Structural and Functional Analysis of Integrin α2I Domain Interaction with Echovirus 1

Abstract Integrins are cell surface receptors for several microbial pathogens including echovirus 1 (EV1), a picornavirus. Cryo-electron microscopy revealed that the functional domain (α2I) of human α2β1 integrin binds to a surface depression on the EV1 capsid. This three-dimensional structure of EV1 bound to α2I domain provides the first structural details of an integrin interacting with a picornavirus. The model indicates that α2β1 integrin cannot simultaneously bind both EV1 and the physiological ligand collagen. Compared with collagen binding to the α2I domain, the virus binds with a 10-fold higher affinity but in vitro uncoating of EV1 was not observed as a result of attachment of α2I. A molecular model, constructed on the basis of the EV1-integrin complex, shows that multiple α2β1 heterodimers can bind at adjacent sites around the virus 5-fold symmetry axes without steric hindrance. In agreement with this, virus attachment to α2β1 integrin on the cell surface was found to result in integrin clustering, which can give rise to signaling and facilitate the initiation of the viral entry process that takes place via caveolae-mediated endocytosis.

Numerous clinically important viruses, including adenoviruses, hantaviruses, and picornaviruses, use the members of the integrin family as their cellular receptors. The natural ligands of the integrins include many extracellular matrix proteins, and the physiological function of these receptors is to mediate cell adhesion. Integrins are heterodimers, consisting of an ␣-subunit and a ␤-subunit. Following ligand binding, they generate cellular signals leading to the formation of cytoskeletal connections and specific cellular responses (1). It is likely that the general properties of integrins have facilitated their utilization as receptors for a number of viruses. However, the mechanisms by which viruses have evolved to utilize cell surface integrins for binding and internalization remain largely unknown because insufficient structural information has been available on the details of virus-integrin interactions.
Echovirus 1 (EV1), 1 a member of the Picornaviridae family, uses the ␣ 2 ␤ 1 integrin (VLA-2) as its receptor for cell entry (2) through caveolae-mediated endocytosis (3). In humans, echovirus infections are associated with meningitis, encephalitis, rash, respiratory infections, diarrhea, and even fatal illness in newborns (4). The picornavirus particle is composed of a singlestranded, infectious RNA molecule, which is packed into an icosahedral capsid consisting of 60 protomers, each containing four viral proteins (VP1-4) (5). A surface depression termed the canyon is located around each of the twelve 5-fold axes of the capsid in many picornaviruses. The canyon has been shown to be a binding site for receptor molecules that belong to the immunoglobulin superfamily (6).
The binding site of EV1 on ␣ 2 ␤ 1 integrin has been located to the inserted domain of the ␣-subunit (␣ 2 I) (7), which also forms the binding site of the physiological ligand collagen. The residues identified as essential for EV1 binding are mostly positioned on a face of ␣ 2 I different from the collagen binding site (8,7). The binding mechanisms of these two ligands to ␣ 2 ␤ 1 integrin are evidently different: in contrast to collagen binding to the integrin, the attachment of EV1 has been reported to be independent of divalent cations (9).
Here, we have determined the structure of the complex of EV1 and ␣ 2 I domain by cryo-electron microscopy (cryo-EM) and compared the interactions in the receptor-virus complex with those made to collagen (10). The interaction with ␣ 2 I domain does not appear to directly induce uncoating of the virus, as is the result of poliovirus receptor interactions. Moreover, following the binding of EV1, we have observed clustering of ␣ 2 ␤ 1 on the plasma membrane, which is essential for the activation of integrin-related signaling pathways (11). These mechanisms may have key implications for initiation of infection by viruses that interact with cell surface integrins.

EXPERIMENTAL PROCEDURES
Preparation of Viruses and Receptors-Human ␣ 2 integrin I domain was produced as a glutathione S-transferase-␣ 2 I (GST-␣ 2 I) fusion protein as described earlier by Ivaska et al. (12). A soluble construct of two poliovirus receptor (PVR) molecules linked together by an Fc antibody fragment (PVR-Fc) was prepared as described by Xing et al. (13). EV1 (Farouk strain) and poliovirus 1 (PV1, Sabin strain) were obtained from the American Type Culture Collection (ATCC) and propagated in African green monkey kidney (GMK) (ATCC) cells. To obtain radioactively labeled preparations of EV1 and PV1, infected GMK cells were incubated in the presence of [ 35 S]methionine (50 Ci/ml; Amersham Biosciences) in Eagle's minimal essential medium (MEM) deficient in Lmethionine (Invitrogen Life Technologies, Inc.). The viruses were purified by sucrose gradient sedimentation as described previously (14) and suspended in phosphate-buffered saline containing 2 mM MgCl 2 . Purified virus preparations were examined by neutralization using serotype-specific antisera, and the infectivity of the viruses was determined by a plaque titration assay. Protein concentration of the EV1 preparation was measured using BCA Protein Assay Kit (Pierce Biotechnology), and the purity of the capsid proteins was analyzed on a 12% SDS-PAGE gel.
Biochemical Characterization of EV1-␣ 2 I Domain Interaction-The blocking effect of the expressed receptors on virus infectivity was tested using a plaque assay. First, EV1 (200 plaque forming units; PFU) were incubated alone, with different concentrations (0 -1000 nM) of recombinant fusion protein GST-␣ 2 I, or with GST for 1 h at 37°C. Corresponding incubations were done with PV1 and PVR-Fc. The mixtures were transferred onto confluent GMK cells for 30 min at 37°C, and, following the removal of unbound virus the cells were overlaid with 0.5% carboxymethylcellulose. After 2 days, the cells were stained with crystal violet before the counting of the plaques. The efficiency of infection was evaluated as the ratio of the number of plaques obtained in the presence of receptor molecules to the number obtained by infecting with the virus alone.
Sucrose gradient sedimentation analysis was used to study whether the binding of GST-␣ 2 I to EV1 can trigger conformational changes in the viral capsid. 20,000 cpm of 35 S-labeled EV1 was incubated with 500 nM GST-␣2I for 1 h at 37°C and centrifuged in a linear 5-20% sucrose gradient at 150,000 ϫ g in a SW41Ti rotor (Beckman) for 2 h at 4°C. Similarly, 35 S-labeled PV1 was incubated with 500 nM PVR-Fc. Empty viral capsids (80 S particles), used as a sedimentation control, were formed by incubation of EV1 at 56°C for 5 min. 500-l fractions were collected after centrifugation, and their radioactivity was measured in a scintillation counter.
A solid phase binding assay was used to examine the binding affinity of GST-␣ 2 I to EV1 and to type I collagen. Microtiter plate wells were precoated with type I collagen (Sigma-Aldrich Inc.) or with EV1. Various concentrations of GST-␣ 2 I were allowed to bind at 37°C for 1 h. The amount of bound GST-␣ 2 I was measured using an europium-labeled anti-GST antibody and time-resolved fluorescence (15). An equation analogous to the Michaelis-Menten equation was used to estimate the apparent K d .
Cryo-electron Microscopy and Image Reconstruction-The virus-receptor complex was prepared by mixing EV1 and GST-␣ 2 I fusion protein at 4°C for 3 h at a molar ratio of 1:300 (five molecules of GST-␣ 2 I per receptor binding site) in a buffer containing 10 mM HEPES, pH 7.5, 50 mM NaCl, 20 mM MgCl 2 , and 2 mM ␤-mercaptoethanol. The mixture was applied to a holey carbon grid and quickly plunged into liquid ethane surrounded by liquid nitrogen. Cryo-electron micrographs were recorded at a magnification of ϫ45,000 with a Philips CM120 electron microscope equipped with a Gatan 626 cryo-transferring system. The selected micrographs were digitized on a Zeiss Phodis scanner with a pixel size of 3.11 Å in the specimen space.
The image data were first processed from images of the highly defocused micrograph, with a low pass filter of 32 Å, so that the GST-␣ 2 I density could easily be identified. The origins and the orientations of the ␣ 2 I domain-bound EV1 particles were first estimated with the self-common lines method. The coefficient of correlation was assessed to measure the occupancy of ␣ 2 I domain binding and used as a criterion to exclude the image data with unsaturated receptor binding. A stable preliminary three-dimensional structure with strong receptor density was obtained after a few cycles of center and orientation refinement. This reconstruction was further refined with the data from the close focus images. Iterative single particle reconstructions were used to aid particle selection and to help fine-tune the intermediate model maps. To avoid bias, the real-space correlation of bound ␣ 2 I domain was no longer used. The resolution of the final reconstruction was validated by the R-factor. At the final stages of refinement, 75 selected individual particles were combined to calculate a 25 Å resolution map. The handedness of the solution was determined by matching the capsid features between the complex reconstruction, and the native density map calculated from the native virus crystal structure as previously described (13).
Molecular Modeling-The crystal structure of unliganded ␣ 2 I domain was fit manually into a difference density map, computed by subtracting the crystal structure-based density of native EV1 (Protein Data Bank (PDB) entry 1EV1) (5) from the EV1-␣ 2 I complex map, using the interactive mode of the software BODIL (www.abo.fi/fak/mnf/bkf/ research/johnson/bodil.html). An exhaustive search of the local translational and rotational space was then performed in BODIL. The sampling parameters were as follows: 10°rotational step around 321 axes spaced 7.9°apart, 3 Å translational step and 7 translational steps along each of the three Cartesian axes, giving an even sampling of the solu-tion space. These parameters resulted in 3.96 million transformations. The merit of fit was based on the sum of map density values at individual C ␣ atom positions for each transformation of ␣ 2 I domain in the search. Positions with C ␣ atoms overlapping the high-resolution crystal structure of EV1 were discarded, and the remaining solutions were sorted according to their fit to the cryo-EM density. The correspondence of the N terminus of ␣ 2 I domain with the linker density to the GST protein was used as a register to filter the search results. The ten highest scoring solutions were essentially identical to the manual fit (all pairwise root mean-squared deviations, computed for the C ␣ atoms, were less than 2.3 Å), which was chosen as the model described here.
A comparative model of the ␣ 2 ␤ 1 integrin heterodimer was built using the crystal structure of the ␣ V ␤ 3 ectodomain (PDB entry 1JV2) (16) as the template, although it does not contain an ␣ I domain. Sequence alignments of ␣ 2 /␣ V and ␤ 1 /␤ 3 integrin chains were based on automatic multiple sequence alignments (17,18) of mammalian integrins and further refined manually before building the three-dimensional coordinates using Modeller 6.1 (19). The crystal structure of ␣ 2 I domain was included in the model, and its relative orientation to the ␤-propeller domain in the ␣-subunit was modeled manually based on steric considerations. The model comprises the wide head region of the integrin (1): three N-terminal domains of the ␣-subunit (␤-propeller domain, I domain, and Ig-like domain) and two N-terminal domains of the ␤-subunit (␤ I domain and Ig-like domain).
Confocal Immunofluorescence Microscopy-In vivo crosslinking of ␣ 2 ␤ 1 was induced in SAOS-2 human osteosarcoma cells (ATCC) that normally lack endogenous ␣ 2 integrin but were stably transfected with an ␣ 2 integrin gene (␣ 2 ␤ 1 -SAOS) as described earlier (12). The cells were first incubated with an antibody against ␣ 2 integrin (MCA2025 from Serotec Inc.) and then with a cross-linking Alexa488-conjugated anti-mouse IgG antibody (Molecular Probes Inc.), each for 1 h on ice. Cells were then transferred to 37°C for 15 min to allow for integrin clustering. Similarly, the ␣ 2 ␤ 1 -SAOS cells were incubated with EV1 for 15 min at 37°C after virus binding onto cells for 1 h at 4°C. After the cells were fixed with 4% paraformaldehyde, the immunofluorescent staining of ␣ 2 ␤ 1 integrin and EV1 and confocal microscopy using a Zeiss LSM510 instrument were carried out as described previously (3) RESULTS

␣ 2 I Domain Inhibits EV1 Infection but Does Not Induce Virus
Uncoating in Vitro-Incubation of recombinant GST-␣ 2 I fusion protein with EV1 prior to the plaque assay was found to block EV1 infection of GMK cells in a dose-dependent manner, whereas GST alone had no detectable neutralization effect (Fig. 1A). At a physiological temperature, a concentration of 100 nM GST-␣ 2 I reduced EV1 infection by 40% compared with incubation of EV1 on GMK cells without the fusion protein. A GST-␣ 2 I concentration of 300 nM inhibited infection by 80%. For comparison, PV1 infection was inhibited by over 80% in the presence of 10 nM PVR-Fc (Fig. 1A), showing that the efficiency of inhibition of infection by GST-␣ 2 I is lower than that of the divalent PVR-Fc at 37°C. Incubation of PV1 with GST or GST-␣ 2 I caused no inhibition of infection (data not shown).
A sucrose gradient sedimentation assay was used to identify possible conformational alterations of the viral capsid structure after interaction with the receptor. At 4°C, the EV1 capsid remained intact and sedimented at 160 S, whereas at 37°C, a significant portion of EV1 spontaneously formed empty particles, which sedimented at 80 S (Fig. 1B). The presence of GST-␣ 2 I fusion protein altered the sedimentation of EV1 slightly (Fig. 1B), but the virus did not form 80 S particles (which lack both VP4 and viral RNA); these sedimentation coefficients correspond to the subviral particles obtained from PVR-bound poliovirus (13,20,21). Incubation of PV1 alone either at 4°C or at 37°C for 15 min yielded only intact 160S particles (data not shown), whereas the attachment of PVR-Fc to PV1 dramatically altered the conformation of the poliovirus capsid and induced the release of viral RNA, resulting in 80 S particles (Fig. 1B). Thus, the in vitro binding of a soluble receptor fragment to EV1 does not induce significant conformational changes in the viral capsid but rather inhibits formation of 80 S empty particles. ␣ 2 I Domain Binds within the Virus Canyon-We have determined the structure of the ␣ 2 I domain bound to EV1 by cryo-EM. The complex was prepared by mixing virus with the GST-␣ 2 I fusion protein for 3 h at 4°C. The fusion protein tends to form intermolecular disulfide bonds, which resulted in aggregates that were observed in cryo-electron micrographs (data not shown). Thus, ␤-mercaptoethanol was added to the buffer to a final concentration of 2 nM during incubation in order to maintain the fusion protein in a monomeric form. Consequently, virus-receptor complexes appeared in micrographs as individual spherical particles and the three-dimensional structure of the complex was reconstructed based on icosahedral symmetry.
The reconstruction of the virus-receptor complex revealed an EV1 particle decorated with 60 copies of ␣ 2 I domain (Fig. 2B). The bound receptor is located roughly halfway between the viral 2-fold and 5-fold axes. The receptor density extends to a radius of 180 Å, possesses a density value similar to that of the viral proteins, and has a volume that is sufficient to accommodate only the ␣ 2 I domain of the fusion protein. A peptide linker of 20 residues connects the GST domain to the N terminus of the ␣ 2 I domain in the fusion protein. The lack of discernible GST density is likely due to the flexibility of this hinge region. When the density map was examined at low isodensity con-tours (0.3-0.5 ), the linker could be traced to a tubular density extending from the ␣ 2 I domain toward the viral 3-fold axis (not shown). A difference map was generated by subtracting the crystal structure based density for native EV1 ( Fig. 2A) from the cryo-EM density map of the virus-receptor complex (Fig.  2B). The unliganded ␣ 2 I structure (PDB entry 1AOX) (22) was docked into this density-difference map. The crystal structure of the ␣ 2 I domain in complex with a synthetic collagen-like triple helical peptide (PDB entry 1DZI) (10) has also been solved, showing a large conformational change at the collagen binding site. Since collagen was not present in our experimental conditions, the unliganded ␣ 2 I domain crystal structure was presumed more representative and was chosen for fitting to the difference density.
In the resulting model, ␣ 2 I domain binds within the canyon on the EV1 surface with extensive contacts to the outer canyon wall (Fig. 2C), but the domain is not in intimate contact with the inner canyon wall. In the docked position, both the N and C termini are exposed to the solvent and point away from the virus surface. This agrees with the situation in ␣ 2 ␤ 1 integrin where both termini of the ␣ 2 I domain are connected with the seven-bladed ␤-propeller structure in the ␣-subunit (16). The ␣3 helix of the ␣ 2 I domain together with the connecting loops interact with the capsid protein VP2 from one protomer (on the The treatment of EV1 at 37°C for 1 h or at 56°C for 3 min caused formation of 80 S empty capsids. While a shift in sedimentation was observed when the virus was incubated with GST-␣ 2 I at 37°C for 1 h, no 80 S particles were formed. In contrast, PV1 80 S particles were generated in the presence of PVR-Fc. Incubation of PV1 alone either at 4°C or at 37°C for 15 min yielded only intact 160 S particles (data not shown). left in Fig. 2C), while the end opposite to the N and C termini contacts VP3 from a neighboring protomer (on the right in Fig.  2C). The metal ion-dependent adhesion site (MIDAS) points toward the canyon floor but it is not in close contact with the virus.
The Virus-Receptor Interface-There are in total three charged residues from the ␣ 2 I domain (Lys 201 , Asp 219 , and Arg 288 ) lining the binding interface that could form complementary electrostatic interactions with EV1 (respectively with Glu 2162 -Asp 2163 , Lys 3230 , and Glu 1273 ; viral residues are numbered sequentially starting from 1001, 2001, 3001, 4001 for VP1, VP2, VP3, and VP4, respectively). Studies of EV1 binding to chimeric and mutated ␣I domains have suggested that residues 199 -201, 212-216, and 289 are required for EV1 binding (23,24). Our fitting places residues 199 -201 (Thr-Tyr-Lys) of the ␣ 2 I domain in contact with the E-F loop located between the E and F strands of VP2, where Lys 201 can form a favorable electrostatic contact with two negatively charged residues of the virus: Glu 2162 and Asp 2163 (Fig. 2D). The side chain of Tyr 200 of ␣ 2 I is positioned to stack with the side chain of His 2164 from EV1. ␣ 2 I domain interacts with two adjacent viral protomers, placing residues 212-216 (QTSQY) above the interface with the phenolic ring of Tyr 216 pointing toward the viral surface (Fig. 2C). Arg 288 and Asn 289 in the ␣C-␣6 loop of ␣ 2 I may interact with the C terminus of VP1 from the second protomer, where the positively charged side chain of Arg 288 points toward the negatively charged Glu 1273 side chain of the virus (Fig. 2D). The point mutation N289G is known to relax the binding specificity of ␣ 2 I toward different collagen types (24), even though this residue is not in close contact with the collagen triple helix (10). This mutation may result in the alteration of the dynamics of the ␤E-␣6 loop of the ␣ 2 I domain, a region involved in a large conformational shift upon collagen binding, and it is likely that mutation of N 289 influences virus binding through a conformational effect.
Collagen and EV1 Cannot Bind to the ␣ 2 I Domain Simultaneously-In our model of the EV1-␣ 2 I complex, the MIDAS motif of ␣ 2 I domain, where collagen binds (10), points toward the canyon without direct contact with the viral surface. Superposition of the crystal structure of the ␣ 2 I domain in complex with the collagen-like peptide ( Fig. 2A) (10) on the EV1-␣ 2 I complex (Fig. 3B) shows that the collagen triple helix would occupy space overlapping that occupied by EV1 when bound to the ␣ 2 I domain. Thus, the model strongly suggests that the ␣ 2 I domain cannot bind both collagen and EV1 simultaneously and that, therefore, EV1 must compete with native ligands for free ␣ 2 ␤ 1 molecules. From the solid phase binding assay, the apparent K d for the ␣ 2 I domain bound to EV1 is 2 nM (Fig. 3C), an affinity 10 times greater than that of the ␣ 2 I domain bound to collagen type I (ϳ20 nM). Again, the model structure for the EV1-␣ 2 I complex and the crystal structure of ␣ 2 I in complex with the collagen-like triple-helical peptide are consistent with these experimental results: the solvent-accessible surface area of ␣ 2 I buried when bound to EV1 (850 Å 2 ) is substantially larger than that buried by the collagen-like peptide (359 Å 2 ).
EV1 Binding Activates Integrin Clustering-EV1 attachment the ␣ 2 ␤ 1 integrin is known to initiate the subsequent internalization of both the virus and the integrin in caveolae (3), suggesting that concomitant integrin signaling may take place. The key question concerning the activation of signaling is whether the ligand can cause the clustering of integrins on the cell surface. In previous studies, integrin clustering has been induced with antibodies, which triggers the activation of signaling pathways and leads to the accumulation of cytoskeletal proteins at the sites of the clusters (11). We investigated the distribution of ␣ 2 ␤ 1 integrin on the cell surface by confocal immunofluorescence microscopy after incubating cells with antibodies and with EV1. Formation of ␣ 2 ␤ 1 clusters was observed after treating ␣ 2 ␤ 1 -SAOS cells with an anti-␣ 2 antibody and a secondary cross-linking antibody (Fig. 4, A and B). Incubation of the cells with EV1 for 15 min caused a redistribution of ␣ 2 ␤ 1 similar to that induced by the antibody treatment (Fig.  4, C and D). Controls, where cells were incubated with either the anti-␣ 2 integrin antibody or an irrelevant anti-␣ V integrin antibody without cross-linking, did not result in changes in the distribution of ␣ 2 ␤ 1 (data not shown).
A comparative model for the ␣ 2 ␤ 1 integrin heterodimer (Fig.  5A) was constructed based on the crystal structures of ␣ V ␤ 3 and ␣ 2 I domain (16,22) and then docked onto the EV1 structure by superimposing ␣ 2 I of the ␣ 2 ␤ 1 model on the ␣ 2 I domain of the virus-receptor complex structure. The five binding sites on the viral pentamer are able to accommodate the entire integrin molecule simultaneously without steric hindrance, despite the presence of the bulky 90 Å by 60 Å head of the ␣ 2 ␤ 1 heterodimer (Fig. 5B). Our three-dimensional model of the EV1-␣ 2 I complex is compatible with the observed virus-induced integrin clustering: while the molecular model demonstrates that there is no steric obstruction to ␣ 2 ␤ 1 binding at adjacent protomers, binding of ␣ 2 ␤ 1 to sites separated by one or more protomers on the EV1 surface may also place the integrins in sufficient proximity to result in signal-inducing clusters.

DISCUSSION
The three-dimensional structure of EV1 bound to ␣ 2 I domain provides the first structural details of an integrin interacting with a picornavirus. The reconstruction shows that EV1 uses the canyon region for receptor binding, similarly to poliovirus and the major group of human rhinoviruses (HRVs), whose receptors, PVR and intercellular adhesion molecule 1 (ICAM-1), belong to the immunoglobulin superfamily (IgSF) (13,25,26). In contrast, the low density lipoprotein receptor (LDLR), a receptor for the minor group of HRVs, and decay-accelerating factor (DAF), a receptor for echovirus 7, bind outside the canyon (27,28). Despite binding into the canyon, the interactions of ␣ 2 I domain with the canyon surface of EV1 differ from those of the IgSF receptors. The ␣ 2 I domain structure is larger and more globular than the elongated virus-binding domains of ICAM-1 and PVR. When we superposed the ␣ 2 I-EV1 complex onto the complex between ICAM-1 and rhinovirus 16 (29), the ␣ 2 I domain footprint on the EV1 surface at the bottom of the canyon overlaps that of ICAM-1 on rhinovirus 16; however, the footprint of the ␣ 2 I domain is larger and the ␣ 2 I domain makes more extensive contacts with the outer canyon wall.
The effects of receptor binding on the conformation of EV1 are different from poliovirus and the major group of HRVs. It has been shown that binding of the soluble fragment of PVR to poliovirus and ICAM-1 to HRV is able to mediate conformational changes in the capsids of both viruses in vitro, resulting in the release of their genomes (21,30). Our density gradient assay data suggest that in the case of EV1, attachment of the GST-␣ 2 I fusion protein is not sufficient to directly induce uncoating. Whether the action of ␣ 2 ␤ 1 integrin on EV1 in vivo is identical to that of the GST-␣ 2 I fusion protein and whether other cellular molecules are involved in EV1 uncoating requires further investigation. EV1 binding to ␣ 2 ␤ 1 represents the interaction of an ␣ Idomain containing integrin receptor with a picornavirus. Several other picornaviruses are known to use integrins as cellular receptors, but they recognize heterodimers that contain the ␣ V subunit (no I domain) and interact with the arginine-glycineaspartic acid (RGD) motif in the viral capsid protein VP1. These picornaviruses include coxsackievirus A9 and echovirus 9, which belong to enteroviruses, human parechoviruses and foot-and-mouth disease viruses (31)(32)(33)(34)(35)(36)(37). Interactions of adenoviruses with integrins have been studied thoroughly. Adenoviruses attach to the cells first through a binding receptor (coxsackievirus-adenovirus receptor or CD46) (38, 39) and subsequently recognize ␣ V integrins with the RGD motif located in the viral penton protein (40). Binding of multiple FIG. 3. Integrin ␣ 2 I cannot bind collagen and EV1 simultaneously. A, the crystal structure of the complex between ␣ 2 I domain and a collagen-like triple helical peptide (␣ 2 I domain drawn as a blue ribbon, collagen peptide as a red space-filling model). B, superposition of the structure of the complex of ␣ 2 I domain and a collagen-like peptide on the docked position of ␣ 2 I domain in the EV1-receptor complex. There is extensive overlap between the location of collagen and the virus (gold and light yellow surfaces), showing that these two integrin ligands cannot bind simultaneously. C, a solid phase binding assay for GST-␣ 2 I binding to type I collagen or EV1 in the presence of 2 mM Mg 2 Cl shows that EV1 binds to GST-␣ 2 I with higher affinity compared with type I collagen. integrins to the virus particle causes clustering, which is known to induce integrin signaling that facilitates the internalization of the viral particle. Another example of virus-induced integrin signaling is the binding of Kaposi's sarcomaassociated herpes virus (KSHV/HHV-8) to ␣ 3 ␤ 1 integrin, which results in the activation of focal adhesion kinase (FAK) (41).
Crystal structures of two RGD-containing viruses have been determined, foot-and-mouth disease virus (FMDV) (PDB entry 1FOD) (42) and coxsackievirus A9 (CAV9) (PDB entry 1D4M) (14), as well as the structure of their receptor ␣ V ␤ 3 integrin in complex with an RGD peptide (PDB entry 1L5G) (43). RGDcontaining peptides bind at the interface of the ␤ I domain of the ␤-chain and the ␤-propeller domain of the ␣-chain, coordinating tightly to a metal ion of MIDAS in the ␤ I domain. In the FMDV capsid, the RGD sequence is located within the G-H loop of VP1, whereas in CAV9 it is near the C terminus of VP1 in a region that is not defined in the crystal structure of the virus. The details of receptor interactions with RGD-containing vi-ruses must differ from those of EV1 with ␣ 2 ␤ 1 . Nonetheless, examination of the crystal structures of FMDV and RGD in complex with ␣ V ␤ 3 by superposition of the RGD motifs, as well as modeling of the RGD-containing segment of CAV9 (not shown), suggests the possibility of integrin clustering since these two viruses can also accommodate multiple integrin heterodimers simultaneously at adjacent binding sites about the 5-fold axes.
EV1 accommodates ␣ 2 ␤ 1 integrin in an orientation where the MIDAS motif faces the canyon floor, making it impossible for the receptor to bind collagen and virus simultaneously. The MIDAS surface is not in close contact with the virus structure, which is consistent with the finding that ␣ 2 I domain can attach to EV1 in the absence of Mg 2ϩ (9), in contrast to metal ion dependent interactions with collagen. The structure is also in agreement with the observation that a cyclic RKKH-containing octapeptide (12,44), which binds to the MIDAS region (44) and blocks collagen binding to ␣ 2 I domain (12), can bind the ␣ 2 I domain simultaneously with EV1 and even increases the binding of ␣ 2 I domain to the virus. There is sufficient space in our model at the MIDAS surface to accommodate the RKKH peptide. The enhancement of the binding probably reflects the shielding of the negatively charged residues around MIDAS by the peptide as well as the increase in the contact area between ␣ 2 I domain and EV1 mediated by the peptide.
Since the model of the EV1-␣ 2 I complex precludes the simultaneous binding of collagen and EV1 to ␣ 2 I domain, it implies that the virus must bind to cells that have ligand-free ␣ 2 ␤ 1 integrin heterodimers on the cell surface or compete with native ␣ 2 ␤ 1 ligands. Thus, the relatively low affinity of ␣ 2 ␤ 1 integrin for its physiological ligands may be one of the factors that have driven receptor selection and allowed EV1 to successfully adapt to exploit this receptor for cell attachment. ␣ 2 ␤ 1 integrin is usually expressed on the basolateral surfaces of cells but, interestingly, some cells such as skin keratinocytes may express ␣ 2 ␤ 1 on surfaces that do not face basement membranes or any known ␣ 2 ␤ 1 ligands (45). Both cellular entry routes may FIG. 4. Attachment of EV1 to the cell supports the clustering of integrins. A, immunofluorescent labeling of ␣ 2 ␤ 1 integrin (green) on ␣ 2 ␤ 1 -SAOS cells after 1 h of incubation with an anti-␣ 2 -integrin antibody. B, ␣ 2 ␤ 1 integrin 15 min after antibody-mediated cross-linking. C, ␣ 2 ␤ 1 on ␣ 2 ␤ 1 -SAOS cells after 15 min of incubation with EV1. D, co-localization of ␣ 2 ␤ 1 integrin (green) and EV1 (red) is seen as yellow color. Bars, 10 m.

FIG. 5. The EV1 pentamer can support clustering of integrins.
A, a comparative model of the ␣ 2 ␤ 1 integrin heterodimer with three domains of the ␣-chain (blue) and two domains of the ␤-chain (gold). B, five copies of the model (blue) were placed without steric hindrance around the virus (gold) 5-fold axis by superimposing ␣ 2 I of the ␣ 2 ␤ 1 model on the ␣ 2 I domain of the virus-receptor complex structure. play an important role in the pathogenesis of the clinical disease, since infection by EV1 is thought to take place primarily in the respiratory and gastrointestinal epithelium, but secondary virus replication occasionally is observed in various target organs.
In our model, all five binding sites around an EV1 5-fold axis can be occupied by integrin molecules without steric hindrance. This finding suggests that simultaneous binding of several ␣ 2 ␤ 1 integrins to the viral particle might also occur in vivo. Indeed, integrin clustering was observed during attachment of EV1 onto the cell surface, supporting the idea that this phenomenon leads to signaling events that may facilitate the entry of the virus. A similar phenomenon has been described in adenoviruses where multiple binding of integrins to the penton base pentamer induces clustering and gives rise to cellular signaling, having an important role in the internalization process (40). Likewise, EV1-induced integrin clustering can lead to the activation of signaling pathways and facilitate internalization of the virus via the caveolar pathway (46).
In conclusion, we report the structure of echovirus 1 in complex with the virus-binding domain of its receptor, ␣ 2 ␤ 1 integrin. To our knowledge, this is the first structural insight into picornavirus-integrin interactions. Interestingly, EV1 is the only picornavirus known to utilize ␣ 2 ␤ 1 integrin in cellular interactions and to use caveolae-mediated endocytosis in the entry into the host cells (3), suggesting that the receptor interaction may largely determine the internalization route. The integrin molecule binds to the surface depression surrounding the 5-fold axes of EV1 in an analogous but not identical manner when compared with the interaction of receptors of the immunoglobulin superfamily with polioviruses and rhinoviruses. The latter interactions give rise to in vitro uncoating, which is not seen after binding of ␣ 2 I to EV1. Whether the whole integrin molecule and additional receptors are needed for uncoating remains to be studied. It is possible that uncoating occurs later during caveolar entry, whereby the ␣ 2 I domain may stabilize the virus structure, insuring against premature genome release before EV1 has entered the targeted cell. The binding affinity of EV1 to ␣ 2 I is significantly higher than that of collagen, suggesting that the virus is able to efficiently compete for occupied basolateral binding sites. Both molecular modeling and in vivo experiments show that binding of an EV1 particle to multiple integrin molecules can cause integrin clustering, likely leading to cellular signaling, a central integrin function, as well as committing the virus-receptor complex to cell entry via the caveolar pathway.