The δ Subunit of Retinal Rod cGMP Phosphodiesterase Regulates the Membrane Association of Ras and Rap GTPases*

Post-translational modifications of GTPases from the Ras superfamily enable them to associate with membrane compartments where they exert their biological activities. However, no protein acting like Rho and Rab dissociation inhibitor (GDI) that regulate the membrane association of Rho and Rab GTPases has been described for Ras and closely related proteins. We report here that the δ subunit of retinal rod phosphodiesterase (PDEδ) is able to interact with prenylated Ras and Rap proteins, and to solubilize them from membranes, independently of their nucleotide-bound (GDP or GTP) state. We show that PDEδ exhibits striking structural similarities with RhoGDI, namely conservation of the Ig-like fold and presence of a series of hydrophobic residues which could act as in RhoGDI to sequester the prenyl group of its target proteins, thereby providing structural support for the biochemical activity of PDEδ. We observe that the overexpression of PDEδ interferes with Ras trafficking and propose that it may play a role in the process that delivers prenylated proteins from endomembranes, once they have undergone proteolysis and carboxymethylation, to the structures that ensure trafficking to their respective resident compartments.

GTPases of the Ras superfamily, including the Ras, Rho/Rac/ Cdc42, and Rab families, play a central role in the control of numerous essential biological functions such as proliferation, differentiation, cell morphology, movement, intracellular trafficking, and gene expression (1)(2)(3). Most Ras-related proteins are ubiquitously expressed and have been conserved through evolution. They bind GDP and GTP with high affinity and carry an intrinsic GTPase activity. Regulatory proteins such as GEFs 1 or GTPase-activating proteins that, respectively, stimulate their ability to exchange GDP for GTP, or to hydrolyze GDP, enable them to act as molecular switches between their inactive GDP-bound and active GTP-bound forms in response to various extracellular stimuli and exert their biological activities (4).
Following their synthesis as soluble precursors, proteins of the Ras superfamily become associated with intracellular membrane compartments via post-translational modifications of their COOH termini (5). As it was shown for Ras many years ago, the association of these proteins with membranes is essential for their biological activity (6). Proteins of the Ras and Rho/Rac/Cdc42 families contain a CAAX sequence (where C is a Cys, A an aliphatic residue, and X any amino acid) at their COOH-terminal extremity that targets them for their posttranslational modifications. The first step consists in prenylation in the cytosol of the Cys of the CAAX motif by a C-15 farnesyl moiety, in the case of Ras and Rap2A, or by a C-20 geranylgeranyl group for most other GTPases (7). Prenylation addresses these proteins to endomembrane compartments (the endoplasmic reticulum and Golgi) (8), which correspond to the intracellular localization of the prenyl CAAX protease that removes the three COOH-terminal AAX residues, and the carboxymethyl transferase that modifies the now COOH-terminal prenyl-cysteine (9 -12). Other cellular proteins that carry a COOH-terminal CAAX sequence are processed by the same mechanisms; this group comprises proteins of the nuclear envelope (lamins A and B), the ␥ subunits of heterotrimeric G proteins, as well as several proteins involved in phototransduction such as rhodopsin kinase, the ␣ subunit of transducin and both ␣ and ␤ catalytic subunits of the cGMP phosphodiesterase from retinal rods and cones (7).
Upstream of the CAAX motif, farnesylated GTPases such as Ras carry a second signal consisting of a stretch of basic residues (as is the case for the K-Ras(4B) protein) or additional cysteines serving as palmitoylation sites (one for N-Ras or two for Ha-Ras) that are necessary for the association of these proteins with the plasma membrane (13). Recent evidence demonstrates that these signals also play a role in the trafficking of newly synthesized Ras proteins to the plasma membrane, as well as in their targeting to membrane subdomains. Palmitoylation addresses Ha-and N-Ras proteins to cholesterol-rich rafts via the exocytotic pathway, whereas K-Ras(4B) proceeds to the non-raft plasma membrane via a yet uncharacterized route (8,14,15).
Despite these new advances, the question of how the newly prenylated, and therefore hydrophobic yet still cytosolic, proteins transit in the cytosol to reach membranes remains open. Newly synthesized Rab proteins are escorted to their cognate membrane compartment, after prenylation by Rab-geranylgeranyltransferase, by Rab Escort Protein, a subunit of the enzyme specifically dedicated to this task (16). Rho-and Rab-GDIs, respectively, enable Rho and Rab GTPases to shuttle between membrane compartments through the cytosol due to their the ability to solubilize processed proteins from mem-* This work was supported in part by a grant from the Association pour la Recherche contre le Cancer. 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.
§ Recipient of fellowships from the Ministère de l'Education Nationale et la Rechercche and from the Fondation pour la Recherche Médicale.

EXPERIMENTAL PROCEDURES
Expression Constructs-Full-length human PDE␦ cDNA was obtained by screening a Jurkat T cell cDNA library in the yeast two hybrid system with the entire coding sequence of Rap2B as a bait; yeast two-hybrid experiments were performed as described with GTPase baits fused to the COOH terminus of the GAL4 DNA-binding domain, and preys (PDE␦ and effector controls) expressed from pGAD vectors (20). pRK5-PDE␦, the coding sequence for the human PDE␦ protein, was amplified by PCR using Pfu DNA polymerase (Stratagene) with oligonucleotides 5Ј ATTAGGATCC GCCATGTCAG CCAAGGACGA GCGGG 3Ј and 5Ј TCACGTCGAC TCAAACATAG AAAAGTCTCA CT 3Ј as forward and reverse amplimers, respectively. The resulting product was digested with BamHI and SalI and cloned into the pRK5 expression vector cut with the same restriction enzymes; pRK5myc-PDE␦ encoded an epitope recognized by the 9E10 anti-Myc antibody NH 2 -terminal of the PDE␦ coding sequences (20). PDE␦ cloned into the bacterial expression vector pET15b (Novagen) (21) and into the baculovirus transfer vector pFastBacHT (Invitrogen) was a generous gifts of A.-M. Marzesco and A. Zahraoui. pEXV-HRas was generated by inserting the Ha-Ras coding sequence into the SmaI site of the pcEXV-3 vector as described in Ref. 9. Vectors encoding GFP fused to wild type (pGFP-NRas) or C181S (pGFP-NRasC181S) and wild type K-Ras(4B) (pGFP-KRas) were generous gifts from E. Choy and M. Philips (8); those expressing GFP fused to or the COOH terminus of either wild type (pGFP-HRasCt) or the double mutant C181S,C184S Ha-Ras (pGFP-HRasCtC181SC184S) (15) were generous gifts from J. Hancock. The absence of mutation in all constructs was verified by DNA sequencing.
Production of His-PDE␦ Protein and Antibodies-Recombinant bacterial (His) 6 -PDE␦ was produced in Escherichia coli BL21(DE3) transformed with pET15b-PDE␦ and purified on nickel-chelating beads as indicated (21); antibodies were raised in rabbits against the recombinant protein.
A recombinant baculovirus encoding (His) 6 -PDE␦ was generated from pFastBacHT-PDE␦ using the Bac-to-Bac protocol (Invitrogen). Sf21 cells were grown in SF900II/TC100 (V/V) medium in spinner suspensions to a density of 2.5 10 6 cells/ml and were infected with recombinant baculovirus for 72 h. The infected cells were harvested by centrifugation, washed in phosphate-buffered saline (PBS), and lysed by a 1-h incubation at 4°C in PBS containing 1% CHAPS, 1 mM phenylmethylsulfonyl fluoride, 1 g/ml leupeptin, 1 g/ml pepstatin, and 1 g/ml aprotinin. After centrifugation at 100,000 ϫ g for 45 min at 4°C, the supernatant was incubated for 1 h with nickel-nitrilotriacetic acid resin (Qiagen); the resin was washed with PBS containing 20 mM imidazole, and PDE␦ was eluted with PBS containing 200 mM imidazole. The protein was concentrated in a Centricon YM-10 (Millipore) and further purified on a Superdex 75 gel filtration column. Fractions containing PDE␦ were pure as estimated by SDS-PAGE; they were concentrated as above, snap-frozen in liquid N 2 , and stored at Ϫ80°C.
Solubilization of GTPases from Membranes in Vitro-HeLa cells were transfected by electroporation with 10 g of pRK5-HRas vector and further cultured for 36 h; membranes were prepared from transfected and untransfected HeLa cells as follows. Cells were lysed in hypotonic buffer containing 25 mM Hepes, pH 7.5, and protease inhibitors by 100 strokes of the tight-fitted pestle of a Dounce homogeneizer. The post-nuclear supernatant obtained after centrifugation at 1000 ϫ g for 3 min at 4°C was submitted to ultracentrifugation at 100,000 ϫ g for 30 min in a Beckman TLA 45 rotor; the membrane pellet was washed once in the same buffer containing 0.1 M NaCl. Membranes were loaded with GDP or Gpp(NH)p for 15 min at 30°C in an exchange buffer containing 25 mM Tris-HCl, pH 7.5, 50 mM NaCl, 1 mM dithiothreitol, and 2 mM EDTA in the presence of 0.1 mM GDP or Gpp(NH)p and centrifuged at 100,000 ϫ g for 30 min at 4°C. Membranes (400 g of protein) were resuspended and incubated for 30 min at 30°C in a solubilization buffer containing 25 mM Tris-HCl, pH 7.5, 50 mM NaCl, 1 mM dithiothreitol, 5 mM MgCl 2 , and 0.1 mM GDP or Gpp(NH)p in the presence of 1 g of (His) 6 -PDE␦ protein purified from insect cells. Solubilized proteins were recovered by centrifugation at 100,000 ϫ g for 30 min and detected by Western blotting using rabbit affinity purified anti-Rap1 (22) and anti-Rap2 (23) antibodies. Mouse monoclonal antibodies directed against Ras, and rabbit polyclonal anti-RalB antibodies, were from Transduction Laboratories.
Overexpression of the PDE␦ Protein-3 ϫ 10 7 HEK293 cells were electroporated with 10 g of pRK5myc-PDE␦ or empty pRK5 vectors. 36 h later, cells were lysed, and membranes were prepared as above. 75 g of protein from the cytosol and membranes were analyzed by Western blotting using rabbit affinity-purified anti-Rap1 (22), anti-Rap2 (23), anti-Ras (24), and mouse monoclonal 9E10 anti-Myc (Roche Molecular Biochemicals) antibodies. When indicated, the cytosol was adjusted to 1% Triton X-114, and the detergent-enriched phase was recovered by centrifugation as described below. Protein was precipitated with trichloroacetic acid and detected by Western blotting.
Effects of PDE␦ on the Post-translational Processing of Ras-HeLa cells were electroporated as above with 4 g of pEXV-Hras and 6 g of pRK5-mycPDE␦ or pRK5myc vectors. 15 h later, they were incubated for 1 h in Dulbecco's modified Eagle's medium without methionine and cysteine supplemented with 5% of fetal calf serum and pulse-labeled for 20 min with 0.1 mCi/ml 35 S-labeled Protein Labeling Mix (PerkinElmer Life Sciences) in the same medium. Cells were then washed twice with PBS and chased for various amounts of time in complete culture medium. At the indicated times, they were lysed by a 10-min incubation on ice in a buffer containing 25 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 1% Triton X-114. After a 2-min centrifugation at 1000 ϫ g to eliminate large debris, the supernatants were warmed for 2 min at 37°C and centrifuged for 2 min at 13,000 ϫ g at room temperature to separate the aqueous phase from the detergent-enriched phase. The aqueous phase was then adjusted to 1% Triton X-114 and Ras proteins were immunoprecipitated from both phases by an overnight incubation with rat anti-Ras monoclonal antibody Y13-238 at 4°C. Complexes were recovered with goat anti-rat antibodies bound to protein A-Sepharose, extensively washed, and analyzed by SDS-PAGE followed by fluorography.
Similarity Searches and Protein Modeling-Searches within nonredundant data base (NR) were performed using PSI-BLAST at NCBI (25). Guidelines for the use of hydrophobic cluster analysis (HCA), which was used to refine alignments, are given in (26). Modeling of the PDE␦ structure was performed using Modeler-4 (27). Visualization of three-dimensional structures was done using SwissPDBviewer (www.expasy.ch/spdbv/mainpage.html).

The PDE␦ Protein Interacts with GTPases Containing an
Intact COOH Terminus-We had previously isolated PDE␦ in yeast two-hybrid screens with several GTPases of the Ras superfamily as baits (20,28). As depicted in Table I, PDE␦ interacted with Ras and Rap proteins, but not with Ral (neither RalA nor RalB), and it also interacted with Rho family members RhoA, RhoB, and Rnd1. Analysis of various mutants of these GTPases revealed that PDE␦ interacted as efficiently with full-length wild type, activated (Ras Val 12 , Rnd1 Val 12 , and RhoA Val 14 ) or dominant negative (Rap1A Asn 17 , Rap2B Asn 17 , and Ras Ala 15 ) forms of the GTPases. In contrast, PDE␦ only interacted with those proteins containing an intact COOH terminus, irrespective of other mutations that may also affect their nucleotide binding properties: Table I shows that it neither interacted with proteins carrying a COOH-terminal truncation, nor with the Ras mutant Val 12 3 Ser 186 where the Cys target of farnesylation has been replaced by a Ser.
Solubilization of Ras Family GTPases by PDE␦-Previous reports had shown that purified PDE␦ protein was able to solubilize the two ␣ and ␤ catalytic subunits of retinal rod cell cGMP phosphodiesterase, as well as the Rab13 GTPase from membranes in vitro (19,21); we sought to determine whether PDE␦ could exert a similar activity on Ras family GTPases. Indeed, Fig. 1 shows that recombinant PDE␦, purified from a baculovirus/insect cell expression system, extracted a significant proportion of endogenous Rap1 and Rap2, as well as ectopically expressed Ha-Ras proteins from the membranes of HeLa cells. Its activity was not dependent on the nucleotide (GDP-or GTP-bound) state of the GTPases, consistently with the interaction properties depicted in Table I. However, membrane extraction by PDE␦ was selective among GTPases, since it neither affected the Ras family GTPase RalB (Fig. 3), nor the more distantly related Rab6 protein (Ref. 21 and results not shown).
Numerous previous studies in mammalian cells have established that at steady state, nearly all of the endogenous Ras and Rap proteins are membrane-bound. We investigated whether the overexpression of PDE␦ would affect the membrane association of Ras family proteins in vivo. Fig. 2A shows that the ectopic overexpression of PDE␦ in HEK293 cells indeed created a sizeable cytosolic pool of Rap1, Rap2 proteins; a similar effect, albeit to a lesser extent, was also observed with Ras.
Overexpression of PDE␦ Does Not Interfere with the Posttranslational Processing of Ras Family GTPases-Since our results show that the overexpression of PDE␦ affected Rap and Ras localization at steady state, one possibility is that this could have resulted from an interference with the post-trans-lational processing of Ras family GTPases, resulting in the cytosolic accumulation of immature proteins. We evaluated processing by the ability of prenylated Ras and Rap proteins to partition into the detergent-rich phase of the detergent Triton X-114, as well as to migrate faster by SDS-PAGE (9, 23). Fig.  2B shows that the overexpression of PDE␦ resulted in the appearance in the cytosol of a form of Rap1 that partitioned in the detergent-rich phase of Triton X-114; moreover, the electrophoretic mobility of cytosolic Rap and Ras proteins was the same as for the membrane-associated proteins ( Fig. 2A). In addition, we investigated the effects of PDE␦ overexpression on the kinetics of Ha-Ras processing in a pulse-chase experiment (Fig. 3). As expected, newly synthesized Ha-Ras protein was hydrophilic, and by 3 h of chase the vast majority of the protein was processed; overexpression of PDE␦ did not alter in any Hf7c yeast expressing Ras superfamily proteins fused to the COOH terminus of the DNA-binding domain of GAL4 were mated with Y187 yeast expressing potential partners fused to the COOH terminus of the activation domain of GAL4. The capacity of diploids to grow in the absence of histidine was scored. Similar results were obtained by measuring ␤-galactosidase activity. ND, not done; fl, full-length; tr, COOHterminal truncation at residue 168; wt, wild type. Rap2A was systematically truncated at residue 168, to avoid the very strong transactivation exhibited by the full-length protein alone. As a positive control, we checked that Ras and Rap proteins were able to interact with the Ras/Rap binding domain of Ral-GDS (20) and that RalA and RalB were able to interact with their effector protein RLIP (not shown). DNA In vitro solubilization of Ras and Rap GTPases from membranes. Membranes were prepared from untransfected HeLa cells (for the analysis of Rap1, Rap2, and RalB) or cells that had been transfected with a pRK5-HRas expression vector. They were loaded with GDP or GppN(H)p and then incubated for 30 min at 30°C with recombinant (His) 6-PDE␦ (ϩ lanes) or buffer alone (Ϫ lanes). Solubilized material was then recovered by ultracentrifugation. GTPases were detected by Western blotting with rabbit affinity-purified anti-Rap1 and anti-Rap2 antibodies, as well as mouse monoclonal antibodies directed against Ras. Lanes labeled T represent the membranes prior to incubation with buffer or PDE␦.

FIG. 2.
Overexpression of the PDE␦ protein in HEK293 cells creates a soluble pool of GTPases. HEK293 cells were transfected with pRK5-myc or pRK5-mycPDE␦ constructs. 36 h later, they were lysed and the particulate (Mb) and cytosolic (Cyt) fractions were isolated. A, 75 g of protein from each fraction were separated by SDS-PAGE. GTPases and PDE␦ protein were detected by Western blotting with rabbit affinity-purified anti-Rap1, anti-Rap2, and anti-Ras antibodies and mouse anti-Myc 9E10 antibodies. B, the cytosolic fraction from cells overexpressing or not PDE␦ was adjusted to 1% Triton X-114, and material partitioning in the detergent phase was analyzed by Western blotting for the presence of Rap1. significant way the extent or kinetics of Ha-Ras processing.
PDE␦ Creates a Cytosolic Pool of Ras GTPases during Their Trafficking to the Plasma Membrane-We next investigated whether PDE␦ might play a role in the trafficking process that leads newly synthesized cytosolic Ras proteins to their final destination, the plasma membrane, by following the subcellular localization of ectopically expressed GFP-Ras fusion proteins; to this end, we used full-length N-and K-Ras(4B) proteins (GFP-NRas and GFP-Kras, respectively), as well as the COOH-terminal region of Ha-Ras fused to the COOH terminus of GFP (GFP-HRasCt). As described previously (8,15), GFP-Ras fusion proteins initially associated with endomembranes in a perinuclear compartment (not shown). At this step, Ras proteins have undergone farnesylation, proteolytic removal of the three COOH-terminal residues, and carboxymethylation of the COOH-terminal farnesylcysteine (8). After 13 h of expression, GFP-Ras proteins still labeled endomembrane compartments, and staining of the plasma membrane became clearly visible in control cells (Fig. 4, b, f, and j). In contrast, cells overexpressing PDE␦ exhibited a strong cytosolic GFP-Ras labeling in addition to the plasma membrane signal (Fig. 4, d, h, and l). It is remarkable that this effect of PDE␦ was similar for Ha-and N-Ras proteins that are palmitoylated and transported to the plasma membrane via the exocytotic pathway, as well as for K-Ras(4B) that carries a polybasic sequence and proceeds to the plasma membrane via a yet uncharacterized route.
Since PDE␦ was able to act on the nonpalmitoylated K-Ras(4B) protein, we investigated whether it could also act on partially processed nonpalmitoylated Ha-and N-Ras proteins, at some intermediate step of their trafficking. To that end, we analyzed its effects on GFP fused with a mutant of N-Ras in which the cysteine target for palmitoylated had been mutated to serine; such a protein cannot reach the plasma membrane and accumulates in the endoplasmic reticulum as seen by its co-localization with PDI, a marker of this compartment (8) (see Fig. 5, a-c). Fig. 5, d and e, show that the overexpression of PDE␦ caused the bulk of this prenylated and nonpalmitoylated GFP-NRas fusion protein to accumulate in the cytosol, despite the fact that the endoplasmic reticulum remained intact (Fig.  5f). A similar result was obtained with a nonpalmitoylated mutant of the COOH-terminal region of Ha-Ras fused to GFP (data not shown).
PDE␦ Exhibits Striking Structural Similarities with RhoGDI-This property of PDE␦ to extract prenylated Ras family proteins is reminiscent of the well characterized action of RhoGDI and RabGDI proteins on Rho and Rab family GTPases, respectively. We therefore searched whether PDE␦ exhibited any similarity with these proteins using the PSI-BLAST program at NCBI. At convergence by iteration 2, similarities were observed with the RhoGDI sequence from fission yeast as well as with its human counterpart (24 and 20% identity over 78 amino acids, respectively, as indicated by the arrows in Fig. 6A). To investigate whether this observed sequence relationship could correspond to structural similarities, we completed the sequence analysis by using information pertaining to secondary structure organization (hydrophobic cluster analysis or HCA (26), not shown) as well as three-dimen- sional experimental observations available for RhoGDI, such as the structure reported for its complex with the Cdc42 GTPase (18). GDI proteins are organized in two domains (Fig.  6B, left). The COOH-terminal geranylgeranyl-binding domain folds into an immunoglobulin-like ␤-sandwich (blue) whose opposite ␤-sheets display a hydrophobic pocket in which the geranylgeranyl moiety of the Cdc42 GTPase inserts; this domain is responsible for the ability of RhoGDI to extract Cdc42 from membranes. An amino-terminal region (green), comprising a helix-loop-helix, is likely to be responsible for the inhibitory effect of RhoGDI on GDP dissociation and GTP hydrolysis by Cdc42 through interactions with the switch I and II regions.
A good correspondence was observed between PDE␦ and RhoGDI, in particular for the two last strands (H and I) and the loop between them (three conserved aspartic acids) (see Fig.  6A). HCA analysis enabled us to extend the similarity between PDE␦ and RhoGDI to the very NH 2 terminus of PDE␦, which noticeably corresponds to the NH 2 -terminal end of the RhoGDI immunoglobulin domain. Accordingly, PDE␦ would lack the NH 2 -terminal regulatory arm characteristic of RhoGDI.
Buried hydrophobic positions of the RhoGDI Ig-like domain, in particular those corresponding to core secondary structures (strands A, D, E, and H in Fig. 6, A and B), are conserved in PDE␦, strongly supporting a shared structure for the gera- nylgeranyl-binding domain of RhoGDI and PDE␦. Moreover, some particular characteristics of the RhoGDI Ig-like fold are also maintained in PDE␦, such as the presence of two additional small strands between strands A and D. Remarkably, most of the amino acids lining the geranylgeranyl-binding pocket of RhoGDI are conserved in PDE␦ (stars on Fig. 6A, violet side chains on Fig. 6B). Only amino acids contributing to the external part of the pocket and belonging the NH 2 terminus of RhoGDI (Leu 11 and Ile 14 ) are absent in PDE␦. Another GTPase-interacting region of RhoGDI is conserved in PDE␦, involving the conserved Asp 185 in the H-I loop (Asp 147 in PDE␦) whose side chain forms a hydrogen bond with the guanidinium group of Cdc42 Arg 66 in the Cdc42⅐RhoGDI complex (Fig. 6B).
In light of these relationships, we constructed a model for the three-dimensional structure of PDE␦ (Fig. 6B, right). This model proposes that PDE␦ should adopt the Ig-like fold of RhoGDI and retain its hydrophobic geranylgeranyl-binding pocket. It predicts that PDE␦ should retain the function of the Ig-like domain of RhoGDI, namely binding prenylated GTPases, allowing their release from cellular membranes; in contrast, the absence in PDE␦ of a domain corresponding to the of the NH 2 -terminal arm of RhoGDI should imply that the action of PDE␦ would be independent of the GDP-or GTPbound state of GTPases, in accordance with our experimental results (see Fig. 1).

DISCUSSION
In this manuscript, we show that the ␦ subunit of cGMP phosphodiesterase from retinal rod cells, PDE␦, which is ubiquitously expressed (19,21), can act in vitro as well as in vivo to solubilize Ras as well as the closely related Rap proteins from the membranes of mammalian cells. Proteins able to exert such an activity on GTPases from the Rho and Rab families, RhoGDI (18) and RabGDI (17) respectively, have already been identified. However this is the first time that such an activity has been described to act in vitro as well as in vivo on Ras and Rap proteins.
The sequence of PDE␦ is highly conserved among mammals, since the bovine, canine, and murine proteins only differ from the human one by one nonconservative (T68A) and zero to three conservative changes; a search for homologous proteins in invertebrate species revealed the existence of orthologues in Drosophila melanogaster and Caenorhabditis elegans, respectively, sharing 61 and 69% identity with human PDE␦ (Fig. 7A,  upper panel). Such a high evolutionary conservation strongly suggests that the function of PDE␦ has been conserved as well. Searches for the conservation of sequence and structural features of PDE␦ with other proteins revealed striking similarities with the established structure of RhoGDI, as visualized in its complex with the Cdc42 GTPase (18). This allowed us to build a model for the three-dimensional structure of human PDE␦ suggesting that it may also adopt a general Ig-like fold very similar to that of RhoGDI. More importantly, our model suggests that the pocket lined with hydrophobic residues in RhoGDI that binds the geranylgeranyl group of Cdc42 is likely to be conserved in PDE␦, hence providing structural support for our functional data showing that PDE␦ is able to extract prenylated proteins such as Ras family GTPases from membranes.
A further search for proteins closely related to PDE␦ revealed significant similarities with the UNC-119/RG4 group of proteins from mammals, zebrafish, C. elegans, and D. melanogaster (Fig. 7A, lower panel). These proteins, of yet unknown function, were identified on the basis of their high expression in the C. elegans nervous system as well as in mammalian photoreceptor cells (29,30). Besides an unrelated NH 2 -terminal region of 50 -90 residues (depending on the species), they exhibited significant sequence similarities with PDE␦ (47% sim-ilarity, 22% identity for the human proteins), extending through to their COOH termini. Alignment of UNC119/RG4, PDE␦, and RhoGDI of human origin revealed that the structural features conserved between PDE␦ and RhoGDI, namely the elements responsible for their Ig-like fold and the presence of hydrophobic residues at conserved positions, which could line the surface of a prenyl-binding pocket, are also conserved in UNC119/RG4 (Fig. 7B). Hence we propose that UNC119/ RG4, PDE␦, and RhoGDI define a novel family of proteins conserved through evolution, that interact with prenylated proteins, and whose function is to regulate their association with membranes.
One major difference between PDE␦ and RhoGDI is that the latter exhibits a strong functional preference for GDP-bound Rho GTPases (31,32), whereas PDE␦ is able to solubilize both the GDP-and GTP-bound form of Ras and Rap. Such a distinction may have important implications for the physiological processes involving the function of PDE␦, such as possibly the trafficking of prenylated proteins. The NH 2 -terminal domain of RhoGDI, by interacting with switch I and switch II regions of the Cdc42 GTPase, is responsible for this selectivity, as well as the inhibition of GDP dissociation and GTPase-activating protein-stimulated GTPase activity (18). The fact that there is no equivalent region in PDE␦ supports our data that it extracts GTPases from membranes independently of the nucleotide (GDP or GTP) bound and predicts that PDE␦ should neither affect the rate of nucleotide binding and dissociation, nor the GTPase activity of Ras family proteins. UNC-119/RG4 proteins present an NH 2 -terminal domain, upstream of the Ig-like core, that exhibits the same length and low content in hydrophobic residues as RhoGDI (not shown). By analogy, it is tempting to propose that this region could play a role in the selectivity of UNC119/RG4 proteins for their prenylated targets.
Another distinctive feature between PDE␦ and RhoGDI is the large size and sequence divergence of the loop connecting ␤-sheets B and C (Fig. 6, A and B). This region is highly exposed in both structures and could be involved in their interaction with other nonprenylated proteins. Indeed, PDE␦ has been reported to interact with the retinis pigmentosa GTPase regulator (33) and the Arf-like protein Arl3 (34). However the role of such interactions, and possible positive or negative interference with the membrane-extracting ability of PDE␦, are presently unknown.
RhoGDI and PDE␦ exhibit a very different spectrum of activity: while RhoGDI is only active on GTPases of the Rho family, PDE␦ is able to exert its membrane extraction activity on a wide variety of proteins, all of which are prenylated on their COOH-terminal extremity (Table II). Indeed, PDE␦ had previously been shown to be able to solubilize the ␣ and ␤ catalytic subunits from retinal rod cGMP phosphodiesterase (19), which are, respectively, modified by the C15 farnesyl and C20 geranylgeranyl isoprenoids, as well as the geranylgeranylated Ras-related Rab13 protein (21). We additionally established that PDE␦ is active on Ras and Rap proteins of the Ras family and that it is also active on the Rho family protein RhoA (data not shown). All of these proteins, including the Rab family protein Rab13, carry a COOH-terminal CAAX sequence that directs them to be modified by a single farnesyl or geranylgeranyl group; they are further processed by proteolysis of the three COOH-terminal residues, and carboxymethylation of the prenylcysteine, modifications that have been shown to be necessary for binding to PDE␦ in vitro (35). In contrast, Rab4 and Rab6, which, respectively, exhibit COOH-terminal CGC and CSC sequences and are modified by two geranylgeranyl groups, are not targets of PDE␦ (21). Surprisingly, we show that Ral proteins (RalA and RalB), which are closely related to Ras, cannot interact with PDE␦ in the yeast two-hybrid system (Table I) and that RalB is not extracted from membranes in vitro by recombinant PDE␦ (Fig. 1). Their lack of interaction with PDE␦ could indicate that they are not processed in vivo like other CAAX-containing proteins; this is, however, unlikely since their COOH-terminal sequences (CCIL for RalA and CCLL for RalB) are identical or closely related to those of the ␣ and ␤ catalytic subunits of retinal rod cGMP phosphodiesterase (CCIQ and CCIL, respectively, see Table II), which are indeed extracted from membranes by PDE␦ (19). Alternatively, it is possible that the COOH-terminal region of Ral could adopt a conformation that prevents its interaction with PDE␦; this point could be investigated experimentally by using Ras/Ral chimeric proteins.
Upstream from the CAAX sequence, some GTPases that are susceptible to the action of PDE␦ such as K-Ras(4B), Rap1A, and RhoA carry a polybasic sequence that has been shown in the case of K-Ras(4B) to address the newly synthesized protein via a yet undescribed trafficking pathway toward the nonraft plasma membrane (8,15). Others, such as Ha-Ras, N-Ras, Rap2 (A and B), and RhoB carry one or two palmitoylation sites upstream from the prenylation site (see Table II). Using mutants, we have shown that PDE␦ is active on the nonpalmitoylated forms of Ha-and N-Ras proteins. However, since palmitoylation, in contrast with prenylation, is not a stable modification in cells and exhibits a mean turnover time of less than 30 min (36), a significant proportion of cellular Ha-Ras, N-Ras and Rap2 proteins should be depalmitoylated at steady state. It is therefore conceivable that in our experiments, PDE␦ only extracted nonpalmitoylated Ras and Rap2 proteins. We attempted to address this question directly by comparing the ability of recombinant PDE␦ to extract wild type and nonpalmitoylated mutants of Ha-and N-Ras fused to GFP from the membranes of transfected HeLa cells; however, due to the relative instability of the mutant proteins, we were unable to perform comparative experiments. In the case of RhoGDI, the reported structure for the geranylgeranyl-binding pocket is unlikely to accommodate a nearby palmitate group, and a re-FIG. 7. PDE␦, RhoGDI, and UNC119/RG4 form a novel protein family. A, the alignment among PDE␦ (upper part) and UNC119/RG4 proteins (lower part) from various species was deduced from the PSI-BLAST results and refined manually, in particular using HCA, for the most NH 2 -terminal region. Identities are shown as white on a black background, similarities are shaded gray (white and black letters for hydrophobic and nonhydrophobic amino acids, respectively). Sequences that could not be aligned are indicated by number of residues; they correspond to the variable loop included between RhoGDI strands B and C depicted in Fig. 6A. A highly conserved area is observed around the FGF motif, corresponding to strand F in Fig. 6A, which contains critical residues of the hydrophobic pocket. Identifiers are those of the Swissprot data base (CNRD corresponds to PDE␦), except for P17/HUMAN (GenBank TM identifier: 2695883), CG9296/DROME (GenBank TM identifier: 7297442), and U119b/BRARE and U119c/BRARE (GenBank TM identifiers: 14587076 and 14587078). The total length of each sequence is indicated within brackets. BRARE, Danio rerio (zebrafish); CAEEL, C. elegans; CAEBR, Caenorhabditis briggsae; CANFA, Canis familiaris; DROME, D. melanogaster. B, alignment of human RG4, PDE␦, and RhoGDI proteins. Annotations are the same as in Fig. 6A. cent report establishes that a palmitoylation site inserted into the COOH-terminal sequence of RhoA blocks RhoGDI binding (32). Given its structural similarities with RhoGDI, it is possible that PDE␦ is only active on the nonpalmitoylated form of Ras family proteins.
The effects of PDE␦ overexpression on Ras trafficking provide some insight into the possible physiological role of this ubiquitously expressed protein. Contrarily to the case of Rho and Rab proteins, there is no significant cytosolic pool of Ras and Rap proteins, which indicates that the main function of PDE␦ is probably not to regulate the level of membrane bound GTPases available for signal transduction. Following their synthesis as cytosolic precursors and prenylation in the cytosol, Ras proteins associate with endomembranes such as the endoplasmic reticulum and Golgi which constitute the site of their proteolysis and carboxymethylation (8). One possibility is that PDE␦ binds the newly prenylated proteins in the cytosol and escorts them to the endomembranes. This is, however, unlikely, since PDE␦ only binds prenylated proteins after they have undergone proteolysis and carboxymethylation (35). Moreover, a recent report suggests that the prenylated Rab acceptor protein PRA1, which interacts and co-localizes with Ha-Ras and RhoA proteins in the Golgi compartment (37) could actually fulfill that role, i.e. escort newly prenylated proteins from the cytosol to endomembanes. Observation of later trafficking steps shows the overexpression of PDE␦ induces the formation of an important cytosolic pool of Ras proteins, but that some Ras proteins nevertheless reach the plasma membrane under those conditions (see Fig. 4). We have shown that the same effect of PDE␦ overexpression is observed with all three Ras proteins, despite their distinct trafficking pathways from endomembrane compartments to the plasma membrane (15). We therefore surmise that PDE␦ should intervene at some common step along this pathway and propose that it may play a role in the process that delivers prenylated proteins from endomembranes, once they have undergone proteolysis and carboxymethylation, to the structures that ensure trafficking to their respective resident compartments. II PDE␦ interaction and posttranslational modification of target proteins Prenylated cysteines are in red, palmitoylated cysteines in green and underlined, and polybasic sequences in blue and underlined. F, farnesyl; GG, geranylgeranyl; palm, palmitoylation sites; memb extraction, extraction from membranes; overexpr, experiment performed with ectopically expressed GTPase; GFP, microscopic analysis of GFP fusion proteins; WB, Western blot assessment of membrane extraction by purified PDE␦ endog, in vitro extraction from membranes of endogenous GTPases by recombinant PDE␦ (Fig. 1); nd, not done. a In the case of RalA and RalB, the possibility that both cysteines of the CCXL motif could be prenylated is hypothetical. b Palmitoylation of both cysteines has been demonstrated for Ha-Ras and remains hypothetical for Rap2 (A and B) as well as for RhoB and Rndl. c Due the absence of sensitive and specific antibodies that discriminate between simultaneously expressed Ras and Rap isoforms, biochemical extraction of the endogenous protein from membranes by recombinant PDE␦ in vitro may concern one or more isoforms.