Evidence That the Integrin β3 and β5 Subunits Contain a Metal Ion-dependent Adhesion Site-like Motif but Lack an I Domain*

The amino-terminal domain of each integrin β subunit is hypothesized to contain an ion binding site that is key to cell adhesion. A new hypothesis regarding the structure of this site is suggested by the crystallization of the I domains of the integrin αL and αM subunits (Lee, J.-O., Rieu, P., Arnaout, M. A., and Liddington, R. (1995) Cell 80, 631–638; Qu, A., and Leahy, D. J. (1995) Proc. Natl. Acad. Sci. U. S. A. 92, 10277–10281). In those proteins, an essential metal ion is bound by a metal ion-dependent adhesion site (MIDAS). The MIDAS is presented at the apex of a larger protein module called an I domain. The metal ligands in the MIDAS can be separated into three distantly spaced clusters of oxygenated residues. These three coordination sites also appear to exist in the integrin β3 and β5 subunits. Here, we examined the putative metal binding site within β3 and β5 using site-directed mutagenesis and ligand binding studies. We also investigated the fold of the domain containing the putative metal binding site using the PHD structural algorithm. The results of the study point to the similarity between the integrin β subunits and the MIDAS motif at two of three key coordination points. Importantly though, the study failed to identify a residue in either β subunit that corresponds to the second metal coordination group in the MIDAS. Moreover, structural algorithms indicate that the fold of the β subunits is considerably different than the I domains. Thus, the integrin β subunits appear to present a MIDAS-like motif in the context of a protein module that is structurally distinct from known I domains.

Integrins are ␣␤ heterodimers that mediate cell adhesion (1,2). Integrins participate in development and tissue remodeling and are linked to several diseases. The integrins bind to many adhesive and extracellular matrix proteins. The focal points of this study are the ␣v␤3 and ␣v␤5 integrins, both of which recognize the Arg-Gly-Asp (RGD) 1 tripeptide motif. The ␣v␤3 integrin binds to at least nine adhesive proteins and has two important biological functions. First, ␣v␤3 mediates the adhesion of osteoclasts to the bone surface (3), an event often considered to be the first step in bone resorption (4). Second, the ␣v␤3 integrin is expressed on the surface of angiogenic endothelial cells, where it is required for cell survival and further vessel development (5)(6)(7). It has been suggested that inhibitors of the ␣v␤3 integrin could be applied as antagonists of osteoporosis and tumor angiogenesis. The biological function of the ␣v␤5 integrin is less clear. This integrin can mediate cell adhesion to vitronectin. The ␣v␤5 integrin is also required for the internalization of adenovirus (8,9), and it may be associated with angiogenesis (7).
All integrins require divalent cations to bind their ligands. An important clue to the structural basis for ion binding was revealed by the crystal structures of the I domains from the integrin ␣ L and ␣ M subunits (10,11). Each I domain spans approximately 200 residues and is homologous to an "inserted" domain in a number of other proteins including von Willebrand factor (12). In ␣ L and ␣ M , the I domain is necessary and sufficient for ligand contact. These I domains contain a metal binding site called a MIDAS (metal ion-dependent adhesion site). This ion binding site consists of five liganding residues that can be separated into three groups. Each group of coordinating residues is located at separate positions within the primary amino acid sequence (10,11). The first coordination group consist of the DXSXS sequence, where D is aspartate, X is any amino acid, and S is serine. The aspartate and both serines coordinate with metal ion. The second group, or coordination point, is a single threonine located 69 amino acids from the DXSXS. The third group is comprised of a single aspartate 102 residues from the DXSXS.
Interestingly, the DXSXS sequence is also present in the integrin ␤ subunits (13), suggesting that they may also contain the MIDAS (10). If correct, this would provide a common structural basis for the regulation of all integrins by divalent metal ions. It would also imply that all integrins are regulated in a similar manner by metal ion. Despite this hypothesized similarity, integrins behave differently with respect to metal ions. For example, we recently demonstrated that the type of divalent ion present in the culture media regulates the way that the ␣v␤3 and ␣v␤5 integrins are organized on the cell surface (14). In fact, the same ion can direct the two integrins to completely different locations on the cell. This distinction suggests that the metal binding site within the ␤ subunits is likely to have subtle but important structural differences that have an impact on receptor function.
To examine the hypothesized ion binding site within ␤3 and ␤5, the putative metal coordinating residues within each sub-unit were mutated to alanine. Results presented here are the first to show that Asp-119 and Asp-217 within ␤3 are important for the binding of soluble ligand to ␣v␤3. The homologous aspartic acids within ␤5 are also key to soluble ligand binding. Interestingly though, mutations at these aspartic acids do not completely abrogate cell adhesion. Their mutation decreases cell adhesion and reduces the apparent affinity of the integrin for metal ion. This finding indicates that each aspartic acid is likely to be part of a metal ion binding site that controls ligand contact. The similarity in spacing and function of these aspartates to the metal ligands in ␣ L and ␣ M indicate a similarity between the integrin ␤ subunits and the MIDAS motif found in the I domains. However, evidence is presented here which argues that the fold of the ␤ subunits is distinct from that of the I domains. The PHD structural algorithm predicts that the ␤3 and ␤5 subunits have little structural similarity with the I domains.

EXPERIMENTAL PROCEDURES
Sequence Alignment and Structural Predictions-Amino acid sequences of ␤3 (residues 107-292) and ␤5 (residues 109 -296) were each aligned with the I domain sequences of ␣ L (residues 125-310) and ␣ M (residues 128 -318) using the multiple sequence alignment program, ClustalW (15). Without further manipulation, the putative metal ligands were identified in each ␤ subunit by comparison to the known metal coordination sites within the MIDAS motifs of ␣ L and ␣ M .
The structures of ␤3, ␤5, and the I domains of ␣ L and ␣ M were analyzed using the PHD algorithm (16,17). The algorithm compares the environment of a single residue within a data base of known crystal and NMR structures and then assigns the probability that a residue is in a helix or a sheet. The algorithm also assigns a reliability index from 0 to 9 to assist in distinguishing residues which could be present in either structure. In this analysis, structural predictions for individual residues were only accepted when the reliability index was greater than five. The amino acid sequences encompassing residues 107-292 in ␤3, residues 109 -296 in ␤5, residues 125-310 in ␣ L , and residues 128 -318 in ␣ M were subjected to this analysis. Individual structural elements in a predicted protein were considered to be a match (compared with elements in the known crystal structure of ␣ L or ␣ M ) only if at least 50% of the residues in the element were predicted to be in the correct position and of the correct structure (helix or sheet).
Cell Lines and FACS Analysis-Human embryonic kidney 293 cells were obtained from ATCC and maintained in Dulbecco's modified Eagle's medium (BioWhittaker) supplemented with 10% fetal calf serum (Irvine Scientific). Mutated constructs of human ␤3 or ␤5 cDNA in plasmid pcDNA3 were transfected into 293 cells using DOTAP transfection reagent (Boehringer Mannheim). Following selection in 500 g/ml geneticin (Life Technologies, Inc.) for approximately 3 weeks, the top 5% of the positive fluorescent population of cells was obtained by sterile FACS using either the anti-␣v␤3 monoclonal antibody (mAb) LM609 or the anti-␣v␤5 mAb P1F6. Cells expanded from the sorted population were continuously monitored for high expression of transfected integrin throughout the duration of the study. FACS analysis was performed using standard protocols. Cells were incubated with mouse primary antibody against ␣v␤3 or ␣v␤5 (5 g/ml), washed, and then incubated with fluorescein isothiocyanate (FITC)-conjugated goat anti-mouse secondary antibody (Caltag). All cells used for binding studies exhibited stable integration of the ␤ subunit cDNA into the genome.
Antibodies and Synthetic Peptides-The anti-␣v␤3 mAb LM609 was purchased from Chemicon. Nonspecific mouse IgG was obtained from Calbiochem. Monoclonal antibody P1F6 (anti-␣v␤5) was purchased from Becton Dickinson. Synthetic peptides with sequences GRGDSP and SPGDRG were obtained from Coast Scientific.
Binding of Fab-9 to ␣v␤3 Expressed on 293 Cells-Fab-9 is an RGDcontaining, synthetically engineered antibody that has been optimized through phage display to bind to ␤3-integrins (18,19). Fab-9 has at least a thousand-fold lower affinity for ␣v␤5 (18) and does not bind specifically to 293 cells transfected with ␤5. 293 cells expressing the ␤3 mutants were harvested from tissue culture flasks with 0.2 mM EDTA in phosphate-buffered saline (EDTA/PBS) and washed with cold Binding Buffer (Hanks' balanced salt solution lacking MgCl 2 , CaCl 2 , and MnCl 2 (Life Technologies, Inc.), 50 mM HEPES, pH 7.4, 3 mg/ml bovine serum albumin) supplemented with 0.5 mM MgCl 2 and 0.05 mM MnCl 2 . These cation conditions were optimized for Fab-9 binding. Fab-9 was labeled with Na 125 I (Amersham Corp.) using IODO-GEN (Pierce) to approximately 40,000 cpm/ng. Cells (3 ϫ 10 6 /ml) were incubated with increasing concentrations of 125 I-Fab-9 at 14°C for 70 min. Nonspecific binding was measured by including 20 mM EDTA in the incubations (18), although prior study in this lab has shown that competition with an excess of unlabeled Fab-9 yields nearly identical values. Following incubation, free 125 I-Fab-9 was separated from cell-bound ligand by centrifugation through 20% sucrose, 50 mM Tris-buffered saline, pH 7.4, at 14,000 rpm for 3 min in disposable microcentrifuge tubes (Fisher). The bottoms of the tubes were cut off and counted in a gamma counter. All data represent the average of triplicate measurements. All assays were repeated at least twice yielding identical results. The affinity of ␣v␤3 for Fab-9 was calculated by Scatchard analysis (20).
To measure the apparent affinity of integrin for Mg 2ϩ , Fab-9 binding was measured as a function of Mg 2ϩ concentration. The apparent affinity for cation was taken as the concentration of cation that supported half-maximal ligand binding.
Binding of Vitronectin to 293 Cells Expressing ␣v␤5-Ligand binding to cells expressing ␣v␤5 and mutant forms of ␤5 was measured with 125 I-vitronectin (Vn) using a tracer format (21). For that purpose, vitronectin was purified from human serum as described previously (22) and labeled to high specific activity (150,000 cpm/ng) with Na 125 I. Cells expressing the mutants of ␤5 were harvested with EDTA/PBS and washed with Binding Buffer containing 0.5 mM MgCl 2 and 0.02 mM CaCl 2 . Cells (10 7 /ml) were incubated with 0.5 nM 125 I-Vn and increasing concentrations of unlabeled Vn in cation-supplemented Binding Buffer at 14°C for 70 min. Specific binding was determined by subtracting the EDTA-sensitive binding from total binding. The affinity of ␣v␤5 for vitronectin was derived using Scatchard analysis (20).
To measure the apparent affinity of integrin for divalent cation, the binding of 125 I-Vn was measured across a range of metal ion concentration. The apparent affinity for cation was determined as the concentration of ion at which half-maximal ligand binding was observed.
Cell Adhesion Assays-Cell adhesion assays were performed as described previously (23). Briefly, Fab-9 (50 nM) or vitronectin (6 nM) were coated on 96-well plates by an overnight incubation at 4°C. The plates were then blocked with 1% bovine serum albumin. Cells (1 ϫ 10 5 cells/well) were harvested with EDTA/PBS and resuspended in Binding Buffer containing the appropriate cations as described for soluble ligand binding. Cells were allowed to adhere at 37°C for 45-60 min. Non-adherent cells were removed by gentle washing, and adherent cells were detected by a colorimetric assay for lysosomal acid phosphatase (24). Color absorbance was detected at A 405 nm .
Measuring the Binding of Antibody AP5-The binding of mAb AP5 to ␣v␤3 on 293 cells was measured as described previously (25). Briefly, cells were incubated with 50 g/ml FITC-conjugated AP5 in the presence of varying concentrations of Ca 2ϩ and analyzed by FACS. Mean fluorescence intensity was determined per 10,000 cells.

Sequence and Structural Comparison between the I Domain of ␣ M and the Amino-terminal
Domain of ␤3 and ␤5-As a first step toward characterizing the putative metal binding site in ␤3 and ␤5, their sequences were compared with the I domain of ␣ M . The sequence of ␤3 (residues 107-292) was aligned with the I domain of ␣ M using the multiple sequence alignment program ClustalW (15). The sequence of ␤5 was incorporated into this alignment using the extensive identity between ␤3 and ␤5. The DXSXS motif and an aspartic acid residue representing the third coordination group in the MIDAS align well in all three proteins. The ␤ subunits diverge from the MIDAS motif at the middle metal coordinating position (Thr-209 in ␣ M ). The ␤ subunits contain a small disulfide-bonded loop in this region, a structure absent in the I domains. Within this disulfide-bonded loop, ␤3 contains two threonine residues (Thr-182 and Thr-183) that could be homologues of the coordinating threonine in ␣ M . Within the same disulfide-bonded loop of ␤5, Asn-186 and Ser-190 could potentially ligand with metal ion.
To examine further the homology between the I domain and the integrin ␤3 and ␤5 subunits, structural predictions were made using the PHD algorithm under strict conditions (16,17). As shown in Fig. 1, the algorithm correctly predicted 11 of 13 structural elements in ␣ M and 10 of 13 elements in the I domain of ␣ L (not shown). These findings validate the fidelity of the algorithm. The algorithm predicted that ␤3 contains only 3 of the 13 structural elements in the I domains and that ␤5 may contain up to 2 of the I domain elements. Consequently, the overall fold of this domain in the ␤ subunits is likely to be significantly different than that of the known I domains.
The Metal Ion Preference of ␣v␤3 and ␣v␤5 Is Similar to the I Domain of ␣ M -As a second step in assessing the similarity between the metal binding site in the ␤ subunits and the MIDAS, the ion preference of ␣v␤3 and ␣v␤5 was compared with that of ␣ M (26). The I domain of ␣ M will bind ligand in a series of metals, although Ca 2ϩ and Ba 2ϩ are largely ineffective in supporting binding. Therefore, we tested the same panel of metals for the ability to support ligand binding to ␣v␤3 and ␣v␤5. This analysis was done by measuring the binding of ligand to each ␣v-integrin as a function of the type of divalent metal ion present in the binding buffer. All metals were tested at a concentration of 1 mM to be consistent with the prior study of the I domain of ␣ M (26). As shown in Fig. 2, most metal ions tested support the binding of vitronectin to ␣v␤5. However, Ca 2ϩ and Ba 2ϩ supported only minimal binding to ␣v␤5. A similar metal preference was observed for the binding of ligand to ␣v␤3 (not shown). The only significant difference between the two ␣v-integrins was the inability of Cd 2ϩ to support ligand binding to ␣v␤3 (not shown). These data show that ligand binding to ␣v␤3, ␣v␤5, and the I domain is supported by similar metals.
Expression of the Mutant Forms of ␤3 and ␤5 in 293 Cells-To probe the structure of the ion binding site within each ␤ subunit, we mutated putative metal liganding residues to alanine. In ␤3 these are Asp-119, which represents the DXSXS sequence; Thr-182 and Thr-183, which are hypothesized to make up the second coordination group; and Asp-217, which is thought to comprise the third coordinating group. Within ␤5 the putative coordinating residues are Asp-121, Asn-186, Ser-190, and Asp-220. Because of their proximity to the last putative coordination residue, we also mutated Glu-220 within ␤3 and Glu-223 within ␤5. Each cDNA construct was used to transfect 293 cells. Following antibiotic selection and FACS sorting, transfected cells were found to express nearly equivalent levels of each of the mutated integrins on the cell surface (Fig. 3). One mutant, ␤3 E220A, was not expressed on the cell surface, even though repeated attempts were made to transfect this mutant. To confirm proper heterodimer forma-  (15). The positioning of the MIDAS residues in ␣ L and ␣ M are identical, so only ␣ M is shown. The amino acid residues that coordinate metal ion in ␣ M and the predicted coordinating residues in ␤3 and ␤5 are boxed. These same regions were subjected to structural prediction using the PHD algorithm (16,17). The predicted structures were compared with the actual structure of ␣ M as determined by crystallographic data (10). ␤ strands are represented as hatched rectangles; ␣ helixes are shown as shaded rectangles, and turns/random coils are left open. Secondary structural elements according to the ␣ M crystal structure are labeled as ␤ sheet strands A-F and ␣ helixes 1-7 (10). Crystal structure is abbreviated as X and predicted structure as P. The position and length of each element in the figure is shown to scale. tion, the mutant integrins were immunoprecipitated from lysates of 125 I-labeled cells using antibodies against ␣v␤3 (LM609) and ␣v␤5 (P1F6). Each antibody immunoprecipitated ␣ v and the relevant ␤ subunit in an approximate 1:1 stoichiometry, confirming that the mutated subunits complex with ␣ v (data not shown).
Assessing the Ligand Binding Function of Mutated Forms of ␤3-The ligand binding function of each mutated form of ␣v␤3 was measured using the model ligand Fab-9 which has been characterized previously (18,19,27). This ligand was chosen because the binding affinity of soluble vitronectin for ␣ v ␤ 3 on 293 cells was too low to yield reproducible binding data. In these binding studies, the metal concentration was set to an optimal level. Wild-type ␣ v ␤ 3 bound to 125 I-Fab-9 with an affinity of 9 Ϯ 3 nM (n ϭ 10). Within the detectable range of binding, the mutation of ␤3 residues D119A and D217A abolished binding of soluble Fab-9 to the cell surface (Fig. 4). Surprisingly, cells expressing the T182A and T183A mutations bound to 125 I-Fab-9 with an affinity identical to that of wildtype ␣v␤3. To determine whether the mutations T182A and T183A had a more subtle effect on cation-dependent ligand binding, we measured their apparent affinities for Mg 2ϩ . The apparent affinities of T182A or T183A for Mg 2ϩ , as reported by Fab-9 binding, were identical to wild-type ␣v␤3 (Table I). Thus, unlike the corresponding threonines within ␣ L and ␣ M , neither of the candidate threonines within ␤3 appear to be crucial metal ligands.
Cell adhesion is a multimeric interaction between clustered integrins and a non-diffusable matrix. Therefore, it can often proceed even when the affinity between integrin and ligand is very low. Consequently, we measured the effect of mutations within the ␤3 subunits on cell adhesion to immobilized Fab-9. Surprisingly, cells expressing ␤3 mutated at Asp-119 and Asp-217 adhered to Fab-9 (Fig. 5), even though they failed to bind soluble ligand. The mutated forms of ␤3 supported adhesion to a level that usually reached approximately 40% that of wildtype ␤3. More importantly, both mutations also exhibited a shift in the apparent affinity for metal ion. This was measured by determining the level of metal ion that supported halfmaximal adhesion. The study was done with Mn 2ϩ because it has the highest affinity for the integrin. The D119A mutation exhibited an apparent affinity for ion that was approximately 6 -10-fold lower than that of wild-type ␤3. The mutation at Asp-217 was even more deleterious, exhibiting an apparent affinity for metal that was 15-20-fold lower than wild-type ␤3. These are the first data to demonstrate that mutations at putative metal-coordinating residues within an integrin ␤ subunit shift the ion response curve. This can be interpreted to indicate that Asp-119 and Asp-217 contribute to metal binding affinity.
Assessing the Ligand Binding Function of Mutant Forms of the ␤5 Subunit-The binding of vitronectin to wild-type ␣v␤5 on 293 cells was initially characterized in conditions containing 500 M Mg 2ϩ and 20 M Ca 2ϩ . Under these cation concentrations, the binding of vitronectin was specific and saturable with a K d of 9 nM. Vitronectin binding to cells expressing wild-type ␣v␤5 could be completely inhibited with function-blocking mAb P1F6 or GRGDSP peptide (data not shown). Each mutant of ␣ v ␤ 5 was evaluated for its ability to bind soluble vitronectin. The binding of vitronectin was assayed as a function of the concentrations of Mg 2ϩ or Mn 2ϩ (Fig. 6). The titration of Mn 2ϩ was carried out to only 5 mM because artifactual binding of vitronectin was detected above this concentration. In this experiment, the data are expressed as a percentage of maximal binding to wild-type ␣v␤5 which was always measured in parallel. Cells expressing alanine mutations at Asp-121 and Asp-220 failed to bind to soluble Vn in either Mg 2ϩ or Mn 2ϩ . In the radioligand binding assay that was employed, we were only able to detect vitronectin binding to integrin when the K d was below 500 nM. Since wild-type ␣v␤5 has a K d of 9 nM for soluble  figure). Cells were maintained in antibiotic selection and FACS-sorted to obtain a population that expressed high levels of integrin. Here, FACS was used to analyze the expression level of the ␣v␤3 or ␣v␤5 heterodimer on each sorted cell population. Cells were incubated with a nonspecific mouse IgG (clear peak) or an antibody (shaded peak) that binds to the ␣v␤3 (mAb LM609) or ␣v␤5 (mAb P1F6) complex. The binding of primary antibody was detected with secondary antibody conjugated to FITC. Kidney 293 cells mock-transfected with the pcDNA3 expression vector did not exhibit any shift in fluorescence (not shown).
vitronectin, we conclude that mutations at Asp-121 and Asp-220 cause at least a 55-fold reduction in the affinity of the integrin for vitronectin. These data are consistent with the role of each aspartate in metal coordination and with nearly identical data obtained for ␤3 (see above). The mutation of Asn-186 and Ser-190 to alanine had no effect on ligand binding. Thus, we were unable to identify a residue in ␤5 that corresponds to the metal coordinating threonine (Group 2) in ␣ L and ␣ M . It is also interesting to note that the mutation of Glu-223 to alanine eliminated the ability of ␣v␤5 to bind soluble vitronectin. Although this residue is not homologous to any of the metal ligands in the MIDAS, our data indicate that it has a role in ligand binding function. It may, in fact, substitute for the missing second metal ligand (see "Discussion").
The ability of each mutant form of ␣v␤5 to mediate adhesion to immobilized vitronectin was also measured (Table II). The substitution of alanine at Asp-121 and Glu-223 of ␤5 resulted in complete abrogation of cell adhesion, whereas alanine substitutions at ␤5 residues Asn-186 and Ser-190 had no effect on the ability of the cells to adhere to vitronectin. In contrast, ␤5 containing D220A mediated cell adhesion, although the absolute level of adhesion at saturation was lower than wild-type ␣v␤5. The apparent affinity of this mutant form of ␣v␤5 for metal ion was 5-50-fold lower than that exhibited by wild-type ␣v␤5 (45-62 versus 1-10 M). This observation is consistent with a role for Asp-220 in coordinating metal ion and is also consistent with the fact that the homologous residue in ␤3 (Asp-217) contributes to metal binding affinity.
Mutations at Asp-119 and Asp-217 of ␤3 Fail to Disrupt the Function of the Inhibitory Ca 2ϩ -binding Site-Integrins contain two classes of ion binding sites, one that promotes ligand binding, called a Ligand Competent site, and another that inhibits ligand binding, called an Inhibitory site (27)(28)(29)(30). The monoclonal antibody AP5 binds to the amino-terminal domain of the ␤3 subunit and reports the occupation of the Inhibitory Ca 2ϩ -binding site (25). As an extension of the present study, we measured the effect of each point mutation within ␤3 on the sensitivity of the binding of the AP5 antibody to Ca 2ϩ . As shown in Fig. 7, the binding of AP5 to wild-type ␣v␤3 and to both ␤3 D119A and ␤3 D217A was blocked by Ca 2ϩ . Thus, Asp-119 and Asp-217 are not part of the Inhibitory ion binding site. DISCUSSION The primary objectives of this study were to examine the possibility that the amino-terminal portion of the integrin ␤  subunit contains a MIDAS-like metal binding site and to assess whether this motif in the integrin ␤ subunit is positioned at the apex of an I domain structure. The simplest step in this analysis involved a comparison of the two structures. The I domains and the amino-terminal portion of the integrin ␤ subunits have similar hydropathy profiles (10) and also exhibit some sequence homology, particularly at residues known to ligand with metal. Both observations suggest the potential for a common fold.
Here, we examined this possibility in more detail using the PHD algorithm, which generates a predicted structure based on the propensity of individual residues within a given local environment to exist in a helix, a sheet, or a disordered loop. Importantly, the algorithm relies on known crystal and NMR structures to predict tertiary structure from the primary sequence. It is reported to have a success rate of approximately 70% (17). The PHD algorithm correctly predicted 10 of 13 structural elements within the I domains of ␣ L and ␣ M , attesting to its ability to identify the major elements within an I domain. In contrast, the algorithm predicted that only 2-3 of the 13 I domain elements are present in the corresponding positions of ␤3 and ␤5. Although the ␤ subunits appear to have some sequence similarity with the I domains, an in-depth analysis using a sophisticated algorithm suggests that the threedimensional structure of the integrin ␤ subunits is likely to be significantly different from that of the I domains. Based on this analysis it does not appear that the integrin ␤ subunits contain an I domain-like region. This does not exclude the possibility that a metal-binding MIDAS motif could be presented in the context of a different backbone structure. Therefore, a series of biochemical studies were performed to further assess metal and ligand binding to the ␣v␤3 and ␣v␤5 integrins. As a first step, the ion specificity of the ligand binding event was tested. Indeed, both integrins have an ion preference that is remarkably similar to that reported for the I domain of ␣ M (26). Although some minor differences exist between ␣v␤3 and ␣v␤5, transition state metal ions like Co 2ϩ and Mn 2ϩ , as well as the cation Mg 2ϩ , support ligand binding. Divalent ions like Ca 2ϩ and Ba 2ϩ were far less effective. Although we know the regulation of ligand binding to ␣v-integrins to be complex and that it can involve regulation by two classes of ion binding sites (27,28,30,32), this simple test shows that the ligand binding event for ␣v-integrins has an ion specificity that is more similar to that of an I domain (26) than to an EF-hand (33).
A more detailed analysis of metal binding involved the mutation of the putative metal coordinating residues within ␤3 and ␤5. This approach identified two distantly spaced aspartic acid residues that greatly influence receptor function. These are Asp-119 and Asp-217 in ␤3 and Asp-121 and Asp-220 in ␤5. By sequence alignment, each aspartate appears to be a homologue of metal ligands in the MIDAS motifs of ␣ L and ␣ M . Substitution of any of these aspartates with alanine reduces the affinity of the ␣v-integrins for soluble ligands by at least 50-fold. Interestingly, mutation of Asp-119 and Asp-217 within ␤3 and Asp-220 in ␤5 did not completely abrogate receptor function because integrins with these mutations could still mediate cell adhesion. Despite the inability of each mutated integrin to bind soluble ligand, the ability of mutants at ␤3 residues Asp-119 and Asp-217 and ␤5 Asp-220 to mediate cell adhesion proves that these aspartates are not absolutely essential for ligand contact. It is important to emphasize that cell adhesion to an immobilized substratum is the summation of multivalent receptor-ligand contacts brought about by integrin clustering. In addition the ligand is immobilized and cannot freely diffuse; therefore, cell adhesion can often be observed even when the affinity between ligand and integrin is too low to measure in soluble ligand binding assays. Thus, another interpretation of this result is that each aspartate contributes to ligand binding affinity. We believe this to be a reasonable inference especially since FACS analysis indicates that each mutant is expressed on the cell surface at a level equivalent to FIG. 5. Cells expressing ␣v␤3 mutants adhere to Fab-9. Human kidney 293 cells expressing mutant ␣v␤3 were allowed to adhere to immobilized Fab-9 at 37°C as a function of Mn 2ϩ concentration. Adhesion was inhibited by inclusion of mAb LM609 or GRGDSP peptide (not shown). Each data point represents the mean of triplicate wells and is expressed as a percentage of maximal adhesion for each cell line. This experiment was repeated five times yielding similar results. the wild-type integrin. However, because the mutant integrins fail to interact with soluble ligand, we are unable to provide a quantitative measure of the difference in ligand binding affinity.
It is also key to assess whether the mutations at putative metal coordination sites alter the affinity of the integrin for metal ion. Unfortunately, the inability to generate milligrams of recombinant integrin, and the relatively low affinity of the integrin for ion, makes a direct measure of this parameter nearly impossible. We were, however, able to assess metal binding affinity indirectly by measuring the apparent affinity of the integrin for metal ion as reported by ligand binding. This was accomplished by measuring cell adhesion across a range of metal ion. From this analysis it is evident that the mutation of Asp-119 and Asp-217 in ␤3, and Asp-220 in ␤5, reduces the apparent affinity of each integrin for metal ion. Mutation of each aspartic acid lowered the apparent affinity for either Mn 2ϩ or Mg 2ϩ by 10 -20-fold. This is the first evidence we are aware of in which an aspartic acid within an integrin ␤ subunit has been shown to influence metal ion affinity. The simplest interpretation of this finding is that each of these aspartic acid residues is part of a metal ion binding site. Without a direct measure of metal binding affinity to each mutant, and without a three-dimensional structure, these aspartates cannot be unequivocally assigned as metal ligands. Yet, because each of the aspartate residues in question aligns well with metal ligands in the MIDAS motif, this finding strongly implies a similarity in the way the two protein modules ligand with ion at the first and third coordination groups.
This study also identifies an important distinction between the metal ligands in the MIDAS and in the ␤ subunits. In the MIDAS motif, the second coordination group is a single threonine that coordinates with bound metal. Based on the alignment presented in Fig. 1, we hypothesized that the analogous threonine in ␤3 was at residue Thr-182 or Thr-183. Interestingly, the ␤5 subunit lacks this threonine, and we originally hypothesized that this difference in sequence at a metal ligand was key to the way in which ␤3 and ␤5 differentially organize on the cell surface in response to metal ions (14). However, the data presented here indicate neither Thr-182 nor Thr-183 within ␤3 makes a significant contribution to metal-dependent ligand binding, nor do Asn-186 and Ser-190 within ␤5. Thus, the distinct organization patterns of ␣v␤3 and ␣v␤5 on the cell surface do not appear to be related to a difference in metal coordination in this region of the ␤ subunit. The inability to identify a metal ligand within the ␤ subunits that is analogous to the second coordination group in the MIDAS is also a clear distinction in the way the two ion binding sites are structured.
Each integrin ␤ subunit contains the sequence DXPE. The aspartate in this motif corresponds to Asp-217 in ␤3 and Asp-220 in ␤5. Here, we present evidence that the glutamic acid in this motif is important for integrin function. Transfection of 293 cells with a cDNA in which ␤3 Glu-220 is mutated to alanine failed to yield a cell line in which the ␣v␤3 heterodimer was expressed. The simplest interpretation of this observation is that ␤3 Glu-220 is required for proper folding or for assembly of a heterodimer with ␣v. In contrast, the mutation of Glu-223 of ␤5 to alanine allows association with ␣v and expression on the cell surface but eliminates ligand binding function. Other reports in the literature also point to the region surrounding the DXPE motif as an important domain that may be part of the ligand binding cleft. Two independent studies showed that synthetic peptides encompassing ␤3 residues 211-222 and 217-231 could block ligand binding to the ␣IIb␤3 integrin (31,34).
In conjunction with the present study, these data suggest a hypothesis regarding the RGD binding site. Collectively the two lines of data indicate that the DXPE sequence may come together with the DXSXS motif to form a metal binding site that is also part of the RGD binding cleft. In this respect, the integrin ␤ subunits appear to contain a site that is similar to the MIDAS motif, where metal-coordinating residues are distantly spaced in the primary sequence but come together in the tertiary structure of the protein to make contact with a metal ion. This similarity must be interpreted in the context of several key differences between the ␤ subunits and the I domains.
Structural algorithms indicate that the ␤ subunits lack similarity to the I domains, so the ␤ subunits are likely to contain a MIDAS within a different backbone. Within the putative metal binding domain, the ␤ subunits contain two disulfide bonds, whereas the I domains do not. The two domains function differently as well. The ␣v␤3 and ␣v␤5 integrins bind the RGD motif and require the association of both subunits for this function.
A final objective of the present study was to classify the putative ion binding site within the ␤ subunits. There are two classes of metal ion binding sites on ␣v␤3 and on ␣5␤1 (25,27,30). These two cation binding sites have opposing effects on ligand binding. One class of site(s), called Ligand Competent sites, must be occupied for ligand to bind (23,27,30). The second class of sites are called Inhibitory sites because, when occupied with Ca 2ϩ , these sites interfere with ligand binding. The Inhibitory sites are allosteric to the ligand binding site and FIG. 7. Calcium inhibits the binding of AP5 to WT ␣v␤3 and to the D119A and D217A mutants. Human 293 cells expressing the noted integrin were incubated with 50 mg/ml FITC-labeled AP5. The concentrations of Ca 2ϩ were varied from 4 to 500 mM. The mean fluorescence intensity of 10,000 cells is presented for each point. Binding of AP5 to mutants T182A and T183A was also inhibited by Ca 2ϩ (not shown).

TABLE II
The effect of point mutations in ␤5 on cell adhesion to vitronectin Cell lines expressing point mutations in ␤5 were allowed to adhere to immobilized vitronectin in the presence of a range of Mg 2ϩ or Mn 2ϩ . The apparent affinities for metal ion were determined as the metal concentration at which half-maximal adhesion occurred. The range of apparent affinities that are listed were compiled from at least three separate experiments. Adhesion of cells containing the ␤5 mutations was always compared with the adhesion of 293 cells expressing wild-type ␣v␤5.  In each report, similar mutations were made in other integrins, yielding similar data. It should be noted, however, that the interpretation of the data are somewhat different. Based on the data presented here, we are reluctant to classify the amino-terminal regions of ␤3 and ␤5 as "I domains." As discussed above, we believe there is sufficient reason to suspect that the ␤3 and ␤5 subunits bind to metal using a MIDAS-like motif but that the backbone of the domain containing the MIDAS is structurally distinct from the known conformation of the I domains.