A Novel Membrane-anchored Rab5 Interacting Protein Required for Homotypic Endosome Fusion*

The ras-related GTPase rab5 is rate-limiting for homotypic early endosome fusion. We used a yeast two-hybrid screen to identify a rab5 interacting protein, rab5ip. The cDNA sequence encodes a ubiquitous 75-kDa protein with an N-terminal transmembrane domain (TM), a central coiled-coil structure, and a C-terminal region homologous to several centrosome-associated proteins. rab5ip lacking the transmembrane domain (rab5ipTM(−)) had a greater affinity in vitro for rab5-guanosine 5′-O-2-(thio)diphosphate than for rab5-guanosine 5′-3-O-(thio)triphosphate. In transfected HeLa cells, rab5ipTM(−) was partly cytosolic and localized (by immunofluorescence) with a rab5 mutant believed to be in a GDP conformation (GFP-rab5G78A) but not with GFP-rab5Q79L, a GTPase-deficient mutant. rab5ip with the transmembrane domain (rab5ipTM(+)) was completely associated with the particulate fraction and localized extensively with GFP-rab5wt in punctate endosome-like structures. Overexpression of rab5ipTM(+) using Sindbis virus stimulated the accumulation of fluid-phase horseradish peroxidase by BHK-21 cells, and homotypic endosome fusion in vitro was inhibited by antibody against rab5ip. rab5ipTM(−) inhibited rab5wt-stimulated endosome fusion but did not inhibit fusion stimulated by rab5Q79L. rab5ip represents a novel rab5 interacting protein that may function on endocytic vesicles as a receptor for rab5-GDP and participate in the activation of rab5.

Membrane trafficking in eukaryotic cells is regulated by a group of ras-related GTPases known as rabs. Each member of this protein family is 23-25 kDa in mass and tightly associated with membranes via C-terminal geranylgeranyl modifications. Each membrane-bound cellular compartment appears to contain a discrete set of rabs, and these proteins appear to regulate vesicle tethering to target membranes and SNARE 1 complex assembly (1).
There are several rabs associated with the endocytic pathway, including rab5 (2), rab4 (3), rab7 (4), and rab11 (5,6). rab5 is a key regulator of endocytosis, because it is rate-limiting for homotypic endosome fusion (2,7). rab5 mutants defective in guanine-nucleotide binding (rab5 S34N or rab5 N133I ) act in a dominant negative fashion when expressed in vivo, causing the accumulation of small vesicles at the cell periphery, probably by inhibiting vesicle fusion events that form large sorting endosomes (7). In contrast, overexpression of the GTPase-defective mutant rab5 Q79L causes the formation of enlarged endosomes by enhanced stimulation of endosome fusion (8). rab5 Q79L stimulates endosome fusion in vitro to a greater extent than rab5 wild type (8,9), and consistent with this, hydrolysis-resistant GTP analogues also stimulate fusion in vitro (10). A rab5 mutant that binds xanthine nucleotide instead of guanine (rab5 D136N ) stimulates endosome fusion in the presence of xanthosine 5Ј-3-O(thio)triphosphate, proving that nucleotide is acting through rab5 and not some other GTPase (11,12). Thus rab5-GTP is essential for endosome fusion, but hydrolysis of GTP itself is not required: the function of the GTPase probably is to maintain a dynamic equilibrium between rab5-GDP and rab5-GTP (11). This equilibrium may be regulated in part by GTPase activation proteins (GAPs) such as p120 rasGAP (13), and by the tumor suppressor tuberin, product of the TSC2 complex (14).
Effectors that interact with rab5 include rabaptin-5, a cytosolic protein that is recruited to endosomes by rab5-GTP (15,16) and inhibits GTP hydrolysis (11). Another rab5 interacting protein, rabex-5, is a 60-kDa cytosolic nucleotide exchange factor with homology to a yeast vacuolar sorting protein, VPS9 (17). rab5-GDP also interacts with guanine nucleotide dissociation inhibitor (GDI), a protein that binds all rabs in their GDP-bound states, inhibits the release of GDP, and extracts rab-GDP from target membranes into the cytosol (18). rab5-GDP is loaded by GDI onto endosomal membranes, where rabex-5, as part of a complex with rabaptin-5, stimulates GDP release and subsequent exchange for GTP. rab5-GTP appears to recruit other proteins, including syntaxin-13 and N-ethylmaleimide-sensitive factor, which form a complex that is dis-* 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.
To learn more about the function of rab5 and the regulation of its activity, we employed a yeast two-hybrid screen to identify other proteins that interact with rab5. Here we report the characterization of a membrane-bound protein that may regulate the activity of rab5 by a specific interaction with rab5-GDP.

EXPERIMENTAL PROCEDURES
Materials-HeLa and BHK-21 cells were obtained from the American Type Culture Collection (Rockville, MD). HeLa cells were cultured in Dulbecco's modified Eagle's medium with 10% fetal bovine serum (Life Technologies, Inc.) and BHK-21 cells in modified Eagle's medium-␣ with 10% fetal bovine serum. A mouse monoclonal antibody against the Xpress (Xp) epitope tag was obtained from Invitrogen (Carlsbad, CA), and an anti-c-myc monoclonal antibody (9E10) from Berkeley Antibody Co. (Berkeley, CA). Goat anti-mouse IgG conjugated to Texas Red was obtained from Molecular Probes (Eugene, OR), and paraformaldehyde was obtained from Electron Microscopy Sciences (Ft. Washington, PA). Horseradish peroxidase (HRP) and FuGENE were obtained from Roche Molecular Biochemicals. The QuikChange sitedirected mutagenesis kit was purchased from Stratagene (La Jolla, CA). All other reagents were from Sigma unless otherwise noted.
Cloning a rab5 Interacting Protein cDNA-A yeast two-hybrid screen for rab5 interacting proteins was performed according to published procedures (21,22). A human rab5A cDNA (9) was first subcloned into the two-hybrid vector pGBT9 (21) to fuse rab5 onto the C terminus of the yeast GAL4 DNA-binding domain, then a prenylation defect (rab5 C212S/C213S ) was created by site-directed mutagenesis. The yeast reporter strain Y190 (MATa, leu2-3, 112ura3-52, trp1-901, his3-D200, ade2-101) was transformed with pGBT9-rab5 C212S/C213S , then with a human B-lymphocyte cDNA library cloned into the vector pACT (a gift of S. Elledge). Double transformants were plated on synthetic media lacking leucine, tryptophan, and histidine, and prototrophic colonies were then tested by a replica filter assay for ␤-galactosidase activity. Approximately 10 ϫ 10 6 transformants were screened with a transformation efficiency of 8.3 ϫ 10 4 /g of library DNA.
Additional rab5 test baits differing in their prenylation status (rab5 C213S and rab5 C212S ) were created by base substitutions, and a C-terminal truncation (pGBT9-rab5 TR ) was generated by changing the arginine codon at position 197 to a stop codon. Negative control bait plasmids pGBT9-lamin, and pGBT9-p53 were obtained from CLON-TECH (Palo Alto, CA). Other baits were constructed by insertion of appropriate cDNA fragments into the two-hybrid vector pAS (23): a human rab4 cDNA (9), and rat rab3D (24) and human rab6 (25) open reading frames (ORFs) that had been generated by add-on polymerase chain reaction (PCR).
Extension of the initial two-hybrid cDNA clone (rab5ipTM(Ϫ): see below) was performed by 5Ј rapid amplification of cDNA ends (RACE) PCR using a rab5ip primer (5Ј-AGGGTGTCCTCGTGGCTCAG-3Ј), together with the adapter-primer AP1 to prime adapter-ligated cDNA from human placenta (Marathon, CLONTECH). The Advantage PCR polymerase mix (CLONTECH) was used as recommended by the manufacturer, except that 5% Me 2 SO was added to the reaction. The RACE PCR fragment (rab5ipTM(ϩ)) was purified and subcloned into the pTAg vector (Novagen, Madison, WI).
Prediction of Structural Properties, Search for Protein Homologues, and Multiple Alignment-The predicted translation product of the rab5ip cDNA was examined for the presence of protein structural features. Coiled coils were predicted by analysis with the program Coils, version 2.2 (26), and transmembrane regions were predicted using TMpred (27). A sequence similarity search was performed using the BLAST 2.0 server at the National Center for Biotechnology Information. Multiple sequence alignments of homologous proteins were done using Clustal W version 1.7 (28).
In Vitro Binding of rab5ipTM(Ϫ) and GST-rab Fusion Proteins-To load rab5 with nucleotides, purified GST-rab fusion proteins (1.5 M) were diluted into chelating buffer (100 l) consisting of 20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM EDTA, and 0.1% CHAPS. After incubation at room temperature for 2 min, guanosine 5Ј-O-2-(thio)diphosphate (GDP␤S) or guanosine 5Ј-3-O-(thio)triphosphate (GTP␥S) was added to a final concentration of 1.5 M, and the incubation was continued for 150 min for GST-rab5 wt or 30 min for GST-rab5 mutants. MgCl 2 was then added to 20 mM, and the proteins were kept on ice until used. Purified GST-rab fusion proteins (7.5 g) were incubated with 50 l of NTA-agarose bound with 6His-Xp-rab5ipTM(Ϫ) in buffer A (20 mM Tris-HCl, pH 8.0, 5 mM MgCl 2 , 2 mM mercaptoethanol) supplemented with 150 mM NaCl, 0.1% CHAPS, and 20 mM imidazole at 4°C for 1 h. The resin was washed with 50 mM imidazole and 500 mM NaCl in buffer A, and the proteins were eluted with 250 mM imidazole and 150 mM NaCl in buffer A. The eluted proteins were separated on 12% SDSpolyacrylamide gels and stained with Coomassie Brilliant Blue. Dried gels were scanned to create digitized images, which were then quantified using NIH Image (v. 1.62).
Northern Blotting-Multiple Tissue Northern blots containing 2 g/ lane of poly(A) ϩ RNA from various mouse and human tissues were obtained from CLONTECH. A probe was generated by PCR amplification of a 566-base pair rab5ip fragment spanning amino acids residues 293-481 of rab5ip. The purified fragment was 32 P-labeled by random priming (29) and used to probe the blots under high stringency conditions.
Anti-rab5ip Antibody-A rab5ip peptide sequence SLSLTLQKEG-VIGVTEEQV (amino acid positions 469 -487) of rab5ip was identified as unconserved among rab5ip-related data base proteins (Fig. 1B). This sequence was added to the N terminus of the peptide QYIKANSKFIG-ITELKK, the universal T-cell epitope of tetanus toxoid (30), and used to immunize rabbits. The antiserum was affinity-purified using the peptide antigen lacking the tetanus toxoid sequence, and the purified antibody was designated #9916. Another rab5ip peptide, SS-NWQKEAMRLERLELRQG (amino acid residues 246 -264 of rab5ip) was identified using the Jameson-Wolf algorithm as having a high antigenic index (31), and used to generate an antipeptide antiserum, designated #9816.
Immunoblotting of rab5ip-pcDNA3.1/6His-Xp-rab5ipTM(ϩ) and pcDNA3.1/6His-Xp-rab5ipTM(Ϫ) were expressed by transient transfection of HeLa cells for 48 h using 17 l of FuGENE and 10 g of DNA per 100-mm dish. To prepare membranes and cytosol, the cells were washed with phosphate-buffered saline (PBS), harvested into 10 mM potassium acetate, 10 mM HEPES, pH 7.3, with Complete protease inhibitor (Roche Molecular Biochemicals) and swelled on ice for 10 min. The cells were then lysed by 10 passages through a 23-g needle, and the potassium acetate was adjusted to 25 mM and the HEPES to 125 mM. The homogenate was centrifuged at 3000 ϫ g at 4°C for 20 min to remove nuclei and intact cells, and the supernatant was centrifuged at 66,000 rpm in a Beckman TLA100.3 rotor for 30 min at 4°C to pellet the membranes. For whole cell extracts, washed cells were quickly dissolved in Laemmli buffer (32) prior to electrophoresis. Samples were separated by SDS-polyacrylamide gel electrophoresis (PAGE), transferred to Immobilon-P membranes (Millipore), and probed with the rabbit antipeptide antibody #9916 against rab5ip or anti-Xp monoclonal antibodies. Bound antibodies were detected by chemiluminescence (Pierce).
Fluid Phase Endocytosis-A Sindbis virus recombinant, expressing rab5ipTM(ϩ) or GFP-rab5 Q79L , was constructed using methods described previously (33). BHK-21 cells were plated in 24-well clusters at a density of 4 ϫ 10 4 cells/cm 2 , then infected 24 h later with recombinant virus or with empty virus vector. 5 h after infection, the cells were incubated in 5 mg/ml HRP in serum-free medium with 1% bovine serum albumin for 60 min at 37°C. The monolayers were washed six times with ice-cold PBS and lysed in PBS with 0.1% Triton X-100. HRP activity was measured in a reaction with 100 mM KH 2 PO 4 , pH 5, with 1 mM 2,2Ј-azino-bis(3-ethylbenzthiazoline-6-sulfonic acid) and 0.02% H 2 O 2 in 96-well plates using an automated plate reader. The V max /g of protein in each lysate was determined and normalized to HRP uptake by empty-vector-infected cells.
Immunofluorescence Microscopy-HeLa cells growing on poly-D-lysine-coated no. 1 glass coverslips in 35-mm cluster wells were transfected using 2 g of plasmid DNA and 3 l of FuGENE per well. For cotransfections, 1 g of each plasmid DNA was used. 48 h after the addition of FuGENE⅐DNA complexes, the cells were washed with PBS containing 0.12% sucrose (PBSS) and fixed with 4% paraformaldehyde in PBSS at 4°C for 10 min, followed by a wash with PBSS. All subsequent steps were done at room temperature, with PBSS used for washes. The fixed cells were incubated in 0.34% L-lysine, 0.05% sodium m-periodate in PBSS for 20 min, then washed and permeabilized with 0.2% Triton X-100. After further washing, the cells were blocked with 10% heat-inactivated goat serum for 15 min. Monoclonal antibodies were diluted to 5 g/ml in PBSS with 0.2% heat-inactivated goat serum and 0.05% Triton X-100 then added to the cells and left overnight. The cells were washed with PBSS three times before labeling with rabbit anti-mouse IgG Texas Red (1:200 dilution) using the same procedure as for the primary antibody. Coverslips were mounted in Mowiol (Calbiochem, La Jolla, CA) and viewed with a Molecular Dynamics Multiprobe 2001 confocal imaging system attached to a Zeiss Axiovert 100 microscope. The absence of bleed-through from GFP emission into the Texas Red channel was verified by examining HeLa cells transfected with GFP-rab5 only. The images were labeled, sized, and assembled into panels using Adobe Photoshop (version 5.0) then printed on a Codonics NP-1600 dye diffusion printer.

RESULTS
Cloning of rab5ip cDNA-By screening a human B-lymphocyte cDNA library with prenylation-defective rab5 as bait, we identified a cDNA encoding a protein that we have termed rab5ip. Yeast expressing this cDNA as a fusion with the GAL4 transcription activation domain (TAD) were transformed with other plasmids expressing GAL4 DNA-binding domain (DBD) fusion proteins to assess the specificity of this interaction (Table I). The expression of the rab5ip-TAD fusion by itself, or together with nonspecific DBD fusion proteins (p53, nuclear lamin) produced no detectable ␤-galactosidase activity. No activity was detected when rab5ip-TAD fusion was coexpressed with rab6, rab3D, or rab4 DBD fusions, indicating specificity for interaction with rab5. The presence of a rab5 C terminus or its prenylation was not necessary for the interaction of rab5ip with rab5. In fact, a quantitative enzyme measurement indicates that lack of rab5 C-terminal prenylation results in greater ␤-galactosidase activity (Table I).
Sequence and Structural Features of rab5ip-The 5Ј-RACE PCR procedure was performed to obtain a longer rab5ip cDNA. The augmented sequence contains an open reading frame encoding a 550-residue polypeptide of M r 62,200 (Fig. 1A). A transmembrane region near the N terminus of this ORF is predicted by TMpred (27), and three regions of potential coiledcoil structure were identified by Coils (26). Such coiled-coil structures could mediate interactions of rab5ip with other proteins (35). The rab5ip sequence does not resemble any known rab5 interacting protein (rabex-5, rabaptin-5, or EEA1) or any other rab interacting protein identified to date.
A search of sequence data bases using the rab5ip nucleotide sequence showed that rab5ip is a novel protein; however, there are dozens of human expressed sequence tags (ESTs) with similarities of Ͼ95% clustered in the 3Ј-untranslated region of the rab5ip cDNA. It is unclear whether there are multiple related rab5ip genes or if there has been repetitive cloning of a small number of rab5ip cDNAs as ESTs. Because these ESTs were obtained from human cDNA libraries of diverse tissue sources, it appears that rab5ip is a ubiquitous protein and is expressed in several different developmental stages and tissues (also see below). No homologues were detected in the mouse, Drosophila melanogaster or Saccharomyces cerevisiae data bases. At least part of the human rab5ip gene has been sequenced (clone accession no. AL021707), and is located on chromosome 22 (22q12-22q13). A number of apparent rab5ip homologues were identified in nucleotide and protein data bases by searching with the rab5ip amino acid sequence. Most of these were similar only in the coiled-coil domain of rab5ip, and are probably not true homologues. However, four proteins show similarity in the C-terminal domain. In the nucleotide data base are cDNAs KIAA0810 (36), encoding a protein of unknown function, and an open reading frame (mug13) from chromosome 5 of Arabidopsis thaliana (37). Two other proteins homologous to the rab5ip C-terminal region are Schizosaccharomyces pombe Sad1 (38), a spindle-pole body-associated protein, and UNC-84 (39), which may facilitate nuclear-centrosome interactions in Caenorhabditis elegans. An alignment of rab5ip-related sequences within this Sad1 homology region is shown in Fig. 1B. Proteins containing the Sad1 homology region share, in addition, the presence of at least one putative transmembrane domain and centrally located regions potentially capable of forming coiled coils (Fig. 1C).
The rab5ip sequence reported here is identical to the Cterminal portion of KIAA0668, with the latter potentially including 191 additional amino acid residues at the N terminus. In KIAA0668, none of the proximal methionine codons (ATG) in-frame with rab5ip possess the classical (most efficient) Kozak consensus sequence (CCPuCCATGG) (40,41). The first methionine codon within the KIAA0668 open reading frame that is followed by a guanine nucleotide (i.e. in the ϩ4 position) starts the rab5ipTM(ϩ) sequence (Fig. 1A). However, immuno- In all tests, the other two-hybrid partner ("prey") was pACT-rab5ipTM(Ϫ), the originally selected clone that lacks the transmembrane domain. pGBT9-rab5 by itself was negative for ␤-galactosidase expression. ϩ, interaction between rab5ipTM(Ϫ) (the prey) and the bait plasmid, indicated by ␤-galactosidase positive (blue) yeast colonies on indicator plates. Ϫ, no detectable interaction, indicated by a white colony.

Bait plasmid
Interaction ␤-Galactosidase activity a mol ONPG/min/cell Ϫ ND a ␤-Galactosidase activities were determined by the colorimetric conversion of ONPG (o-nitrophenyl ␤-D-galactopyranoside) in yeast extracts as described (22). b ND, not determined.
FIG. 1. Amino acid sequence of rab5ip and its alignment with other proteins. A, predicted amino acid sequence of rab5ip. The N-terminal end point of the original rab5ip two-hybrid clone (rab5ipTM(Ϫ)) and the potential translation start of rab5ipTM(ϩ) are indicated. Also shown are potential translation starts derived from the KIAA0668 cDNA (57). The probable initiation codon begins the predicted 75-kDa ORF, as suggested by immunoblot (see Fig. 4A). The most likely transmembrane domain, predicted using TMpred (27), is highlighted in black. Coiled coils were predicted using Coils (27) with a window of 21; shown underlined is the extent of polypeptide chain with a coil-forming probability of 0.25 or higher. The Sad1 homology region is in the open box, and the peptides used to generate anti-peptide antibodies are highlighted in gray. The boldfaced numbers indicate residue positions with respect to the predicted 75-kDa ORF. B, alignment of Sad1 homologous regions from several data base proteins: KIAA0810 is a human brain cDNA clone (57); UNC-84, from C. elegans (39), and Sad1, from S. pombe (38) are both centrosome-associated proteins; mug13 is an open reading frame from the A. thaliana genome (37). Identical residues are highlighted in black, and similar ones are highlighted in gray. The first residue of this region in rab5ip is number 470 of the 75.0-kDa ORF in A. C, predicted structures of proteins with homology to Sad1. Mug13 is not shown, because its cDNA has not been characterized.
blotting data suggest that rab5ip is the 75-kDa protein initiated at the upstream methionine codon shown in Fig. 1A (see below).
Interaction between rab5ip and rab5 in Vitro-An in vitro assay was used to examine the interaction between rab5ip and rab5. 6His-Xp-rab5ipTM(Ϫ) was expressed in E. coli, bound to NTA-agarose beads, then incubated with GST-rab fusion proteins purified in the presence of GDP. The beads were eluted with buffer containing 250 mM imidazole, and the eluates were examined by SDS-PAGE and Coomassie Blue staining to measure GST-rab5 coelution with rab5ipTM(Ϫ). GST-rab5-GDP bound efficiently to rab5ipTM(Ϫ), whereas GST alone or GST-rab4-GDP did not bind ( Fig. 2A). GST-rab5-GDP did not bind to NTA-agarose beads lacking rab5ipTM(Ϫ) (not shown). To determine the nucleotide dependence of binding, GST-rab5 wt , GST-rab5 Q79L , and GST-rab5 S34N were loaded with either GTP␥S or GDP␤S prior to incubation with rab5ipTM(Ϫ) beads. GST-rab5 wt (Fig. 2B, lanes 1 and 2) and GST-rab5 Q79L (Fig. 2B,  lanes 5 and 6) bound with GDP␤S interacted with rab5ipTM(Ϫ) more efficiently than did GST-rab5 bound with GTP␥S. GST-rab5 S34N did not bind rab5ipTM(Ϫ) detectably in the presence of either nucleotide (Fig. 2B, lanes 3 and 4). These results are quantified in Table II.
Distribution of rab5ip mRNA in Mammalian Tissues-A segment of the rab5ip coding sequence was amplified by PCR and used as a hybridization probe of Northern blots to determine the expression pattern of rab5ip in human and mouse tissues (Fig. 3). A discrete rab5ip mRNA was detected in all tissues, and the mRNA length (ϳ4000 nucleotides) is consistent with any of the open reading frames shown in Fig. 1A. rab5ip transcripts were uniformly prevalent in human tissues, whereas in mouse tissues there was more variability in abundance.
Immunoblot Analysis of rab5ip-An antipeptide antibody was raised against a sequence that is not conserved among known rab5ip homologues, located between the third coiled-coil region and the Sad1 domain (Fig. 1A). This antibody (#9916) recognizes a protein of 70 -75 kDa on an immunoblot of HeLa cell membranes (Fig. 4A), consistent with the 75.0-kDa ORF indicated at the top of Fig. 1A. To show that this protein corresponds to the cloned rab5ip, whole cell extracts prepared from untransfected HeLa cells or cells transfected with pcDNA3.1/6His-Xp-rab5ipTM(Ϫ) were probed using the #9916 antibody. In both extracts, an endogenous rab5ip band was detected, but in addition, the transfected cells expressed a smaller protein encoded by the cloned cDNA (Fig. 4B). To determine whether rab5ip lacking the putative trans- FIG. 3. Tissue distribution of rab5ip mRNA. A PCR product was generated using the rab5ip cDNA as template with primers bracketing amino acid residues 292-480 of the coding sequence. The probe was labeled by random priming and hybridized to Multiple Tissue Northern blots (CLONTECH) of RNA from mammalian tissues under high stringency conditions. 2 g of poly(A) ϩ RNA were loaded in each lane. A, human tissues: 1, heart; 2, brain; 3, placenta; 4, lung; 5, liver; 6, skeletal muscle; 7, kidney; 8, pancreas. B, mouse tissues: 1, heart; 2, brain; 3, spleen; 4, lung; 5, liver; 6, skeletal muscle; 7, kidney; 8, testis.

TABLE II
Quantification of nucleotide effect on rab5-rab5ip interaction Gels such as those in Fig. 2B, from three independent experiments, were scanned and quantified as described under "Experimental Procedures." The densities of the GST-rab5 and rab5ipTM(Ϫ) bands in each lane were measured, and the ratios GST-rab5/rab5ipTM(Ϫ) were calculated. GST 2. Interaction between rab5ipTM(؊) and rab5 in vitro. A, 6His-Xp-rab5ipTM(Ϫ) was purified from E. coli and bound to NTA-agarose beads, then incubated with GST-rab5 (lane 1), GST-rab4 (lane 2), or GST alone (lane 3), washed, and eluted. The eluates were examined by SDS-PAGE followed by Coomassie Blue staining. B, nucleotide dependence of rab5-rab5ip interaction. GST-rab5 was loaded with nucleotide as described under "Experimental Procedures" then incubated with 6His-Xp-rab5ipTM(Ϫ) bound to NTA beads. The beads were washed and eluted, and the eluates were examined as described in A. membrane domain was cytosolic, HeLa cells transfected with pcDNA3.1/6His-Xp-rab5ipTM(Ϫ) or pcDNA3.1/6His-Xp-rab5ipTM(ϩ) were fractionated into cytosol and membranes then probed with the anti-Xpress monoclonal antibody. rab5ipTM(ϩ) was detected only in the membrane fraction, whereas rab5ipTM(Ϫ) was found in both membrane and cytosol (Fig. 4C).
The subcellular localization of rab5ipTM(ϩ) with respect to rab5 was examined by cotransfecting HeLa cells with pEGFP-rab5 wt together with pcDNA3.1/6His-Xp-rab5ipTM(ϩ) then treating with cycloheximide for 2 h prior to fixation to enrich in overexpressed protein transported out of the Golgi. GFP-rab5 wt and rab5ipTM(ϩ) overlapped substantially, although some GFP-rab5 wt -containing vesicles were relatively depleted of rab5ipTM(ϩ) (Fig. 7). Similar results were obtained using rab5ipTM(ϩ) tagged with the c-myc epitope (data not shown). These results indicate that rab5ip localizes largely to a rab5positive population of endosomes.
Stimulation of Endocytosis by rab5ip Overexpression-Because rab5ip preferentially binds rab5-GDP, it is possible that rab5ip is involved in activating rab5. If so, the overexpression of rab5ip might stimulate endocytosis. To test this hypothesis, the rab5ipTM(ϩ) cDNA was inserted into a Sindbis virus vector, and recombinant virus was generated for infection of BHK-21 cells. The steady-state level of HRP accumulation by 60 min after infection was used as an index of endocytic activity. Cells infected with Sindbis-rab5ipTM(ϩ) recombinant internalized more HRP than did cells infected with vector alone but at a level similar to that of cells infected with Sindbis-rab5 Q79L (Fig. 8).
In Vitro Fusion-An antiserum was raised against a rab5ip peptide (residues 128 -146) located between the first two coiledcoil regions and having a high predicted antigenic index. This antibody (designated #9816) recognized rab5ip in immunoblots of extracts from cells overexpressing rab5ipTM(ϩ) (data not shown). In vitro endosome fusion was significantly inhibited by the addition of antiserum #9816 but not by the addition of an irrelevant antibody (Fig. 9). Furthermore, the inhibitory effect of anti-rab5ip antiserum was observed when endosome membranes were incubated with anti-rab5ip antiserum prior to fusion, but no inhibition was observed if the cytosol was preincubated with the antiserum instead (Fig. 9). Similar results were obtained using the affinity-purified antibody #9916 (data not shown).
Because rab5ip appears to interact preferentially with rab5-GDP, it seemed likely that rab5ip acts at an early step in endosome fusion, perhaps in activating rab5 at the endosome membrane. If this were so, we expected that recombinant rab5ipTM(Ϫ) would inhibit endosome fusion in vitro by sequestering rab5-GDP and preventing rab5 interaction within endosomes. Indeed, the addition of recombinant rab5ipTM(Ϫ), but not heat-inactivated rab5ipTM(Ϫ), strongly inhibited endosome fusion in a dose-dependent fashion (Fig. 10A). This inhibition occurred only if rab5ipTM(Ϫ) was added to fusion reactions early in incubation (Fig. 10B), further suggesting that rab5ip acts during the step of rab5 activation. The addition of  (12), with 1 mg/ml cytosol and endosome membranes. Anti-rab5ip antiserum #9816 or irrelevant IgG were used at a 100 g/ml concentration. 1, complete system; 2, complete system with irrelevant IgG; 3, complete system with anti-rab5ip; 4, membranes pretreated with anti-rab5ip, washed by centrifugation, and then used in the fusion assays; 5, cytosol pretreated with anti-rab5ip, then antibody removed using protein A-Sepharose prior to fusion reaction. Each point is the average of three determinations made in two separate experiments. more rab5 wt had little effect on the inhibition of endosome fusion by rab5ipTM(Ϫ), but the addition of rab5 Q79L completely reversed the inhibition (Fig. 10C). DISCUSSION We describe an rab5 interacting protein (rab5ip) that is distinct from other such proteins previously described. First, rab5ip is an integral membrane protein, unlike other rab5 interacting proteins (rabaptin-5, EEA1, and rabex-5), which are recruited to early endosome membranes by rab5-GTP (15,17). Second, rab5ip appears to interact preferentially with rab5-GDP, a property also distinct from EEA-1, rabex-5, and rabaptin-5.
The membrane association of rab5ip is indicated by the presence of a predicted transmembrane region near the N terminus of the protein (Fig. 1A), and immunoblot analysis shows that rab5ip is found in the particulate fraction (Fig. 4). rab5ip lacking the transmembrane region (rab5ipTM(Ϫ)) was also associated with membranes fractionated from transfected HeLa cells, although some rab5ipTM(Ϫ) appeared to be cytosolic (Fig. 4). It is likely that rab5ipTM(Ϫ) can associate with membranes in vivo, probably through rab5, but the association may not be completely preserved during fractionation of broken cells. It is also possible that overexpression of rab5ipTM(Ϫ) saturates membrane-bound rab5 and accumulates in the cytoplasm. 6His-Xp-tagged rab5ipTM(ϩ) was found exclusively in the membrane fraction in transfected HeLa cells, as was endogenous rab5ip (Fig. 4).
Several lines of evidence indicate that rab5ip binds rab5-GDP. First, rab5ipTM(Ϫ) binds preferentially in vitro to GST-rab5 wt or GST-rab5 Q79L loaded with GDP␤S (Fig. 2B). Second, rab5ipTM(Ϫ) interacts in vivo with GST-rab5 wt , but there is much less interaction with GST-rab5 Q79L (Fig. 6), a mutant that is predominantly GTP-bound in vivo (8). This conclusion is further supported by the finding that rab5ipTM(Ϫ) strongly localizes with GFP-rab5 G78A (Fig. 6), a mutant that, by analogy with the cognate mutant of p21ras (43), is most likely locked in the GDP conformation.
rab5ip has several regions of potential coiled-coil structure (Fig. 1A) that could be involved in protein-protein interactions, such as those that mediate the formation of SNARE complexes between membrane fusion partners (45). Another structural feature of note is a C-terminal region homologous to S. pombe Sad1, a protein that is associated with the spindle-pole body (Fig. 1B). Sad1 is also a transmembrane protein, and current evidence suggests that it is an anchor for molecular motors, serving to position the nucleus by migration along cytoplasmic microtubules (38). Another protein homologous to Sad1, UNC-84, is required for nuclear migration in C. elegans and may facilitate nuclear-centrosomal interactions (39). It is thus possible to speculate that a part of rab5ip's function involves an interaction with the cytoskeleton, possibly the microtubules. On the endocytic pathway, microtubules are abundant in the recycling compartment (46), which itself localizes with the perinuclear microtubule organizing center (47). Microtubules are required for trafficking from the sorting endosome to the late endosome (48), and microtubule disruption by nocodazole lowers the rate of transferrin receptor endocytosis in some studies (49,50). There is no evidence for an involvement of microtubules in homotypic endosome fusion (51). However, the movement of transferrin receptors from sorting to recycling endosomes is inhibited by nocodazole (52), a trafficking step that is also blocked by overexpression of activated rab5 (5). Of great interest is the recent finding that rab5 promotes the association of endosomes with microtubules and endosome motility (53). Additional studies are needed to determine whether rab5ip interacts with microtubules or some other element of the cytoskeleton.
Our evidence supports a role for rab5ip in the activation of rab5. Overexpression of rab5ipTM(ϩ) enhances the accumulation of a fluid phase marker by BHK-21 cells, to an extent similar to that caused by overexpression of rab5 Q79L (Fig. 8).
Although we could not detect changes in the rates of receptor endocytosis or recycling (transferrin and ␤ 2 -adrenoreceptor), GFP-rab5 wt -expressing Chinese hamster ovary cells show enlarged endosomes when infected with the Sindbis rab5ipTM(ϩ) recombinant (data not shown). It is possible that small differ- ences in receptor trafficking kinetics that are difficult to measure can cause the increased accumulation of fluid phase marker and changes in the morphology of GFP-rab5-containing vesicles. Additional evidence, consistent with a role for rab5ip in stimulating GDP release and activating rab5, is that rab5ipTM(Ϫ) preferentially binds rab5-GDP both in vitro (Fig.  2) and in vivo (Fig. 6). Moreover, the inhibition of endosome fusion by rab5ipTM(Ϫ) is not overcome by the addition of recombinant rab5 wt , probably because rab5 purified from E. coli is predominantly GDP-bound (9). rab5 Q79L can overcome the inhibition of fusion by rab5ipTM(Ϫ), because this rab5 mutant, in addition to showing a decreased rate of GTP hydrolysis (2.8-fold), also has an increased rate of nucleotide release (3.6-fold) (9). Thus rab5 Q79L more rapidly dissociates from GDP and reduces the proportion of rab5 able to interact with rab5ipTM(Ϫ). In support of this, we have observed that there is no inhibition of endosome fusion by rab5ipTM(Ϫ) when the reactions are carried out in the presence of GTP␥S (data not shown).
How might rab5ip contribute to the activation of rab5? Cytosolic rab proteins are escorted to and loaded onto endosome membranes by GDI (54). The release of GDI and the association of rabs with membranes are then followed by the exchange of GDP for GTP. It has been proposed that the displacement of GDI and subsequent nucleotide exchange are mediated by separate factors (55). The existence of a GDI-displacement factor is supported by the finding that yeast Sec4p (a rab homologue) in a complex with GDI is not an efficient substrate for nucleotide exchange promoted by the yeast exchange factor Dss4p (56). Similarly, the rabex-5-promoted nucleotide exchange by rab5 is inefficient when rab5-GDP is complexed to GDI (17). rab5ip could serve as a receptor for rab5-GDP, displacing GDI and then presenting rab5-GDP to the exchange factor rabex-5. Additional studies with multiple rab5 interactive proteins are needed to further test this hypothesis.