. 6, G, I, K, M,  and O, right panels). In contrast to phototransduction proteins, the ARL3 GEF ARL13b, which has an N-terminal palmitoylation motif and in C. elegans is palmitoylated (63), did not mislocalize in the ret Arl3 Ϫ/Ϫ ONL (Fig. 6, P and Q). ARL13b localization in the mutant retina is restricted to the apical IS and is presumably membrane-associated. The absence of outer segments as early as P10 in central ret Arl3 Ϫ/Ϫ photoreceptors suggests that ARL3, a known microtubule-interacting protein, may regulate a late stage of ciliogenesis or intraflagellar transport. ret Arl3 Ϫ/Ϫ Versus rod Arl3 Ϫ/Ϫ Photoreceptor Degeneration-At P15, photoreceptor degeneration in ret Arl3 Ϫ/Ϫ mice proceeds faster than in rod Arl3 Ϫ/Ϫ mutants due to embryonic expression of Cre and lack of outer segments in the central retina (compare Figs. 3, I-L, and 6, E-Q). At 1 month, ret Arl3 Ϫ/Ϫ and rod Arl3 Ϫ/Ϫ photoreceptors degenerate rapidly with profound accumulation of PDE6, GRK1, and rod T␣ in the remaining inner segments and ONL (Fig. 7, A-C). One-monthold ret Arl3 Ϫ/Ϫ and rod Arl3 Ϫ/Ϫ retinas reveal only 3-4 or 5-6 rows of ONL nuclei (Fig. 7, A-C, and E), respectively, and at 2 months the mutant retinas are both nearly completely degenerated (Fig. 7F).

Rescue of Ciliogenesis and Protein Trafficking in ret Arl3
Ϫ/Ϫ Photoreceptors-AAV particles expressing ARL3-EGFP were injected into the subretinal space of P15 WT and ret Arl3 Ϫ/Ϫ mice. Retinas harvested at 2 months post-injection (Fig. 8,  A-D) show significant rescue of the ONL layer (4 -5 rows of nuclei in treated versus 1 row in untreated retinas) (Fig. 8, A-D,  middle and right panels, E). Moreover, treated central ret Arl3 Ϫ/Ϫ photoreceptors developed inner and outer segments, demonstrating rescue of ciliogenesis. Although outer segments were shorter in treated retinas, outer segment proteins trafficked normally and did not mislocalize (Fig. 8, A-D,  middle panels).
Trafficking of INPP5E to the Golgi-INPP5E is present in the cilia of hTert-RPE cells, cilia of mouse kidneys and cerebellum (64), and axonemes of mouse embryonic fibroblasts (26). INPP5E is farnesylated and was suggested to travel to the OS in a PDE␦-dependent manner (25). In WT (data not shown) and ret Arl3 ϩ/Ϫ mouse rods, however, INPP5E locates to the Golgi of inner segments and is absent in the outer segment (Fig. 9, A and  E). At P10 when OS are forming, INPP5E is present in the inner segment proximal to centrioles, and rhodopsin localizes exclusively to the OS (Fig. 9, A and B, and enlargements). At P15 when OS are more developed, separation of rhodopsin and INPP5E fluorescence is most evident (Fig. 9, E and F). The enlargements reveal that centrioles and connecting cilia, identified by EGFP-CETN2, do not colocalize with INPP5E (Fig. 9E,  bottom panel). In ret Arl3 Ϫ/Ϫ animals at P10 (Fig. 9, C and D) and P15 (Fig. 9, G and H), rhodopsin mislocalizes to the inner segments as outer segments are absent. INPP5E fluorescence of the ret Arl3 Ϫ/Ϫ IS appears significantly reduced (Fig. 9, C and G) compared with ret Arl3 ϩ/Ϫ IS (Fig. 9, A and E), suggesting that INPP5E trafficking from ER-to-Golgi may in part be ARL3/ PDE␦-dependent.
We generated an EGFP-INPP5E expression construct to avoid misinterpretation due to antibody artifacts. Neonatal electroporation of the construct and colabeling with a Golgi marker, anti-giantin, demonstrates that INPP5E distributes to the ER surrounding the nucleus and Golgi apparatus (Fig. 9, I and J), and it is excluded from WT and mutant outer segments ( Fig. 9, M and N). In ret Arl3 Ϫ/Ϫ retina (Fig. 9, K and L), antigiantin-labeled Golgi is retained in the degenerating ONL, and rhodopsin, unable to traffic, colocalizes with INPP5E (Fig. 9, O  and P). The results show that farnesylated INPP5E traffics to the Golgi and not the outer segment of both WT and mutant photoreceptors.

Discussion
ARL3 is a small GTPase colocalizing with microtubules in HeLa cells and brain (40). The best-known function of ARL3, in the GTP-bound form, is that of a cargo displacement factor that evicts lipidated cargo from its lipid-binding protein (37,38,65). ARL3-dependent lipid-binding proteins are PDE␦ and UNC119 featuring immunoglobulin-like ␤-sandwich structures that can accommodate lipid side chains with high affinity. However, germline Arl3 Ϫ/Ϫ mice exhibited abnormal development of renal, hepatic, and pancreatic epithelial tubule structures, implicating ARL3 in ciliogenesis (46) and syndromic ciliopathy. To explore the role of ARL3 in photoreceptors, we generated rod-and retina-specific Arl3 knockouts. An advantage of the rod-specific knock-out is that Cre recombinase is expressed relatively late (ϾP7), after ciliogenesis has occurred, allowing rod Arl3 Ϫ/Ϫ rod outer segments to form normally. We observed significant mistrafficking of lipidated proteins as early as P15 (Fig. 3, H-L, right panels), signaling the onset of an RP-like photoreceptor degeneration. Rhodopsin and other transmembrane proteins are unaffected by loss of ARL3. In the WT scenario, lipidated cargo, docked to the ER membrane post-biosynthesis, is extracted by its lipid-binding protein (PDE␦ or UNC119) and forms a soluble complex (Fig. 10A). The complex interacts with its CDF (ARL3-GTP), expelling cargo from its binding site and delivering it to the destination membrane. The phenotype of the rod-specific Arl3 knockout is consistent with loss of CDF function (Fig. 10B). In the absence of ARL3, lipidated cargo is retained and not delivered to the destination membrane, leading to accumulation of lipidated proteins in the IS and ONL (Figs. 3, H-L, and 10B). Non-delivery of PDE6 to the outer segment poses significant problems by upsetting the PDE6/guanylate cyclase cGMP regulatory system and shifting free cGMP to higher levels. As a result, rods degenerate with an RP-like phenotype that eventually affects cones as well (66).
Homozygous RP2 null alleles in humans cause severe X-linked RP (XLGRP), whereas Rp2 deletion in mice promotes a slowly progressing retinal degeneration (41,67). We proposed that an abundance of dominantly active ARL3-GTP, stabilized by negligible intrinsic GTPase activity, attenuates the ability of PDE␦ to interact with prenylated cargo (41). Interaction of ARL3-GTP with its PDE␦ complex constricts the lipid-binding site and impairs the binding of cargo (alternatively, for UNC119, the binding pocket widens). Failure to extract cargo from the ER steadily lowers the concentration of lipidated proteins in the OS, initiating a slowly progressing photoreceptor degeneration. The rod Arl3 Ϫ/Ϫ trafficking defects are distinct from the Rp2 Ϫ/Ϫ mouse defects where ARL3 is locked in the GTP-bound form thereby impeding extraction of PDE6 and GRK1 from the ER and leading to the degradation of the lipidated proteins (41). Over- expression of dominantly active ARL3 in C. elegans and Leishmania also causes ciliogenesis defects (30,68), but mistrafficking of lipidated proteins has not been invoked as causal.
The severe ciliopathy phenotype of Arl3 germline knockouts prompted us to investigate the consequence of ARL3 deletion during embryonic and early postnatal development, before the onset of ciliogenesis. Using transgenic EGFP-CETN2 as centriole and connecting cilium marker, we observed that connecting cilia (transition zones) and axonemes were undetectable in the central P10 ret Arl3 Ϫ/Ϫ retina, and the outer segments never form (Figs. 5 and 6, D and E). As an immediate consequence, transmembrane and peripheral membrane outer segment polypeptides accumulated in the inner segment (Fig. 6,  D-Q). Although severe retinal degeneration was obvious at P10, ciliogenesis in central ret Arl3 Ϫ/Ϫ photoreceptors could be ameliorated by viral expression of ARL3 tagged with EGFP (Fig. 8).
Although rescue of degeneration was only partial, inner segment and shortened outer segments were present in treated retinas demonstrating rescue of ciliogenesis and protein trafficking. Compartmentalization of ARL3 in the IS/basal body area and of its GEF, ARL13b, in the OS suggests that ARL3 must be switched into its GTP-bound active conformation to enable ciliogenesis. Accordingly, mutations in the human ARL13b gene disabling GEF activity on ARL3 and germline knock-out of the mouse Arl13b gene cause Joubert syndrome (44,45).
The ret Arl3 Ϫ/Ϫ phenotype is strikingly similar to a retinaspecific knock-out of KIF3A (obligatory subunit of the heterotrimeric kinesin-2, KIF3) and IFT88 (particle of the IFT-B complex) (60). KIF3 and IFT88 are responsible for IFT of tubulin and axoneme building blocks. In ret Kif3a Ϫ/Ϫ rod and cone photoreceptors, ciliogenesis was disabled as early as P6 in the central retina, with a delay in the peripheral retina. A link between ARL3 and intraflagellar transport was observed in Leishmania and C. elegans. In Leishmania, ARL3 is required for flagella formation (69), and in C. elegans ARL13b and ARL3 regulate IFT through a tubulin deacetylase (HDAC6) pathway (28,68). In Arl3 mutant worms, which have no GEF activity and presumably lack ARL3-GTP, IFTA, and IFTB, particles dissociate and disrupt IFT (28) suggesting that ARL3-GTP may be required for assembly of IFT complexes.
Finally, we investigated the trafficking of INPP5E in photoreceptors, dependence on ARL3/PDE␦, and function as a phosphatidylinositol-5Ј-phosphatase. Mutations in the human INPP5E gene are linked to Joubert and MORM syndromes (26,64), and depletion of INPP5E in the adult mouse by tamoxifen induction causes rapid photoreceptor degeneration (26), suggesting a key role for phosphoinositide metabolism in ciliary maintenance. INPP5E, originally cloned from Golgi membranes, was shown to be predominantly located to the COS-7 Golgi (70). Among ciliated cells, INPP5E colocalized with acetylated ␣-tubulin in hTert-RPE cells, localized to cilia of P10 mouse kidneys and P5 cerebellum (64), and is concentrated in the axoneme of mouse embryonic fibroblasts (26). In IMCD3 cells, targeting to Golgi was proposed to be PDE␦/ARL3-dependent, whereas targeting to cilia was ARL13b-dependent (71). Absence of PDE␦ prevented delivery of INPP5E to mouse embryonic fibroblast cilia, and a PDE6D null allele was linked to Joubert syndrome (25).
INPP5E was absent in mouse photoreceptor outer segments and apical inner segments and instead localized to the proximal inner segment and Golgi (Fig. 9, A-H). This is surprising, as INPP5E localizes along the axoneme of ciliated cells (26). Furthermore, PDE␦-mediated trafficking of INPP5E to the OS was suggested to be impaired in patients carrying a PDE6D null allele (25). Absence of INPP5E in the OS and its localization to the Golgi implies a critical role for INPP5E in the latter organelle. INPP5E levels in the ret Arl3 Ϫ/Ϫ inner segment/Golgi were reduced, suggesting that ER-to-Golgi trafficking of INPP5E may be ARL3/PDE␦-dependent. ARL13b, localizing exclusively to WT and ret Arl3 Ϫ/Ϫ OS (Fig. 6, P and Q), was suggested to target INPP5E to the OS (71). Distinct localizations of ARL13b in the OS and INPP5E in the Golgi discount this putative pathway.
The physiological role of INPP5E in the photoreceptor Golgi is intriguing as the Golgi is an organelle in which phosphatidylinositol 4-phosphate (PI4P) is enriched (72)(73)(74). PI4P could be generated by INPP5E-mediated hydrolysis of 5-phosphate bound to the inositol ring in phosphatidylinositol 4,5-bisphosphate or by 4Ј-phosphorylation of phosphatidylinositol by phosphatidylinositol 4-kinase. Interestingly, phosphatidylinositol 4-kinase knockdown abrogated Golgi to plasma membrane trafficking in mammalian cells (75) suggesting a key role for PI4P in vesicular trafficking. Reduction of INPP5E in the Golgi could deplete PI4P and impair Golgi-to-OS trafficking of membrane proteins, including rhodopsin, a protein that is essential for OS elaboration and vision.
Author Contributions-C. H. G., J. M. F., and H. Z. generated data; Z. W. generated the AAV virus expressing ARL3-EGFP; C. D. G. generated the polyclonal anti-ARL3 antibody; H. Z. and W. B. designed the knock-out strategy; C. H. G. and W. B. wrote the paper.