The Role of α and β Chains in Ligand Recognition by β7 Integrins*

Integrins αEβ7and α4β7 are involved in localization of leukocytes at mucosal sites. Although both αEβ7 and α4β7utilize the β7 chain, they have distinct binding specificities for E-cadherin and mucosal addressin cell adhesion molecule-1 (MAdCAM-1), respectively. We found that mutation of the metal ion-dependent adhesion site (MIDAS) in the αE A-domain (D190A) abolished E-cadherin binding, as did mutation F298A on the A-domain surface near the MIDAS cleft. A docking model of the A-domain with E-cadherin domain 1 indicates that coordination of the αE MIDAS metal ion by E-cadherin Glu31 and a novel projection of Phe298 into a hydrophobic pocket on E-cadherin provide the basis for the interaction. The location of the binding site on the αE A-domain resembles that on other integrins, but its structure appears distinctive and particularly adapted to recognize the tip of E-cadherin, a unique integrin ligand. Additionally, mutation of the β7 MIDAS motif (D140A) abolished αEβ7 binding to E-cadherin and α4β7-mediated adhesion to MAdCAM-1, and α4 chain mutations that abrogated binding of α4β1 to vascular cell adhesion molecule-1 and fibronectin similarly reduced α4β7interaction with MAdCAM-1. Thus, although specificity can be determined by the integrin α or β chain, common structural features of both subunits are required for recognition of dissimilar ligands.

Integrins are heterodimeric glycoproteins consisting of noncovalently associated ␣ and ␤ subunits that play diverse roles in cell-cell and cell-matrix interactions. Integrins of the ␤ 1 , ␤ 2 , and ␤ 7 subfamilies are critical for leukocyte homing (1,2). For example, binding of integrin ␣ 4 ␤ 7 to mucosal addressin cell adhesion molecule-1 (MAdCAM-1) 1 on intestinal endothelial cells is required for the homing of naive lymphocytes to Peyer's patches (2). Integrins are also thought to play a role in microenvironmental homing or retention of lymphocytes at particular sites within a tissue (2). For instance, integrin ␣ E ␤ 7 binds to epithelial cadherin (E-cadherin) and is involved in retention of lymphocytes within the epithelium (3)(4)(5)(6).
Many integrin ligands are members of the immunoglobulin superfamily (such as VCAM-1 and MAdCAM-1) or possess domains that resemble the immunoglobulin fold (e.g. E-cadherin and the type III repeats of fibronectin). A unifying feature of these ligands is an integrin-binding surface containing an exposed acidic amino acid. For example, ␣ 4 ␤ 1 and ␣ 4 ␤ 7 bind to VCAM-1 and MAdCAM-1, respectively, on the face of domain 1 formed by the C, F, and G ␤-strands, centered on an aspartate residue in the loop connecting the C and D strands (7)(8)(9). Domain 1 of E-cadherin contains a glutamate residue at the tip of the BC-loop essential for ␣ E ␤ 7 binding (10,11), and the tenth type III repeat of fibronectin has an acidic residue within an RGD sequence on the FG-loop required for ␣ 5 ␤ 1 binding (12). Other integrin ligands such as collagen I and iC3b that are not known to adopt immunoglobulin-like folds also possess vital acidic residues for ␣ 2 ␤ 1 and ␣ M ␤ 2 binding, respectively (13,14).
Approximately one-third of known ␣ chains (␣ 1 , ␣ 2 , ␣ 10 , ␣ 11 , ␣ D , ␣ E , ␣ L , ␣ M , and ␣ X ) contain an inserted domain of ϳ200 amino acids related to the A-domains of von Willebrand factor. For those integrins tested, the A-domain is critically involved in ligand binding. For instance, isolated recombinant ␣ 2 , ␣ L , ␣ M , and ␣ D A-domains share ligand binding properties with their parent integrins (15)(16)(17)(18). The crystal structures of a number of integrin A-domains (␣ 1 , ␣ 2 , ␣ L , and ␣ M ) reveal a common "Rossmann fold" in which a six-stranded parallel ␤-sheet is flanked by seven ␣-helices (19 -22). Five conserved amino acids (DXSXS, T, and D) come together with two or three water molecules at the top of the domain to chelate a metal ion in what has been termed the metal ion-dependent adhesion site (MIDAS). Mutation of these residues abolishes metal binding to the ␣ M A-domain (23) and severely curtails or eliminates ligand binding activity for all integrins tested (23)(24)(25)(26)(27)(28). In one crystal form of the ␣ M A-domain, a glutamate residue from a neighboring A-domain penetrates the MIDAS cleft and provides a sixth coordination site for the metal ion (19,29). This "open" conformation of the A-domain is proposed to reflect the active ligand-bound conformation. Because most integrin ligands possess a critical exposed acidic residue, bona fide ligand recognition may use a similar mechanism (19). Crystal structures in which no ligand mimetic is bound in the MIDAS cleft have a "closed" conformation that likely reflects the inactive form of the A-domain (29,30). Comparison of the unbound ␣ 2 A-domain with the recently crystallized complex of this A-domain bound to a collagenous triple peptide confirms the model (31). This complex and mutational analyses of the ␣ 2 , ␣ L , and ␣ M integrins have revealed other residues on the upper surface of the A-domains surrounding the MIDAS cleft that contribute to the affinity and specificity of ligand binding (26,29,(32)(33)(34).
This mechanism provides an elegant explanation of ligand binding by A-domain-containing integrins. However, most integrins do not possess an A-domain in the ␣ chain, yet they are also dependent upon acidic residues in their ligands and have a similar requirement for divalent cations. A potential explanation is the proposal that all integrin ␤ chains have a region that adopts an A-domain-like fold (19,35,36). Indeed, all ␤ chains possess a relatively conserved region of ϳ250 amino acids that contains a DXSXS sequence, which, together with conserved downstream oxygenated residues, could form a complete MIDAS site. In fact, mutation of such residues in ␤ 1 , ␤ 2 , ␤ 3 , ␤ 5 , and ␤ 6 integrins abolishes ligand binding activity (35,(37)(38)(39)(40)(41)(42)(43)(44)(45). However, it remains likely that non-A-domain-containing integrin ␣ chains also contribute directly to ligand binding. The seven repeats present in the N-terminal region of all integrin ␣ chains are proposed to fold into a "␤ -propeller" of seven four-stranded ␤-sheets (46), and mutations in the ␣ 4 chain on the upper surface of the predicted second, third, and fourth blades of the propeller, for example, diminish the binding of ␣ 4 ␤ 1 to VCAM-1 (47)(48)(49)). An alternative model proposes that this region of the ␣ chain forms "EF-hand-type" domains, but also postulates a role in ligand binding (50).
Here, we analyze the role of ␣ and ␤ chains in the ligand binding activity of the ␤ 7 integrins ␣ E ␤ 7 and ␣ 4 ␤ 7 . This is important for a number of reasons. First, the ligand for ␣ E ␤ 7 , E-cadherin, is a unique integrin counter-receptor. Although a critical glutamate residue (Glu 31 ) required for integrin binding has been identified, its structural context is unlike that of other ligands of integrins with ␣ chain A-domains (10,11). Such integrins bind to counter-receptors that contain critical acidic residues presented on relatively flat surfaces (e.g. ␤ 2 integrin ligands ICAM-1 and -2) or on collagenous structures (9,31). However, Glu 31 in E-cadherin is on a highly exposed loop more akin to similar loops containing critical acidic residues in VCAM-1, MAdCAM-1, and fibronectin that bind to integrins lacking an ␣ chain A-domain (11). Also, ␣ E has an "X"-domain of 55 amino acids immediately N-terminal to the A-domain that has 18 charged residues within a stretch of 21 amino acids and contains a post-translational cleavage site (see Fig. 1A) (51). For these reasons, there exists doubt as to the role of the A-domain in E-cadherin binding. Therefore, it is of great interest to determine if the ␣ E A-domain is involved and what role the X-domain plays. Second, the ␤ 7 chain must be important in defining binding specificity since ␣ 4 ␤ 7 binds to MAdCAM-1, whereas ␣ 4 ␤ 1 does not (52)(53)(54); yet ␣ E ␤ 7 and ␣ 4 ␤ 7 both utilize the ␤ 7 chain, but they have non-overlapping ligand binding specificities for E-cadherin and MAdCAM-1, respectively (54). 2 Also, in a number of animal models of colitis and graft-versushost disease, the administration of antibodies to either ␣ 4 ␤ 7 or ␣ E ␤ 7 ameliorates the development of intestinal inflammation (55)(56)(57)(58), and ␣ E ␤ 7 potentially has a role in rejection of epithelial tissue within allografts (59). Therefore, there is considerable interest in the structural basis for ligand recognition by these integrins because of their important role in leukocyte trafficking and to aid the design of therapeutic small molecule inhibitors.
Adhesion Assays-Adhesion assays were carried out as described previously (5). Briefly, 96-well microtiter plates were coated with goat anti-human IgG polyclonal antibody and blocked with 1% BSA in TBS. Wells were then coated with human IgG1 Fc-containing proteins in 50 l of TBS and 1 mM CaCl 2 and blocked with 1% BSA, 2% human serum, and 1 mM CaCl 2 in Hepes-buffered saline. Alternatively, transfected CHO cells were grown to near confluence in 96-well tissue culture plates. Transfected L1-2 cells, CHO cells (released with 0.5 mM EDTA in phosphate-buffered saline), or K562 cells were labeled with calcein/AM (Molecular Probes, Inc., Eugene, OR), and adhesion assays were carried out at 37°C in Hepes-buffered saline with 0.1% BSA and 1 mM each MnCl 2 , MgCl 2 , and CaCl 2 for 15 min (K562 cells) or 1 mM each MgCl 2 and CaCl 2 (15 min for CHO cells and 60 min for L1-2 cells). To correct for variations in integrin expression level between clones, we established that the percent adhesion seen with a panel of K562 clones was linearly dependent on the mean fluorescence intensity (MFI) of wild-type ␣ E ␤ 7 on the surface determined by flow cytometry with anti-␤ 7 mAb Fib504 (correlation coefficient ϭ 0.91). The adhesion of each mutant clone was determined as a percentage of that seen for a clone expressing wild-type ␣ E ␤ 7 at a comparable level (Ϯ50% wild-type MFI). Then, % wild-type adhesion ϭ 100 ϫ (% adhesion of test cells/MFI on test cells)/(% adhesion of K562- Biotinylation of E-cadherin-Fc-Purified E-cadherin-Fc or human IgG1 at an equal concentration was dialyzed into 10 mM sodium borate with 0.5 mM CaCl 2 (pH 8.4). After incubation with a 33-fold molar excess of aminohexanoylbiotin-N-hydroxysuccinimide ester (Zymed Laboratories Inc., South San Francisco, CA) for 50 min at 25°C, the biotinylated protein was dialyzed into TBS with 1 mM CaCl 2 (pH 7.4).

Model of the ␣ E Integrin A-domain-
To facilitate selection of residues for mutagenesis and to aid interpretation of the results, a three-dimensional model of the ␣ E chain A-domain was generated. A FASTA screen (75) of the Swiss Protein Database with the human ␣ E A-domain protein sequence revealed that it was most closely related to the A-domain of human ␣ M (CD11b), with 38% identity over 192 amino acids. The alignment of the ␣ E A-domain sequence with that of ␣ M revealed a single insertion of two amino acids in ␣ E compared with ␣ M (Fig. 1A). Substitution of human ␣ E residues into the human ␣ M crystal structure in the open conformation (1IDO (19)) produced a model of the ␣ E A-domain (Fig. 1B). No major changes in the backbone structure are necessary to accommodate the ␣ E protein sequence, and the conserved MIDAS residues Asp 190 , Ser 192 , Ser 194 , Thr 261 , and Asp 294 form a potential metal ion coordination site in a cleft at the top of the domain, as in known integrin A-domain structures. A notable difference from the ␣ M (and ␣ L ) structure in the model is the extended ␣ 1 -␤ B loop at the base of the A-domain due to the insertion of two amino acids mentioned above. The length of this loop is thus more similar to that in the ␣ 1 and ␣ 2 A-domains, but is unique in that a disulfide bond that may rigidify the region is predicted between the ␣ 1 -␤ B loop and the start of the ␤ B -strand ( Fig. 1, A and B).
Generation of K562 Cells Expressing Mutant ␣ E ␤ 7 Integrins-To localize sites on the human ␣ E ␤ 7 integrin that are involved in binding to E-cadherin, we generated a series of site-directed mutants of both ␣ (Fig. 1A) and ␤ chains. Mutations were made in the DXSXS MIDAS motifs of both ␣ E (D190A) and ␤ 7 (D140A) chains and in a number of residues predicted by the model to be solvent-exposed at the top of the ␣ E A-domain surrounding the MIDAS site (G193A, D199A, G230A/V231A, F298A, and E325A). Further mutations were made in prominent surface-exposed residues in other regions of the ␣ E A-domain based upon the model (R202A/D205A, D240A, P311H/E345A/T346A, and Y354W). The double-basic posttranslational cleavage site within the X-domain was removed by mutation (R159S/R160S); a deletion of a single residue within the highly charged region of the X-domain was generated (⌬Glu 176 ); and a final mutant lacking the entire charged region of the X-domain was produced (⌬Glu 163 -Glu 180 ).
After transfection of K562 cells with these mutant constructs together with the appropriate wild-type partner chain cDNA, we obtained cells expressing cell-surface ␣ E ␤ 7 as determined by flow cytometry. K562 cell clones expressing the mutant forms of ␣ E ␤ 7 were obtained by autocloning (see "Experimental Procedures"), and a number of clones expressing each mutant at similar levels as determined by surface staining with anti-␤ 7 integrin mAb Fib504 were selected for further analysis.
To confirm that the mutations introduced into ␣ E ␤ 7 did not prevent heterodimer formation, cells expressing each mutant were surface-labeled with biotin and immunoprecipitated with anti-␤ 7 mAb Fib504 ( Fig. 2A). The ratio of co-immunoprecipitated ␣ E chain to ␤ 7 chain as determined by densitometry was within 2-fold of that determined for wild-type ␣ E ␤ 7 in all cases, with the exception of ␣ E (R159S/R160S)␤ 7 , confirming that these mutations do not have a major impact on heterodimer formation. The mutation R159S/R160S appeared to reduce the amount of ␣ E chain co-immunoprecipitated with the anti-␤ 7 antibody. However, this mutation removes two arginine residues from the ␣ E chain that are possible targets of biotinylation and might affect the efficiency of labeling. Therefore, immunoprecipitation was performed after 125 I surface labeling of K562-␣ E (R159S/R160S)␤ 7 cells, and this revealed that the normal proportion of ␣ E to ␤ 7 chain was present. Moreover, the R159S/ R160S mutation, as expected, prevented the majority of the ␣ E chain from being cleaved, resulting in the larger size of the ␣ E chain (ϳ175 kDa) compared with the wild type (150 kDa) under reducing conditions (Fig. 2B). Under nonreducing conditions, both the wild type and the R159S/R160S mutant have an ␣ E chain of ϳ180 kDa (data not shown). A low level of residual cleavage of the mutant chain was seen that may be due to inefficient cleavage of the chain at a single arginine residue located two residues N-terminal of the bona fide dibasic site (see Fig. 1A). Immunoprecipitation from K562-␣ E (⌬Glu 163 -Glu 180 )␤ 7 cells demonstrated the presence of a smaller ␣ E chain, as expected, resulting from loss of the charged region (data not shown). However, this mutant, which displayed evidence of proteolytic degradation, was expressed at a low level, and less ␣ E chain was precipitated with an anti-␤ 7 chain mAb than for the other mutants. Another mutant with a deletion of the entire X-domain was not expressed at the cell surface and therefore was excluded from the analysis. This suggests that large deletions within the X-domain have some impact on the ability of the ␣ E chain to associate with ␤ 7 and to be transported to the cell surface.
Mapping of anti-␣ E ␤ 7 Monoclonal Antibody Epitopes Defines a Likely Binding Site for E-cadherin-To probe the overall conformation of the mutant ␣ E ␤ 7 integrins on the cell surface and to map antibody-binding epitopes, the reactivity of each mutant with a panel of anti-␣ E ␤ 7 mAbs was determined by flow cytometric analysis and compared with that of anti-␤ 7 mAb Fib504 as a reference (see "Experimental Procedures"). mAbs Fib21, Fib30, and Fib504, which recognize three distinct epitopes on the ␤ 7 chain (65), all bound equally well to all the mutant ␣ E ␤ 7 integrins (Fig. 3A). Thus, none of the mutations, including that of the ␤ 7 chain MIDAS motif (D140A), had any detectable influence on the conformation of the ␤ 7 chain. This analysis also supported the use of Fib504 as a reference mAb for subsequent analysis of anti-␣ E mAbs. All the mutant ␣ E ␤ 7 integrins were recognized as well as the wild type by multiple mAbs that recognize different epitopes of ␣ E ␤ 7 and that do not recognize denatured protein in immunoblotting (Refs. 63 and 71 and this study), suggesting that the mutant ␣ E ␤ 7 integrins maintain their basic structure. Indeed, no significant difference was seen in the recognition by mAbs LF61 and ABB1.3 of any of the mutants (Fig. 3B).
Interestingly, removal of the cleavage site within the unique ␣ E X-domain (␣ E (R159S/R160S)␤ 7 ) reduced the binding of mAb BerACT-8 by 70%, whereas all the other mAbs tested bound normally (Fig. 3B). Immunoprecipitation of ␣ E (R159S/ R160S)␤ 7 with BerACT-8 suggests that the residual binding of this mAb is due to recognition of the small proportion of ␣ E (R159S/R160S)␤ 7 with a cleaved ␣ E chain (data not shown). These findings suggest that the loss of BerACT-8 binding to the R159S/R160S mutant is not directly due to the loss of the two arginine residues, but rather that BerACT-8 recognizes an epitope that is fully exposed only when the X-domain is cleaved. However, the precise location of the BerACT-8 epitope was not determined in this study.
Mutation of the MIDAS Motif of the ␣ E or ␤ 7 Chain Abolishes Adhesion to E-cadherin-The functional activity of the ␣ E ␤ 7 mutants was tested by measuring the adhesion of the K562 transfectants to E-cadherin-Fc fusion protein immobilized on microtiter plate wells as described previously (5,11). Thirty to forty percent of K562-␣ E ␤ 7 WT cells adhered to E-cadherin-Fc at saturating concentrations, whereas K562 control cells did not adhere (data not shown). Strikingly, mutation of the DX-SXS sequence to AXSXS within the MIDAS motif of either the ␣ E (D190A) or ␤ 7 (D140A) chain completely abolished detectable adhesion of transfected K562 cells to E-cadherin-Fc-coated wells (Fig. 4A). Thus, the coordination of metal ions by the MIDAS residues of the ␣ E A-domain and the proposed A-like domain of ␤ 7 are likely to be critical for ␣ E ␤ 7 -mediated adhesion to E-cadherin.

Phe 298 at the Top of the ␣ E A-domain Is Critical for
Adhesion to E-cadherin-K562 cells expressing ␣ E ␤ 7 containing mutations G193A, D199A, R202A/D205A, G230A/V231A, D240A, P311H/E345A/T346A, E325A, and Y354W in the ␣ E A-domain adhered to E-cadherin-Fc similarly to K562-␣ E ␤ 7 WT transfectants (Fig. 4A). Assays conducted at lower coating concentrations of E-cadherin-Fc or without manganese also did not reveal any differences from the wild type (data not shown). In contrast, K562-␣ E (F298A)␤ 7 cells did not adhere to E-cadherin-Fc at any concentration tested (p Ͻ 0.01 versus the wild type by Dunnett's multiple comparison test). Phe 298 is predicted to be prominently exposed at the top of the ␣ E A-domain close to the MIDAS metal ion coordination site and thus is likely to be involved directly in binding to E-cadherin. The lack of structural data for the potential A-like domain of integrin ␤ chains precluded a reliable similar analysis of the ␤ 7 chain.
Neither removal of the cleavage site from the X-domain of the ␣ E chain (R159S/R160S) nor deletion of a single residue within the charged region (⌬Glu 176 ) altered adhesion to Ecadherin-Fc (Fig. 4A). The results suggest that the fine structure of the charged region of the X-domain is not crucial for binding to E-cadherin and that post-translational cleavage in this region is not a prerequisite for adhesion to E-cadherin. Furthermore, an antiserum to the charged region of the X-domain that recognizes ␣ E ␤ 7 on the cell surface by flow cytometry does not inhibit the adhesion of K562-␣ E ␤ 7 WT transfectants to E-cadherin-Fc (data not shown), indicating that this region does not participate directly in the binding of ␣ E ␤ 7 to E-cadherin.

Direct Binding of Soluble E-cadherin-Fc to K562-␣ E ␤ 7 Transfectants Confirms and Extends Definition of the E-cadherin-
binding Site-Adhesion assays are not simply a measure of direct receptor to ligand binding because stable adherence of cells to immobilized ligands may require post-receptor events including modulation of the cytoskeleton. It is possible that mutation of ␣ E ␤ 7 might affect signaling through the integrin and such downstream events rather than alter the direct binding to E-cadherin. To rule out this possibility and to provide a quantitative assay for direct binding of ␣ E ␤ 7 to E-cadherin, we made use of the ability of soluble E-cadherin-Fc to bind directly to ␣ E ␤ 7 (5). The binding of biotinylated E-cadherin-Fc to the surface of K562-␣ E ␤ 7 transfectants could be detected by flow cytometry using a neutravidin-conjugated Alexa 488 secondary reagent (Fig. 4B). No such binding was seen to K562 control cells. Saturation of binding was achieved between 50 and 100 g/ml E-cadherin-Fc (data not shown). The binding of 100 g/ml biotinylated E-cadherin-Fc to K562 cells transfected with each mutant form of ␣ E ␤ 7 was assessed. In each case, the background binding of biotinylated human IgG1 was subtracted, and the results are expressed as the percentage of the binding seen to wild-type ␣ E ␤ 7 after correcting for the level of surface expression determined using anti-␤ 7 chain mAb Fib504.
The results (Fig. 4C) confirm those obtained in adhesion assays. Mutation of the MIDAS motif of either the ␣ E or ␤ 7 chain completely abrogated the ability of E-cadherin-Fc to bind ␣ E ␤ 7 (p Ͻ 0.01 versus the wild type by Dunnett's multiple comparison test). Also, the critical importance of Phe 298 for binding to E-cadherin was reinforced. No direct binding of E-cadherin-Fc to ␣ E ␤ 7 could be detected if this amino acid was mutated to alanine (p Ͻ 0.01). In addition, the direct binding assay highlighted the previously undetected role of Glu 325 in binding to E-cadherin. Mutation of this amino acid to alanine unexpectedly increased the binding of E-cadherin-Fc to ␣ E ␤ 7 2-fold over the wild type (p Ͻ 0.01). This residue is exposed on the surface of the ␣ E A-domain near the MIDAS site, and it forms a crucial part of the epitope of the adhesion-blocking mAb ␣E7-2 (Fig. 3B). As in the adhesion assays, mutation of other residues in the A-domain, mutation of the X-domain cleavage site (R159S/R160S), and removal of a single amino acid within the charged region (⌬Glu 176 ) did not alter E-cadherin-Fc binding. Mutation of the MIDAS Motif of the ␤ 7 Chain Abolishes ␣ 4 ␤ 7 -mediated Adhesion to MAdCAM-1-A role for the MIDAS motif within the potential A-like domain of the ␤ 7 chain has been proposed for ␣ 4 ␤ 7 -mediated adhesion to MAdCAM-1 based on ␤ 1 /␤ 7 chimera analysis and mapping of the adhesionblocking antibody ACT-1 epitope to this region of the ␤ 7 chain (60). Transfection of both human ␣ 4 and ␤ 7 chains into K562 cells conferred the ability to adhere to MAdCAM-1-Ig fusion protein, but not to human IgG1, as described by others (60). Adhesion to MAdCAM-1-Ig could be blocked by anti-human ␣ 4 ␤ 7 mAb ACT-1, and K562 control cells were unable to adhere to MAdCAM-1-Ig (data not shown). After transfection of K562 cells with ␣ 4 and the ␤ 7 chain containing the D140A mutation within the MIDAS motif, we obtained clones that expressed the mutant ␣ 4 ␤ 7 (D140A) as assessed by flow cytometry. A comparison of the staining obtained with mAbs to at least two different epitopes on the ␣ 4 chain (B5G10, B5E2, and HP1/2) (76), three different epitopes on the ␤ 7 chain (Fib21, Fib30, and Fib504) (65), and an antibody that recognizes cell-surface ␤ 7 chain only in the context of ␣ 4 ␤ 7 (ACT-1) (60,77) showed that the binding of these mAbs was unaffected and confirmed that the D140A mutation did not grossly affect the structure of ␣ 4 ␤ 7 (Fig. 5A). In addition, immunoprecipitation of wild-type ␣ 4 ␤ 7 and ␣ 4 ␤ 7 (D140A) with either mAb ACT-1 or Fib504 demonstrated that heterodimerization was not affected (Fig. 5B) (data not shown). Despite this, K562-␣ 4 ␤ 7 (D140A) cells were unable to adhere to MAdCAM-1-Ig, even when the mutant integrin was expressed at higher levels than wild-type ␣ 4 ␤ 7 (Fig. 5C). Thus, an intact MIDAS motif within the A-like domain of the ␤ 7 chain is required for the adhesion of ␣ 4 ␤ 7 to MAdCAM-1 as well as for the adhesion of ␣ E ␤ 7 to E-cadherin.
Mutations of the ␣ 4 Chain That Reduce ␣ 4 ␤ 1 -mediated Adhesion Also Reduce ␣ 4 ␤ 7 -mediated Adhesion to MAdCAM-1-Ig-A number of residues in the second, third, and fourth repeats of the ␣ 4 chain have been implicated in the binding of ␣ 4 ␤ 1 to VCAM-1 and fibronectin (47)(48)(49). To determine if the same residues are involved in the binding of ␣ 4 ␤ 7 to MAdCAM-1, we transfected CHO cells with previously described mutant ␣ 4 chains together with the wild-type ␤ 7 chain. CHO cells were used in this case because K562 cells were found to express endogenous human ␣ 4 chain at significant levels upon transfection with ␤ 7 (data not shown). We generated CHO cells expressing ␣ 4 ␤ 7 with single amino acid substitutions in the ␣ 4 chain (Y187A, G190A, and Y202A) (48) or in which predicted loops in the second and third repeats of the ␣ 4 chain were replaced with corresponding regions of the ␣ 5 chain (R2, residues 112-131; and R3a, residues 151-164) (49). All the mutants were expressed at levels greater than that of wild-type background adhesion to the same concentration of human IgG1 was subtracted. Each bar represents the mean Ϯ S.E. pooled from four to nine determinations in triplicate from two to four independent experiments involving two to four different K562 clones. The results were combined as described under "Experimental Procedures." The lack of a bar indicates that no adhesion above that to IgG1 could be detected. *, p Ͻ 0.01 versus the wild type by Dunnett's multiple comparison test. B, E-cadherin-Fc directly binds to K562-␣ E ␤ 7 WT cells, but not to K562 control cells. The binding of 50 g/ml biotinylated human E-cadherin-Fc or human IgG1 to transfected K562 cells was determined by flow cytometry in the presence of 1 mM each MnCl 2 , MgCl 2 , and CaCl 2 , using a neutravidin-conjugated Alexa 488 secondary reagent. C, the direct binding of 100 g/ml E-cadherin-Fc to K562-␣ E ␤ 7 mutants was analyzed as described for B. Binding is expressed as the percentage of binding obtained with anti-␤ 7 mAb Fib504 as described under "Experimental Procedures." Each bar represents the mean Ϯ S.D. determined from two to four different K562 clones analyzed in two to three separate experiments. Asterisks indicate that no binding above that to IgG1 could be detected (p Ͻ 0.01 versus the wild type by Dunnett's multiple comparison test); double asterisks indicate a significant increase in binding over the wild type (p Ͻ 0.01). ⌬E176, ⌬Glu 176

DISCUSSION
Despite the distinctive nature of E-cadherin as the only integrin ligand known that is a cadherin and the presence of unique structural features within the ␣ E chain that could contribute to ligand binding, the E-cadherin-binding site includes the A-domain of the ␣ E chain, and its location appears similar to that defined for other integrin-ligand interactions. Mutation of the conserved MIDAS motif within the ␣ E A-domain (D190A) abolishes E-cadherin binding. This is consistent with the requirement for magnesium or manganese ions (5) and an exposed acidic residue within E-cadherin for ␣ E ␤ 7 binding (10, 11) and strongly suggests a direct role for a metal ion coordinated by the MIDAS motif in the interaction. This result also shows not alter the binding of anti-␣ 4 ␤ 7 (ACT-1), anti-␣ 4 (B5G10, B5E2, and HP1/2), and anti-␤ 7 (Fib21, Fib30, and Fib504) mAbs to K562-␣ 4 ␤ 7 transfectants. Antibody binding was determined as described under "Experimental Procedures." None of the antibodies bound to K562 control cells. Each bar represents the mean Ϯ S.D. determined from four different K562 clones analyzed in three separate experiments. B, the ␤ 7 mutation D140A does not alter ␣ 4 ␤ 7 heterodimer formation. K562 cells transfected with wild-type ␣ 4 ␤ 7 or ␣ 4 ␤ 7 (D140A) or K562 control cells were surface-labeled with biotin. Proteins immunoprecipitated with anti-␣ 4 ␤ 7 mAb ACT-1 were detected as described in the legend to Fig. 2. C, the mutation D140A in the ␤ 7 chain MIDAS motif of ␣ 4 ␤ 7 abolishes detectable adhesion to MAdCAM-1-Ig. The adhesion of K562 transfectants to human MAdCAM-1-Ig or IgG1 was determined as described under "Experimental Procedures." The expression level of ␣ 4 ␤ 7 (D140A) on the cell surface by flow cytometry was higher than that of the wild type. The data are the mean Ϯ S.D. (n ϭ 3) and are representative of three separate experiments with two different clones. that, contrary to previous concepts (9), A-domain-containing integrins do not require the critical acidic residue of the ligand to be on a flat surface since Glu 31 of E-cadherin is on the highly exposed BC-loop at the top of domain 1 in a context most similar to that of the RGD sequence in fibronectin, rather than on a fairly level surface at the end of the C strand as in ICAM-1 and -2.
In contrast, the mutation R159S/R160S within the extra X-domain of the ␣ E chain prevents the normal post-translational cleavage of the ␣ E chain, but does not alter the binding of E-cadherin to ␣ E ␤ 7 . Deletion of a single amino acid within the charged region of the X-domain also has no detectable effect on E-cadherin binding. Since an anti-X-domain antiserum also fails to influence adhesion to E-cadherin, we speculate that the ␣ E X-domain does not contribute directly to the E-cadherinbinding interface. The X-domain may be important in optimizing the orientation of the A-domain for ligand binding since it is located at the junction between the second blade of the predicted ␤-propeller and the A-domain of ␣ E . Such a role would be consistent with the presence of an X-domain in murine and rat ␣ E chains, but its low sequence homology and poor length conservation compared with human ␣ E (78,79). Alternatively, the X-domain might be involved in interactions with yet unidentified ␣ E ␤ 7 ligands.
Residues predicted from our ␣ E A-domain model to reside in loops close to the MIDAS cleft also were implicated in Ecadherin binding and suggest a wider interaction surface that is involved in ligand recognition at the top of the A-domain. The binding of five blocking monoclonal antibodies to ␣ E ␤ 7 mapped to a cluster of residues (Gly 193 , Gly 230 /Val 231 , Phe 298 , and Glu 325 ) encircling the MIDAS on the surface of the A-domain (Fig. 7A). The mutation F298A abolished detectable binding of E-cadherin to ␣ E ␤ 7 and is located in the ␤ D -␣ 5 loop. This hydrophobic residue is predicted to be prominently exposed on the top surface of the A-domain (Fig. 7B). The mutation E325A, in the ␤ E -␣ 6 loop also at the top of the A-domain (Fig. 7B), FIG. 6. Mutations of the ␣ 4 chain that reduce ␣ 4 ␤ 1 -mediated adhesion also reduce ␣ 4 ␤ 7 -mediated adhesion to MAdCAM-1-Ig. A, the mutations of ␣ 4 do not prevent ␣ 4 ␤ 7 heterodimer formation. Transfected CHO cells were surface-labeled with biotin, and proteins were immunoprecipitated with anti-␤ 7 mAb Fib504 as described in the legend to Fig. 2. To correct for different levels of integrin expression (see below) and to ease comparison between clones, varying ECL exposures of the different mutants, immunoprecipitated in the same experiment, are shown. Note that changes in the number and location of lysine and arginine residues, the major targets for biotinylation, may alter the efficiency of ␣ 4 chain labeling for the R2 and R3a mutants. B, the adhesion of CHO transfectants to human MAdCAM-1-Ig at 0.1 g/well was determined as described under "Experimental Procedures." In all cases, the expression level of the mutant integrin on the cell surface by flow cytometry was higher than that of the wild type (MFI with anti-␣ 4 ␤ 7 mAb ACT-1: 43 for the wild type, 285 for Y187A, 62 for G190A, 72 for Y202A, 253 for R2, 377 for R3a, and 3 for the control). The data are expressed as the mean percentage Ϯ S.D. (n ϭ 3) of adhesion obtained compared with CHO-␣ 4 ␤ 7 WT cells and are representative at least two separate experiments involving each clone. The mean percentage adhesion of all CHO transfectants to human IgG1 was ϳ1%, and that of CHO-␣ 4 ␤ 7 WT cells to MAdCAM-1-Ig was 39%. C, CHO cells expressing wild-type or mutant ␣ 4 integrins were cultured to near confluency in 96-well tissue culture plates, and fluorescently labeled L1-2 cells expressing MAdCAM-1 (4 ϫ 10 5 cells/well) were added and incubated for 1 h at 37°C. Data are shown as percentage Ϯ S.D. (n ϭ 3) of added cells that were adherent. Parent L1-2 cells showed only background level binding. *, significant decrease in adhesion versus wild-type ␣ 4 ␤ 7 by Dunnett's multiple comparison test (p Ͻ 0.01).
increased the direct binding of E-cadherin to ␣ E ␤ 7 2-fold and thus also may be close to the cadherin-binding surface.
Using this information, we docked a model of human Ecadherin (11) onto the ␣ E A-domain. Importantly, we utilized a model of the ␣ E A-domain based on the open form of the ␣ M A-domain, as this is likely to be the active conformation (31). The critical Glu 31 residue in the BC-loop of E-cadherin domain 1 was maintained at a distance suitable for interaction with the magnesium ion chelated by the A-domain MIDAS motif. The E-cadherin molecule was then rotated relative to the A-domain structure to optimize the interaction and to minimize steric interference. Strikingly, the best orientation of the two structures places Phe 298 of the ␣ E A-domain into a hydrophobic pocket formed between the BC-and FG-loops of E-cadherin domain 1 (Fig. 8A). This interaction and the coordination of the MIDAS metal ion by Glu 31 in E-cadherin provide the most important contributions to the contact between the two molecules. This is consistent with the finding that the mutation E31A in E-cadherin and the F298A and MIDAS mutation D190A in ␣ E ␤ 7 all abolish detectable binding between the two proteins. Our previous study identified further residues in Ecadherin that contribute to ␣ E ␤ 7 binding (11). Interestingly, mutation of Asn 27 , Lys 30 , and Glu 89 in the BC-and FG-loops of E-cadherin reduces adhesion of K562-␣ E ␤ 7 cells to E-cadherin-Fc. These amino acids are all found at the proposed interface between E-cadherin and the A-domain (Fig. 8B). Asn 27 and Glu 89 could potentially form a hydrogen bond and a salt bridge, respectively, with Arg 331 in the A-domain, and Lys 30 in Ecadherin could form another salt bridge with Asp 196 in the A-domain. Arg 331 was not mutated in this study, but the model predicts that it may be an important component of the Ecadherin-binding site. This proposal is consistent with the marked reduction (Ͼ90%) that mutation of Glu 89 causes in K562-␣ E ␤ 7 adhesion to E-cadherin (11). We also examined the interaction of E-cadherin with an ␣ E A-domain model in the closed conformation that is likely to represent the inactive form (30,31). E-cadherin residues Asn 27 , Lys 30 , and Glu 89 all interact with a region of the A-domain that undergoes a marked conformational change upon transition between the two forms (19, 29 -31) (Fig. 8B). Indeed, in the active form of the A-domain, Asp 196 and Arg 331 both move into more favorable positions for interaction with E-cadherin. Glu 325 is also in this region of the A-domain, but is not predicted to make direct contacts with E-cadherin. This raises the possibility that the E325A mutation may increase binding to E-cadherin because it favors adoption of the active conformation.
The potential projection of Phe 298 into a distinct pocket in E-cadherin domain 1 suggests a novel mode of integrin-ligand interaction that contributes to the specificity of ␣ E for E-cadherin. The equivalent of Phe 298 in ␣ L , Thr 243 , is involved in binding to ICAM-1 and -2, but forms part of a relatively flat binding surface (20,32,33), and His 258 of ␣ 2 is involved in collagen I binding, but forms a hydrogen bond with the main chain of collagen (31,34). Also, the ␣ 1 and ␣ 2 integrin A-domains contain an extra four-amino acid ␣-C helix in the ␤ E -␣ 6 loop that markedly alters the topology of the A-domain surface and regulates access to the collagen-binding site (21,22,31,34,80). Although hydrophobic residues are involved in all integrin A-domain interactions examined, the penetration of a prominent aromatic residue into a hydrophobic pocket of the ligand is not predicted for ␣ L binding to ICAM-1 or -2 (20,33,81) and is not involved in ␣ 2 A-domain binding to collagen (31). Interestingly, the amino acids implicated in binding of the ␣ E A-domain to E-cadherin are similar to residues on the closely related A-domain of ␣ M that are involved in binding iC3b (see Fig. 1A). The residue equivalent to Phe 298 in the ␣ M A-domain is also phenylalanine, and the mutation F246K abolishes ␣ M binding to iC3b (29), whereas the F246A mutation reduces binding by 50% (26). Strikingly, the mutation D273K, the residue in the ␣ M A-domain corresponding to Glu 325 , also leads to a doubling of ligand binding activity, in this case for iC3b (29). The results FIG. 7. Residues at the E-cadherin-binding site of the ␣ E A-domain. Shown are space-filling models of the ␣ E A-domain in the open conformation viewed from the top. The MIDAS residues predicted to coordinate magnesium are colored pink. In A, residues colored red are those that significantly reduce the binding of the mAbs shown in parentheses. Mutation of residues in blue had no detectable effect on mAb binding. Since the five mAb indicated all block E-cadherin binding to ␣ E ␤ 7 , this provides a preliminary mapping of the E-cadherin-binding site to the upper surface of the ␣ E A-domain surrounding the MIDAS cleft. In B, mutation of residues colored red abolishes detectable binding, and the mutation E325A in green increases the binding of Ecadherin to ␣ E ␤ 7 , confirming the location of the E-cadherin-binding site on the ␣ E A-domain. Mutation of residues in blue did not have a significant influence on E-cadherin binding.
suggest that the ␣ E and ␣ M A-domains share common features for recognition, despite the dissimilarity of their ligands.
We found residues that could be mutated in the ␣ E A-domain with no detectable influence on E-cadherin binding, but that are important for ligand binding by ␣ M and ␣ L A-domains. The mutations G193A, D199A, and G230A/V231A in ␣ E did not cause a significant change in E-cadherin binding, whereas the mutations G143M, D149K, and E178A/E179A in ␣ M all reduced iC3b binding by Ͼ90% (29), and M140Q, E146A, and T175A in ␣ L all reduced ICAM-1 or -2 binding (32,33). It is possible that in some instances, the more dramatic amino acid substitutions made in the studies of ␣ M ␤ 2 and ␣ L ␤ 2 are responsible for the detection of more important residues. For example, the M140Q mutation reduced ICAM-1 binding by 65% (32), but M140A had no effect on ICAM-1 or -2 binding in another study (33). Alternatively, it may be that E-cadherin has a smaller footprint on the ␣ E A-domain than iC3b on the ␣ M A-domain or ICAM on the ␣ L ␤ 2 A-domain. Indeed, the docking model of E-cadherin with the ␣ E A-domain does suggest limited contact between the molecules. This is due in part to the position of the critical Glu 31 residue of E-cadherin at the tip of the first domain (Fig. 8A, inset), in marked contrast to the flat surfaces of ICAM-1 and -2 that bind the ␣ L A-domain. It also possible that other residues on the surface of the ␣ E A-domain that we did not mutate in this study contribute to E-cadherin binding. However, these results implicate common ligand-binding loops on the surface of all A-domains that surround the MIDAS cleft, but highlight integrin-specific differences that are critical to defining ligand binding specificity.
We also found that mutation of residues in the ␣ 4 chain of  30 , and Glu 89 , which were suggested by previous mutagenesis (11) to contact ␣ E ␤ 7 , are indicated. The human ␣ E A-domain models are described under "Results," and the model of human E-cadherin was described previously (11). ␣ 4 ␤ 7 can abrogate binding to MAdCAM-1. The mutations G190A and Y187A substantially reduced adhesion of ␣ 4 ␤ 7expressing cells to MAdCAM-1, whereas the mutation Y202A did not. In addition, the replacement of ␣ 4 residues 112-131 with the corresponding residues of ␣ 5 abolished MAdCAM-1 recognition, whereas a similar substitution of residues 151-164 did not. The residues that influence adhesion lie on the upper surface of the second and third blades of the predicted ␤-propeller (46,49), consistent with a role in ligand binding. Strikingly, this pattern of abrogation of ␣ 4 ␤ 7 -mediated adhesion to MAdCAM-1 is identical to that previously reported for ␣ 4 ␤ 1mediated adhesion to VCAM-1 and the CS-1 fragment of fibronectin (48,49). Although it remains possible that there are residues on ␣ 4 that interact with VCAM-1 and fibronectin but not with MAdCAM-1 or vice versa, we clearly implicate a similar region of the ␣ 4 molecule in ligand binding for both ␣ 4 ␤ 1 and ␣ 4 ␤ 7 .
Since ␣ E ␤ 7 and ␣ 4 ␤ 7 have distinct, non-overlapping ligand binding specificities for E-cadherin and MAdCAM-1, respectively (54), 2 clearly the ␣ chain must determine this difference in ligand binding specificity. Indeed, we have shown that mutations in both the ␣ E and ␣ 4 chains can abrogate the binding of these integrins to their ligands. Similarly, since ␣ 4 ␤ 7 binds to MAdCAM-1, whereas ␣ 4 ␤ 1 binds to VCAM-1 but not MAdCAM-1 (52)(53)(54), in this case, the ␤ chain must be important in defining specificity. As expected, mutation of the MIDAS motif of the ␤ 7 chain (this study) or of the ␤ 1 chain (47) abolishes ligand binding in both cases. Despite this, however, we have demonstrated that the ␤ 7 chain MIDAS is critically involved in ligand binding for both ␣ E ␤ 7 and ␣ 4 ␤ 7 and that the ␣ 4 chain is critical in ligand binding for both ␣ 4 ␤ 7 and ␣ 4 ␤ 1 . In each case, a similar binding site is implicated on the shared chain even when the integrin is engaged in interactions with different counter-receptors. These results indicate that changes in the combinations of ␣ and ␤ chains that make up integrin heterodimers can yield different ligand binding specificities despite the utilization of similar binding surfaces on each chain.
Like the ␤ 1 chain, ␤ 7 can pair with ␣ chains either containing or lacking a classical A-domain. It is known that the MIDAS motif of the ␤ chain is involved in ligand binding in the context of the A-domain-containing integrins ␣ L ␤ 2 and ␣ M ␤ 2 (43)(44)(45) and in ligand binding in the context of ␣ chains that lack A-domains such as ␣ 4 ␤ 1 , ␣ 5 ␤ 1 , ␣ IIb ␤ 3 , ␣ V ␤ 5 , and ␣ V ␤ 6 (35,(37)(38)(39)(40)(41)(42). We have shown for the first time that the MIDAS motif of a single ␤ chain, ␤ 7 , is critical whether it is paired with an A-domain-containing (␣ E ) or A-domain-lacking (␣ 4 ) ␣ chain. The results stress that in every case studied, the ␤ chain has a critical role to play in ligand binding, whatever the nature of the ␣ chain. Because of this, it seems likely that the majority of integrin ligands will possess residues that are critical for interaction with ␣ and ␤ chains. Such a mechanism has been proposed for the binding of the ␣ 5 ␤ 1 integrin to fibronectin, where the RGD sequence in the tenth type III repeat together with a "synergy region" in the ninth type III repeat contribute to integrin binding (82). Similarly, along with the critical CDloop in domain 1 of VCAM-1 and MAdCAM-1, an "accessory site" in domain 2 has been implicated in the binding of the ␣ 4 ␤ 1 and ␣ 4 ␤ 7 integrins (9,(83)(84)(85). For these non-A-domain-containing integrins, it is interesting that although the primary binding site in the ligand includes a critical exposed acidic residue, the accessory site need not. Indeed, it has been suggested that the MIDAS site of the ␤ 1 chain coordinates the acidic residue of the RGD sequence, whereas the synergy site interacts with the ␣ 5 chain (86). For A-domain-containing integrins containing MIDAS motifs in both the ␣ and ␤ chains, it might be expected that two acidic residues within the ligand are vital for ligand binding. In general, however, only one such residue has been identified in these ligands. In some cases, it is possible that the necessary pairing of acidic residues might be achieved by dimerization of ligand. Indeed, both ICAM-1 and E-cadherin are thought to form dimers on the cell surface (87)(88)(89). It also remains possible that the ␤ chain A-like domain may regulate ligand binding by influencing the conformation of the A-domain in the ␣ chain (30), but this is more difficult to reconcile with the similar role of the ␤ 7 chain MIDAS in integrins with and without an ␣ chain A-domain. Further structural data will be required to resolve these issues.