Retinyl Ester Homeostasis in the Adipose Differentiation-related Protein-deficient Retina*

The retinal pigmented epithelium (RPE) plays an essential role in vision, including storing and converting retinyl esters of the visual chromophore, 11-cis-retinal. Retinyl ester storage structures (RESTs), specialized lipid droplets within the RPE, take up retinyl esters synthesized in the endoplasmic reticulum. Here we report studies of mice lacking exons 2 and 3 of the gene encoding adipose differentiation-related protein (Adfp), a structural component of RESTs. We found that dark adaptation was slower in AdfpΔ2-3/Δ2-3 than in Adfp+/+ mice and that AdfpΔ2-3/Δ2-3 mice had consistently delayed clearances of all-trans-retinal and all-trans-retinol from rod photoreceptor cells. Two-photon microscopy revealed aberrant trafficking of all-trans-retinyl esters in the RPE of AdfpΔ2-3/Δ2-3 mice, a problem caused by abnormal maintenance of RESTs in the dark-adapted state. Retinyl ester accumulation was also reduced in AdfpΔ2-3/Δ2-3 as compared with Adfp+/+ mice. These observations suggest that Adfp plays a unique role in vision by maintaining proper storage and trafficking of retinoids within the eye.

named after perilipin, adipose differentiation-related protein (Adfp), and tail-interacting protein of 47 kDa (Tip47). PAT domain proteins serve as structural constituents by coating the surface of lipid droplets (4). Although perilipin expression is more limited to adipose and steroidogenic cells (5,6), Adfp is expressed throughout the body, mainly in nonadipogenic tissues such as liver, lung, muscle, testes, and eye (7)(8)(9). Overexpression of Adfp can promote the accumulation of lipid in cultured cells (10,11). Consistent with the proposed lipid storage function of Adfp in nonadipogenic tissues, Adfp-deficient mice have a reduced TAG content in the liver but fail to exhibit a pathological phenotype under normal conditions (12). Currently, the in vivo functions of nonadipogenic lipid droplets are poorly understood, and those of Adfp remain unknown. A recent cell culture study suggests that Adfp plays a role in modulating the turnover rate of lipid esters (13), but the physiological consequences of this effect require further investigation.
In the eye, the retinal pigmented epithelium (RPE) serves as the major storage site for retinyl esters, precursors of the visual chromophore 11-cis-retinal (14). After transfer to the RPE from the circulation or photoreceptors, all-trans-retinol is converted into all-trans-retinyl esters by lecithin-retinol acyltransferase (Lrat) (15,16). These esters are key substrates for Rpe65 that catalyzes their hydrolysis and isomerization to the 11-cis configuration (reviewed in Refs. 15,16). Thus, these esters are essential for generation of the visual chromophore, 11-cis-retinal, that links to opsin-forming rhodopsin and cone visual pigment (14). Retinyl esters are synthesized in the endoplasmic reticulum (ER) of the RPE where Lrat is localized (17) and then are transported to and temporarily stored in retinyl ester storage structures (RESTs or retinosomes), a type of lipid droplet unique to the RPE. Mechanisms by which RESTs located proximal to the plasma membrane take up newly synthesized retinyl esters have yet to be characterized (9). For the RPE to maintain proper retinoid homeostasis in the visual cycle, trafficking and storage of all-trans-retinyl esters must be well balanced to support the rapid turnover of retinyl esters needed for continuous replenishment of visual chromophores. In previous studies we gained some insight into how the quantity and quality of retinyl esters in the RESTs are regulated by enzymes involved in the visual cycle. RESTs overaccumulate in Rpe65 Ϫ/Ϫ mice but are absent in Lrat Ϫ/Ϫ mice (9). Moreover, Adfp specifically localizes to RESTs in the RPE (9) where its expression level correlates with the amount of retinyl esters (9). These observations suggest a specific role for Adfp in retinoid storage and metabolism within the eye. However, they do not explain the mechanisms underlying retinyl ester uptake by the RESTs.
Insight into the roles of REST proteins should increase understanding of overall REST function as well. Here we investigated whether Adfp contributes to the storage and transport of visual cycle retinoids. Our observations not only reveal an essential role for Adfp in the transport and storage of visual retinoids, they also suggest that Adfp promotes the accumulation of retinyl esters by serving as a transient storage site. This Adfp-mediated binding is unusual in that it does not limit the turnover of all-trans-retinyl esters and regeneration of 11-cisretinal, a property required for rapid visual pigment regeneration. Our studies support the hypothesis that Adfp is a novel retinyl ester-binding protein functioning in vitamin A homeostasis of the retina.

EXPERIMENTAL PROCEDURES
Animals-Mice were maintained under complete darkness or in a 12-h light/12-h dark environment. Animal procedures and experiments were approved by the Case Western Reserve University Animal Care Committees and conformed to both the recommendations of the American Veterinary Medical Association Panel on Euthanasia and the Association of Research for Vision and Ophthalmology.
Targeting Constructs and Transgenesis-Adfp ⌬2-3/⌬2-3 mice were generated by standard procedures (Ingenious Targeting, Inc., Rochester, NY). An ϳ11.5-kb region used to construct the targeting vector was cloned from a positively identified BAC clone. The region was designed such that the short homology arm extended ϳ1.5 kb 3Ј to exon 3. The long homology arm started at the 5Ј side of exon 2 and was 7.6 kb long. The neo cassette replaced 2.1 kb of the gene, including exons 2 and 3 that contained the ATG start codon in exon 2. The targeting vector was confirmed by restriction analysis after each modification step and by sequencing with the primers N1 (5Ј-TGCGAGGC-CAGAGGCCAGTTGTGTAGC-3Ј) and N7 (5Ј-ATGTGT-CAGTTTCATAGCCTGAAG-3Ј), designed to read from the selection cassette into the long homology arm and the short homology arm, respectively. T7 (5Ј-ATTTAGGTGAC-ACTATAGAACTC-3Ј) and P6 (5Ј-ATTATGCTGAGT-GATATCCCTCT-3Ј) primers were annealed to the vector sequence and read into the 5Ј-and 3Ј-ends of the BAC subclone. Ten g of the targeting vector was linearized by NotI and transfected by electroporation into 129 SvEviTL1 embryonic stem cells. After selection in medium containing G418 and neomycin, surviving colonies were expanded, and PCR analysis was performed to identify clones that had undergone homologous recombination. Recombinant clones were identified by the 2.0-kb amplified PCR band using primer pairs A1 (5Ј-CATTA-GGGAGTTCGAAGTGTTGG-3Ј) and N1 (5Ј-TGCGAGGC-CAGAGGCCACTTGTGTAGC-3Ј). A1(5Ј-CATTAGGGAG-TTCGAAGTGTTGG-3Ј)/AT2(5Ј-TGGCCTTTGGTACAA-TCAGAGTC-3Ј), and AT1(5Ј-TGGAGAGTCTAGTAGT-TCCTC-3)/N1(5Ј-TGCGAGGCCAGAGGCCACTTGTGT-AGC-3Ј) were used as internal controls with expected size fragments of 1.5 and 1.8 kb, respectively. Correctly targeted embryonic stem cell lines were microinjected into C57BL/6 mouse blastocysts. A total of six male chimeras (90 -100% agouti) and seven female chimeras (100% agouti) were produced from these injections. Chimeric mice were then mated to effect germ line transmission.
Immunoblotting-Mammary gland tissue from Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 female mice with 1-day-old pups was collected and homogenized in 136 mM NaCl, 11.4 mM sodium phosphate, pH 7.4, containing protease inhibitor with a Dounce homogenizer. Protein concentrations were determined by the Bradford assay. Each sample (15 g of protein) was separated by a 10% SDS-PAGE, transferred onto a polyvinylidene difluoride membrane, and probed with rabbit anti-Adfp C-terminal antibody (a kind gift from Dr. McManaman) followed by horseradish peroxidase-conjugated secondary antibody. The signal was visualized by applying SuperSignal West Pico chemiluminescent substrate (Pierce) and subsequent photography. Heart and white adipose tissue were collected and homogenized as above, separated by 10% SDS-polyacrylamide gel, and transferred. Rabbit anti-Tip47 N-terminal antibody (Novus Biologicals, Littleton, CO) was used to probe Tip47 followed by alkaline phosphataseconjugated secondary antibody and developed by incubating with Western Blue stabilized substrate (Promega Corp., Madison, WI). ␤-Tubulin was detected by E7 antibody (Developmental Studies Hybridoma Bank at Iowa University) as internal control.
Histology-Eyecups for light microscopy were fixed in 2% glutaraldehyde, 2% paraformaldehyde for 18 h, infiltrated with 20% sucrose in 0.1 M sodium phosphate, pH 7.4, and then embedded in 33% OCT compound (Miles) diluted with 20% sucrose in 0.1 M sodium phosphate, pH 7.4. Thin sections were cut at 5 m.
Electron Microscopy-Mouse eyeballs were fixed with triple aldehyde/DMSO fixative for 2 h at room temperature. Specimens were thoroughly rinsed in 0.1 M phosphate buffer, pH 7.4, and then postfixed for 2 h in an unbuffered 1:1 mixture of 2% osmium tetroxide and 3% potassium ferricyanide. After rinsing with distilled water, specimens were soaked overnight in an acidified solution of 0.25% uranyl acetate. After another rinse in distilled water, they were dehydrated with increasing concentrations of ethanol, passed through propylene oxide, and embedded in an Epon mixture. Thin sections (80 nm) were cut with an RMC MT 6000-XL ultramicrotome. Sections were sequentially stained with acidified methanolic uranyl acetate and lead tartrate and examined with a JEOL 1200EX electron microscope.
ERG-Full-field ERG recordings, flicker ERGs, and singleflash recordings after intense constant illumination were performed by previously published methods (21).
Retinoid and A2E Analyses-Experimental procedures involving extraction, derivatization, and separation of retinoids from dissected mouse eyes have been described (21).
Two-photon Microscopy-Two-photon excitation microscopy was performed with a Leica TCS SP2 scanning head (Leica) attached to a DM IRBE2 inverted microscope stand. LCS three-dimensional software (Leica) was used for data acquisition. Laser pulses from a mode-locked Ti:Sapphire laser (Chameleon TM -XR, Coherent, Mountain View, CA) were focused on the sample by an HCX PL APO 40ϫ oil immersion objective lens (NA ϭ 1.25, Leica). Autofluorescence from the sample (385-545 nm) was collected by the objective lens, separated from the excitation light by a dichroic mirror, filtered through custom-made filters (HQ 465/160, from Chroma Technology Corp., Rockingham, VT), and directed to a photomultiplier tube detector (R 9624, Hamamatsu). The objective lens was kept at 37°C by an objective lens heater (PeCon, Germany). A temperature-controlled microscope stage (PeCon, Germany) maintained the reaction at 37°C. For ex vivo imag-ing, mouse eyes or eyecups were located at the center of a glassbottomed 35-mm dish (MatTek Corp.) and perfused with oxygenated (95% O 2 , 5% CO 2 ) artificial cerebral spinal fluid (119 mM NaCl, 2.5 mM KCl, 1.3 mM MgCl 2 , 2.5 mM CaCl 2 , 1.0 mM NaHPO 4 , 11 mM glucose, 22.6 mM NaHCO 3 ) at 37°C. For in vivo observations, an anesthetized mouse was laid on the temperature-controlled microscope stage, and the right side of the eye was located on the cover glass of the microscope (44-mm diameter, 0.16-mm thickness; Carl Zeiss MicroImaging, Inc.). A small amount of cyanoacrylate glue was applied between the sclera and cover glass to minimize lateral movement of the eyes. In this configuration, the retina was imaged at the periphery by the laser penetrating through the sclera while the emission fluorescence was collected coming back into the microscopic objective lens. In case of a slight movement of the RPE cell layer, the same area of the retina was traced using the unique texture of the RPE cell layer formed by the randomly arranged singleand dual-nucleated RPE. In most experiments, thoroughly dark-adapted mice were exposed to intense 10-ms flashes that bleached 60% of the visual pigment.
Image Analyses-ImageJ software (National Institutes of Health) was used to track the motion of retinyl ester aggregates in the RPE. The contrast and brightness of the images were adjusted by the Adobe Photoshop CS2 (Adobe Systems Incorporated, San Jose, CA). Only linear adjustments were applied to the images.

RESULTS
Generation of Adfp-deficient Mice-We generated Adfp-deficient mice to investigate the effects of its ablation on the visual cycle. The targeting construct used for transgenesis consisted of a 5Ј-long arm starting from the 5Ј end of exon 2, the neomycin selection (Neo) cassette, and a 3Ј-short arm extending from the 3Ј end of exon 3 (Fig. 1A). By homologous recombination, the Neo cassette replaced exon 2 containing the ATG initiation site and exon 3, and therefore mice deficient in exon 2 and 3 of the Adfp gene are referred as Adfp ⌬2-3/⌬2-3 mice. Appropriate recombination was confirmed by a PCR-based method and Southern blotting analysis (Fig. 1B, top and bottom). Genomic Southern blotting analysis of Adfp ϩ/ϩ , Adfp ϩ/⌬2-3 , and Adfp ⌬2-3/⌬2-3 mouse genomes revealed the expected sizes of EcoRI-digested DNA fragments probed by 32 P-labeled Adfp cDNA, i.e. a 7.5-kb fragment from Adfp ϩ/ϩ mice, a 3.5-kb fragment from Adfp ⌬2-3/⌬2-3 mice, and a mixture of both fragments in Adfp ϩ/⌬2-3 mice. To confirm the lack of Adfp expression in Adfp ⌬2-3/⌬2-3 mice, we used RT-PCR, immunoblotting, and immunofluorescence techniques to characterize Adfp gene products in several tissues (Figs. 1 and 2). Expression of the full-length Adfp was abolished at the RNA level in both the RPE and white adipose tissue of Adfp ⌬2-3/⌬2-3 mice (Fig. 1C, top).
Recently, an Adfp ⌬2-3/⌬2-3 mouse line generated by another group (12) was shown to express a short isoform of Adfp in the lactating mammary gland (22). We found that the same short Adfp mRNA, which contains all the exons except exons 2 and 3, was still expressed in the RPE of Adfp ⌬2-3/⌬2-3 mice (supplemental Fig. 1). To understand the expression of the Adfp gene product in the RPE, we employed an antibody directed against the C-terminal region of Adfp (22). We confirmed that this C-terminal specific antibody recognized both the full-length and the short isoforms of Adfp in the mammary glands (Fig. 1C, bottom). As reported previously (9), Adfp ϩ/ϩ mice expressed Adfp in RPE cells ( Fig. 1, C and D) that we found to be located specifically to lipid droplet-like structures ( Fig. 1D, left, red signals indicated by arrows). In the RPE of Adfp ⌬2-3/⌬2-3 mice, the C-terminal specific antibody showed a faint diffused signal in the cytoplasmic area (Fig. 1D). Faint staining also was observed in the inner retina of Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice. These faint signals are possibly derived from N-terminally truncated Adfp suggesting that the short isoform of Adfp fails to bind to RESTs and is unstable in the RPE of Adfp ⌬2-3/⌬2-3 mice.
Expression Profiles of Six PAT Domain Proteins in the RPE-The tissue distribution of Adfp was tested by RT-PCR with primer pairs that specifically amplify Adfp from its start to stop codons. In addition to the expected size product, we also amplified a short DNA fragment from testicular tissue of Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice (Fig. 2, A and B). The full coding region of this cDNA then was cloned by 5Ј-and 3Ј-rapid amplification of cDNA ends PCR. Sequence analysis revealed that this cDNA encodes a novel PAT family protein with a predicted amino acid sequence comprising 159 residues (supplemental Fig. 2A). Because this coding region showed great similarity to that of Adfp (77.4% sequence homology for amino acids and 90% for nucleotides) (supplemental Fig. 2B), we named the protein Adfp2. Adfp2 is encoded by a gene located 27 kb downstream of the Adfp gene on the minus strand of chromosome 4C4 (Fig. 2C) suggesting that Adfp2 diverged from Adfp by recent gene duplication events. Adfp2 was expressed in the testis of Adfp ⌬2-3/⌬2-3 and Adfp ϩ/ϩ mice (Fig. 2) implying that removal of exons 2 and 3 from the Adfp gene had no effect on Adfp2 expression (Fig. 2C). Based on our analysis of the expressed sequence tag data base (www.ncbi.nlm.nih.gov/), Adfp2 also is expressed in the skin (GenBank TM accession number AA727312). Adfp2 seems to be absent from the human genome. With Adfp2 and the previously reported five other members (23)(24)(25)(26)(27), mice have six members of the PAT domain protein family in their genomes. Because Adfp2 is not expressed in the eye (Fig. 2B), we did not investigate it further.
We then studied the expression of PAT domain family members in the eye (Fig. 3). Previously it was shown that Tip47 is expressed in mouse RPE (28). Here we found that this protein is mainly localized to the cytoplasm of the RPE, both in Adfp ⌬2-3/⌬2-3 and Adfp ϩ/ϩ mice (Fig. 3A). Immunofluorescence microscopy failed to reveal perilipin, S3-12, and MLDP in the RPE of Adfp ⌬2-3/⌬2-3 and Adfp ϩ/ϩ mice (Fig. 3, B-D). Moreover, S3-12 and MLDP were not detectable in the RPE by immunoblotting analysis, and the perilipin mRNA level was about 0.1% that in white adipose tissue based on quantitative PCR analysis (data not shown). Thus the RPE expresses only a limited repertoire of PAT family proteins, namely Adfp and Tip47. Importantly, there was no apparent up-regulation of Tip47 in the absence of Adfp (Fig. 3E).
Delayed Dark Adaptation of Adfp ⌬2-3/⌬2-3 Mice-Adfp was restricted to the RPE (Fig. 1D), where it co-localized with retinyl esters (9). We characterized the rate of dark adaptation in Adfp ⌬2-3/⌬2-3 mice to determine whether Adfp is involved in the visual cycle. Two-month-old Adfp mice were exposed to intense light activation (500 cd⅐m Ϫ2 ) for 3 min, a light stimulus that photoactivates about 80 -90% of rhodopsin. After activation, recovery of the photoreceptor response was assessed by following a-wave amplitudes. a-wave amplitudes recovered more slowly in Adfp ⌬2-3/⌬2-3 mice than in Adfp ϩ/ϩ mice (Fig.  4A), indicating that Adfp ⌬2-3/⌬2-3 mice are deficient in photoreceptor dark adaptation. A more prominent delay in recovery was observed for these same BALB/c background mice at the age of 1 year (Fig. 4C). Adfp ⌬2-3/⌬2-3 mice with a C57BL/6 background also showed a similar delay in dark adaptation (supplemental Fig. 3). It is well established that polymorphisms of the Rpe65 gene affect the rate of dark adaptation (29). To exclude a possible contribution of Rpe65 gene polymorphisms to the delayed dark adaptation noted in Adfp ⌬2-3/⌬2-3 mice, we confirmed the Rpe65 genotypes of our Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice. Regardless of Adfp genotype, all mice with BALB/c backgrounds retained the Leu residue at position 450, whereas mice with a C57BL/6 background had a Met residue at this position. Therefore, the observed difference in dark adaptation between Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice was not because of reported Rpe65 polymorphisms.
Despite the observed delay in dark adaptation, scotopic and photopic ERG responses were comparable in 2-month-old Adfp ⌬2-3/⌬2-3 and Adfp ϩ/ϩ mice (Fig. 4, top row). But in 1-year-old Adfp ⌬2-3/⌬2-3 mice, attenuations in ERG responses were noted (Fig.  4D). These affected both a-wave and b-wave amplitudes under scotopic conditions and b-wave amplitudes under photopic conditions. Disruption of Adfp in mice did not affect the global histology of the retina up to the age of 1 year (Fig. 4E). No differences in the lengths of photoreceptor outer segments were discerned between Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice (Fig. 4F). The thickness of the outer nuclear layer and the number of nuclei were similar in both Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice (Fig. 4G), indicating the absence of photoreceptor degeneration in Adfp ⌬2-3/⌬2-3 mice. Thus, it is unlikely that the observed delay in dark adaptation was caused by global structural changes in photoreceptor neurons or the photoreceptor-RPE interface.
Neither was the delay in dark adaptation caused by a defect in phototransduction machinery because   SEPTEMBER 5, 2008 • VOLUME 283 • NUMBER 36 scotopic and photopic ERG responses were normal in both sets of 2-month-old mice. Nonetheless, a deficiency or variation in visual cycle components similar to that seen in Adfp-deficient mice often results in delayed dark adaptation of photoreceptor cells (21, 29 -32). Our results are consistent with a proposed role of Adfp in the visual cycle. Visual Cycle in Adfp ⌬2-3/⌬2-3 Mice-Adfp specifically binds to RESTs, and the amount of this protein is well correlated with the accumulation of retinyl esters in the RPE (9). All-trans-retinyl esters are substrates for Rpe65, the isomerohydrolase responsible for generation of visual chromophore (33)(34)(35). Therefore, we studied how an Adfp gene deficiency might affect the metabolism and storage of retinoids by quantifying retinoids in the eye by normal phase HPLC. Under dark-adapted conditions, the amount of all-trans-retinyl esters in Adfp ⌬2-3/⌬2-3 mice was ϳ25% that found in Adfp ϩ/ϩ animals (Fig. 5A). Photoactivation of rhodopsin (ϳ90% of total amount present in the retina) led to rapid isomerization of 11-cis-retinal to all-trans-retinal in both Dark-adapted mice were exposed to intense illumination (500 cd⅐m Ϫ2 ) for 3 min, and recovery of a-wave amplitudes was monitored with singleflash ERG (Ϫ0.2 log cd⅐s⅐m Ϫ2 ) every 5 min for 60 min. The recovery rate was significantly attenuated in Adfp ⌬2-3/⌬2-3 mice (*, p Ͻ 0.01 compared with Adfp ϩ/ϩ mice; n ϭ 5 each). B, ERG responses recorded from 8-to 10-week-old mice. Adfp ⌬2-3/⌬2-3 and Adfp ϩ/ϩ mice under scotopic (left) and photopic (right) conditions. Both a-and b-wave amplitudes are plotted as a function of light intensity (n ϭ 5). C and D, ERG responses of 1-year-old mice. C, recovery of a-wave amplitudes after constant light activation. Dark-adapted mice were exposed to intense illumination (500 cd⅐m Ϫ2 ) for 3 min, and recovery of a-wave amplitudes was monitored with single-flash ERG (Ϫ0.2 log cd⅐s⅐m Ϫ2 ) every 5 min for 60 min. Further delay in dark adaptation was noted in aged animals than in young animals (see A for comparison), and the delay was more pronounced in Adfp ⌬2-3/⌬2-3 mice than in Adfp ϩ/ϩ mice (*, p Ͻ 0.01 versus Adfp ϩ/ϩ mice; n ϭ 5 each). D, ERG responses recorded from Adfp ⌬2-3/⌬2-3 and Adfp ϩ/ϩ mice under scotopic (left) and photopic (right) conditions. The a-and b-wave amplitudes are plotted as a function of light intensity (n ϭ 5 each). Attenuations in ERG responses were noted for both a-wave and b-wave amplitudes under scotopic conditions and b-wave amplitudes under photopic conditions. E, cross-section of eyecups from 1-year-old Adfp ϩ/ϩ (left) and Adfp ⌬2-3/⌬2-3 mouse (right). Sections were located on the superior side and ϳ1.0 mm from the optic nerve head. F, lengths (m) of ROS were compared between Adfp ϩ/ϩ (open circles) and Adfp ⌬2-3/⌬2-3 (closed circles) mice at the age of 1 year. x axis indicates the distance from the optic nerve head in mm. No significant differences were observed between Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice. G, thicknesses (m) of outer nuclear layers were compared (n ϭ 4) between 1-year-old Adfp ϩ/ϩ (open circles) and Adfp ⌬2-3/⌬2-3 (closed circles) mice. x axis indicates the distance from the optic nerve head in mm. No significant differences were observed between Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice. Data are presented as means Ϯ S.D. OS, photoreceptor outer segments; ONL, outer nuclear layer; IS, inner segment; OPL, outer plexiform layer; INL, inner nuclear layer; IPL, inner plexiform layer; GCL, ganglion cell layer.
Aggregation of Newly Synthesized All-trans-retinyl Esters in Adfp ⌬2-3/⌬2-3 Mice-All-trans-retinol generated by photoreceptors is transported to the RPE where it is esterified to form all-trans-retinyl esters (36). Because Adfp is localized to RESTs in the RPE, we tested whether the observed delay in retinoid processing stems from dysfunctional handling of retinyl esters by the RPE. Two-photon microscopy (see Ref. 37) allowed us to monitor the changes in retinyl ester fluorescence at subcellular resolution and to determine the effect of Adfp on this process. As reported previously (9), photoactivation of rhodopsin resulted in a time-dependent increase of retinyl ester fluorescence in pre-existing RESTs under ex vivo conditions (Fig. 6A, top row). But in dark-adapted Adfp ⌬2-3/⌬2-3 mice, retinyl esters were diffusely localized within the cells (Fig. 6A, bottom row). After light exposure of Adfp ⌬2-3/⌬2-3 mouse eyes, a gradual increase in the number of these retinyl ester aggregates was observed in the same intracellular area of the RPE occupied by the smooth ER. At 30 min after light exposure, the number of retinyl ester aggregates (or RESTs) in Adfp ⌬2-3/⌬2-3 mouse was ϳ1.5 times greater than that in Adfp ϩ/ϩ mouse (supplemental Fig. 4). However, the amount of retinyl esters per each aggregate is lower in Adfp ⌬2-3/⌬2-3 mouse than in Adfp ϩ/ϩ mice (supplemental Fig. 4). To understand the ultrastructural relationship between these newly formed aggregates and the ER, we analyzed RPE from Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice by transmission electron microscopy. Mice were exposed to an intense light flash, and eyes were fixed 30 min later. Under this light condition, Adfp ⌬2-3/⌬2-3 mice had significant amounts of retinyl esters (about 72% the amount of Adfp ϩ/ϩ ) in their eyes (Fig.  5A). In Adfp ⌬2-3/⌬2-3 mice, a population of vacuole-like structures (Fig. 6B, left, arrows) was seen with a distribution similar to that of retinyl ester aggregates observed by two-photon microscopy. Only a small number of such structures were  detected in Adfp ϩ/ϩ mice (Fig. 6B, right). Intriguingly, these vacuole-like structures were encased in two layers of membranes and always surrounded by smooth ER (Fig. 6B, left inset). This morphological feature is reminiscent of the previously noted complex association of lipid droplets and ER membranes in cultured cells (2).
Retinyl ester storage structures of Adfp ϩ/ϩ mice show similar distributions over time under ex vivo conditions (Fig. 6A, top  row), suggesting that they are relatively immobile. We used high resolution time-lapse imaging of RPE after a light flash of anesthetized animals for direct monitoring of retinyl ester mobility in Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice. In Adfp ϩ/ϩ mice, RESTs moved only slightly, mostly up to 0.5 m/min (Fig. 6, C and D; supplemental movie 1). But Adfp ⌬2-3/⌬2-3 mice evidenced highly mobile retinyl ester aggregates that moved at rates of up to 1.5 m/min (Fig. 6, C and D, supplemental movie 2). These observations suggest that retinyl ester trafficking differs between Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice such that Adfp ⌬2-3/⌬2-3 animals tend to accumulate these compounds and thereby promote their aggregation in the middle of the ER.

DISCUSSION
Here we show that Adfp, a specific component of the RESTs in the RPE, is both involved in the transport and storage of retinyl esters and required for normal dark adaptation of photoreceptors. Our data suggest a specific role of Adfp on the distribution and localization of RESTs that cannot be counterbalanced by other members of the PAT domain protein family. Thus, this study clearly demonstrates that RESTs function to sustain the normal rate of vision recovery after photoactivation of visual pigments by regulating the storage and transport of retinol/retinyl esters in the eye. As reported recently, generation of biologically active retinoic acid also is regulated by the sequestration of retinol as retinyl esters by Lrat during zebrafish development (38). Our studies show that compartmentalization by Adfp binding is an essential process for regulating the temporal and spatial distribution of biologically active vitamin A derivatives. Possibly PAT domain proteins can increase the cellular storage capacity of retinyl esters and regulate the equilibrium between retinyl esters and retinol in tissues other than the eyes.
Role of Adfp in Retinoid Compartmentation and Transport within the Eye-The RPE plays a unique role in the processing of retinoids and storage of retinyl esters (1). Lrat, the major enzyme responsible for retinyl ester formation in several tissues, including hepatic stellate cells and the RPE (19,20), also is required for lipid droplet formation in hepatic stellate cells (20) and the RPE (supplemental Fig. 5). In the absence of Lrat, fluorescent RESTs did not form in the RPE and Adfp localized diffusively in the cytoplasm (9), but in the presence of Lrat, Adfp localized to RESTs often found proximal to the plasma membrane (9). Because of this unique retinyl ester-dependent translocation, we postulated that Adfp is involved in the transport of retinyl esters. Lrat predominantly localizes to the smooth ER where esterification of all-trans-retinol occurs (17). Because the subcellular sites of retinyl ester synthesis and storage are distinct, an uncharacterized mechanism must exist that facilitates the ER to REST translocation of retinyl esters essential for normal visual cycle functioning.
Direct imaging of retinyl esters in ex vivo and in vivo mouse eyes revealed that Adfp is essential for normal intracellular transport of retinyl esters in the RPE. Our current model of Adfp function in the visual cycle is depicted in Fig. 8. In darkadapted Adfp ϩ/ϩ mice, RESTs are well maintained and capable of serving as the "destination" of newly synthesized retinyl esters (Fig. 8A). But in dark-adapted Adfp ⌬2-3/⌬2-3 mice, these structures are poorly maintained such that newly synthesized retinyl esters form aggregates instead (Fig. 8B). Those retinyl ester aggregates increase in number after the release of alltrans-retinol from the rod outer segments (ROS); however, the FIGURE 7. Adfp increases the amount of retinyl esters in Rpe65 ؊/؊ mice. A, retinyl ester levels (pmol per eye) were compared between Rpe65 Ϫ/Ϫ and Adfp ⌬2-3/⌬2-3 Rpe65 Ϫ/Ϫ mice at the age of 3-4 months. Adfp knock-out mice had significantly less all-trans-retinyl esters in the eye (*, p Ͻ 0.01). B, two-photon image of retinyl esters in the RPE. Accumulations of retinyl ester aggregates were seen in RESTs of both Rpe65 Ϫ/Ϫ and Adfp ⌬2-3/⌬2-3 Rpe65 Ϫ/Ϫ mice at the age of 3-4 months but were less prominent in Adfp ⌬2-3/⌬2-3 Rpe65 Ϫ/Ϫ mice. retinyl ester content per aggregate is lower in Adfp ⌬2-3/⌬2-3 mice than in Adfp ϩ/ϩ mice (supplemental Fig. 4). These observations are in line with our model that Adfp increase the capacity of each aggregate to store retinyl esters. Considering the specific localization of Adfp to RESTs (9), these imaging results indicate that Adfp is required for the maintenance of RESTs during dark adaptation and that well maintained RESTs are essential for proper transport of retinyl esters in the RPE.
Retinyl esters were decreased in Adfp ⌬2-3/⌬2-3 mice, consistent with an essential function of Adfp in the maintenance of RESTs. Several previous studies have linked Adfp to the accumulation of lipid, both in cultured cells and in vivo. Overexpression of Adfp led to accumulation of lipid droplets in hepatic stellate cells in vitro (10,11). Also, diet-induced fatty liver was significantly reduced in Adfp-deficient mice (12) suggesting that Adfp promoted pathological accumulation of these lipid droplets. Exactly how Adfp achieves these effects in vivo remains unclear. Adfp was shown to reduce association of adipose triacylglycerol lipase (AGTL) with lipid droplets (13) leading to the hypothesis that reduced lipid ester hydrolysis in the presence of Adfp might directly contribute to the overall increase in TAG accumulation. However, this AGTL-related mechanism cannot directly account for the function of Adfp in the RPE, where Rpe65 is the only enzyme known to be responsible for the hydrolysis/isomerization of all-trans-retinyl esters (33)(34)(35). So if Adfp protects retinyl esters from Rpe65-catalyzed hydrolysis, the absence of Rpe65 in Adfp ⌬2-3/⌬2-3 Rpe65 Ϫ/Ϫ should not affect the amount of retinyl esters. But instead Adfp increased the amount of retinyl esters in the RPE of mice with Rpe65 Ϫ/Ϫ backgrounds. Another remarkable observation was that the rate of 11-cis-retinal synthesis did not differ signifi-cantly between Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice. Because formation of 11-cis-retinoid and hydrolysis of all-trans-retinyl esters are coupled reactions catalyzed by Rpe65, our results suggest that Adfp-mediated storage of retinyl esters is independent of the enzymatic activity of Rpe65.
Our previous and current studies suggest that Adfp is a retinyl ester-binding protein and that their interaction results in mutual stabilization. Thus, less accumulation of retinyl esters occurred in the absence of Adfp (Figs. 5 and 7), whereas induced accumulation of retinyl esters led to up-regulation of Adfp (9). An interaction of Adfp with retinyl esters also is indicated by the localization of Adfp on the surface of the RESTs (9). Intriguingly, the quantity of Adfp is tightly regulated by the ubiquitin/proteasome pathway that actively degrades Adfp in the absence of fatty acids (39). Pharmacological inhibition of proteasome degradation led to increased accumulation of cellular lipids that correlated with increased Adfp levels (39). Possibly up-regulated Adfp promotes accumulation of retinyl esters in the RPE by a similar mechanism, as illustrated by more accumulation of retinyl esters in Adfp ϩ/ϩ Rpe65 Ϫ/Ϫ mice as compared with Adfp ⌬2-3/⌬2-3 Rpe65 Ϫ/Ϫ mice. These genetic studies provide evidence that direct interactions between Adfp and retinyl esters facilitate compartmentalization of retinyl esters to the RESTs.
Transport and compartmentalization of retinyl esters confined to RESTs should reduce the concentrations of retinyl ester in the ER where retinyl ester biosynthesis occurs. In Adfp ⌬2-3/⌬2-3 mice, increases of retinyl ester in the ER would reduce the rate of retinyl ester synthesis by Lrat via product inhibition. Such potent product inhibition was shown previously with the retinyl ester analog, all-trans-retinyl bromoacetate (40).
Esterification of all-trans-retinol by Lrat is essential for transport of all-trans-retinol between photoreceptor cells and the RPE (9,19) so inhibition of Lrat activity should have a negative impact on this process (Fig. 8). Consistent with this concept, we observed delayed transport of all-trans-retinol and all-transretinal in Adfp ⌬2-3/⌬2-3 mice. Reduction in Lrat activity also should cause less accumulation of retinyl esters, as observed in Adfp ⌬2-3/⌬2-3 mice, but without affecting the processing of retinyl esters by Rpe65. Collectively, our findings are consistent with Adfp-mediated transport and compartmentalization in enhancing the storage of retinyl esters.
Another possible role for Adfp is its direct binding to hydrophobic substances to increase the rate of their uptake. Adfp expressed in COS7 cells enhanced the rate of long chain fatty acid uptake (41), and recombinant Adfp can bind to fatty acids in vitro (42). Adfp can be modified by an acyl group (43), and this hydrophobic residue may promote its interaction with other hydrophobic compounds such as retinyl esters and alltrans-retinol. It will be interesting to investigate the possible role of Adfp in the direct transport of all-trans-retinol.
Delayed Dark Adaptation and Retinoid Transport in Adfp ⌬2-3/⌬2-3 Mice-Delayed dark adaptation is one of the common phenotypes in mice with a deficiency or variation in particular visual retinoid cycle components. For example, polymorphisms in the Rpe65 gene largely affect the rate of 11-cisretinal regeneration (29). In other cases, delayed dark adapta-  tion can occur without any observable difference in the rate of 11-cis-retinal regeneration. In general, mouse models deficient in retinol dehydrogenases show delayed dark adaptation (21,31,32). Yet another example is the ABCR transporter-deficient mouse (30). Despite delayed processing of all-trans-retinal to retinol, the rate of 11-cis-retinal generation is not affected in all these mutant mice. But the degree of delay in dark adaptation is well correlated with the rate of all-trans-retinal clearance from ROS. All-trans-retinal may slow dark adaptation by binding to opsin and activating the transducin cascade (44,45). All-transretinal in the nanomolar range can also bind to cGMP-gated channels, and this has been suggested to slow dark adaptation by inhibiting the reopening of these channels (46,47). All these observations are consistent with the idea that delayed clearance of all-trans-retinal is the direct cause of the slow dark adaptation observed in Adfp ⌬2-3/⌬2-3 mice.
We also noticed that the delay in dark adaptation was more prominent in 1-year-old than in 2-month-old Adfp ⌬2-3/⌬2-3 mice (Fig. 4, A and C). From those observations, it is apparent that aging affects the visual performance of mice and that Adfp is required for the rapid restoration of photoresponses in old animals. Our recent studies suggest that clearance of all-transretinal and regeneration of 11-cis-retinal are delayed by aging in mice. 5 We assume that the attenuation of photoresponses in 1-year-old Adfp ⌬2-3/⌬2-3 mice (Fig. 4D) were caused by abnormalities in retinoid metabolism during the process of aging. Our experimental conditions for scotopic ERG can lead to subtle photoactivation of rhodopsin and generate free all-transretinal, which is more difficult to clear in older animals. Thus, the Adfp ⌬2-3/⌬2-3 mouse could serve as a model to study agedependent changes of dark adaptation and retinoid metabolism.
Possible Compensation for Adfp Deficiency by Other PAT Domain Family Members-This study ( Fig. 2 and supplemental Fig. 2) and previous studies indicate that six members of the PAT domain family exist in the mouse genome (4,26,27,48). Among them, only Adfp and Tip47 are expressed in the RPE (Figs. 1 and 3). Thus the RPE serves as a unique in vivo system to test the function of Adfp and Tip47 without interference from other PAT family members. In Adfp-null embryonic fibroblasts, Tip47 can compensate for the function of Adfp in increasing the storage of TAG (49). Possibly the relatively mild eye phenotype observed in Adfp ⌬2-3/⌬2-3 mice is because of functional compensation by Tip47. Intriguingly, aggregates of retinyl esters in Adfp ⌬2-3/⌬2-3 mice were more mobile than those in Adfp ϩ/ϩ mice. Two possible mechanisms could account for the increased movements of retinyl ester aggregates as follows. 1) Adfp functions in tethering RESTs proximal to the plasma membrane. 2) Lack of Adfp induces the random movement observed in Adfp ⌬2-3/⌬2-3 mice. The lack of Adfp promotes the interaction of retinyl ester aggregates with other proteins, which then facilitate such movement. Lipid droplets are associated with cytoskeletal components, such as vimentin and tubulin (50,51). In Drosophila melanogaster, the PAT domain protein, LSD2, has been proposed to interact directly with Klar, a proposed coordinator of microtubule-based transport (52). An interesting working hypothesis is that the interaction of mammalian lipid droplets with motor proteins is mediated by PAT domain proteins. Nocodazole, a potent inhibitor of tubulin polymerization, inhibits the fusion of small lipid droplets in NIH3T3 cells (51), suggesting active transport of lipid droplets. It is therefore feasible that differential interactions of Tip47 and Adfp with microtubule motor components effect the observed changes in the mobility and transport of retinyl esters in Adfp ϩ/ϩ and Adfp ⌬2-3/⌬2-3 mice. Future studies of the eye phenotypes in either Tip47 Ϫ/Ϫ or double mutant Adfp ⌬2-3/⌬2-3 -Tip47 Ϫ/Ϫ mice should be revealing in this regard. The essential function of Tip47 in recycling mannose 6-phosphate receptor (53) could hamper the generation of live Tip47 Ϫ/Ϫ mice, although a conditional knock-out of this gene in the eye would provide a feasible option to study the role of PAT proteins in the cell biology of lipid droplets.