Molecular Basis of Ligand Recognition by Integrin α5β1

The NH2-terminal portion (putative ligand-binding domain) of α subunits contains 7 homologous repeats, the last 3 or 4 of which possess divalent cation binding sequences. These repeats are predicted to form a seven-bladed β-propeller structure. To map ligand recognition sites on the α5 subunit we have taken the approach of constructing and expressing αV/α5 chimeras. Although the NH2-terminal repeats of α5 and αV are >50% identical at the amino acid level, α5β1 and αVβ1show marked differences in their ligand binding specificities. Thus: (i) although both integrins recognize the Arg-Gly-Asp (RGD) sequence in fibronectin, the interaction of α5β1 but not of αVβ1 with fibronectin is strongly dependent on the “synergy” sequence Pro-His-Ser-Arg-Asn; (ii) α5β1 binds preferentially to RGD peptides in which RGD is followed by Gly-Trp (GW) whereas αVβ1 has a broader specificity; (iii) only α5β1 recognizes peptides containing the sequence Arg-Arg-Glu-Thr-Ala-Trp-Ala (RRETAWA). Therefore, amino acid residues involved in ligand recognition by α5β1 can potentially be identified in gain-of-function experiments by their ability to switch the ligand binding properties of αVβ1 to those of α5β1. By introducing appropriate restriction enzyme sites, or using site-directed mutagenesis, parts of the NH2-terminal repeats of αV were replaced with the corresponding regions of the α5 subunit. Chimeric subunits were expressed on the surface of Chinese hamster ovary-B2 cells (which lack endogenous α5) as heterodimers with hamster β1. Stable cell lines were generated and tested for their ability to attach to α5β1-selective ligands. Our results demonstrate that: (a) the first three NH2-terminal repeats contain the amino acid sequences that determine ligand binding specificity and the same repeats include the epitopes of function blocking anti-α subunit mAbs; (b) the divalent cation-binding sites (in repeats 4–7) do not confer α5β1- or αVβ1-specific ligand recognition; (c) amino acid residues Ala107–Tyr226 of α5(corresponding approximately to repeats 2 and 3) are sufficient to change all the ligand binding properties of αVβ1 to those of α5β1; (d) swapping a small part of a predicted loop region of αV with the corresponding region of α5 (Asp154-Ala159) is sufficient to confer selectivity for RGDGW and the ability to recognize RRETAWA.

Integrins are ␣,␤-heterodimeric transmembrane receptors that play central roles in cell adhesion, migration, differentiation, and survival (1). More than 20 distinct integrin heterodimers have been identified, and each integrin recognizes a different set of extracellular matrix or cell-surface proteins. The major sequences recognized by integrins within their ligands have been shown to be short motifs that contain a critical Asp or Glu residue, such as Arg-Gly-Asp (RGD). 1 Analysis of the three-dimensional structures of integrin ligands has revealed that these recognition motifs are displayed in surfaceexposed sites (2,3).
Ligand-binding sites within integrins are less well characterized. Although ligand recognition is known to involve the NH 2 -terminal regions of both ␣ and ␤ subunits (4,5), unresolved questions include (i) what is the precise location of the ligand-binding sites? and (ii) how is the specificity of integrinligand binding determined?
The tertiary structure of integrins has not yet been determined; however, high quality structure predictions have been made for the ligand-binding domains of both subunits. These predictions are supported by extensive biochemical analyses (6 -9). The NH 2 -terminal portion of integrin ␣ subunits has been shown to contain seven homologous repeats, each 60 -70 amino acid residues in length. Repeats 4 -7 (or in some integrins repeats 5-7) contain putative divalent cation-binding sites (10). The seven NH 2 -terminal repeats are predicted to fold cooperatively into a seven-bladed ␤-propeller (11). Each blade of the propeller contains four ␤-strands connected by loops of varying length; these strands are tilted such that the connecting loops are either on the lower or upper surfaces of the propeller. Loops between the first and second, and between the third and fourth ␤-strands lie on the lower surface of the propeller, whereas loops between the second and third ␤-strands, and between the fourth ␤-strand of one blade and first ␤-strand of the next blade lie on the upper surface. The divalent cation-binding sites are predicted to lie on the lower surface of the propeller.
An inserted (I or A) domain of about 200 amino acid residues is present in about one-third of integrin ␣ subunits, lying between repeats 2 and 3. In ␣ subunits that contain an A-domain, this module contains the major sites involved in ligand binding (2,9,(12)(13)(14). Ligands have been shown to interact with the top face of this domain through a metal ion-dependent adhesion site (MIDAS) motif (15)(16)(17)(18). Although the A-domain is present in only a subset of ␣ subunits, the region of the ␤ subunit that participates in ligand recognition has been predicted to have a tertiary fold similar to that of an A-domain (19 -22). This domain may also interact with ligand through a MIDAS site (19 -25).
A large number of studies have investigated the location of ligand-binding sites in ␣ subunits that lack an A-domain but no consensus has emerged. Cross-linking of a peptide from the ␥ chain of fibrinogen to ␣ IIb ␤ 3 showed that the major site of interaction was in the fifth NH 2 -terminal repeat (26). An RGDcontaining peptide also cross-linked to ␣ V ␤ 3 via a site in the divalent cation binding repeats, although some cross-linking was also observed to the second and third repeats (27). In addition, recombinant proteins containing repeats 4 -7 of ␣ IIb or ␣ 5 have been shown to possess ligand binding activity (28,29). In contrast to these findings, the epitopes of function blocking anti-␣ subunit mAbs have been localized to repeats 1-3 (4, 8, 30 -36), and mutations in repeats 2-4 of ␣ 3 , ␣ 4 , ␣ 5 , and ␣ IIb have been shown to perturb ligand recognition (6,(33)(34)(35)(36)(37)(38)(39).
Integrin ␣ 5 ␤ 1 is a widely distributed cell surface receptor for fibronectin, and has served as a prototype for the study of integrin-ligand interactions. The central cell-binding domain (CCBD) of fibronectin contains a number of repeated modules termed fibronectin type III repeats. An RGD sequence in the tenth type III repeat is the major binding site for ␣ 5 ␤ 1 (40,41); however, ␣ 5 ␤ 1 differs from other fibronectin-binding integrins, such as those of the ␣ V family, in that ligand recognition is also strongly dependent on binding of a synergy sequence (Pro-His-Ser-Arg-Asn) in the ninth type III repeat (42)(43)(44). In addition, although several closely related integrins (such as ␣ V ␤ 1 , ␣ V ␤ 3 , ␣ V ␤ 5 , and ␣ IIb ␤ 3 ) recognize the RGD sequence, the nature of the amino acid residues COOH-terminal to RGD has a major effect on the affinity and specificity of integrin binding. For example, ␣ 5 ␤ 1 shows strong selectivity toward peptides containing the sequence RGDGW, whereas ␣ V integrins bind well to peptides containing the sequences RGDXF or RGDXR (where X is frequently Ser or Thr) (45,46). In addition, a specific ligand peptide for ␣ 5 ␤ 1 , Arg-Arg-Glu-Thr-Ala-Trp-Ala (RRETAWA), has been isolated from a Cys-X 7 -Cys phage display library (47). Although the RRETAWA sequence appears unrelated to RGD, the binding sites for RRETAWA and RGD on ␣ 5 ␤ 1 are closely overlapping because a peptide containing the RRETAWA sequence acts as a direct competitive inhibitor of RGD binding to ␣ 5 ␤ 1 (36).
A possible criticism of previous integrin mutagenesis studies is that only a loss of ligand binding was demonstrated, therefore it is possible that these mutations had an indirect effect on ligand recognition by perturbing integrin structure. Here we have taken the approach of constructing and expressing chimeric ␣ V /␣ 5 subunits to determine the regions of ␣ 5 that when introduced into ␣ V cause gain of ligand-binding function. Our results show that replacing the second and third repeats of ␣ V with those of ␣ 5 converts ␣ V ␤ 1 into a receptor with the same ligand binding specificity as ␣ 5 ␤ 1 . Furthermore, we demonstrate that a short putative loop region connecting the second and third repeats is involved in the binding of RRETAWA and RGDGW-containing peptides. Our findings are consistent with the ␤-propeller model and appear to rule out a direct role for the divalent cation-binding sites in ligand recognition.
Mutagenesis-Full-length clones of human ␣ V and ␣ 5 were gifts from D. Cheresh (Scripps Research institute, La Jolla, CA) and K. Yamada (National Institute of Dental Research, NIH, Bethesda, MD), respectively. A 1.3-kilobase BamHI/HindIII fragment of ␣ V was subcloned into pUC118. The following oligonucleotides (listed 5Ј to 3Ј) were used to introduce restriction enzyme sites into ␣ V by site-directed mutagenesis using the GeneEditor kit (Promega, Southhampton, UK): TTTGATGA-CAGCTACCTAGGTTATTCTGTGGC (AvrII site), GAGCATCTGTGAG-GGCCCATGGGGATAAAATTTTGGC (NcoI site), GGACAGGGATTTT-GCCAAGGAGGCTTCAGCATTGATTTTAC (BglI site). The mutations introducing the AvrII and BglI sites were silent; however, the mutation introducing the NcoI site caused changes to the amino acid sequence and so an additional base change was made to convert the amino acid sequence at this site to that of ␣ 5 (S100KQ to A107HG).
The presence of the mutations was verified by restriction enzyme digestion of miniprep DNA (Qiagen) prepared from individual clones. To reconstitute full-length ␣ V the mutated 5Ј BamHI/HindIII fragment of ␣ V was ligated with a 3Ј HindIII/XbaI fragment of ␣ V into pCDNA3 cut with BamHI and XbaI. To construct the ␣ V /␣ 5 (F1-G232) chimera, a BamHI/AvrII fragment of ␣ 5 was ligated with an AvrII/ApaI fragment of ␣ V into pCDNA3 cut with BamHI and ApaI. To construct the ␣ 5 / ␣ V (F1-G223) chimera, a BamHI/AvrII fragment of ␣ V was ligated with an AvrII/XhoI fragment of ␣ 5 into ␣ 5 in pCDNA3 cut with BamHI and XhoI. To construct the ␣ V /␣ 5 (A107-G232) chimera, a BamHI/NcoI fragment of ␣ V was ligated with a NcoI/AvrII fragment of ␣ 5 and an AvrII/ ApaI fragment of ␣ V into pCDNA3 cut with BamHI and ApaI. To construct the ␣ V /␣ 5 (A107-C164) chimera, a BamHI/NcoI fragment of ␣ V was ligated with a NcoI/BglI fragment of ␣ 5 and a BglI/XbaI fragment of ␣ V into pCDNA3 cut with BamHI and XbaI. To construct the ␣ V / ␣ 5 (C164 -G232) chimera, a BamHI/BglI fragment of ␣ V was ligated with a BglI/AvrII fragment of ␣ 5 and an AvrII/ApaI fragment of ␣ V into pCDNA3 cut with BamHI and ApaI. The constructs were verified by DNA sequencing. Chimeras were designated according to the position of the restriction site in the corresponding amino acid sequence.
The following oligonucleotides (listed 5Ј to 3Ј) were used to exchange amino acid residues within putative loop regions of ␣ V with the corresponding amino acid residues in Oligonucleotides were purchased from MWG Biotech (Southampton, UK) or from PE-Applied Biosystems (Warrington, UK). Restriction enzymes were from New England Biolabs (Hitchin, UK) or Roche Molecular Biochemicals (Lewes, UK).
Coupling of Cyclic Peptides to IgG-Rabbit IgG (3 mg) was dissolved in 1 ml of Dulbecco's PBS without Ca 2ϩ and Mg 2ϩ (PBSϪ). To this solution, approximately 0.5 mg of bis(sulfosuccinimidyl)suberate (Pierce, Chester, UK) dissolved in 1 ml of PBSϪ was added. The mixture was incubated for 5 min at room temperature, and then cyclic peptide (1-1.5 mg dissolved in 1 ml of PBSϪ) was added. After incubation of the mixture for 5 min at room temperature, unreacted peptide and cross-linker were removed by dialysis against PBSϪ. The dialysate was centrifuged at 13,000 ϫ g for 15 min, and stored in aliquots at Ϫ70°C.
Transfection-Chinese hamster ovary cells B2 variant (50) (a gift from R. L. Juliano, University of North Carolina, NC) were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 2 mM glutamine, and 1% non-essential amino acids (growth medium). Cells were detached using 0.05% (w/v) trypsin, 0.02% (w/v) EDTA in PBS, and plated overnight into 6-well culture plates (Costar). 2 g of wild-type ␣ V or chimeric ␣ V /␣ 5 DNA/well was used to transfect the cells using LipofectAMINE PLUS reagent (Life Sciences) according to the manufacturer's instructions. 24 h post-transfection, cells were removed from the wells using Hank's balanced salt solution containing 5 mM EDTA and tested for transient expression of transfected integrin subunits by flow cytometry. 48 h post-transfection, cells were passaged into 75-cm 2 flasks (Costar) and the medium was supplemented with 0.7 mg/ml G418 (Life Sciences). G418-resistant colonies were harvested after 10 -14 days. The cell population was incubated with mAb P3G8 or mAb11, and then with anti-mouse or anti-rat IgG-coated magnetic beads (Dako) to select for cells expressing wild-type or chimeric integrins. Cells were then cloned by limiting dilution to obtain clones with a high level of expression. Chinese hamster ovary-B2 cells transfected with wild-type human ␣ 5 were a gift from Y. Takada (Scripps Research Institute, La Jolla, CA). The reactivity of cloned cells with a panel of anti-␣ 5 and anti-␣ V mAbs was assessed using flow cytometry, using rat IgG or mouse IgG as controls.
Cell Attachment Assay-Chinese hamster ovary-B2 cells, or cells transfected with chimeric or wild-type integrins were detached using 0.05% (w/v) trypsin, 0.02% (w/v) EDTA in PBS, washed with 150 mM NaCl, 25 mM HEPES, pH 7.4, 1 mg/ml BSA, resuspended in the same buffer with 1 mM MgCl 2 and 1 mM CaCl 2 (buffer A) to a concentration of 5 ϫ 10 5 /ml, and incubated at 37°C for 20 min. For experiments examining the effect of anti-␣ 5 or anti-␣ V mAbs on cell attachment, cells were preincubated with mAbs (10 -50 g/ml) at 37°C for 20 min. Assays were performed in 96-well microtiter plates (Costar, High Wycombe, UK). Wells were coated for 60 min at room temperature with 100-l aliquots of ligands diluted with Dulbecco's PBS. Sites on the plastic for nonspecific cell adhesion were then blocked for 40 -60 min with 100 l of 10 mg/ml heat-denatured BSA. The BSA was removed by aspiration and the wells were washed once with buffer A. 100-l aliquots of the cells were then added to the wells and incubated for 25-30 min at 37°C in a humidified atmosphere of 5% (v/v) CO 2 . To estimate the reference value for 100% attachment, cells in quadruplicate wells coated with poly-L-lysine (Sigma) (500 g/ml) were fixed immediately by direct addition of 100 l of 5% (w/v) glutaraldehyde for 30 min at room temperature. Loosely adherent or unbound cells from experimental wells were removed by aspiration, the wells washed once with 200 l of buffer A, and the remaining bound cells were fixed as described above for reference wells. The fixative was aspirated, the wells were washed twice with 200 l of PBS, and attached cells were stained with Crystal Violet (Sigma) as described previously (51). The absorbance of each well at 570 nm was then measured using a multiscan enzyme-linked immunosorbent assay reader (Dynatech). Each sample was assayed in quadruplicate, and attachment to BSA (Ͻ2% of the total) was subtracted from all measurements. Each experiment shown is representative of at least three separate experiments.

RESULTS
The First Three NH 2 -terminal repeats of the ␣ Subunit Determine the Specificity of Ligand Recognition-To identify the regions of the ␣ 5 subunit that determine the ligand binding specificity of ␣ 5 ␤ 1 , chimeric ␣ V /␣ 5 subunits were generated in which segments of ␣ V were removed and replaced with the corresponding homologous region of ␣ 5 . Previous studies have suggested that sites important for ligand recognition by ␣ 5 lie in either NH 2 -terminal repeats 1-3 (35,36) or in the divalent cation binding repeats 4 -7 (29). Therefore, initially two chimeras were constructed. The first, ␣ V /␣ 5 (F1-G232), 2 contained repeats 1-3 of ␣ 5 with the remainder of the subunit being ␣ V . The second chimera, ␣ 5 /␣ V (F1-G223), was complementary to the first in that it contained repeats 1-3 of ␣ V with the remainder of the subunit having the sequence of ␣ 5 (Fig. 1). Note that the ␣ V /␣ 5 (F1-G232) chimera contains the divalent cation binding sites of ␣ V (in repeats 4 -7), while the ␣ 5 /␣ V (F1-G223) chimera contains the divalent cation-binding sites of ␣ 5 .
Chimeric and wild-type subunits were expressed on the surface of ␣ 5 -deficient Chinese hamster ovary cells (B2 variant, Ref. 50). Stably transfected cell lines were obtained by G418 selection and dilution cloning. Using immunoprecipitation of cell lysates (not shown) each wild-type or chimeric ␣ subunit was found to be associated with a ϳ130-kDa ␤ subunit, which is the expected size for hamster ␤ 1 . Since ␤ 3 is not expressed by Chinese hamster ovary cells (52), and no ␤ 5 or ␤ 6 (molecular mass ϳ90 kDa) was found in association with the ␣ subunits, we concluded that the ␣ subunits formed heterodimers solely with hamster ␤ 1 . In agreement with the findings of Takagi and co-workers (53) we found low levels of endogenous ␣ V ␤ 5 on the Chinese hamster ovary-B2 cells. However, this endogenous FIG. 1. Schematic representation of integrin ␣ V /␣ 5 chimeras. The seven NH 2 -terminal repeats are represented by squares, with the connecting loops represented by short rectangles. The COOHterminal portion of each subunit is represented by a long rectangle. Chimeras consist of a backbone of ␣ V (filled squares and rectangles) or ␣ 5 (open squares and rectangles) from which regions have been replaced with the homologous region from the other subunit. In this paper we use the term "repeat" in the sense of structural repeat, each repeat corresponding to one "blade" of the ␤-propeller. The structural repeats are offset relative to the sequence repeats, with the first 14-amino acid residues forming the last part of the seventh structural repeat (11). Not drawn to scale.
The results (Figs. 2-4) showed that cells expressing the ␣ V /␣ 5 (F1-G232) chimera exhibited a strong dependence on the presence of the synergy site for adhering to fibronectin, similar to that observed for cells expressing wild-type ␣ 5 . Cells expressing ␣ V /␣ 5 (F1-G232) showed much higher levels of attachment to the *CRGDGWC* sequence than to *CRGDGRC*, and gained the ability to attach to *CRRETAWAC*, similar to the results obtained with cells expressing wild-type ␣ 5 . In contrast, cells expressing ␣ 5 /␣ V (F1-G223) showed only a weak dependence on the presence of the synergy region for attaching to fibronectin, comparable to that observed for cells expressing wild-type ␣ V . Cells expressing this chimera showed approximately equal levels of attachment to *CRGDGWC* and *CRG-DGRC*, and lacked the ability to attach to *CRRETAWAC*, similar to the results obtained for cells expressing wild-type ␣ V .
In summary, the results showed that ␣ V /␣ 5 (F1-G232)␤ 1 had the same ligand-binding specificity as ␣ 5 ␤ 1 , while ␣ 5 /␣ V (F1- G223)␤ 1 had the same ligand binding specificity as ␣ V ␤ 1 . Hence, ligand binding specificity appears to be determined by sequences within the first three NH 2 -terminal repeats of the ␣ subunit. Since the ␣ V /␣ 5 (F1-G232) chimera contains the divalent cation-binding sites of ␣ V (in repeats 4 -7), and the ␣ 5 / ␣ V (F1-G223) chimera contains the divalent cation-binding sites of ␣ 5 , it can be concluded that the divalent cation-binding sites do not play a role in determining the ligand binding specificity.
Cells expressing chimeric or wild-type subunits were analyzed for expression of the epitopes of anti-␣ 5 and anti-␣ V mAbs using flow cytometry (Table I). The results showed that the ␣ V /␣ 5 (F1-G232) subunit contained the epitopes of the functionblocking anti-␣5 mAbs JBS5, SAM-1, SAM-2, 16, and P1D6, lacked the epitopes of function-blocking ␣ V mAbs, but expressed the epitopes of the non-function blocking anti-␣ V mAbs P3G8 and LM142. Conversely, the ␣ 5 /␣ V (F1-G223) subunit contained the epitopes of the function blocking anti-␣ V mAbs 14D9.F8, 17E6, and 69.6.5, lacked the epitopes of function blocking ␣ 5 mAbs, but expressed the epitopes of the non-function blocking anti-␣ 5 mAbs 11 and VC5. These results show that the epitopes of function blocking mAbs anti-␣ 5 mAbs lie within the first three NH 2 -terminal repeats of ␣ 5 ; similarly the epitopes of function blocking anti-␣ V mAbs lie within repeats 1-3 of ␣ V . In contrast, the epitopes of non-function blocking mAbs lie outside the first three repeats. Because the level of epitope expression was similar to that of wild-type ␣ V or wildtype ␣ 5 subunits, it is likely that the chimeras were folded correctly.
Identification of a Minimal Region of the ␣ 5 Subunit That Confers Ligand Binding Specificity-Since cells expressing the ␣ V /␣ 5 (F1-G232) chimera had the same ligand binding properties as cells expressing wild-type ␣ 5 , we attempted to narrow down the region of ␣ 5 required to switch ligand binding specificity by constructing ␣ V /␣ 5 chimeras containing smaller segments of ␣ 5 (see Fig. 1). A chimera ␣ V /␣ 5 (A107-G232), which contains essentially the second and third repeats of ␣ 5 , was expressed at high levels. FACS analysis with a panel of anti-␣ 5 or anti-␣ V mAbs showed that the ␣ V /␣ 5 (A107-G232) chimera retained the epitopes of all the function blocking anti-␣ 5 mAbs tested (Table I), although the binding of JBS5 was decreased compared with the ␣ V /␣ 5 (F1-G232) chimera. Chimeras containing only the second or third repeat of ␣ 5 (␣ V /␣ 5 (A107-C164) and ␣ V /␣ 5 (C164-G232)) were expressed at lower levels than ␣ V /␣ 5 (A107-G232), and failed to react, or reacted only weakly, with function blocking anti-␣ 5 and ␣ V mAbs (data not shown).
Since it could not be demonstrated that these two chimeras were folded correctly, they were not studied further.
Cells expressing the ␣ V /␣ 5 (A107-G232) chimera were tested for their ability to attach to ␣ 5 ␤ 1 -or ␣ V ␤ 1 -selective ligands (Fig.  5). The results showed that cells expressing ␣ V /␣ 5 (A107-G232) showed a strong dependence on the synergy region for binding to the CCBD of fibronectin, displayed preferential recognition of RGDGW over RGDGR, and possessed the ability to attach to RRETAWA. Therefore, amino acid residues 107-232 of ␣ 5 were found to be sufficient to confer on ␣ V ␤ 1 the ligand binding specificity of ␣ 5 ␤ 1 . In addition, since the amino acid sequence Asp 227 -Gly 232 of ␣ 5 is identical to the corresponding sequence Asp 218 -Gly 223 in ␣ V , residues that confer ligand binding specificity must be contained within Ala 107 -Trp 226 of ␣ 5 .
failed to attach to *CRRETAWAC* (Figs. 8D and 9D). In striking contrast, cells expressing the ␣ V /␣ 5 (D154 -A159) chimera showed strong selectivity for *CRGDGWC*, and possessed the ability to attach to *CRRETAWAC*, identical to the results for cells expressing wild-type ␣ 5 (Figs. 8B and 9B, compare Figs. 3C and 4C). Therefore, these results showed that replacing Gln 145 -Asp 150 of ␣ V with Asp 154 -Ala 159 of ␣ 5 conferred on ␣ V ␤ 1 two of the ligand binding properties of ␣ 5 ␤ 1 : selectivity for the RGDGW sequence and the ability to recognize RRETAWA.

DISCUSSION
In this report we have sought to identify the regions of the integrin ␣ subunit that are involved in ligand recognition using ␣ V /␣ 5 chimeras. Our major findings are as follows: (i) the first three NH 2 -terminal repeats contain the epitopes of function blocking anti-␣ subunit mAbs and the amino acid sequences that determine ligand binding specificity; (ii) the divalent cation-binding sites (in repeats 4 -7) do not determine the specificity of ligand recognition; (iii) the amino acid sequence Ala 107 -Trp 226 of ␣ 5 (corresponding approximately to the second and third repeats) is sufficient to confer on ␣ V ␤ 1 the ligand binding properties of ␣ 5 ␤ 1 ; (iv) swapping a 6-amino acid sequence from a predicted loop region of ␣ V with the corresponding region of ␣ 5 (Asp 154 -Ala 159 ) is sufficient to confer on ␣ V ␤ 1 selectivity for RGDGW and recognition of RRETAWA.
In this study we observed a close correspondence between the regions of the ␣ subunit involved in determining the specificity of ligand recognition and those that contained the epitopes of function blocking mAbs. This finding supports previous evidence that the epitopes of function blocking mAbs are proximal to sites involved in ligand binding (5,(33)(34)(35)(36)54). In addition, these data lend strong support to the ␤-propeller model, which predicts that the ligand-binding sites lie on the upper face of the ␤-propeller (11). Recently, we have mapped some of the residues that form part of the JBS5, mAb 16, and P1D6 epitopes by substituting residues in human ␣ 5 with the corresponding residues from mouse ␣ 5 . Ser 85 was found to contribute to the JBS5 epitope, Glu 126 and Leu 128 to the mAb 16 epitope, and Leu 212 to the P1D6 epitope (8). All these residues are predicted to lie on the upper face of the ␤-propeller domain. The results of the present study, including decreased binding of JBS5 to the ␣ V /␣ 5 (A107-G232) chimera, are in good agreement with these data.
Two loop swapping mutants, ␣ V /␣ 5 (K125-D130) and ␣ V / ␣ 5 (D154 -A159), failed to react with the function blocking anti-␣ V mAb 69.6.5. However, since they did react with two other function blocking anti-␣ V mAbs it is likely that the structure of the NH 2 -terminal repeats had not been grossly perturbed. Hence, a probable explanation is that these mutations disrupted the 69.6.5 epitope. Therefore, residues that form part of the 69.6.5 epitope probably lie in the 2-3 loop of repeat 2 and in the 4-1 loop between repeats 2 and 3. These two loops are predicted to lie adjacent to each other on the upper face of the ␤-propeller. The observation that the ␣ V /␣ 5 (D154 -A159) chimera bound ␣ 5 ␤ 1 -selective ligands but retained the epitopes of 17E6 and 14D9.F8 is consistent with previous findings that the epitopes of most function blocking mAbs are close to but not directly overlapping with sites involved in ligand recognition (5,8,33,54).
It is intriguing that the ␣ V /␣ 5 (Y186 -F187) chimera was expressed only at very low levels, whereas the ␣ V /␣ 5 (Y186 -F187, Q218 -Y226) chimera, in which residues from two loops were exchanged, was expressed normally. The reasons for this are unclear. However, it is likely that there is an interaction between these two loops since they are predicted to lie adjacent to each other in the ␤-propeller model (11). Mutation of Phe 177 -Tyr 178 of ␣ V could have a destabilizing effect on the tertiary structure, leading to loss of expression at the cell surface. An interaction between the two exchanged loops in the ␣ V / ␣ 5 (Y186 -F187, Q218 -Y226) chimera may compensate for this destabilizing effect, thereby allowing normal levels of expression.
Although our results show that the divalent cation sites in the ␣ subunit do not influence the specificity of ligand binding, it is possible that the divalent cation binding repeats of ␣ 5 and ␣ V could have an identical function in ligand recognition. For example, these repeats may be involved in interacting with the RGD sequence, which is a property common to both ␣ 5 ␤ 1 and ␣ V ␤ 1 . However, it is noteworthy that none of the function blocking mAbs mapped to repeats 4 -7. Indeed, it has been shown that a mAb directed against the ␣ L subunit whose epitope includes the divalent cation-binding site of the fifth repeat has no effect on ligand recognition by ␣ L ␤ 2 (7). Additionally, in a study of ␣ 5 /␣ 6 chimeras it was found that swapping parts of the divalent cation binding repeats of ␣ 5 with the corresponding regions of ␣ 6 (which does not recognize RGD) did not perturb recognition of RGD or the synergy sequence (55). Taken together with evidence that mutations at the divalent cation-binding sites do not affect ligand binding (56), we favor the hypothesis that these sites have a purely regulatory role. Finally, there is strong evidence that the ␤ subunit, rather than the ␣, contains the major site of RGD recognition (5,20,(22)(23)(24)(25)(57)(58)(59)(60).
In two reports, recombinant proteins containing repeats 4 -7 of ␣ 5 or ␣ IIb have been shown to support RGD-dependent ligand binding (28,29). Importantly, however, it has yet to be demonstrated whether the specificity and affinity of ligand binding of these recombinant fragments is the same of that of the native integrin. Cross-linking of ligand peptides to ␣ IIb ␤ 3 and ␣ V ␤ 3 also indicated that these peptides interacted (at least in part) with the divalent cation binding repeats (26,27). Further studies will be required to resolve these inconsistencies, and to show if the findings for ␣ 5 ␤ 1 in this report can be generalized to all non-A-domain containing integrins. However, the position of ligand recognition sites in the ␣ 5 subunit deduced here (the second and third repeats) is strikingly similar to the location of the A-domain in ␣ subunits that contain this module (between the second and third repeats). This observation suggests an attractive general model of receptor-ligand interactions in both A-domain containing and non-A-domain containing integrins (61,62). Our findings are also consistent with previous mutagenesis data for ␣ 5 ␤ 1 and other integrins (33)(34)(35)(36)(37)(38)(39).
In a study of ␣ V /␣ IIb chimeras (4), the minimal region of the ␣ IIb subunit required to switch the ligand binding specificity to that of ␣ IIb was found to be Leu 1 -Pro 334 (corresponding approximately to the first five repeats). A chimera containing only the first 3 repeats of ␣ IIb (Leu 1 -Phe 223 ) expressed the epitope of a function blocking anti-␣ IIb ␤ 3 mAb but, in contrast to the results from the present study, this chimera did not interact with ␣ IIb ␤ 3 -specific ligand mimetic mAbs or a peptidomimetic. The reason for this discrepancy is unclear. However, a possible explanation is that the ␣ V /␣ 5 (F1-G232) and ␣ 5 /␣ V (F1-G223) chimeras used here contain all of the putative loop between repeats 3 and 4, whereas the ␣ V /␣ IIb (L1-F223) chimera lacks most of this loop. Mutation of Asp 224 in this loop of ␣ IIb to Val has been shown to lead to loss of ligand-binding function by ␣ IIb ␤ 3 (39). Importantly, in the ␣ V /␣ IIb (L1-F223) chimera Asp 224 of ␣ IIb is replaced by Thr 210 of ␣ V . Hence, the ␣ V /␣ IIb (L1-F223) chimera may lack one or more additional residues necessary for ligand binding specificity, whereas these residues are contained in the ␣ V /␣5(F1-G232) and ␣ 5 /␣ V (F1-G223) chimeras.
In the same study of ␣ V /␣ IIb chimeras (4), it was shown that the ␣ V /␣ IIb (R140 -P334)␤ 3 receptor did not interact with ␣ IIb ␤ 3specific ligands. However, this chimera is missing much of the second repeat of ␣ IIb . In contrast, the ␣ V /␣ 5 (A107-G232) chimera used here contains nearly all of the second repeat of ␣ 5 . Therefore, we propose that ␣ V /␣ IIb (R140 -P334) lacks residues in the second repeat required for ligand binding specificity. In further support of our finding that the second and third NH 2terminal repeats determine the specificity of ligand recognition, these repeats show the greatest degree of sequence divergence between ␣ subunits. In contrast, the divalent cationbinding repeats are more similar in sequence (7,63).
None of loop swapping mutations led to a "gain" of synergy sequence binding (i.e. a change from the weak dependence on the presence of the synergy sequence for binding to fibronectin seen for ␣ V ␤ 1 to the strong dependence seen for ␣ 5 ␤ 1 ). It is possible that the residues involved in high affinity recognition of the synergy sequence lie outside the regions exchanged. Alternatively, however, the overall fold of second and third repeats may determine the shape of the fibronectin-binding pocket. It is noteworthy that in addition to ␣ 5 ␤ 1 recognizing the synergy sequence of fibronectin more strongly than ␣ V ␤ 1 , it also recognizes the RGD site more weakly. Both these characteristics were evident in the ␣ V /␣ 5 (A107-G232) chimera. Interestingly, this chimera retained all the epitopes of the function blocking anti-␣ 5 mAbs; thus the second and third NH 2 -terminal repeats appear to have the same conformation as those of wild-type ␣ 5 . We therefore propose that the overall fold of the FIG. 6. Alignment of human ␣ 5 and ␣ V sequences for NH 2 -terminal repeats 2-4 (R2-R4). Sequences were aligned using Clustal W. Predicted ␤-strands in ␣ 5 and ␣ V are underlined, and the sequence Ala 107 -Tyr 226 of ␣ 5 is shown in bold. Residues in putative loop regions of ␣ V between the second and third ␤-strands, and between the fourth and first ␤-strands (shown boxed) were chosen for swapping with the corresponding residues in ␣ 5 . Assignment of ␤-strands and loops is based on an alignment of the sequence of human ␣ 5 with that of human ␣ 4 by Irie et al. (6).  second and third repeats determines the shape of the fibronectin-binding pocket, and slight differences in the shape of this pocket between ␣ V ␤ 1 and ␣ 5 ␤ 1 may lead to the differential recognition of the synergy and RGD sequences by these two integrins. Support for the suggestion that the overall fold of repeats 2-3 is important comes from our finding that ␣ V /␣ 5 chimeras containing only the second repeat, or only the third repeat, of ␣ 5 appeared to be misfolded.
Based on the structure of the central cell-binding domain of fibronectin (64,65), the recognition site of synergy sequence on the ␣ 5 subunit is predicted to lie a distance of ϳ35 Å away from the RGD-binding site. As we show in an accompanying paper (66), the loop between the second and third repeats of ␣5 lies within ϳ7 Å of the RGD-binding site; this places constraints on possible locations of the RGD and synergy sequence recognition sites on ␣ 5 ␤ 1 . Based on these constraints, the divalent cation- binding sites are too far away from this loop to participate in RGD binding. However, an interaction of the RGD sequence with the MIDAS site of the A-domain-like region of the ␤ subunit is possible if this domain lies adjacent to repeats 2 and 3 of the ␣ subunit, as previously proposed (5). Two potential positions for the synergy sequence recognition site on the ␣ 5 subunit are close to the center of the ␤-propeller, or near the junction of repeats 3 and 4. In support of the latter suggestion, the epitope of P1D6, an anti-␣ 5 mAb that blocks recognition of the synergy region but not the RGD sequence (5), has been shown to contain Leu 212 in repeat 3 (8). 3 In summary, we have defined a minimal domain of the ␣ 5 subunit (Ala 107 -Trp 226 ) that determines the specificity of ligand binding, and identified a sequence that plays a critical role (Asp 154 -Ala 159 ). Our findings are consistent with the ␤-propel- 3 In support of this finding, the chimera containing only the third repeat of ␣ 5 (␣ V /␣ 5 (C164 -G232)) was weakly reactive with P1D6 but not with other function-blocking anti-␣ 5 mAbs (A. P. Mould, unpublished results). ler model and appear to exclude a function for the divalent cation binding repeats in ligand recognition. In the accompanying paper (66) we determine further the molecular basis of the interaction of ␣ 5 ␤ 1 with RGDGW and RRETAWA.
Acknowledgments-We thank K. Yamada and S. Aota for the fibronectin III 6 -10 clones, the human ␣ 5 clone, and mAbs 11, 13, and 16. We are grateful to D. Cheresh for the human ␣ V clone and for mAb LM142, S. Goodman for mAbs 17E6 and 14D9.F8, R. Juliano for Chinese hamster ovary-B2 cells, Y. Takada for Chinese hamster ovary-B2 cells transfected with human ␣ 5 , and L. Hall and R. Slater for DNA sequencing.
Addendum-While this manuscript was under review, a publication from Banères and co-workers (Banères J. L., Roquet, F., Martin, A., and Parello, J. (2000) J. Biol. Chem. 275, 5888 -5903) presented new data showing that a recombinant fragment of ␣ 5 spanning the divalent cation-binding repeats can interact with fibronectin. The fragment lacks the second and part of the third repeat of ␣ 5 , which we have identified here as playing an essential role in ligand recognition. Further analysis of the specificity and affinity of ligand binding, and of the extent to which such "minimized" integrins reproduce the functions of the native receptor, is clearly required to resolve this apparent discrepancy. In a separate publication (Puzon-McLaughlin, W., Kamata, T., and Takada ligand-mimetic mAbs against ␣ IIb ␤ 3 have been shown to be contained within the second and third repeats of ␣ IIb and the A-domain-like region of ␤ 3 . Our data are fully consistent with this latter study.