a4 Integrin Binding Interfaces on VCAM-1 and MAdCAM-1 INTEGRIN BINDING FOOTPRINTS IDENTIFY ACCESSORY BINDING SITES THAT PLAY A ROLE IN INTEGRIN SPECIFICITY*

From the ‡Wellcome Trust Centre for Cell-Matrix Research, School of Biological Sciences, University of Manchester, 2.205 Stopford Building, Oxford Road, Manchester, M13 9PT, United Kingdom, ¶Zeneca Pharmaceuticals, Alderley Park, Macclesfield, Cheshire, SK10 4TF, United Kingdom, the iBritish Biotech Ltd., Watlington Road, Cowley, Oxford OX4 5LY, United Kingdom, and the **Laboratory of Molecular Biophysics, Oxford Centre for Molecular Sciences, Department of Biochemistry, University of Oxford, South Parks Road, Oxford OX1 3QU, United Kingdom

Under normal conditions, leukocytes exhibit a weakly adhesive phenotype, however, at sites of inflammation or at specialized lymphoid tissues, leukocyte adhesion receptor activity is modulated and cells become capable of interacting with ligands expressed on the lumenal surface of the vasculature. Leukocyte integrin receptors and endothelial immunoglobulin superfamily cell adhesion molecules (IgCAMs) make key contributions to the events that facilitate leukocyte emigration into the tissues (1,2).
Thus, integrin-IgCAM interactions appear to share several common features: a characteristic Ig-fold and a dominant acidic peptide motif flanked by non-contiguous accessory residues. Key questions that remain are: what are the principal features of IgCAM structures that govern receptor selection and specifically does the dominant LDV-like motif encode integrin receptor specificity or are additional residues required to discriminate between integrin receptors?
Using the VCAM-1 crystal structure (29) as a basis for a mutagenesis study we have generated IgCAM chimeras and investigated the link between IgCAM LDV motif degeneracy and integrin binding. Using this data we have further characterized the interaction of ␣4␤1 and ␣4␤7 with VCAM-1 and have identified additional key binding/regulatory residues both within and outside the C-D loop of Ig domain 1. Our results provide evidence of distinct but overlapping ␣4␤1/␣4␤7 adhesion footprints that extend over both of the most membrane distal domains of VCAM-1 and provide an insight into the mode of integrin-IgCAM specificity.
CAM-Fc Cloning and Expression-Truncated CAM-Fc chimeras consisting of Ig domains 1 and 2 of human VCAM-1 or murine MAdCAM-1 fused to the hinge and Fc region of human IgG1 were constructed as follows. pCDM8-VCAM (a gift of J. Clements) and pCDM8-MAdCAM (a gift of Dr. Peter Altevogt, German Cancer Research Center, Heidelberg, Germany) were used as templates for the polymerase chain reaction amplification of DNA encoding 5Ј and 3Ј restriction sites, the 5Ј-untranslated region, Ig domains 1 and 2 of VCAM-1 or MAdCAM-1 and a 3Ј splice donor consensus sequence (35). Amplified DNA was cloned into the phagemid vector pUC118 (VCAM) or pUC119 (MAdCAM) and sequenced. CAM cDNA constructs were then subcloned using EcoRI (VCAM) or EcoRI/BamHI (MAdCAM) into the pIg mammalian expression vector (a gift of D. Simmons, Institute of Molecular Medicine, Oxford, UK) encoding a consensus splice acceptor sequence and the hinge and Fc portion of human IgG1 downstream of the multiple cloning site.
Mutations were introduced into CAM cDNAs by the method of Kunkel et al. (41) using the Mutagene phagemid mutagenesis kit (Bio-Rad, Hemel Hemstead, Herts, UK). Mutants were confirmed by sequencing. A list of mutagenic primers (size range 20 -74-mer) is available on request.
COS-1 cells in exponential phase were trypsinized and washed twice with divalent cation-free phosphate-buffered saline and resuspended in cation-free phosphate-buffered saline to a concentration of 6 -8 ϫ 10 6 cells/750-l aliquot. Triplicate aliquots were mixed with 10 -20 g of pIg-CAM DNA, prepared by alkaline lysis and anion-exchange resin (Qiagen, Dorking, Surrey, UK), and chilled on ice for 15-20 min in 0.4-cm gap electroporation cuvettes. Cells were electroporated in a Gene Pulser II (Bio-Rad) electroporator (parameters: 950 microfarads, 0.25 kV). Triplicate transfectants were pooled and resuspended in culture medium and seeded into 3 ϫ 225-cm 2 flasks each containing 35 ml of culture medium. Cells were cultured for 16 h in a humidified atmosphere of 5% (v/v) CO 2 at 37°C. Culture supernatants were then replaced with Dulbecco's minimal essential medium, 2 mM glutamine, 1% fetal calf serum (precleared of immunoglobulins with protein G-Sepharose, Pharmacia, St. Albans, Herts, UK) and cells cultured for 7-10 days. Expressed proteins were isolated from conditioned media by mixing with protein A-Sepharose (Pharmacia) for 16 h and eluting bound proteins with 0.1 M citric acid, pH 3.5, into a neutralizing volume of 1 M Tris, pH 9.0. CAM-Fc proteins were dialyzed against 25 mM HEPES, 150 mM NaCl, pH 7.4 (HEPES-buffered saline, HBS).
Enzyme-linked Immunosorbent Assay of CAM-Fc Proteins-Assays were carried out in 96-well plates (Costar, Northumbria Biologicals, Cramlington, UK). Wells were coated for 60 min at room temperature with 50 l of CAM-Fc (10 g/ml) and nonspecific sites blocked with 5% (w/v) bovine serum albumin, 150 mM NaCl, 10 mM Tris, pH 7.4, for 60 min. Plates were washed and 50-l aliquots of primary mAbs (10 g/ml) in HBS were added to triplicate wells and incubated for 60 min at room temperature. Bound antibody was detected with rabbit anti-mouse IgG (Fc-specific) alkaline phosphatase-conjugated secondary antibody (1: 1000 dilution in HBS; Sigma, Poole, Dorset, UK) and chromogenic substrate, p-nitrophenyl phosphate (Sigma).
Cell Spreading Assay-Cell spreading assays were performed as described previously (42,43). Assays were carried out in 96-well plates. Wells were coated for 60 min at room temperature with 50 l of CAM-Fc diluted with HBS. Wells were blocked for 60 min at room temperature with 100 l of 10 mg/ml heat-denatured bovine serum albumin (42). A375-SM cells were detached with 0.05% (w/v) trypsin, 0.02% (w/v) EDTA, resuspended to 1 ϫ 10 5 /ml in Dulbecco's minimal essential medium, 25 mM HEPES and allowed to recover for 10 min at 37°C. 100-l aliquots of cells were added to each well and plates incubated in a humidified atmosphere of 7% (v/v) CO 2 for 60 min at 37°C. Cells were fixed by the addition of 20 l of 50% (v/v) glutaraldehyde and monitored for the degree of spreading using phase-contrast microscopy as described (43). Each data point was obtained by counting at least 300 cells/well from a number of randomly selected fields. No cell spreading was observed on wells coated only with heat-denatured bovine serum albumin.
Cell Attachment Assay-96-well plates were coated for 60 min at room temperature with 50 l of CAM-Fc diluted with HBS and blocked with heat-denatured bovine serum albumin (see above). JY cells were centrifuged at 1400 ϫ g for 3 min, washed in Dulbecco's minimal essential medium, 25 mM HEPES (assay medium), resuspended at 1 ϫ 10 6 /ml in assay medium and allowed to recover at 37°C for 10 min. 50-l aliquots of cells were added to CAM-Fc-coated wells containing 50 l of HBS, 0.2 mM MnCl 2 and incubated in a humidified atmosphere of 7% (v/v) CO 2 for 20 min at 37°C. Cells attached to untreated wells were fixed directly with 20 l of 50% (v/v) glutaraldehyde to enable quantitation of cell number. Nonspecifically attached cells were aspirated from experimental wells and the residual cells fixed with 5% (v/v) glutaraldehyde. Fixed cells were then washed three times with H 2 O and stained for 60 min with 0.1% (w/v) crystal violet in 0.2 M MES, pH 6. Excess stain was removed with three water washes before cells were lysed with 10% (v/v) acetic acid and the released stain measured spectrophotometrically at A 560 nm .

RESULTS
The predicted C-D loop sequences of VCAM-1, MAdCAM-1, and ICAMs 1-3 have all been implicated as sites for integrin binding and, without exception, possess an acidic residue essential for receptor interaction. Alignment of these sequences reveals a consensus integrin-binding motif (Fig. 1). Despite only subtle variations within this consensus and the use of a relatively conserved Ig domain structure, integrins are capable of discriminating between IgCAM ligands. Although the two related integrins, ␣4␤1 and ␣4␤7, can both bind VCAM-1 and MAdCAM-1, there is a ligand preference between the two integrins; ␣4␤1 is primarily a receptor for VCAM-1 whereas ␣4␤7 is primarily a receptor for MAdCAM-1. The C-D loops of VCAM-1 and MAdCAM-1 differ in two major respects: (i) the VCAM-1 loop is composed of 8 amino acids, the MAdCAM-1 loop only 6, and (ii) two positions flanking the active site motif differ in MAdCAM-1, Gly for Gln and Ser for Pro (see Fig. 1). We therefore tested the hypothesis that IgCAM selection by these two integrins is regulated by C-D loop sequence differences.
For this purpose truncated VCAM-1/MAdCAM-1 proteins containing the two N-terminal Ig domains were synthesized as fusion proteins joined to the Fc region of human IgG 1 . Expression of IgCAM protein in a soluble form provides a major advantage over cell-surface expressed molecules since it is possible to examine the dose dependence of adhesive activity in a quantitative manner and determine relative activities. Since VCAM-1 and MAdCAM-1 Ig domain 1 C-D loops are predicted to differ in size, two VCAM-1/MAdCAM-1 chimeras and two MAdCAM-1/VCAM-1 chimeras were constructed, each chimera differing with respect to their parent molecule and C-D loop size (see Fig. 2). CAM-Fc fusion proteins were purified and tested for their ability to support ␣4␤1-dependent or ␣4␤7-dependent cell adhesion and structural integrity monitored using a panel of mAbs (Fig. 3, Table I). VCAM-1-Fc supported ␣4␤1dependent cell spreading, but MAdCAM-1 was a poor ligand confirming previous reports (23,44). When VCAM/MAdCAM C-D loop chimeras were tested for ␣4␤1 binding activity, one chimera (VCAM-Fc chimera B) demonstrated a super-adhesive phenotype compared with the native VCAM-Fc construct (Fig.  3). In contrast, the VCAM-Fc chimera A failed to support cell adhesion as did both MAdCAM-Fc chimeras. Similarly ␣4␤7-dependent assays revealed that both MAdCAM-Fc chimeras demonstrated very low adhesive activities. However, again VCAM-Fc chimera B supported increased ␣4␤7 binding compared with native VCAM-Fc (Fig. 3). Although local structural perturbations may be responsible for the reduced binding activities of some of the chimeras, as a whole these data suggest that the VCAM-1 C-D loop is not entirely responsible for preferential recognition of VCAM-1 by ␣4␤1 compared with MAd-CAM-1. However, interestingly, they do suggest that the MAd-CAM-1 C-D adhesion loop represents a more ideal ␣4 integrin binding structure.
We next attempted to identify those residues of the VCAM-1 C-D loop that were responsible for ␣4␤1/␣4␤7 binding activity and thus provide a molecular explanation for the super-adhesive activity of the MAdCAM-1 adhesion loop. The x-ray crystal structure of VCAM-1 domain 1 C-D loop (29) was analyzed and those residues that were both conserved between species (human, mouse, and rat) and were solvent-accessible were targeted for mutagenesis. Residues Arg, at position 36 of the mature protein (Arg 36 ), Gln 38 , Ile 39 , Asp 40 , Pro 42 , and Leu 43 met these criteria and VCAM-Fc proteins, mutated at each of these positions, were expressed, purified, and tested for their ability to promote ␣4␤1/␣4␤7-dependent cell adhesion and for their structural integrity (Figs. 4-6, Table I). The D40A mutation (alanine substituted for aspartate at position 40) has been previously shown to abolish VCAM-1-␣4␤1 binding and was used as a control (16,18,19,21,22). In contrast to published data (19), substitution of glutamate for aspartate (D40E) at this position markedly reduced ␣4␤1-dependent cell spreading, but did not abolish it (Fig. 4). Mutation R36A was found to have no effect on ␣4␤1-dependent cell adhesion, however, R36E markedly reduced activity. Mutations at position 38 that intro- duced either a positive or negative charge (Glu or Arg) also reduced cell adhesion. Mutations at positions 39 and 43 where aliphatic side chains were replaced with charged residues (Arg or Lys) also had negative effects on VCAM-1 activity and, in the case of mutant L43K, abolished ␣4␤1 binding (Fig. 4).
We further investigated the role of amino acids Glu 38 and Ile 39 in ␣4␤1-VCAM-1 interactions by making further amino acid alterations at these positions. Gln 38 mutated to either a glycine or a leucine demonstrated a super-adhesive phenotype with significant changes in the concentration of protein required for half-maximal cell spreading (2-4-fold decrease, Fig.  5). Thus, removing a hydrophilic side chain at this position, or replacing it with an aliphatic group, conferred not a negative effect but instead augmented ␣4␤1-VCAM-1 binding. These data are consistent with the effects observed with one of the VCAM-1/MAdCAM-1 chimeras (VCAM-Fc chimera B), where the equivalent amino acid to VCAM-1 Gln 38 in the MAdCAM-1 C-D loop is glycine. The effect of increasing hydrophobicity at position 38 was further tested by the introduction of a large aromatic residue, phenylalanine. Intriguingly, ␣4␤1-dependent cell adhesion was still supported with only a 2-fold increase in the concentration for half-maximal spreading (Fig. 5). Two further mutations at position 39 were performed to test the effects of subtle changes in the aliphatic side chain characteristics, I39A and I39V. Remarkably, the smaller alanine side chain severely affected ␣4␤1 binding ability with a marked reduction in maximal spreading; however, a valine at this position only slightly affected activity (Fig. 5).
␣4␤7 adhesion data largely mirrored that of ␣4␤1, however, no super-adhesive phenotype was obtained with mutations at position Gln 38 . In addition, D40E abolished ␣4␤7-dependent attachment and mutants Q38F and I39V markedly reduced ␣4␤7 binding (Fig. 6). MAdCAM-1 C-D loop residues were also tested for their contribution to ␣4␤1 and ␣4␤7 binding. Residues Gly 39 , Asp 41 , and Leu 44 were mutated and expressed as MAdCAM-Fc fusion proteins. ␣4␤1-dependent cell adhesion was abolished and ␣4␤7-dependent cell attachment severely affected by all four mutations, reinforcing the role of this region of the molecule in integrin binding ( Fig. 7 and data not shown). Of note, however, is the finding that the MAdCAM-D41E-Fc mutant had a very low ␣4␤7 binding activity, in marked contrast to the equivalent VCAM-1 mutation (D40E), which could support a significant level of ␣4␤1-mediated cell spreading.
The results described above clearly suggest that residues outside the VCAM-1 C-D loop play a role in integrin selection. We therefore attempted to map the adhesion "footprints" of ␣4␤1 and ␣4␤7 in an effort to identify a difference in binding requirements and an explanation for the preference for VCAM-1 by ␣4␤1. Since the VCAM-1 C-D loop appears to be central to ␣4 integrin binding we reasoned it was likely that residues that comprise this "face" of the Ig domain 1 might be capable of participating directly in integrin binding. In addition, since VCAM-1 Ig domain 2 has been predicted to play a role in ␣4␤1 binding from studies with ICAM-1 chimeras, this region was also targeted for mutagenesis (17). Those residues that constitute the predicted integrin-binding face that are both conserved between species and solvent-accessible (as revealed by the x-ray crystal structure, Ref. 29) are highlighted in Fig. 8. These residues were mutated to either alanine or another amino acid (often to swap or introduce charge) and expressed as VCAM-Fc fusion proteins. All mutants were examined for structural perturbations using anti-VCAM-1 mAbs in enzyme-linked immunosorbent assay (Table I). Only mutation V47K affected binding of both domain 1 mAbs. VCAM-Fc mutants were then tested for their ability to support ␣4␤1-dependent spreading or ␣4␤7-dependent attachment (Figs. 9 and 10, respectively).
For clarity and visual presentation the effects of mutations were classified according to the maximal level of adhesion attained and the concentration required to reach half-maximum adhesion. "Severe" effects produced Ͻ50% of the maximal level of spreading on native VCAM-Fc and the protein concentration required for half-maximal spreading was either not reached or was Ͼ7-fold higher than native VCAM-Fc. "Marked" effects were classified as those that reached 60 -70% maximum native adhesion with a 3-4-fold increase in concentration required for half-maximal concentration. "Slight" effects produced 80 -100% of maximum spreading with only a 2-fold increase in concentration required for half-maximal adhesion. In addition, if adhesion was augmented (maximum adhesion un-   changed but protein concentration required for half-maximum adhesion decreased two-fold or more) then a mutant VCAM-Fc was described as "super-adhesive" (see Fig. 11). Only two residues outside the C-D loop, when mutated, were found to affect ␣4␤1-VCAM-Fc binding severely, Leu 70 and Glu 87 . Several residues, however (Thr 72 , Glu 81 , Asp 143 , Ser 148 , and Glu 150 ), markedly affected ␣4␤1 interactions. Mutations of amino acids at six further positions resulted in a slight perturbation of ␣4␤1-dependent adhesion (Lys 46 , Val 47 , Ser 68 , Ser 77 , Glu 155 , and Thr 157 ). Interestingly, three amino acids outside the C-D loop (Thr 74 , Leu 80 , and Lys 147 ), when mutated to alanine, conferred a super-adhesive phenotype on VCAM-Fc. In contrast, ␣4␤7 appeared to differ in its sensitivity to mutation of residues outside the C-D loop region. Only one residue in Ig domain 1 severely affected ␣4␤7 binding (Glu 87 ) and mutations of Leu 70 had no effect on ␣4␤7 binding (Fig. 10). Furthermore, Ig domain 1 mutations that markedly affected ␣4␤1 binding (Thr 72 and Glu 81 ) had no effect on ␣4␤7-mediated cell attachment. One of the main differences between ␣4␤1and ␣4␤7-VCAM-1 interactions appeared to lie in Ig domain 2; mutation of three residues at positions 143, 148, and 150 severely affected ␣4␤7-dependent cell adhesion. In addition, residues that markedly affected ␣4␤7 binding were also located in domain 2 (Lys 152 and Glu 155 ). A further difference was the super-adhesive activity of T151A for ␣4␤7. The key integrin-binding residues identified in VCAM-1 domain 2 correspond to the CЈ-E loop region, a region in MAdCAM-1 domain 2 highlighted to be of potential functional importance (29). We therefore directly tested the role of this region of MAdCAM-1 in ␣4 integrin binding by engineering a deletion mutant in which the central acidic region (143-150) of the highly extended Ig domain 2 CЈ-E loop had been deleted. The resultant mutant (MAdCAM-⌬CЈ-E-Fc) was then assayed for ␣4␤1 and ␣4␤7 binding activity (Fig. 12). In both experiments, MAdCAM-⌬CЈ-E-Fc failed to support ␣4␤1 or ␣4␤7-mediated cell adhesion, confirming the importance of this region for both ␣4 integrin interactions. DISCUSSION Using IgCAM chimeras and a structure-guided site-directed mutagenesis strategy, we have analyzed the molecular basis of ␣4 integrin binding to the IgSF members VCAM-1 and MAd-CAM-1. All VCAM-1 and MAdCAM-1 mutants were produced as recombinant IgG Fc fusion proteins permitting quantitative analysis of purified mutant binding properties.
The main findings reported here are that the C-D loops of these IgCAMs are only partially responsible for integrin specificity, and that ␣4␤1/␣4␤7 integrin binding footprints on VCAM-1 differ in their relative dependence on residues within VCAM-1 Ig domains 1 and 2. In addition, we report that several amino acids within VCAM-1 modulate integrin interactions with residues at predicted integrin contact sites. These findings for the first time demonstrate a direct role for VCAM-1 Ig domain 2 and suggest a mechanism for integrin selection by IgCAMs.
Integrin-IgCAM interactions are dependent on a relatively conserved acidic peptide motif presented as a surface-exposed structure supported on a conserved Ig-domain scaffold. Despite only slight variations in this binding motif, integrin-CAM specificities are exquisite. This suggests two possibilities, either slight variations within this motif are sufficient to discriminate integrin receptors or this motif might represent a general binding structure and integrin selection may be governed by distal sites. We investigated the role of IgCAM C-D loop motifs in regulating integrin specificity by making chimeras between VCAM-1 and MAdCAM-1 (the former is primarily a ligand for ␣4␤1, the latter for ␣4␤7). ␣4␤1-dependent cell spreading assays revealed that MAdCAM-1 chimera Fc fusion proteins expressing either of two VCAM-1 C-D adhesion loop variants failed to demonstrate increased binding to ␣4␤1 over native MAdCAM-1, indicating that the VCAM-1 C-D adhesion loop alone was insufficient to confer full ␣4␤1 binding activity. Intuitively, complementary VCAM-1 (MAdCAM C-D loop) chimeras might therefore be expected to possess lower ␣4␤1 bind-ing activity compared with native VCAM-Fc. However, one chimera (VCAM Fc chimera B) demonstrated super-adhesive ␣4␤1 binding activity. When both sets of chimeras were tested for ␣4␤7 binding activity, essentially identical results were obtained; MAdCAM-1 chimeras with the VCAM-1 C-D adhesion loop supported reduced levels of cell attachment whereas the super-adhesive VCAM-Fc chimera B again demonstrated a slight increase in the ability to bind integrin relative to native VCAM-Fc. These data reinforce the importance of CAM C-D loops in integrin binding and suggest that ␣4 integrins have a distinct preference for the C-D loop structure presented by MAdCAM-1 Ig domain 1. Furthermore, since ␣4␤1 has a greater affinity for VCAM-1 compared with MAdCAM-1 this would suggest that ␣4␤1-VCAM-1 binding is enhanced (or MAdCAM-1 binding is attenuated) by sites outside the C-D loop.
Despite several studies on ␣4␤1-VCAM-1 interactions, ␣4␤1 contact sites within the VCAM-1 C-D loop are not fully understood. In an effort to enhance our understanding of integrin-IgCAM interactions we made a detailed analysis of the VCAM-1 and MAdCAM-1 C-D loops. To aid our study we analyzed the x-ray crystal structure of VCAM-1 (29) and identified those C-D loop residues that were both conserved between species (mouse, rat, and human) and surface-exposed, reasoning that functionally important residues would be conserved and accessible to integrin.
␣4 integrin-dependent adhesion data obtained with purified mutant C-D loop VCAM-Fc fusion proteins has extended our understanding of ␣4 integrin binding requirements. Mutation of Arg 36 to alanine had no effect, while replacement with a glutamate markedly affected ␣4␤1 interactions. This indicates that the physical characteristics of the arginine side chain are not absolutely required for integrin-VCAM-1 engagement, however, a negative charge in this position (glutamate) appears to adversely affect the recognition of adjacent residues. Analysis of the x-ray crystal structure of VCAM-1 in this region reveals that Arg 36 is in close proximity to the acidic side chain of Asp 40 (see Fig. 8) and therefore might feasibly influence the pK a of the crucial Asp 40 carboxyl group and consequently integrin active site engagement. Alternatively, alterations at position 36 might affect the conformation of neighboring side chains within the C-D loop. Mutagenesis data on amino acid Gln 38 suggest an important role in ␣4␤1-VCAM-1 binding, since changing the glutamine to a charged residue (glutamate or lysine) severely affected ␣4␤1 binding. However, removal of the glutamine side chain (Q38G) was found to convert VCAM-1 into a super-adhesive ␣4␤1 ligand and replacement of glutamine with an aliphatic residue (leucine) resulted in a further enhancement of integrin binding activity. When a phenylalanine was substituted into this position a slight decrease in ␣4␤1 binding capacity was observed. At least two explanations can be drawn from these data: first, part of the integrin active site might comprise a restricted pocket with hydrophobic characteristics and thus have a preference for interacting with hydrophobic residues. Alternatively, residues in position 38 might not be directly recognized by the ␣4␤1 active site but might confer subtle structural alterations on the remainder of the C-D loop, thus modifying integrin binding. It is notable that the MAdCAM-1 C-D loop possesses a glycine in a homologous position to Gln 38 , providing an explanation for the super-adhesive phenotype of VCAM-Fc chimera B.
Residues at VCAM-1 positions 39, 40, and 43 are critical for ␣4␤1 binding. Several mutations of Ile 39 were tested: replacing isoleucine with a charged residue or an alanine residue severely affected ␣4␤1 binding ability, however, replacement with valine only resulted in a slight decrease in activity. These data are again consistent with a hydrophobic pocket at or near the integrin-ligand-binding site. Charged residues or side chains too small to fill this pocket result in compromised integrin interactions, while longer, branched side chains such as valine appear to be sufficient to permit essentially unperturbed ␣4␤1 binding. Previous reports in which Asp 40 was replaced with alanine, asparagine, or lysine resulted in abolition of VCAM-1 adhesive activity (16,18,19), however, mutation of Asp 40 to glutamate was reported to have no detectable affect on ␣4␤1 binding (19). Here, we observed a marked effect on ␣4␤1-VCAM-1 binding when Asp 40 was mutated to glutamate. In addition VCAM-1 mutants R36A and P42G have been reported previously to perturb ␣4␤1 binding (19,22), an explanation for these discrepancies is unclear. Previous mutagenesis studies coupled with sequence alignment data have identified the ␣4 binding motif within VCAM-1 as IDSP; however, we report here that Leu 43 is crucial for integrin-VCAM-1 binding and therefore propose that the adhesion motif should be extended to include this residue: i.e. IDSPL.
Analysis of the effects of residues in the C-D loop region on ␣4␤7-VCAM-1 binding revealed essentially the same results as ␣4␤1 binding, however, there were several key differences: (i) the Q38G mutation slightly reduced ␣4␤7 interactions, while Q38L only slightly enhanced ␣4␤7 binding, (ii) mutant I39V had markedly reduced ␣4␤7 binding activity, and (iii) D40E abolished ␣4␤7-VCAM-1 interactions. These data suggest sub- All atoms are represented as spheres proportional to predicted Van der Waal's radii. Atoms are colored according to type: gray, carbon; red, oxygen; blue, nitrogen; and yellow, sulfur. Atoms colored cyan are from residues that are conserved between species of VCAM-1 (human, mouse, and rat), predicted to be solvent-accessible according to x-ray crystal structure data, and located on the Ig domain 1 C-D loop face. C as B except the structure has been rotated anti-clockwise by 90°.
tle differences between the ligand-binding sites of ␣4␤1 and ␣4␤7 and imply that ␣4␤1 may possess a slightly deeper pocket to accommodate the crucial acidic group at position 40.
Mutation 1-Fc chimera A possessed ␣4␤7 binding activity, suggesting that the increased loop size of this construct comprised a more ideal adhesion structure than MAdCAM-G39Q-Fc. The importance of this region of MAdCAM-1 has recently been confirmed by mutagenesis and peptide inhibition studies: MAdCAM-1 residues Leu 40 , Asp 41 , Thr 42 , and Leu 44 were found to be required for full ␣4␤7 binding activity (45).
These data demonstrate for the first time the effects of performing full IgCAM C-D loop swaps. However, partial swaps have been investigated previously; the central portion of the ICAM-1 Ig domain 1 C-D loop has been used to replace the analogous region of VCAM-1 and was found to have no negative effect on ␣4␤1 binding (18). Our data with the same mutant in our VCAM-Fc construct (Q38IDS/GIET) demonstrated identical integrin binding activity (data not shown). Analysis of the effects of point mutations described above now provides an explanation for these observations: the overall phenotype of the Q38IDS/GIET mutation is a composite effect of Q38G (superadhesive) and D40E (marked negative effect), thus the overall effect produces a native phenotype. We further investigated the ability of the ICAM-1 (and ICAM-3) C-D loop sequences to substitute for that of VCAM-1 by performing whole loop swaps (8 VCAM-1 C-D loop amino acids replaced with 8 ICAM-1 (or ICAM-3) C-D loop amino acids, see Fig. 1). The resulting constructs failed to bind two anti-VCAM-1 mAbs (4B2 and 1E10, see Table I) and failed to bind ␣4␤1 (data not shown). These data suggest that the ICAM-1 and ICAM-3 C-D loops, although similar to that of VCAM-1 in terms of sequence and function, may adopt markedly different conformations.
Taken together, these data provide an insight into the struc- tural requirements of ␣4 integrins for the C-D loops of VCAM-1 and MAdCAM-1 and also identify mutant forms of VCAM-1 capable of conferring a super-adhesive phenotype. In an effort to identify those sites outside the C-D loop that differentially affect ␣4␤1/␣4␤7 binding, we performed an extensive survey of the contribution of VCAM-1 Ig domain 1 and 2 amino acids. We targeted residues that were conserved between species, surface-exposed, and confined to the C-D loop face of the molecule according to the VCAM-1 x-ray crystal structure (29, Fig. 8). When purified mutant VCAM-Fc proteins were tested for their ability to bind ␣4␤1 and ␣4␤7 in cell-based assays, striking differences between ␣4␤1 and ␣4␤7 binding were observed. Although both integrins had an acute requirement for residues within the C-D loop, their respective adhesion footprints over VCAM-1 Ig domains 1 and 2 differed in two major respects. First, key Ig domain 1 residues required for ␣4␤1-VCAM-1 interactions outside the C-D loop included Leu 70 , Thr 72 , Glu 81 , and Glu 87 but only Glu 87 appeared to be involved in ␣4␤7-VCAM-1 binding. Second, mutation of amino acids at three positions in Ig domain 2 (Asp 143 , Ser 148 , and Glu 150 ) had a marked, but not severe effect, on ␣4␤1 binding whereas these residues have a much more pronounced affect on ␣4␤7 binding (Fig. 11). Furthermore, mutation of two residues, Lys 152 and Glu 155 , had a marked effect on ␣4␤7 binding reinforcing the importance of this region for ␣4␤7-VCAM-1 binding. In addition, several residues were found to augment integrin binding with effects analogous to the Q38G/Q38L mutants. In all instances augmentation sites (Thr 74 , Leu 80 , Lys 147 (both ␣4␤1 and ␣4␤7), and Thr 151 (␣4␤7 only)) lie spatially close to residues that affect integrin interactions (Fig. 11). The identification of Thr 151 as an augmentation site in the case of ␣4␤7-VCAM-1 binding further implies that residues in this region of VCAM-1 Ig domain 2 have a greater role for ␣4␤7 interactions compared with ␣4␤1. Although the mechanism of integrinbinding augmentation is unclear, it is likely that changes at these positions confer slight structural changes which lead to enhanced integrin recognition of nearby contact sites. The main evidence to support this hypothesis comes from the finding that replacement of the native residue with amino acids of very different physical characteristics are capable of producing the same result, e.g. L80A, L80K (Fig. 9).
Overall these data suggest an important shift in the "binding footprint" between ␣4␤1 and ␣4␤7. ␣4␤1 primarily binds to VCAM-1 Ig domain 1 via a group of residues comprised of the FIG. 11. Space-fill representation of VCAM-1 Ig domains 1-2 with target residues (see Fig. 8) color coded according to the effects of mutation on adhesion. All atoms are represented as spheres proportional to predicted Van der Waal's radii. All other residues are "ghosted." Color code for effect of mutations on integrin binding: green, no effect; yellow, slight; orange, marked; red, severe; purple, augmentation (see text for effect classification). A, effects of mutations on ␣4␤1 binding; and B, effects of mutations ␣4␤7 binding.
C-D adhesion loop and three further adjacent Ig domain 1 sites with a contribution from several Ig domain 2 residues, whereas ␣4␤7, apart from residues within the Ig domain 1 C-D loop has a reduced requirement for sites within Ig domain 1 but a pronounced requirement for residues that locate to the CЈ-E loop/E ␤-strand of Ig domain 2. Conflicting data regarding the role of Leu 70 for ␣4␤7 binding have been reported: VCAM-L70N was found to abolish both ␣4␤1 and ␣4␤7 interactions (22).
Since both Leu 70 mutants tested here, L70A and L70K perturbed ␣4␤1 but not ␣4␤7 binding, an explanation for this discrepancy is not clear, however, enzyme-linked immunosorbent assay data suggest that the L70N mutant may perturb structure since two anti-functional mAbs bind with reduced affinity relative to a native VCAM-1 construct (56 and 42%; Ref. 22).
One of the most important implications of our findings is that the CЈ-E loop of Ig domain 2 contributes to the regulation of integrin-IgCAM specificity. This hypothesis is strikingly reinforced by sequence alignment studies between VCAM-1 and MAdCAM-1 (29). The CЈ-E loop of MAdCAM-1 Ig domain 2 is highly acidic and contains a sequence of three consecutive glutamate residues (murine MAdCAM-1, Ref. 24) or five consecutive glutamate residues (human MAdCAM-1, Ref. 51). Since ␣4␤7 binding is highly sensitive to the removal, by sitedirected mutagenesis, of acidic residues from the CЈ-E loop of VCAM-1 Ig domain 2, this suggests a key role for this region in governing MAdCAM-1-␣4␤7 interactions. We therefore examined the effect of truncating the MAdCAM-1 Ig domain 2 CЈ-E loop by removing residues 143-150 (including the acidic sequence of amino acids predicted to be inserted residues in comparison to VCAM-1 Ig domain 2 CЈ-E loop, Ref. 29). In both ␣4␤1and ␣4␤7-dependent assays, cell binding activity was abolished, confirming the role of this region in integrin interactions (Fig. 12).
Comparison of integrin-IgCAM interactions with other ligands reveals striking parallels. Several integrins interact with the extracellular matrix protein fibronectin including ␣5␤1, ␣V␤1, ␣IIb␤3, and ␣V␤3 and all four integrins require the tripeptide motif RGD displayed by the 10th type III repeat of fibronectin (FnIII(10)) (6 -10). Site-directed mutagenesis, domain deletion, and peptide-based approaches have subse-quently identified a second binding site (sequence PHSRN, termed the synergy site) located in the 9th type III repeat (46 -52). The x-ray crystal structure of this region of fibronectin (FnIII(7-10)) has recently been solved and reveals a tandem array of fibronectin type III repeats; each composed of an anti-parallel array of ␤-strands analogous to Ig domains (53). The RGD motif lies in the F-G loop of FnIII(10) and the synergy sequence is found in the C-E loop and part of E ␤-strand. Comparison of the arrangement of active sites in fibronectin with that of VCAM-1 highlights three similarities: (i) both sets of adhesion motifs are on surface-exposed loops, (ii) both molecules display active sites on the same face of adjacent domains, and (iii) the functional groups of the key amino acids in each motif are approximately 30 -40 Å apart. Thus, two different integrin ligands appear to share a broadly similar "binding topology" composed of a dominant adhesion motif (IDSPL, VCAM-1, and RGD, fibronectin) coupled to second "synergy site" that may regulate integrin specificity (Ig domain 2 CЈ-E loop-E strand, VCAM-1 and III(9) C-E loop and part of E ␤-strand, fibronectin).
Historically, peptide-based approaches have dominated the identification of integrin ligand active sites, and this has led to the perception that integrins principally recognize discrete peptide motifs such as RGD and LDV. However, data, largely provided by site-directed mutagenesis studies, is now accumulating to suggest that integrin recognition sites on ligands are more complex and are comprised of a dominant acidic peptide flanked by additional contact residues. In the future it will be of interest to assess the relative contribution of functionally important residues in providing the free-energy of interaction between integrin and ligand, an approach successfully used to identify residues responsible for human growth hormone-human growth hormone receptor interactions (54). In turn these findings, together with the use of reporter mAbs (such as those that recognize integrin neo-epitopes arising as a result of ligand engagement), might provide clues as to the connection between ligand engagement by integrin and the conformational responses that occur prior to signal transduction across the plasma membrane.