EHD2 and the novel EH domain binding protein EHBP1 couple endocytosis to the actin cytoskeleton.

Here we identified two novel proteins denoted EH domain protein 2 (EHD2) and EHD2-binding protein 1 (EHBP1) that link clathrin-mediated endocytosis to the actin cytoskeleton. EHD2 contains an N-terminal P-loop and a C-terminal EH domain that interacts with NPF repeats in EHBP1. Disruption of EHD2 or EHBP1 function by small interfering RNA-mediated gene silencing inhibits endocytosis of transferrin into EEA1-positive endosomes as well as GLUT4 endocytosis into cultured adipocytes. EHD2 localizes with cortical actin filaments, whereas EHBP1 contains a putative actin-binding calponin homology domain. High expression of EHD2 or EHBP1 in intact cells mediates extensive actin reorganization. Thus EHD2 appears to connect endocytosis to the actin cytoskeleton through interactions of its N-terminal domain with membranes and its C-terminal EH domain with the novel EHBP1 protein.

Components of the clathrin-dependent process whereby receptors and transporters are removed from the cell surface by endocytosis into intracellular vesicles have been well characterized (1). Following coating, invagination, and scission of internalized membrane, early endosomes containing the GTPase Rab5 and the endosomal protein EEA1 are formed, directing cargo to various destinations including the recycling endosome for transit back to the plasma membrane or to late endosomes and lysosomes where degradation of proteins occurs (2). Although many of these steps have been studied in detail, a major gap in understanding endocytosis relates to the role of the actin cytoskeleton. In yeast, genetic analysis clearly shows a requirement of the actin cytoskeleton for endocytosis (3,4). This connection has been further strengthened by the identification of proteins in the Saccharomyces cerevisiae endocytic pathway that also bind actin or indirectly regulate actin. One of these proteins, Sla2p (5), binds actin directly through a talinlike domain, whereas two others, Pan1p (6) and Abp1p (7), are activators of actin assembly nucleated by Arp2/3 complex. Genetic confirmation of the functional significance of these interactions provides strong support for the concept that actin polymerization can be coupled to the endocytic machinery, generating forces necessary for endocytosis.
In mammalian cells the link between endocytosis and actin dynamics is less clear. Introducing actin-perturbing drugs or mutant forms of the Rho family of small GTPases disrupts endocytosis in some cells but not in all cases (8 -11). However, analogous to studies in yeast, the case for a role of the actin cytoskeleton in mammalian endocytosis has been strengthened recently by identification of proteins that modulate actin dynamics and also interact with endocytic components (12,13). These include mouse Abp1 (7) and mHip1R (14), a protein closely related to the human Huntington interacting protein Hip1, and the yeast Sla2p, which bind actin directly. Two other proteins, intersectin (15) and syndapin (16), bind N-WASP, which in turn induces actin polymerization via Arp2/3. All of these proteins also bind components of the endocytic machinery such as clathrin-associated proteins or dynamin and thus could coordinately regulate actin assembly and trafficking events. However, potential roles played by actin filaments at different stages of endocytosis and the structures of complexes at the interface of endocytic components and the actin cytoskeleton remain to be clarified.
Using the recycling of GLUT4 glucose transporter proteins in cultured adipocytes as a model system, we identify here two novel proteins, EHD2 1 and EHBP1, that appear to connect actin dynamics with endocytosis. In this system, insertion of GLUT4 into the plasma membrane in response to insulin is accompanied by its rapid retrieval by the clathrin-dependent pathway (17)(18)(19). EHD2 is detected in membranes that contain some of the intracellular GLUT4. Through its EH domain, which is similar to those present in many proteins involved in endocytosis, EHD2 binds NPF motifs in EHBP1. Both EHD2 and EHBP1 also connect to the cytoskeleton through an acidic motif in the former and through a CH domain in the latter. High expression of either EHD2 or EHBP1 impairs endocytosis of GLUT4 in cultured adipocytes and transferrin in COS cells. Furthermore, loss of the endogenous EHD2 or EHBP1 mediated by siRNA gene silencing inhibits transferrin endocytosis, indicating that these interacting proteins function at an early stage of endocytosis near the plasma membrane by providing functional linkages to the actin cytoskeleton.

EXPERIMENTAL PROCEDURES
Materials-Anti-EHD2 antibody was generated in rabbits using a GST-EHD2 fusion protein as an antigen and the IgG fraction isolated from rabbit serum. Anti-EHBP1 antibody was prepared by immunizing rabbits with a GST fusion protein containing residues 360 -715 of EHBP1 sequence. Anti-EHBP1 was then affinity-purified from serum, using a cyanogen bromide-conjugated EHBP1 (GST-(360 -715)) column. In order to test the specificity of anti-EHD2 and anti-EHBP1 antibodies, immunoblot analysis from lysates of cells expressing GFP-empty vector, GFP-EHD2, or GFP-EHBP1 were performed, using pre-immune or immune serums. Anti-EHD2 or anti-EHBP1, but not the pre-immune serum, recognizes specifically expressed GFP-EHD2 or GFP-EHBP1 proteins, respectively. Anti-EHD2-rhodamine red-and anti-EHBP1-Alexa488-conjugated antibodies were generated using protein labeling kits (Molecular Probes) following the manufacturer's instructions. Goat anti-GLUT4 polyclonal antibody (C-20) and mouse anti-GLUT4 monoclonal antibody (clone IF8) used in whole mount and in freeze deep-etch EM analyses were purchased from Santa Cruz Biotechnology and Biogenesis, respectively. Rabbit anti-actin polyclonal antibody was a gift from Dr. Christine Chaponnier (University of Geneva, Switzerland). Mouse anti-Myc epitope (clone 9E10) monoclonal antibody was from NeoMarkers, Inc. Rabbit anti-HA polyclonal antibody was produced as described (20). Rhodamine-phalloidin, rhodamine-transferrin, goat anti-rabbit-Alexa350-conjugated and goat anti-mouse-Alexa594-conjugated antibodies were from Molecular Probes. Goat anti-rabbit FITCconjugated antibody was from BIOSOURCE International (Camarillo, CA). The antibodies conjugated to 6-and 12-nm gold particles used in immunoelectron microscopy analysis were purchased from Jackson ImmunoResearch.
Constructs-The murine EST encoding mouse EHD2 (GenBank™ accession number AI787872) was obtained from Genome Systems (St. Louis, MO). The plasmid DNA was isolated and sequenced, and the full-length mEHD2 was used to generate the different constructs used in this study. The plasmids expressing GFP-EHD2 (aa 1-543), GFP-⌬EH-EHD2 (aa 1-444), GFP-⌬EH-⌬A-EHD2 (aa 1-428), and GFP-EH (aa 444 -543) were constructed by PCR amplification using primers creating XhoI and BamHI sites at the 5Ј and 3Ј ends, respectively. The PCR products were subcloned in-frame with a pEGFP-C2 vector. The plasmids expressing HA-EHD2, HA-⌬EH-EHD2, or HA-EH were similarly constructed by PCR amplification using primers creating KpnI and BamHI sites at the 5Ј and 3Ј ends, respectively. The PCR products were then subcloned in-frame with a 3XHA-pCMV5 vector. The two overlapping ESTs (aa 1-310; accession number BAA91391 and aa 235-1196; accession number BAA74926) encoding human EHBP1 were obtained from Kazusa DNA Research Institute and from Institute of Medical Science, University of Tokyo, Japan, respectively. The plasmid DNAs were isolated and sequenced, and the two ESTs were fused to generate the complete EHBP1 cDNA sequence. The plasmid expressing the full-length EHBP1 (aa 1-1196) was constructed by PCR amplification using primers creating KpnI and XmaI sites at the 5Ј and 3Ј ends, respectively. The PCR products were subcloned in-frame with a 3XHA-pCMV5 or a pEGFP-C3 vector to express HA-EHBP1 or GFP-EHBP1. For antibody production and in vitro pull-down assays, full-length or fragments of EHD2 or EHBP1 were expressed as glutathione S-transferase (GST) fusion proteins in Escherichia coli. EHD2, EHBP1 (aa 360 -715), and the NPF motifs from EHBP1 (aa 227-407) were PCRamplified by using primers generating BamHI and XhoI restriction sites at the 5Ј and 3Ј ends, respectively. The PCR products were subcloned in-frame with a pGEX5x3 vector. The Myc-GLUT4-EGFP construct has been described previously (21). EGFP-EEA1 was generated as described (22). All the constructs were sequenced for verification prior to transfections.
Cell Culture and Transfection of Differentiated 3T3-L1 Adipocytes-3T3-L1 fibroblasts were grown to confluency and differentiated as described previously (23). Differentiated 3T3-L1 adipocytes were transfected by electroporation as described (24). The cells were then re-plated and allowed to recover for 24 h before serum starvation for 3 h and stimulation with insulin. For the myc-GLUT4-GFP internalization assays, adipocytes were cotransfected with 50 g of myc-GLUT4-EGFP in combination with either 150 g of pCMV5-3XHA-EHD2, 3XHA⌬EH-EHD2, or 3XHA-EH. The effect of cortical actin rearrangement by expression of the GFP-EHD2 constructs was assayed in 3T3-L1 adipocyte electroporated with 50 g of the plasmid encoding the indicated construct.
siRNA-induced Degradation of EHD2 and EHBP1-The siRNA species purchased from Dharmacon were designed to target the following cDNA sequences: scrambled, 5Ј-CAGTCGCGTTTGCGACTGG-3Ј; EHD2-siRNA, 5Ј-AAGAAAGAGATGCCCACGGTGTT-3Ј; EHBP1-siRNA 1, AAGCTCTTGCCACCAGCAGCATT-3Ј; EHBP1-siRNA 2, AA-GAGGAGAAGGCGGCAAAAATT-3Ј. Either 20 nmol of scrambled siRNA, 10 nmol of EHD2-siRNA, or 10 nmol of each of the EHBP1-siRNA species were electroporated into 3T3-L1 fibroblasts as described (24). Briefly, fibroblasts were detached from culture dishes with 0.25% trypsin in PBS, washed twice, and resuspended in PBS. Half of the cells from one 150-mm dish were then mixed with siRNA, which was delivered to the cells by a pulse of electroporation with a Bio-Rad gene pulser II system at the setting of 0.18 kV and 950-microfarad capacitance. Using a Cy3-tagged siRNA for lamin A/C, we showed previously that Cy3-siRNA was introduced with virtually 100% efficiency into cells using this method, and nearly all cells showed loss of nuclear lamin A/C (24). After electroporation the cells were reseeded in 12-well plates and allowed to rest for 48 h. Alexa594-transferrin uptake was then assayed as described below. A portion of these cells was analyzed for EHD2, EHBP1, and actin by Western blotting.
Mass Spectral Analysis, Northern and Western Blotting-Rat fat cell subcellular fractions were prepared using a differential centrifugation procedure described previously (25). Briefly, the plasma membrane fraction was obtained after 20 min of centrifugation at 16,700 ϫ g followed by centrifugation through sucrose. The high density microsomes were obtained by centrifuging the 16,700 ϫ g supernatant at 38,700 ϫ g for 20 min, and the low density microsomes were obtained by spinning the 38,700 ϫ g supernatant at 150,000 ϫ g for 90 min. The supernatant from the 150,000 ϫ g centrifugation was concentrated using a Centriprep apparatus to obtain the cytosolic fraction. These fractions have been characterized in detail previously (52). The isolation and fractionation of GLUT4-containing vesicles were carried out as described (25). The proteins were resolved by SDS-PAGE and visualized by silver staining (Bio-Rad), and the bands were excised. The peptide sequences were determined by matrix-assisted laser desorption ionization-time of flight-mass spectrometry analysis as described (25). To examine the subcellular distribution of endogenous EHD2 in adipocytes, subcellular fractions were prepared as described before, and 50 g of fractions indicated in Fig. 1D were loaded onto 10% polyacrylamide gels and transferred to nitrocellulose membranes. EHD2 was detected by immunoblotting, using anti-EHD2 antibody. As depicted in Fig. 1D, a 65-kDa band, corresponding to the molecular weight of EHD2, was detected by anti-EHD2 but not pre-immune serum (data not shown). The EHBP1 protein was detected by immunoblotting with affinity-purified anti-EHBP1 antibody. For Northern blot analysis, total RNA from 3T3-L1 fibroblasts, differentiated adipocytes, and different mouse tissues was isolated as described (20,23). 10 g of total RNA was resolved on a 1.2% denaturing gel and was transferred to a Nytran membrane. DNA probe corresponding to nucleotides 1606 -1881, unique to mouse EHD2, was prepared by PCR amplification, labeled with [␣-32 P]dCTP, and hybridized to the membrane at 42°C. A phosphor screen was scanned using an Amersham Biosciences Storm 860 Scanner.
Rhodamine-Transferrin Uptake and Immunofluorescence Microscopy-To assay transferrin uptake, COS-1 cells were transfected, using the calcium precipitation method, with either empty EGFP vector, GFP-EHD2, GFP-⌬EH-EHD2, or GFP-EH constructs, and after 24 h the cells were serum-starved in Krebs-Ringer/HEPES ϩ 0.5% bovine serum albumin ϩ 2 mM sodium pyruvate for 2 h. They were then incubated in the presence of 5 g/ml transferrin-rhodamine for 5 min. The cells were then fixed with 4% formaldehyde and observed using confocal microscopy. To visualize endogenous actin and EHD2, 3T3-L1 adipocytes were fixed with 4% formaldehyde in phosphate-buffered saline (PBS), permeabilized, and blocked with 0.5% Triton X-100 and 1% fetal bovine serum for 20 min. Cells were incubated with primary antibodies for 2 h and with rhodamine-phalloidin and FITC-conjugated secondary antibody for 30 min. To analyze Myc-GLUT4-GFP internalization in adipocytes, cells were first cotransfected with plasmid DNA(s) as indicated. Following this, adipocytes were treated with 0.2 M insulin for 45 min, washed with iced-cold PBS, and incubated on ice for 1 h with mouse anti-Myc monoclonal antibody. The cells were then washed with ice-cold PBS, and the myc-GLUT4-GFP internalization was initiated by warming the cells at 37°C. At different time points, cells were fixed, permeabilized, and immunostained with Alexa594-labeled antimouse secondary antibody to visualize myc-GLUT4-GFP. The cells expressing the HA-tagged proteins were visualized by immunostaining with polyclonal anti-HA antibody and Alexa350-conjugated secondary antibody. Images were taken with an Olympus IX-70 microscope with CCD camera and then processed using Metamorph software.
Live Cell Imaging-For live cell imaging COS-1 cells were transfected with GFP-tagged constructs, and after 24 h, cells were serumstarved for 2 h and then incubated for 7 min with Texas Red transferrin. Cells were washed with KHR buffer (25 mM HEPES, 125 mM NaCl, 5 mM KCl, 1.3 mM CaCl 2 , 1.2 mM MgSO 4 ) followed by a 30-min chase with unlabeled transferrin. Cells were imaged using high speed, three-dimensional microscopy as described (22). The microscope was configured to 133 nm/pixels using a ϫ60 objective, and the laser illumination was configured to provide a 488-nm excitation wavelength with a flux on specimen of ϳ12 watts/cm 2 . Exposure times of 4 (GFP) and 15 ms (Texas Red) were used to acquire each of 21 optical sections, spaced by 250 nm. Each set of 21 optical sections was acquired in less than 1 s, allowing 20 ms for each 250-nm shift in focus. Stacks were acquired every 20 s for 30 continuous minutes. The haze originating from light sources outside the in-focus plane of the cell was reduced by image restoration. Stacks were projected into single two-dimensional images, which were concatenated into a QuickTime video format. After imaging the cells were fixed, permeabilized, and stained with anti-HA and Alexa350-conjugated secondary antibodies in order to verify coexpression of GFP-EEA1 and HA-EHD2. Enlargement of early endosomes and increased cytosolic localization of GFP-EEA1 was induced by coexpressing HA-EHD2 (data not shown). The same phenotype was observed in the live cell coexpressing GFP-EEA1 and HA-EHD2 (Fig. 4, bottom panel).
Electron Microscopy and Immunogold Labeling Analysis-Wholemount electron microscopy analysis was performed as described (20). Double immunogold labeling with different primary antibodies (rabbit anti-EHD2 with goat anti-GLUT4 or, as a negative control, rabbit non-immune IgG with goat non-immune IgG, all antibodies used at 10 g/ml) and specific secondary antibodies conjugated to 6 and 12 nm gold particles were carried out on parafilm sheets at room temperature. After immunogold labeling, the cell preparations were fixed and processed as described (20). For immunogold labeling of EHD2, GLUT4, and actin at plasma membrane, lawns from 3T3-L1 adipocytes were generated as described (23). Immediately after preparation, membrane lawns were fixed on coverslips with 3.7% formaldehyde and labeled with primary antibodies, followed by 6-and 12-nm gold-tagged secondary antibodies. Samples were fixed with 2.5% glutaraldehyde in PBS for 30 min. An alternate method (26), which employs the use of the transitional solvent hexamethyldisilazane for optimally drying soft tissues, has been used widely to preserve the fine surface details of soft tissue specimens without subjecting the specimens to the extreme pressures of Critical Point Drying or the prolonged periods at high vacuum needed for freeze-drying. Whole-mount grids were examined on a Philips CM 10 transmission electron microscope at 80 kV.
Expression and Purification of Recombinant Proteins-The expressed fusion proteins were isolated using glutathione-agarose beads. To isolate recombinant protein, the fusion GST constructs were used to transform competent BL-21 cells, and the resulting fusion proteins were expressed by induction with isopropyl-␤-thiogalactoside. EHD2, ⌬EH-EHD2, and NPF-containing motifs of EHBP1 fusion proteins were FIG. 1. EHD2 identified in GLUT4-containing membranes and its localization in 3T3-L1 adipocytes. A, sequences of EHD2 peptides obtained from GLUT4-containing membranes purified by equilibrium density gradient centrifugation. GLUT4-containing membranes were isolated from primary adipocyte low density microsome fractions by a velocity sedimentation method as described (25). Fractions (8 -18) containing the insulin-sensitive pool of GLUT4-membrane (4 -11) were then combined and pelleted, resolved by a 5-15% gradient SDS-PAGE, and visualized with silver stain. Bands were excised and subjected to proteolytic digestion prior to analysis by mass spectrometry. Peptides from an EH domain containing protein are identified at the top of A. Arrows indicate other proteins identified. B, overall structure and deduced sequence of EHD2 protein. Within the EHD2 sequence shown, the P-loop motif is boxed with a dashed line, the region corresponding to a predicted coiled-coil is underlined, and the region corresponding to the EH domain is boxed with a solid line. C, multitissue Northern blot of EHD2. 10 g of total RNA from different mouse tissues (top left panel) and 3T3-L1 fibroblasts (F) and adipocytes (A) (top right panel) were resolved on a denaturing formaldehyde gel. The RNA was then transferred to a Nytran® membrane and probed with PCR-amplified, random primer labeled EHD2 specific probe, corresponding to the region 1606 -1881 in the mouse EHD2 cDNA. Bottom panels depict the corresponding agarose gels showed in the top panel, stained with ethidium bromide to detect total RNA in each samples, as a loading control. EHD2 mRNA is indicated by the arrowhead. D, left panel, EHD2 protein is enriched in plasma membrane fractions. 50 g of cytosol, high density microsomes (HDM), low density microsomes (LDM), and plasma membrane (PM) subcellular fractions from rat fat cells were immunoblotted for EHD2. Arrow indicates EHD2 protein. Right panel, EHD2 localizes in ring-shaped membrane structures localized close to the cell surface. Differentiated 3T3-L1 adipocytes were fixed, permeabilized, and stained with rabbit anti-EHD2 polyclonal antibody and FITC-conjugated secondary antibody to visualize endogenous EHD2 protein. E, ring-shaped structures containing EHD2 colocalize with cortical F-actin. 3T3-L1 adipocytes were fixed, permeabilized, and stained with anti-EHD2 antibody as described above and rhodamine phalloidin to visualize endogenous EHD2 and F-actin, respectively. Images in D and E were acquired by focusing at the bottom of the cell with an Olympus IX-70 inverted microscope. Overlay images (merge) show in yellow colocalization of EHD2 (arrowheads) and F-actin (arrows). Enlargement of a selected area is depicted.
isolated from bacterial lysates prepared as described (27). The other GST fusion proteins were isolated from bacterial lysates prepared according to the standard procedures.

Identification of EHD2 in Membrane Fractions from 3T3-L1
Adipocytes-As a means of identifying proteins that may be involved in regulating the trafficking of GLUT4-containing membranes, a purified membrane fraction from primary rat adipocytes shown to be highly enriched in GLUT4 (25) was subjected to SDS-PAGE, and many of the silver-stained bands were analyzed by tryptic hydrolysis and mass spectrometry. Among a large number of proteins identified previously in GLUT4-containing membranes (25), a protein band of about 65 kDa was found to contain peptides identical to a human isoform of the EHD2 protein (Fig. 1A). DNA sequences encoding this isoform, EHD2, were reported previously in the context of a genomics study without further characterization (28), and the murine homolog has not been identified previously.
Mouse EHD2 belongs to a group of four closely related proteins expressed in human cells that present a P-loop near the N terminus, followed by a coiled-coil region and an EH domain at the C terminus (28). This family of proteins has been suggested to function in endocytosis and membrane trafficking through binding of their EH domains to partner proteins with NPF motifs (29 -31). Sequence analysis of the full-length mEHD2 cDNA, derived from an EST clone obtained from Genome Systems, revealed the expected deduced domain structure for this protein as depicted in Fig. 1B. The expression profile of mEHD2 reveals high abundance in lung, fat, and skeletal and heart muscle (Fig. 1C). The latter three tissues are insulinsensitive and uniquely express GLUT4 glucose transporters. Also, mEHD2 mRNA is greatly elevated upon differentiation of mouse 3T3-L1 fibroblasts to adipocytes in culture (Fig. 1C). These expression data suggest a functional significance of EHD2 expression in the context of GLUT4 trafficking. Consistent with this possibility, recent studies with EHD1 and its Caenorhabditis elegans homolog RME-1 suggest that these proteins are part of the molecular machinery responsible for recycling of receptors to the plasma membrane (32)(33)(34). Furthermore, connections between EHD1 and endocytosis of IGF-1 receptor in Chinese hamster ovary cells has been described (35). A role for an EHD4-related protein, Pincher, in mediating FIG. 2. Ultrastructural analysis of EHD2, GLUT4, and actin filament localization in 3T3-L1 adipocytes. A, ring-shaped structures containing EHD2 associate with plasma membrane sheets. Plasma membrane lawns from 3T3-L1 adipocytes were generated as described under "Experimental Procedures," fixed, permeabilized and stained with anti-EHD2 polyclonal antibody and FITC-conjugated secondary antibody to visualize endogenous EHD2 protein. The arrows show EHD2 associated with plasma membranes. Image was acquired with an Olympus IX-70 inverted microscope. B, EHD2 and GLUT4 colocalize in the plasma membrane. Plasma membrane lawn sheets prepared from insulin-treated cells were fixed, permeabilized, and stained with polyclonal anti-EHD2 and monoclonal anti-GLUT4 antibodies and colloidal gold conjugated secondary antibodies to visualize EHD2 (12-nm gold particles) and GLUT4 (6-nm gold particles). The samples were then fixed and processed for goldpalladium replica EM, using the technique described under "Experimental Procedures." The inset shows a low magnification area of a plasma membrane lawn. EHD2 and GLUT4-positive ring structures were observed in this region, and B shows high magnification of the squared area. Arrowheads and arrows indicate EHD2 (12-nm gold particles) and GLUT4 (6-nm gold particles), respectively. C, EHD2 and GLUT4 containing membranes associated with cortical actin. Plasma membrane lawns were prepared as described above. C, plasma membrane lawns were stained with a monoclonal anti-GLUT4 (6-nm gold particles) and a polyclonal anti-actin (12 nm gold particles) antibody, or D, stained with a monoclonal anti-GLUT4 (6-nm gold particles) and polyclonal anti-EHD2 (12-nm gold particles). E, visualization of EHD2 and GLUT4 by whole-mount EM of a peripheral region of an insulin-stimulated 3T3-L1 adipocyte. Differentiated 3T3-L1 adipocytes were grown on Formvar-coated gold grids. Top panel, the arrows outside of the cell (e) shows the edge of the insulin treated 3T3-L1 adipocyte. The bottom panel in F represents high magnification of the squared area in E and shows EHD2 (arrows) and GLUT4 (arrowhead) colocalization. endocytosis and trafficking of the nerve growth factor receptor TrkA in PC12 cells has also been reported recently (36). Thus, these recent reports combined with the findings in Fig. 1 suggest the possibility that EHD2 is a novel mouse protein that functions in some aspect of the trafficking of membranes containing GLUT4 and other recycling proteins.
EHD2 and GLUT4-containing Vesicles Localize Near the Plasma Membrane-The subcellular localization of EHD2 was determined using a polyclonal antibody against GST-EHD2 in biochemical and morphological approaches. As shown in Fig.  1D (left), subcellular fractionation of adipocytes followed by immunoblot analysis revealed that most of EHD2 protein is associated with plasma membrane and microsomal fractions. In contrast, very low amounts of this protein were detected in the cytosol. Consistent with the immunoblot analysis depicted in Fig. 1D (left), indirect immunofluorescence of 3T3-L1 adipocytes using the same anti-EHD2 antibody revealed that endogenous EHD2 localizes near the plasma membrane ( Fig. 1, D , right, and E, left). By focusing at the bottom of the cultured adipocytes, we detected EHD2 protein in tubulovesicular, ring shaped structures with different diameters that appear to be associated or connected to the plasma membrane (Fig. 1D,  right). In C. elegans intestine cells, a similar ring-shaped tubulo-endosomal system near the surface containing RME-1 was also described (32).
To determine the relationship between the vesicular system containing EHD2 and cytoskeletal elements in that region, 3T3-L1 adipocytes were double-stained with anti-EHD2 and phalloidin to detect EHD2 and F-actin, respectively. As shown in Fig. 1E, a close association between patches of cortical Factin and EHD2 was observed. Thus, F-actin structures appear to localize at the proximity of the EHD2-containing membranes beneath the cell surface. To better visualize this region, plasma membrane lawns were prepared from insulin-treated cultured adipocytes. When imaged by indirect immunofluorescence (Fig.  2A), membrane lawns revealed vesicular structures containing EHD2, similar to those detected by focusing at the bottom of the cells (Fig. 1, D and E). By immunogold staining, these vesicular structures were found to be positive for EHD2 and GLUT4 and appear to be linked to the plasma membrane through actin filaments (Fig. 2, C and D).
Further analysis of the peripheral region of insulin-treated 3T3-L1 adipocytes by whole-mount electron microscopy revealed a complex network of tubulovesicular membranous structures beneath the cell surface (Fig. 2E). Both anti-EHD2 (large particles) and anti-GLUT4 (small particles) showed binding to these structures in close proximity (Fig. 2F). This filamentous network of intracellular membranes at the cell periphery appears to be actin-based, because they disappear when these cells are treated with the actin-depolymerizing agent latrunculin B (20). These results suggest that EHD2 and GLUT4 colocalize in vesicles near the plasma membrane connected with cortical F-actin.
Disruption of EHD2 Function Inhibits Transferrin and GLUT4 Endocytosis-The data described above, combined with the known function of other EH domain-containing proteins in endocytosis (29,31), suggest a possible role of EHD2 in movement of GLUT4 away from the cell surface during its recycling. To test this hypothesis, we examined the effects of truncated constructs of EHD2, which would be predicted to act in a dominant negative manner (37). The uptake of rhodamineconjugated transferrin in COS-1 cells transfected with GFP fusion constructs of full-length EHD2 or EHD2 devoid of its EH domain (⌬EH-EHD2) was monitored. As shown in Fig. 3, 5 min after addition of rhodamine-transferrin most of the fluorescence signal observed in control cells expressing EGFP or EH domain alone is already in a tightly localized juxtanuclear position. In contrast, between 70 and 80% of cells expressing either native GFP-EHD2 or GFP-⌬EH-EHD2 display no detectable rhodamine-transferrin in the perinuclear area (Fig. 3, A  and B).
We next examined whether the steady-state distribution of transferrin receptors was affected by expression of EHD2 constructs. The overall steady-state distribution of transferrin receptor detected by immunofluorescence with anti-transferrin receptor monoclonal antibody was unaffected by expression of EHD2 proteins in COS-1 cells. Both untransfected and EHD2expressing cells depict similar transferrin receptor distribution (data not shown). Thus, the steady-state distribution of transferrin receptor was not altered dramatically by expression of EHD2 constructs. These data suggest that disruption of endogenous EHD2 function by high expression of EHD2 constructs impairs the internalization of transferrin or its movement from early sorting endosomes to the juxtanuclear recycling endosome.
In order to determine at which step transferrin receptor internalization is compromised by high expression of EHD2 proteins, the trafficking of Texas Red-labeled transferrin was analyzed in live COS-1 cells using high speed, three-dimensional digital imaging microscopy (22). When cells expressing the GFP-tagged EEA1 were incubated with the Texas Redlabeled transferrin for 7 min and then chased with unlabeled transferrin for the following 30 min, an intense Texas Redbased fluorescence was detected in the GFP-EEA1-labeled early endosomal membranes. In contrast, in cells expressing high levels of EHD2, the transferrin signal in early endosomes and at the perinuclear area was virtually undetectable (Fig. 4). These data indicate that EHD2 functions in the earliest steps of clathrin-mediated endocytosis prior to movement of cargo to early endosomes.
We next examined the consequences of the expression of EHD2 and ⌬EH-EHD2 proteins on GLUT4 internalization in 3T3-L1 adipocytes. In order to investigate whether EHD2 plays a role in GLUT4 movement from the cell surface to the perinuclear storage compartment, we employed both the HA-tagged EHD2 (HA-EHD2) and HA-⌬EH-EHD2, as dominant inhibitory constructs. These constructs were cotransfected into cultured adipocytes with a plasmid encoding GLUT4 fused at its C terminus to EGFP and tagged with a Myc epitope in its exofacial loop (21). In initial experiments, 3T3-L1 adipocytes expressing myc-GLUT4-GFP (control) or coexpressing myc-GLUT4-GFP and either HA-EHD2 or HA-⌬EH-EHD2 or HA-EH constructs were analyzed for effects on translocation of GLUT4 to the cell surface in response to insulin. No effect of the EHD2 constructs could be observed on the intense staining by anti-Myc of cell rims induced by the action of insulin (Fig. 5). These data indicate that the expression of EHD2 proteins do not interfere with insulin action on GLUT4 exocytosis in cultured adipocytes.
In order to monitor plasma membrane myc-GLUT4-GFP movement back to the perinuclear area, cells were washed to remove insulin, labeled with anti-Myc antibody at 4°C, warmed at 37°C to allow myc-GLUT4-GFP internalization, fixed at increasing time points, and the Myc signal visualized. At 40 min post-insulin removal, 85% of 3T3-L1 adipocytes expressing only myc-GLUT4-GFP showed a marked decrease of anti-Myc signal at the cell rims, with parallel increases in the perinuclear area (Fig. 5, A and B, control), consistent with previous results (38). In contrast, expression of the HA-tagged EHD2 constructs exerted a marked inhibitory effect on myc-GLUT4-GFP internalization, with more than 50% of the cells still presenting Myc signal at the rims or in vesicles near the cell surface (Fig. 5). These results are consistent with the hypothesis that EHD2 function is required for endocytic trafficking between plasma membrane and early endosomes.
Identification and Cloning of an EHD2-binding Protein, EHBP1-Our results showing EHD2 likely functions in endocytosis prompted us to seek partner proteins that interact with its EH domain. Extracts of 3T3-L1 adipocytes were incubated with agarose beads coated with GST alone or GST-EH domain, and the bound proteins were resolved by SDS-PAGE and silverstained. A protein with a molecular mass of 140 kDa (p140) was detected associated with the GST-EH domain fusion but not with GST alone. The molecular identity of p140 was determined by mass spectral analysis. Molecular weights of two peptide fragments derived from p140 ( 472 AYDGFASIGISR 483 and 788 VLLEQAR 794 ) coincided with those from the human gene KIAA0903. We named this novel protein EHBP1, for EH domain binding protein 1. We cloned the human gene by obtaining two overlapping ESTs with the coding sequences of EHBP1 from amino acids 1-310 and 235-1196. These two ESTs were fused to generate the complete EHBP1 cDNA sequence. EHBP1 contains 1196 amino acids, predicting a protein of 136 kDa. EHBP1 contains sequences with exact matches to the two tryptic peptides identified by mass spectrometry (Fig. 6B, underlined sequences).
EHBP1 contains a CAAX (where AA is aliphatic amino acid) box at its C terminus and in its N-terminal region five closely spaced NPF motifs. NPF motifs are known to bind EH domains (29,31) and have been identified in many proteins that play a role in the endocytic process (29,30). We therefore examined the potential interaction of these NPF motifs with the EH domain of EHD2. Extracts of adipocyte plasma membranes and low density microsomes were incubated with GST alone or with a GST fusion protein containing the NPF motifs of EHBP1 (GST-NPF, amino acids 227-407). As depicted in Fig. 6A, EHD2 was found to bind to GST-NPF but not to GST alone. This interaction is mediated by the EH domain of EHD2, because HA-EHD2 and HA-EH, but not HA-⌬EH-EHD2, bind NPF motifs from EHBP1 (Fig. 6A). These results are consistent with a direct interaction of EHBP1 and EHD2 in vitro, mediated by their respective NPF motifs and EH domain. As shown if Fig. 6A, subcellular fractionation of COS-1 cells followed by immunoblot using anti-EHBP1 antibody showed that EHBP1 protein is associated mostly with cytosol and plasma membrane fractions. Consistent with these data, endogenous EHD2 partially colocalizes with EHBP1 in ruffled membranes at the cell periphery (Fig. 6, C and D). Interestingly, EHD2 and EHBP1 appear to define discrete structures that are quite distinct at the bottom of the cell (left panel of Fig. 6D). Colocalization of EHD2 and EHBP1 occurs most markedly as the optical plane is raised (right panels of Fig. 6D).
To determine the relationship between EHD2, EHBP1, and F-actin in these ruffled membranes, COS-1 cells were costained with anti-EHD2, anti-EHBP1, and phalloidin. The cortical actin cytoskeleton in these ruffled membrane appears to be in close association with EHD2 and EHBP1, as shown in Fig. 6E.
In an attempt to evaluate the functional relevance of EHD2/ EHBP1 interactions, experiments were conducted to evaluate cortical actin organization following disruption of this interaction. Expression of high levels of GFP-EH domain of EHD2 or GFP-NPF motifs (amino acids 1-340) of EHBP1 caused a marked disruption of the cortical actin rearrangements (Fig. 7). Conversely, expression of GFP-EHD2 or GFP-⌬EH-EHD2 promotes cortical actin rearrangement in adipocytes. Thus, these results suggest that normal cortical F-actin assembly requires interaction between these two novel proteins in 3T3-L1 adipocytes. Surprisingly, we could not detect decreases in transferrin internalization in COS-1 cells expressing GFP-EH (Fig. 3). However, it should be noted that actin-perturbing drugs disrupt endocytosis in some cells but not in all cases (8 -11).
The interaction of EHBP1 with EHD2 prompted us to investigate whether EHBP1 also plays a regulatory role in endocytosis. Experiments were performed similar to those described in Fig. 3 to evaluate endocytosis and movement of rhodaminelabeled transferrin to the perinuclear area of COS cells expressing high levels of EHBP1. Uptake and concentration of the labeled transferrin in the juxtanuclear region of these cells was readily observed in control cells, whereas GFP-EHBP1 expression disrupted this phenotype (Fig. 8, lower panel).
Quantification of these data showed that 80% of control cells expressing GFP alone displayed strong transferrin signal near the nucleus, whereas only about 20% of cells expressing high levels of GFP-EHBP1 did so (Fig. 8, right panel). We also performed experiments to examine whether the expression of GFP-EHBP1 affects the distribution of transferrin receptor detected with anti-transferrin receptor antibody. No differences in untransfected versus GFP-EHBP1 transfected COS-1 cells could be detected (data not shown).
Interestingly, the NPF motifs in EHBP1 are located near a region with sequence similarity to calponin homology (CH) domains, which have been found to bind and cross-link F-actin when present in oligomeric proteins (39,40). This domain structure of EHBP1 suggests a function that connects the process of endocytosis through binding EHD2 as well as membranes via the CAAX box motif with the actin cytoskeleton through its CH domain. Consistent with this function, high expression of HA-EHBP1 in COS-1 cells induced a marked cortical actin phenotype (Fig. 8, top panel). Eighty percent of the cells transiently transfected with full-length HA-EHBP1 exhibited membrane ruffling and prominent actin-rich lamellipodia on the cell surface. These results are consistent with the conclusion that EHBP1 operates with EHD2 to regulate receptor-mediated endocytosis as well as actin cytoskeletal rearrangement in COS-1 cells.
Loss of Endogenous EHD2 or EHBP1 Mediated by siRNA Gene Silencing Inhibits Transferrin Endocytosis-Next we transfected EHD2-or EHBP1-targeted siRNA into cultured fibroblast cells to induce specific degradation of EHD2 or  FIG. 6. EHBP1, a novel EHD2 interacting protein that regulates actin rearrangement and endocytosis. A, upper panel, interaction of recombinant NPF motifs from EHBP1 protein with EHD2 from adipocyte plasma membrane (PM) and low density microsomes (LDM) fractions. GST alone or GST-NPF (25 g) in agarose beads were incubated with 0.5 mg of the indicated subcellular fractions solubilized in buffer containing EHBP1 mRNA, and transferrin uptake was then assayed in these cells. Forty eight hours after transfection, the EHD2 or EHBP1 protein levels were significantly decreased in 3T3-L1 fibroblasts transfected with siRNA specific to EHD2 or EHBP1 mRNA, respectively, but not scrambled siRNA (Fig. 9A). Loss of the EHD2 or EHBP1 proteins resulted in a marked inhibi-  tion of transferrin uptake in fibroblasts (Fig. 9B). Quantification of these data shows that 85% of cells transfected with scrambled siRNA displayed transferrin associated with endosomes, whereas only 40 -50% of the cells with reduced levels of endogenous EHD2 or EHBP1 did so (Fig. 9B). The effects of the loss of the endogenous EHD2 or EHBP1 on transferrin endocytosis were also examined by measuring fluorescent transferrin accumulation. Control cells or cells in which EHD2 or EHBP1 protein levels were reduced by siRNA-mediated gene silencing were incubated with Alexa594-labeled transferrin for 0 -45 min, solubilized in SDS sample buffer, and the extracts analyzed by SDS-PAGE. The resulting immunoblots were scanned using the appropriate laser line of a Storm 860 Phos-phorImager. As depicted in Fig. 9C, depletion of EHD2 or EHBP1 proteins significantly reduced the amount of transferrin that accumulated in fibroblasts. These results indicate that EHD2 and EHBP1 are required for optimal endocytic trafficking of transferrin.
Expression of EHD2 Induces Cortical Actin Rearrangement-In the course of the above studies we noted that expression of EHD2 or ⌬EH-EHD2 in adipocytes and COS-1 cells induces a marked effect on cortical actin, as does EHBP1 (Figs. 7,8,and 10). Up to 60% of COS cells transiently transfected with full-length GFP-EHD2 exhibit less stress fibers than controls as well as prominent actin-rich filopodia at the cell surface. Furthermore, massive cortical actin formation and filopodia induction was observed in 90% of cells expressing ⌬EH-EHD2. In contrast, untransfected cells or cells expressing EH domain alone display very few filopodia. These surprising results indicate that the EHD2 protein devoid of its EH domain is more effective than full-length EHD2 in promoting actin polymerization, suggesting a regulatory role for the EH domain in this function.
The marked cortical actin rearrangement induced by EHD2 expression prompted us to investigate the molecular mechanism by which EHD2 induces this phenotype. Analysis of the sequences of the EHD protein family revealed a conserved cluster of acidic amino acids and a tryptophan residue just outside of the N-terminal end of their EH domain (Fig. 10A). Similar sequences are also presented in WASP-like proteins, cortactin and the yeast EH domain protein Pan1 (6,41,42). These sequences are required for interaction of those proteins with Arp2/3 complex, a key regulator of actin-filament nucleation (41,43). To test the hypothesis that EHD2 acidic region is required to induce actin rearrangement, we expressed in COS-1 cells GFP-⌬EH-EHD2, which induced a massive cortical actin phenotype, or GFP-⌬EH-⌬A-EHD2, a construct in which both the EH domain and the conserved acidic putative Arp2/3binding site (amino acids 428 -444) have been deleted. As depicted in Fig. 10B, COS-1 cells expressing GFP-⌬EH-EHD2 presented marked actin rearrangement, with many filopodia and spikes over the cell surface. In contrast, neither untransfected cells nor cells expressing GFP-⌬EH-EHD2 without the acidic region (GFP-⌬EH-⌬A-EHD2) show significant cortical actin rearrangement and or filopodia and microspikes (Fig.  10B). These results suggest that the cluster of acidic amino acids 428 -444 in EHD2 is required for the actin rearrangement that we find mediated by this protein. Taken together, these data support the hypothesis that EHD2 modifies actin dynamics through its putative Arp2/3-binding site as well as through interaction with EHBP1, which contains the actinbinding CH domain. DISCUSSION The data presented here strongly implicate two novel proteins, EHD2 and EHBP1, in the complex process of endocytosis in mammalian cells. EHD2, a mouse protein we identify that contains a N-terminal P-loop and coiled-coil domain, is a member of a large class of proteins that contain one or more EH domains and function at early steps of membrane retrieval. Because EHD2 is most highly expressed in lung and fat cells among mouse tissues we studied, much of the present work focused on murine 3T3-L1 adipocytes as a convenient model system for assessing its function. EHD2 localization in these cultured adipocytes is consistent with a role in endocytosis, as it is concentrated in actin-based membranes near the cytoplasmic face of the plasma membrane ( Figs. 1 and 2). The endocytosis of GLUT4 glucose transporters and their movement to perinuclear membranes was inhibited by expression of a putative dominant negative construct of EHD2 devoid of its EH domain (Fig. 5). High expression of native EHD2 also disrupts this GLUT4 internalization process, which is likely due to titration of partner proteins involved in endocytosis or motility of internalized membranes. These data and the similar domain structure of EHD2 compared with proteins of the endocytic machinery directly link its function to some stage of the endocytosis process.
Further experiments presented here document EHD2 function at an early stage in the pathway of endocytosis. For these studies transferrin internalization in COS cells was employed as a means to assess whether the block in endocytosis due to EHD2 disruption occurs prior to formation of early endosomes. By using live cell imaging techniques, we found that high expression of EHD2 prevented transferrin movement into endosomes that harbor the early endosome marker EEA1 (Fig. 4). These data position EHD2 function at the stage of formation of coated pits, their invagination, or at a stage immediately after scission of coated vesicles. The results presented here are also consistent with a role for endogenous EHD2 and its interacting protein EHBP1 in endocytosis. Loss of EHD2 or EHBP1 significantly inhibits transferrin internalization (Fig. 9). Thus, attenuating EHD2 or EHBP1 function by dominant negative constructs (Figs. 3-5 and 8) or by siRNA gene silencing (Fig. 9) disrupts endocytosis. Interestingly, recent studies with a related human isoform of mouse EHD2, hEHD1, was found to apparently regulate recycling of transferrin receptors back to the plasma membrane (32, 33). Our results argue against a role of EHD2 in promoting early endosome recycling back to the plasma membrane because little or no transferrin enters early endosomes when endogenous EHD2 is disrupted (Fig. 4). Targeting of EHD2 to sites on membranes is likely through the N-terminal P-loop or coiled-coil regions because both native EHD2 and mutant EHD2 devoid of the EH domain localize to sub-plasma membrane structures, whereas the EH domain alone does not (Figs. 3, 4, 7, and 10). Thus we hypothesize that EHD2 connects to specific proteins in the plasma membrane through one or both these N-terminal regions and to downstream elements through its C-terminal region (Fig. 11).
A key finding presented here is that at least two distinct motifs are present in the C terminus of EHD2 that might link this protein to the actin cytoskeleton, an acidic and putative Arp2/3 complex binding motif and the EH domain itself which binds the novel CH domain-containing protein EHBP1 (Figs. 6 and 10). The EHBP1 protein contains a CH domain most similar to that of the type 1 CH domain of ␣-actinin, which binds F-actin (39). Data obtained in the last few years indicate that the actin cytoskeleton is involved in membrane retrieval and movement of endosomes to intracellular destinations in mammalian cells (for reviews, see Ref. 12, 13, and 44). Variable results have been obtained when actin function in the initial budding, invagination, and scission steps of the clathrin-dependent endocytosis pathway was tested, but in some cell types one or more of these steps do require F-actin for optimal function (8,10,11). Downstream movements of internalized vesicles have been shown to require actin filaments, including endocytic trafficking in polarized epithelial cells (8,45). Actin accumulation has been found around coated vesicles when endocytosis is disrupted at the synapse (46), and motility of endocytic vesicles through actin tails has been observed (47). Thus, the linkages of EHD2 to the actin cytoskeleton through interaction with EHBP1 provides strong evidence that EHD2 functions at some point in endocytosis at which F-actin facilitates the process.
The results obtained by three-dimensional microscopy revealed localization of these proteins at the sub-plasma membrane region of the cells (Fig. 6, D and E). Results of this deconvolution-based microscopic analysis and Z series of images from the cell periphery indicate more precisely that endogenous EHD2 and EHBP1 proteins associate in ruffled membranes, a region of intense actin dynamics and endocytosis (see QuickTime movies 1 and 2 in Supplemental Material). These data reveal that EHD2 and EHBP1 localize to adjacent structures and that colocalization is observed only in places where these structures intersect. Although the resolution of such experiments is in the 200-nm range, the data are consistent with the hypothesis that only subpopulations of EHD2 and EHBP1 directly interact (at the interface of the structures in which they reside), but that the majority of these protein molecules are not directly interacting in these adjacent structures at any given time. Thus, the results depicted in Fig. 6, D and E, also provide a rationale for why it is difficult to coimmunoprecipitate these proteins (data not shown).
It is noteworthy that the EH domain of EHD2 appears to play a key role in EHD2 action on actin polymerization through the short acidic sequence (Fig. 10). EHD2 stimulation of actin polymerization is significantly increased upon deletion of its EH domain, whereas the additional deletion of the short acidic domain reverses this increase (Fig. 10). Thus, EHD2 may exist in at least two conformations regarding its ability to promote actin polymerization. In the less active conformation, the EH domain may occlude the short acidic and putative Arp2/3binding site. Upon binding of ligands containing NPF motifs such as EHBP1, the EH domain may be modulated to expose the Arp2/3-binding site, triggering the actin polymerization process. This speculative model is similar to that proposed for N-WASP activation, in which its masked VCA domain is exposed and able to interact with Arp2/3 upon binding of Cdc42 and phosphatidylinositol 4,5-bisphosphate to N-WASP (48,49).
What might be the role of F-actin in the mechanism of EHD2 and EHBP1 function? The Arp2/3 complex acts to nucleate actin filaments, leading to polymerization of F-actin (41,43), whereas the function of type I CH domains is to bind F-actin, causing cross-linking of actin filaments when CH domains are present in oligomeric proteins (39). In eliciting both of these actin-regulatory pathways, the EHD2 and EHBP1 proteins might be expected to promote juxtaposition of membranes with actin filaments, as schematized in Fig. 11. Recent considerations of steps in the endocytic pathway where actin may play important roles include spatial organization of the endocytic machinery (12,44). Indeed, recent results show that depolymerization of actin is associated with increased lateral move- FIG. 11. Hypothesis: EHD2 and the novel EHD2-interacting protein, EHBP1, couple clathrin-dependent endocytosis to F-actin. Shown is a model depicting EHD2 and EHBP1 binding to membranes of the endocytic system through the N-terminal regions of EHD2 and the CAAX box motif of EHBP1, respectively. EHD2 is hypothesized to direct coupling of endocytic membranes to F-actin by promoting actin rearrangement via a short acidic and putative Arp2/ 3-binding motif in the EHD2 C-terminal region adjacent to its EH domain. Coupling to the actin cytoskeleton by EHD2 also occurs indirectly through its binding of EHBP1, which contains a type 1 CH domain similar to other CH domains that bind F-actin. See text for further details. ments of clathrin-coated pits (50). Deformation or invagination of the plasma membrane, the fission step, and vesicle movement away from the plasma membrane may also theoretically involve the actin cytoskeleton (44,51). Our data at this stage cannot discriminate among these possibilities, nor do we have direct evidence that supports the hypothesis that EHD2 and EHBP1 bind to different membranes (Fig. 11) versus the same membrane (not shown). However, this study adds two new components to the puzzle of how the endocytic machinery operates in association with the actin cytoskeleton. Future experiments will be needed to define the mechanism or mechanisms by which the functions of F-actin are coordinated with those of the EHD2 and EHBP1 proteins.