The small GTPase RAB28 is required for phagocytosis of cone outer segments by the murine retinal pigmented epithelium

RAB28, a member of the RAS oncogene family, is a ubiquitous, farnesylated, small GTPase of unknown function present in photoreceptors and the retinal pigmented epithelium (RPE). Nonsense mutations of the human RAB28 gene cause recessive cone-rod dystrophy 18 (CRD18), characterized by macular hyperpigmentation, progressive loss of visual acuity, RPE atrophy, and severely attenuated cone and rod electroretinography (ERG) responses. In an attempt to elucidate the disease-causing mechanism, we generated Rab28−/− mice by deleting exon 3 and truncating RAB28 after exon 2. We found that Rab28−/− mice recapitulate features of the human dystrophy (i.e. they exhibited reduced cone and rod ERG responses and progressive retina degeneration). Cones of Rab28−/− mice extended their outer segments (OSs) to the RPE apical processes and formed enlarged, balloon-like distal tips before undergoing degeneration. The visual pigment content of WT and Rab28−/− cones was comparable before the onset of degeneration. Cone phagosomes were almost absent in Rab28−/− mice, whereas rod phagosomes displayed normal levels. A protein–protein interaction screen identified several RAB28-interacting proteins, including the prenyl-binding protein phosphodiesterase 6 δ-subunit (PDE6D) and voltage-gated potassium channel subfamily J member 13 (KCNJ13) present in the RPE apical processes. Of note, the loss of PDE6D prevented delivery of RAB28 to OSs. Taken together, these findings reveal that RAB28 is required for shedding and phagocytosis of cone OS discs.

Mammalian photoreceptors renew one-tenth of the OS 3 membrane daily by adding new disc membrane at the OS base and removing senescent components distally through disc shedding and phagocytosis by the adjacent retina pigmented epithelium (1,2). Rod outer segment (ROS) disc shedding and phagocytosis are synchronized by the circadian clock, with the peak of activity shortly after light onset (3). Mammalian cone photoreceptors undergo similar OS shedding and phagocytosis (4,5), but peak activity varies, depending on species (5)(6)(7)(8). Much progress has been achieved in identifying molecular components of the phagocytosis pathway, including the OS-located "eat me" signal phosphatidylserine (PS) (9), RPE membrane receptors ␣V␤5 integrin (10) and receptor tyrosine kinase MerTK (11), and scavenger receptor CD36 (12). The vast majority of data derives from study of rodent rods, as mouse and rat retinas are ϳ97% rod-dominant (13).
Prenylated Rab GTPases are master regulators of membrane vesicular trafficking, and mutations in human RAB genes are associated with multiple human inherited diseases (14). Rab5 isoforms have been shown to be present in RPE phagosomes, where their function is elusive (15). Absence of geranylgeranylation at the Rab27A C terminus, caused by a defect in REP1 (Rab escort protein 1), is associated with choroideremia, a rare X-linked retina degeneration (16,17). RAB28 is a distal member of the Rab family carrying a C-terminal CAAX box motif instead of the common Rab geranylgeranylation motif (18,19) and the first Rab GTPase linked directly to human retina disease (20). Nonsense mutations in Rab28 cause cone-rod dystrophy 18 (CRD18), presenting with foveal hyperpigmentation and atrophy, with diminished cone and rod electroretinography (ERG) responses (20 -22).
Whereas the molecular basis underlying CRD18 has been obscure, our study suggests that RAB28 is an essential factor in cone-specific disc shedding and phagocytosis. In Rab28 germ line knockout mice, which recapitulate the human CRD18 phe-notype, we found that mutant cones elongate and form enlarged tips as early as P14, suggesting a disc shedding impairment. RAB28 traffics to the photoreceptor OS in a complex with PDE6D. Further, RAB28 interacts with KCNJ13, a potassium channel in the RPE associated with Leber congenital amaurosis (LCA16); KCNJ13 may collaborate with RAB28 to facilitate COS phagocytosis.

The murine Rab28 gene and splice variants
The mouse Rab28 gene consists of nine exons and expresses three splice variants, V1-V3. The variants share amino acids 1-191 encoded by exons 1-6 but differ in their C-terminals, encoded by exons 8, 7, and 9, respectively (Fig. 1A). RAB28V1 (221 amino acids) and V2 (220 amino acids) carry a CAAX box sequence, where X ϭ Q predicts farnesylation. RAB28V3 (191 amino acids) has a shorter C-terminal region and is not prenylated. The polyclonal RAB28 antibody used in this study recognized both V1 and V2 recombinant proteins (Fig. 1B) and thus should also recognize V3. Recombinant variants V1 and V2 are present in the HEK293 cell body, and V2 appears to be translocated selectively into the nucleus (Fig. 1C).

Generation of Rab28 gene trap (GT) and knockout mice
Rab28 GT/GT mutant mice carrying a GT cassette in intron 2 and a floxed exon 3 were generated using a EUCOMM cell line. Correct gene targeting was verified by sequencing of PCR-amplified 3Ј and 5Ј recombination arms and FRT and LoxP sites (Fig. 1, D and E). To avoid interference of the GT cassette on nearby gene expression, we generated a Rab28 floxed allele (Rab28 fl/ϩ ) by crossing Rab28 GT/GT with flippase mice; the knockout allele (Rab28 ϩ/Ϫ ) was generated subsequently by mating Rab28 fl/ϩ with CMV-Cre mice. Rab28 Ϫ/Ϫ mice are viable and fertile. RAB28 protein variants of 25-kDa mobility are undetectable in Rab28 Ϫ/Ϫ retina lysates by immunoblotting (Fig. 1F); variant V3 is not expressed at detectable levels in WT samples. In WT retina cryosections, RAB28 is expressed in the RPE and all photoreceptor compartments, most strongly in outer segments (Fig. 1G, left); no variant is detectable in Rab28 Ϫ/Ϫ retina cryosections (Fig. 1G, right). When electroporated into Rab28 Ϫ/Ϫ rods, mCherry-RAB28V2 distributes throughout the cell (Fig. 1H).

Rab28 ؊/؊ mice recapitulate human CRD18
Photoreceptor function was evaluated by scotopic and photopic ERG testing over a period of 13 months. At 3 months of age, both scotopic (rod) and photopic (cone) responses are significantly reduced in Rab28 Ϫ/Ϫ mice, with greater reduction in photopic b-wave amplitudes. Mutant scotopic a-wave amplitudes are reduced to ϳ70% of control (Fig. 2, A and C), whereas the photopic b-wave amplitude is Ͻ50% (Fig. 2, B and D). Retina dystrophy of human CRD18 patients is typically diagnosed in early childhood (20).
Immunolabeling of Rab28 Ϫ/Ϫ retina with antibodies directed against guanylate cyclase 1 (GC1) or mouse cone arrestin revealed fast degeneration of cones, accompanied by progressive ROS shortening and a mild reduction of ONL thick-ness (Fig. 3A). The cone number is slightly reduced (88% of control) at 1 month (Fig. 3, A and F), but severely reduced at 3 months, and very few cones are detectable at 13 months (Fig.  3A, bottom panels). Accordingly, the scotopic a-wave amplitude declines slowly starting at 1 month, and by 13 months, it is 42% of WT controls (Fig. 3B). Photopic ERGs decline much faster and are attenuated at all tested time points, at 65% of control at 1 month and undetectable at 13 months (Fig. 3C). ONL thickness in the central retina is reduced to 87, 78, and 73% of controls by 3, 6, and 13 months of age (Fig. 3D), respectively, whereas mutant retina OS lengths are reduced to 69, 65, and 50% at 3, 6, and 13 months (Fig. 3E). Cone densities are reduced to 88, 71, 35, and 12% of control levels at 1, 3, 6, and 13 months, respectively (Fig. 3F).

Rab28 ؊/؊ cones have elongated OS and enlarged OS tips
All examined rod proteins (rhodopsin, transducin, PDE6, and CNGA1) were correctly localized to the ROS of 1-monthold mutants (Fig. 4A, bottom panels). By contrast, visual pigments ML-opsin and S-opsin mislocalized in part to the IS, ONL, and OPL of mutant cones ( Fig. 4 (C and D), red arrows). We further observed that mutant COS are elongated, stretching toward the apical RPE edge, and frequently display a bulge at the distal tip revealed by CNGA3 and cone opsin immunolabeling ( Fig. 4 (B-D), white arrows). Whereas WT COS taper toward their tips (Fig. 4F), mutant COS exhibit enlarged, balloon-like tips carried by elongated and slender outer segments (Fig. 4, G and H). One-month-old mutant COS (21.7 Ϯ 4.9 m) average 43% longer than control COS (15.1 Ϯ 1.6 m) (Fig. 4I). Slight elongation of mutant COS and occasional COS tip enlargement are first observed shortly after eye opening (P12), which suggests a defect in COS membrane shedding and RPE phagocytosis (Fig. 4E).
A second observation gleaned from immunohistochemistry was that levels of prenylated OS proteins (G protein-coupled receptor kinase 1 (GRK1), cone PDE ␣-subunit (cPDE␣Ј), and cone transducin ␥-subunit (cT␥)) appear reduced in mutant cones ( Fig. 5 (A-C), bottom panels). Reduction of GRK1 levels in mutant cones could be observed as early as P14 (Fig. 5D), whereas GRK1 levels in ROS appear comparable at both P14 and 1 month of age (Fig. 5, A and D). Western blotting using 7-week-old retina lysates showed that Rab28 Ϫ/Ϫ COS protein levels, including cPDE␣Ј and GRK1, are comparable with heterozygous controls (Fig. 5, E and F). The reduction of cT␥ and cPDE␥ was statistically significant (Fig. 5F). Western blotting with 3-week-old retina lysates did not display significant differences in COS protein levels (results not shown).

Phagocytosis defects of Rab28 ؊/؊ cones
To evaluate COS membrane shedding defects, we harvested P24 WT and mutant retinas 1.5 h after light onset and labeled fixed cryosections with combined ML-and S-opsin antibodies. At this time, cone opsin-positive phagosomes (green) are located within the apical RPE microvilli or RPE somata (Fig. 6A, white arrows; cone sheaths identified using peanut agglutinin (PNA; red)). The number of COS phagosomes is reduced in mutant retina where the COS tip enlargement is already obvious (Fig. 6B, arrows). Rab28 Ϫ/Ϫ COS phagosome counts ( Fig.   RAB28 and cone dystrophy 6C) revealed a 5-fold reduction (7.5 Ϯ 2.6 per 2-mm retina) compared with WT controls (39.8 Ϯ 6.0 per 2-mm retina). In contrast, the rod phagosome density is comparable in P24 WT and Rab28 Ϫ/Ϫ retinas (  (Fig. 6F). These results are interpreted to indicate that Rab28 Ϫ/Ϫ COS membrane shedding is defective, whereas ROS phagocytosis proceeds normally.

Accumulation of material in Rab28 ؊/؊ COS tip
Transmission EM of 1-month-old Rab28 Ϫ/Ϫ and heterozygous control retinas documented elongated cones with disor-ganized distal membranes in mutants (Fig. 7, A and B). At the Rab28 Ϫ/Ϫ RPE/photoreceptor interface, we observed RPE microvilli-wrapped structures filled with distorted membrane stacks or vacuolar membranes (Fig. 7, C and D). These structures appear in different shapes (round and oblique) at least 2 m in size (Fig. 7, F and G), indicating buildup of materials at the OS tip. The size of the vacuoles (Fig. 7 (F and G), yellow arrows) and the shapes of membrane stacks (Fig. 7 (F and G), Figure 1. Generation of Rab28 ؊/؊ mice. A, schematic representation of the mouse Rab28 gene and three splice variants, V1-V3. The mouse Rab28 gene consists of nine exons (yellow rectangles). Dashed gray, red, and green lines indicate three types of mRNA splicing. V1, V2, and V3 share exons 1-6 but differ in the C terminus, which is encoded by exons 8, 7, and 9, respectively. aa, amino acids. B, Western blotting of HEK293 lysates expressing mCherry-RAB28V2 (lane 1) and EGFP-RAB28V1 (lane 2). The polyclonal antibody used in this study recognizes both variants (red arrow); the faster moving polypeptide is nonspecific. C, mCherry-RAB28V2 and EGFP-RAB28V1 expression pattern in HEK293 cells. V1 and V2 colocalize in the cytoplasm, and V2 is also present in the nucleus. Top and right small panels, orthogonal views of y and z planes at the lines indicated. Scale bar, 20 m. D, schematic representation of the Rab28 gene, the location of the gene trap in intron 2, the floxed allele, and the knockout allele. The En2SA-IRES-LacZ-pGK-Neo GT cassette is flanked by two FRT sites (green triangle) and is inserted into intron 2 by homologous recombination. Exon 3 is flanked by two LoxP sites (red triangle). Small arrows represent the approximate position of primers for genotyping. E, PCR genotyping of the 3Ј and 5Ј recombination arm, third LoxP, second FRT/LoxP, and first FRT sites in ES cells. D1 and A5, two ES cell clones and the derived mice with identical genotyping results. The right-most panel represents a PCR genotyping result of one litter of Rab28 ϩ/Ϫ ϫ Rab28 ϩ/Ϫ pups. Primer sequences are listed in Table S1.

RAB28 and cone dystrophy
arrowheads) within these bulbous structures also vary. As these structures correspond to the bulges of COS tips seen in Rab28 Ϫ/Ϫ immunostaining (Fig. 4), the variable-size vacuoles and distorted membranes stacks may represent different stages of COS degeneration. Additionally, there are increased RPE pigment deposits in Rab28 Ϫ/Ϫ compared with Rab28 ϩ/Ϫ controls ( Fig. 7 (C and D), dashed arrows). By contrast, ROS disc morphology appears to be normal in Rab28 Ϫ/Ϫ rods (Fig. 7H).

RAB28 interacts with KCNJ13, an RPE microvillus protein
Co-immunoprecipitation (co-IP) using bovine retina lysates followed by MS was performed to identify interacting proteins. We identified dozens of potential interactors (t test, p Ͻ 0.05, RAB28 IP/control IP ratio Ͼ3, Table S2). Top candidates include the prenyl-binding protein PDE6D, which has been reported previously to interact with RAB28 (25). Three other notable candidates are KCNJ13, mutations of which are associated with Leber congenital amaurosis (LCA16) (26); MEF2D, a transcription factor required for photoreceptor development and maintenance (27); and GLUT1 (glucose transporter 1), mediating RdCVF (rod-derived cone viability factor)-induced cone survival (28). In selective Western blotting verification, we successfully verified the interaction of RAB28 with KCNJ13, MEF2D, GLUT1, VAC14, and PDE6D, but failed to validate GRK1 and CAND1 (Fig. 8A), indicating high reliability of our co-IP results.
KCNJ13 is located at the root of WT apical RPE microvilli (29) and overlaps with the cone sheath marker, PNA (Fig. 8B). This pattern is preserved in Rab28 Ϫ/Ϫ (Fig. 8C). Because the possible alteration of KCNJ13 localization on cone-associated microvilli could be masked by the overwhelming signal of rodassociated microvilli, we determined the KCNJ13 localization in Rab28 Ϫ/Ϫ ; Nrl Ϫ/Ϫ mice. KCNJ13 is located at the RPE microvilli that span the entire short OS in Nrl Ϫ/Ϫ control animals (Fig. 8D). However, in Rab28 Ϫ/Ϫ ; Nrl Ϫ/Ϫ double knockout mice, KCNJ13 is partially trapped at the base of apical RPE microvilli (Fig. 8E), suggesting a role of RAB28 in KCNJ13 microvilli trafficking.

Discussion
RAB28 has been suggested to participate in lysosomal delivery of endocytosed surface protein in unicellular trypanosomes (31), angiotensin-induced NF-B nuclear delivery and activation in rat endothelia (32), glucose transporter 4 plasma membrane trafficking in mouse skeletal muscle (33), and intraflagellar transport in Caenorhabditis elegans (34).
In this study, we show that Rab28 germ line knockout mice recapitulate the human CRD18 phenotype. In human CRD patients, the RAB28 gene carries nonsense mutations in exons 5 and 6, truncating all known splice variants (20). CRD may be

RAB28 and cone dystrophy
accompanied by visual acuity deterioration, absence of photopic ERG responses, and myopia as early as age 16. In the Rab28 Ϫ/Ϫ mouse model, photopic b-waves are reduced at 1 month and nearly absent by 1 year of age, which is roughly consistent with human CRD (Figs. 2 and 3).
We found that Rab28 Ϫ/Ϫ COS are elongated to the length of rod outer segments, and the mutant tips are expanded to balloon-like structures and appear unable to shed membranes to be phagocytosed by the RPE. In view of impaired disc shedding at the distal cone outer segment, continued overproduction of new disc membranes at the proximal end may provide a jammed gate impenetrable for prenylated proteins (see the model in Fig. 10). The absence of membrane shedding is supported by dramatic reduction of cone pigment-containing phagosomes (Fig. 6). Bulbous OS tips and lack of phagosome production was also observed in the Nrl Ϫ/Ϫ mouse, a model for enhanced S-cone syndrome (24). Rab28 Ϫ/Ϫ COS elongation and enlarged COS tips were observed shortly after eye opening (Fig. 4). We propose that CRD18 is caused by impaired membrane shedding at the distal COS and/or failed phagocytosis by the RPE. The resulting accumulation is detrimental for cone survival. The molecular identities of components involved in human and mouse disc shedding are currently unclear. Human cone membrane shedding, observed in vivo by adaptive optics

RAB28 and cone dystrophy
optical coherence tomography (8), was characterized by abrupt and transient loss of the COS tip. Average duration of tip loss was ϳ9 min, and average length loss was reported as 2.1 m.

RAB28 and cone dystrophy
lier and is more severe. Another notable observation is that both Rab28 Ϫ/Ϫ and Nrl Ϫ/Ϫ showed COS disorganization at the disc periphery. The Rab28 Ϫ/Ϫ disc edge shows merger/swelling or disassembly (Fig. 7B), whereas the Nrl Ϫ/Ϫ disc edge reveals interconnection (24), suggesting that the COS disc edge is vulnerable to the accumulation of oxidative materials.
Bulbous Rab28 Ϫ/Ϫ OS tips are only observed in mutant cones, as mutant rods are initially unaffected and shed discs exhibit a normal phagosome count (Fig. 6). The reason for this specificity is unknown. ROS phagocytosis is triggered by diurnal surface exposure of the anionic phospholipid phosphatidylserine at the OS tip (9), which is recognized by multiple RPE membrane receptors and their ligands, including integrin ␣V␤5 and ligand MFG-E8 (10, 35), MerTK and ligand proteins S and Gas6 (11, 36 -38), and CD36, which directly binds PS (12). Following substrate-receptor binding, the shed ROS tip is engulfed and ingested by the RPE apical microvillus. MerTK signaling plays a critical role in initiating this process (11). Whether COS phagocytosis is subjected to similar molecular regulation is unknown. Phagocytosis requires that internalized COS phagosomes move along the microvillus before entering the RPE soma (39). However, as both ␣V␤5 integrin and CD36 localize at the RPE soma (10,12), their participation in COS recognition and binding is unlikely. A recent large-cohort genetic study of inherited retina diseases found that MerTK mutations exist in a small percentage of CRD patients (40), suggesting a conserved MerTK-dependent mechanism between ROS and COS phagocytosis. In agreement with this, MerTK is expressed at apical RPE microvilli (38). As RPE signals actively participate in delineating the position of disc shedding (e.g. the localized PS exposure) at ROS tips (9), the failed Rab28 Ϫ/Ϫ shedding phenotype may indicate a communication defect between RPE microvilli and mutant COS tips.
Our protein-protein interaction screen identified the RPE microvilli-resident protein, KCNJ13, as a RAB28 interactant. Recessive mutations of KCNJ13 are associated with LCA16 in humans, with an undefined mechanism (26). Kcnj13 mutant mice generated with the CRISPR-Cas9 system showed mosaic expression of KCNJ13 in RPE cells and photoreceptor degeneration resembling LCA (41). RPE cells null for KCNJ13 can survive, but KCNJ13 expression is required for RPE cells to maintain photoreceptor survival (41). We found that KCNJ13 delivery to RPE microvilli in Rab28 Ϫ/Ϫ ; Nrl Ϫ/Ϫ double knockout is partially reduced (Fig. 8), providing a clue that RPE phagocytosis may be impaired as well. We speculate that absence of RAB28 may alter the KCNJ13 activity, which in turn affects the function of RPE microvilli and cone phagocytosis.
A second important RAB28 interactant is PDE6D, a prenylbinding protein. A Pde6d Ϫ/Ϫ mouse develops recessive CRD in which prenylated OS proteins (cPDE␣Ј, GRK1) are absent in the COS (30). We showed previously that RAB28 trafficking to the ROS is ARL3-GTP-dependent (42). Our finding that farnesylated RAB28 is unable to traffic to the OS in the absence of PDE6D (Fig. 9C) raises the interesting point that the cone-rod dystrophy in PDE6d mutant mice may result partly from the absence of RAB28 in cones. A PDE6D null allele in human patients was associated with Joubert syndrome and shown to impair trafficking of INPP5E, a phosphatidylinositol polyphosphate phosphatase involved in phosphatidylinositol metabolism (43).
The Rab28 Ϫ/Ϫ mouse model provides evidence that CRD18 is caused by a defect in cone disc shedding. Owing to continuous renewal of membranes containing phototransduction components at the proximal OS and the absence of disc shedding, mutant OSs stretch toward the RPE and expand their tips to a balloon-like structure. Cone pigment-containing phagosomes are dramatically reduced. However, because RPE and cones collaborate in phagocytosis and RAB28 is normally expressed in the RPE, a role of the Rab28 Ϫ/Ϫ in defective RPE phagocytosis cannot be excluded.

Animals
Animal procedures were approved by the University of Utah institutional animal care and use committee and were conducted in compliance with the National Institutes of Health Guide for Care and Use of Laboratory Animals.

RAB28 and cone dystrophy Generation of Rab28 ؊/؊ mouse
Rab28 embryonic stem (ES) cells (clones EPD0688_3_D01 and HEPD0677_7_A05), in C57BL/6N genetic background, JM8A3.N1 subline) containing a gene trap in intron 2 were acquired from EUCOMM (Helmholtz Zentrum, Munich, Germany). ES cell blastocyst injection, generation of chimera, and generation/breeding of heterozygous (GT/ϩ) mice were performed at the University of Utah transgenic core. The rd8 mutation (44) was removed by crossing our Rab28 GT/ϩ animals with C57BL/J WT mice. We crossed Rab28 GT/GT with flippase mice (C57BL/6 background) to generate animals with a floxed allele (Rab28 fl/ϩ ) (Fig. 1D). Exon 3 was deleted by crossing Rab28 fl/fl with CMV-Cre mice (C57BL/6 background) to generate Rab28 ϩ/Ϫ and Rab28 Ϫ/Ϫ mice. Mice were maintained under 12-h cyclic dark/light conditions. ES cell and mouse tail genomic DNA was extracted using a standard protocol (45). Routine genotyping was performed using genomic DNA prepared with HotSHOT (46). PCR genotyping primers and other primers (see below) are listed in Table S1.

RNA extraction, RT-PCR, and plasmid construction
Retina tissues were dissected from WT mice, and total RNA was extracted using TRIzol reagent (Invitrogen). Reverse tran-scription was performed using SuperScript II reverse transcriptase (with random primers) following the manufacturer's instructions. Full-length RAB28V1 and RAB28V2 were amplified by PCR and cloned into pEGFP-C1 and pmCherry-C1 (Clontech), respectively. PDE6D-EGFP was kindly provided by Dr. Houbin Zhang (University of Electronic Science and Technology of China, Chengdu). Nonprenylated RAB28V2-C217A was generated using the Phusion site-directed mutagenesis kit (Thermo Fisher Scientific). All plasmids were verified by Sanger sequencing.

ERG
ERG was performed on 1-, 3-, 6-, and 13-month-old Rab28 Ϫ/Ϫ and WT control animals (n ϭ 5 each group) using an UTAS E-3000 universal electrophysiological system (LKC Technologies) as described (47). Briefly, mice were darkadapted overnight, anesthetized by intraperitoneal injection of ketamine (100 g/g body weight) and xylazine (10 g/g body weight) in 0.1 M PBS, and loaded onto a recording platform with body temperature maintained at 37 Ϯ 0.5°C. After dilating pupils with 1% tropicamide solution (Bausch & Lomb Inc., Tampa, FL), ERG responses were recorded from five mice of each genotype per time point. For scotopic ERG, mice were  (Table S2)

RAB28 and cone dystrophy
tested at intensities ranging from Ϫ1.63 to 2.38 log cd s/m 2 . For photopic ERG, a rod saturating background light of 1.3979 log cd s/m 2 was applied for 20 min before and during recording at Ϫ0.01 to 1.86 log cd s/m 2 . Bactitracin ophthalmic ointment (Perrigo, Minneapolis, MN) was routinely applied to the eye to prevent infection after ERG testing, and animals were kept on heating pads until fully recovered before being returned to cages. Peak amplitudes for both a-and b-waves were used for analysis using a one-way ANOVA test.

Measurement of ONL thickness and OS length
Average ONL thickness and OS length were measured based on DAPI and anti-GC1 antibody fluorescence (48). Cone density was measured based on cone arrestin staining, and COS length was measured based on CNGA3 staining.

In situ phagosome counting
Retina sections collected at 1.5 h after light onset were stained with anti-rhodopsin antibody or mixed anti-ML-opsin and S-opsin antibodies. Rhodopsin-positive phagosomes were counted as described (35). Cone opsin-positive phagosomes were counted from entire retina sections.

RAB28 co-immunoprecipitation
Bovine retinas (n ϭ 5) were pooled and homogenized in 50 ml of lysis buffer (50 mM Tris-Cl, pH 7.6, 120 mM NaCl, 0.5% IGEPAL CA-630, 1 mM phenylmethylsulfonyl fluoride, 1ϫ protease inhibitor) using a Dounce homogenizer. Cell debris was removed by centrifugation at 3000 ϫ g, 4°C, for 5 min. The cleaned samples were centrifuged at 20,000 ϫ g, 4°C, for 30 min, and the supernatants were collected. The supernatants were precleared with 500 l of 50% protein G-Sepharose 4 fast flow beads (GE Healthcare) for 30 min. Precleared lysates were divided into 10 aliquots (5 ml each). Half of the aliquots were each incubated with goat anti-RAB28 primary antibody (10 g), and the other half were incubated with normal goat IgG (10 g) for 1 h 45 min at 4°C with gentle agitation, followed by incubation with protein G-Sepharose beads for 45 min. Beads were collected by centrifugation at 2300 ϫ g for 1 min and washed six times in lysis buffer without protease inhibitors and once in 1ϫ PBS. Immunoprecipitated proteins were eluted using an improved soft elution method, which minimally coelutes immunoglobulin (51). The eluted proteins (200 l) were incubated with 1 ml of acetone at Ϫ20°C overnight, precipitated by centrifugation at 10,000 ϫ g for 10 min at 4°C, washed with 75% ethanol, and air-dried. Samples were subjected to quantitative MS.

Quantitative MS
Protein samples were subjected to in-solution tryptic cleavage, as described previously (52). LC-MS/MS analysis was performed using an Ultimate3000 nano-RSLC system (Thermo Scientific) coupled to an Orbitrap Fusion Tribrid mass spectrometer (Thermo Scientific) by a nanospray ion source. Tryptic peptide mixtures were injected automatically and loaded at a flow rate of 30 l/min in 0.1% TFA in HPLC grade water onto a nano trap column (300-m inner diameter ϫ 5-mm precolumn, packed with Acclaim PepMap100 C18, 5 m, 100 Å; Thermo Scientific). After 3 min, peptides were eluted and sep-arated on the analytical column (75-m inner diameter ϫ 25 cm, Acclaim PepMap RSLC C18, 2 m, 100 Å; Thermo Scientific) by a linear gradient from 2 to 30% of buffer B (80% actetonitrile and 0.08% formic acid in HPLC-grade water) in buffer A (2% acetonitrile and 0.1% formic acid in HPLC-grade water) at a flow rate of 300 nl/min over 85 min. Remaining peptides were eluted by a short gradient from 30 to 95% buffer B in 5 min. Analysis of the eluted peptides was performed by an LTQ Fusion mass spectrometer. From the high-resolution MS prescan at a resolution of 120,000 with a mass range of 335-1500, the most intense peptide ions were selected for fragment analysis if they were at least doubly charged. The normalized collision energy for higher-energy collisional dissociation was set to a value of 30, and the resulting fragments were detected in the ion trap. The lock mass option was activated; the background signal with a mass of 445.12003 was used as lock mass (53). Every ion selected for fragmentation was excluded for 20 s by dynamic exclusion. MS/MS data were analyzed using the Max-Quant software (version 1.5.2.8) (54). As digesting enzyme, trypsin/P was selected with maximal 2 missed cleavages. Cysteine carbamidomethylation was set for fixed modifications, and oxidation of methionine and N-terminal acetylation were specified as variable modifications. The data were analyzed by label-free quantification with a minimum ratio count of 2. The first search peptide tolerance was set to 20, the main search peptide tolerance was set to 4.5 ppm, and the requantify option was selected. For peptide and protein identification, the Bos taurus subset of the Uniprot database (Release 2013_09) was used, and contaminants were detected using the MaxQuant contaminant search. A minimum peptide number of 2 and a minimum length of 7 amino acids was tolerated. Unique and razor peptides were used for quantification. The match between run options was enabled with a match time window of 0.7 min and an alignment time window of 20 min. The statistical analysis, including ratio, t test, and significance calculation, was done using the Perseus software (55).

HEK293 cell culture and transfection
HEK293 cells (ATCC) were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 1ϫ penicillin/streptomycin in a standard CO 2 incubator. Cells were transfected with plasmid DNA using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Two days following transfection, cells were fixed and imaged for protein localization, lysed in radioimmune precipitation assay buffer for Western blotting, or lysed in co-IP lysis buffer (see above) for co-IP experiments.

Transmission EM
1-and 2.5-month-old WT and Rab28 Ϫ/Ϫ retinas were fixed 2 h in fixative (2% glutaraldehyde, 1% paraformaldehyde in 0.1 M sodium cacodylate buffer, pH 7.4) at 4°C, post-fixed for 1 h in RAB28 and cone dystrophy 1% osmium tetroxide, and stained en bloc with uranyl acetate. Washed specimens were dehydrated through an ascending series of methanol, dried in propylene oxide, and infiltrated overnight with Epon resin mix/propylene oxide (1:1) mixture, followed by 100% Epon resin for 2 days. Specimens were embedded in plastic, and the plastic was cured by incubation in a 60°C oven for 2 days. Blocks were trimmed, and 1-m-thick sections were cut to identify and orient photoreceptors near the optic nerve. Ultrathin sections at 60 nm were cut, placed onto slot grids with carbon-coated Formvar film (EMS, Hatfield, PA), post-stained with uranyl acetate followed by lead citrate, and finally examined using a JOEL electron microscope at 75 kV.

Statistics
Data are presented as mean Ϯ S.D., where n represents the number of mice analyzed (ERG, retina measurement, phagosome count), or experiment repeats (RAB28 co-IP). Statistical comparisons (significance level set at p Ͻ 0.05) were performed using one-way ANOVA for all experimental data.