A Minimized Human Integrin α5β1 That Retains Ligand Recognition

Two isolated recombinant fragments from human integrin α5β1 encompassing the FG-GAP repeats III to VII of α5 and the insertion-type domain from β1, respectively, are structurally well defined in solution, based on CD evidence. Divalent cation binding induces a conformational adaptation that is achieved by Ca2+ or Mg2+ (or Mn2+) with α5 and only by Mg2+ (or Mn2+) with β1. Mn2+ bound to β1 is highly hydrated (∼3 water molecules), based on water NMR relaxation, in agreement with a metal ion-dependent adhesion site-type metal coordination. Each fragment saturated with Mg2+ (or Mn2+) binds a recombinant fibronectin ligand in an RGD-dependent manner. A conformational rearrangement is induced on the fibronectin ligand upon binding to the α5, but not to the β1 fragment, based on CD. Ligand binding results in metal ion displacement from β1. Both α5 and β1 fragments form a stable heterodimer (α5β1 mini-integrin) that retains ligand recognition to form a 1:1:1 ternary complex, in the presence of Mg2+, and induces a specific conformational adaptation of the fibronectin ligand. A two-site model for RGD binding to both α and β integrin components is inferred from our data using low molecular weight RGD mimetics.

are a family of structurally and functionally related adhesion receptors that participate in cell-cell and cellextracellular matrix (EM) interactions (1). All integrins are heterodimers of non-covalently associated ␣ and ␤ subunits. At the functional level, the interactions between integrins and their EM protein ligands involve the following: (i) the extracellular integrin heteromeric "head" that encompasses the Nterminal halves from both ␣ and ␤ subunits and hosts the ligand-recognition pocket with a variety of binding sites on each subunit (2); (ii) short specific amino acid sequences (adhesion motifs) from the EM ligands. The prototype for these adhesion motifs is the Arg-Gly-Asp (RGD) sequence that is present in fibronectin, fibrinogen, vitronectin, and other EMadhesive proteins (3). The exact location of these binding loci in the integrin subunits, as well as their respective role on ligand binding energy and specificity, still remains an open question. The N-terminal half of the integrin ␣ subunits is characterized by the presence of seven N-terminal repeats of about 60 amino acids each (4,5). Some of the ␣ subunits include an insertion domain, or I-domain, about 200 residues in length, between repeats II and III (2). The homologies between repeats I and VII essentially include the FG and GAP consensus sequences, so that these repeats are also referred as FG-GAP repeats (6). Three to four of these repeats (i.e. repeat IV or V to repeat VII) display sequences that resemble the EF-hand consensus sequence found in various divalent cation-binding proteins (7). However, the integrin EF-hand type sequences are systematically devoid of an acidic residue at their relative position 12, a highly invariant Glu residue in the typical EF-hands, that is replaced by a non-polar residue in integrins (8). Isolated recombinant integrin fragments encompassing repeats III to VII of ␣ IIb (9) and IV to VII of ␣ 5 (10) have been shown to mimic the ligand-binding features (divalent cation and RGD dependence) that are observed with native integrin receptors, indicating that the divalent cation-binding domains in the ␣ subunits are part of the ligand-recognition pocket of integrins. Repeats III and IV in the ␣ subunits have also been shown to be involved in cell spreading and in assembly of focal contacts at the cell surface (11).
The N-terminal regions of the integrin ␤ subunits are characterized by a conserved domain that displays sequence homology with the I-domain found in several integrin ␣ subunits (12). This I-type domain in the integrin ␤ subunits includes the following: (i) a functional cation-binding site that displays strong similarities at the level of its metal-coordinating residues with the MIDAS site of the ␣ subunit I-domains (13,14); (ii) a totally conserved DDL motif, close to the MIDAS-type site that is apparently involved in the recognition of RGD-containing ligands (15); (iii) possibly a specific sequence responsible for heterodimer formation; in the case of the human ␤ 3 I-type domain this specific sequence corresponds to the segment ␤ 3 -(275-280) with the unique hexapeptide sequence, VGSDNH, that is responsible for species-restricted ␣␤ heterodimer assembly (16). As tentatively proposed by McKay et al. (16), the sequences in the ␤ subunits that are topologically related to the ␤ 3 -(275-280) sequence are likely to participate in heterodimer assembly for the different integrins.
Divalent cations play a central role in ligand recognition by the integrins (17). Both ␣ and ␤ subunits are responsible for divalent-cation binding (2). In the ␣ subunit, the divalent cation-binding sites are located at the level of the EF-hand type sequences. By using isolated recombinant EF-hand domains from ␣ IIb (9) and ␣ 5 (10), a stoichiometric ratio of four cations bound per protein has been inferred, in agreement with the occurrence of four EF-hand-type sequences in both integrin ␣ subunits (2). These four divalent cation-binding sites are distributed in two classes of sites (each class with two sites), differing by their affinities with the low affinity sites regulating ligand recognition (9,10). Divalent cation binding to the integrin ␤ subunits is also intimately linked to integrin function. Apparently, a single cation-binding site occurs in the ␤ subunits at the level of their MIDAS-type sequences (13,14). Based on model studies, D'Souza et al. (18) proposed that divalent cation binding to the ␤ 3 subunit promotes a ligandcompetent conformation and that the initially protein-bound cation is released after formation of the ligand-receptor complex. This mechanism has not yet been fully assessed with a native receptor or with an isolated ␤ I-type domain. Evaluation by equilibrium dialysis of the number of divalent cation-binding sites in an intact integrin receptor, ␣ IIb ␤ 3 , amounts to a total of 4 -5 (19), in agreement with the aforementioned distribution of 4 sites in the ␣ subunits and 1 site in the ␤ subunits.
So far the respective role of these divalent cation-binding sites, and their possible synergy, on ligand recognition by the integrins have not been fully elucidated. McKay et al. (16) have reported a recombinant ␣␤ heterodimer formed of both segments ␣ IIb -(1-233) and ␤ 3 -(111-318) that binds Arg-Gly-Asp-Trp (RGDW), as a peptide ligand, in the presence of Ca 2ϩ . Interestingly, this minimized ␣ IIb ␤ 3 integrin does not contain any of the ␣ IIb EF-hand cation-binding sites; its ␣ component is represented exclusively by repeats I to III. It is likely that this mini-integrin only contains part of the ligand recognition pocket in the absence of the ligand-competent EF-hand domain from its ␣ IIb subunit. A model has been recently proposed (10) in which the RGD tripeptide sequence from the protein ligand acts as a dual motif with two oppositely charged groups, a negative one (the carboxylate from Asp) and a positive one (the guanidinium from Arg), that interact with distinct structural elements on each of both ␣ and ␤ subunits. Such a model would account for the binding of the RGDW tetrapeptide by the miniintegrin ␣ IIb ␤ 3 , as reported by McKay et al. (16). One possibility for retaining ligand-binding capacity in an RGD-dependent manner could be related to the fact that this ␣ IIb ␤ 3 miniintegrin includes the RGD-binding sequence DDL conserved in all integrin ␤ subunit I-type domains (15). However, the three N-terminal FG-GAP repeats, I to III, that are present in the ␣ component of the ␣ IIb ␤ 3 mini-integrin could also contain ligandbinding sites unidentified so far. As emphasized by Weber (20), protein interactions need to viewed as the convergence of cooperative effects of many inputs rather than a sum of fixed independent influences. In this respect, the RGD adhesive sequence displays two complex charged groups, a guanidinium and a carboxylate (which will give rise to electrostatic as well as hydrogen bonding) and several hydrogen bond acceptors and donors from the tripeptide main chain that will elicit a variety of contacts with elements from the integrin receptor. These elements can be rather interspersed in the amino acid sequences from both ␣ and ␤ subunits. Since McKay et al. (16) expressed their mini-integrin directly as a heterodimeric assembly, a strict delineation of the ligand-binding properties of each subunit fragment, taken separately, as well as of their respective cation requirements, could not be achieved.
To carry out such an analysis, we cloned and expressed in Escherichia coli two recombinant proteins from the ␣ and ␤ subunits of human integrin ␣ 5 ␤ 1 , i.e. ␣ 5 -(160 -448) and ␤ 1 -(121-329). Both fragments form a stable and soluble ␣␤ miniheterodimer that is ligand-competent in an RGD-and cationdependent manner. Based on previous results with a minimal EF-hand domain of ␣ 5 (10), we expressed here a novel ␣ 5 fragment as a more integrated structural unit with regard to its capacity of combining ligand and cation binding. In doing so, we were guided by previous results indicating that the FG-GAP repeat III in ␣ 5 converges toward the ligand-binding pocket of ␣ 5 ␤ 1 (21). Moreover, the ␤ subunit I-domain is likely to lie close to repeat III of the ␣ subunits in native integrin receptors, indicating that repeat III participates to heterodimer stability (22). Besides the whole EF-hand domain (repeats IV through VII), for which cation-dependent ligand binding has been previously demonstrated (10), the ␣ 5 component of our minimized human integrin ␣ 5 ␤ 1 also includes the totality of repeat III, whereas the ␤ 1 component of this mini-integrin corresponds to the totality of the I-type domain. We establish here through a detailed study of the binding to ␣ 5 -(160 -448) and ␤ 1 -(121-329) of (i) divalent cations (Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ ), (ii) RGD-containing ligands (fibronectin fragments and synthetic peptides), and (iii) simple compounds mimicking the RGD motif (n-butyrate and guanidinium chloride) that these isolated integrin fragments taken separately display cation-dependent ligand-binding properties that are closely related to those observed with the ␣ 5 -(160 -448)⅐␤ 1 -(121-329) mini-integrin assembly itself. Cloning, Expression, and Purification of ␣ 5 -(160 -448)-The DNA sequence coding for residues 160 -448 of human ␣ 5 was produced by PCR using the full-length ␣ 5 cDNA (25) as a template. The following oligonucleotides were used as primers: 5Ј-GCA CGT CAT ATG GGC CAG ATC CTG TCT GCC-3Ј and 5Ј-CG GGATCC TCA CCC CTC TAA GCT GCA GCT-3Ј. These primers were designed to generate an NdeI restriction site at the 5Ј end, and an in-frame stop codon, as well as a BamHI restriction site, at the 3Ј end. Amplification was performed in a Perkin-Elmer thermal cycler using Taq polymerase (Perkin-Elmer) and proceeded through a cycle of denaturation at 94°C (1 min), annealing at 55°C (1 min), and extension at 72°C (2 min), for a total of 30 cycles. The PCR fragment was purified using the Qiaquick gel extraction kit (Qiagen, Chatsworth, CA), digested with NdeI and BamHI, and inserted into the pET15b vector (Novagen, Madison, WI). This vector allows expression of the corresponding sequence fused with a HisTag N-terminal segment that includes the hexahistidinyl segment, as well as the thrombin cleavage site. The clone obtained was confirmed by dideoxy sequencing (26). E. coli BLR(DE3) host strain (Novagen) was transformed with the pET15b-␣ 5 -(160 -448) vector. The corresponding recombinant protein was purified from the insoluble fraction obtained after bacterial lysis, as described previously for ␣ 5 -(229 -448) (10). The His-tagged protein was digested with thrombin under conditions similar to those described for ␣ 5 -(229 -448) (10) to remove the oligohistidine N-terminal appendix. The protein obtained includes the extraneous N-terminal tetrapeptide GSHM resulting from thrombin cleavage.

Materials and Buffers-Buffer
Cloning, Expression, and Purification of ␤ 1 -(121-329)-The DNA sequence coding for residues 121-329 of human ␤ 1 (25) was produced by PCR using a human cDNA library (CLONTECH, Palo Alto, CA) as a template. The following oligonucleotides were used as primers: 5Ј-GCA CGT CCA TGG TAT CCC ATT GAC CTC TAC CTT-3Ј and 5Ј-CG GGA TCC CTA GGA AAG GGA ATT GTA TGC ATC AAT-3Ј. These primers were designed to generate an NcoI restriction site at the 5Ј end and an in-frame stop codon, as well as a BamHI restriction site, at the 3Ј end.
Amplification was performed in a Biometra thermal cycler (Whatman) using Pfu polymerase (Stratagene, La Jolla, CA) and proceeded through a cycle of denaturation at 94°C (45 s), annealing at 53°C (45 s), and extension at 72°C (2 min), for a total of 30 cycles. The PCR fragment was purified using the Qiaquick gel extraction kit (Qiagen, Chatsworth, CA), digested with NcoI and BamHI, and inserted into the pET32a vector (Novagen, Madison, WI). This vector allows expression of the corresponding sequence fused with thioredoxin (Trx) and a hexahistidinyl segment at the N terminus of the target protein. The clone obtained was confirmed by dideoxy sequencing (26). The E. coli AD494(DE3) host strain (Novagen) was transformed with both the pET32b-␤ 1 -(121-329) expression vector and a pET9a vector containing the GroESL-coding region. Cultures of the transformed strain were grown overnight at 30°C in Terrific Broth medium containing 100 g/ml ampicillin and 50 g/ml kanamycin, diluted 1 in 100 into the same culture medium, and incubated with shaking at 30°C to an absorbance of 0.8 at 600 nm. Induction of protein expression was achieved by addition of IPTG to a final 1 mM concentration and further incubation for 4 h at 30°C. Cells were pelleted by centrifugation, resuspended in 25 mM Tris-HCl, 500 mM KCl, 5 mM imidazole, 0.005% (v/v) Triton X-100, 0.5 mM PMSF at pH 8, and lysed by sonication. The suspension was centrifuged (27,000 ϫ g for 30 min), and the supernatant was recovered. The DNA was precipitated with protamine sulfate (27), and the protein supernatant was dialyzed against 12.5 mM sodium borate, 150 mM KCl, 0.2 mM PMSF, 10% glycerol, pH 8.3, and loaded onto a Ni-NTA agarose column (Qiagen) previously equilibrated in the same buffer. TrxTag-␤ 1 -(121-329) was eluted with a linear imidazole gradient (0 -200 mM). The protein was eluted at ϳ150 mM imidazole.
Enterokinase Cleavage-The Trx-tagged ␤ 1 -(121-329) protein was dialyzed against 20 mM Tris-HCl at pH 8 containing 150 mM NaCl and 2.5 mM CaCl 2 . Digestion was carried out at 21°C at an enzyme-tosubstrate ratio of 10 units of enterokinase (Sigma) per mg of cleavable protein. The reaction was stopped by the addition of PMSF to a final concentration of 2 mM. The time needed for an efficient cleavage was determined by a time course SDS-PAGE analysis (not shown). The digestion mixture was dialyzed against 20 mM Tris-HCl, 400 mM KCl, 2 mM PMSF, 10% glycerol at pH 8 and loaded onto a Ni-NTA column. The unbound protein was dialyzed against 12.5 mM sodium borate, 150 mM KCl, 0.2 mM PMSF, pH 8.3, and loaded onto a DEAE-Trisacryl column (IBF, France). Elution was carried out with a linear KCl gradient (0.15 to 1.5 M KCl). The protein obtained includes the extraneous N-terminal dipeptide AM resulting from the cleavage by enterokinase. Desalting was carried out on a Sephadex G-50 column (Amersham Pharmacia Biotech) equilibrated in buffer A.
Molecular Weight Determinations by Mass Spectrometry-The purified proteins were analyzed by electrospray ionization-mass spectrometry using a Fisons VG Trio2000 mass spectrometer equipped with an ion spray source. Molecular weights were determined using the electrospray ionization deconvolution software from Fisons.
Amino Acid Sequencing-N-terminal amino sequencing was carried out by automated step-by-step Edman degradation using a gas phase amino acid sequencer with an on-line PTH analyzer (Applied Biosystems Inc.), as described by Speicher (28). Ten steps were performed.
Protein Concentrations-UV absorbance measurements at 276 nm (Cary 118 UV spectrometer) were used to determine accurate protein concentrations of the stock solutions. The molar absorptivities of the different recombinant proteins used throughout this work were determined by the procedure of Gill and von Hippel (29). Molar absorptivities (expressed in liters M Ϫ1 ⅐cm Ϫ1 ) at 276 nm are 28,600 and 21,350 for the fully Mg 2ϩ -loaded forms of ␣ 5 -(160 -448) and ␤ 1 -(121-329), respectively. No real difference was observed between the Mg 2ϩ -loaded proteins and the corresponding apoforms with regard to their UV molar absorptivities at 276 nm (in contrast with the CD effects reported under "Results"). The molar UV absorptivities for the Fn ligands used in this work had been previously determined (10). In general, no aggregation effects were observed in the 300 -390 nm region of the UV spectra for the different proteins used in this work (if any, this was usually discarded through ultracentrifugation). By adopting the borate buffer A, at pH 8.3, aggregation effects were negligible with the integrin apoforms, whereas this is not totally the case for ␣ 5 -(160 -448) at lower pH values.
Preparation of the Protein Apoforms-The apoforms of ␣ 5 -(160 -448) and ␤ 1 -(121-329) were prepared by precipitation of the fully Mg 2ϩloaded forms using trichloroacetic acid 3% (v/v) as a precipitating agent, as described previously in the case of the EF-hand Ca 2ϩ /Mg 2ϩ -binding parvalbumin (30). The precipitated protein was washed once in 3% trichloroacetic acid and twice with distilled water with no apparent loss of protein in the supernatant. Although the metal content in the apoforms was not directly measured, the CD experiments reported in this work show that these forms are practically metal-free. Metal-free distilled water was used (Ca 2ϩ impurities estimated in the 2 M range by flame spectrophotometry).
Circular Dichroism Measurements-CD spectra were recorded at 25°C with a dichrograph CD6 (Jobin Yvon, Paris, France). The spectra are the average of five scans recorded using a bandwidth of 2 nm, a step-width of 0.1 nm, and a 0.5 s averaging time per point. Cell path lengths (Hellma, France) were 1.00 Ϯ 0.01 mm (far-UV region, wavelengths Ͻ250 nm) or 10.00 Ϯ 0.01 mm (near-UV region, wavelengths Ͼ250 nm). Protein concentrations in the 10 M range were used for all experiments. The UV absorption spectra (390 to 230 nm) of the protein samples were systematically recorded before the CD measurements and showed no contribution of aggregation effects (see under "Protein Concentrations"). All CD measurements were carried out in the borate buffer A at pH 8.3 which is optically transparent within all the wavelength range instrumentally accessible. Since pH values above 7-8 give rise to formation of hydroxides with Mn 2ϩ solutions, the CD experiments using this divalent cation were carried out in cacodylate buffer B at pH 6.0. We verified using the Mg 2ϩ -loaded proteins in either buffer A or B that changing the pH did not affect the CD spectra. However, we preferentially used the borate buffer A in most CD experiments since (i) the apoform of ␣ 5 -(160 -448) was found to be more soluble at basic pH values than at lower pH values and (ii) in contrast to the cacodylate buffer B, the borate buffer A offered the possibility to carry out CD measurements at the lowest wavelength accessible in the 185-190 nm range. All CD experiments were carried out in duplicate (the observed variations between two independent experiments are illustrated in Fig.  2C). Absolute ellipticity values were inferred from standardization with d10-camphorsulfonic acid (Fluka, Switzerland) according to the procedure of Venyaminov and Yang (31). Molar ellipticities are given in degree⅐cm 2 dmol Ϫ1 of mean amino acid residue (M r ϭ 115).
Secondary Structure Predictions-Deconvolution of the far-UV CD profiles was carried out using a standard computer procedure (32). The CD data used in the deconvolution included the far-UV range 190 -250 nm (spectra in Fig. 1, A and C, only display data down to 195 nm). Although the signal-to-noise ratio in the 190 -195 nm region is less favorable than at higher wavelengths, a complete set of the far-UV data down to 190 nm was systematically included in the deconvolution procedure. It is known that the shape of the CD spectra of proteins in the low wavelength range is important for correctly predicting the content of ␣-helical residues (31). Besides the strong double minimum at 222 and 208 -210 nm, the ␣-helix displays a stronger maximum at 191-193 nm (31).
PRE Measurements-Water proton T 1 relaxation times were measured at 100 MHz (Bruker spectrometer AC100) and 23°C by the standard inversion-recovery (180°--90°) method with ␤ 1 -(121-329) solutions in buffer B at constant protein concentration (0.1 mM) and at Mn 2ϩ concentrations ranging from 0.1 to 1 mM (metal-cation titration). To avoid isotope effects on the T 1 measurements, no D 2 O was added to the NMR samples so that the T 1 measurements were carried out in the absence of frequency field lock. The reproducibility of our measurements was checked by comparing the curves R 1 (see definition below) versus [Mn 2ϩ ] determined at different times with standard Mn 2ϩ solutions. The relative variations in slope of these straight lines did not exceed a few percent. No attempt was made to deoxidize the solutions (according to a previous study with parvalbumin, the paramagnetic effects of dissolved oxygen are small, although not totally negligible (37)). Relaxation rates, R 1 (in s Ϫ1 ), are defined as the inverse of the longitudinal relaxation times T 1 . The dimensionless enhancement factor ⑀ corresponds to the ratio (R 1 * Ϫ R 1 *°)/(R 1 Ϫ R 1°) , where the exponents * and°denote the presence of protein and the absence of metal cation, respectively (38). For the aquocation itself, ⑀ is identical to 1. The characteristic enhancement factor of the metal-macromolecule complex is defined by ⑀ b ϭ lim ⑀ at [P] ϭ ϱ and directly depends on the hydration of the paramagnetic cation in the protein-bound state (defined by q*, i.e. the number of water molecules surrounding the proteinbound cation) as well as on the dynamics and kinetics of the water molecules surrounding the paramagnetic cation (38). Upon binding of the paramagnetic cation to the macromolecule, the number of water molecules surrounding the cation decreases, i.e. q* Ͻ q (for aqueous Mn 2ϩ q ϭ 6 (39)). If this were the only parameter to change, then there would be a decrease in the relaxation rate of the water protons in the presence of protein, resulting in ⑀ b Ͻ1. However, due to the increase in the dipolar correlation times c of the water protons upon binding to the macromolecule and/or a decrease in the chemical exchange lifetimes M * of these protein-bound water molecules, it is usually found that ⑀ b Ͼ1 (enhancement). Under fast exchange conditions on the NMR time scale, i.e. M * Ͻ Ͻ T 1,M *, ⑀ b is independent on M *. The intrinsic enhancement factor ⑀ b then corresponds to the ratio (q*/q)⅐(f 1 ( c *)/f 1 ( c )), assuming that the metal-cation to water-proton distance remains unchanged from that in aqueous solution (38). Since f 1 ( c *) Ͼ f 1 ( c ) with macromolecules, then ⑀ b Ͼ1.
Protein Interaction Assays-A protein affinity chromatography assay was used to evaluate the binding of the integrin recombinant fragments to the fibronectin ligand. The assay includes two steps, a complex formation step in solution and an immobilization step on a Ni-NTAagarose support (see Fig. 5A). In the first step, HisTag-␣ 5 -(160 -448), TrxTag-␤ 1 -(121-329), or the HisTag-␣ 5 -(160 -448)⅐␤ 1 -(121-329) binary complex were mixed with the different recombinant fibronectin fragments (3Fn7-10 or 3Fn10 -11) at a 1:1 molar ratio (for buffer conditions, see legends to Figs. 5, 6, and 8) with protein concentrations in the 10 M range in a final volume of 2 ml. In the second step, the protein mixture was deposited on a column of Ni-NTA (usually 1 ml of wet support corresponding to 10 -20 mg of immobilized Ni 2ϩ ions). Washing of the column after the immobilization step was carried out with three column volumes of incubation buffer to remove all unbound material. In all cases, a control assay was carried out in parallel with a standard protein (thioredoxin fused with a C-terminal oligohistidine sequence) with no capacity of IN or Fn binding (complete elution of unbound IN or Fn proteins occurred during the first two eluting volumes). The proteins that remained bound to the agarose support were identified by SDS-PAGE analysis on a 10 -20% polyacrylamide gel. This immobilization assay was also used to analyze the interactions between the ␣ 5 and ␤ 1 recombinant fragments. In this case, the immobilization assay was carried out, as described above, using an equimolar mixture of HisTag-␣ 5 -(160 -448) and ␤ 1 -(121-329) (for buffer conditions, see legend to Fig. 8).
Chemical Cross-linking-Before complex formation, the IN and Fn components were dialyzed against buffer C. After centrifugation (80,000 ϫ g for 30 min), the recombinant IN and Fn molecules were mixed at a 1:1 molar ratio (at protein concentrations in the 10 M range), incubated at room temperature for 2 h, and then subjected to chemical cross-linking by addition of dithiobis(succinimidyl propionate) (Pierce; 125 mM stock solution in N,N-dimethylformamide) to a final concentration of 0.5 mM (40). The reaction was stopped by addition of glycine to a final concentration of 50 mM. The time necessary for optimal cross-linking was previously determined by a time course SDS-PAGE analysis (not shown). The mixtures of cross-linked species were loaded on a Sephacryl S-100 HR column (2.6 ϫ 100 cm; Amersham Pharmacia Biotech), and elution was carried out with buffer C at a 50 ml/h flow rate (41). The individual molecular species in the cross-linked complexes were identified by SDS-PAGE under reducing conditions (40).
Divalent Cation-binding Measurements-Quantitation of the amount of Ca 2ϩ ions bound to ␣ 5 -(160 -448) was carried out by densitometry (public domain NIH Image software) of the electrophoretic band labeled with the fluorescent quinoline Ca 2ϩ indicator quin2, as described previously (10). CD-monitored titrations with Ca 2ϩ or Mg 2ϩ were carried out using the apoforms of both integrin recombinant fragments, ␣ 5 -(160 -448) and ␤ 1 -(121-329). Titrations were carried out by adding aliquots of either CaCl 2 or MgCl 2 stock solutions in buffer A (at concentrations in the 10 to 500 mM range) to the apoprotein solution at 10 M protein concentration. Under these conditions, the protein concentration was kept nearly constant over the entire range of cation concentrations (the maximum dilution effect did not exceed 2% of the initial protein concentration).

Production and Structural Characterization of
␣ 5 -(160 -448) and ␤ 1 -(121-329) E. coli strain BLR(DE3) was transformed with the pET15b-␣ 5 -(160 -448) vector, and protein expression was induced by the addition of IPTG. Independently of temperature and IPTG concentration, only a minor part of the expressed His-tagged ␣ 5 -(160 -448) protein was found in the soluble fraction of the bacterial lysate, whereas the major part remained in the insoluble fraction. The His-tagged ␣ 5 -(160 -448) recombinant protein was purified from the insoluble fraction and refolded as a Ni-NTA-immobilized protein, as described previously for the related protein ␣ 5 -(229 -448) (10). The His-tagged protein was subsequently cleaved with thrombin. As previously reported (10), cleavage yields by thrombin were optimized using urea concentrations in the 1 M range. The protein recovered after refolding displays both far-and near-UV CD spectra identical to those of the protein purified from the soluble fraction (data not shown), indicating that a correct refolding is achieved under the conditions used. Under these conditions, about 6 mg of ␣ 5 -(160 -448) were obtained from 1 liter of bacterial culture. To test for both correct synthesis in the bacterium and correct thrombin cleavage, we determined the exact mass of the recombinant protein by electrospray ionization-mass spectrometry. A mass of 30,478.8 Ϯ 0.5 daltons was obtained, in good agreement with the calculated value of 30,479.4. Partial N-terminal amino acid sequencing, including the 10 first residues, gave the expected sequence, GSHMGQILSA, with the four N-terminal residues representing the appendix left after thrombin cleavage of HisTag-␣ 5 -(160 -448).
To maximize the production of soluble protein, the Cyscontaining ␤ 1 -(121-329) protein was expressed as a fusion protein with thioredoxin (42). Many target proteins that are normally produced in an insoluble form in E. coli tend to become more soluble when fused with this protein. Moreover, by combining expression of the target protein fused with thioredoxin with the AD494(DE3) host strain (43), production of soluble proteins containing disulfide bonds is facilitated in the E. coli cytoplasm. We thus used this expression system since Cys-241 and Cys-281 in ␤ 1 are likely linked by an intramolecular disulfide bond, based on previous results obtained with the ␤ subunit of ␣ IIb ␤ 3 (44). To improve the production of properly folded ␤ 1 -(121-329), we also used a co-expression system in which the target protein is expressed along with chaperones GroES and GroEL. The role of the GroESL complex in catalyzing the correct folding of heterologous proteins synthesized in E. coli is well documented (45). Under the conditions used, about 20% of the total protein was found in the soluble fraction of the bacterial lysate. Soluble ␤ 1 -(121-329) was subsequently purified and the Trx tag removed through enterokinase cleavage. Under these conditions, about 5 mg of purified protein were obtained from 1 liter of bacterial culture. A mass of 23,544.0 Ϯ 0.5 daltons was determined by electrospray ionization-mass spectrometry, thus correctly matching the calculated value of 23,544.6. Partial N-terminal amino acid sequencing (10 first residues) gave the expected sequence, AMYPIDLYYL, with the two N-terminal residues representing the appendix left after enterokinase cleavage of TrxTag-␤ 1 -(121-329). We note that the yields of expressed recombinant ␤ 1 -(121-329) largely exceed those observed with recombinant ␣ 5 -(160 -448), if one considers that the soluble fraction used in this work only represents about 20% of the total ␤ 1 protein expressed in the AD494(DE3) host strain. No attempts were made at this stage to refold the ␤ 1 protein from the major insoluble fraction.
Both recombinant proteins, ␣ 5 -(160 -448) and ␤ 1 -(121-329), were characterized by CD. Both far-and near-UV regions of the CD spectra of these proteins are presented in Fig. 1. Based on the profiles observed in the far-UV (wavelength Ͻ250 nm), it appears that (i) ␣ 5 -(160 -448) displays the characteristic spectral features (Fig. 1A) of a protein containing both ␣-helical and ␤-stranded residues; (ii) ␤ 1 -(121-329) displays the characteristic features (Fig. 1C) of a helical protein with a strong minimum at 208 nm (with a similar molar ellipticity value as in ␣ 5 -(160 -448), compare Fig. 1, A and C); however, the minimum at 222 nm displays a lesser (absolute) molar ellipticity than in ␣ 5 -(160 -448) denoting a relatively strong contribution from ␤-stranded residues. The secondary structure predictions with both recombinant proteins are summarized in Table I.
The near-UV CD spectrum (wavelength Ͼ250 nm) of ␣ 5 -(160 -448), as shown in Fig. 1B, is similar in shape and inten-sities to that previously reported for ␣ 5 -(229 -448) (10). Four negative bands are observed between 255 and 285 nm, with well resolved minima at 257, 262, and 268 nm, for the Phe aromatic chromophores (14 Phe residues in total) and at 280 nm for the Tyr aromatic chromophores (16 Tyr residues in total). The CD band associated with the unique Trp-406 is positive and centered at ϳ290 nm. As shown in Fig. 1D, the near-UV CD spectrum of ␤ 1 -(121-329) markedly differs from that of ␣ 5 -(160 -448), although both proteins display rather similar aromatic amino acid compositions (compare Fig. 1, B and D): (i) the region above 275 nm in Fig. 1D that includes the Tyr and Trp bands (three resolved positive bands with maxima at 282, 290 and 293 nm) is far more complex than the corresponding aromatic region in the CD spectrum of ␣ 5 -(160 -448); and (ii) the Phe-associated bands are essentially unresolved (broad bands or shoulders at about 262 and 268 nm, in both profiles 1 and 2). The fact that in the near-UV CD spectrum of ␤ 1 -(121-329) the Phe-associated CD bands are embedded in a monotonous CD profile likely translates the occurrence of a chiral disulfide motif (46).
As stated above, both cysteinyl residues, Cys-241 and Cys-281 in ␤ 1 -(121-329), are likely to form a disulfide bridge in the folded structure (44). Upon addition of 2-mercaptoethanol in excess (non-chiral reducing agent), profile 2 was transformed into profile 3 (Fig. 1D). As a general characteristic, profile 3 displays low intensity CD bands for all the aromatic chromophores from Phe, Tyr, and Trp. However, although not intense, the characteristic Phe bands at 262 and 268 nm appear resolved in profile 3 in the absence of any contribution from a disulfide group. All together, the conversion from profile 2 to profile 3 indicates that upon reduction of the disulfide bridge, there is a net decrease in the rotational strengths of all chromophores in the near-UV CD spectrum of ␤ 1 -(121-329). It is not clear whether the disulfide contribution covers the entire range of the spectrum from 250 to 300 nm, as a broad featureless negative band, with possibly an extremum in the 260 -270 nm range, as observed in disulfide-containing proteins (46). Besides the disappearance of the CD contribution from the intrinsically chiral S-S chromophore, it is likely that the exposure to the solvent of the different aromatic chromophores in the ␤ 1 -protein changes upon cleavage of the disulfide bond. The strong reduction in the intensity of the positive CD bands above 275 nm in profile 3, as compared with profile 2 (Fig. 1D), likely translates a reduction in the rotational strengths of the Tyr and Trp bands rather than uniquely the disappearance of the S-S contribution. This point will not be considered further, and we tentatively conclude, based on the CD data of Fig. 1D, that an intramolecular disulfide bridge is present in ␤ 1 -(121-329), between Cys-241 and Cys-281, and that this disulfide bridge participates in the tertiary fold of the protein.

Conformational Dependence of the Recombinant Integrin Fragments upon Divalent Cation Binding
␣ 5 -(160 -448)-As shown in Fig. 1A, both the apoform and the fully Mg 2ϩ -loaded form of ␣ 5 -(160 -448) display superimposable far-UV CD spectra (profiles 1 and 2, respectively), indicating that the secondary structure is independent of the binding of divalent cations. In contrast, a marked difference is observed in the near-UV between the apoprotein and the fully Mg 2ϩ -loaded form (Fig. 1B). The observed increase in absolute rotational strength of both Phe and Trp CD bands upon cation binding likely translates a rearrangement of the hydrophobic interior of the protein leading to a more compact structure in comparison to the apoform, in agreement with a previous observation with ␣ 5 -(229 -448) (10). In contrast the Tyr CD bands remain invariant upon divalent cation binding. The minimum of this composite band (16 Tyr residues in total) lies at ϳ280 nm thus denoting that most of the Tyr residues are solventexposed in both the apoform and the fully Mg 2ϩ -loaded form (46). This observation markedly differs from what is observed with a typical EF-hand protein, such as calmodulin, for which Ca 2ϩ binding affects both Tyr and Phe chromophores (47).
As shown in Fig. 2, titration of the ␣ 5 -(160 -448) apoform was carried out using three different divalent cations, Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ . All titration profiles giving the variation of the intensity of the negative Phe band at 268 nm as a function of cation concentration are biphasic (transitions I and II). Identical profiles are observed for the three cations used, indicating that there is no selectivity in the binding by the protein of cations that markedly differ by their ionic radii. The biphasic profile is also observed using the Phe band with a minimum at 262 nm (data not shown). A stoichiometric value of four Ca 2ϩ ions bound per protein molecule was inferred for ␣ 5 -(160 -448) using the fluorescence quin2 procedure adapted for quantitative measurements (10), in agreement with the occurrence of 4 EF-hand type cation-binding sites in the ␣ 5 amino sequence. The titration profiles in Fig. 2 are superimposable with those previously reported with ␣ 5 -(229 -448) using Ca 2ϩ and Mg 2ϩ (10). Since all titration assays in this work, as well as those previously reported with the minimal EF-hand domain of ␣ 5  (10), were carried out under identical conditions, we conclude that both isolated proteins, ␣ 5 -(229 -448) and ␣ 5 -(160 -448), display identical K d values for each class, I and II (high and low affinity, respectively), of divalent cation-binding sites, i.e. ϳ30 and ϳ120 M, respectively, independently of the cation itself (Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ ). ␤ 1 -(121-329)-No difference is observed between the far-UV CD spectra of the apoprotein and the fully Mg 2ϩ -loaded form of ␤ 1 -(121-329) (profiles 1 and 2, respectively; Fig. 1C), thus indicating that the basic organization of the isolated ␤ 1 I-type domain is independent of cation binding. In contrast, a marked difference is apparent between both protein forms in the near-UV, at least in the 250 -270 nm range (Fig. 1D), indicating that cation binding is associated with a conformational adaptation of the protein hydrophobic core. In this case, there is a reduction of the (absolute) intensity at 268 nm upon Mg 2ϩ binding. If the spectrum in this region is dominated by the Phe chromophores, such a reduction in the molar ellipticity at 268 nm could be interpreted as due to the induction of an open form of ␤ 1 -(121-329) upon divalent cation binding, with some of its Phe residues becoming more exposed to the external environment. It must be noted, however, that the CD spectrum of ␤ 1 -(121-329), in the 255-270 nm range, corresponds to a composite profile with contributions from both Phe and S-S chromophores (see above) so that the observed variations between profile 1 (apoform) and profile 2 (fully Mg 2ϩ -loaded form) may include contributions from both chromophores without an exact delineation between these two types of contributions being readily feasible. As expected, profile 1 was restored (not shown) upon addition of EDTA in excess (50 mM) to the Mg 2ϩ -loaded form of ␤ 1 -(121-329).
As shown in Fig. 3, we then carried out a titration of the apoform of ␤ 1 -(121-329) with the three divalent cations, Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ , used with ␣ 5 -(160 -448). The ellipticity variations at 268 nm with ␤ 1 -(121-329) resulted in a monophasic profile upon progressive addition of either Mn 2ϩ or Mg 2ϩ (Fig.  3A), thus indicating the occurrence of a single class of cationbinding sites. The absolute molar ellipticity variations observed with Mg 2ϩ or Mn 2ϩ at saturation are identical, suggesting that both cations induce a practically identical conformational variation upon binding to the protein. In contrast, even in the presence of a large excess of Ca 2ϩ (up to 100 mM), the near-UV CD spectrum of the apoform of ␤ 1 -(121-329), i.e. profile 1 in Fig. 1D, remained unchanged (CD data with Ca 2ϩ not shown). This appears as a striking difference with what is observed with the ␣ 5 protein fragment for which no  Fig. 1B), and [] to the molar ellipticity at a given cation concentration. Note (see also "Experimental Procedures") that the titration by Mn 2ϩ was carried out at pH 6.0 (cacodylate buffer B), whereas the titration with Ca 2ϩ and Mg 2ϩ was carried out at pH 8.3 (borate buffer A). The initial protein concentration was 10 M in all experiments, and the successive points along the titration profiles correspond to a progressive dilution of the solution not exceeding 2% (see "Experimental Procedures"). To establish the correspondence between the titrations at both pH values, the assays with Ca 2ϩ and Mg 2ϩ were carried out at pH 6.0 and 8.3 ( Ⅺ and E, respectively). Only some points (Ⅺ) of the titration profiles at pH 6.0 are given in A and B. All CD experiments were carried out in duplicate. For the observed variations between two independent experiments see C (E and q).

FIG. 3. CD-monitored divalent cation binding to ␤ 1 -(121-329).
A, normalized molar ellipticity variations upon binding of Mn 2ϩ (E) and Mg 2ϩ (q) to the apoprotein. The changes in molar ellipticity are presented as a function of the cation concentration for the Phe band at 268 nm. [] 0 and [] max correspond to the molar ellipticities of the apoform and of the cation-saturated form, respectively, and [] to the molar ellipticity at a given cation concentration. The Mn 2ϩ titration was carried out at pH 6.0 (cacodylate buffer B), whereas the titration with Mg 2ϩ was carried out at pH 8.3 (borate buffer A). In both cases, the initial protein concentration was 10 M, and the successive points along the titration profiles correspond to a progressive dilution of the solution not exceeding 2% (see "Experimental Procedures"). To establish the correspondence between the titrations at different pH values, the assay with Mg 2ϩ was also carried out at both pH values, 6.0 and 8.3, yielding practically identical results. For the sake of comparison, some points of the titrations at pH 6.0 (Ⅺ) are given in A. B, normalized molar ellipticity changes upon binding of Ca 2ϩ to the Mg 2ϩ -loaded ␤ 1 -(121-329) protein; initial conditions: [protein] ϭ 0.01 mM and [Mg 2ϩ ] ϭ 0.6 mM. selectivity was observed for the binding of the different divalent cations (Fig. 2). We hypothesized that the ionic radius likely plays a major role in the effects observed upon cation binding to ␤ 1 -(121-329). Indeed, a large cation, Ba 2ϩ (1.34-Å ionic radius), induced no effect on the near-UV CD spectrum of the ␤ 1 -protein, whereas a small cation, Zn 2ϩ , induced a similar effect as the one observed with Mg 2ϩ or Mn 2ϩ with ionic radii below 0.8 Å (CD data with Ba 2ϩ and Zn 2ϩ not shown). So far the stoichiometry of cation binding by ␤ 1 -(121-329) has not been determined. This originates from the fact that Ca 2ϩ binds to the recombinant ␤ 1 protein with low affinity (see below); the quin2 assay, described above in the case of ␣ 5 -(160 -448), could thus not be used. It is accepted that the integrin ␤ subunits display a single cation-binding site resembling the MIDAS site found in the ␣ subunit I-domains (13,14). Assuming that there is a single cation-binding site in ␤ 1 -(121-329), K d values of ϳ5 and ϳ85 M were inferred from the titration profiles of Fig. 3A for Mn 2ϩ and Mg 2ϩ , respectively. These values are closely related to the apparent K d values reported for the ␤ 3 MIDAStype site in ␣ v ␤ 3 , i.e. 1-10 and 80 -100 M for Mn 2ϩ and Mg 2ϩ , respectively (14).
Two possible explanations can be envisaged for the apparent lack of any conformational effect (based on near-UV CD evidence) when ␤ 1 -(121-329) is brought in the presence of Ca 2ϩ : (i) either ␤ 1 -(121-329) does not bind Ca 2ϩ or (ii) bound Ca 2ϩ is unable to induce any conformational adaptation of the protein.
In order to assess if one or both possibilities applied to ␤ 1 -(121-329), we carried out a CD-monitored Ca 2ϩ titration of the fully Mg 2ϩ -loaded protein. As shown in Fig. 3B, a full reversal of the Mg 2ϩ -induced CD effects was observed for Ca 2ϩ concentrations above 50 mM as judged by the molar ellipticity variations at 268 nm. In fact, above this Ca 2ϩ concentration, the entire near-UV CD spectrum of ␤ 1 -(121-329) becomes superimposable (not shown) to that of the apoprotein (profile 1 in Fig. 1D), thus suggesting that Mg 2ϩ is released from the ␤ 1 protein when Ca 2ϩ binds to it. This situation is reminiscent of what is observed with the homologous ␣ subunit I-domains where Ca 2ϩ displaces radioactive 54 Mn 2ϩ from the ␣ L recombinant I-domain when added in excess (48,49). Our Ca 2ϩ /Mg 2ϩ competition experiment in Fig. 3B suggests that Ca 2ϩ competes with Mg 2ϩ for binding to ␤ 1 -(121-329). Based on the competition profile of Fig. 3B, it appears that Ca 2ϩ differs from Mg 2ϩ by an affinity less than 2 orders of magnitude, but Ca 2ϩ induces no significant conformational change, if any, in ␤ 1 -(121-329).
We then used the PRE NMR approach to understand further some of the mechanistic aspects of the interaction of divalent cations with ␤ 1 -(121-329). This method, as briefly presented under "Experimental Procedures," is based on the fact that paramagnetically induced relaxation of water protons is markedly enhanced if the water molecules belong to the coordination sphere of a protein-bound paramagnetic cation in comparison to the relaxation observed with the free paramagnetic cation (38). In the case of ␤ 1 -(121-329), Mn 2ϩ was selected as a convenient paramagnetic probe due to its elevated electronic magnetic moment and to the fact that it interacts with the protein with a relatively high affinity (see above). As shown in Fig. 4, the progressive addition of Mn 2ϩ to the ␤ 1 -(121-329) apoform is accompanied by an enhancement of the water relaxation rates (1/T 1 ). An enhancement factor, ⑀ ϭ 2.1 (for a definition of this dimensionless factor see "Experimental Procedures"), was observed at the latest point of the titration at 1 mM Mn 2ϩ . Based on K d ϭ 5 M, as given above, an intrinsic enhancement factor ⑀ b (defined under "Experimental Procedures") of about 12.5 can be estimated (assuming a stoichiometric ratio of 1). Such an estimation assumes fast exchange conditions on the NMR time scale for water exchange, a condition that appears to apply to the present situation (38). This ⑀ b value accounts for a relatively large hydration of the ␤ 1 -bound Mn 2ϩ ion with at least 3 water molecules. 2 This conclusion is in agreement with the structural organization of the MIDAS site in the I-domains of ␣ L and ␣ M where the coordination of Mn 2ϩ involves 3 water oxygen atoms as part of the regular octahedral coordination of the central cation (50,51). Our PRE results in Fig. 4 therefore suggest a strong structural resemblance between the divalent cation-binding site of ␤ 1 -(121-329) and the MIDAS site of the integrin ␣ I-domains, in agreement with a previous proposal (13,14).
We then analyzed the effects of adding Ca 2ϩ to the Mn 2ϩloaded ␤ 1 -(121-329) protein by PRE. Addition of Ca 2ϩ in excess (to a final concentration of 100 mM; see Fig. 3B), using the last point of the Mn 2ϩ titration in Fig. 4, resulted in a relaxation rate (1/T 1 ) identical to that of free Mn 2ϩ . This clearly establishes that Ca 2ϩ competes with Mn 2ϩ for binding to ␤ 1 -(121-329). The reversal of the Mg 2ϩ -induced conformational effect observed on the CD spectrum of ␤ 1 -(121-329) when an excess of Ca 2ϩ is added to the fully Mg 2ϩ -loaded protein (Fig. 3B) can thus be interpreted as due to the release of Mg 2ϩ upon Ca 2ϩ binding to the protein. Whether both divalent cations, Ca 2ϩ and Mg 2ϩ , bind to the same site of ␤ 1 -(121-329), or not, remains an open question (see "Discussion").
The trichloroacetic acid-precipitated apoforms of ␣ 5 -(160 -448) and ␤ 1 -(121-329) used in this work (see under "Experimental Procedures") need some consideration in terms of their structural integrity. The fact that both apoforms are characterized by far-UV CD spectra identical to those of the corresponding cation-loaded native forms (see Fig. 1) and that conformational reversibility is observed, based on near-UV CD evidence, for cation binding and removal, establishes that the trichloroacetic acid-precipitated proteins retain the basic folding features of the native proteins. However, it would be necessary to 2 An ⑀ b of ϳ17 was found for the Mn 2ϩ -loaded third site of the typical EF-hand Ca 2ϩ /Mg 2ϩ -binding parvalbumin (76); according to x-ray crystallographic evidence (66) this low affinity site retains 5 water molecules directly coordinating the central Mn 2ϩ ion. In contrast, ⑀ b ϭ 2.3 for Gd 3ϩ bound to the largely dehydrated high affinity Ca 2ϩ /Mg 2ϩ -binding sites of parvalbumin (76). investigate if cation removal through trichloroacetic acid precipitation is really identical to cation removal through competition using a cation-chelating agent. 3 EDTA readily generates the apoforms of both proteins ␣ 5 -(160 -448) and ␤ 1 -(121-329). However, it is known that the apoforms of Ca 2ϩ /Mg 2ϩ -binding proteins bind EDTA in a selective manner, and this binding of a metal-chelating molecule biases the interaction studies with metal cations using EDTA-generated protein apoforms (52).

Ligand-binding Properties of ␣ 5 -(160 -448)
and of ␤ 1 (121-329) The immobilization assay depicted in Fig. 5A was used to assess the ligand binding capacities of both recombinant proteins, ␣ 5 -(160 -448) and ␤ 1 -(121-329). The ligand used was a fibronectin recombinant fragment, 3Fn7-10 (23), that contains the RGD adhesion motif located in the type III module 3Fn10, as well as the synergistic regions located in 3Fn9 (53). As shown in Fig. 5B, His-tagged ␣ 5 -(160 -448) binds 3Fn7-10 in the presence of either Mg 2ϩ or Ca 2ϩ . The stoichiometric ratio of the complex was determined through covalent cross-linking (see "Experimental Procedures") using the untagged proteins, as previously reported (10). At saturating concentrations of Mg 2ϩ , a major peak appears by gel filtration (data not shown) with an estimated mass of about 70 kDa, in agreement with the formation of the 1:1 complex ␣ 5 -(160 -448)⅐3Fn7-10 (calculated mass 72.5 kDa). Recombinant ␤ 1 -(121-329) also binds 3Fn7-10 in the presence of Mg 2ϩ in excess (Fig. 5C) to form a 1:1 complex (stoichiometric ratio inferred from cross-linking and size-exclusion chromatography using the untagged ␤ 1 -protein; data not shown). As shown in Fig. 5C, the ␤ 1 -(121-329)⅐3Fn7-10 binary complex is formed in the presence of Mg 2ϩ (or Mn 2ϩ , not shown), whereas no interaction occurs in the presence of Ca 2ϩ , as judged by the immobilization assay.
shown). A first interpretation of the results presented in Fig. 6 is that the 10th type III module in Fn, with its RGD adhesion sequence, primarily interacts with the ␤ 1 I-type domain, whereas this adhesion motif in Fn has no interaction with the isolated ␣ 5 EF-hand domain, as judged by the immobilization procedure used here. However, this does not preclude that the RGD motif in 3Fn10 can still interact directly with the ␣ 5 subdomain if the integrin-ligand interactions at play involve a restricted number of intermolecular contacts (see "Discussion"). Conversely, our results in Fig. 6 indicate that the adhesive regions in 3Fn9 (acting in synergy with RGD) primarily interact with the ␣ 5 EF-hand domain, in agreement with previous results concerning the ligand-binding specificities of both ␣ and ␤ subunits of integrin ␣ 5 ␤ 1 (55,56).

Functional Role of Cation binding to ␤ 5 -(121-329)
It has been recently shown that ligand binding to the ␣ 2 I-domain results in metal ion displacement to generate a metalfree I-domain-ligand complex (57). A similar mechanism was initially proposed for the ␤ subunit I-type domains (18). By using the isolated ␤ 1 -(121-329) protein, we investigated by PRE the binding status of Mn 2ϩ in the presence of the RGD peptide, G(Pen)*ELRGDGWC*. The choice of a small RGD ligand, instead of a fibronectin fragment, such as 3Fn7-10, was guided by the fact that a small ligand will not introduce any significant change in the local correlation times of the water molecules surrounding Mn 2ϩ upon complex formation (see under "Experimental Procedures"), whereas this might not be the case with a high molecular weight ligand. 3 A solution of Mn-loaded ␤ 1 -(121-329) was titrated with the G(Pen)*ELRGDGWC* peptide, and water relaxation rates were measured as a function of peptide concentration (Fig. 7). We used the last point in the Mn 2ϩ titration of Fig. 4 to initiate a PRE-monitored peptide titration. As apparent in Fig. 7, the initial enhancement factor ⑀ ϭ ϳ2.1 progressively returns in a monophasic manner to the value ⑀ ϭ 1 characteristic of free Mn 2ϩ . In contrast, no variation of the water relaxation rate was observed upon addition of the control peptide GAC*VRLNSLAC*GA (54), at least in the range of concentrations used. The PRE results observed with the cyclic RGD-containing peptide appear as direct proof that binding of this RGD ligand to the Mn 2ϩ -saturated ␤ 1 protein is characterized by a complete release of Mn 2ϩ to form a metalfree ligand-integrin complex. It is likely that this is also the case of Mg 2ϩ upon binding of an RGD ligand, such as 3Fn7-10, to ␤ 1 -(121-329), although we presently provide no direct evidence in this respect (see "Discussion").

␣␤ Heterodimer Formation
We used the two-step immobilization assay described under "Experimental Procedures" to determine if the ␣ 5 and ␤ 1 recombinant fragments form a stable ␣␤ heterodimer (or miniintegrin). HisTag-␣ 5 -(160 -448) and untagged ␤ 1 -(121-329) were incubated at a 1:1 molar ratio in the presence of saturating concentrations of Mg 2ϩ ions and loaded on the Ni-NTA agarose column. Under these conditions, both fragments were immobilized on the column, based on SDS-PAGE (Fig. 8A, lane  3), establishing that a stable complex is formed between both integrin fragments. Interestingly, no complex was observed (SDS-PAGE data not shown) between ␤ 1 -(121-329) and His-Tag-␣ 5 -(229 -448) (10) that only includes repeats IV to VII but not repeat III.
The stoichiometry of the ␣ 5 -(160 -448)⅐␤ 1 -(121-329) complex was determined through covalent cross-linking (Fig. 9A). At saturating concentrations of Mg 2ϩ , a major peak appears by gel filtration with an estimated mass of ϳ50 kDa (Fig. 9A), in agreement with the formation of a 1:1 ␣ 5 ␤ 1 complex (calculated value 54 kDa). The divalent cation dependence of the interaction between both recombinant integrin fragments is clearly apparent in Fig. 9A. Indeed, in the presence of an excess EDTA (50 mM) no band at 50 kDa was observed on the gel filtration profile, thus establishing that the ␣ 5 -(160 -448)⅐␤ 1 -(121-329) binary complex is not formed under these conditions. This is in agreement with previous results indicating that divalent cations are required so that the the association between the ␣ and ␤ integrin subunits is maintained (58).

Interactions between the ␣␤ Heterodimer and Fibronectin
We analyzed the interaction between the minimized ␣␤ heterodimeric complex and 3Fn7-10 using the two-step immobilization assay. As shown in Fig. 8B, a stable ternary complex   Fig. 9B), thus establishing the RGD dependence of the interaction between the 3Fn7-10 ligand and the minimized ␣ 5 ␤ 1 integrin. We also investigated the competition between n-butyrate, as well as guanidinium chloride, and the RGD-containing 3Fn7-10 ligand for binding to the ␣ 5 -(160 -448)⅐␤ 1 -(121-329) heterodimer. As shown in Fig. 8B  (lanes 6 and 7), the ternary complex was totally dissociated in the presence of 50 mM n-butyrate, whereas no dissociation was observed with GdnHCl at the same concentration.
The stoichiometry of the ternary complex was determined through covalent cross-linking. As shown in Fig. 9B, the crosslinked ternary complex displays an estimated mass of ϳ85 kDa, indicating that one 3Fn7-10 ligand molecule is bound per heterodimer to form a ternary 1:1:1 complex (calculated mass 90 kDa). The presence of a small amount of free ␣ 5 ␤ 1 miniintegrin in the cross-linking experiment in the presence of Mg 2ϩ in excess (see Fig. 9B) likely denotes an imbalance in the initial input ratios rather than a dissociation of the complex, since no free Fn ligand was concomitantly observed in the final reaction mixture. In the presence of the RGD peptide, G(Pen)*ELRGDGWC*, the ternary complex is not formed, and no cross-linked ternary complex was thus visualized. Both molecular species, the initial binary complex ␣ 5 -(160 -448)⅐␤ 1 -(121-329) and free 3Fn7-10, are primarily visualized on the chromatographic profile (Fig. 9B). We nevertheless note a minor amount of ternary complex in this profile, and this is certainly due to the relative affinities of both RGD-containing ligands, 3Fn7-10 and G(Pen)*ELRGDGWC*, for the ␣ 5 ␤ 1 mini-integrin.

Conformational Adaptation of the Integrin and Ligand Components in the Complexes
We summarize in Fig. 10 our CD observations (near-UV region) with regard to the conformational changes induced upon formation of the different binary complexes and the ternary ligand-receptor complex described above. In all cases, comparisons (difference spectra) were made using the Mg 2ϩloaded forms of both isolated integrin fragments (there is no divalent cation dependence on the conformation of the fibronectin ligand, as initially reported; see Ref. 10). For the sake of clarity, we present in Fig. 11A an illustration of the different conformational states adopted by both ␣ and ␤ components that make up the minimized ␣ 5 ␤ 1 integrin, as well as by the Fn ligand, as a complement to the CD results presented in Fig. 10. It thus appears that (i) the binary complex ␣ 5 -(160 -448)⅐3Fn7-10 involves changes at the level of both Phe and Trp bands without affecting the Tyr bands (Fig. 10A). Similar effects were already reported (10) in the case of the related binary complex ␣ 5 -(229 -448)W406L⅐3Fn8 -10, in which the ␣ 5 integrin component displays a single mutation of its unique Trp-406 (Trp to Leu point mutation). In this case, the differences observed with the Trp-associated CD bands specifically translate conformational variations of the Fn molecule since it is only this Fn fragment that contains the Trp residues. We note that the intensities observed for the Trp-associated CD bands in the difference spectrum "7 ؊ (3 ؉ 5)" (see Fig. 10A) are closely related to those previously observed with the ␣ 5 -(229 -448)W406L⅐3Fn8 -10 complex. This suggests that the spectral changes induced at the level of the Trp chromophore upon formation of the binary ␣ 5 ⅐Fn complex is essentially due to conformational changes that occur within the Fn ligand molecule and not in the ␣ 5 molecule. (ii) The binary complex ␤ 1 -(121-329)⅐3Fn7-10 is characterized by a null difference spectrum "8 ؊ (4 ؉ 5)", as shown in Fig. 10B. The latter result establishes that the structure of the isolated ␤ 1 -(121-329) in the binary complex is similar to that of the Mg 2ϩ -loaded form of the ␤ 1 -protein, although Mg 2ϩ is released from its site upon formation of the ligand-integrin complex (see above). The Mg 2ϩ -loaded conformation of ␤ 1 -(121-329) therefore remains "frozen" in the complex although the ␤ 1 -component is metalfree (see "Discussion"). (iii) The binary complex ␣ 5 -(160 -448)⅐␤ 1 -(121-329) is characterized by a null difference spec-trum "6 ؊ (3 ؉ 4)" as shown in Fig. 10C, thus indicating that the formation of the ␣ 5 -(160 -448)⅐␤ 1 -(121-329) heterodimer does not involve any major conformational rearrangement of both components. (iv) The ternary ligand-receptor complex, ␣ 5 -(160 -448)⅐␤ 1 -(121-329)⅐3Fn7-10, is characterized by a difference spectrum (Fig. 10D) that affects both Phe and Trp chromophores in a similar manner as already observed in the case of the binary complex ␣ 5 -(160 -448)⅐3Fn7-10 (Fig. 10A). The difference spectrum "9 ؊ (3 ؉ 4 ؉ 5)" is practically superimposable to the 7 ؊ (3 ؉ 5) spectrum, suggesting that the addition of 3Fn7-10 to the soluble mini-integrin is characterized by a specific conformational change of the fibronectin ligand and not by a change of the integrin heterodimer. It appears that this change is essentially induced by the interaction with the integrin ␣ component (see comment i above).
We note, as a general comment, that the relevance of the data presented in Fig. 10 derives from the accuracy with which the molar UV absorptivities were determined (see "Experimental Procedures") for both recombinant integrin proteins ␣ 5 -(160 -448) and ␤ 1 -(121-329), as well as for the Fn ligand proteins (10).

DISCUSSION
As presented under "Results," ␣ 5 -(160 -448) adopts well defined structural features corresponding to ϳ30 and 25% of ␣-helical and ␤-stranded residues, respectively (Table I). In comparison with a previous CD analysis (10) of ␣ 5 -(229 -448), representing the minimal EF-hand type domain of ␣ 5 , the secondary structure percentages obtained with ␣ 5 -(160 -448) indicate that this isolated protein includes a significantly larger number of ␣-helical and ␤-stranded residues (see Table  I). If we assume that the central EF-hand type domain 229 -448 in ␣ 5 -(160 -448) adopts the same structural organization than in the isolated ␣ 5 -(229 -448) fragment (10), then the additional number of helical residues in the former should be located the additional third repeat.
The production of an isolated ␤ 1 I-type domain appeared of the highest interest to assess the role of this domain on the structure and function of integrin ␣ 5 ␤ 1 . The expression of such a ␤ integrin region has been recently viewed as a complex one due, among others, to the occurrence of disulfide bonding (59). As reported under "Results," the production of the isolated ␤ 1 I-type domain as a thioredoxin-fused recombinant protein in an E. coli host strain compatible with disulfide bonding, in conjunction with the co-expression of bacterial chaperones, is characterized by relatively good yields of a soluble, natively folded, recombinant protein including an intramolecular disulfide bridge. The CD analysis of ␤ 1 -(121-329) indicates that this isolated ␤ 1 I-type domain corresponds to a highly organized structure, including up to 65% of ␣-helix and ␤-strand residues in total (see Table I). This experimentally derived secondary structure organization can be compared with the ones inferred from theoretical models used for predicting the folding of the integrin ␤ subunit I-type domains. A first model has been proposed (13,59) in which the ␤ I-type domain displays a tertiary fold closely resembling that determined for the I-domains of ␣ M , ␣ L , and ␣ 2 by x-ray crystallography (12,60,61). The tertiary fold in the ␣ M I-domain includes 38% of helical residues and 27% of ␤-stranded residues (60). These values appear very close to those experimentally measured in this work by CD with the isolated ␤ 1 I-type domain (see Table I). However, Lin et al. (14) have recently concluded that the fold of the ␤ subunit I-type domains is expected to differ significantly from the one observed with the ␣ I-domains. According to their predictions, the secondary structure organization would correspond to much a lower content of ␣-helix and ␤-strand residues (ϳ21 and 11%, respectively), in contrast with our experimental results with ␤ 1 -(121-329) (see Table I). Based on the CD evidence reported in this work, ␤ 1 -(121-329) undergoes a conformational adaptation upon Mg 2ϩ (or Mn 2ϩ ) binding (Fig. 3A). The observed variations essentially affect the 250 -280 nm range of the CD spectrum, whereas the intensity of the Trp-and Tyr-associated bands in the 280 -300 nm range is only slightly altered (see Fig. 1D). The reduction in the absolute CD signal intensity at 268 nm can be interpreted as an enhanced exposure to the external medium of some of the Phe residues upon Mg 2ϩ (or Mn 2ϩ ) binding. Such a conformational adaptation upon divalent cation binding is reminiscent of what has been previously observed with the integrin ␣ subunit I-domains. Based on crystallographic evidence (51,62), two different conformational states have been defined for the human ␣ L I-domain as follows: (i) a "closed" conformation in which all Phe residues are buried in the protein structure, and (ii) an "open" form that is characterized by the exposure of some of the Phe residues to the solvent. The transition from 2 to 4, in Fig. 11A, illustrates such a cation-induced conformational rearrangement in the case of the isolated ␤ 1 I-type domain.
A remarkable observation is that Ca 2ϩ is unable to induce any conformational adaptation of the isolated ␤ 1 -(121-329) protein in contrast to Mg 2ϩ or Mn 2ϩ . However, Ca 2ϩ can expel a ␤ 1 -bound Mn 2ϩ ion from its binding site when it is present in excess, based on the PRE results presented in Fig. 4. The differences observed between both cations, Ca 2ϩ and Mn 2ϩ , could be linked to the difference in their ionic radii. As indicated by site-directed mutagenesis (13,14,63,64), the putative cation-binding sites in the I-type domains of the ␤ 1 , ␤ 2 , and ␤ 3 subunits present a strong sequence homology with the MIDAS sites in the ␣ I-domains with regard to the number and the chemical nature of the metal-coordinating residues (see Table  II). Our PRE measurements with the isolated ␤ 1 I-type domain also lead to the conclusion that the divalent cation-binding site in this integrin domain adopts a MIDAS-type topology. This conclusion is based on the observation of a relatively large hydration of the Mn 2ϩ ion when bound to the ␤ 1 I-type domain (3 water molecules are likely to coordinate the central cation; see "Results"), in agreement with the known hydration of Mn 2ϩ in the crystal structures of the isolated I-domains of ␣ L and ␣ M (50,51). Coordination in proteins of "small" cations, such as Mg 2ϩ or Mn 2ϩ , versus "large" cations, such as Ca 2ϩ , differs by the number of coordinating oxygen atoms. Hexa-coordination is generally the rule for Mg 2ϩ and Mn 2ϩ , whereas hepta-coordi-nation is largely observed for Ca 2ϩ (65)(66)(67)(68). It could be that hepta-coordination is not adapted to the intrinsic features of the MIDAS site. No crystal structure of a Ca 2ϩ -substituted MIDAS site has been reported so far. Two mutually excluding possibilities can be envisaged to explain the effects of Ca 2ϩ on cation binding by ␤ 1 -(121-329) as well as on the conformation of this isolated ␤ 1 I-type domain as follows: (i) upon replacement of the small cation Mg 2ϩ (or Mn 2ϩ ) by Ca 2ϩ within the ␤ 1 MIDAS-type site, a rearrangement of the coordination sphere is induced to accommodate an additional coordinating atom (conversion from hexa-to hepta-coordination), whereas the protein conformation is reverted to an apoform-like conformation; (ii) Ca 2ϩ binds to a distinct cation-binding site, with reduced affinity, and this additional site could share one of its cationcoordinating residues with the adjacent MIDAS-type site (hypothesis of overlapping Mg 2ϩ -and Ca 2ϩ -binding sites). At the structural level, the latter possibility is reminiscent of what is observed with the typical EF-hand parvalbumin, in which one of the high affinity divalent cation-binding sites displays a satellite (low affinity) site with both sites sharing a common coordinating residue (66). If such an overlap between sites occurs in ␤ 1 , it would be expected that, upon binding, Ca 2ϩ will significantly alter the affinity of ␤ 1 -(121-329) for Mg 2ϩ (or Mn 2ϩ ) so that the initially bound cation will be released from ␤ 1 . Such a mechanism requires validation by further structural studies with the isolated ␤ 1 -(121-329) protein reported here.
Both isolated ␣ 5 -(160 -448) and ␤ 1 -(121-329) fragments, taken separately, bind 3Fn7-10 in a divalent cation-dependent manner. However, whereas all three divalent cations, Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ (see Fig. 2), promote the same ligand-competent conformation of ␣ 5 -(160 -448), it is only Mg 2ϩ (or Mn 2ϩ ), and not Ca 2ϩ , that is responsible for promoting ligand recognition in the case of the isolated ␤ 1 -(121-329) protein. Such a behavior is closely related to that of the ␣ M I-domain for which Ca 2ϩ is ineffective in promoting binding of the iC3b ligand, in contrast to Mg 2ϩ (48). This suggests that, as in the case of the ␣ subunit I-domains (69), the open and closed conformations of ␤ 1 -(121-329), in the presence and absence of Mg 2ϩ , respectively, correspond to the active and inactive forms of the protein, as judged by its ligand-binding capacity. The higher affinity of ␤ 1 -(121-329) for Mn 2ϩ than for Mg 2ϩ (see Fig. 3), which is not observed with the ␣ 5 subunit (see Fig. 2), could explain the initial observations of Gailit and Ruoslahti (17) indicating that Mn 2ϩ affords a 2-3-fold increase of the binding of fibronectin to its ␣ 5 ␤ 1 receptor.
The stability of the ␣ 5 -(160 -448)⅐3Fn7-10 and ␤ 1 -(121-329)⅐3Fn7-10 binary complexes, in the presence of n-butyrate and GdnHCl, as summarized in Table III, deserves some comments with regard to the possible intermolecular contacts at play between the fibronectin RGD-containing ligand and both ␣ 5 and ␤ 1 fragments. The fact that the isolated binary complex ␤ 1 ⅐Fn is dissociated into its components in the presence of  1ZON; ␣ M , PDB entries 1IDO and 1JLM; ␣ 2 , PDB entry 1AOX) that are involved in divalent cation coordination in the MIDAS site are given in bold. The putative cation-coordinating residues in the ␤ subunit I-type domains (13,63,64) are also given in bold. The underlined DDL sequence in the ␤ subunit I-type domains corresponds to a putative RGD-binding motif (15). a Swiss-Prot accession numbers for ␣ L , ␣ M , and ␣ 2 are P20701, P11215 and P20701, respectively. b PIR accession number is B27079. c Swiss-Prot accession numbers P05107 and P05106 are for ␤ 2 and ␤ 3 , respectively.
d Relative positions of the metal-coordinating residues in the ␤ EFhand type sequences (positions are given relative to the first metalcoordinating residue AspT1).

Binary complexes
Ternary complex GdnHCl and this is not the case of the binary complex ␣ 5 ⅐Fn is a strong indication that the interactions between the RGD ligand and the ␣ or ␤ subunit essentially differ by their chemical nature, in agreement with a model previously proposed (10). In this model, the RGD motif interacts with the integrin ligand-binding sites by anchoring its aspartyl carboxylate 4 and its arginyl guanidinium on the ␣ subunit and on the ␤ subunit, respectively (Fig. 11C). As previously noted (10), other carboxylic acids with a lesser number of carbon atoms were not effective in dissociating the ␣ 5 ⅐Fn binary complex, and this could translate that n-butyrate interacts in a specific manner with the RGD-binding site within the ␣ 5 EF-hand domain. The fact that GdnHCl is effective in dissociating the ␤ 1 ⅐Fn binary complex and is ineffective in the case of the ␣ 5 ⅐Fn binary complex (Table III) also suggests that the effects of such a simple guanidinium compound are rather specific. The observation that n-butyrate also dissociates the ␤ 1 ⅐Fn complex (Table  III) can be rationalized if one assumes that the chemical group in the ␤ subunit that directly interacts with the RGD motif is an integrin carboxylate group (Fig. 11C), possibly originating from one Asp residue from the highly conserved DDL motif that has been identified as an integrin RGD-binding site (15). There is apparently a paradox between the proposed dual binding of the RGD motif by each subunit of the integrin ␣ 5 ␤ 1 , as illustrated in Fig. 11C, and the results obtained for the interaction between 3Fn10 -11 (containing the RGD motif but devoid of the synergistic regions in 3Fn9) and the isolated ␣ 5 and ␤ 1 fragments (see Fig. 6). If we conclude, in the absence of any stable ␣ 5 -(160 -448)⅐3Fn10 -11 binary complex (as judged by our immobilization assay), that there is no direct contact between the ligand adhesion RGD motif and the integrin ␣ 5 subunit, then it is not easy to rationalize the observed competition between the cyclic RGD peptide, G(Pen)*ELRGDGWC*, and 3Fn7-10 for binding to isolated ␣ 5 -(160 -448) (see Fig. 5B, lane 5). It was previously concluded (55,56) that the ␣ 5 subunit is essentially involved in interactions with the synergistic regions found in the 9th type III module of Fn. A possibility accounting for our observation that the cyclic peptide G(Pen)*ELRGDGWC* readily dissociates the binary complex between 3Fn7-10 and ␣ 5 -(160 -448) would then be that RGD, in the small cyclic peptide, also directly competes with the Fn synergistic regions for binding to ␣ 5 . This is a possibility that cannot be ruled out presently. If RGD was to interact solely with ␣ 5 through an intermolecular metal-coordination bond, as shown in Fig. 11C, then this contact would certainly not be sufficient, in energetic terms, to ensure the stability of the binary complex ␣ 5 -(160 -448)⅐3Fn10 -11, in the absence of other stabilizing contacts (synergistic contacts). However, a direct contact between the Asp carboxylate from RGD and the metal cation bound to one 5 of the EF-hands in ␣ 5 (Fig. 11C) would be highly relevant to an accurate positioning of the RGD-containing ligand within the binding pocket of the integrin receptor.
If most of the binding energy between ␣ 5 -(160 -448) and 3Fn7-10 originates from other contacts besides this assumed specific RGD contact with ␣ 5 , what then is the exact role of n-butyrate in dissociating this binary complex? The synergistic pentapeptide motif PHSRN in 3Fn9 (53) displays several structural elements in common with the RGD motif, i.e. an Arg residue as well oxygenated polar side chains (from Ser and Asn). Based on mutation studies, Aota et al. (53) concluded that the arginine residue of PHSRN is important for cell-adhesive function. If this is so, Arg in PHSRN might then interact with a negatively charged residue (carboxylate) from ␣ 5 , and this interaction would be weakened by the presence of n-butyrate concomitantly with the weakening of the contact between RGD-Asp and the integrin EF-hand. 5 The inefficiency of GdnHCl to dissociate the ␣ 5 ⅐Fn binary complex (Fig. 5B, lane 7) could result from the fact that Arg, in the synergistic PHSRN sequence, is only one among the different stabilizing intermolecular contacts in the complex. However, such conclusions still remain speculative, and it appears necessary to evaluate the contributions of the different intermolecular contacts to the global binding energy on a quantitative basis. Although energetically weak, some of the aforementioned interactions could strongly influence the specificity of ligand recognition by the integrin receptor.
By using Tb 3ϩ as a luminescence probe, Dickeson et al. (57) have recently shown with the isolated I-domain of ␣ 2 that the presence of a metal cation is required for initial binding of a collagen-derived peptide, and subsequently ligand binding results in cation displacement from the integrin domain thus generating a metal-free ligand-receptor complex. This mechanism, in which a cation promotes a ligand-competent conformation and the cation is subsequently released upon formation of the ligand-receptor complex also applies apparently to the case of the ␤ I-type domains, as initially suggested by D'Souza et al. (18). Our results with the isolated I-type domain of ␤ 1 are in agreement with this "cation displacement" mechanism, as indicated by the PRE experiment reported in Fig. 7. Such a cation release from the ␤ I-type domains upon formation of the ligand-integrin complex remains basically unexplained in mechanistic terms. It must be noted that the ␤ I-type domains are systematically characterized by the occurrence of a totally conserved DDL sequence that is adjacent to the metal-binding DXSXS sequence (see Table II). Pasqualini et al. (15) showed that the C*WDDLWLC* cyclic peptide, including the conserved ␤ DDL tripeptide sequence (Table II), is ligand-competent and recognizes RGD-containing fibronectin fragments, thus suggesting that the DDL sequence could correspond to an essential RGD-binding element in the integrin ␤ subunits. 6 An interesting feature of the DDL motif is that it systematically includes the highly hydrophobic residue Leu that is adjacent to the Asp-Asp dipeptide (see Table II). Such a hydrophobic environment could help stabilize a salt bridge between the guanidinium (from the RGD arginyl residue) and a carboxylate (from DDL in the ␤ subunit) by increasing the pK a of the interacting carboxylic group (70).
If the integrin DDL sequence plays such a central role in ligand recognition, an attractive mechanism accounting for the release of Mn 2ϩ from ␤ 1 -(121-329), as induced by RGD ligand binding (as well as by Ca 2ϩ binding), would be that the DDL motif in the ␤ I-type domains acts by itself as a metal-coordinating element. As previously noted by D'Souza et al. (18), the second aspartyl residue in DDL occupies the relative position 9 with regard to Asp-1 in the MIDAS-like DXSXS sequence (see Table II). This conserved "Asp-9" could act, in the integrin ␤ subunits, as a specific metal-coordinating residue, as is the case of the aspartyl residue usually found at the relative position 9 in the typical EF-hand loops (7). One possibility would be that Asp-9 in the integrin ␤ subunits switches from a metalcoordinating state (contributing to Mg 2ϩ or Mn 2ϩ coordination) to a ligand-binding state, thus resulting in the loss of the initially bound Mg 2ϩ or Mn 2ϩ ion after binding of the RGD ligand. This hypothetical proposal assumes that the divalent cation-binding site in the ␤ I-type domains departs somewhat from the typical MIDAS coordination in the sense that Asp-9 in the ␤ subunits could contribute to the hexa-coordination of Mg 2ϩ (or Mn 2ϩ ). There is evidence for the occurrence of a long range metal-coordinating residue in the integrin ␤ I-type domains, i.e. Asp-217 in ␤ 3 (13) and Asp-232 in ␤ 2 (63), as in the MIDAS sites (see Table II). However, there is no evidence for the occurrence of another long range metal coordination as observed with Thr in the MIDAS sites of the ␣ I-domains (Table  II). As reported by Tozer et al. (13), substitution of Thr-197 in ␤ 3 (a candidate for the missing MIDAS metal-coordinating Thr residue) did not affect ligand binding function by integrin ␣ IIb ␤ 3 , thus suggesting that no metal-coordinating Thr exists in the MIDAS-type site of this ␤ subunit. It might be that the totally conserved Asp-9 in the integrin ␤ subunit sequences acts as a direct metal-coordinating residue in place of the missing MIDAS Thr residue. Such an hypothesis needs to be tested through site-directed mutagenesis at the level of the DDL sequence. 3 Similarly, Ca 2ϩ could interact with the DDL motif thus rendering Asp-9 unavailable for Mg 2ϩ (or Mn 2ϩ ) binding and therefore accounting for the release of these small divalent cations, in the presence of Ca 2ϩ in excess. If Asp-9 effectively acts as an additional Mg 2ϩ (Mn 2ϩ )-coordinating residue in ␤ 1 (in the absence of any RGD ligand), then the number of protein residues involved in metal coordination would amount to five (see Table II). This is in apparent contradiction with our PRE results that indicate that Mn 2ϩ bound to ␤ 1 -(121-329) remains hydrated by 3 water molecules at least. However, the MIDAS site itself in the ␣ I-domains that involves five protein metal-coordinating residues (see Table II) still retains 3 water molecules directly coordinating Mn 2ϩ (first sphere), as shown by x-ray crystallography (50,51). Two among the five metal-coordinating residues of ␣ L (i.e. Asp-137 and Thr-206; see Table II) do not directly coordinate the central cation and are relayed by one water molecule.
Based on the CD evidence presented in this work, it appears that ␤ 1 -(121-329) does not undergo any apparent conformational change upon formation of the ␤ 1 -(121-329)⅐3Fn7-10 binary complex, although the formation of this complex is likely to be accompanied by the loss of the initially bound Mg 2ϩ ion, as illustrated in Fig. 11A. Strictly speaking, our results only established the loss of Mn 2ϩ from ␤ 1 -(121-329) upon binding of an RGD peptide (Fig. 7). Extension to the case of Mg 2ϩ and macromolecular RGD ligands appears substantiated. However, there is still a somewhat puzzling situation with regard to the conformational features of ␤ 1 -(121-329) in the different complexes investigated in this work. Indeed, if we assume that the apoform of ␤ 1 -(121-329) is present in the binary complex 8 (labeled ␤ 0 in Fig. 11A), a non-null CD difference spectrum 8 ؊ (4 ؉ 5) would then be expected in contrast to what is observed experimentally (see Fig. 10B). We are thus led to the conclusion that, after loss of the metal cation from ␤ 1 -(121-329), the resulting metal-free state (or ␤ 0 ) within the binary complex 8 (see Fig. 11A) corresponds to a conformation that is practically identical, based on the CD data, to that observed with the isolated Mg 2ϩ -loaded form of ␤ 1 -(121-329). One possibility would be that 3Fn7-10 stabilizes the metal-induced ligandcompetent conformation of ␤ 1 -(121-329) within the binary com-plex 8, even after release of the protein-bound Mg 2ϩ ion, as illustrated in Fig. 11A. The observation that the CD difference spectrum, 9 ؊ (3 ؉ 4 ؉ 5) (Fig. 10D), is practically identical to the CD difference spectrum, 7 ؊ (3 ؉ 5) (Fig. 10A), also implies that the ␤-component in the ternary complex 9 adopts a similar conformation as the one observed with the Mg 2ϩ -loaded form of ␤ 1 -(121-329), although it is probable that Mg 2ϩ does not remain bound to the ␤ 1 -component in the ternary complex 9 (see transition 6 to 9 in Fig. 11A). The ionic content of ␤ 1 -(121-329) in this ternary complex still remains an open question. Strictly speaking, we did not establish in this work if any divalent cation is displaced from the ␤-component of the mini-integrin 6 upon interaction with the Fn RGD-containing ligand (conversion from 6 to 9 in Fig. 11A). However, evidence is available that release of Mn 2ϩ from the intact ␣ IIb ␤ 3 integrin occurs upon ligand binding (18), thus suggesting that the ␤ 1 -component in our ternary complex 9 (Fig. 11A) is presumably devoid of any bound divalent cation.
The spectral changes induced at the level of the aromatic chromophores upon formation of the binary complex ␣ 5 ⅐Fn (complex 7 in Fig. 11A), as well as of the ternary complex ␣ 5 ␤ 1 ⅐Fn (complex 9 in Fig. 11A), appear to be associated rather exclusively with conformational changes that occur within the hydrophobic regions of the Fn ligand molecule (see "Results"). Such a conclusion derives from the fact that both CD difference spectra, in Fig. 10, A and D, are quantitatively very close to the difference spectrum previously reported (10) with a mutant form of the isolated ␣ 5 EF-hand domain in which the single Trp-406 residue was substituted by Leu. One possible explanation accounting for the specific conformational deformability of the Fn ligand can be found in the fact that both type III modules of fibronectin, 3Fn9 and 3Fn10, are involved in intermolecular contacts with the integrin ␣ subunit (see above). The Fn adhesive motifs, RGD (in 3Fn10) and PHSRN (in 3Fn9), are distant by 30 -40 Å in the crystal structure of free 3Fn7-10 (23). An analysis of the shape of the minimal ligand-competent EF-hand domain, ␣ 5 -(229 -448), by small angle neutron scattering 7 in solution indicates that this domain corresponds to a globular protein with a nearly spherical distribution of its mass and only includes intramolecular distances well below 30 Å. An adaptation of the Fn ligand, resulting in a reduction of the distance between both sites, RGD in 3Fn10 and PHSRN in 3Fn9, would then be required to lead to a productive complex with the integrin ␣ subunit. This could occur through bending of the multi-module Fn filament at the level of the hinge region between both modules 3Fn9 and 3Fn10. This mechanism appears substantiated by a recent NMR analysis indicating that the isolated 3Fn9 -10 fragment in solution displays a certain degree of flexibility (71). A bending of the 3Fn9 -10 motif at the level of its hinge region could be responsible for changes within the cores of the 9th and/or 10th type III modules of Fn. These structural changes, although subtle, would thus be detected in the near-UV region of the CD spectra (Phe and Trp aromatic chromophores). Site-directed mutagenesis of the Trp residues in Fn7-10 (one tryptophan per type III module), in combination with CD and fluorescence studies, will certainly allows us to better delineate the role of the hydrophobic regions within each Fn type III module on the formation and stability of the integrin-fibronectin complex. 3 The question that finally arises is to know whether all the 7 The shape factor of ␣ 5 -(229 -448) does not exceed 1.1 (the shape factor corresponds to the ratio between the experimental radius of gyration and the gyration radius calculated for a spherical distribution of the protein mass); small angle neutron scattering results to be published (J.-L. Banères, ?. Calmettes, and J. Parello, unpublished information). conformational states identified in this work (see Fig. 11), using our soluble minimized ␣ 5 ␤ 1 integrin (as well as its isolated ␣and ␤-components), are relevant to the function of this integrin receptor under physiological conditions. It appears unlikely that the apoforms, 1 and 2 (Fig. 11), will play any physiological role if one takes into consideration the concentration range of divalent cations in the extracellular environment. One of the remarkable features derived from this work is that the isolated ␤ 1 -(121-329) protein is unable to recognize its Fn ligand in the presence of Ca 2ϩ . This suggests that the Ca 2ϩ to Mg 2ϩ ratio in vivo will be essential for switching the integrin ␤-component from a ligand-competent conformation (high affinity complex) to a ligand-incompetent one (low affinity complex), as illustrated in Fig. 11B. In the extracellular medium where both divalent cations, Ca 2ϩ and Mg 2ϩ , are present at similar concentrations (in the millimolar range), it can be anticipated, based on the measured affinity constants in vitro, that the ␤-integrin subunits will be essentially substituted by Mg 2ϩ within their MIDAS-type sites, whereas the ␣-integrin subunits, with their EF-hand type sites, will correspond to mixed states of occupation by both cations, Ca 2ϩ and Mg 2ϩ . Such hybrid ␣(Ca 2ϩ /Mg 2ϩ )⅐␤(Mg 2ϩ ) complexes (see 6 in Fig.  11B) are expected to be ligand-competent in vivo, as suggested by the results reported in this work. Ligand binding by ␣(Ca 2ϩ / Mg 2ϩ )⅐␤(Mg 2ϩ ) will certainly involve the loss of the Mg 2ϩ ion from the ␤ subunit, whereas the EF-hand type sites of the ␣ subunit will remain occupied by divalent cations (transition 6 to 9 in Fig. 11B). If an imbalance of the Ca 2ϩ to Mg 2ϩ ratio in vivo occurs with Ca 2ϩ concentrations reaching values comparable to the concentrations used in our in vitro studies, the physiologically active form 6 might be converted into an inactive ␣(Ca 2ϩ )⅐␤(Ca 2ϩ ) form (state 6 in Fig. 11B) in which both subunits ␣ and ␤ are occupied by Ca 2ϩ . Liberation of Ca 2ϩ from mineralized bone can lead to an increase of free Ca 2ϩ concentration at the level of the osteoclast up to 40 mM (72). As established in this work, the Ca 2ϩ -loaded form of the ␤ 1 I-type domain is ligand-incompetent (see Fig. 5C, lane 4). It is thus concluded that the Ca 2ϩ -substituted binary complex 6, if present in vivo, will only recognize the RGD-containing Fn ligand through its ␣-component to form a ternary complex 9 with a lower stability than the active ternary complex 9 (Fig. 11B). In this respect, we observed that 3Fn7-10 is still recognized, under in vitro conditions, by the minimized ␣ 5 ␤ 1 integrin in the presence of Ca 2ϩ exclusively, based on our immobilization assay (SDS-PAGE data not shown). However, in the presence of Ca 2ϩ , the immobilized complex, 3Fn7-10⅐␣ 5 ␤ 1 , is more susceptible to dissociation by a competing RGD-containing peptide than is the case in the presence of Mg 2ϩ . 3 One possible explanation is that the binding of 3Fn7-10 to the mini-integrin receptor as a whole becomes significantly weaker in the presence of Ca 2ϩ than in the presence of Mg 2ϩ . It is not clear whether Ca 2ϩ is released from the ␤ subunit upon formation of the low affinity complex 9, by analogy with what is likely to occur upon formation of the high affinity complex 9, in the presence of Mg 2ϩ . Our results with the isolated ␣ 5 and ␤ 1 fragments thus provide a possible explanation for the observed decrease in the affinity of the native ␣ 5 ␤ 1 integrin for fibronectin in the presence of Ca 2ϩ (73). The inefficiency of the ␤ subunit to bind Fn in the presence of Ca 2ϩ (Fig. 5C, lane 4) is likely to be responsible for such a decrease in the affinity of the native integrin for fibronectin. It thus appears that a shift in the relative concentrations of extracellular Ca 2ϩ and Mg 2ϩ that would favor the binding of Ca 2ϩ by the integrin receptor, under in vivo conditions, could limit adhesion, as suggested by Hu et al. (74) in the case of integrin ␣ V ␤ 3 in the bone. As illustrated in Fig. 11B, we anticipate that the conversion of 6 into 6 upon substitution of Mg 2ϩ by Ca 2ϩ within the ␤ I-domain-type region is accompanied by a conformational change of this ␤ domain which adopts an apoform-like structure based on our studies with our soluble ␣ 5 ␤ 1 mini-integrin and its isolated ␣ and ␤ components (see Fig. 11A). Such a prediction needs to be validated experimentally using the difference CD spectroscopy approach used here. 3 Knowledge of the exact conformational features that are associated with divalent cation and ligand interactions needs to await the determination of three-dimensional structures at atomic resolution. Our present study, however, with the soluble mini-integrin ␣ 5 ␤ 1 allowed us to delineate an extended repertoire of conformational events (see Fig. 11) that likely play a direct role in the regulation of EM-cell adhesion.
Recently, Cierniewski et al. (75) have reported the expression of human ␤ 3 -(95-373) in E. coli and established that Ca 2ϩ inhibits the binding of fibrinogen to the isolated ␤ fragment, in contrast to intact ␣ IIb ␤ 3 . As noted by these authors (75), recombinant ␤ 3 -(95-373) binds fibrinogen in the presence of Mn 2ϩ and at low concentrations of Ca 2ϩ , whereas high concentrations of Ca 2ϩ abolish the interaction. These observations are totally consistent with our results with recombinant ␤ 1 -(121-329) establishing that Ca 2ϩ inhibits ligand recognition through release of Mn 2ϩ (and probably Mg 2ϩ ) from the isolated ␤ 1 I-type domain (see "Results"). As discussed above, Ca 2ϩ might interfere with the Mg 2ϩ /Mn 2ϩ -binding site on the ␤ 1 I-type domain by eliciting a common coordinating residue so that Ca 2ϩ , at high concentration, will be able to destabilize the primary MIDAS-type site (with high affinity for Mg 2ϩ or Mn 2ϩ ) by occupying a vicinal site (with low affinity for Ca 2ϩ ). The stoichiometry of Ca 2ϩ binding to the ␤ I-type domain is not known presently.
As a general comment, we note that both mini-integrins, ␣ 5 -(160 -448)⅐␤ 1 -(121-329), described here, and ␣ IIb -(1-233)⅐␤ 3 -(111-318), previously reported by McKay et al. (16), essentially differ by their ␣ components. Both ␣␤ heterodimeric assemblies only display the ␣ repeat III in common. Besides repeat III, our ␣ 5 ␤ 1 mini-integrin displays a full EF-hand type domain (repeats IV through VII), whereas this Ca 2ϩ /Mg 2ϩ -binding domain is completely absent from the ␣ IIb ␤ 3 mini-integrin. Both mini-integrins essentially include a similar ␤ I-type domain taking into account the sequence homologies between both ␤ 1 and ␤ 3 domains (see Table II). Although both mini-integrins display RGD-dependent ligand recognition, it is not clear if they do have the same cation-binding requirements. The lack of the ␣ IIb EF-hand domain in the ␣ IIb ␤ 3 mini-integrin, combined with the RGD-dependent recognition of fibrinogen by the minimized integrin receptor, would suggest that the Ca 2ϩ /Mg 2ϩbinding domain of ␣ IIb is not required for ligand recognition. However, Gulino et al. (9) using the recombinant construct ␣ IIb -(171-464), i.e. repeats III to VII, showed that this isolated domain from ␣ IIb is ligand-competent and recognizes fibrinogen upon occupation of all four sites by divalent cations (Ca 2ϩ , Mg 2ϩ , and Mn 2ϩ ). Our isolated ␣ 5 EF-hand domain, ␣ 5 -(160 -448), that also encompasses repeats III to VII recognizes the RGD-containing Fn ligands in a divalent cation-dependent manner (Ref. 10 and this work). However, it has been shown that the isolated fragment ␣ 5 -(229 -448) that only encompasses repeats IV to VII is similarly ligand-competent in an RGD-and divalent cation-dependent manner (10). This result emphasizes the central role of the ␣ 5 EF-hand type sites as part of the integrin ligand recognition pocket. 5 It is likely that in both integrins ␣ IIb ␤ 3 and ␣ 5 ␤ 1 , the ␣ EF-hand domains play a similar role with regard to their cation-and ligand-binding properties. In this respect, the ␣ IIb ␤ 3 mini-integrin reported by McKay et al. (16) lacks such a contribution from the ␣ IIb subunit. However, the ligand competence of the mini-integrin ␣ IIb ␤ 3 can be explained based on the model of Fig. 11C, since part of the interactions involved in the stabilization of the ligand-integrin complex (interaction between the RGD guanidinium and the DDL site in the ␤ subunit) is still possible between the ␣ IIb ␤ 3 mini-integrin and the RGD ligand.
In closing, the soluble ␣ 5 ␤ 1 mini-integrin, ␣ 5 -(160 -448)⅐␤ 1 -(121-329), with ϳ500 residues out of a total of ϳ1800 in the native ␣ 5 ␤ 1 integrin, needs to be viewed as a molecular assembly mimicking most, if not all, of the essential interactions that are encountered between the extracellular part of the integrin receptor and its RGD-containing fibronectin ligand. We note, however, that both recombinant fragments ␣ 5 -(160 -448) and ␤ 1 -(121-329) from human integrin ␣ 5 ␤ 1 lack the potential Nlinked glycosylated moieties, and this could affect the binding properties of the fibronectin ligands in comparison to the glycosylated native receptor. Of particular interest is the fact that in ␣ 5 three potential glycosylation sites (i.e. Asn-216, Asn-266, and Asn-275) are flanking the EF-1 domain in repeat IV. This EF-hand is essential for ligand recognition. 5 Our minimized ␣ 5 ␤ 1 construct with two structurally well defined domains, from both its ␣ and ␤ subunits, apparently includes all cationbinding sites that occur in the native ␣ 5 ␤ 1 heterodimeric integrin and certainly most of the structural elements that make up the fibronectin-binding pocket, as well as the essential elements that are involved in ␣␤ heterodimerization. This soluble recombinant ␣ 5 ␤ 1 integrin therefore appears to be well adapted for analyzing the molecular bases of integrin-ligand recognition at the structural and pharmacological levels.