Identification and characterization of all-trans-retinol dehydrogenase from photoreceptor outer segments, the visual cycle enzyme that reduces all-trans-retinal to all-trans-retinol.

Retinol dehydrogenase (RDH), the enzyme that catalyzes the reduction of all-trans-retinal to all-trans-retinol within the photoreceptor outer segment, was the first visual cycle enzymatic activity to be identified. Previous work has shown that this enzyme utilizes NADPH, shows a marked preference for all-trans-retinal over 11-cis-retinal, and is tightly associated with the outer segment membrane. This paper reports the identification of a novel member of the short chain dehydrogenase/reductase family, photoreceptor RDH (prRDH), using subtraction and normalization of retina cDNA, high throughput sequencing, and data base homology searches to detect retina-specific genes. Bovine and human prRDH are highly homologous and are most closely related to 17-beta-hydroxysteroid dehydrogenase 1. The enzymatic properties of recombinant bovine prRDH closely match those previously reported for RDH activity in crude bovine rod outer segment preparations. In situ hybridization and RNA blotting show that the PRRDH gene is expressed specifically in photoreceptor cells, and protein blotting and immunocytochemistry show that prRDH localizes exclusively to both rod and cone outer segments and that prRDH is tightly associated with outer segment membranes. Taken together, these data indicate that prRDH is the enzyme responsible for the reduction of all-trans-retinal to all-trans-retinol within the photoreceptor outer segment.

All vertebrate visual pigments consist of a chromophore, 11-cis-retinal, bound covalently to an integral membrane protein, opsin. Photoisomerization of the chromophore from 11-cis to all-trans is the initiating event in vision, inducing a conformational change in the attached protein that renders it competent to activate a photoreceptor-specific G-protein. Visual pigment activation and inactivation occur on time scales of hundreds of microseconds and hundreds of milliseconds, respectively (1). Regeneration of the visual pigment requires release of all-trans-retinal and binding of a new molecule of 11-cis-retinal, a process that occurs on a time scale of minutes (2)(3)(4)(5).
Photobleaching of visual pigments was discovered by Kuhne (6), who observed that light-exposed rhodopsin undergoes a rapid conversion from pink to yellow followed by a slow conversion to a colorless, UV-absorbing product. Wald and Hubbard (7,8) demonstrated that the yellow product of visual pigment bleaching is all-trans-retinal that is subsequently reduced enzymatically to the colorless product, all-trans-retinol. Kuhne (6) made the further observation that efficient visual pigment regeneration requires close apposition of the neural retina and the retinal pigment epithelium (RPE). 1 This requirement is now known to reflect an obligatory movement of all-trans-retinol from the rod outer segment (ROS) to the RPE, followed by isomerization from the all-trans-to the 11-cis-configuration, oxidation of 11-cis-retinol to 11-cis-retinal, and a return of 11-cis-retinal from the RPE to the outer segment ( Fig.  1; Refs. 2 and 9 -11). The joining of 11-cis-retinal and opsin regenerates the visual pigment and completes the cycle of events, which together are referred to as the visual cycle. The rate-limiting step in the visual cycle at high light levels appears to be the conversion of all-trans-retinal to all-trans-retinol within the outer segment (4,5).
Although the chemical transformations that comprise the visual cycle are understood in outline, many of the relevant proteins remain poorly characterized, and several were unknown until recently. Within the outer segment, recent work (12,13) suggests that a photoreceptor-specific ABC transporter, ABCR, facilitates the enzymatic reduction of all-transretinal to all-trans-retinol by transporting all-trans-retinal and/or its Schiff base adduct with phosphatidylethanolamine (PE) from the lumenal to the cytosolic face of the disc membrane. Mutations in ABCR are responsible for autosomal recessive Stargardt disease (14 -17), an early onset macular dystrophy associated with impaired dark adaptation and yellow deposits in the retina and RPE, and for a subset of cases of autosomal recessive retinitis pigmentosa (18). Mutations in ABCR may also represent a risk factor for age-related macular degeneration (19). These retinal disease phenotypes and the corresponding physiologic defects observed in ABCR knockout mice (13) underscore the importance of efficiently converting all-trans-retinal to all-trans-retinol within the outer segment. * This work was supported by the Howard Hughes Medical Institute and the NEI, National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM  Retinol dehydrogenase (RDH), the enzyme that catalyzes the reduction of all-trans-retinal to all-trans-retinol (by convention named for the reverse reaction), was the first visual cycle enzyme to be identified (8). Photoreceptor RDH utilizes NADPH (20 -22), shows a marked preference for all-transretinal over 11-cis-retinal (22,23), and is tightly associated with the ROS membrane (22)(23)(24)(25)(26). Despite 50 years of study, minimal progress has been made in purifying photoreceptor RDH (24,26), and our current knowledge of its enzymatic properties is based on studies of its activity in ROS homogenates, detergent extracts, or minimally enriched preparations (20 -26).
In this paper we report the amino acid sequence of bovine and human photoreceptor RDH (prRDH) as deduced from cDNA and genomic DNA sequences, the exclusive localization of prRDH to rod and cone outer segments, and the enzymatic properties of recombinant prRDH produced in transfected mammalian cells.

EXPERIMENTAL PROCEDURES
Reagents-11-cis-Retinal was a gift from Hoffmann-La Roche; other reagents used in the RDH assays were obtained from Sigma. Retinoids were stored under argon in the dark at Ϫ80°C, and their integrity was monitored by examining an absorbance spectrum prior to each experiment.
Construction and Analysis of a Normalized and Subtracted Bovine Retina-minus-Brain cDNA Library-Total RNA was isolated from bovine retina and frontal cortex using the guanidinium-phenol method (27). Poly(A) ϩ RNA was purified by oligo(dT) chromatography (Fast-Trak II; Invitrogen), and 2 g each of bovine retina and frontal cortex poly(A) ϩ RNA was used for first strand cDNA synthesis with 20 units of avian myeloblastosis virus-reverse transcriptase and 200 units of Superscript II (Life Sciences). Subtraction and normalization were performed using bovine retina cDNA as tester and bovine frontal cortex cDNA as driver as described ((28) PCR-Select kit; CLONTECH) with minor modifications to the published protocol. In brief, the retina cDNA was divided into two portions and ligated to two different adapters that were modified to encode an EcoRI restriction site 5Ј-CTAATACGACT-CACTATAGGGCTCGAGCGGCCGCCCGAATTCGT and 5Ј-CTAATAC-GACTCACTATAGGGCAGCGTGGTCGCGGAATTCGT. Ligation efficiency was assessed by PCR of glyceraldehyde-3-phosphate dehydrogenase sequences using primers 5Ј-ACCACTGTCCACGCCATCAC and 5Ј-AGTGGTCGTTGAGGGCAATG that amplify a segment of the bovine glyceraldehyde-3-phosphate dehydrogenase sequence that lacks an RsaI site. The two portions of retina cDNA were denatured and annealed in the presence of 20-fold excess driver cDNA and 20% PEG 8000. In a second cycle of annealing the two primary hybridization samples were mixed without denaturing, and a 20-fold excess of freshly denatured driver cDNA was added. After the second round of annealing, those double-stranded retina cDNA segments that were flanked by two different adapters were amplified in two rounds of PCR (Advantage cDNA PCR; CLONTECH), digested with EcoRI, size-selected for fragments greater than 300 bp by agarose gel electrophoresis, and cloned into the EcoRI site of gt10.
To identify and eliminate abundant cDNA clones, the cDNA library was plated at a density of 300 plaques per 10-cm plate, and nitrocellulose filter replicas were prepared and hybridized under standard conditions with a probe prepared from the subtracted and normalized retina cDNA. Approximately 50% of the clones showed little or no hybridization, and 1000 of these were individually picked into 96-well trays in ϳ200 l of lambda dilution buffer (10 mM Tris-HCl, pH 8.0, 10 mM MgSO 4 ). Five microliters of the phage eluate was used for insert amplification in a 30-l PCR with gt10-specific primers 5Ј-CTTTT-GAGCAAGTTCAGCCTGGTTAAGT and 5Ј-GAGGTGGCTTATGAG-TATTTCTTCCAGGGTAA. Twenty microliters of each PCR product was incubated with 5 l of 7.5 M ammonium acetate containing 0.4 mg/ml yeast tRNA and 25 l of isopropyl alcohol for 30 min on a shaker platform followed by a 30-min centrifugation at 2,000 ϫ g at 4°C. The precipitated PCR products were washed with 70% ethanol, dried, and resuspended in 20 l of 5 mM Tris, pH 8.0, 0.5 mM EDTA and analyzed on a 1.5% agarose gel. Approximately 90% of the PCR showed single inserts, and these were subjected to automated sequence analysis. Sequences were compared with the GenBank TM nonredundant data base using the BLASTN and BLASTX programs and to the GenBank TM EST data base using the BLASTN program.
cDNA and Genomic DNA Clones-Multiple independent cDNAs encoding bovine and human prRDH were isolated by DNA hybridization to oligo(dT)-primed cDNA libraries from adult human retina (29) and adult bovine retina. The bovine retina cDNA library was prepared in ZAP Express (Stratagene) from poly(A) ϩ RNA. The central 95% of the coding region of mouse prRDH was amplified by PCR from a mouse eye cDNA library (30) using primers based on the bovine prRDH sequence. The cloned murine prRDH PCR product was used for Northern blot and in situ hybridization. The human PRRDH gene was isolated from a Sau3AI partial-digest human genomic library in EMBL3 (a gift of Y. Wang) by hybridization with a human prRDH cDNA probe. The complete sequence of the bovine prRDH coding region was deduced from the sequences of three independent cDNA clones. The sequence of the human prRDH coding region was deduced from two independent cDNA clones and from sequencing the six exons from cloned genomic DNA.
Chromosomal Localization-Chromosome mapping of human The principal chemical transformations that constitute the visual cycle occur in the photoreceptor outer segment (upper) and retinal pigment epithelium (lower). Current evidence indicates that the RPE is required for the rod visual cycle; whether cones require the RPE for chromophore isomerization is uncertain (Redmond et al. (67)).
prRDH was performed using the GeneBridge 4 and Stanford TNG radiation hybrid panels (Research Genetics). The GeneBridge 4 panel was screened by PCR with a primer pair from the 3Ј-noncoding sequences of human prRDH that produce a 147-bp fragment (forward primer, 5Ј-AACAACCAGACCTCTTCATTCCAC; reverse primer, 5Ј-CA-GAGGAGTCTATGCCTGAAAATAAAG). Statistical analysis using RH-MAPPER localized the PRRDH gene between marker WI-8049 and marker WI-7557 at a distance of 3.05 centiRay 3,000 from the latter, but with a less-than-significant log-of-the-odds score of 0.54. The Stanford TNG panel was screened by PCR with a primer pair from intronic sequences flanking human prRDH exon 5 that produce a 275-bp fragment (forward primer, 5Ј-GAGTGGTTGGGGCATCAGACTTACT-3Ј; reverse primer, 5Ј-GCAACATACCTGAGCCACGTCTTGG-3Ј). Statistical analysis using the Stanford Genome Center Radiation Hybrid server localized prRDH near marker SHGC-15369 on chromosome 19 with a log-of-the-odds score of 5.66 at a distance of 33 centiRay 10,000 . SHGC-15369 is not ordered on the chromosome map, but it is linked to marker WI-7557, the human gene for DNA cytosine-(5)-methyltransferase (GenBank TM accession number X63692) on chromosome 19p13.3-13.2. (Genome Data Base locus D19S830). To verify further the chromosomal location of prRDH, a fragment from the 3Ј-untranslated region of human prRDH cDNA was used to probe a genomic Southern blot from a somatic cell hybrid panel (Oncor). This analysis assigned prRDH to either chromosome 1 or chromosome 19.
Northern Blot Hybridization-RNA was isolated from adult rat or bovine tissues using the guanidinium thiocyanate method (27). For rapid screening of the tissue specificity of bovine EST clones, RNA dot blots were prepared with total RNA from retina, heart, liver, lung, kidney, spleen, skeletal muscle, cerebellum, and cerebrum and hybridized with the PCR-amplified inserts described above. For Northern blotting, 20 g of total RNA from each tissue was resolved by agarose gel electrophoresis in the presence of formaldehyde and transferred to a nylon filter. Filters were hybridized under standard conditions with either the 0.5-kb BamHI-BglII fragment of the bovine prRDH cDNA (for the bovine RNA blot) or the PCR-amplified mouse prRDH open reading frame (for the rat RNA blot) and washed in 0.1ϫ SSC at 65°C.
In Situ Hybridization-For in situ hybridization to bovine retina, the cornea and lens were removed, and the retina was fixed for 1 h at 4°C with 4% paraformaldehyde in PBS, cryoprotected overnight in 30% sucrose, frozen in OCT, and sectioned at 20 m. A 1.2-kb fragment of bovine prRDH cDNA containing the complete coding region was subcloned into pBluescript (Stratagene) and transcribed in vitro using T3 or T7 RNA polymerases to prepare antisense or sense probes, respectively. In situ hybridization was performed with digoxigenin-labeled riboprobes as described previously (30) except that high stringency washes and RNase treatment were used as described by Wilkinson et al. (31). In situ hybridization to sections of whole mouse eye was performed by the same method.
Antibody Production and Purification-A fusion protein containing amino acids 5-312 from human prRDH fused to the C terminus of glutathione S-transferase was produced in Escherichia coli, purified by preparative SDS-polyacrylamide gel electrophoresis, and used for immunization of rats. An analogous fusion containing the same human prRDH segment fused to the maltose-binding protein (MBP; New England Biolabs) was produced in E. coli and purified by amylose affinity chromatography. The purified MBP fusion protein was used for affinity purification of immune sera after covalent coupling to Affi-Gel 10 matrix (Bio-Rad).
A second immunogen was prepared using a pair of synthetic Cterminal peptides from bovine prRDH. Both peptides contain the sequence CFRTPVWPR but differ in having either a free carboxylate or an amide at the C terminus. The two peptides were mixed at a 1:1 ratio, cross-linked to BSA carrier protein with glutaraldehyde (32), and used to immunize rabbits. The antiserum was affinity-purified using the peptide mixture coupled to an Affi-Gel-10 matrix.
Immunoblotting-ROS were prepared by sucrose gradient centrifugation from dark-adapted bovine retinas or from human retinas obtained from the Lions Eye Bank in Denver, CO, as described (33). To prepare total retinal protein, bovine retinas were resuspended in PBS, 1% Nonidet P-40 containing protease inhibitors (1 g/ml each chymostatin, leupeptin, antipain, and pepstatin A and 1 mM phenylmethylsulfonyl fluoride), incubated on ice for 15 min, and centrifuged at 5,000 ϫ g for 10 min to remove nuclei and debris. Protein concentrations were determined using the Bradford assay (Bio-Rad) or estimated by comparison to a dilution series of BSA following gel electrophoresis and Coomassie Blue staining. Immunoblots were sequentially incubated with affinity-purified rat anti-human prRDH antibodies followed by horseradish peroxidase-conjugated goat anti-rat antibody (The Jack-son Laboratory). Bound antibodies were visualized using the ECL detection system (Amersham Pharmacia Biotech).
Immunohistochemistry-Macaque retinas were collected following cardiac perfusion with 4% paraformaldehyde in PBS and overnight fixation of isolated eyecups in the same buffer. After cryoprotection in 8% sucrose and embedding in OCT, 10 -20-m frozen sections were prepared, fixed for 30 s in acetone, incubated for 1 h in PBS containing 0.05% SDS and 5% normal goat serum, and incubated in PBS containing affinity-purified primary antibody at 1:50 in PBS containing 0.05% SDS and 5% normal goat serum overnight at 4°C, and washed 3 times in PBS containing 0.3% Triton X-100 followed by 3 washes in PBS. Biotinylated, fluorescein-conjugated, or Texas Red-conjugated goat anti-rat secondary antibody (Vector) was diluted 1:200 to 1:500 in PBS with 5% normal goat serum and incubated for 1 h at room temperature. Biotinylated antibody was detected using horseradish peroxidase and the avidin-biotin complex method (Vector) with diaminobenzidine substrate. Rhodamine-conjugated peanut agglutinin (Vector) was used at a concentration of 5 g/ml. Confocal images were collected on a Zeiss LSM 510.
Production of Bovine prRDH in Transfected Cells-The bovine prRDH open reading frame was modified by site-directed mutagenesis to contain a HindIII site in place of the stop codon and then fused in frame to a DNA segment encoding six histidines followed by the Cterminal 9 codons of bovine rhodopsin (the latter is recognized by mAb 1D4 (34)). This modified prRDH open reading frame was expressed in transiently transfected 293 cells using a pCIS vector (35). Twenty four hours after transfection, cells were disrupted in a Polytron homogenizer, and the membranes were purified by centrifugation onto a sucrose shelf and then concentrated by centrifugation as described (36). Membrane pellets from 293 cells transfected with RDH and from untransfected 293 cells were resuspended in 0.4 ml of 50 mM HEPES, pH 6.5, 140 mM NaCl, 3 mM MgCl 2 , 2 mM EDTA at a protein concentration of 175 g/ml and stored in aliquots at Ϫ80°C.
RDH Enzyme Assays-RDH reactions for HPLC analysis were performed as follows. Four-fold serial dilutions of membranes were pre- For spectrophotometric analysis of the RDH reaction, the following components were first assembled, aliquoted, and frozen on dry ice: 1) 3.5 g of 293 membranes, 1% BSA, 20% glycerol, and 1% CHAPS in buffer B (200 l aliquots); and 2) NADPH or NADH at variable concentrations, 40 M all-trans-retinal, 0.1% BSA, 2% glycerol, and 0.1% CHAPS in 2ϫ buffer B (220-l aliquots). Each RDH reaction was started by thawing 1 aliquot of each type and then mixing them together in a quartz cuvette. The cuvette was placed in a water-jacketed cuvette holder maintained at 32°C, and spectra were recorded between 300 and 600 nm with a Uvikon 860 spectrophotometer over the ensuing 20 min. Under these conditions, the rate of conversion of retinal to retinol proceeds linearly for at least the first 30 min of the reaction.
HPLC Separation of Retinoids-Incubation in 0.25 M O-ethylhydroxylamine for 15 min at room temperature quantitatively converts retinal to O-ethyl retinal oxime, which can be readily resolved from retinol by C-18 reverse phase HPLC (37). Following the 15-min incubation, retinoids were extracted with 80 l of hexane, dried under argon, dissolved in 100 l of methanol, and resolved by HPLC on a 25-cm C-18 column with 100% methanol as the mobile phase. The column was monitored at 335 nm, a wavelength at which the extinction coefficients of all-trans-retinol and all-trans-retinal oxime are equal.

RESULTS
Identification and Primary Structure of prRDH, a Novel SDR-To identify genes expressed specifically in the retina, we constructed a library from bovine retina cDNA that had been simultaneously normalized and subtracted with bovine brain cDNA. One thousand cDNA clones were identified that contained low abundance sequences, as judged by their failure to hybridize to a probe consisting of the pool of normalized and subtracted retina cDNA. Approximately 0.5 kb of sequence was determined from each of these clones. A comparison of these 1,000 partial sequences with the GenBank TM data base re-vealed a novel sequence with significant homology to 17-␤hydroxysteroid dehydrogenase type 1 (17-␤-HSD1), a member of the short chain dehydrogenase/reductase (SDR) family. This novel sequence was of interest because of the possibility that it codes for the long-sought photoreceptor all-trans-retinol dehydrogenase. As described below, this appears to be the case, and we therefore refer to this clone as prRDH (photoreceptor retinol dehydrogenase).
A search of the expressed sequence tag (EST) data base with the original prRDH sequence resulted in the identification of only one EST with a nearly identical DNA sequence (clone 76 g1; GenBank TM accession number w22972). Subsequent searches with a full-length prRDH sequence identified a second EST (GenBank TM accession number w22947, clone 76e7) that derives from the 3Ј-untranslated region. Both ESTs are part of a collection of 5,000 ESTs sequenced from a human retina cDNA library that was not normalized. 2 These data suggest that prRDH expression may be limited to the retina.
Isolation of cDNA clones from human and bovine retina cDNA libraries and characterization of their complete coding sequences revealed open reading frames of 311 and 312 codons, respectively, encoding proteins with calculated molecular masses of 34 kDa (Fig. 2A). The longest human cDNA clone is missing the first 8 codons that were subsequently identified by sequencing human prRDH genomic clones (see below). The predicted human and bovine proteins show 85% mutual identity and 48% identity with 17-␤-HSD1, the most similar sequence in the current GenBank TM data base (Fig. 2B).
The predicted prRDH amino acid sequence contains all the motifs present in the SDR superfamily of enzymes including the invariant YXXXK motif containing the catalytic Tyr-155 (in the human prRDH numbering) and the highly conserved nucleotide cofactor binding motif GXXXGXG at amino acids 12-18 (38). The presence of serine at position 14 within the nucleotide cofactor binding motif suggests a preference for NADP/NADPH binding. In many SDRs that utilize NAD/NADH, the analogous position is occupied by aspartate that most likely excludes the 2Ј-phosphate of NADP/NADPH (39). Other amino acids that are predicted from the crystal structure of the 17-␤-HSD1-NADPH complex to interact with NADPH (40) are also conserved as follows: Ser-14, Ser-15, Ile-17, Arg-40, Leu-65, Val-67, Asn-90, Gly-92, Thr-190, and Phe-192.
To determine the intron-exon structure of the human PRRDH gene, overlapping clones were isolated from a Sau3AI partial digest human genomic DNA library. The relative location of the 6 exons and a restriction map of the surrounding 20.5 kb of human genomic DNA, obtained from 4 overlapping clones, are depicted in Fig. 3. Sequencing of each of the coding exons confirms the amino acid sequence deduced from the human cDNAs. The human PRRDH gene was mapped to chromosome 19p13 using radiation hybrid and chromosome hybrid cell lines. At present, no human retinopathies map to this region.
Expression of prRDH in Photoreceptors-The tissue distribution of prRDH transcripts was examined by Northern blot hybridization to total RNA from various bovine tissues ( B and C) and mouse retina (not shown). In retinas from both species, prRDH transcripts are present exclusively in the outer nuclear layer, indicating expression in photoreceptors. Although the uniform prRDH hybridization pattern in the outer nuclear layer clearly indicates expression in rods, the relative paucity of cones in these species and the partial superposition of rod and cone hybridization signals in 20-m sections does not permit a clear conclusion to be drawn from these data regarding prRDH expression in cones.
Localization of prRDH to Photoreceptor Outer Segments-To localize the prRDH protein within the retina, polyclonal antibodies were raised against a glutathione S-transferase-human prRDH fusion protein produced in E. coli (Fig. 5A). Immunoblots of non-nuclear proteins from whole bovine retinas or proteins from sucrose gradient purified ROS show that affinitypurified anti-prRDH antibodies bind to a single polypeptide with an apparent molecular mass of ϳ40 kDa that is greatly enriched in ROS (Fig. 5B). Since little or no prRDH signal is detected in the total retina sample and the ROS purification produces a ϳ20-fold enrichment of ROS protein relative to total retinal protein, this analysis suggests that virtually all of the prRDH resides within outer segments.
Optimal detection of prRDH in the ROS samples requires boiling of the samples, a treatment that results in aggregation of opsin, the major ROS protein. We assume that boiling facilitates detection of prRDH on Western blots because it removes comigrating opsin that would otherwise interfere with binding of prRDH to the filter or limit access to the antibody. We note that the constant appearance of the prRDH Western blot signal at ϳ40 kDa before and after boiling of the protein sample in SDS-gel buffer demonstrates that the signal obtained with the anti-prRDH antibodies is not derived from cross-reactivity with opsin, which, as noted above, is converted from monomers to multimers upon boiling (Fig. 5C). A quantitative analysis of the prRDH Western blot signal obtained with human ROS and a dilution series of an MBP-human prRDH fusion protein indicates that prRDH represents ϳ0.2-0.5% of ROS protein. This abundance implies that the 13-and 25-fold purifications of RDH from bovine ROS reported by previous workers (24,26) resulted in preparations that were far from homogeneous.
As a second method for determining the subcellular localization of prRDH, macaque retinas were immunostained with affinity-purified anti-prRDH antibodies. As seen in Fig. 5D, prRDH is detected exclusively in photoreceptor outer segments. Similar results were obtained with human retinas (data not shown). Double labeling with rabbit anti-ABCR, a marker for ROS (41), and rat anti-prRDH shows that in the macaque retina all ROS contain prRDH (not shown). prRDH is also present in cone outer segments, as seen most readily in the peripheral retina where the large cone inner segments permit a clear distinction between cones and rods (Fig. 6A). As further confirmation of this localization, Fig. 6, B, C, and D, shows a section of peripheral macaque retina cut in the plane of the retina at the level of the outer segments. The section was double-labeled with peanut agglutinin (rhodamine), which binds specifically to the extracellular matrix sheaths of cones (42) and anti-prRDH (fluorescein). These analyses show that prRDH is present in both rod and cone outer segments in the primate retina.
Enzymatic Activity of Recombinant prRDH-Bovine prRDH was tagged at its C terminus with 6 histidine residues followed by the C-terminal 9 residues of bovine rhodopsin (which constitute the epitope for mAb 1D4 (34)), and the epitope-tagged prRDH was produced in transiently transfected 293 cells (Fig.  5A). Following homogenization and ultracentrifugation, virtually all of the expressed prRDH could be recovered in the membrane pellet as assayed by Western blotting with either mAb 1D4 or affinity-purified rat anti-human prRDH antibodies.
To determine whether recombinant prRDH can reduce alltrans-retinal to all-trans-retinol, crude membrane preparations from untransfected 293 cells or 293 cells transfected with bovine prRDH were solubilized in CHAPS and incubated in the presence of 100 M NADPH and 50 M all-trans-retinal. The substrate and product of the reaction were resolved by HPLC following conversion of retinal to retinal-O-ethyl oxime (37). In this system retinol elutes at ϳ6 min and retinal-O-ethyl oxime elutes at ϳ7 min. The product of the RDH reaction appears to be all-trans-retinol as it is indistinguishable from pure alltrans-retinol in HPLC retention time and absorption spectrum. As seen in Fig. 7, A-F, membranes from 293 cells expressing prRDH have substantial RDH activity, whereas untransfected 293 cell membranes have minimal activity. However, when 11-cis-retinal is used as a substrate instead of all-trans-retinal, membranes from 293 cells expressing prRDH show ϳ20-fold lower activity (Fig. 7, G-M). This 20:1 ratio of conversion efficiency probably underestimates the specificity of prRDH for the all-trans-isomer since several percent of the 11-cis-isomer would be predicted to isomerize thermally to the all-trans configuration during the 1 h in which it is dissolved in detergent solution during preparation and incubation of the reactions (43). The specificity of prRDH for all-trans over 11-cis-retinal ensures that the visual cycle is not short-circuited by adventitious reduction of 11-cis-retinal in the outer segment prior to its recombination with opsin.
The RDH reaction can also be monitored spectrophotometrically, a method that is more convenient for kinetic experiments (Fig. 8). In Fig. 8, A-C, the difference spectra are generated by subtracting a single reference spectrum measured at an early time point from spectra measured at later times. In the difference spectra, the negative peak at ϳ330 nm and the positive peak at ϳ380 nm arise from the time-dependent increase in all-trans-retinol ( max ϳ325 nm) and the concomitant decrease in NADPH ( max ϳ340 nm), all-trans-retinal ( max ϳ385 nm), and protonated Schiff bases of all-trans-retinal ( max ϳ440 nm). The difference spectra in Fig. 8A confirms that reduction of all-trans-retinal occurs in the presence of membranes from prRDH-transfected cells and not in the presence of membranes from untransfected cells.
Previous analyses of RDH activity in crude ROS preparations indicate that NADPH serves as a cofactor but NADH does not (22,44). Recombinant prRDH shows the same specificity (Fig. 8B), as predicted by the amino acid sequence alignment described above. Measurements of the initial reaction velocity ( Fig. 8C) as a function of NADPH concentration gives a K m value for NADPH of ϳ9 M (Fig. 8D), in good agreement with previous estimates of 3.5 (24) and 13 M (22) for the enzyme activity in crude ROS membranes. Although most of the recombinant prRDH polypeptide in transfected 293 cell membranes is insoluble in 1% CHAPS (as measured by Western blotting), ϳ60% of the enzymatic activity is solubilized by CHAPS addition (not shown). This suggests that a large portion of the recombinant protein expressed in 293 cells may be aggregated and/or denatured.
Membrane Association of prRDH-All-trans-RDH activity in ROS is membrane-associated and can be solubilized with mild detergents (25). As noted above, prRDH activity obtained from transfected 293 shows similar behavior. Western blotting of human ROS membranes with anti-prRDH antibodies shows that treatment of sonicated ROS membranes with 5 M urea, 2 M NaCl, water, or 0.2 M sodium carbonate, pH 10.5, fails to release detectable quantities of prRDH from the membrane, whereas addition of Triton X-100 to 1% solubilizes greater than 90% of prRDH (not shown). Although the prRDH sequence shows no regions of extended hydrophobicity that might be indicative of a membrane-embedded domain, the C-terminal 14 amino acids of human and bovine prRDH contain three conserved cysteines (Fig. 2A). These residues are not found in other members of the SDR family. Their location suggests that they may serve as sites of lipid attachment, a mode of membrane association that is found in a number of ROS proteins including cGMP phosphodiesterase and transducin (45)(46)(47). Consistent with the possibility that the C terminus of prRDH is modified and/or proteolytically processed, we observe that antibodies against a synthetic peptide corresponding to the Cterminal 9 amino acids of bovine prRDH recognize a significant fraction of the recombinant bovine prRDH expressed in transfected COS cells but do not recognize bovine prRDH in isolated

DISCUSSION
Identification and Properties of prRDH-This paper reports the identification and characterization of a photoreceptor-specific all-trans-retinol dehydrogenase, prRDH. prRDH was initially identified as a candidate retina-specific dehydrogenase/ reductase using cDNA subtraction and normalization, high throughput sequencing, and data base homology searches. prRDH is a member of the short chain dehydrogenase/reductase superfamily and is most closely related to 17-␤-hydroxysteroid dehydrogenase 1. Its amino acid sequence is well conserved among mammals, and in humans it is encoded by a single copy gene on chromosome 19. The properties of recombinant bovine prRDH match those previously reported for RDH activity in crude bovine ROS preparations or detergent extracts with respect to membrane association and substrate specificity (NADPH versus NADH and all-trans-retinal versus 11-cis-retinal). Northern blot and in situ hybridization show that, among the tissues and cell types assayed, prRDH transcripts are found exclusively in photoreceptors. Western blotting and immunocytochemistry show that within the retina prRDH localizes exclusively to rod and cone outer segments. This high specificity in cell-type expression and subcellular localization is a characteristic of many outer segment proteins, and it suggests that like other outer segment proteins prRDH is specialized for visual function.
One question that arises from the present work is whether prRDH is the only outer segment RDH relevant to the visual cycle. Various RDH activities have been reported in crude retinal homogenates and ROS preparations, and it has long been known that other SDR family members, such as horse liver alcohol dehydrogenase, can efficiently reduce retinal to retinol (21,48). Some of the RDH activities present in the retina utilize NADH, but current evidence suggests that these activities do not reside in outer segments (20 -23, 44). The strongest evidence on this point comes from experiments in which small quantities of all-trans-retinal were generated in situ by limited light exposure of dark-adapted bovine ROS, and RDH activity was monitored by the transfer of 3 H from [ 3 H]NADPH or [ 3 H]NADH to all-trans-retinal (22). Under these conditions, ROS RDH activity exhibits an absolute requirement for NADPH.
A novel short chain RDH that is distinct from prRDH and is referred to as retRDH, was identified recently from a retina cDNA library (49). retRDH is also expressed in a variety of nonretinal tissues. Like prRDH, the enzymatic properties of recombinant retRDH match those of the outer segment RDH activity. At present, the precise localization of retRDH within the retina is unclear; an anti-retRDH monoclonal antibody shows immunoreactivity in cones but not rods, whereas in situ hybridization shows a signal in rods but not cones (49). Both methods also label cells within the inner retina. Whether prRDH acts alone within the outer segment or is partially redundant with retRDH in rods and/or cones remains to be determined.
Physiologic Significance of the Reduction of All-trans-retinal-There are several reasons for believing that efficient reduction of all-trans-retinal is essential for optimal photoreceptor function. First, all-trans-retinal can combine in vitro with opsin to create a complex that has significant G-protein (transducin) coupling activity (50 -53). Although this activity has yet to be directly demonstrated in intact photoreceptors by electrophysiological measurements (reviewed in Ref. 54), a large body of electrophysiological and psychophysical data point to the persistence of visual pigment bleaching products as the source of the noise that limits the kinetics of dark adaptation (55)(56)(57). Second, by virtue of its ability to form Schiff bases with free amines in proteins and with PE in the outer segment disc membrane, all-trans-retinal has the potential to interfere with protein function and to remain sequestered on the lumenal face of the outer segment discs. By contrast, all-trans-retinol would be expected to readily diffuse across the ROS disc and plasma membranes. Third, there is strong circumstantial evidence that one of the major constituents of lipofuscin within the aging human RPE, a di-retinal adduct referred to as A-2E, forms within the outer segment following the sequential condensation of two all-trans-retinal molecules with the ethanolamine of PE (58 -60). A2-E carries a fixed positive charge, and it accumulates within phagosomes in the RPE, presumably following engulfment and digestion of outer segments (61,62). Fourth, photoreceptor outer segments can efficiently generate NADPH from NADP via the hexose monophosphate pathway (20,63), an arrangement that presumably evolved to supply NADPH for reduction of all-trans-retinal. It will be of interest to determine whether prRDH interacts with the enzymes associated with the hexose monophosphate shunt, at least one of which, glyceraldehyde-3-phosphate dehydrogenase, is a relatively abundant ROS protein (64).
Given the clear importance of rapidly reducing all-transretinal following visual pigment photoactivation, it seems paradoxical that this reaction constitutes the rate-limiting step of the visual cycle in vivo. In the dark adapted rat eye, a flash of light sufficient to bleach ϳ10% of the visual pigment leads to a measurable accumulation of all-trans-retinal that persists for approximately 30 min (4). In the mouse eye, constant illumination sufficient to produce a 30 -40% bleach of visual pigment results in all-trans-retinal accumulation to approximately 30% of total retinoids, whereas all-trans-retinol accumulates to only ϳ5% of total retinoids (5). During a period of dark recovery following 60 min of steady illumination, the concentration of all-trans-retinal declines with a half-life of approximately 5 min (5).
One mechanistic explanation for the apparent inefficiency of the RDH reaction is that it may be limited by the accessibility of all-trans-retinal. This idea follows from the likely role of ABCR in transporting or flipping all-trans-retinal and/or its Schiff base adducts with PE from the lumenal to the cytosolic face of the disc membrane to make it available to RDH (12,13), which presumably acts only on all-trans-retinal substrates that reside on the cytosolic face of the bilayer. In this regard it is of interest that the RDH activity present in ROS has been shown to reduce efficiently exogenously added Schiff base adducts of retinaldehyde (retinylidene imines) to retinol (21). By contrast, horse liver alcohol dehydrogenase can reduce free retinal but does not act on Schiff bases of retinal. These data suggest that Schiff bases of retinal and PE may represent a physiologically relevant substrate for hydrolysis and reduction in the ROS. In light of these considerations, it will of interest to determine whether prRDH interacts with ABCR or colocalizes with ABCR at the disc rim (41,65) or whether transport of all-trans-retinal by ABCR is coupled to its reduction by prRDH. One teleological explanation for the apparent inefficiency of the prRDH reaction (at least among terrestrial mammals) is that the observed reaction rate may be sufficient to accommodate the rate of dark adaptation required under most natural lighting conditions. With the exception of the very recent experience of our own species, the only light source of relevance to terrestrial mammals and their ancestors has been the sun (and its reflected light from the moon). The rate of change in the intensity of natural illumination at dawn and dusk is relatively slow, approximately a factor of 10 every 10 min (66). It therefore follows that there should be little or no selective pressure for the retina to attain dark adaptation kinetics very much greater than this rate, and indeed the sensitivity of the human eye increases by approximately a factor of 10 every 5 min during dark adaptation (66). This line of reasoning suggests that those animals that experience rapid changes in the level of ambient illumination, for example aquatic mammals and birds that dive for food at depths where there is little light, might have increased rates of dark adaptation and a corresponding increase in the rate of reduction of all-trans-retinal.
Implications for Retinal Disease-The importance of the visual cycle in human retinal function is underscored by the recent identification of mutations in four different visual cycle genes in a variety of autosomal recessive human retinopathies: ABCR in Stargardt disease and retinitis pigmentosa, as noted in the Introduction (14 -18); RPE65, an abundant RPE protein that is required for isomerization of retinal from all-trans-to 11-cis-(67), in early onset retinal degenerations (68 -70); cellular retinaldehyde-binding protein in retinopathies associated with night blindness, maculopathy, and yellow/white deposits in the fundus (71)(72)(73); and 11-cis-retinol dehydrogenase, the RPE enzyme that oxidizes 11-cis-retinol to 11-cis-retinal (74), in patients with delayed dark adaptation and yellow/white deposits in the fundus (75). If, as the present work suggests, prRDH is the principal enzyme responsible for the reduction of all-trans-retinal in the outer segment, then mutations in the PRRDH gene may be responsible for or may increase the risk and/or severity of any of a variety of inherited retinopathies that would ultimately be referable to defects in this step in the visual cycle.