Rab Proteins Form in Vivo Complexes with T w o Isoforms of the GDP-dissociation Inhibitor Protein (GDI)*

GTPases of the Rab family play a key role in the regu- lation of vesicular transport in eukaryotic cells. Several accessory proteins that regulate their GDPIGTP cycle as well as their subcellular localization have been identified within the past few years. The best known is Rab3A GDP dissociation inhibitor protein (GDI), originally identified as an inhibitor of GDP dissociation from RabSA, a Rab protein specifically expressed in neuronal and neuroendocrine cells. Recent studies have pointed out a role of Rab3A GDI as a chaperone of several Rab proteins during their cycling between cytosol and membranes and Rab3A GDI has been considered so far as a general regulator of Rab function. However, cDNAs encoding potential isoforms of this protein, called GDI fl and GDI-2, have been recently isolated. In this study, we have characterized cytosolic Rab protein complexes in various cell types and tissues using Mono Q chromatography. We show that in rat brain and in insulin-secreting RINm5F cells, the majority of Rab proteins are complexed with Rab3A GDI. In contrast, in Chinese hamster ovary cells, they are mainly complexed to a protein that we have identified as GDI p. In rat liver cytosol, Rab proteins form complexes with both isoforms. We also show that the proportion of Rab

GTPases of the Rab family play a key role in the regulation of vesicular transport in eukaryotic cells. Several accessory proteins that regulate their GDPIGTP cycle as well as their subcellular localization have been identified within the past few years. The best known is Rab3A GDP dissociation inhibitor protein (GDI), originally identified as an inhibitor of GDP dissociation from RabSA, a Rab protein specifically expressed in neuronal and neuroendocrine cells. Recent studies have pointed out a role of Rab3A GDI as a chaperone of several Rab proteins during their cycling between cytosol and membranes and Rab3A GDI has been considered so far as a general regulator of Rab function. However, cDNAs encoding potential isoforms of this protein, called GDI fl and GDI-2, have been recently isolated. In this study, we have characterized cytosolic Rab protein complexes in various cell types and tissues using Mono Q chromatography. We show that in rat brain and in insulin-secreting RINm5F cells, the majority of Rab proteins are complexed with Rab3A GDI. In contrast, in Chinese hamster ovary cells, they are mainly complexed to a protein that we have identified as GDI p. In rat liver cytosol, Rab proteins form complexes with both isoforms. We also show that the proportion of Rab proteins complexed with either isoform depends on the relative abundance of Rab3A GDI and GDI p in the cytosol. These findings suggest that GDI isoforms are either redundant or could be involved in the fine control of Rab function.
Over 30 members of the ras-related Rab GTPase family have been identified within the past few years in a wide variety of eukaryotic cells, including yeast Saccharomyces cerevisiae (Sec4 and Ypt proteins) (for review, see Refs. 1 and 2). Being associated with distinct compartments of both the biosynthetic/ secretory and the endocytic pathways, Rab proteins are thought to play an important role in the regulation of vesicular traffic. They cycle between a cytosolic "inactive" (GDP-bound) form and a membrane-associated "active" (GTP-bound) form. This cycle could specify accurate docking and/or fusion of transport vesicles with their acceptor membranes (for review, see * This study was supported in part by grants from the Human Frontier Science Program (RG-380/92), the Universite Pierre et Mane Curie, and an Economic European Community Concerted Action Grant BI02-CT92-0205. C. Yang and V. I. Slepnev contributed equally to this work. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "aduertisement" in  Refs. [3][4][5]. An unresolved question is to know how Rab proteins fulfill their function in the context of proteins such as the v-and "ARES recently shown to be involved in the docking/fusion machinery (6,7).
Several accessory proteins which regulate the nucleotide state of Rab proteins and their subcellular localization have been identified. They include the mammalian protein Mss4 and the yeast protein Dss4 which exhibit guanine-nucleotide exchange factor-like activities (8)(9). A putative smg p25AIRab3A guanine-nucleotide exchange factor has also been identified in brain cytosol (10). Less is known about the proteins that might stimulate the very low intrinsic GTPase activity of Rab proteins. Although Rab GTPase-activating protein-like activities have been detected in various cell or tissue extracts (10, 111, only Gyp 6 which encodes for a GTPase-activating protein specific for Ypt6, the yeast homolog of Rab6, has been cloned (12). Another protein, rabphilin, which shares homology with the synaptic vesicle integral membrane protein synaptotagmin, can inhibit GTPase activating protein-stimulated GTPase of Rab3A, but its precise function is still unclear (13).
So far, the best characterized Rab accessory protein is Rab3A GDI.' This protein was originally discovered as an inhibitor of the release of GDP from the isoprenylated form of smg p25N Rab3A, a Rab protein preferentially expressed in neuronal and neuroendocrine cells (14). Rab3A GDI forms cytosolic complexes with Rab3A and in vitro inhibits the binding to and promotes the dissociation of the GDP-bound form of Rab3A from membranes (15,16). In contrast to the cellular specificity of Rab3A, Rab3A GDI mRNA is expressed in all tissues, although at lower levels than in the brain (17). In addition, Rab3A GDI is not specific for this protein as it has also been shown to inhibit GDP release from Sec4 and Rabll and to extract a broad range of Rab proteins from membranes (18-20). Furthermore, cytosolic complexes formed between Rab proteins and Rab3A GDI have been detected in various cell types, including insulin-secreting RINm5F and Chinese hamster ovary (CHO) cells (16,21).
Based on the above results, it has been suggested that Rab3A GDI acts as a general regulator of Rab function (20). Its role would be to both inhibit GDP release and to act as a chaperone of Rab proteins during their cycle between cytosol and membranes. However, two new cDNAs coding for GDI-like proteins have been very recently cloned from a rat brain library and a mouse skeletal muscle library, referred to as GDI p and GDI-2, respectively (22,23). GDI p has been shown to inhibit GDP release from Rab3A and Rabll and GDI-2 can extract Rab4 and Rab5 from membranes (22,23). These findings then raise the question of whether differences in function between Rab3A The abbreviations used are: GDI, GDP dissociation inhibitor protein; QFF, Sepharose Q Fast Flow; EGS, ethylene glycol-bis(succinic acid N-hydroxysuccinimide ester); CHAPS, 3-[(3-cholamidopropyl)dimethylammoniol-l-propanesulfonic acid; CHO, Chinese hamster ovary.

Rub Proteins Complexed with GDI Isoforms
GDI and these GDI-like proteins exist. Cells-CHO cells were routinely grown on Petri dishes in a-minimal essential medium (TechGen, France) containing 7.5% fetal calf serum (Biological Industries, Israel) and antibiotics (100 IU penicillin and 100 pg of streptomycixdml). To prepare large amounts of cytosol, CHO cells were grown in roller flasks. RINm5F cells (a gift of Dr. Claes Wollheim, Geneva, Switzerland) were grown in suspension in spinner flasks (Bellco) in RPMI 1640 medium (ICN) supplemented with 1% newborn calf serum, 25 mM Hepes, pH 7, and antibiotics.
To construct cells expressing myc epitope-tagged Rab6, CHO cells were cotransfected with myc-tagged human Rab6 cDNA (24) inserted into the pKC3 vector (a gift of Dr. Jean-Pierre Abastado, Institut Pasteur) and with the pSV2-histidinol plasmid (a gift of Dr. Bernard Malissen, Marseille-Luminy, France). Transfected cells were selected in culture medium containing 7.5 mg/ml histidinol and cloned by limit dilution 2 weeks after transfection. The clone that was used in these experiments expressed myc epitope-tagged Rab6 at about &fold the endogenous level of Rab6, as determined by immunoblotting (data not shown).
Antibodies-Rabbit antibodies raised against RablA, Rab2, Rab4A, and Rab6 proteins were described elsewhere (25, 26). Rabbit anti-Rab3A GDI antibody (raised against a synthetic COOH-terminal peptide of Rab3AGDI) was kindly provided to by Dr. Marino Zerial (EMBL, Germany) and the rabbit antibody against Rab3A was a gift of Dr.
F r a q o i s Darchen (IBPC, Paris). We generated for this study a rabbit antibody against purified histidine-tagged Rab3A GDI (a gift of Dr. Marino Zerial). This antibody was shown to cross-react with GDI /3.
Mouse monoclonal antibody 9E10, reactive with the epitope EQKLI-SEEDLN derived from the c-myc proto-oncogene product (27), was used to detect myc epitope-tagged Rab6. Preparation of Cytosols-CHO and RINm5F cells were trypsinized when grown on Petri dishes or collected by centrifugation when grown in suspension and washed twice i n phosphate-buffered saline. All subsequent steps were carried out at 4 "C. Cells were resuspended in Hepes buffer (30 mM Hepes, pH 7.5, 5 mM MgCI,, 1 mM EGTA, 1 mM phenylmethylsulfonyl fluoride, and a mixture of protease inhibitors consisting of 1 pg/ml final concentration of leupeptin, aprotinin, chymostatin, antipain, and pepstatin) and homogenized using a Barreltype homogenizer as described previously (25).
Mouse brains and rat livers were obtained immediately after slaughter, washed several times with ice-cold Hepes buffer, and then homogenized with a Dounce homogenizer. Cell or tissue homogenates were centrifuged for 10 min a t 600 x g to remove cell debris, nuclei, and aggregates and then for 1 h at 4 "C at 150,000 x g . High speed supernatants were considered as cytosols and the corresponding pellets a s membrane fractions.
Mono Q Chromatography-Cytosols were filtered through a 0.22-pm filter (Millipore) and then applied to a 1-ml Mono Q HR 5/5 column.
After washing with 2 ml of Hepes buffer, elution was performed with a linear gradient of NaCl(O-400 m~) in the same buffer at a flow rate of 0.4 mumin. 0.5-ml fractions were collected and analyzed for their content in small GTP-binding proteins and Rab proteins. Cross-linking Experiments-About 1 mg of cytosolic proteins from CHO, 0.1 mg of cytosolic proteins from mouse brain, and aliquots of the different fractions obtained after Mono Q chromatography were preincubated with 0.1 mM GDP for 15 min at 30 "C. 5 mM EGS (ethylene glycol bis([succinic acid N-hydroxysuccinimide ester)) (Sigma) was then added for 45 min at 30 "C. T h e reaction was stopped by the addition of 10 m~ monoethanolamine.
Purification of GDI p from CHO Cells-About 250 mg of cytosolic proteins were prepared from 3 x lo9 CHO cells grown in roller flasks.
This extract was loaded onto a 15-ml QFF column equilibrated in buffer A (30 m~ Hepes, pH 7.5, 5 mM MgCl,, 1 mM EGTA, and 0.1 mM phenylmethylsulfonyl fluoride). After washes with 30 ml of buffer A, proteins were eluted with 150 ml of a linear gradient of NaCl (0-400 mM). 5-ml fractions were collected. 50 p1 of each fraction were used to test their ability to extract myc epitope-tagged Rab6 from membranes (see above). Fractions 7-12 (corresponding to a NaCl concentration between 100 and 150 mM) were pooled and dialyzed overnight at 4 "C in buffer A containing 0.6% CHAPS. This material was then loaded onto a 5-ml QFF column equilibrated in buffer A containing 0.6% CHAPS. After washing with 10 ml of buffer A, proteins were eluted with 50 ml of a linear gradient of NaCl (0-500 mM). 2-ml fractions were collected and dialyzed overnight in buffer A. Fractions 10-14 (corresponding to a NaCl concentration between 100 and 150 mM) were found to be active in removing myc epitope-tagged Rab6 from membranes and pooled. After a 2-h dialysis against buffer B (20 mM Tris/HCl, pH 8.0, 2 mM MgCI,, and 0.6% CHAPS), they were applied to a 1-ml Mono Q column equilibrated in buffer B. The column was developed with 10 ml of a linear gradient of NaAc (0-750 mM). 0.4-ml fractions were collected and dialyzed in buffer A. The activity to extract myc epitope-tagged Rab6 from membranes was detected in fractions 4-11. After concentration on a Sartorius 10 filtration unit, 900 pl of pooled fractions were loaded onto a Superdex 75 column equilibrated with buffer A supplemented with 200 mM NaCI. The column was run a t 100 pl/min and 500-pl fractions were collected. The GDI activity was detected in fractions 5-7, corresponding to a molecular mass around 50 kDa. Further purification was achieved through two successive Mono Q chromatographies. The first one (Mono Q HR 5/51 was developed with 10 ml of a linear gradient of NaCl (0-250 mM). The second one (Mono Q PC 1.6/5) was developed with 1.8 ml of a linear gradient of NaCl (50-300 mM). 50-111 fractions were collected. Aliquots were then boiled in Laemmli's buffer and analyzed by SDS-PAGE. A 1-pl aliquot of each fraction was tested at the same time for their ability to remove myc epitope-tagged Rab6 from membranes. Peptide Mapping a n d Sequence Analysis-About 5 pg of partially purified GDI-like protein from CHO cytosol was subjected to 10% SDS-PAGE. After coloration with Amido Black (0.003%), the protein band was excised, extensively washed with distilled water, and digested with the lysine-specific endoproteinase overnight at 37 "C in the presence of 0.03% SDS. The digested protein was subjected to high performance liquid chromatography on a DEAE-C18 column. The column was developed with 2 4 5 % acetonitrile gradient in 0.05% trifluoroacetic acid. About 30 peptides were obtained and three of them were sequenced using an automated gas-phase sequencer.
SDS-PAGE, GTP Overlay Assay, a n d Immunoblotting-Proteins were fractionated on SDS-PAGE and electroblotted onto nitrocellulose. GTP overlay assay was performed as described previously (28). Rab3A GDI, GDI p, and Rab proteins were detected using specific antibodies and peroxidase-labeled anti-rabbit IgG antibody (ECL protocol), alkaline phosphatase-labeled anti-rabbit or anti-mouse IgG antibody (Promega), or '251-ProteinAas described previously (25). Specific bands were quantitated by scanning immunoblots and autoradiograms on a Master Scan (Bionis).
Release of myc Epitope-tagged Rab6 from Membranes-Membranes were prepared from CHO cells expressing myc epitope-tagged Rab6 as described above and washed once in Hepes buffer to eliminate contaminant cytosolic proteins. 7 pg of membrane proteins were then incubated with aliquots of the different fractions collected throughout the purification procedure of GDI p for 45 min at 30 "C in 25 mM Hepes, pH 7.5,2 mM MgCl,, 0.3 M sucrose, 0.2 mM GDP, and protease inhibitors. The samples were centrifuged a t 4 "C for 15 min at 150,000 x g in a TLlOO Beckman centrifuge. Supernatants were then boiled in Laemmli's sample buffer (291, electrophoresed, and blotted onto nitrocellulose. myc epitope-tagged Rab6 was revealed using the 9E10 antibody. Ituo-dimensional Gel Electrophoresis-A combination of isoelectric focusing and SDS-PAGE was used to resolve proteins in two dimensions as essentially described by O'Farrell (30) with the following minor modifications: for isoelectric focusing, a mixture of 1% Pharmalyte, pH 4 4 (Pharmacia) and 1% Servalyte, pH 6-8 (Serval was used; for the second dimension, 12% polyacrylamide-SDS gels were run. Bovine serum albumin was added in all samples as internal standard. Other Methods-Protein concentration was estimated by the method of Bradford using rabbit IgG as standard (31).

RESULTS
Rab proteins occur as complexes in cytosol, as detected by gel filtration or sucrose gradient sedimentation (16,21). It is generally thought that these complexes consist of Rab and Rab3A GDI proteins in a 1:l ratio. Since cDNAs encoding potential candidates for Rab3A GDI isoforms have been recently isolated (22,231, we decided to characterize in further detail these complexes in order to determine whether cytosolic Rab proteins are complexed with Rab3A GDI or with isoforms of this protein. For this purpose, we chose to analyze various cytosols prepared from different cell and tissue extracts by Mono Q chromatography. Small GTP-binding proteins present in the eluted fractions were detected by the GTP overlay assay. The presence in these fractions of several Rab proteins (RablA, Rab2, Rab3A, Rab4A, and Rab6) and that of Rab3A GDI was also monitored by using specific antibodies raised against these proteins. Under the experimental conditions used in this study, the majority (>95%) of the small GTP-binding proteins were retained on the Mono Q column (data not shown).

GDI in Brain Cytosol and to Another Protein in CHO Cytosol-
The chromatography of brain cytosol (Fig. lA) showed that the majority of small GTP-binding proteins, as well as most of RablA, Rab3A, Rab4A (data not shown), and Rab6 co-eluted with Rab3A GDI at an ionic strength of 250-350 mM NaCI. To test whether GTP-binding proteins and Rab proteins were complexed with Rab3A GDI in these fractions, they were cross-linked with EGS. After cross-linking, the majority of small GTP-binding proteins migrated with an apparent molecular mass of around 80 kDa (Fig. 1 B ) . The presence of Rab3A GDI in this band was revealed by immunoblot analysis (data not shown). The GTP overlay assay preferentially reveals proteins of the Rab and Ral families (32). In addition, Ral proteins do not interact with Rab3A GDI. The -80-kDa species then most likely correspond to cross-linked 1:l complexes formed between Rab3A GDI (which migrates on SDS gels with an apparent molecular mass of 55 kDa, see Fig. 3B) and Rab proteins (21-28 kDa). It should be pointed out that Rab proteins eluted from the Mono Q column in two peaks (Fig. lA). Most of RablA and Rab6, as well as of Rab4A (data not shown), were detected in the first one (fractions 11-13), whereas the majority of Rab3A was detected in the second peak (fractions 15-17). This elution profile can be explained by the fact that Rab3A is one of the most acidic Rab proteins (PI 4.7) as determined by two-dimensional gel electrophoresis (33). Complexes formed between Rab3A and Rab3A GDI are therefore expected to elute at a higher ionic strength than Rab3A GDI complexed with other members of the Rab family. This experiment confirms that Rab proteins are complexed with Rab3A GDI in brain cytosol, as has been previously documented (16). It also indicates that Rab proteins do not dissociate from Rab3A GDI at the ionic strength allowing the elution of complexes from the Mono Q column. This was further supported by in vitro studies showing that Rab3A GDI.Rab complexes are stable up to 0.5 M NaCl concentration (data not shown). Mono Q chromatography thus provided an accurate  Fig. 1 (lower panel). The amounts of Rab3A GDI and GTP-binding proteins found in the different fractions were plotted on the graph shown in the upper panel. B , aliquots of fractions 8 and 12 from the Mono Q column as well as crude (unfractionated) CHO and brain cytosols were cross-linked with EGS and resolved by SDS-PAGE. GTP-binding proteins present in fraction 8 migrate as 70-kDa complexes while they migrate at around 80 kDa after cross-linking fraction 12. C, cross-linked fractions 9, 10, and 12 were immunoprecipitated with the monoclonal anti-myc epitope antibody and blotted to nitrocellulose. The presence of myc-epitope tagged Rab6 was then revealed by the GTP overlay assay. method to analyze cytosolic Rab complexes.
The chromatography on the same Mono Q column of a cytosol prepared from CHO cells gave a very different profile ( Fig. 2A). In order to facilitate the detection of Rab complexes, we used a CHO cell line stably expressing myc epitope-tagged Rab6 a t about 5-fold the endogenous level of Rab6. The majority of small GTP-binding proteins present in CHO cytosol was eluted from the Mono Q column at lower ionic strength (around 100-150 mM NaCl, corresponding to fractions 6-9) than in brain cytosol (Fig. 2 A ) . Only a minor fraction of GTP-binding proteins (around 10%) was found to co-elute with Rab3A GDI (fractions [11][12]. To demonstrate the presence of complexes in the different fractions, they were cross-linked with EGS and analyzed by the GTP overlay assay (Fig. 2 B ) . Cross-linking of fraction 12 generated complexes of the same size as those detected in crosslinked brain cytosol and as such probably correspond to Rab proteins complexed with Rab3A GDI. In contrast, cross-linking of fraction 8, in which the majority of the small GTP-binding proteins were recovered, generated complexes of smaller size migrating with an apparent molecular mass around 70 kDa. Similar size complexes were obtained after cross-linking unfractionated cytosol (Fig. 2B). This suggests that the majority of Rab proteins were not complexed with Rab3A GDI but to another protein with a smaller molecular weight. This was confirmed by using specific antibodies against several Rab proteins. As shown in Fig. 2 A , RablA as well as endogenous Rab6, eluted with the bulk of small GTP-binding proteins. Rab2 and Rab4A were also found in fractions 6-9 (data not shown). myc epitope-tagged Rab6, which contains 4 additional acidic residues, was eluted from the Mono Q column at a higher NaCl concentration than endogenous Rab6 (fraction 10). After crosslinking of proteins present in fraction 10 with EGS, the majority of myc epitope-tagged Rab6 was detected in a band corresponding to a -70-kDa complex (Fig. 2C). In contrast, crosslinking of fraction 12, in which only a minor amount of myc epitope-tagged Rab6 was present, but which contains Rab3A GDI, revealed the presence of Rab6 in a -80-kDa complex. This result demonstrates that the majority of myc epitope-tagged Rab6 was complexed to a protein with a smaller apparent molecular mass than Rab3A GDI.
Purification of the Protein Complexed with Cytosolic Rab Proteins in CHO Cells-The above result prompted us to purify the protein which forms cytosolic complexes with Rab proteins in CHO cells. About 250 mg of cytosolic proteins prepared from wild-type CHO cells were loaded onto a QFF anion exchange column and eluted with a linear gradient of NaCl. Since one of the properties of Rab3A GDI is to induce the dissociation of Rab proteins from membranes (20), we tested each fraction for its ability to remove membrane-bound myc epitope-tagged Rab6. As expected, fractions which contained Rab3A GDI were able to dissociate Rab6 from membranes. However, a significant release of myc epitope-tagged Rab6 was obtained in fractions which did not contain detectable levels of Rab3A GDI, suggest- ing the existence of another GDI-like protein (data not shown). These fractions were then pooled and subjected to further purification through a combination of Mono Q and gel filtration columns (see "Experimental Procedures"). Fig. 3A shows the protein profile of the fractions collected after the last step of the purification procedure. A GDI-like activity was found to correlate with the presence in these fractions of a protein migrating -47 kDa. No Rab3A GDI was detected by immunoblotting in these fractions (data not shown).
The -47-kDa protein was then excised, digested with the Lys-specific endoproteinase, and the resulting peptides were purified by high pressure liquid chromatography. The chromatographic profile of the digested protein was found to be close to that of Rab3A GDI (data not shown). However, several peptides were different and some of them were microsequenced. We show in Fig. 4 the amino acid sequences of three of these peptides. Of the 32 amino acids sequenced, no difference was found with the predicted sequence of GDI p whose cDNA has been recently cloned from a rat brain library (22). Only two differences were found with GDI-2, a novel GDI isoform characterized from mouse skeletal muscle (22). On the other hand, we found more differences (7 out of 32) between the sequenced peptides and those corresponding to Rab3A GDI (bovine) or GDI CY and GDI-1, the rat and mouse counterparts of Rab3A GDI, respectively (17,22,23). The protein which forms cytosolic complexes in CHO cells is then most likely GDI p.
GDI p from CHO cells migrates with a slightly faster mobility on SDS-PAGE than Rab3A GDI (Fig. 3B). On the other hand, rat GDI p has the same predicted molecular weight as Rab3A GDI (51,000) (22). We found, however, that Escherichia coli expressed mouse GDI p, whose cDNA has been recently cloned in our laboratory: migrates -at 47 kDa (data not shown). This indicates that GDI /3 and Rab3A GDI do not have the same apparent masses as determined by SDS-PAGE. This observation can explain why cross-linked cytosolic Rab protein complexes were found to be of smaller size in CHO cells (-70 kDa) as compared with complexes formed with Rab3A GDI in brain cytosol (-80 m a ) . In addition, the predicted isoelectric point of rat GDI p (PI 5.7) is more basic than that of bovine Rab3A GDI (PI 4.8), as calculated by the Genetics Computer Group software. Comparable values were obtained by two-dimensional gel electrophoresis (see Fig. 7). This likely explains why Rab.GDI p complexes were eluted from the Mono Q column at lower ionic strength than Rab/2 Rab3A GDI complexes ( Figs. 1 and 2).

Depending on Cell npes, a Variable Fraction of Rab Proteins
Are Complexed with Either GDI p or Rab3A GDZ-In order to extend the above observations, we analyzed various cytosols by Mono Q chromatography. Fig. 5 illustrates the chromatographic profile of a cytosol prepared from insulin-secreting RINm5F cells. As in the case of brain cytosol, the majority of small GTP-binding proteins co-eluted with Rab3A GDI at an ionic strength around 250 mM. Especially, Rab3A, which is expressed in this cell line, was recovered in the same fraction (fraction 16) as the one in which most of Rab3Apresent in brain cytosol was collected (Fig. lA). These results confirm a previous report showing that the majority of cytosolic Rab proteins are complexed with Rab3A GDI in RINm5F cells (16). However, a minor amount of small GTP-binding proteins including RablA, Rab6, and Rab4A (data not shown), did not co-elute with Rab3A     Rab3A was found in the same fraction (fraction 16) after chromatography of brain cytosol (Fig. L4). However, a minor amount of small GTPbinding proteins eluted at the same ionic strength as the majority of them in CHO cytosol (fractions 7-9).
GDI, but eluted at about the same NaCl concentration as the majority of small GTP-binding proteins present in CHO cytosol (fractions 8 and 9).
The elution profile of small GTP-binding proteins present in rat liver cytosol was found to be intermediate between the one obtained for CHO cytosol and the one found for brain cytosol (Fig. 6). Indeed, about half of the total small GTP-binding proteins, including about half RablA and Rab6 as well as Rab2 and Rab4A (data not shown), eluted at the same ionic strength as the majority of these proteins in CHO cytosol (fractions 6-9). The other half of the GTP-binding proteins co-eluted with Rab3A GDI and was recovered in the same fractions in which small GTP-binding proteins present in brain cytosol were collected (first peak corresponding to fractions 11-13 in Fig. lA). In addition, cross-linking of small GTP-binding proteins present in fractions 6-9 and 11-13 generated -70-and -80-kDa size complexes, respectively (data not shown).
These results indicate that a minority of Rab proteins in RINm5F cells and about half of the total in rat liver cytosol were not complexed with Rab3A GDI. The fact that these complexes have the same apparent molecular mass (-70 kDa) and eluted at the same ionic strength as in CHO cytosol strongly FIG. 7. Analysis of fractions 7-9 and 11-12 of rat liver cytosol obtained after Mono Q chromatography by high resolution two-dimensional electrophoresis and immunoblotting. Aliquots of combined 7-9 and 11-12 fractions obtained as described in the legend to Fig. 6 were resolved by two-dimensional gel electrophoresis and transferred to nitrocellulose. GDI proteins present in these fractions were revealed using an antibody recognizing both Rab3A GDI and GDI p (see Fig. 8). In fractions 11-12, the GDI protein migrates similarly as Rab3A GDI while in fractions 7-9 it displays the tric point as GDI p. Stars indicate the same electrophoretic mobility and isoelecposition of bovine serum albumin used as internal marker (PI 5.5).
Rab Proteins Complexed with GDI Isoforms IEF 1 e suggests that Rab proteins were interacting with GDI p. To confirm this point, fractions 7-9 and 11-12 obtained after chromatography of rat liver cytosol were analyzed by high resolution two-dimensional gel electrophoresis and immunoblotting using an antibody reacting against both Rab3A GDI and GDI p (see Fig. 8 for the characterization of this antibody). As shown in Fig. 7, a protein displaying the same electrophoretic mobility and isoelectric point as GDI p purified from CHO cells was detected in fractions 7-9. On the other hand, the protein that forms complexes with Rab proteins in fractions 11-12 was found to be, as expected, Rab3A GDI.
A Correlation Exists between the Level of Expression of Rab3A GDI and GDI and the Proportion of Rab Proteins Complexed with Them-In an attempt to understand why Rab proteins are complexed with either Rab3A GDI or GDI /3, we next estimated the relative proportion of these two proteins in the different cytosols that were tested above. These experiments were performed with a polyclonal antibody generated against Rab3A GDI which also recognizes GDI p, as illustrated in Fig. 8A. This antibody allowed the detection of both proteins in CHO extracts (Fig. 8B). In brain, liver, and RINm5F cytosols, the antibody identified two bands migrating with the same electrophoretic mobility as Rab3A GDI and GDI p. The level of expression of Rab3A GDI was found to be much lower in liver and CHO cytosols as compared with brain and RINm5F cells. In contrast, GDI p appeared to be expressed at the same levels in brain, RINm5F, and liver cytosols (Fig. 8, B and C ) , confirming previous studies showing that GDI p mRNA is expressed a t a similar level in a broad range of tissues (22).
However, GDI p was about 4-fold more abundant than Rab3A GDI in CHO cytosol. Interestingly, the above observations point out that a correlation exists between the relative amount of both GDI isoforms and the proportion of Rab proteins complexed with them. When Rab3AGDI is far more abundant than GDI p, such as in brain or RINm5F cells, the majority of Rab proteins form complexes with Rab3A GDI. In contrast, they appear to be complexed with either GDI p or Rab3A GDI when both proteins are expressed at about the same levels, such as in rat liver cytosol. Remarkably, when GDI /3 is the more abundant GDI, such as in CHO cytosol, it can complex the majority of Rab proteins. Therefore, the simplest interpretation of the   1 and 3 ) and purified Rab3A GDI (lanes 2 and 4 ) were transferred to nitrocellulose after SDS-PAGE. Proteins were stained with Ponceau S (lanes Ponceau S) or with an antiserum raised against purified bovine Rab3A GDI (lanes immunoblot). This antiserum recognizes both proteins. B , 5 1. 18 of mouse brain, CHO (expressing myc epitope-tagged RabG), rat liver, and RINm5F cytosols were transferred to nitrocellulose, blotted with the same antiserum, and revealed by ECL. C, the amounts of Rab3A GDI and GDI p present in these cytosols were quantitated using "'1-Protein A.
above results is that the majority of Rab proteins can form complexes with either GDI isoform and do so in proportion to their relative abundance.

DISCUSSION
Recent studies have illustrated that the function of GDI could be to chaperone Rab proteins throughout their cycle between cytosol and membranes. For instance, Rab3A GDI has been shown to mediate in vitro specific membrane association of Rab5 and Rab9 proteins (34,35). It is also likely that another important role of GDI is to remove the GDP-bound form of Rab proteins from membranes when they have performed their function, thus completing the Rab cycle (35). These studies have, however, been performed with Rab3A GDI. In this study, we provide the first evidence that Rab proteins form, in fact, in vivo complexes with at least another GDI isoform, GDI p. Interestingly, the proportion of Rab proteins found as cytosolic complexes with either GDI isoform depends on the tissue and cell type. In rat brain and insulin-secreting RINm5F cells, cytosolic Rab proteins are mostly complexed with Rab3A GDI, in good agreement with previous findings (16). In contrast, Rab proteins are mainly complexed with GDI p in CHO cells. An intermediate situation was found for rat liver cytosol, in which a roughly equivalent proportion of Rab proteins appear to be complexed with either Rab3A GDI or GDI 0. Interestingly, a correlation exists between the ratio between Rab3A GDI and GDI 0 and the proportion of Rab proteins complexed with either isoform.
I t is not clear at the moment how many isoforms of GDI exist in mammals. rub GDI p was isolated concomitantly with rub GDI Q from a rat brain cDNA library (22). While GDI only exhibits 86% amino acid identity with bovine Rab3A GDI, identity between rat GDI a and bovine Rab3A is 99%. Both rat GDI Q and bovine Rab3A GDI mRNA are highly expressed in brain and at much lower levels in other tissues (22). In contrast GDI p appears to be expressed at the same level in all tissues, including brain. These results then indicate that at least two GDIs exist in rat, where GDI a is the rat counterpart of bovine Rab3A GDI. The same seems to be true in man, GDI / 3 being 94% identical to a protein referred to as human GDI by Takai and co-~orkers.~ Another cDNA encoding a GDI-like protein, named GDI-2, has also been recently cloned from a mouse skeletal muscle library. GDI-2 exhibits 86.7% amino acid sequence identity with Rab3A GDI. The sequence identity between rat Rab GDI p and mouse GDI-2 is very high (94%), both mRNA transcripts have the same size (2.4 kilobase pairs) and are ubiquitously expressed (22,23). In addition, the apparent molecular mass of GDI-2 determined by SDS-PAGE is around 46 kDa, which is exactly what we found for GDI fi in this study. However, whether or not GDI p and GDI-2 are the same proteins is still unclear and future experiments will be needed to clarify this issue. In any case, they appear to be more closely related to each other than to Rab3A GDVGDI a. The important questions that now arise are why two (or three) different isoforms of GDI exist in the cells and whether or not they fulfill different functions. An important finding in this study is that the majority of Rab proteins seems to be able to form complexes with either GDI isoform depending on their relative abundance. This is especially striking in rat liver cytosol in which roughly half of the small GTP-binding proteins, including RablA, Rab2, Rab4A, and Rab6, are complexed with Rab3A GDI and the remaining with GDI p. These observations raise the possibility that GDI isoforms might be interchangeable to form complexes with the majority of Rab proteins. Why then do GDI isoforms exist? Several hypotheses can be envisaged. First, Rab3A GDVGDI Q might regulate Rab proteins specifically expressed in cells in which this GDI isoform is the most abundant. On the other hand, GDI PIGDI-2, which seems M. Asada, K. Kaibuchi, and Y. Takai, unpublished results.
to be expressed at similar levels in all cells and tissues tested so far, could serve as a regulator of ubiquitous Rab proteins. An obvious candidate for a protein regulated by Rab3A GDI is RabSAitself, a Rab protein that appears to be expressed in cells or tissues displaying high levels of Rab3A GDI, such as neuronal, neuroendocrine, and some endocrine cells (36)(37)(38). We found, however, in this study that Rab3A GDI forms complexes with most of the Rab proteins in brain and RINmSF cells. In light of what is known so far about GDI function, it is difficult to imagine how Rab3A GDI can be a regulator of a particular Rab and at the same time form complexes with many other Rab proteins.
Another possibility is that GDI isoforms specifically interact with a subset of Rab proteins. It will then be of interest to precisely determine the afiinity constants of both GDI isoforms for several Rab proteins. Nevertheless, the fact that Rab binding (formation of cytosolic complexes) correlates with the proportion of a given GDI argues against major differences in affinity constants of GDI fi and Rab3A GDI for the majority of Rab proteins. It has also been shown that Rab3A GDVGDI a and GDI fi display the same activity (inhibition of GDP release) on Rab3A and Rabll proteins (22). Alternatively, Rab/Rab3A GDI and Rab/GDI p complexes might have different affinities for other(s) protein(s), such as a GDI-displacement factor or an exchange protein thought to be present on cellular membranes and to interact with Rab/GDI complexes (34,35). Finally, GDI isoforms might selectively interact with different regulatory proteins. It will be of interest, for instance, to determine whether GDI p is phosphorylated in uiuo, as previously shown for Rab3A GDI (39).
In conclusion, the recent discovery of one or two new GDI isoforms has rendered the understanding of GDI function more complicated than previously thought. Clearly, at least two GDI isoforms can regulate Rab function. Further experiments will be required to test whether GDI isoforms are redundant or whether they fulfill different functions.
ing us with the antibody against Rab3A GDI and to Dr. FranGois