Photoreceptor cGMP Phosphodiesterase δ Subunit (PDEδ) Functions as a Prenyl-binding Protein*

Bovine PDEδ was originally copurified with rod cGMP phosphodiesterase (PDE) and shown to interact with prenylated, carboxymethylated C-terminal Cys residues. Other studies showed that PDEδ can interact with several small GTPases including Rab13, Ras, Rap, and Rho6, all of which are prenylated, as well as the N-terminal portion of retinitis pigmentosa GTPase regulator and Arl2/Arl3, which are not prenylated. We show by immunocytochemistry with a PDEδ-specific antibody that PDEδ is present in rods and cones. We find by yeast two-hybrid screening with a PDEδ bait that it can interact with farnesylated rhodopsin kinase (GRK1) and that prenylation is essential for this interaction. In vitro binding assays indicate that both recombinant farnesylated GRK1 and geranylgeranylated GRK7 co-precipitate with a glutathione S-transferase-PDEδ fusion protein. Using fluorescence resonance energy transfer techniques exploiting the intrinsic tryptophan fluorescence of PDEδ and dansylated prenyl cysteines as fluorescent ligands, we show that PDEδ specifically binds geranylgeranyl and farnesyl moieties with a Kd of 19.06 and 0.70 μm, respectively. Our experiments establish that PDEδ functions as a prenyl-binding protein interacting with multiple prenylated proteins.

Because PDE␦ is able to regulate the association of Rab13 to the membrane, which is similar to the function of GDI (GDP dissociation inhibitor), similarities among PDE␦, RhoGDI, and RabGDI were investigated (23). By analysis of the predicted secondary structure of PDE␦ and the crystal structure of RhoGDI, it was found that PDE␦ and RhoGDI shared striking structural similarity, whereas the sequence identity is very low. However, a number of amino acid residues within the geranylgeranyl binding pocket of RhoGDI are conserved particularly well in PDE␦. The crystal structure of PDE␦ verified the presence of a hydrophobic domain packed by two opposite ␤-sheets, and the overall ␤-sandwich fold is identical to that of RhoGDI (24). The binding of Arl2 to PDE␦ was found to be independent of lipids, suggesting that PDE␦ can interact with proteins in two distinct ways: 1) through a lipid binding pocket and 2) ␤-sheet/␤-sheet interactions (24).
In terms of prenylated ligand specificity, PDE␦ is more promiscuous than GDI. Although only GTPases of the Rho family have been shown to interact with RhoGDI, PDE␦ interacts with various prenylated proteins including protein kinases, PDE subunits, and GTPases. To examine interactions of PDE␦ with components present in the retina, we screened a y2h retina expression library with a PDE␦ bait and specifically explored PDE␦ interactions with two photoreceptor-specific protein kinases: rhodopsin kinase GRK1 (farnesylated) and its homologue GRK7 (geranylgeranylated). We found that the prenyl groups covalently linked to their C-terminal cysteines mediate GRK/PDE␦ interactions. We then investigated interactions among PDE␦, dansylated farnesyl, and geranylgeranyl cysteines in the absence of polypeptides and found that PDE␦ interacts with farnesyl but to a lesser extent with geranylgeranyl side chains. We conclude that PDE␦ is a promiscuous prenyl-binding protein targeting hydrophobic prenylated C termini of a variety of polypeptides.

EXPERIMENTAL PROCEDURES
Polyclonal Antibody Preparation and Confocal Immunolocalization-Anti-PDE␦ antibody (directed against the N-terminal peptide-MSAKDERAREILRGFKLC supplied by Brent Rollmann, Affinity BioReagents, Inc., Golden, CO) was purified by affinity chromatography. To verify specificity by immunoblotting, a small piece of fresh bovine retina was sonicated briefly in PBS and centrifuged at 20,000 ϫ g for 10 min at 4°C to remove insoluble debris. An aliquot of the supernatant was subjected to 14% SDS-PAGE and transferred to nitrocellulose, and the membrane was blotted using anti-PDE␦ polyclonal antibody as the primary antibody and HRP-anti-rabbit IgG as secondary antibody and visualized by chemiluminescence (ECL kit, PerkinElmer Life Sciences). To determine the photoreceptor localization of PDE␦, fresh bovine retina was fixed 2 h in freshly prepared 4% paraformaldehyde in 0.1 M phosphate buffer, pH 7.4, rinsed in phosphate buffer, cryoprotected in phosphate buffer containing 30% sucrose, and frozen. Sections (12-m thick) were cut, blocked 1 h in 0.1 M phosphate buffer containing 0.1% Triton X-100 and 10% normal goat serum, and incubated with primary antibodies: anti-PDE␦ polyclonal antibody (1:400 dilution), Rho1D4 monoclonal antibody (1:500), and 7G6 monoclonal antibody (1:200 dilution). After rinsing in 0.1 M phosphate buffer for 10 min (three times), the sections were incubated with secondary antibodies tagged with fluorescein isothiocyanate or rhodamine. Immunolocalization images were captured using a Zeiss LSM 510 confocal microscope set to an optical slice of 0.9 m.
Construction of Bovine Retina Yeast Two-hybrid cDNA Library-Fresh bovine eyes were purchased from a local vendor (Dale T. Smith & Sons Meat Packing Co., Draper, UT). Retinas were dissected immediately and stored in liquid nitrogen, and bovine retina mRNAs were isolated using a FastTrack TM 2.0 kit (Invitrogen). Purified mRNA quality was estimated by 2.2 M formaldehyde, 1% agarose gel electrophoresis followed by Northern blotting. 10 g of retina mRNA was used to construct the library. First-strand cDNAs were synthesized using ran-dom primers (ZAP Express cDNA synthesis kit, Stratagene). EcoRI adapters (Stratagene) were added to both ends of the double-stranded cDNAs. After the adapter ligation, the double-strand cDNAs were phosphorylated at the EcoRI sites (T4 polynucleotide kinase, Stratagene) and then purified using cDNA size-fractionating columns (CHROMA SPIN TM -400 columns, Clontech), which removed unligated adapters, unincorporated nucleotides, and most short fragments (Ͻ300 bp). The purified, adapter-ligated double-strand cDNAs were ligated into the EcoRI-digested, dephosphorylated pGADT7 vector. Transformation of the recombinant plasmids (bovine retina cDNAs in pGADT7 vector) into electro-competent Escherichia coli DH5 cells (Clontech) was performed using a Bio-Rad gene pulser. The original library contains ϳ10 6 -independent clones. The titer of the first amplified library is 2 ϫ 10 9 cfu/ml. Construction of PDE␦ Bait and Library Screening-The full-length coding region of bovine PDE␦ was amplified using the bovine retina library (described above) as template and cloned into vector pGBKT7 (Clontech) in-frame with the upstream GAL4 DNA binding domain, and the bait construct was sequenced. Competent yeast cells were prepared using a LiAc/single-stranded carrier DNA/polyethylene glycol yeast transformation method (www.umanitoba.ca/faculties/medicine/biochem/gietz/2HS.html). The AH109 yeast colony was inoculated into 5 ml of YPD (2% Difco peptone, 1% yeast extract, 2% glucose) medium supplemented with 0.003% adenine hemisulfate and grown at 30°C overnight. The overnight culture was transferred to 50 ml of warm YPDA (YPD supplemented with adenine hemisulfate) medium with a final cell density of 5 ϫ 10 6 /ml and grown at 30°C on a shaker until the cell density reached 2 ϫ 10 7 cells/ml. The cells were harvested by centrifugation and washed once with 20 ml of sterile water followed by a wash with 1.0 ml of 0.1 mM LiAc. The cells then were resuspended in 100 mM LiAc with a final volume of 0.5 ml. Subsequently, the cells were aliquoted into 10 Microfuge tubes and pelleted and the LiAc was removed. For each 1.5 ml of Microfuge tube, the transformation mixture (consisting of 240 l of polyethylene glycol 3350 (50% w/v), 36 l of 1.0 M LiAc, 50 l of predenatured salmon sperm DNA (2.0 mg/ml), 5 g of bovine library DNA, and 5 g of bovine PDE␦ bait DNA) was added to achieve a total volume of 360 l. The tubes were vortexed vigorously to completely disperse the cells. The tubes were incubated at 30°C for 30 min and heat-shocked at 42°C for 30 min. The cells were pelleted and resuspended in double-distilled H 2 O and spread on synthetic dropout medium plates lacking tryptophan, histidine, adenine, and leucine (THAL medium). The plates were incubated at 30°C. After 4 -5 days, the colonies growing up from the selective plates were transferred to the same selective plates but containing 0.2 mM ␣-X-gal. Blue colonies were inoculated into media, and the DNA from each positive clone was sequenced using T3/T7 primers.
Expression of PDE␦ and Purification GST-PDE␦ Fusion Protein-Full-length mouse PDE␦ cDNA was amplified by RT-PCR and cloned in-frame into pGEX-2T expression vector (Amersham Biosciences). The expression construct was verified by DNA sequencing and transformed into E. coli ER2566 (New England Biolabs). A transformant was inoculated into LB medium and grown overnight at 37°C. The culture was transferred to fresh LB medium with a ratio of 1:100 and continued growing until the A 600 reached 0.6 -0.8. Isopropyl-1-thio-␤-D-galactopyranoside was added to the culture with a final concentration of 0.1 mM. The culture was moved to a 30°C shaker and grown for another 5 h. The cells were harvested, resuspended in PBS containing 0.5% Triton X-100, and disrupted by sonication. The lysate was cleared by centrifugation. The supernatant was loaded onto a GSTrap column (Amersham Biosciences). After extensive washing with PBS, the protein was eluted by PBS containing 1 mM reduced glutathione. Recombinant GST-PDE␦ purity was checked by SDS-PAGE, and its quantity was measured (protein assay kit, Bio-Rad).
In vitro Expression of GRK1/GRK7 and Binding Assay-GRK1 and GRK7 were each expressed in High-Five (H5) insect cells (25). Cultures (5-ml) were centrifuged, and the cells were resuspended in 1 ml of PBS containing 0.1% Triton X-100, 0.1% Tween 20, and protease inhibitor mixture (Roche Applied Science). Cell debris was pelleted by re-centrifugation while the supernatant was divided between two Microfuge tubes. In each tube, 20 l of glutathione-conjugated-agarose slurry and 4 g of GST protein (or equal amount of GST-PDE␦ protein) were added to supernatant. The tubes were incubated 2 h at room temperature on a rocking platform. The agarose was pelleted and washed (three times) with 1 ml of PBS containing 0.1% Triton X-100. Bound proteins were eluted by the addition of 40 l of 2ϫ sample buffer and boiling for 2 min. The supernatants and eluted proteins were separated by SDS-PAGE followed by immunoblot detection with anti-GRK1 monoclonal antibody G8 (from K. Palczewski, University of Washington-Seattle) or anti-GRK7 polyclonal antibody UU45APC (25).

Synthesis of N-Acetyl S-Farnesyl Cysteine Methyl Ester (AFCME) and
N-Acetyl S-Geranylgeranyl Cysteine Methyl Ester (AGGCME)-AFCME and AGGCME were synthesized as described previously and stored as a stock solution in Me 2 SO ( Fig. 1) (26).

Synthesis of N-Dansyl S-Geranylgeranyl Cysteine Methyl Ester (DG-GCME) and N-Dansyl S-Farnesyl Cysteine Methyl Ester (DFCME)-
S-Geranylgeranyl cysteine methyl ester and S-farnesyl cysteine methyl ester were prepared according to a method reported previously (27). To synthesize DGGCME, a solution containing 1.32 mg (4.91 mol) of dansyl chloride, S-geranylgeranyl cysteine methyl ester (2 mg, 4.91 mol), and N,N-diisopropylethylamine (1 l, 7.5 mol) in 0.3 ml of N,N-dimethylformamide was stirred at room temperature overnight. The reaction mixture was concentrated and purified on SiO 2 (ethyl acetate-hexane 1:4). Finally, 1.92 mg (61%) of oil was obtained. Similarly, 0.34 mg of DFCME was prepared. The molecular structures of both compounds were confirmed by 1 H NMR. The fluorescence properties were determined by a SPECTRAmax Gemini XS multiplate spectrofluorimeter.
Competition Assay-For competition experiments, an excess amount of compounds in stock solution were diluted into 100 M PBS separately and incubated with 7 g of GST-PDE␦ overnight at room temperature. The preincubated GST-PDE␦ was mixed with H5 cell lysate containing expressed GRK1 or GRK7 in the presence of excess AFCME or AGGCME. The mixtures were incubated for 1 h at room temperature. The pull-down assays were carried out as described above.
Fluorescence Resonance Energy Transfer (FRET) to Determine Dissociation Constants of PDE␦ Binding to Prenyl Groups-For FRET assay of PDE␦ in a total volume of 100 l of PBS, 2.8 M PDE␦ was mixed with 10 M DFCME or DGGCME and incubated overnight at room temperature. DFCME or DGGCME without protein was used as control. The fluorescence emission spectrum was recorded with a fixed ex ϭ 282 nm from 315 to 550 nm.
To determine the dissociation constant of PDE␦ binding to DFCME, a series of increasing concentrations of DFCME in 100 l of PBS was excited at 282 nm and the fluorescence spectrum for each concentration was measured. The values of fluorescence intensity at ϭ 505 nm were retrieved and plotted against the concentrations of the fluorescence probe. Their relations can be described in Equation 1, where F is the measured relative fluorescence intensity, C the concentration of probe, and k 1 stands for a coefficient that was determined by curve fitting using Origin software. DFCME at different concentrations then was incubated with or without 2.8 M PDE␦ in 100 l of PBS at room temperature for 4 h. The fluorescence spectrum between 315 and 540 nm for each mixture was measured. The dissociation constant of PDE␦ binding to DFCME was fit to Equation 2 using Origin software, where F t is the emission fluorescence of the protein and probe mixture at 505 nm, F f is the fluorescence of DFCME without adding protein, C t is the concentration of added DFCME, k 2 is the coefficient, K d is the dissociation constant, and P t is the initial concentration of added protein. The dissociation constant of PDE␦-binding DGGCME was determined in a similar way with the exception that the protein and fluorescence probe mixture were incubated overnight.

PDE␦ Is Expressed in Rods and
Cones-Multiple tissue Northern blots showed that many tissues express the PDE␦ gene at low levels, but among the tissues tested, the retina expressed the highest levels (10,12). At the protein level, the PDE␦ polypeptide immunolocalized to rod but not cone outer segments (10). In view of our hypothesis that PDE␦ may interact with multiple prenylated proteins, we re-investigated the distribution of PDE␦ in the bovine retina with a polyclonal antibody raised against an N-terminal polypeptide of the bovine sequence (Fig. 2, A-F). In a Western blot of bovine retina extract, anti-PDE␦ recognized the 17-kDa PDE␦ polypeptide nearly exclusively (Fig. 2G). Although immunolabel attributable to PDE␦ is present at low levels throughout all of the layers of transversely sectioned retina (results not shown), the most intense label is associated with photoreceptors. Specifically, strong signals originate in the rod inner segments and rod outer segments colocalizing with rhodopsin (Fig. 2, D-F), consistent with previous results (10). However, PDE␦ was also observed in cone photoreceptors (Fig. 2, A-C), including outer segments and synaptic pedicles. Cone photoreceptors were identified specifically with an anti-cone arrestin antibody (mAb 7G6) (28), a widely used marker for primate and bovine cones. The presence of PDE␦ in the cytoplasm of cones is demonstrated clearly when cones are double-labeled with mAb 7G6 and anti-PDE␦ antibody (Fig. 2F). A similar result was obtained with a second monospecific antibody raised against recombinant PDE␦ (FL, results not shown). Like PDE␦, cone arrestin is a soluble protein that associates with membranes under certain conditions. PDE␦ is distributed throughout the cone cells, consistent with cytoplasmic distribution. Results of our immunolocalization studies show that PDE␦ is present in bovine rod and cone photoreceptors and in much lower concentrations in other retinal cell types as well (data not shown), supporting the hypothesis that PDE␦ function is not limited to its association with rod PDE.
Identification of Retina Proteins Interacting with PDE␦-We next used full-length bovine PDE␦ (mouse and bovine PDE␦ polypeptides are 98.5% identical) as a bait to identify PDE␦interacting proteins by screening a y2h bovine retina cDNA library. In total, we screened approximately 4 million clones and found that Ͼ1,000 colonies were able to grow on selective medium and turned blue on selective THAL medium containing ␣-X-gal. We randomly picked 95 colonies and isolated the library plasmids (Table I) from these yeast clones that contained both PDE␦ bait plasmid and library plasmid. Sequencing and in-frame translation revealed several clones encoding Arl2 and PDE␣ that were known PDE␦-interacting partners, and seven clones represented partial sequences of GRK1 (rhodopsin kinase). These seven clones probably originated from two independent clones that corresponded to 165 (GRK1L) and 69 (GRK1S) amino acid residues at the C terminus of GRK1, respectively. Full-length bovine GRK1 consists of 561 amino acid residues with a predicted mass of 63 kDa. To verify spec- ificity of the interaction between the C termini of GRK1 and PDE␦, the two independent GRK1 clones (encoding the 165 and 69 amino acid residues of the GRK1 C terminus, respectively) were co-transformed into yeast with bovine RG4/Unc-119, a retina-dominant protein that shares 23% identity with PDE␦ in its C-terminal 153 amino acids (29,30). We found that partial GRK1 polypeptides encoded by these two clones only interacted with PDE␦ but not with RG4/Unc-119, suggesting that PDE␦ specifically interacts with GRK1.
In Vitro Interaction of PDE␦ with GRK1 and GRK7-To test whether the farnesylation of GRK1 is required for its interaction with PDE␦, the cysteine residue of the CAAX box motif of GRK1L was mutated to serine, a mutation that disables farne-sylation of GRK1L at the C terminus. The library plasmidencoding mutant GRK1L was cotransformed into yeast with PDE␦ bait. The yeast cells did not grow on the selective THAL medium. This result suggests that farnesylation is essential and required for the interaction between PDE␦ and GRK1. Similarly, Rab13 did not bind to PDE␦ when the CAAX box was deleted (12). To further confirm the interaction between PDE␦ and full-length GRK1 protein, we performed in vitro pull-down assays with PDE␦ expressed in E. coli as a GST fusion protein (Fig. 3A). Because the post-translational farnesylation is required for the interaction between PDE␦ and GRK1L in yeast, we expressed GRK1 in H5 cells (these cells prenylate cysteines of the CAAX box motif, whereas bacteria are unable to perform this C-terminal modification). In addition, we also expressed GRK7, another photoreceptor-specific G protein-coupled receptor kinase that was recently identified in several species (25,31,32). Upon the incubation of GST-PDE␦ protein with the crude extract of H5 cells infected by GRK-expressing viruses, GRK1 (Fig. 3A) and GRK7 (Fig. 3B) bound to GST-PDE␦ immobilized by glutathione-agarose, whereas the recombinant kinases could not be coprecipitated with GST alone. These results demonstrate that PDE␦ can bind specifically both GRK1 and GRK7 in vitro.  Prenyl Chains Compete with the Binding of PDE␦ to GRK1 or GRK7-We suspected that farnesyl or geranylgeranyl side chains alone may be capable of binding to PDE␦. Therefore, we designed experiments in which recombinant GRK1 and GRK7 were incubated with PDE␦ in the presence of excess of isoprenoid compounds. AFCME and AGGCME, representing the modified C termini of prenylated proteins, were synthesized (see "Experimental Procedures"). GST-PDE␦ was preincubated with these two synthetic compounds, and the pull-down binding assay was performed as described above. The presence of AFCME could completely abolish the binding of GST-PDE␦ to (farnesylated) GRK1 and strongly compete with the binding of GST-PDE␦ to (geranylgeranylated) GRK7. AGGCME also strongly competed with the binding of GST-PDE␦ to GRK1 but only weakly prevented the binding of GST-PDE␦ to GRK7 (Fig.  3). In a control experiment, ACME (the backbone of AFCME and AGGCE) did not compete (Fig. 3). These results demonstrate that PDE␦ interacts with GRK1 and GRK7 predominantly through prenyl moieties at their C termini. The shorter farnesyl side chain (C15) has higher binding affinity for PDE␦ than the longer geranylgeranyl chain (C20). The inability of AGGCME to chase GRK7 binding to GST-PDE␦ suggests that additional binding sites exist on the GRK7 polypeptide that stabilizes the interaction.
Binding Constants of PDE␦ Bound to Prenyl Side Chains-Our experiments indicate PDE␦ can function as a prenyl-binding protein. Therefore, we decided to use FRET to measure the PDE␦/prenyl-chain dissociation constants, taking advantage of intrinsic tryptophan fluorescence of the PDE␦ polypeptide.
When PDE␦ is excited at 282 nm, a strong tryptophan fluorescence can be observed with a peak emission at ϭ 335 nm. The four tryptophan residues present in PDE␦ are distributed uniformly in its binding pocket and serve as reporters for ligand binding. For the PDE␦ FRET assay, we synthesized two isoprenoid compounds in which dansyl was conjugated to S-farnesyl cysteine methyl ester and S-geranylgeranyl cysteine methyl ester, representing the C termini of prenylated proteins. Fluorescence spectra of synthetic DFCME and DGGCME indicated that the maximal excitation is around 340 nm, close to the maximal emission of PDE␦. The maximal fluorescence emission of both DFCME and DGGCME is at ϭ 505 nm.
The FRET assay was carried out by mixing proper amounts of PDE␦ with DFCME or DGGCME and exciting the mixture at ϭ 282 nm. Significant energy transfer was seen for DFCME almost immediately after the mixing (Fig. 4A), whereas FRET was observed for DGGCME only after extended incubation. As expected, concomitant with FRET, quenching of the intrinsic  Fig. 1A but GRK1 replaces GRK1. C, competition of GRK1 binding to PDE␦ by prenyl groups. GST-PDE␦ was incubated with excess amount of AF-CME or AGGCME. The preincubated GST-PDE␦ then was mixed with crude extract of insect cell expressing GRK1 for pull-down assay in the presence of excess amount of AFCME and AGGCME, respectively. The coprecipitated proteins along with GST-PDE␦ were analyzed by Western blot using anti-GRK1 antibody. Me 2 SO, the solvent, and ACME, the backbone of the isoprenoid compounds, were used as controls. The four lanes represent the solvent or the different compounds used for competition. Lane 1, DMSO; lane 2, ACME; lane 3, AFCME; lane 4, AGGCME. D, identical sequence of experiments in which GRK1 was replaced by GRK7. Note that competition of geranylgeranylated GRK7 is weaker, consistent with the higher K d values.
FIG. 4. Binding of dansyl farnesyl and dansyl geranylgeranyl to PDE␦. A, FRET assay of PDE␦ binding to DFCME. 9 M DFCME in the presence or absence of 2.8 M PDE␦ was excited at ϭ 282 nm simultaneously, and the emission fluorescence spectrum from 315 to 530 nm was recorded. The intrinsic fluorescence of PDE␦ was also measured. f, the fluorescence from DFCME probe only; E, the fluorescence from the protein and DFCME mixture; OE, the intrinsic fluorescence from protein without DFCME. B, non-linear curve fitting of the binding of PDE␦ to DFCME or DGGCME. Various concentrations of fluorescence probe were mixed with 2.8 M PDE␦ or a control buffer without protein. For each concentration, the fluorescence probe or the probe-protein mixture was excited at ϭ 282 nm and the fluorescence spectra were recorded. Each point in the figure represents the difference between the relative fluorescence intensity from probe-protein mixture and from the probe alone at ϭ 505 nm for each corresponding concentration of fluorescence probes. The dissociation constants were fit by computer software as described under "Experimental Procedures." tryptophan fluorescence of PDE␦ by DFCME or DGGCME was observed (Fig. 4A) because of the energy transferred from PDE␦ to the fluorescent probes. No FRET was observed for the mixture of PDE␦ with a control dansyl probe (dansyl alanine cycloheximide, no prenyl chain was present). Thus, PDE␦ can specifically bind farnesyl or geranylgeranyl. We next set out to determine the binding affinity of PDE␦ to DFCME or DG-GCME. When a series of concentrations of fluorescent probes were incubated with PDE␦, different intensities of FRET were seen (Fig. 4B). Calculations (see "Experimental Procedures") of the dissociation constants (K d ) revealed that half-maximal binding of PDE␦ to farnesyl is at 0.70 Ϯ 0.27 M, whereas that of PDE␦ to geranylgeranyl is at 19.06 Ϯ 2.41 M. The binding constant of farnesyl to PDE␦ is higher than that of another prenyl-binding protein, RhoGDI (K d ϭ 4.8 M), whereas geranylgeranyl has approximately 10 times less affinity for PDE␦ compared with RhoGDI (K d ϭ 1.6 M) (26). DISCUSSION We show that PDE␦ can interact with both farnesyl (C15) and geranylgeranyl (C20) side chains in the absence of polypeptides. The following independent results support this hypothesis. 1) Almost all of the PDE␦-binding proteins identified by y2h screening are prenylated (as exceptions, the N-terminal region of RPGR and Arl2/Arl3 have no lipid modification). 2) Mutations in the CAAX box motif-disabling prenylation eliminate the binding of GRK1 to PDE␦, suggesting that for some interacting partners the lipid attachment is essential. 3) An excess of prenyl side chains prevents stable interactions between GST-PDE␦ and prenylated GRKs in GST pull-downs assays. 4) Dansylated farnesyl and geranylgeranyl chains stably bind to PDE␦ with binding constants in the millimolar range (dansyl alone is unable to bind). The binding affinity between PDE␦ and farnesyl is comparable to RhoGDI-binding farnesyl, whereas the binding affinity between PDE␦ and geranylgeranyl is much weaker, consistent with our competition assay data (Fig. 3). An analysis of the prenyl-binding data revealed that PDE␦ binds farnesyl and geranylgeranyl groups with a stoichiometry of 1:1. Thus, it is very likely that GRK1 and GRK7 bind to PDE␦ with a ratio of 1:1, whereas for multiple subunit proteins with more than one prenyl tail such as rod PDE, it is predicted that each subunit can interact with one PDE␦.
In addition to these experiments, other groups showed that the deletion of 13 amino acid residues at the C terminus of Rab13 eliminated binding to PDE␦ (12) and prenylated/carboxymethylated peptides corresponding to the C termini of PDE catalytic subunits blocked PDE/PDE␦ interaction while non-prenylated polypeptides had no effect. Recent crystallographic data of PDE␦ show that PDE␦ forms a hydrophobic pocket between two ␤-sheet propellers, providing a structural basis for the binding between PDE␦ and prenyl chains (24). The hydrophobic pocket within PDE␦ is shallower than the similar pocket in RhoGDI, which is designed to bind geranylgeranylated proteins of the Rho family, consistent with weaker binding of geranylgeranyl of PDE␦. Despite weaker PDE␦/geranylgeranyl association, the interaction between PDE␦ and geranylgeranylated GRK7 appears stronger than the interaction between PDE␦-GRK1 because neither the excess of farnesyl nor that of geranylgeranyl could completely compete off PDE␦ binding to GRK7, whereas excess of farnesyl could completely compete off PDE␦ binding to GRK1. It is conceivable that within GRK7, there exists a second PDE␦ binding site apart from its geranylgeranylated C terminus, which may have higher affinity for PDE␦ than its C terminus. A second binding site in GRK7 is undefined at the molecular level but may be akin ARl2/Arl3 interaction sites or similar to interaction with RPGR, which has a high affinity PDE␦ binding site (K d ϭ 90 nM) within its N-terminal half (13,18). It has been shown that PDE␦ has C-terminal SRV and FYV motifs that may be involved in interacting with PDZ (PSD95, Dig, ZO-1) domaincontaining proteins (12). When PDE␦ C-terminal sequences were deleted, interaction with prenylated targets was disrupted (13) and colocalization with vesicular structures was disabled (12). Therefore, it is conceivable that PDE␦ has several major interacting domains in addition to the hydrophobic pocket binding to prenylated proteins.
Although both PDE␦ and RhoGDI can bind prenyl groups and they share structural similarity in their hydrophobic pocket, their distinct sequences suggest that they assume different functions. RhoGDI was originally identified as a Rho GTPase-binding protein. RhoGDI prefers the GDP-bound form of Rho family proteins, serving as Rho GDP dissociation inhibitor while PDE␦ prefers Arl2 and Arl3 in the active form with GTP bound. As small GTPases, ARF proteins participate in a variety of intracellular transport process. Arl2 and Arl3 may interact with PDE␦ to target a set of prenylated proteins to their destination membrane. The localization of PDE␦ in the cytoplasm of cells (Fig. 2) supports this hypothesis. Thus, one function of PDE␦ may be that of a soluble transport factor interacting with a large number of prenylated proteins steered by a number of factors like Arl proteins. Protein trafficking is particularly important for photoreceptors where high turnover of outer segments housing the phototransduction machinery requires unusually active protein transport. In ϳ100 million photoreceptors present in the human retina (see webvision. med.utah.edu/facts.html), an entire outer segment must be replaced once per week, an enormous task for protein transport and correct membrane targeting. Since many of the membraneassociated proteins participating in phototransduction are prenylated (rod and cone PDE, GRK1/7, T␥ subunit), it is conceivable that PDE␦, being more abundant in photoreceptors than in other cell types, may exert a prominent role in membrane targeting of prenylated phototransduction components in photoreceptors. Consistent with this model, preliminary experiments with a PDE␦ knock-out mouse 2 suggest that in the absence of PDE␦ GRK1 remains mostly in the inner segment where biosynthesis occurs and is not transported to its outer segment destination.