The FYVE domain of early endosome antigen 1 is required for both phosphatidylinositol 3-phosphate and Rab5 binding. Critical role of this dual interaction for endosomal localization.

Early endosome antigen 1 (EEA1) is 170-kDa polypeptide required for endosome fusion. EEA1 binds to both phosphtidylinositol 3-phosphate (PtdIns3P) and to Rab5-GTP in vitro, but the functional role of this dual interaction at the endosomal membrane is unclear. Here we have determined the structural features in EEA1 required for binding to these ligands. We have found that the FYVE domain is critical for both PtdIns3P and Rab5 binding. Whereas PtdIns3P binding only required the FYVE domain, Rab5 binding additionally required a 30-amino acid region directly adjacent to the FYVE domain. Microinjection of glutathione S-transferase fusion constructs into Cos cells revealed that the FYVE domain alone is insufficient for localization to cellular membranes; the upstream 30-amino acid region required for Rab5 binding must also be present for endosomal binding. The importance of Rab5 in membrane binding of EEA1 is underscored by the finding that the increased expression of wild-type Rab5 increases endosomal binding of EEA1 and decreases its dependence on PtdIns3P. Thus, the levels of Rab5 are rate-limiting for the recruitment of EEA1 to endosome membranes. PtdIns3P may play a role in modulating the Rab5 EEA1 interaction.

Early endosome antigen 1 (EEA1) is 170-kDa polypeptide required for endosome fusion. EEA1 binds to both phosphtidylinositol 3-phosphate (PtdIns3P) and to Rab5-GTP in vitro, but the functional role of this dual interaction at the endosomal membrane is unclear. Here we have determined the structural features in EEA1 required for binding to these ligands. We have found that the FYVE domain is critical for both PtdIns3P and Rab5 binding. Whereas PtdIns3P binding only required the FYVE domain, Rab5 binding additionally required a

30-amino acid region directly adjacent to the FYVE domain. Microinjection of glutathione S-transferase fusion constructs into Cos cells revealed that the FYVE domain alone is insufficient for localization to cellular membranes; the upstream 30-amino acid region required for
Rab5 binding must also be present for endosomal binding. The importance of Rab5 in membrane binding of EEA1 is underscored by the finding that the increased expression of wild-type Rab5 increases endosomal binding of EEA1 and decreases its dependence on PtdIns3P. Thus, the levels of Rab5 are rate-limiting for the recruitment of EEA1 to endosome membranes. PtdIns3P may play a role in modulating the Rab5 EEA1 interaction.
Early endosome antigen 1 (EEA1) 1 is a protein that was originally identified as the primary antigen in a patient with a subacute form of lupus erythematosus (1). Cloning and biochemical characterization of EEA1 revealed distinct structural features, including an extensive region of coiled-coil, an IQ calmodulin binding motif, and two Zn 2ϩ finger motifs, one at the amino terminus, and one at the extreme carboxyl terminus (2). This latter motif is conserved among several yeast, Caenorhabditis elegans, and mammalian proteins and was named "FYVE," after four proteins that contain this motif (2). The FYVE of EEA1 and of other proteins has been found to bind directly to PtdIns3P, a major product of PI3 kinase activity in mammalian cells (3)(4)(5)(6).
The activity of PI3 kinase has been strongly implicated in the control of endosomal membrane traffic. Treatment of cells with PI3 kinase inhibitors results in a decrease in transferrin receptor recycling, platelet-derived growth factor receptor downregulation, and transcytosis in epithelial cells (7)(8)(9)(10)(11)(12). In addition, wortmannin inhibits endosome fusion in in vitro assays (13,14). These effects coincide with an inhibition of EEA1 binding to early endosomal membranes (3,15), which accounts for the inhibition of endosome fusion in vitro, and may account for some of the wortmannin-induced phenotypes observed in intact cells.
Rab5 has been recognized as a critical regulatory factor in endosome fusion; a mutation in Rab5, which renders it resistant to GTP hydrolysis (Rab5Q79L), results in enlarged endosomes in vivo and enhanced fusion in in vitro endosome fusion assays (16,17). Conversely, mutations in Rab5, which render it resistant to exchange, result in small endosomes and inhibition of endocytosis (16,17). A relationship between Rab5 and PI3 kinase with respect to endosome fusion was first indicated by the ability of Rab5Q79L to rescue an in vitro assay of homotypic endosome fusion from a wortmannin block (14). The nature of this relationship was clarified with the identification of EEA1 as a critical effector protein downstream of both Rab5 and PI3 kinase (18). Exogenous EEA1 added to an in vitro endosome fusion assay was sufficient to rescue fusion both from a wortmannin block and from a block induced by removal of Rab5 by the addition of GDP dissociation inhibitor (19). From these studies it has been proposed that EEA1 is a critical factor required for endosome fusion, where it acts as a docking/tethering protein between vesicles to promote soluble N-ethylmaleimide-sensitive fusion attachment receptor (SNARE)-mediated fusion reactions (20,21).
Although it is clear that EEA1 binds directly to both PtdIns3P and to Rab5-GTP in vitro, the functional role of a dual interaction at the endosomal membrane is not clear. EEA1 co-localizes with Rab5 at early endosomes (2) but, unlike Rab5, does not associate with early endosome precursors such as clathrin-coated or adaptin-coated vesicles. It has been proposed that the combinatorial interaction of EEA1 with both Rab5-GTP and PtdIns3P ensures its precise localization to endosomes containing both signals, as opposed to cellular membranes containing exclusively one or the other (18). Alternatively, PtdIns3P and Rab5 may play a regulatory role in EEA1 function. The half-life of PtdIns3P and the duration of the GTP-bound state of Rab5 may influence the rate and extent of the endosome fusion reaction. In this regard, it is worth noting that introduction of a GFP tag at the COOH terminus of EEA1 results in the gross enlargement of early endosomes in vivo, coinciding with a loss of specificity for PtdIns3P binding (4).
To begin to understand the mechanisms and functional significance of the dual interaction of EEA1 with PtdIns3P and Rab5, we have determined the structural features in EEA1 required to bind these ligands. Our studies indicate that the integrity of the FYVE domain in conjunction with a stretch of 30 amino acids upstream are essential for Rab5 association. Thus, the PtdIns3P binding FYVE domain is nested within the Rab5 binding site, which explains the overlapping influence of both PI3 kinase and Rab5 activities on early endosome fusion. Microinjection of diverse constructs of EEA1 into cells revealed that the FYVE domain alone, which binds PtdIns3P in vitro with high affinity, is insufficient for membrane localization. The upstream 30 amino acids were absolutely required for membrane binding, which was further enhanced by the presence of the IQ motif. Thus, additional protein-protein interactions play a critical role in establishing FYVE/PtdIns3P-mediated protein-membrane associations.

Generation of Constructs and Recombinant
Protein-GST1277 was generated as described previously (4). Both GST1306 and GST1336 were generated by polymerase chain reaction amplification corresponding to residues 1306 -1411 and 1336 -1411 of EEA1, respectively, cloned in frame into the BamHI/SalI sites in pGEX-4T1 (Amersham Pharmacia Biotech), and sequenced for verification. Rab5c cDNA was isolated from a 3T3-F442A murine adipocyte cDNA expression library via [␣-32 P]GTP overlay. 2 pCMV5-Rab5c was constructed by EcoRI excision of the 768-base pair Rab5c cDNA insert from Bluescript-Rab5c obtained from cloning procedures and subcloned into EcoRI-digested pCMV5. Correct orientation of Rab5c was determined by restriction analysis and sequencing. The 10xHis-Rab5c construct, corresponding to residues 17-185, lacks both hypervariable regions but has normal function with respect to GTP binding/hydrolysis. It interacts with EEA1 in a nucleotidedependent manner, as does full-length Rab5c expressed as a GST fusion protein. 10xHis-Rab5c was constructed by polymerase chain reaction from the full-length cDNA, subcloned into a modified pET15b vector (Novagen) via unique BamHI/SalI sites, and sequenced for verification. Modifications to the pET15b vector include the addition of polyhistidine tags for a total of ten followed by a thrombin cleavage site aminoterminal to the BamHI site.
To isolate recombinant protein, the fusion constructs were used to transform competent DH5␣ or BL-21 cells (Life Technologies, Inc.), and the resulting fusion proteins were expressed by induction with 1 mM isopropyl-␤-D-thiogalactoside. The expressed fusion proteins were then isolated from bacterial lysates by affinity chromatography using a glutathione-Sepharose column (Amersham Pharmacia Biotech) for GST fusion proteins and a nickel nitrilotriacetic acid-agarose column (Qiagen) for His fusion proteins and were eluted according to the manufacturer's instructions.
Generation of Antibodies-Rab5c antiserum was prepared by injecting the Rab5c carboxyl-terminal peptide QNAAGAPGRTRGVDLQESN coupled to keyhole limpet hemocyanin via an amino-terminal cysteine residue into New Zealand White rabbits. Affinity purified Rab5c antibodies were purified by chromatographing the serum on a GST-Rab5cagarose column. EEA1 antiserum was prepared as above using the untagged form of GST1277 as antigen, and a GST1277 glutathione-Sepharose column to affinity purify the EEA1 antibodies. The untagged form of GST1277 was generated by incubation with thrombin at a 1:2000 (wt/wt) ratio at 4°C for 12-48 h in 50 mM Tris, pH 8, 2 mM CaCl 2 , and 0.1% mercaptoethanol, followed by incubation with glutathione-Sepharose to remove uncleaved fusion protein, and subsequent chromatographic steps to remove thrombin.
In Vitro Binding of GST Fusion Proteins-10xHis-Rab5c was incubated with 100 M GDP␤S or GTP␥S in A buffer (20 mM HEPES, pH 7.2, 100 mM NaCl, 0.5 mM MgCl 2 , 2 mM EDTA, and 1 mM dithiothreitol) at 30°C for 30 min followed by the addition of MgCl 2 to a final concentration of 20 mM. 5 g of GST fusion protein were bound to glutathione-Sepharose beads that had been preblocked in 5% nonfat dry milk in TBST (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Tween 20). Beads were then washed and incubated with 2 g of preloaded 10xHis-Rab5c for 1 h at 4°C in 300 l of cytosol buffer containing 10 mg/ml bovine serum albumin and 0.1% Tween 20. 5 mM N,N,NЈ,NЈ-tetrakis(2-pyri-dylmethyl)ethylenediamine (TPEN) (Sigma) were included in the incubation where indicated. Following centrifugation, the pellets were washed four times in 1 ml of A buffer, 20 mM MgCl 2 , 0.1% Tween 20. Bound proteins were resolved by SDS-polyacrylamide gel electrophoresis (PAGE) and transferred to nitrocellulose membrane, which were blocked in 5% dry milk, TBST followed by incubation with anti-His monoclonal antibody (Amersham Pharmacia Biotech). Filters were then incubated with horseradish peroxidase-conjugated goat anti-mouse antibody (Promega), which was detected by Renaissance enhanced luminol reagent (NEN Life Science Products).
Liposome Binding Assay-Liposome binding assays were performed as described previously (3) with minor modification. PS/PI liposomes were composed of 50% phosphatidylserine and 50% phosphatidylinositol (Avanti), whereas PI-3P liposomes contained 1% PtdIns3P (Matreya), 50% phosphatidylserine, and 49% phosphatidylinositol. Liposomes (100 mol of total lipid) were incubated for 15 min at room temperature with 5 g of GST fusion protein in 100 l of 50 mM HEPES, pH 7.2, 100 mM NaCl, and 1 mM MgCl 2 either in the absence or presence of 5 mM TPEN. After centrifugation, liposome pellets were analyzed by SDS-PAGE and Coomassie Blue staining.
Cell Culture and Transfection-Cos cells and 3T3-L1 fibroblasts were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (Life Technologies, Inc.). Chinese hamster ovary (CHO) fibroblasts grown in F12 medium containing 10% fetal bovine serum were transfected with empty pCMV5 or pCMV5-Rab5c using the calcium phosphate precipitation method. After 60 h, the cells were trypsinized and reseeded in medium containing 385 g/ml Geneticin (Life Technologies, Inc.). Surviving clones were isolated and recloned by limiting dilution. Overexpression of Rab5c protein in different clones was assessed by immunofluorescence microscopy and by immunoblotting cell lysates prepared from these clones with affinity purified anti-Rab5c antibodies.
Microinjection of Cells-Cos cells were grown to 40 -50% confluence on gridded coverslips and used for single cell microinjection in a 35-mm dish containing Dulbecco's modified Eagle's medium with 10% fetal calf serum. The cells were impaled using an Eppendorf model 5171 micromanipulator and injected with approximately 0.1 pl of a solution containing either GST1277, GST1306, or GST1336 fusion protein at various concentrations ranging from 0.1 to 1 mg/ml along with 20 g/ml rhodamine-coupled dextran, using an Eppendorf model 5246 transjector. Co-injection of fluorescent dextrans (Molecular Probes) was used to identify microinjected cells. Following microinjection of between 100 and 200 cells/coverslip within 30 min, the coverslips were placed in fresh medium for 2 h, incubated with or without 50 nM wortmannin for 10 min, and subsequently processed for immunofluorescence.
Immunofluorescence-Stable transfectants were grown to 40 -50% confluence on coverslips and incubated for 10 min without or with wortmannin as indicated. Coverslips were washed twice with cold phosphate-buffered saline, fixed in 4% formaldehyde/phosphate-buffered saline for 10 min at 4°C, and permeabilized with 0.2% Triton X-100/ phosphate-buffered saline for 10 min at 4°C. Cells were then blocked with 1% fetal bovine serum/0.5% Triton X-100 in phosphate-buffered saline for 30 min at 4°C. EEA1 was detected with human antiserum reactive with EEA1 (3) at a 1:10,000 dilution. Rab5c was detected with affinity purified rabbit anti-Rab5c antibody at approximately 10 g/ml. Microinjected GST fusion proteins were detected with anti-GST monoclonal antibody (Upstate Biotechnology) at 3 g/ml. Incubation with fluorescein isothiocyanate-conjugated goat anti-human antibody (Zymed Laboratories Inc.), rhodamine-conjugated goat anti-rabbit antibody, or fluorescein isothiocyanate-conjugated goat anti-mouse antibody (Jackson ImmunoResearch Laboratories) was used to visualize the proteins on a Zeiss Axiovert 35 M microscope.
Cell Fractionation-CHO cells were grown in 150-mm dishes. The cells were rinsed twice and swollen by incubation (10 min) in a 10-fold dilution of cytosol buffer (25 mM HEPES, pH 7, 125 mM potassium acetate, 2.5 mM magnesium acetate, 0.2 M sucrose, 1 mM dithiothreitol, 1 mM ATP, 5 mM creatine phosphate, 0.01 mg/ml creatine phosphokinase with proteinase and phosphatase inhibitors) in water. Cells were homogenized by repeated passage through a 27-g needle and centrifuged at 1000 ϫ g for 5 min. The post nuclear supernatants (PNS) were then incubated for 15 min at 37°C either with or without 50 nM wortmannin in the absence or presence of 100 M GTP␥S. Supernatant was then separated from particulate structures by centrifugation at 100,000 ϫ g for 15 min. Equal amounts of protein from the starting PNS and subsequent supernatants were separated by SDS-PAGE and analyzed by immunoblotting with anti-EEA1 polyclonal antiserum. 2 R. A. Heller-Harrison, manuscript in preparation.

The Binding of EEA1 to Rab5 and PtdIns3P Occurs through
Overlapping Structural Regions-We investigated the structural features in EEA1 that determine its binding to Rab5 and to PtdIns3P. We have previously shown that the carboxylterminal 134 amino acids of EEA1, expressed as a GST fusion protein (here referred to as GST1277), will bind specifically to liposomes containing PtdIns3P (4). In addition, this region has been shown in vitro to bind directly to Rab5-GTP (18). This region also contains an IQ motif, a putative calmodulin binding domain composed of extensive coiled-coil, which probably contributes to protein-protein interactions including homodimerization (22). To identify the minimal sequence within this region required to bind Rab5, a series of deletion constructs were generated in frame with GST (Fig. 1A), and their ability to associate with Rab5 was analyzed (Fig. 1B). The constructs were immobilized on glutathione-Sepharose beads and incubated with a 10xHis-tagged Rab5c protein that had been preloaded with either GTP␥S or GDP␤S. As expected, GST1277 bound to Rab5 in a GTP-dependent fashion. Interestingly, GST1306, which lacks the IQ motif and contains significantly less of the regions of predicted coiled-coil, also bound Rab5GTP effectively. A further deletion of 30 amino acids (GST1336) abolished binding (Fig. 1B). Thus, the 30-amino acid region upstream of the FYVE domain is the minimal region necessary for Rab5 binding.
To determine whether this 30-amino acid region was sufficient for Rab5 binding, a GST fusion protein encompassing amino acids 1306 -1357 was generated, and its ability to associate with Rab5-GTP was determined. This construct was unable to pull down either Rab5-GTP or Rab5-GDP (data not shown), indicating that additional structural elements are necessary for the interaction.
The ability of GST1277, GST1306, and GST1336 to bind PtdIns3P was also determined using a liposome binding assay (Fig. 1C). All three fusion proteins bound tightly and specifically to liposomes composed of 50% phosphatidylserine, 49% PtdIns, and 1% PtdIns3P (PI-3P), as opposed to liposomes composed of only 50% phosphatidylserine and 50% PtdIns (PS/ PI) (Fig. 1C). These results indicate that the FYVE domain in these constructs is completely functional and independent of the structural elements required to bind Rab5-GTP.
To determine whether the three-dimensional structure of the FYVE domain was required for Rab5 binding, we examined the ability of GST1277 to interact with Rab5 after treatment with the zinc chelator, TPEN. The ability of GST1277 to interact with Rab5 in a GTP-dependent manner was abrogated by the removal of Zn 2ϩ (Fig. 2A). As expected, Zn 2ϩ chelation also abrogated the binding of GST1277 to PtdIns3P (Fig. 2B). Thus, these data indicate that an integral FYVE domain, together with amino acids 1306 -1336 of EEA1, are necessary for Rab5-GTP binding.
Structural Elements Upstream of the Rab5 and PtdIns3P Binding Domain Enhance Binding to Endosomes-To directly compare the interactions observed in vitro to the interactions required for endosomal localization in intact cells, the GST1277, GST1306, and GST1336 fusion proteins were microinjected into COS cells, and their subcellular distribution was visualized after fixation by immunofluorescence with an anti-GST antibody (Fig. 3). GST1277 localized to punctate and ringshaped vesicular structures throughout the cytoplasm (Fig. 3A,  top left panel). Most of the injected protein was bound to endo-

FIG. 1. Delineation of the Rab5 binding site on EEA1.
A, schematic representation of the carboxyl terminus of EEA1 and deletion constraints thereof fused to GST at the indicated sites. B, the GST fusion proteins depicted above were immobilized on glutathione-Sepharose and incubated with 10xHis-Rab5c, previously loaded with GTP␥S or GDP␤S as described under "Materials and Methods." Protein bound to the Sepharose beads was analyzed by Western blotting with an anti-His antibody. C, GST fusion proteins were incubated with liposomes composed of 50% phosphatidylserine and 50% PtdIns (PS/PI) or 50% phosphatidylserine, 49% PtdIns, and 1% PtdIns3P (PI3P). Protein bound to liposomes was detected by PAGE and Coomassie Blue staining.

FIG. 2. A functional FYVE domain is required for Rab5-GTP binding.
A, 10xHis-Rab5c preloaded with GTP␥S or GDP␤S was incubated with immobilized GST1277, either in the presence or absence of 5 mM TPEN, and bound 10xHis-Rab5c was detected by Western blotting as in Fig. 1B. B, GST1277 was incubated with either PS/PI or PI-3P liposomes in the presence or absence of 5 mM TPEN, and the amount of GST protein bound was measured by PAGE and Coomassie Blue staining. somes, with very little diffuse staining observed, which is emphasized by comparison to dextran localization within the same cell (Fig. 3A, top right panel). Wortmannin treatment of cells injected with GST1277 completely reverses endosomal localization, indicating that PtdIns3P production is crucial for membrane binding (Fig. 3A, bottom left panel). Binding to PtdIns3P is not sufficient, however, because microinjected GST1336, which in vitro binds to PtdIns3P with an apparent affinity similar to that of GST1277 (Fig. 1C), failed to localize to endosomal membranes (Fig. 3B, bottom left). Even at much higher concentrations this construct remained cytosolic, as determined by the similarity between the fluorescence images of the GST and dextran signals (Fig. 3B, bottom panels).
Interestingly, the localization of GST1306, which in vitro binds Rab5 as well as PtdIns3P (Fig. 1, B and C), was substantially more diffuse than the localization of GST1277 (compare Fig. 3, A and B, top left panels). The number and intensity of the vesicular structures marked with GST1306 increased with the concentration of protein injected (not illustrated). In contrast, even at very low concentrations GST1277 consistently localized to punctate vesicles, and diffuse staining was never observed. These results indicate that interactions that are not apparent in in vitro binding assays for Rab5 and PtdIns3P are crucial for efficient endosomal localization within the cell. These interactions may be directly dependent on the IQ motif contained within amino acids 1288 to 1300 of EEA1 and may represent additional protein-protein interactions. That these interactions are not crucial for endosomal localization may explain the failure to detect an effect of mutation of two conserved residues within the IQ motif on the endosomal localization of EEA1 (2).
Wild-type Rab5 Levels Are Rate-limiting for EEA1 Binding to Membranes-The results shown above demonstrate that the simultaneous occurrence of PtdIns3P and Rab5 binding are required for the interaction of the COOH terminus of EEA1 with the endosome. Furthermore, binding of these two ligands occurs through overlapping structural domains. To delineate how endogenous Rab5 and PtdIns3P influence the binding of EEA1 to cellular membranes, we took advantage of our finding that binding of EEA1 to membranes can be observed in broken cells. PNS obtained from pCMV5-transfected CHO cells was placed at 37°C for 20 min, and the membrane fraction was then removed by high speed centrifugation. Approximately 50% of EEA1 was present in the supernatant (Fig. 4A, lane 1). Inclusion of GTP␥S to maximally activate small GTPases markedly enhanced binding to membranes, with only 20% of EEA1 remaining in the supernatant (Fig. 4A, lane 3). Inclusion of wortmannin during the 20-min incubation inhibited binding almost completely, resulting in 90% of EEA1 remaining in the supernatant (Fig. 4A, lane 2). The presence of GTP␥S did not overcome the inhibitory effect of wortmannin on EEA1 binding to membranes, indicating that even with maximal activation of Rab5, EEA1 binding was still dependent on PtdIns3P (Fig. 4A,  lane 4). Thus, at normal Rab5 levels, PtdIns3P is critical for EEA1 binding.
We next examined whether the levels of expression of Rab5 would influence the interaction of EEA1 with endosomes. CHO cells were stably transfected with Rab5c, and the properties of PNS obtained from Rab5c-overexpressing cells were analyzed (Fig. 4B). Overexpression of Rab5c enhanced binding of EEA1 to the membrane, as only 10 -20% of EEA1 could be detected in the supernatant (Fig. 4B, lane 1). GTP␥S further enhanced binding, with virtually a complete disappearance of EEA1 from the supernatants (Fig. 4B, lane 3). Inclusion of wortmannin under these conditions had only a modest inhibitory effect on EEA1 binding (Fig. 4B, lanes 2 and 4). Thus, overexpression of Rab5c not only enhances association of EEA1 with membranes, but renders the interaction insensitive to the levels of PtdIns3P.
The results observed in broken cells are supported by the observation that Rab5c-overexpressing cells display a more intense EEA1 staining pattern localized to peripheral vesicular and ring-shaped tubular structures as compared with cells transfected with vector alone (Fig. 5). In these cells some of the Rab5c co-localized with EEA1 on vesicular structures but most was present in more perinuclear structures (Fig. 5, bottom  right panel). Thus, not all Rab5-positive structures contained EEA1; however, overexpression of wild-type Rab5c markedly enhanced EEA1 binding in vivo and led to larger EEA1-containing structures. These results suggest that endosome fusion is directly regulated by the levels of Rab5, which determines the degree of interaction of EEA1 with membranes.
We next examined the effect of PI3 kinase inhibition on EEA1 membrane association in the Rab5c overexpressing cells. Immunofluorescence analysis revealed the majority of EEA1 in association with membranes both in the presence or absence of wortmannin treatment (Fig. 6), in accordance with the fractionation results shown in Fig. 4. Interestingly, whereas in nontreated cells the overexpressed Rab5 was relatively diffuse and only partially co-localized with EEA1 (Fig. 5, bottom, and  Fig. 6, top), in wortmannin-treated cells Rab5c was found in a tight perinuclear cluster, where it clearly co-localized completely with EEA1 (Fig. 6, compare bottom left and right panels). Thus, wortmannin treatment appears to cause an alteration in the partitioning of Rab5c among intracellular compartments, suggesting that PI3 kinase activity may influence the GTP cycle on Rab5.

DISCUSSION
Recent work from many laboratories has contributed to the identification of the specific molecular elements involved in the process of endosome fusion (20,(23)(24)(25)(26). Among these elements are the small GTPase Rab5, PI3 kinase, and the protein EEA1, which interacts functionally with both of these effectors and is thought to provide a tethering system necessary for membrane fusion. Endosome fusion also depends critically on SNARE proteins, an integral part of the fusion system, which directly mediate the formation of the fusion pore (27,28). How these systems of tethering and fusion, each of which have components that are likely to play regulatory roles, operate in the context of intact cells is currently an unanswered question.
In this manuscript, we have addressed two specific questions related to the mechanisms that regulate endosome fusion in intact cells. First, what is the structural basis for the interaction between Rab5, PtdIns3P, and EEA1, and second, what is the physiological relevance of these interactions in intact cells.
Our results indicate that (a) Rab5 binding by EEA1 requires the integrity of the FYVE domain plus a 30-amino acid NH 2terminal extension; (b) the FYVE domain of EEA1 alone, although capable of high affinity binding to PtdIns3P, cannot bind to endosomes nor to any other membrane system in intact cells; both PtdIns3P and Rab5 binding are required for binding of EEA1 to endosomes; and (c) that the rate-limiting factor for EEA1 association with endosomes in vivo is the level of activated Rab5.
Our conclusion that an efficient interaction between EEA1 and Rab5 requires both the integrity of the FYVE domain and a 30-amino acid sequence to the NH 2 terminus contrasts with that reached by Simonsen et al. (18). In these studies the minimal Rab5 binding region of EEA1 was defined as being comprised of residues 1277-1348, excluding the FYVE domain. It is possible that a weak interaction between this region and a persistently active mutant of Rab5 may be detectable in the two-hybrid system used in these studies but not in a biochemical pull-down assay such as the one used here. However, this weak interaction is clearly functionally insufficient, as mutations in the FYVE domain completely prevent binding of EEA1 to endosomes even in the presence of persistently active Rab5 (2). Overexpression of this mutant form of Rab5 abrogates the normal requirement of PtdIns3P association for EEA1 binding to endosomal membranes (18). Thus the inability of the FYVE domain mutants to localize to endosomes in cells expressing persistently activated Rab5 is likely to be due to the disruption of Rab5 binding.
A dual role of the FYVE domain in mediating the interaction both with PtdIns3P and Rab5 underscores the functional similarity between EEA1 and the yeast protein Vac1p. Point mutations in the FYVE domain of Vac1p abolished its interaction with Vps21p, a Rab GTPase that functions in Golgi to endosome transport (29). Similarly, other Zn 2ϩ binding domains are present in Rab effector proteins and contribute directly or indirectly to the interaction with activated Rab GTPases. For example, Rabphilin-3A contains a Zn 2ϩ binding domain within its Rab3A binding region, and mutations that disrupt Zn 2ϩ binding abolish Rab3A association (30). Comparison of the crystal structure of the FYVE domain of Vps27 with that of the FIG. 5. Overexpression of Rab5c enhances EEA1 association with vesicular structures in vivo. CHO cells stably transfected with either Rab5c or the pCMV5 vector alone were fixed with 4% formaldehyde and double-stained using a human antiserum reactive to EEA1 (left panels) and a rabbit polyclonal antibody reactive to Rab5c (right panels). Primary antibodies were detected using fluorescein isothiocyanate-conjugated anti-human and rhodamine-conjugated anti-rabbit secondary antibodies.
Rab3A-Rabphilin-3A complex suggests a potential interaction between the COOH terminus of Rab5 and the FYVE domain of EEA1 (31).
What is the physiological role of the dual interaction of EEA1 with PtdIns3P and Rab5? PI3 kinase, Rab5, and EEA1 have been recognized as critical components in the control of endosome fusion, but the mechanisms underlying their interplay are not understood. One current hypothesis is that Rab5 provides a mechanism to ensure targeting of EEA1 to early endosomes and not to other cellular membranes that might contain PtdIns3P (18). However, the finding that PtdIns3P binding by the FYVE domain is not sufficient for EEA1 association with any cellular membrane suggests that Rab5 binding is fundamentally required for the binding event itself. Furthermore, overexpression of wild-type Rab5 enhances EEA1 binding to the point where the FYVE-PtdIns3P interaction becomes unnecessary for endosomal localization. These results suggest that the principal role of Rab5 is to direct the membrane binding of EEA1 and that the FYVE-PtdIns3P interaction is likely to be regulatory.
What type of regulatory role might PtdIns3P play in endosome fusion? Experiments using fusion proteins of the COOH domain of EEA1 indicate that both Rab5 and PtdIns3P are required to achieve the binding of EEA1 to endosomal membranes (Fig. 3). These results suggest that these signals play a synergistic role in mediating EEA1 binding to endosomal membranes and in subsequent endosome fusion. The prediction from this conclusion would be that inhibition of PI3 kinase activity with wortmannin would result in an inhibition of endosome fusion. Although this result is indeed observed in in vitro endosome fusion assays (13,14), results in intact cells differ significantly. For example, wortmannin increases the accumulation of the fluid phase marker Lucifer Yellow and of transferrin in early endosomes and causes only a small decrease in the rate of transferrin recycling (9,11,12). Furthermore, a pronounced enlargement and tubulation of endosomes containing these markers is observed after treatment with this toxin (11). This phenotype is not consistent with a block in endosome fusion. In fact, it more closely resembles the phenotype of enhanced endosome fusion produced by expression of persistently activated forms of Rab5 (16).
The discrepancy between the apparent requirement for PtdIns3P to bind EEA1 to membranes and the enhanced fusion of endosomes observed in intact cells after wortmannin treat-ment suggests more than one specific role for PtdIns3P during the process of endocytosis. For example, PtdIns3P might serve to initiate the recycling of EEA1 and other components of the fusion machinery after endosome fusion, thus preventing the formation of large endsomes while promoting the continual formation of new ones. An intriguing possibility supported by results shown here (Fig. 6) is that PI3 kinase activity may be a critical factor in regulating Rab5 function. Whereas in control cells Rab5 was diffusely distributed, in wortmannin-treated cells it was largely concentrated in enlarged juxtanuclear endosomes that also contained EEA1. These enlarged endosomes might arise from a persistent activation of Rab5 resulting in enhanced EEA1 binding and an uncontrolled increase in the number of endosome fusion events. Further studies aimed at better understanding the interplay among PtdIns3P, EEA1, and Rab5 are necessary to clarify these questions.
Microinjection experiments shown here indicate that whereas the presence of both the FYVE and the upstream sequence required for Rab5 binding are sufficient for endosome localization, the presence of the IQ motif amino-terminal to these sites enhances this interaction. This enhanced interaction might be due to oligomerization of the carboxyl terminus of EEA1 with itself, as amino acids 1277-1411 but not 1307-1411 were able to homodimerize in a yeast two-hybrid assay (22). Oligomerization of the protein would result in multiple Rab5-GTP/PtdIns3P binding sites in one complex, increasing its affinity for membranes containing both signals, thereby stabilizing the protein at the membrane. Alternatively, additional interactions with other endosome components might be involved. In the case of Vac1p, interactions with components of the SNARE fusion machinery have been demonstrated (29). Such interactions might also occur in mammalian cells. Whether these interactions directly involve the IQ domain of EEA1 remains to be determined. CHO cells stably transfected with Rab5c were treated (ϩWort) or not (-Wort) with 50 nM wortmannin for 15 min. EEA1 and Rab5c were detected as described in Fig.  5. Arrows show areas where marked colocalization of EEA1 and Rab5c was observed.