Jararhagin-derived RKKH Peptides Induce Structural Changes in α1I Domain of Human Integrin α1β1*

Integrin α1β1 is one of four collagen-binding integrins in humans. Collagens bind to the αI domain and in the case of α2I collagen binding is competitively inhibited by peptides containing the RKKH sequence and derived from the metalloproteinase jararhagin of snake venom from Bothrops jararaca. In α2I, these peptides bind near the metal ion-dependent adhesion site (MIDAS), where a collagen (I)-like peptide is known to bind; magnesium is required for binding. Published structures of the ligand-bound “open” conformation of α2I differs significantly from the “closed” conformation seen in the structure of apo-α2I near MIDAS. Here we show that two peptides, CTRKKHDC and CARKKHDC, derived from jararhagin also bind to α1I and competitively inhibit collagen I binding. Furthermore, calorimetric and fluorimetric measurements show that the structure of the complex of α1I with Mg2+ and CTRKKHDC differs from structure in the absence of peptide. A comparison of the x-ray structure of apo-α1I (“closed” conformation) and a model structure of the α1I (“open” conformation) based on the closely related structure of α2I reveals that the binding site is partially blocked to ligands by Glu255 and Tyr285 in the “closed” structure, whereas in the “open” structure helix C is unwound and these residues are shifted, and the “RKKH” peptides fit well when docked. The “open” conformation of α2I resulting from binding a collagen (I)-like peptide leads to exposure of hydrophobic surface, also seen in the model of α1I and shown experimentally for α1I using a fluorescent hydrophobic probe.

Integrins are a large family of cell surface glycoproteins that mediate cell-cell, cell-extracellular matrix, and matrix-matrix adhesion and transduce bidirectional signals between the cytoplasm and the extracellular matrix or other cells (1). The adhesive function of integrins is important in various physiological processes such as platelet aggregation, inflammation, wound healing, tumor metastasis, cell migration during embryogenesis, viral infections, and other diseases (1,2). Orthologues of human integrins have been identified in the genomes of birds, amphibians, and bony fish but so far not in the urochordates or in other invertebrates (3).
In humans, over 20 different integrin ␣␤ heterodimers are formed from eight different ␤ subunits (2) and 18 different ␣ subunits (4). The x-ray structure of integrin ␣ V ␤ 3 (5) defined much of the heterodimer common to all ␣␤ integrins, where 12 domains assemble forming the "head" region where ligands bind plus two "tails" leading to the transmembrane helices that anchor integrins to the cell surface. The subunit interface between ␣ V and ␤ 3 is mainly formed by a seven-bladed ␤-propeller domain (␣ V ) and an I-like domain (␤ 3 ). The structure of ␣ V ␤ 3 in complex with a cyclic pentapeptide with the arginine-glycineaspartate (RGD) motif showed that the RGD sequence binds into a crevice between the ␤-propeller domain and the ␤I-like domain (6). Upon binding the RGD pentapeptide, the tail regions of ␣ V ␤ 3 come closer to each other, and the ␤-propeller domain rotates slightly.
Nine integrin ␣ domains have an extra ϳ200-residue inserted domain, the ␣I domain (for a review, see Ref. 7). The ␣I domain is predicted to locate on the side and top of the ␤-propeller domain; currently, the only representative structure of an integrin heterodimer, ␣ V ␤ 3 , lacks the ␣I domain (8). In the ␣I domain-containing integrins, the ␣I domain plays a major role in ligand binding, primarily to the metal ion-dependent adhesion site, MIDAS. 1 The collagen binding ␣I domains (␣ 1 , ␣ 2 , ␣ 10 , and ␣ 11 ) are distinguished from the ␣I domains of the immune system (␣ D , ␣ E , ␣ L , ␣ M , ␣ X ) by having an additional ␣-helix, helix C, near MIDAS and can be divided into two groups according to differences in their collagen binding specificities (9). Integrin ␣ 2 ␤ 1 and ␣ 11 ␤ 1 bind best to fibrillar collagens I-III, whereas integrins ␣ 1 ␤ 1 and ␣ 10 ␤ 1 have higher affinity for network-forming collagen IV and for beaded filament-forming collagen VI (10 -12). Furthermore, ␣ 1 ␤ 1 can attach to collagen XIII, whereas ␣ 2 ␤ 1 cannot (13). To date, only one complex structure with "collagen" is available, the ␣ 2 I structure in complex with a collagen (I)-like peptide (14). In comparison with the apo-form of ␣ 2 I (15), a conformational change takes place on binding the * This work was supported by the Academy of Finland, the Graduate School in Informational and Structural Biology, Svenska Kulturfonden, the Sigrid Jusélius Foundation, the Tor, Joe, and Pentti Borgs Memorial Fund, the Foundation of Åbo Akademi University, and the European Community Access to Research Infrastructure Action of the Improving Human Potential Programme (to the EMBL Hamburg Outstation), contract number HPRI-CT-1999-00017. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The atomic coordinates and structure factors ( collagen (I)-like peptide, which may reflect a general feature of ligand binding at MIDAS common to the collagen-binding I domains.
Jararhagin is a snake venom metalloproteinase and a well known inhibitor of ␣ 2 ␤ 1 integrin. Jararhagin and other class P-III snake venom metalloproteinases such as atrolysin A are composed of a metalloproteinase domain, a non-RGD-containing disintegrin domain and a cysteine-rich domain. The exact binding mechanism of snake venom metalloproteinases to ␣ 2 ␤ 1 integrin is unknown. Previous studies have indicated that peptides derived from all three domains can inhibit ␣ 2 ␤ 1 function (16 -19), and therefore, snake venom metalloproteinases might contain several integrin recognition sites. We have shown that a cyclic RKKH peptide derived from metalloproteinase domain of jararhagin recognizes the ␣ 2 I domain and prevents collagen binding to ␣ 2 ␤ 1 integrin (17,18). Already our first results suggested that RKKH might not be specific to ␣ 2 I domain but that it may also bind to another integrin I domain, such as ␣ 1 I (17). Integrin binding to natural ligands, such as collagen, activates cellular signaling, and, based on current knowledge, ligand binding triggers a conformation change in the ␣I domain that leads to changes in the interaction between the ␣ and ␤ subunits, resulting in the separation of their intracellular domains (20). Importantly, jararhagin can trigger ␣ 2 ␤ 1 integrin signaling in cells, and, therefore, jararhagin is a novel tool in the study of the integrin structure-function relationship (21). Here, we have used RKKH peptides to study whether a small peptide can mimic natural ligands and induce the first step in integrin signaling, namely a conformational change in the integrin ␣ 1 I domain.

EXPERIMENTAL PROCEDURES
Peptides-Two different peptides cyclized via terminal cysteine residues were used in the binding studies. CTRKKHDC is directly based on the jararhagin metalloproteinase sequence (17); in the CARKKHDC peptide, threonine of CTRKKHDC was replaced with alanine. All other experiments were made on the CTRKKHDC peptide only. Synthesis of the peptides was described in Ivaska et al. (17).
Binding Assay for Europium-labeled ␣ 1 I-Human recombinant integrin ␣ 1 I was expressed as reported earlier (13). The integrin ␣ 1 I binding assay to collagen I was performed as previously described (13,17); collagen I was precoated on microtiter plate wells, the wells were blocked with bovine serum albumin to prevent nonspecific binding, and ␣ 1 I was allowed to bind to collagen in the absence or presence of inhibitory peptides. Detection of ␣ 1 I binding to collagen was based on collagen-bound europium-labeled ␣ 1 I and time-resolved fluorescence measured by fluorimetry (Delfia model 1232; Wallac/PerkinElmer Life Sciences).
In order to measure the binding of ␣ 1 I to the cyclic peptides, peptides were precoated on amine microtiter plate wells according to the manufacturer's instructions, and the integrin ␣ 1 I binding assay was performed as described previously (17). Wells were blocked with bovine serum albumin, and europium-labeled ␣ 1 I was introduced. Bound ␣ 1 I was detected using time-resolved fluorescence measured by fluorimetry, as above.
Differential Scanning Calorimetry-In order to study the thermostability of ␣ 1 I, differential scanning calorimetry (Nano II; Calorimetry Sciences Corp.) was used. For this purpose, 31.8 nmol of ␣ 1 I in PBS with either 2 mM EDTA or 2 mM MgCl 2 was prepared. In order to examine how the presence of peptide affected the stability of the ␣ 1 I, 95.4 nmol of peptide was added to PBS containing 2 mM MgCl 2 . All samples were incubated for 7 h at room temperature (ϳ23°C) without stirring and were degassed before being transferred to the cell for calorimetric measurements. The reference cell was filled with PBS. The heat capacity was recorded as a function of temperature, between 0 and 100°C at a rate of 1°C/min. The base line was subtracted, and the enthalpy (⌬H), the entropy (⌬S), and the transition midpoint temperature (T m ) were calculated using the software provided by Calorimetry Sciences Corp. The van't Hoff enthalpy (⌬H vH ) was calculated in the usual way according to Equation 1, where R is the gas constant, T m is the temperature at which the base line-subtracted and concentration-normalized heat capacity function (͗⌬C p ͘ tr ) has a maximum, and ͗⌬C p ͘ tr, max is the measured heat capacity at the peak maximum. The value of the ratio ⌬H vH /⌬H cal is often used as a measure of cooperativity. In the case of small globular proteins, a folding/unfolding ratio of 1 is generally assumed to be evidence for a two-state transition (22,23), whereas larger values indicate the presence of additional intermediates along the unfolding pathway. Determination of the Free Mg 2ϩ Concentration-The fluorescent Mg 2ϩ chelator Mag-Fura-2 (Molecular Probes, Inc., Eugene, OR) was used to monitor changes in the concentration of free Mg 2ϩ in the sample buffer (24). For this purpose, 100 nmol of ␣ 1 I was incubated at room temperature in PBS containing 2 mM MgCl 2 for 2 h, in order to saturate the protein with Mg 2ϩ . Excess Mg 2ϩ was then removed, using a PD-10 column (Amersham Biosciences), after which the sample was transferred to a cuvette, and Mag-Fura-2 was added to a final concentration of 2.05 M. The Mg 2ϩ concentration was determined from the ratio of the intensities of emitted light following excitation at 340-and 398-nm wavelengths. The emission wavelength was set to 483 nm, and, as the excitation ratio was being monitored, 297 nmol of CTRKKHDC peptide was added. In order to calibrate the system, 100 nmol of MgCl 2 was injected into the ␣ 1 I solution before the peptide was injected. As a control, we added peptide and MgCl 2 to PBS buffer pretreated in a similar manner as the ␣ 1 I solution.
Fluorescence Spectroscopy-In all of the fluorimetric experiments, we used a Quantamaster 1 steady-state spectrofluorimeter (Photon Technology International) with the slit width set to 4 nm unless stated otherwise. The experiments were conducted in a 10-mm quartz cuvette at room temperature. The solution was kept under constant stirring (300 rpm). The molar ratio of peptide to protein was 3:1 in all fluorimetric experiments.
Tryptophan emission spectra were measured on solutions containing 10 M ␣ 1 I. In order to examine whether the emission spectra of ␣ 1 I was affected by the presence of Mg 2ϩ , measurements were made in PBS with and without 2 mM MgCl 2 . The effect of peptide interactions with ␣ 1 I was studied in the presence of 30 M peptide and 2 mM MgCl 2 . The excitation wavelength was set to 295 nm. All spectra were recorded as the average of three scans. Quenching of tryptophan fluorescence by KI (Sigma) was measured with a solution containing 3.75 nmol of ␣ 1 I in the presence and absence of MgCl 2 and 10.5 nmol of peptide was added to the ␣ 1 I solution with MgCl 2 . The excitation wavelength was set to 295 nm, and the emission due to tryptophan was measured at 328 nm, which is the emission maximum. KI was injected into the cuvette in 50-mol increments from a 5 M stock solution. In order to prevent oxidation of iodide, 0.1 mM sodium thiosulfate (Na 2 S 2 O 3 ; Merck) was included in the KI solution.
Bis-ANS (Molecular Probes, Inc.) is a fluorescent probe that binds specifically to hydrophobic surfaces on proteins (25). Bis-ANS binding to 10 nmol of ␣ 1 I was measured in a PBS solution in the presence and absence of 2 mM MgCl 2 containing 1 nmol of bis-ANS. In order to examine how the presence of Mg 2ϩ affected the fluorescence due to bis-ANS, we added 2 mM MgCl 2 to the Mg 2ϩ -free buffer while monitoring the fluorescence intensity. Changes in hydrophobic surface due to ␣ 1 I-peptide interactions was studied by the addition of 30 nmol of peptide to the 2 mM MgCl 2 -PBS while the fluorescence intensity was being recorded. The excitation wavelength was 370 nm, and the bis-ANS emission was measured at 470 nm. Light scattering was simultaneously measured on another detector (370 nm; a small slit width was used to reduce the amount of light reaching the detector) in order to detect aggregation.
Crystallization and Data Collection-The protein was expressed, purified, and crystallized according to the previously described protocol (26). Briefly, the I domain (Swiss-Prot number P56199; residues 138 -338) of the human integrin subunit ␣ 1 was expressed as a recombinant glutathione S-transferase fusion protein in Escherichia coli and purified by affinity chromatography. The protein was concentrated to 15-30 mg/ml, and initial crystallization conditions were screened with the sparse matrix screen (27) (Hampton Research) using the vapor diffusion method. The initial crystallization conditions were systematically refined and shown to be suitable for x-ray analysis. The crystals were obtained using a drop containing 13 mg/ml protein and 0.25 mM ligand and the well containing 30% (w/v) polyethylene glycol 6000, 100 mM Tris, pH 8.5, 0.2 M sodium acetate, and 15% (v/v) glycerol.
The x-ray data were collected from a single crystal at 100 K, flashfrozen using the Oxford cryosystem at the Turku Center for Biotechnology, using the X11 beamline at the EMBL-Hamburg Outstation and using a MAR CCD detector. The data were indexed, integrated, scaled, and reduced with the XDS software package (28). The crystal has symmetry consistent with the space group P2 1 with cell dimensions a ϭ 37.5 Å, b ϭ 97.6 Å, c ϭ 53.1 Å, ␣ ϭ ␥ ϭ 90°, and ␤ ϭ 103.6°, and there are two ␣ 1 I molecules in the asymmetric unit according to the calculated Matthews coefficient (29) of 2.2 Å 3 /Da. Details from the data collection are presented in Table I.
Structure Determination, Model Building, and Refinement-The ␣ 1 I structure was solved by molecular replacement. CNS (30) was used in the molecular replacement work, whereas the CCP4 suite (31) was used for refinement. The software O (32) was used to display electron density maps and for model building. Before any structure solution was attempted, 5% of the reflections were randomly set aside and used only for calculations of the cross-validated error function R free (33) but never in refinement or map calculations. Since we have already solved an ␣ 1 I structure at 2.3 Å (26) (Protein Data Bank code 1QCY), we used that structure as the MR search model in order to solve the structure of ␣ 1 I at 1.87-Å resolution. For refinement of the structure, the maximum likelihood method implemented in Refmac5 (34) was used within the CCP4 suite. Water molecules, having reasonable hydrogen bonding geometry to the protein, were added to the models in places where the F o Ϫ F c electron density maps had a peak value 3 above the mean.
Finally, the structure was checked with the Whatcheck program in the software package WHATIF (35) and with Procheck (36). The coordinates have been submitted to the Protein Data Bank (code 1PT6).
Peptide Modeling-A model structure of the TRKKHD peptide was created based on the theoretical model of the metalloproteinase jararhagin (18) (Protein Data Bank code 1C9G). A cysteine residue was added to both the amino terminus and the carboxyl terminus, and the peptide was cyclized via a disulfide bond built between them. For molecular dynamics, a simulation box containing the peptide was filled with solvent (water described by the simple point charge model) (37)). The ␦-nitrogen of the histidine residue was protonated, whereas the ⑀-nitrogen was left unprotonated, enabling the formation of stabilizing intramolecular hydrogen bonds as seen in the metalloproteinase structures of adamalysin II from Crotalus adamanteus (Protein Data Bank code 1IAG (38)) and acutolysin A from Agkistrodon acutus (Protein Data Bank code 1BUD (39)). The system was simulated for 1 ns at constant temperature (298 K; the temperature of the solvent and peptide were controlled separately) and pressure (1 bar), both controlled using the Berendsen weak coupling method (40). Bond lengths were constrained using the LINCS algorithm (41). Short range electrostatic interactions were calculated with a cut-off of 1 nm, and long range electrostatic interactions were treated by the fast particle mesh Ewald method (42,43). The system was sampled with a time step of 1 fs using a leapfrog integrator (44). Periodic boundary conditions were used. Molecular dynamics was performed using GROMACS (45,46) and the GROMACS force field. The CARKKHDC peptide was created by exchanging alanine for threonine of the simulated CTRKKHDC.
Integrin ␣ 1 I Modeling-The structure of the "open" conformation of the integrin ␣ 1 I domain was modeled. The sequence of human ␣ 1 I was aligned against the sequence of human ␣ 2 I using the program MALIGN (47,48) in the Bodil modeling environment. 2 Since the structures of apo-␣ 1 I and apo-␣ 2 I are very similar to each other (when equivalent C␣ atoms of the structures are optimally superimposed, the structures differ by a root mean square deviation of 0.8 Å), a model of ␣ 1 I in the "open" conformation was created with the modeling program HO-MODGE in Bodil, using as the template the structure of ␣ 2 I in complex with a collagen (I)-like mimetic peptide (PDB code 1DZI (14)), which is in the "open" conformation. The sequence identity between the ␣ 1 I and ␣ 2 I sequences is high (52%), and experience suggests that the model structure of ␣ 1 I, "open" form, is likely to be of high quality. The carboxyl-terminal helix, located well away from MIDAS, was built manually in Sybyl version 6.9 (Tripos Inc.), since it was truncated in the template structure (1DZI contains residues 142-326 of ␣ 2 I). The final model structure of ␣ 1 I contained residues 142-329. The rotamer library (50) implemented within Bodil was used to optimize the conformations of amino acid side chains.
Ligand Docking and Molecular Dynamics Simulation-The cyclic CTRKKHDC peptide was manually docked to the ␣ 1 I model structure with the aid of a grid map for the NH 3 ϩ probe calculated with the program GRID, version 20 (51). Molecular dynamics simulations with GROMACS were performed with the same settings as for the peptide alone, unless stated otherwise. A proton was attached to the ⑀-nitrogen atom of His 192 , His 257 , and His 260 and to the ␦-nitrogen of His 310 in ␣ 1 I, based on calculations with the program Whatcheck (35). New side-chain rotamers were selected, where appropriate, for the amino acid side chains of the ligand. For computational efficiency, a simulation box with the shape of a dodecahedron was created, containing the receptorligand complex and surrounding solvent (simple point charge model water). Initially, a 10-ps simulation with a time step of 1 fs was performed in order to equilibrate the solvent. During this run, all atomic positions of the receptor-ligand complex were restrained by using a force constant of 1000 kJ mol Ϫ1 nm Ϫ2 per axis. During the next 100 ps of the simulation, the receptor was given more freedom of movement by only restricting the C␣ atoms of the protein main chain, again with a force constant of 1000 kJ mol Ϫ1 nm Ϫ2 . A time step of 2 fs was used in order to reduce the overall computational time needed for the simula- tion. After the simulation was completed, the resulting complex was energy-minimized using Sybyl in order to remove noise resulting from thermal vibrations.

Calculating the Hydrophobic Surface Area of the Model Structures-
The hydrophobic surface area of ␣ 1 I (residues 142-333) was calculated with MOLCAD (Sybyl) for the structure of the "closed" conformation (chain A) without peptide and for the model structure of the "open" conformation with and without CTRKKHDC peptide. Surfaces were created with the Fast Connolly method (52, 53) using a probe radius of 1.4 Å. The lipophilic surface potentials were calculated using the "protein" lipophilicity potential function (54) and the "new" Crippen type atomic partial lipophilicities (55). A global scale with the minimum value set to Ϫ0.3 and the maximum value set to 0.15 was used. Lipophilic surface area was contributed by regions having values greater than either 0.0 or Ϫ0.05; similar results were obtained, and the average percentage change in lipophilic surface area is reported using these two limits.

RESULTS
Binding Assays-The binding of recombinant human integrin ␣ 1 I to jararhagin-derived cyclic peptides was studied using two different binding assays (Fig. 1). The binding of ␣ 1 I to collagen I was tested in the absence and presence of CTRKKHDC and CARKKHDC. Significant inhibition of ␣ 1 I binding was detected, CARKKHDC peptide being the better inhibitor (Fig. 1A). The second binding assay was performed using peptide-coated microtiter plate wells, and the binding of ␣ 1 I to the peptides was measured. Both peptides bound to ␣ 1 I directly, and CARKKHDC bound better than CTRKKHDC (Fig. 1B).
Calorimetric and Fluorimetric Studies-Differential scanning calorimetry was used to measure the effects of Mg 2ϩ and CTRKKHDC peptide on the stability of ␣ 1 I (Fig. 2). EDTAtreated ␣ 1 I was least stable, with T m ϭ 43.9 Ϯ 0.1°C, ⌬H ϭ 190.9 kJ/mol, and ⌬S ϭ 0.60 kJ/K⅐mol (Table II). The addition of MgCl 2 stabilized ␣ 1 I considerably, raising the T m to 46.4°C. Both ⌬H (199.3 kJ/mol) and ⌬S (0.62 kJ/K⅐mol) also increased in the presence of MgCl 2 . The addition of peptide to the Mg 2ϩenriched ␣ 1 I solution led to a further increase in the T m , to 46.9°C, as well as increases in ⌬H (220 kJ/mol) and ⌬S (0.69 kJ/K⅐mol). In addition to providing information on stability, differential scanning calorimetry also provides information on the unfolding pathway. The unfolding process was cooperative, ⌬H vH /⌬H Ͼ 2 (Table II), whether metal and/or peptide were present or not, indicating that the unfolding of ␣ 1 I cannot be represented by a simple two-state process but includes additional intermediate states.
Mag-Fura-2 is a fluorescent chelator of Mg 2ϩ , useful for measuring the amount of free Mg 2ϩ in solution. Before the addition of Mag-Fura-2, ␣ 1 I was saturated with Mg 2ϩ , and excess free Mg 2ϩ was removed with a PD-10 desalting column. Mag-Fura-2 was then added to the solution, and the free Mg 2ϩ concentration was monitored (Fig. 3). After 200 s of base line measurement, MgCl 2 was added, and at 400 s, CTRKKHDC peptide was then added. The addition of peptide to the control buffer caused a shift in the excitation ratio equal to the presence of 65 Ϯ 1.6 nmol of Mg 2ϩ in the solution, whereas less free Mg 2ϩ , 39 Ϯ 0.26 nmol, was seen in the ␣ 1 I solution, supporting the notion that, like ␣ 2 I, the metal ion is required for RKKH peptide binding to ␣ 1 I.
Tryptophan emission spectra were measured in order to detect any changes in the environment near the lone Trp 158 in ␣ 1 I, located at the amino-terminal end of helix 1 and buried between helix 1 and helix 2, in the presence and absence of FIG. 2. Unfolding of ␣ 1 I measured using differential scanning calorimetry. ␣ 1 I (31.8 M) in PBS containing EDTA (curve 1), MgCl 2 (curve 2), or MgCl 2 and peptide (curve 3). ␣ 1 I was incubated for 7 h at room temperature (without stirring) before the temperature scan. FIG. 1. Binding of europium-labeled ␣ 1 I to jararhagin-derived peptides. A, binding of europium-labeled ␣ 1 I to collagen I in the absence and presence of jararhagin-derived peptides. Wells were coated with rat tail collagen I overnight (5 g/cm 2 ). Integrin ␣ 1 I (1 g/ml) was allowed to attach for 3 h in the presence of 2 mM MgCl 2 in PBS, 500 M peptide; wells were then washed three times, and signal was measured due to europium. B, binding of europium-labeled ␣ 1 I to jararhagin-derived peptides. Amine binding microtiter plate wells were coated with peptide (1 mg/ml) in the presence of PBS, pH 8.5, for 1 h. Bovine serum albumin was used as the control. Integrin ␣ 1 I (1 g/ml) was allowed to attach to peptides in the presence of 2 mM MgCl 2 in PBS for 3 h. Wells were washed three times, and europium signal was measured. In all cases, the means Ϯ S.E. of three parallel measurements are reported. MgCl 2 and CTRKKHDC peptide. We found that the addition of Mg 2ϩ led to an increase in the intensity of tryptophan fluorescence (Fig. 4A), which was quenched (16% reduction in fluorescence) by the addition of KI (Fig. 4B). Tryptophan fluorescence of the Mg 2ϩ -free control solution was only marginally quenched (about 4%) by the addition of KI.
The effects of MgCl 2 and peptide on the amount of hydrophobic surface present in ␣ 1 I were probed using bis-ANS, whose fluorescence was measured (Fig. 5A). In order to detect any aggregation caused by peptide binding, light scattering was simultaneously measured using a second detector (Fig. 5B). The addition of MgCl 2 resulted in an increased emission of light from bis-ANS, suggesting that more bis-ANS bound to ␣ 1 I, whereas no change in the scattering of light was seen. The increase in fluorescence intensity was only 2.6 Ϯ 0.2%, but still it suggests that the hydrophobic surface of the protein increased due to interactions with Mg 2ϩ . The addition of the CTRKKHDC peptide to ␣ 1 I (containing MgCl 2 ) led to a more prominent increase in the fluorescence intensity of 10 Ϯ 1%. Peptide addition, however, also led to increased light scattering. The kinetics for the increase in emission and scattering were quite similar, suggesting that they occurred simultaneously (Fig. 5). An increased scattering of light could be caused by aggregation of the proteins, and aggregation could result from the exposure of the hydrophobic surface on ␣ 1 I.
X-ray Structure of ␣ 1 I-Our current refined structure of ␣ 1 I contains two molecules in the asymmetric unit, A and B, respectively with 192 (positions 142-333) and 195 (positions 143-337) residues, a total of 279 water molecules, and one glycerol molecule and one Mg 2ϩ ion per monomer (Table I). For both of the monomers, no electron density was detected for the five carboxyl-terminal residues or for the ligand added to the protein solution. The structure has well defined geometry, and the values for R and R free are 19.1 and 24%, respectively. In the final maps, most of the side chains, even those in loop regions, showed good electron density; this is also reflected by the fact that the B-factors converged to low values (Table I). Residues that showed weak electron density upon visual inspection of the final maps or that participate in crystal contacts are listed in Table I. The overall architecture is the classic ␣/␤ "Rossmann" fold (Fig. 6A). The core of the molecule is a ␤-sheet consisting of six ␤-strands (A-F) with a ϩ1x, ϩ1, Ϫ3, Ϫ1x, Ϫ1x  3. Effects of ligand binding on the concentration of free Mg 2؉ measured using the fluorescent chelator Mag-Fura-2. 100 nmol of ␣ 1 I was incubated at room temperature in PBS containing 2 mM MgCl 2 for 2 h, in order to saturate the protein with Mg 2ϩ . Excess Mg 2ϩ was then removed, using a PD-10 column (Amersham Biosciences), after which the sample was transferred to a cuvette, and Mag-Fura-2 was added to a final concentration of 2.05 M. The Mg 2ϩ concentration was determined from the ratio of the intensities of emitted light following excitation at 340-and 398-nm wavelengths. The emission wavelength was set to 483 nm, and as the excitation ratio was being monitored, 297 nmol of CTRKKHDC peptide was added. In order to calibrate the system, 100 nmol of MgCl 2 was injected into the ␣ 1 I solution before (arrow 1) the peptide was injected (arrow 2). As a control, we added peptide and MgCl 2 to PBS buffer pretreated in a similar manner as the ␣ 1 I solution.
FIG. 4. Intrinsic fluorescence of Trp 158 on ␣ 1 I. A, the emission spectra measured in PBS without Mg 2ϩ (curve 1), with 2 mM MgCl 2 (curve 2), and with 2 mM MgCl 2 and peptide (curve 3). B, Stern-Volmer plots of KI quenching of tryptophan fluorescence in ␣ 1 I with 2 mM EDTA (circles), 2 mM MgCl 2 (squares), and 2 mM MgCl 2 plus CTRKKHDC peptide (triangles). All measurements were performed at room temperature, and the protein to ligand ratio was 1:3. topology (56), and this core is flanked by ␣-helices: helices 1, 3, and 7 on one side and helices 4, 5, and 6 on the other. A short additional ␣-helix, helix C, is also present close to the ligandbinding region.
Model of the "Open" Conformation of Integrin ␣ 1 I with Docked Ligand-Experimental studies (Fig. 5) have indicated that the CTRKKHDC peptide binds to a conformation of ␣ 1 I different from the apo-␣ 1 I structure seen in the absence of ligand (Fig. 6). The ␣ 1 I sequence shares 52% sequence identity with ␣ 2 I, and the structures of the apo-forms are highly similar. The close similarity between the ␣ 1 and ␣ 2 I domains provides a basis for building a high quality model of the ligand-binding form of ␣ 1 I. Near MIDAS, the ligand-bound "open" conformation structure of ␣ 2 I (1DZ1) exhibits substantial conformational differences from the apo-␣ 2 I structure (1AOX): the root mean square deviation calculated for all C␣ atoms of the "closed" versus "open" conformations of ␣ 2 I is 1.6 Å overall and 2.9 Å when the C␣ atoms of helix C plus those C␣ atoms within 10 Å of MIDAS are considered.
The ␣ 2 I "open" conformation (1DZI) differs from the ␣ 2 I "closed" conformation (1AOX) in four major respects (14): 1) the coordination of the metal at MIDAS changes, whereby the metal moves 2.6 Å toward Thr 221 forming a direct bond with it while Asp 254 looses its direct bond to the metal, instead forming a water-mediated bond; 2) helix C unwinds, whereas helix 6 is extended by one extra turn; 3) helix 7 shifts ϳ10 Å away from MIDAS; and 4) the conserved salt bridge formed between Arg 288 and Glu 318 is broken. Each of these differences is also reflected in the modeled structure of the "open" conformation built for ␣ 1 I (Fig. 6B).
Studies on ␣ 2 I (17, 18, 57) have localized the CTRKKHDC peptide-binding site near MIDAS. The peptides were docked to the modeled structure of the "open" conformation of ␣ 1 I, and the resulting complexes suggest that extensive interactions between the 1 CTRKKHDC 8 peptide and ␣ 1 I take place (Fig. 7A): 1) the amino-terminal amine nitrogen atom of Cys 1 forms a hydrogen bond with the side-chain hydroxyl group (OG) of Ser 291 ; 2) the side chain hydroxyl group (OG) of Thr 2 forms a hydrogen bond with the main-chain carbonyl oxygen atom of Ser 291 ; 3) the positively charged guanido group of Arg 3 is involved in an extensive network of hydrophilic interactions with Ser 301 , Glu 297 , and Glu 298 , and the side chains of the latter two residues form a negatively charged patch located between helix 5 and helix 6; 4) the side chains of Lys 4 and Asn 259 interact with each other; and 5) the main-chain carbonyl oxygen atom of Lys 5 binds to His 257 (NE2) and Lys 5 (NZ) and is located within hydrogen bonding distance of both the main-chain carbonyl oxygen atom of Arg 218 and the side chain oxygen atom (OE) of Glu 255 . In addition, the main-chain nitrogen atoms of both Lys 4 and Lys 5 form hydrogen bonds with the backbone carbonyl oxygen atom of Ser 256 . In the metalloproteinase structures, His 92 (corresponding to His 6 of the peptides) stabilizes the loop conformation, forming a hydrogen bond to the main-chain oxygen atom of Lys 243 (corresponding to Lys 4 of the peptides). A similar role is proposed for His 6 in the peptides, which is supported by the effects on binding of the H6A mutant (CTRKKHDNAQC peptide) to ␣ 2 I (17). The same interactions are seen in the model of the CARKKHDC-␣ 1 I complex, with the exception of the sequence difference at position 2 of the peptide; alanine cannot form the hydrogen bond formed by Thr 2 (OG) in the CTRKKHDC peptide with Ser 291 (O). The smaller hydrophobic side chain of alanine is a better fit than threonine to the available space on ␣ 1 I as well as being ideal for interactions with a hydrophobic patch near MIDAS formed by the Phe 295 side chain (Fig. 7B), explaining the better binding of CARKKHDC compared with CTRKKHDC.
Hydrophobic surface areas were calculated for the apo-structure of ␣ 1 I in the "closed" conformation and for the modeled structure of the "open" conformation of ␣ 1 I with and without docked CTRKKHDC peptide. The total surface area of the "open" conformation of ␣ 1 I increases by 10% (no bound peptide) and 12% (bound peptide included; the peptide contributes only about 2% toward the total surface area) with respect to the "closed" conformation. In comparison with the "closed" conformation (Fig. 8A), the "open" conformation ( Fig. 8B) of ␣ 1 I has on average 22% more hydrophobic surface both in the absence and presence of peptide. The increase in hydrophobic surface area appears to be due to the hydrophobic side chains of Ala 280 , Leu 282 , Thr 292 , Val 296 , and Phe 311 at the elongated helix 6 of the "open" conformation. In addition, Tyr 156 and Tyr 285 that are close to each other in the "open" conformation form a hydrophobic patch not seen in the "closed" conformation ( Fig.  8). Similar changes are seen between the x-ray structures of the "closed" (Fig. 8C) and "open" (Fig. 8D) conformations of ␣ 2 I. DISCUSSION Different conformations have been observed in some crystal structures of ␣I domains solved in the presence and absence of ligands. To date, four complexes are available: ␣ 2 I with bound collagen (I)-like peptide (14), ␣ L I with both ICAM-1 and lovastatin (58,59), and ␣ 1 I with a Fab fragment (60). Lovastatin does not bind at MIDAS in ␣ L I, unlike the other I domainbound ligands, but instead to a crevice formed by the carboxylterminal helix 7, stabilizing the helix and locking the I domain into the "closed" conformation. No significant change in conformation comparable with the "open" form was observed in the ␣ 1 I-Fab complex structure although glutamate from the Fab fragment is coordinated to the metal ion at MIDAS.
The structures of the ␣ 2 I-collagen (I)-like peptide complex (14) and the ␣ L I-ICAM1 complex have a conformation different from their "closed" conformation seen in the structures without bound ligand and similar to that observed in ␣ M I with bound Mn 2ϩ (61), referred to as the "open" conformation. The "open" and "closed" conformations are likely to be energetically very similar, allowing the I domain to easily shift between the two states. Thus, in vitro, isolated I domains may exist as an equilibrium between both forms, whereas, in vivo, the conformational state of the heterodimeric integrin is likely to be tightly controlled, since the ligand binding state is linked directly to bidirectional signaling and to other integrin functions.
In the ligand-free structures of both ␣ 1 I and ␣ 2 I, the "closed" conformation is stabilized by a conserved salt bridge formed between Arg 287 (␣ 1 I)/Arg 288 (␣ 2 I) from helix C and Glu 317 (␣ 1 I)/ The peptide is shown as balland-stick representations and covered by transparent surfaces (pink). The surface around Thr 2 is highlighted by yellow (arrow); this region is hydrophobic, and the better binding of the CARKKHDC peptide probably reflects the better compatibility of alanine at that site. A similar mode of docking was obtained for the CARKKHDC peptide (not shown).
Glu 318 (␣ 2 I). Furthermore, the site of collagen and RKKH peptide binding in both ␣ 1 I and ␣ 2 I is partially blocked by Glu 255 (␣ 1 I)/Glu 256 (␣ 2 I), in the loop between strand D and helix 5 near MIDAS, and Tyr 285 (␣ 1 I)/Tyr 285 (␣ 2 I) from helix C. Helix C, corresponding to 283 GSYNR 287 in ␣ 1 I and 284 GYLNR 288 in ␣ 2 I, present in the "closed" conformation, unravels in forming the structure of the "open" collagen-binding conformation of ␣ 2 I (14). In the model structure of the "open" conformation of ␣ 1 I, Glu 255 is bent toward MIDAS in such a way that it would not obstruct the peptide from binding in the docked complexes. Tyr 285 shifts significantly when helix C is unwound in the "open" conformation, and the binding site widens in comparison with the "closed" form ( Fig. 6). This is consistent with experimental results obtained for ␣ 2 I, where the CTRKKHDC peptide bound perfectly well in the absence of the helix C sequence (⌬␣C␣ 2 I), whereas helix C is important for collagen binding (18,57). In ⌬␣C␣ 2 I, the salt bridge that stabilizes the "closed" form cannot be formed, but whether this deletion is sufficient to shift the conformation toward the "open" form is not known at this time. The salt bridge has been disrupted in a E318W mutant of ␣ 2 I, which resulted in a higher affinity for collagen (62).
Five residues near MIDAS in ␣ 2 I, Asp 219 , Glu 256 , Asp 259 , Asp 292 , and Glu 299 , were originally proposed to have a role in RKKH peptide binding, which was confirmed by mutational studies (18,57). Mutation of Glu 256 (Glu 255 in ␣ 1 I), Asp 259 (Asp 258 ), Asp 292 (Ser 291 ), and Glu 299 (Glu 298 ) in ␣ 2 I lowered the binding affinity of ␣ 2 I for the peptide. However, when Asp 219 in ␣ 2 I was mutated to arginine, which is the equivalent residue found in ␣ 1 I (Arg 218 ), the CTRKKHDC peptide bound with higher affinity than to the wild type I domain (18). In the docked complex with ␣ 1 I, the guanido group of Arg 218 could interact with the side chain of Asp 7 in the peptide (in structures of rattlesnake snake venom metalloproteinases (e.g. 1IAG and 1BUD), the corresponding Asp 93 binds a positively charged Ca 2ϩ ion), giving a plausible explanation for the significantly better binding of peptide to ␣ 2 I with the D219R mutation (18). Interestingly, the D219R mutation also changed the ␣ 2 I binding preferences for collagen subtypes to resemble that of ␣ 1 I and ␣ 10 I; the reverse mutant, R218D, in ␣ 1 I and ␣ 10 I led to an ␣ 2 I-like binding pattern (11). In the modeled structure of the "open" form of ␣ 1 I, Arg 218 , Glu 255 , Ser 291 , and Glu 298 would be directly involved in peptide binding (Fig. 7). Asp 258 , the first residue in helix 5, forms a strong interaction with the mainchain nitrogen atom of the amino-terminal residue of helix 4, thus stabilizing the conformation of the peptide-binding site. In our modeled complex, Glu 255 is involved in a network of interactions; Glu 255 (OE) would form a salt bridge with Lys 5 in the peptide, and it might be indirectly bound to the metal ion at MIDAS via an intervening water molecule. Glutamate is conserved at this position in the human ␣I domains and is important for peptide binding to ␣ 2 I (18) but is not required for collagen binding (57).
Studies with differential scanning calorimetry show that human ␣ 1 I can exist in a stable conformation without bound metal, as has been seen for human ␣ M I, rat ␣ 1 I, and human ␣ L I (59,63,64). The stability of the protein increased significantly upon binding Mg 2ϩ , and this shift in the melting temperature (T m ) is directly proportional to the affinity, suggesting that Mg 2ϩ is only loosely bound by ␣ 1 I at MIDAS. The current results support the conclusion that Mg 2ϩ is required to form the ␣ 1 I-peptide complex, as is the case for ␣ 2 I (14, 17), rather than releasing metal ions as was reported for terbium ions (Tb 3ϩ ) on collagen binding to ␣ 2 I (65). The addition of CTRKKHDC peptide to the ␣ 1 I-Mg 2ϩ complex led to a small increase in T m consistent with the expected micromolar affinity of ␣ 1 I for the peptide (for ␣ 2 I, the IC 50 is 1.3 Ϯ 0.2 M for inhibition of collagen binding by the peptide (18)).
Structural changes do occur upon addition of the CTRKKHDC peptide to the ␣ 1 I-Mg 2ϩ complex but not in the vicinity of Trp 158 , since tryptophan fluorescence did not reveal any major conformational change on ligand binding localized near Trp 158 , in agreement with the structural data. The values of ⌬C p (Table II), which reflect the change in exposure of amino acids to the solvent when the protein unfolds (66), do indicate that the native conformations of the ␣ 1 I-Mg 2ϩ complex and the ␣ 1 I-Mg 2ϩ -peptide complex differ from each other. Furthermore, when CTRKKHDC peptide is bound to the ␣ 1 I-Mg 2ϩ complex, an ϳ10% increase in exposed hydrophobic surface was detected (fluorescence due to bound bis-ANS), whereas a small ϳ3% change occurs in the ␣ 1 I structure upon binding Mg 2ϩ . Upon binding CTRKKHDC, a simultaneous increase in light scattering was also observed, suggesting that the newly exposed hydrophobic surface leads to aggregation of ␣ 1 I. In the model structure of the "open" conformation of ␣ 1 I we see a comparable increase in hydrophobic surface in comparison with the structure of the "closed" conformation. These hydrophobic patches locate nearby but not overlapping with the collagen/peptide binding site at MIDAS (Fig. 8), suggesting the possibility of a role in integrin function (e.g. linking two integrin heterodimers in cell-cell adhesion).
The fact that ␣ 1 I domains can bind to each other raises the question of whether two ␣ 1 ␤ 1 integrins can act as homophilic counter receptors. Indeed, many previous studies have indicated that matrix receptor integrins, such as ␣ 2 ␤ 1 , can participate in the formation of cell-cell adhesion sites (67), suggesting their homophilic interactions. Recently, ␣ 1 ␤ 1 was also shown to participate in cell-cell adhesion between smooth muscle cells (68). Experiments with cells transfected with ␣ 2 integrin have indicated that ␣ 2 ␤ 1 -␣ 2 ␤ 1 interaction alone cannot mediate high affinity cell-cell adhesion (69). However, a similar analysis has not been done with cells expressing ␣ 1 ␤ 1 . Model structures of ␣ 1 ␤ 1 and ␣ 2 ␤ 1 (data not shown), based on the structures of ␣ V ␤ 3 , ␣ 2 I (1DZI), and ␣ 1 I (model), did not rule out the possibility of homophilic interactions between two ␣ I domain-containing integrins, since the location of the hydrophobic patches on the heterodimer would permit such interactions.