Atomic Basis for the Species-specific Inhibition of αV Integrins by Monoclonal Antibody 17E6 Is Revealed by the Crystal Structure of αVβ3 Ectodomain-17E6 Fab Complex*

Background: 17E6, a primate-specific mouse mAb that inhibits αV integrins, is in phase II trials for treating cancer. Results: We determined crystal structure of the αVβ3–17E6 Fab complex, revealing the molecular basis of 17E6 specificity and function. Conclusion: 17E6 is an allosteric inhibitor of fibronectin-integrin interaction. Significance: The defined 17E6 epitope may help in developing novel therapeutics targeting related regions in other integrins. The function-blocking, non-RGD-containing, and primate-specific mouse monoclonal antibody 17E6 binds the αV subfamily of integrins. 17E6 is currently in phase II clinical trials for treating cancer. To elucidate the structural basis of recognition and the molecular mechanism of inhibition, we crystallized αVβ3 ectodomain in complex with the Fab fragment of 17E6. Protein crystals grew in presence of the activating cation Mn2+. The integrin in the complex and in solution assumed the genuflected conformation. 17E6 Fab bound exclusively to the Propeller domain of the αV subunit. At the core of αV-Fab interface were interactions involving Propeller residues Lys-203 and Gln-145, with the latter accounting for primate specificity. The Propeller residue Asp-150, which normally coordinates Arg of the ligand Arg-Gly-Asp motif, formed contacts with Arg-54 of the Fab that were expected to reduce soluble FN10 binding to cellular αVβ3 complexed with 17E6. This was confirmed in direct binding studies, suggesting that 17E6 is an allosteric inhibitor of αV integrins.

Integrins are ␣/␤ heterodimeric adhesion receptors that transmit biochemical and mechanical signals bidirectionally across the plasma membrane, thus serving as key communication molecules between the extracellular environment and the cytoskeleton (1). The ␣ and ␤ subunits associate noncovalently into a ligand-binding "head" segment, comprising a ␤-Propeller domain from the ␣-subunit and a von Willebrand Factor A type (vWFA) domain (␤A or I-like) from the ␤-subunit. The head segment sits on top of two leg-like extensions, each possessing a single-pass transmembrane helix, which anchors the receptor to the plasma membrane, followed by two short cytoplasmic C-terminal tails (2). Integrin function is tightly regulated: integrins are normally kept in an inactive state (unable to bind soluble physiologic ligands), thus allowing, for example, leukocytes and platelets to circulate in the blood without aggregation or adhesion to the vessel walls. Physiologic stimuli, acting through the short integrin ␣/␤ cytoplasmic tails, induce allosteric changes in the ectodomain that switch it to a high affinity state, capable of binding soluble ligand, (so called "inside-out" activation) (2). Bound ligand then induces structural rearrangements in the ectodomain, forging links via the ␤ cytoplasmic tail, with the actin cytoskeleton (3). This "outside-in" signaling can be enhanced by lateral clustering of integrins induced by multivalent ligand (4). This promotes the assembly of cytoskeletal and signaling molecular complexes at "focal adhesions," regulatory sites through which forces can be applied on liganded integrins through the cytoskeleton or extracellularly through ECM stiffening or shear stress (2). Inappropriate integrin activation contributes to the pathogenesis of many diseases including cancer growth and metastasis (1,5).
The ␣V-specific mAb 17E6 as IgG or F(abЈ) 2 inhibits cell attachment and spreading of cancer cells in vitro and in vivo and induces cell detachment (10,11), suggesting that it disrupts stable ␣V-ligand interactions (10). Cellular ␣V␤3 that is dimerized by bound IgG or F(abЈ) 2 , forms of 17E6, but not by anti-␣V␤3 mAbs 2E7 (nonblocking anti-␣V) or AP3 (nonblocking anti-␤3), is internalized (12), which may contribute to its inhibitory effects on cell adhesion, and links the 17E6 epitope to adhesion reversal. Although the primary sequence of the ␣V chain is strongly conserved between species, 17E6 does not bind murine ␣V integrins.
To explore the structural basis of its specificity and function, we crystallized the complex of the Fab fragment of 17E6 with the ␣V␤3 ectodomain (13) in the presence of the activating metal ion Mn 2ϩ . The new structure assumed a bent conformation, exhibited some flexibility at the ␣-genu, revealed the structural basis for 17E6 selectivity to the ␣V subunit of primates, and defined the antibody-integrin binding interface. The significance of these findings is discussed.

EXPERIMENTAL PROCEDURES
Reagents and Site-directed Mutagenesis-Restriction and modification enzymes were obtained from New England Biolabs, Invitrogen, or Fisher Scientific. All cell culture reagents were obtained from Invitrogen. Purified ␣V␤3 was generated and purified as described (13). The noninhibitory mAb AP3 (ATCC) detects the ␤3-subunit in all conformations. The Fab fragment of the ␣V-specific mAb 17E6 (10) generated by papain digestion was verified by SDS-PAGE. The Fab fragment of the function-blocking mAb 7E3, which binds to the ␤A domain of the ␤3 subunit (14) was purchased (Eli Lilly and Company, Indianapolis, IN). Wild-type human FN9 -10 and FN10 2 were produced as described (15). Lys-203 3 Ala (K203A) substitution in human ␣V and Lys-145 3 Asn 3 Gln-145 3 Asp (K145N/Q145D) substitution in mouse ␣V were generated by PCR and confirmed by DNA sequencing.
Cell Adhesion Inhibition Assays-Cells were allowed to adhere for 1 h at 37°C on vitronectin-coated surfaces in the presence of serially diluted 17E6 or cilengitide as described elsewhere (10). Cilengitide is an RGD-containing inhibitor of ␣V␤3 and ␣V␤5 integrins (16).
Crystallography, Structure Determination, and Refinement-The ␣V␤3-17E6 Fab complex was crystallized at room temperature by vapor diffusion using the hanging drop method. Data were collected from crystals obtained using a reservoir solution comprising sodium citrate buffer, pH 5.6, 13.5% PEG, 10 mM MnCl 2 , 1% glycerol, 90 mM trimethylamine N-oxide, and 0.3 M sodium nitrate. Crystals were frozen in liquid nitrogen. Glycerol was used as a cryoprotectant, and its concentration in the reservoir solution was increased to 23% in 2% steps. The data were collected at Beamline 19ID at Advanced Photon Source (APS) (Chicago, IL) and were processed using the HKL package (18). The asymmetric unit contained one molecule of the complex. Molecular replacement using Phaser (19) was used for structure determination. Structures of ␣V␤3 (Protein Data Bank code 1jv2) and Fab from Protein Data Bank code 2vdl were used as starting models for molecular replacement. Refinement was performed using the deformable elastic network procedure for low resolution data (20) in the CNS (21). This was followed by refinement in 1.8.4 version of PHENIX (22) using Protein Data Bank structures 3ije and Fab from 2vdl as reference models, Translation/Liberation/Screw (TLS), positional, and individual temperature factors. In addition to the default restraints, automatic optimization of x-ray and stereochemistry/ADP and additional Ramachandran restraints were used in the last cycles. Model building was performed initially using O (23) and later using Coot (24). Structural illustrations were prepared with the Chimera software (25).
EM and Imaging Studies-The ␣V␤3-17E6 complex was applied to glow-discharged continuous carbon grids, rapidly washed in water, and stained with 0.75% uranyl formate. Grids were imaged on a Philips CM10 electron microscope (100 kV; nominal magnification, 41,000ϫ; defocus range, Ϫ1.0 to Ϫ1.8 m). Images were recorded on SO163 film (Kodak) and digitized with a step size of 6.35 m (Nikon Coolscan 8000). Images were subjected to 2-fold pixel averaging to yield 2.76 Å/pixel at the image. Image processing was performed with the EMAN suite (26). 2,530 particles were selected and Contrast Transfer Function (CTF) phase-corrected. Selected particles were subjected to an initial reference-free alignment and averaging and k-means classification using 10 classes. Class averages were refined with seven rounds of multireference classification using the class averages as references as described (27).
mAb Binding to Wild-type and Mutant Cellular ␣V␤3-HEK 293T cells were co-transfected with a plasmid encoding fulllength human WT ␤3 and pcDNA3 plasmids encoding WT or K203A-mutated ␣V, together with Lipofectamine 2000 (Invitrogen) according to the manufacturer's specifications. 48 h after transfection, cells were detached (10 mM EDTA/PBS), washed two times in Hanks' balanced saline solution (HBSS) and once in HBSS containing 1 mM CaCl 2 /1 mM MgCl 2 (HBSS 2ϩ ). 6 ϫ 10 5 cells in 100 l of HBSS 2ϩ were stained with the Fab fragments of 17E6 or 7E3 (10 g/ml; 30 min; 4°C), followed by one wash, and then addition of FITC-labeled anti-mouse Fab (20 g/ml; 30 min; in the dark; 4°C). Samples were again washed once and then fixed (1% buffered paraformaldehyde). To assess expression levels, 6 ϫ 10 5 cells in 100 l of HBSS 2ϩ were incubated with AP3 mAb (10 g/ml; 30 min; 4°C), washed and then stained with allophycocyanin-labeled anti-mouse IgG (20 g/ml). 20,000 cells were analyzed for each sample using a FACSCalibur flow cytometer (BD Biosciences, San Jose, CA). Binding of 17E6 or 7E3 to ␣V␤3-expressing HEK293T was expressed as mean fluorescence intensity, as determined using the CellQuest software (BD Biosciences), and normalized by dividing this mean fluorescence intensity by the mean fluorescence intensity for AP3 binding to WT ␣V␤3 and multiplying by 100.
To assess the effect of mouse to human K145N/Q145D ␣V chimera on 17E6 binding to mouse ␣V-integrin, murine Flip-In-3T3 cells (Invitrogen) were transfected with Flip-In pcDNA5/FRT plasmids encoding mouse or human WT ␣V subunits or mouse ␣V containing the K145N 3 Q145D mutant ␣V. Transfected cells were enriched for 17E6 binding by selecting for adherent cells in each of two rounds of panning on 17E6 coated surfaces. After harvesting (10 mM EDTA/PBS), 1 ϫ 10 6 transfected cells in 100 l of DPBS (Invitrogen) containing 1 mM CaCl 2 /1 mM MgCl 2 and 1% BSA were incubated with 17E6 IgG (10 g/ml) in a total volume of 0.5 ml for 30 -60 min on ice, washed, and then stained with FITC-labeled goat anti mouse IgG (BD Biosciences) for 30 min on ice. After washing, the viable cells were analyzed by flow cytometry.
Ligand Binding Assays-K562 cells (American Type Culture) stably expressing ␣V␤3 (K562-␣V␤3) were generated as described (28) and maintained in Iscove's modified Dulbecco's medium supplemented with 10% heat-inactivated fetal bovine serum, 50 IU/ml penicillin and streptomycin, and 0.5-1.0 mg/ml G418. For ligand binding experiments, cells were harvested and washed two times in HEPES-buffered saline containing 1% BSA and 10 mM glucose (HBG buffer). 1 mM CaCl 2 with 1 mM MgCl 2 or 1 mM MnCl 2 were added to 6 ϫ 10 5 cells in 100 l of HBG buffer and incubated in the presence or absence of a saturating amount of unlabeled 17E6 Fab (30 g/ml for 30 min at 20°C), washed once in HBSS buffer containing the corresponding metal ion, and then incubated with Alexa Fluor 488-labeled wild-type FN10 or FN9 -10 (2.5 g/ml for 30 min in the dark). Cells were washed once in HBSS containing the corresponding ions, fixed (1% buffered paraformaldehyde), and analyzed by flow cytometry.
Molecular Dynamics Simulations-All atom simulations of ␣V Propeller domain and 17E6 Fab were performed with the program NAMD, using a CHARMM27 force field (29 -31). The temperature and pressure of the system were held constant at 310 K and 1 atm, respectively, using the Langevin's piston and Hoover's method, as successfully used previously for modeling integrins (29 -32). The time step was taken as 2 fs. The cutoff distance for nonbonded interactions was 1.2 nm, and the particle mesh Ewald method was used for electrostatic force calculations (29). Hydrogen atom bond length was constrained using SHAKE (33). All systems were minimized, at 20,000 steps, and equilibrated for 40 ns.
Interactions were simulated using the crystal structure of ␣V␤3-17E6 Fab complex. Q145D/K145N and K203A muta-tions were made in the Propeller structure using the software Swiss-Pdb Viewer 4.1.0 (34). The Propeller and Fab molecules were shifted by ϳ1 Å from their original bound state to compare the effects of the mutations. Each of the three systems was solvated in a water box of 188 ϫ 127 ϫ 139 Å in size and ionized with 150 mM NaCl to represent the solvent.
The association constant (k a ) of Propeller-Fab interaction was calculated according to Equation 1, where ⌬G, k B , and T are the free energy of association per mole, Boltzmann constant, and temperature, respectively. Gibbs free energy was calculated according to Equation 2, where U, P, V, S, and T are the interaction energy, pressure, volume, entropy, and temperature (at 310 K), respectively. Interaction energy was computed as the sum of the electrostatic and van der Waals energies. The ratio of k a for mutant to that of wild-type structure was derived according to Equation 3.

RESULTS
17E6 Only Inhibits Primate ␣V Integrins-Established adherent cell lines were examined for sensitivity to 17E6 in cell adhesion-inhibition assays on the ␣V␤3/␣V␤5 integrin ligand vitronectin. Of 18 species tested (primates to avians), only cells from human and monkey were inhibited by 17E6 (IC 50 ϭ 0.13-1.3 nM). IC 50 was not attained up to 300 nM with cells from any other species. This was not due to lack of integrin expression, because the attachment to vitronectin of cells from every species was specifically inhibited by the ␣V␤3/␣V␤5 peptidic inhibitor cilengitide with IC 50 scattered at ϳ1 M (16, 35) ( Table 1 and Fig. 1). We investigated the structural basis for this specificity.
Crystal Structure of Fab in Complex with ␣V␤3 Ectodomain-The water-soluble ectodomain of ␣V␤3 (13) formed a stable complex with the Fab fragment of 17E6, and the complex was crystallized in presence of the integrin activating metal ion Mn 2ϩ . We determined the crystal structure of the ␣V␤3-17E6 Fab complex by molecular replacement and refined the model with data to 3.6Å resolution. The structure contained one integrin-Fab complex in the asymmetric unit, consisting of ␣V residues (Phe-1 to Ile-833 and Leu-872 to Trp-953), and ␤3 residues Gly-1 to Cys-687, Fab heavy chain residues Gln-1 to Arg-218 and light chain residues Asp-1 to Cys-214 ( Fig. 2A). The four metal ion sites at the base of the Propeller and at ADMIDAS of ␤A were occupied, but no metal ion density was present at MIDAS, LIMBS, or at the ␣-genu. Crystallographic data and refinement statistics are given in Table 2. ␣V␤3 in complex with 17E6 Fab assumed a bent conformation. EM class averages of 17E6-bound ␣V␤3 segregated into two different orientations (Fig. 2B) and showed that the integrin also assumed a bent conformation when adsorbed from solution, thus complementing the crystallographic data.
Interactions between ␣V␤3 and 17E6 Fab-17E6 Fab only contacted the ␣V subunit of the integrin heterodimer and did so primarily through its heavy chain. The three heavy chain CDRs (CDR H1-3 ) contacted loop residues in blades 2 and 3 of the ␣V Propeller domain (Fig. 3A). Core interactions involved Gln-145 and Lys-203 from D2A3 loop and D3 strand of the Propeller domain, respectively, which inserted into adjacent hydrophobic pockets in the heavy chain (Fig. 3B) Atomic Basis for Specificity of mAb17E6 for Primate ␣V␤3-Sequence analysis of the species tested in cell adhesioninhibition assays (Table 1) revealed that every species other than humans and macaque lacked Gln-145 3 Asp and expressed Lys-145 3 Asn or in avians Thr-Thr at this site (Fig.  4). By contrast Lys-203 was conserved across species (Fig. 4). Replacing Lys-145 3 Asn with Gln-145 3 Asp in mouse ␣V and expressing this chimeric ␣V with the endogenous mouse ␤-chains in 3T3 cells imparted reactivity of 17E6 with the chimeric mouse cell surface receptor to essentially the same level as full-length human ␣V (Fig. 5A). The importance of the con-   served Lys-203 for 17E6 interaction with ␣V was evaluated in a K203A mutation of human ␣V. K203A mutant ␣V did not affect folding of the integrin as assessed by surface binding of mAbs AP3 and 7E3 (Fig. 5B); however, it was no longer recognized by 17E6 Fab (Fig. 5B). Molecular Dynamic Simulation of ␣V␤3-17E6 Interaction Supports the Structural and Mutational Data-The energy of interaction between the integrin ␣V Propeller and 17E6 Fab was simulated for wild-type Propeller, K203A, and Q145D/ K145N mutants (Fig. 5C). The energy of interaction for wildtype Propeller fluctuated at ϳ500 kCal/mol, but K203A showed an abrupt energy drop to ϳ300 kCal/mol within ϳ0.2 ns after initiation of the simulation, reflecting loss of the salt bridge between Lys-203 and Asp-106 H3 .
The Lys-203-Asp-106 H3 bond distance in the Q145D/ K145N ␣V mutant yielded a minimum distance of ϳ2.5 Å as in wild type during equilibration but showed much larger fluctuations, reflecting less stable binding (Fig. 5D). The energy of interaction of Q145D/K145N with 17E6 Fab dropped, first to ϳ450 kCal/mol then to ϳ300 kCal/mol. Initially, the Asp-146 -Arg-54 H2 bond disappeared in Q145D/K145N structure, replaced by Asp-148 and Asp-150 bonds with Arg-54 H2 (Fig.  5E). These bonds, as well as a Lys-145-Phe-99 H-bond, were lost at 10 -12 ns, with a simultaneous reduction in the energy of interaction between the Q145D/K145N Propeller and 17E6 Fab to ϳ300 kCal/mol (Fig. 5C).
Inserting the average interaction energies over the last 10 ns of simulations for wild-type (Ϫ478 kCal/mol), K203A (Ϫ338 kCal/mol), and Q145K/D146N (Ϫ292 kCal/mol) Propeller into Equation 3 yielded association constant ratios of 2 ϫ 10 Ϫ99 for  Fig. 2A. B, electrostatic potential surface of 17E6 Fab heavy chain at the integrin interface. Two adjacent pockets in Fab-H house human integrin residues Gln-145 and Lys-203. The orientation in B is the same as in A.  K203A, and 8 ϫ 10 Ϫ132 for Q145D/K145N. Thus either Propeller mutation markedly reduced affinity of 17E6, despite maintenance of other ionic and van der Waals bonds between the two interactants. The above biochemical and computational studies thus verify the structural model and underscore the crit-ical roles of Gln-145 3 Asp and Lys-203 in 17E6 recognition of human ␣V␤3. Impact of Bound 17E6 Fab on Fibronectin Binding to Cellular ␣V␤3-Superimposition of the ␣V␤3-17E6 Fab structure onto that of ␣V␤3-FN10 structure (Protein Data Bank code 4mmx), 3 predicts that no clashes would occur between bound 17E6-Fab and FN10 (Fig. 6). In RGD-bound ␣V␤3 structures (35), the ligand Arg inserts into a pocket in the Propeller domain and is stabilized by salt bridges to OD1 and OD2 of Asp-218 and to OD1 of Asp-150. In ␣V␤3-17E6 Fab structure, stability of the ligand Arg in the Propeller pocket is predicted to be weakened by loss of the salt bridge to Asp-150, which contacts Arg-54 H2 instead (Fig. 6, top panel, inset). Furthermore, docking FN10 from each of 10 nuclear magnetic resonance structures of the FN9 -10 fragment (Protein Data Bank code 2mfn) onto that of FN10 in the superimposed ␣V␤3-FN10 and ␣V␤3-17E6 structures suggested that FN9 would clash mainly with the light chain of bound 17E6 Fab, even when the least conflicting FN9 -10 model (model 4.7) was used (Fig. 6). This predicted that binding of FN9 -10 to cellular ␣V␤3 would be inhibited by 17E6 Fab. This was tested experimentally. Binding of soluble fluoresceinated FN10 and FN9 -10 to full-length ␣V␤3 stably expressed on K562 cells was measured in presence or absence of unlabeled 17E6 Fab. 17E6 Fab reduced binding of FN10 to  Mn 2ϩ -activated cellular ␣V␤3 by some 40% and that of FN9 -10 by some 60% (Fig. 7).

DISCUSSION
In this report, we describe the first crystal structure of an integrin ectodomain complexed to a function-blocking mAb, the Fab fragment of 17E6, in presence of Mn 2ϩ , an integrinactivating cation. The new structure allowed an atomic level analysis of the ␣V␤3-17E6 interface. It showed that 17E6 exclusively contacted the ␣V Propeller domain and did so at those critical residues on blades 2 and 3 that precisely correspond to the demonstrated phylogenetic specificity of inhibition of 17E6. The binding site reveals a mechanism by which 17E6, which itself lacks an RGD sequence, allosterically inhibits RGD-based ligand interaction with cellular ␣V integrins.
The integrin in the complex assumes a bent conformation and also in solution, suggesting that the well established integrin activator Mn 2ϩ does not necessarily switch the ␣V␤3 ectodomain to a linear conformation as has been suggested (36). This finding is consistent with our earlier fluorescent lifetime imaging study revealing a bent 17E6-bound cellular ␣V␤3 in Mn 2ϩ (15) and a cryoelectron tomography study showing no dramatic genu extension upon Mn 2ϩ -induced activation of membrane-bound unliganded ␣IIb␤3 integrin (37). It could be argued that binding of 17E6 prevents Mn 2ϩ -induced conformational switching of ␣V␤3. This seems unlikely, because 17E6 Fab bound to ␣V␤3 did not block an increase in hydrodynamic radius of the ectodomain complex triggered by a cyclic RGD ligand. Furthermore, binding of 17E6 did not block a subsequent Mn 2ϩ -induced expression of a conformation-dependent LIBS (15). We also found that despite the crystallization of the ␣V␤3-Fab complex in the presence of 5 mM MnCl 2 , only ADMIDAS but neither MIDAS nor LIMBS of ␣V␤3 were occupied by a metal ion. However, all three sites were occupied by Mn 2ϩ when ␣V␤3 was bound by an RGD ligand (35), suggesting that the latter appears necessary for stable coordination of the metal ion at MIDAS and LIMBS in ␣V␤3.
The ␣V␤3 ectodomain from the complex was superimposable on unliganded ␣V␤3 (Protein Data Bank code 3ije) (15), yielding a C␣ root mean square deviation of 3.44 Å. This improved to 1.53 Å when superimposition excluded Calf-1/2 and ␤TD domains, which gave a root mean square deviation of 4.31 Å in such an alignment. Conversely, superimposition of Calf-1/2 and ␤TD domains exclusive of the rest of the integrin domains in the two structures gave a root mean square deviation of 1.53 Å. These data suggest that a rigid body movement of Calf-1/2 plus ␤TD differentiated the two structures. The origin of this movement is likely the different crystal contacts from symmetry-related molecules in the two structures of 17E6bound versus unliganded ␣V␤3. The ␣V␤3-17E6 Fab structure, compared with the unliganded ␣V␤3 structure, lacks a major stabilizing area of interaction (595 Å 2 ) involving Calf-2 with the Propeller from a symmetry-related molecule. However, at the top of Calf-1, including the ␣-genu, it does have an interaction (207 Å 2 ) with the L-chain from a symmetry-related molecule. These differences likely account for the higher mobility of the Calf-1/2-␤TD domains in the ␣V␤3-17E6 complex, allowing only a partial assignment of their side chains in the present structure. The higher mobility and/or the crystal contact Calf-1 makes with the symmetry-related molecule may also account for a lack of a visible metal ion at the ␣-genu.
17E6 recognizes the ␣V subunit only in human and primate integrins (10,38,39). Using ␣V-␣5 chimeric subunits expressed as heterodimers with ␤1 on viable Chinese hamster ovary cells, Mould et al. (40) mapped the 17E6-binding site in human integrin ␣V to Propeller residues Ala-107-Trp-226. We found that this region, which is identical in sequence to that in rodents except for a K145N substitution of human residues Q145D, contains all the residues that directly contact 17E6 Fab. Binding of 17E6 to chimeric ␣V/␣5 in which the ␣V sequence Gln-145-Asp-Ile-Asp-Ala-Asp was replaced with the corresponding sequence Asp-154 -Phe-Ser-Trp-Ala-Ala from integrin ␣5 subunit (40) indicated that Gln-145 is the critical residue accounting for primate specificity and that it could be replaced by an Asp but not a Lys (as in the rodent sequence). This interpretation is consistent both with the crystal structure and with mutational analysis. In the crystal, an Asp replacing Gln-145 could form the same H-bonds with the carbonyl oxygens of Phe-99 and Leu-100 and the amide main chain of Trp-33. In cells, we could show that replacing murine Lys-145 3 Asn with human Gln-145 3 Asp in ␣V was a gain of function mutation for 17E6 binding to mouse ␣V. Interestingly, excepting primates, this pair of residues is also substituted in those many other species, where cell adhesion, as we show here, is not inhibited by 17E6.
How does mAb 17E6 impair ligand binding to ␣V␤3? Superimposing the NMR structures of wild-type FN9 -10 onto FN10 docked into the ␣V␤3-17E6 Fab complex predicted a significant clash of FN9 mainly with the light chain of 17E6, which would hinder binding of soluble FN9 -10 to cellular ␣V␤3. Binding inhibition studies supported this prediction, showing ϳ60% reduction of FN9 -10 binding when 17E6 was bound to cellular ␣V␤3. FN9, although unnecessary for binding of FN to ␣V␤3 (41), may yet clash with bound 17E6, thus reducing binding of FN9 -10. The known flexibility of the FN9 -10 interdo- main linkage (42,43) may explain why inhibition of binding of FN9 -10 to cellular ␣V␤3 by 17E6 is not complete. The modest size of the ␣V␤3/FN10 interface seen in the crystal structure of this complex 3 may make it particularly susceptible to allosteric inhibition.
Mutating Asp-150 in the ␣V Propeller domain to alanine (Asp-150 3 Ala) has been shown to reduce binding of the RGD-containing ligand-mimetic WOW-1 Fab to cellular ␣V␤3 by 37% (44). Our findings show that 17E6 Fab caused a 41% inhibition of binding of soluble RGD-containing FN10 to cellular ␣V␤3. This reduction may be explained by the predicted loss of the salt bridge Asp-150 normally makes with the ligand Arg. Thus 17E6-bound ␣V␤3 appears to functionally behave like the described Asp-150 3 Ala ␣V␤3 mutant (44).
In addition to the inhibition of soluble ligand binding to cellular ␣V␤3 by monovalent 17E6, divalent 17E6 reduces the number of cell surface ␣V␤3 receptors through internalization, which may also contribute to the inhibition of cell adhesion by divalent 17E6 in vitro (11) and retarded growth of model melanoma in nude mice (10). Finally, integrin clustering by multivalent physiological ligands likely orients the integrin in the membrane into ordered arrays, which then encourage the formation of signal transduction complexes (45). Dimerization of ␣V␤3 by divalent 17E6 may interfere with such ordering of the liganded integrin in the membrane, destabilizing ␣V␤3-substrate interactions, altering signal transduction, and contributing to a greater effectiveness of divalent versus monovalent 17E6 in blocking cell adhesion.
Integrins are appealing therapeutic targets (46). The structure of ␣V␤3-17E6 Fab complex reveals the detailed inhibitory interactions between this mAb and its epitope. Changes might be engineered to productively modify the antibody-integrin interaction and potentially be used to guide development of novel therapeutics targeting related regions in other integrins. Because the 17E6 epitope is distinct from the ligand-binding MIDAS face of the ␤-subunit, perhaps 17E6 in combination with ligand mimetic drugs might achieve synergistic or additive therapeutic effects.