Dynamic Structural Changes Are Observed upon Collagen and Metal Ion Binding to the Integrin α1 I Domain*

Background: Integrin α1 I domain undergoes conformational changes upon collagen binding. Results: Deuterium exchange was used to measure the effects of cations, collagen, or an antibody on the α1I solution structure. Conclusion: Full-length collagen and metal ions induce changes that differ in key aspects from previously proposed models for α1I activation. Significance: These studies support a new model for integrin I domain activation. We have applied hydrogen-deuterium exchange mass spectrometry, in conjunction with differential scanning calorimetry and protein stability analysis, to examine solution dynamics of the integrin α1 I domain induced by the binding of divalent cations, full-length type IV collagen, or a function-blocking monoclonal antibody. These studies revealed features of integrin activation and α1I-ligand complexes that were not detected by static crystallographic data. Mg2+ and Mn2+ stabilized α1I but differed in their effects on exchange rates in the αC helix. Ca2+ impacted α1I conformational dynamics without altering its gross thermal stability. Interaction with collagen affected the exchange rates in just one of three metal ion-dependent adhesion site (MIDAS) loops, suggesting that MIDAS loop 2 plays a primary role in mediating ligand binding. Collagen also induced changes consistent with increased unfolding in both the αC and allosteric C-terminal helices of α1I. The antibody AQC2, which binds to α1I in a ligand-mimetic manner, also reduced exchange in MIDAS loop 2 and increased exchange in αC, but it did not impact the C-terminal region. This is the first study to directly demonstrate the conformational changes induced upon binding of an integrin I domain to a full-length collagen ligand, and it demonstrates the utility of the deuterium exchange mass spectrometry method to study the solution dynamics of integrin/ligand and integrin/metal ion interactions. Based on the ligand and metal ion binding data, we propose a model for collagen-binding integrin activation that explains the differing abilities of Mg2+, Mn2+, and Ca2+ to activate I domain-containing integrins.

Because of their critical role in mediating cell/cell and cell/ matrix interactions, integrins are among the most studied families of transmembrane receptors. In particular, great progress has been made in understanding the general structural features of these heterodimeric cell-surface receptors and the conformational changes that regulate affinity modulation, ligand binding, and signaling (1,2). Crystal structures of both intact integrins and functional domains have been critical to this process. Of particular interest is the ␣ subunit I domain that is primarily responsible for mediating ligand binding by integrins containing this domain. Among others, the co-crystal structures of disulfide-"locked" ␣L I domains with ICAM-1 (3), ICAM-3 (4), and ICAM-5 (5) and the ␣2 I domain (␣2I) with a collagen triple-helical peptide (THP) 4 (6) have provided snapshots of the closed (nonligand bound), open (ligand-bound), and intermediate conformations of the ligand-binding regions of these integrins. In many cases, direct structural comparisons between free and ligand-bound integrins (or integrin I domains) cannot be made on the basis of existing data. Even less data are available on the nature of the conformational changes that occur upon integrin-ligand binding in solution. Thus, it is important to develop solution-based methods that allow the direct assessment of the conformational dynamics of integrin/ ligand interactions.
In humans, the 18 ␣ and 8 ␤ integrin subunits pair to form at least 24 known heterodimers (7). Nine of the ␣ subunits contain I domains, which play a central role in ligand binding and selectivity. The four collagen-binding receptors (␣1␤1, ␣2␤1, ␣10␤1, and ␣11␤1) fall into this subset of I domain-containing integrins. Of these, ␣1␤1 (VLA1) and ␣2␤1 (VLA2) have been the most thoroughly studied. The crystal structure of ␣2I has been determined in the presence of a heterotrimeric THP corresponding to its primary ligand, type I collagen (6). These studies, along with structural studies of the ␣ I domains of the ␤2 integrins LFA-1 (4, 5, 8 -11), Mac-1 (12,13), and ␣X␤2 (14), have helped to establish and refine a model for the ligand-induced conformational changes that lead to "outside-in" signaling in I domain-containing integrins (1). In this model, ligand binding to the divalent cation within the I domain metal iondependent adhesion site (MIDAS) induces a shift from the socalled "closed" (unliganded) to the "open" (liganded) conformation and induces allosteric changes in the C-terminal ␣7 helix of the I domain, triggering a large scale rearrangement of the global topology of the receptor. This model was refined to include the concept of an "intrinsic ligand," in which a glutamate from the C terminus of the I domain binds to the MIDAS site within a similar domain within the ␤ subunit (the ␤ I domain or "I-like" domain) (15,16).
VLA1 is a major receptor for basement membrane-type IV collagen and is expressed on hematopoietic, neuronal, and mesenchymal cell types (17). Inhibition of VLA1 function, through the use of genetic knock-outs or pharmacological agents, blocks cell retention and activation in pathogenic sites and leads to reduced inflammation and amelioration of disease in a variety of animal models (for examples, see Refs. 18 -25). Despite the wealth of biological and pharmacological data on VLA1, limited structural information is available for this integrin. Crystal structures of isolated ␣1 I domains (␣1I) have been reported (26 -28), as well as the structure of ␣1I in complex with the function-blocking antibody AQC2 (29,30). Although computational models have been proposed for the ␣1I open conformation (27) and for the ␣1I/collagen interaction site (31), no structures of ␣1I in complex with any type of ligand or ligandmimetic peptide have been reported (2). Recently, the structure of ␣1I containing an activating mutation (E317A) was determined (32). This protein adopted a novel conformation that differed from both the open and closed structures previously published for other integrins, indicating that the activation mechanism of collagen-binding integrins may differ from that of other integrins.
Peptide amide hydrogen-deuterium exchange mass spectrometry (DXMS) has been used to analyze protein/protein, protein/substrate, protein/inhibitor, protein/solvent, and protein/DNA interactions, as well as protein dynamics and protein conformational changes (33)(34)(35)(36)(37)(38)(39)(40). Here, we utilized DXMS to directly study the solution binding and conformational dynamics of the interaction between ␣1I and full-length type IV collagen. DXMS allowed us to investigate the conformational changes that occur when ␣1I binds to collagen under conditions closely resembling the native physiological environment and to test the current structural models for integrin I domain/ ligand interactions that are largely based on static crystallographic data. We also examined conformational changes induced upon binding of the function-blocking ␣1 monoclonal antibody AQC2 and compared them with those mediated by the native ligand. Our results reveal important differences from existing models for integrin I domain/ligand interactions and ligand-induced allosteric changes, and they suggest novel dynamic features of the ␣1I/collagen and ␣1I/metal ion interactions. The methods we developed to characterize the ␣1I domain should be easily adapted to the study of other integrins.

EXPERIMENTAL PROCEDURES
Protein Preparation-These studies utilized a chimeric ␣1I comprising elements of both the rat and human sequences (Fig.  1). The sequence numbering is based on the full-length mature human ␣1 integrin sequence (NCBI Reference Sequence NP_852478.1). Four human residues (Val-213, Gln-214, Arg-215, and Arg-218) were substituted into the MIDAS region of the corresponding rat ␣1I sequence. This form of ␣1I was used to establish the binding sites for anti-human ␣1 integrin antibodies using biochemical methods (41) and was crystallized in complex with a Fab fragment of the function-blocking antihuman ␣1 integrin antibody AQC2 (29,30). The bacterially expressed chimeric ␣1I is significantly more soluble than its fully human GST-␣1I fusion counterpart, making it more suitable for structural studies. Recombinant ␣1I was expressed in Escherichia coli as a glutathione S-transferase (GST) fusion protein and was purified as described previously (41). Samples were analyzed using size exclusion chromatography on a Superdex 200 10/300 GL column using phosphate-buffered saline (PBS) as the mobile phase and compared with a reference standard mixture (Bio-Rad). For studies requiring removal of the GST tag, the protein was cleaved using thrombin and purified as described previously (41). The anti-␣1␤1 monoclonal antibody AQC2 was prepared as described previously (41).
Ligand Binding Studies-The binding of GST-␣1I to collagen was measured using an electrochemiluminescence assay (42). Type IV collagen from human placenta (Sigma) was coated onto Tosyl M-280 Dynabeads (Dynal, A.S., Oslo, Norway) at a ratio of 20 g of collagen/mg of beads, as described previously (42). Beads were blocked by incubation with 8% rat plasma in assay buffer (50 mM Hepes, pH 7.5, 150 mM NaCl, 0.1% Triton X-100) for 15 min at 25°C prior to use. A goat polyclonal antibody against GST (GE Healthcare) was labeled with ruthenium(II) tris(bipyridine)-NHS ester (TAG-NHS Ester; IGEN International, Inc., Gaithersburg, MD) as per the manufacturer's instructions. Serial dilutions of GST-␣1I in assay buffer were incubated with 30 g/ml labeled anti-GST antibody, 0.2 mg/ml collagen-coated beads, and 0.5 mM MnCl 2 , in a final volume of 120 l. After gentle agitation at 25°C for 60 min, 200 l of assay buffer was added, and samples were read in an ORIGEN 5.1 detector according to the manufacturer's instructions (IGEN). To assess the blocking ability of AQC2, serial dilutions of AQC2 were incubated with 5 ng/ml GST-␣1I, 2 g/ml labeled anti-GST antibody, 0.2 mg/ml collagen-coated beads, and 1 mM MnCl 2 in a final volume of 100 l and processed as described above.
Differential Scanning Calorimetry-DSC was performed using an automated capillary DSC (capDSC, GE Healthcare). Protein and reference solutions were sampled automatically from 96-well plates using the robotic attachment. Prior to each sample measurement, two buffer scans were performed to define the base line for subtraction. All 96-well plates containing protein were stored within the instrument at 5°C. Each I domain sample was diluted to 1 mg/ml in relevant buffer. Scans were performed from 10 to 100°C at 2°C/min using the high feedback mode for enhanced peak resolution. Scans were analyzed using the Origin software package supplied by the manu-facturer. Subsequent to the subtraction of reference base-line scans, nonzero sample base lines were corrected using a third order polynomial. The unfolding parameters for the multidomain unfolding profiles of I domain were deconvoluted using the multipeak fitting routine within the software assuming non-two-state unfolding behavior. The measured melting temperature (T m ) for ␣1I in the presence of the chelating agent EDTA ranged from 54.5 to 57.9°C, depending on the protein concentration, buffer composition, and scan rate, but it was reproducible Ϯ1°C within a given set of experimental parameters.
Denaturation Studies-The denaturation of ␣1I as a function of urea was measured by fluorescence spectroscopy (41). Briefly, samples containing 300 g/ml ␣1I in 50 mM Tris-HCl, pH 7.5, 0.1 mM DTT with 10 mM EDTA, 10 mM CaCl 2 , or 1 mM MnCl 2 and with varying amounts of urea were analyzed at 25°C using an excitation wavelength of 280 nm. Emission spectra from 300 to 400 nm were collected using a Spectramax M5 fluorescence spectrometer (Molecular Devices Inc., Sunnyvale, CA). Fluorescence data at 350 nm were plotted as a function of urea and standardized using the change in fluorescence from 0 to 9 M urea for each of the test conditions as a measure of the total fraction folded.
Optimization of Fragmentation Conditions-Prior to conducting the hydrogen-deuterium exchange experiments, test digests prepared with undeuterated buffer in varying concentrations of guanidine hydrochloride (GdnHCl) were made to optimize proteolysis conditions for the best peptide coverage (43). For each sample, 6 l of stock solution of ␣1 I domain (8 mg/ml) was first diluted with 18 l of H 2 O buffer (150 mM NaCl, 8.3 mM Tris-HCl, pH 7.2). The sample was then quenched with 36 l of a quench solution (0°C) of 0.8% formic acid, 17% glycerol, and GdnHCl at final concentrations of 0.05, 0.5, 1.0, 2.0, or 4.0 M. This quenching step lowered the pH to 2.2-2.5 to minimize further hydrogen-deuterium exchange in subsequent deuterated samples. The quench-denaturation process was allowed to proceed on ice for 30 s, after which the sample was stored frozen at Ϫ80°C. Procedures for pepsin digestion for DXMS have been described previously (44 -48). Briefly, a sample was thawed at 5°C and then immediately digested over a protease column (66-l bed volume) filled with porcine pepsin (Sigma; immobilized on Poros 20 AL medium at 30 mg/ml following the manufacturer's instructions) at a flow rate of 100 l/min with 0.05% trifluoroacetic acid. Peptic fragments were collected on a C18 trap column and separated on a C18 reversed phase column (Vydac, catalog no. 218MS5105) with a linear acetonitrile gradient from 6 to 38%. The column effluent was electrosprayed directly into an LCQ Classic (Thermo Finnigan, Inc.) or Q-TOF mass spectrometer (Micromass). Determination of pepsin-generated peptides from MS/MS data sets was facilitated through the use of SEQUEST (Thermo Finnigan, Inc.). This set of peptides was then further verified by DXMS Explorer (Sierra Analytics Inc., Modesto, CA). The peptide coverage maps for the different concentrations of GdnHCl were compared, and the condition with the best coverage map for each individual protein or protein complex was used for subsequent deuterium exchange experiments.
Hydrogen-Deuterium Exchange Experiments-Samples were prepared with three states of hydrogen-deuterium exchange as follows: nondeuterated (ND), partially deuterated, and equilibrium deuterated. All samples were processed as described above. Equilibrium deuterated samples were prepared by incubating protein at room temperature in D 2 O buffer (1% (v/v) formic acid) overnight. Procedures for individual experiments are described below. The hydrogen-deuterium exchange experiments performed with our automated apparatus produces measurements of deuteron incorporation where the standard deviation is less than 2% of the mean of triplicate determinations (49,50).
␣1 I Domain-␣1 I domain solutions (8 mg/ml) were mixed with solutions of CaCl 2 , MnCl 2 , or EDTA to final concentrations of 1, 0.5, or 5 mM, respectively, or with MgCl 2 at either 1 or 10 mM. The solutions were preincubated at room temperature for 10 min and then chilled to 0°C. The ND sample was processed exactly as described under "Optimization of Fragmentation Conditions." Hydrogen-deuterium exchange experiments were initiated by mixing 6 l of preincubated samples with 18 l of D 2 O buffer (8.3 mM Tris, 150 mM NaCl, pH 7.2, in D 2 O). The samples with 10 mM Mg 2ϩ were incubated at 0°C for 10, 30, or 100 s or at 25°C for 30, 100, or 300 s. All other samples were incubated at 0°C for 10, 30, or 300 s or at 25°C for 30, 100, 300, 1000, 3000, 10,000, 30,000, 100,000, or 300,000 s. At the indicated times, the H/D exchange was quenched by adding 36 l of ice-cold quench solution (0.8% formic acid, 16.6% glycerol, and 0.05 M GdnHCl). The samples were then transferred to ice-cooled autosampler vials, frozen on dry ice, and stored at Ϫ80°C. As shown in previous studies (51)(52)(53)(54), incubation at 0°C for 10, 30, or 100 s is equivalent to incubation at 25°C for 1, 3, or 10 s, respectively. For all ribbon diagrams, the data generated at 0°C were converted to the data equivalent at 25°C.
AQC2-␣1 I Domain Complex-The complex between antibody AQC2 and ␣1I was prepared by mixing solutions of ␣1I (7 mg/ml) and antibody (6.4 mg/ml) at a 2:1 molar ratio at room temperature for 30 min and then chilling to 0°C. The ND sample was processed by diluting 25 l of the complex with 75 l of H 2 O Buffer (8.3 mM Tris, 150 mM NaCl, pH 7.2) and then mixed with 100 l of quench solution (0.8% formic acid, 16.6% glycerol, 0.5 M GdnHCl). For deuterated samples, 25 l of preincubated sample was mixed with 75 l of D 2 O buffer and incubated at various time intervals (10, 30, and 100 s at 0°C and 30, 100, 300, 1000, 3000, 10,000, 30,000, 100,000, and 300,000 s at 25°C) before being added to 100 l of quench solution. All samples were frozen on dry ice. Incubation at 0°C for 10, 30, or 100 s is equivalent to incubation at 25°C for 1, 3, or 10 s, respectively.
GST-␣1 I Domain-For the ND sample, 5 l of GST-␣1I (12.7 mg/ml with 10 mM MgCl 2 ) was diluted with 64 l of H 2 O buffer and quenched with 96 l of quench solution (0.8% formic acid, 16.6% glycerol, 0.5 M GdnHCl). Deuterated samples were prepared by diluting 5 l of GST-␣1I with 11 l of H 2 O buffer, followed by mixing with 48 l of D 2 O buffer. Samples were incubated for 10, 100, and 1000 s at 0°C and 300, 1000, 3000, 10,000, 30,000, and 100,000 s at 25°C. At the indicated times, the samples were added to vials containing the 96 l of quench solution. The samples were transferred to autosampler vials and frozen on dry ice. Incubation at 0°C for 10, 100, or 300 s is equivalent to incubation at 25°C for 1, 10, or 100 s, respectively.
GST-␣1 I Domain-Collagen Complex-An on-column exchange process was used to eliminate the interference of collagen peptides during data acquisition. The collagen column was created by the following procedure: 5 mg of collagen (type IV collagen from human placenta, Sigma) was dissolved in 2 ml of 50 mM sodium citrate, pH 4.4. 140 mg of POROS AL 20 media (Applied Biosystems), 330 l of 1 M sodium cyanoborohydride, and 3.45 ml of 1.5 M sodium sulfate were added, and the reaction was mixed continuously using a tube shaker at 25°C for 16 h. The beads were washed with 50 mM sodium citrate, pH 4.4, and incubated with 6 ml of 0.1 M ethanolamine, 50 mM sodium citrate, and 330 l of 1 M sodium cyanoborohydride, pH 4.4, at 25°C for 2 h. The beads were further washed using H 2 O Buffer and packed into a 150-l bed volume column at 10 ml/min with H 2 O Buffer. For ND samples, 20 l of 12.7 mg/ml GST-␣1I was loaded onto the collagen column following preequilibration with H 2 O Buffer and incubated at 0°C for 5 min.
To initiate the H/D exchange reaction, 500 l of D 2 O buffer was loaded onto the column and incubated for various times (10, 100, and 1000 s at 0°C and 300, 1000, 3000, 10,000, 30,000, and 100,000 s at 25°C). 500 l of 1% formic acid was applied to the column to quench the reaction and elute the GST␣1I. 150 l of eluent was added to 12.7 l of quench solution. Samples were stored at Ϫ80°C prior to analysis. Incubation at 0°C for 10, 100, or 1000 s is equivalent to incubation at 25°C for 1, 10, or 100 s, respectively. The collagen column was regenerated between runs by washing it with 1% formic acid until the acid-eluted material measured by absorbance at 214 nm (A 214 ) reached base line, followed by equilibration of the column with neutral pH buffer (8.3 mM Tris, 150 mM NaCl, pH 7.2, with appropriate divalent cations). The efficiency of regeneration was confirmed by observation of complete restoration of ligand binding capacity assessed by both saturable binding capacity and the magnitude of material eluted after binding (measured by A 214 ). In control studies, when the acid-washed, re-equilibrated column was again eluted without protein re-loading, less than 2% of the maximum protein binding capacity of the column was eluted, demonstrating that the column had been cleared of previously bound GST-␣1I by the regeneration procedure.
Calculations-DXMS Explorer was used for the data processing and reduction of hydrogen-deuterium exchange experiments. Back-exchange occurs immediately after addition of the quench solution, and corrections for it were determined by previously described methods (48) and as shown in Equation 1, where m(P), m(N), and m(F) are the centroid value of the partially deuterated, ND, and equilibrium deuterated peptide, respectively.
Generation of Ribbon Diagrams-The deuteration levels of peptides were further sublocalized using overlapping peptides (44). First, fragments (f i ) within each peptide were delineated in a set of overlapping peptides according to overlapping regions. At least five residues were included within each fragment to avoid overinterpreting the experimental data. Second, a variable s i,t was defined to represent the number of deuterium of fragment f i values at the on-exchange time point t. Thus, a set of linear equations was established such that the sum of s i,t in a specific peptide was equal to the total number of deuterons observed in DXMS experiments. The best solutions were determined by linear least square fitting, with the following stipulations: (a) that s i,t is Ն0; and (b) for a given fragment f i , mass shift at the longer on-exchange time point cannot be smaller than the mass shift at the shorter on-exchange time point. These fitted values were colored according to the degree of deuterium exchange (ribbon diagrams).
Generation of ␣1I-Collagen Model-A predicted structure of the ␣1I-collagen complex was created by homology modeling with the program MODELLER (55,56). The crystal structure of the ␣2 I-domain in complex with collagen (Protein Data Bank code 1DZI) (6) and a structure of the ␣1 I domain (Protein Data Bank code 1MHP) (30) were used as templates. Type IV collagen was modeled as a trimeric THP (31). Restraints for the MIDAS site and the ␣C helix were taken exclusively from the ␣2-collagen template structure so that the modeled I domain would remain in the open conformation. The model was refined by simulated annealing under MODELLER spatial restraints from the template structures.

RESULTS
Purity and Bioactivity of ␣1 Integrin I Domain Proteins-The ␣1 I domain was expressed as a GST fusion protein comprising elements of both the rat and human ␣1I sequences ( Fig. 1) (41). To fully characterize the properties of this chimeric form of ␣1I, both the GST fusion (GST-␣1I) and ␣1I generated by cleavage of the GST tag were purified ( Fig. 2A). Under reducing conditions, GST-␣1I migrated at ϳ45 kDa ( Fig. 2A, lane 1) and ␣1I migrated at ϳ24 kDa (lane 2), consistent with their predicted molecular masses (51.1 and 24.9 kDa, respectively). Using sizeexclusion chromatography, ␣1I migrated as a monomer (apparent mass ϳ20 kDa), whereas GST-␣1I eluted with an apparent molecular mass corresponding to a trimer (ϳ135 kDa), consistent with the native trimeric structure of GST (57). The purified GST-␣1I protein bound to type IV collagen with an EC 50 of 31 nM (Fig. 2B), and this interaction could be completely blocked by incubation with AQC2 (IC 50 ϭ 64 pM, Fig.  2C). The apparent affinity of chimeric GST-␣1I for collagen FIGURE 1. Amino acid sequence of the chimeric ␣1I protein used in this study. Boxed regions represent N-and C-terminal non-integrin sequences derived from the thrombin cleavage site and the multiple cloning site of the bacterial expression vector, respectively. Numbering is based on the fulllength ␣1 integrin sequence and does not include the vector-derived residues. Underlined residues (Val-213, Gln-214, Arg-215, and Arg-218) are amino acids from the human ␣1I sequence that were inserted in place of the corresponding rat ␣1I residues (Gly-213, Arg-214, Gln-215, and Leu-218) to generate the chimeric molecule.
was identical to what we have observed using the corresponding rat GST-␣1I (data not shown), and is consistent with previously reported affinities for the human GST-␣1I/collagen IV interaction (58,59). The chimeric GST-␣1I was utilized for these studies because the corresponding human protein was poorly soluble under the physiological conditions used to assess its biochemical and functional attributes.
Effect of Divalent Cations on the Stability of ␣1I-All integrins require divalent cations for ligand binding. A previous study, using circular dichroism, demonstrated the stabilizing effect of divalent cations (Mg 2ϩ and Mn 2ϩ ) on ␣1I (41). We have extended these observations using differential scanning calorimetry (DSC) (Fig. 3). The measured melting temperature (T m ) for ␣1I in the presence of the chelating agent EDTA was 57.9°C. A single cooperative unfolding transition was observed (Fig. 3A). No significant change occurred in the presence of 10 mM CaCl 2 alone (T m ϭ 58.5°C) or in 1 mM CaCl 2 , 1 mM MgCl 2 , which approximates physiological cation concentrations (T m ϭ 58.2°C, Fig. 3A). Addition of 10 mM Mg 2ϩ (T m ϭ 61.4°C) or 10 mM Mn 2ϩ (T m ϭ 62.0°C) alone increased the thermal stability of ␣1I significantly. The effects of Mn 2ϩ and Mg 2ϩ were concentration-dependent, as determined in a second set of experiments showing stabilizations of as much as 8 -10°C (Fig. 3B). The EC 50 values for Mn 2ϩ and Mg 2ϩ were 0.16 and 1.8 mM, respectively. These values agree with Mn 2ϩ and Mg 2ϩ concentrations previously shown to support collagen binding to ␣1␤1 (42,60). Similar results were obtained using the GST fusion protein (GST-␣1I), although under some conditions the I domain transition was masked by the GST unfolding transition at ϳ57-59°C (data not shown). To complement the thermal stability measurements, the impact of Mn 2ϩ on susceptibility of ␣1I to the denaturant urea was measured by monitoring intrinsic tryptophan fluorescence (Fig. 3D). ␣1I has a single tryptophan residue (Trp-158), located near the MIDAS site. In the absence of divalent cation, ϳ5 M urea was required for 50% denaturation, and Mn 2ϩ significantly increased the stability of ␣1I domain, to an EC 50 of ϳ7 M urea.
Ca 2ϩ does not support ␣1␤1 collagen binding, and it has been shown to bind with poor (mM) affinity to integrin I domains (61)(62)(63). Unlike Mn 2ϩ and Mg 2ϩ , Ca 2ϩ did not stabilize ␣1I toward thermal denaturation at concentrations up to 100 mM (Fig. 3B). However, at equimolar concentrations (10 mM), Ca 2ϩ completely reversed the stabilizing effect of Mg 2ϩ , indicating that Ca 2ϩ is capable of binding to ␣1I and displacing Mg 2ϩ from the MIDAS site. Interestingly, in urea denaturation studies, addition of 10 mM Ca 2ϩ increased the urea concentration required for half-maximal denaturation to ϳ7 M, similar to the amount of stabilization observed with Mn 2ϩ (Fig. 3D). Similar results were observed with urea denaturation studies using the rat ␣1 I domain (Ref. 41 and data not shown). These results clearly indicate that Ca 2ϩ is capable of binding to the ␣1I MIDAS, even though it does not effectively support collagen ligand binding. The difference between the thermal and chemical denaturation results suggests that Ca 2ϩ binding affects the local environment around the MIDAS such that it can be detected by chemically induced unfolding but not by calorimetry, which is a more global measure of stability.
DXMS of ␣1I-DXMS was utilized to evaluate the structural properties of unliganded ␣1I in solution and to provide a reference for subsequent experiments incorporating bound ligands. The DXMS method measures the solvent accessibility of mainchain amides in defined segments of a protein through a combination of time-dependent deuterium exchange, limited proteolysis, and mass spectrometry (33)(34)(35)(36)(37)(38)(39)(40). Percent deuteration is operationally defined as the number of deuterium ions incorporated into a given peptide at a fixed time, divided by the maximum level incorporated at equilibrium. The DXMS profile of ␣1I in the presence of 10 mM MgCl 2 is shown in Fig. 4. Briefly, a sample of ␣1I was diluted 1:3 into a deuterated buffer and incubated at 25°C for defined time intervals from 1 to 1000 s. The exchange reaction was quenched by lowering the temperature and pH, and samples were digested with pepsin and analyzed using LC-MS. The sequence coverage of ␣1I was 93%. The data generated from 77 overlapping peptides was used to gen- erate a one-dimensional ribbon map representing the average deuterium incorporation throughout the protein sequence (Fig. 4A). This linear map was annotated with the secondary structural elements identified from the ␣1I crystal structure (Fig. 4B). This experiment revealed several regions of ␣1I with very high incorporation of deuterium even at short incubation times, indicative of a highly flexible and dynamic structure. For example, N-terminal residues 134 -143 appeared largely unfolded, exchanging Ͼ80% of the accessible amide protons within 1 s in D 2 O. This is consistent with the lack of any observed density N-terminal to residues 141-143 in ␣1 I domain crystal structures 1QC5 (28), 1PT6 (27), and 1CK4 (26). Similarly, residues at the C terminus (residues 333-336) were also exchanged very rapidly, consistent with the crystallographically determined terminus of the C-terminal ␣7 helix at position 332 (26). A third region comprising residues 291-295, a loop that connects the ␣C and ␣6 helices, also exchanged rapidly. As discussed below, the ␣C helix and adjacent residues play a key role in collagen binding and selectivity by this integrin. The ␣C helix itself (residues 283-287) exchanged more slowly than these other regions, but it still showed Ͼ75% deuterium incorporation within 300 s. The remainder of the ␣1I sequence, including the ␣-helical and ␤-strand regions forming the classic dinucleotide-binding fold, exchanged much more slowly, with most showing Ͻ50% deuterium incorporation even with longer D 2 O incubation times. The exchange data at 100 s were overlaid on the three-dimensional structure of ␣1I (Fig. 4C). From this overlay it is clear that the slowest exchanging regions are within well defined secondary structural elements (e.g. the ␣3 and ␣5 helices, and the ␤E strand), although the faster exchanging regions are contained within or at the interfaces with less structured loops. The DXMS profile of the GST␣1I fusion protein was also determined (Fig. 4A) and was very similar to the untagged ␣1I.
Effect of Divalent Cations on ␣1I Domain Solution Structure-We next evaluated the impact of MIDAS site occupancy with different divalent cations on the DXMS exchange properties of ␣1I. The incorporation of deuterium into ␣1I was assessed in the presence of 5 mM EDTA (apo-␣1I) or in 10 mM Mg 2ϩ , 0.5 mM Mn 2ϩ , or 1 mM Ca 2ϩ (Fig. 5). Relative to apo-␣1I, all three cations significantly decreased the exchange rates in multiple peptides, including residues 157-171, 195-206, and 250 -264, as shown in Fig. 5A. No regions showed increased exchange rates in the presence of cation, suggesting an overall decrease in flexibility throughout the ␣1I structure. Difference maps comparing each of the cation-bound states versus apo-␣1I (Fig. 5B) show that, at 100 s, slower exchange rates were observed over the majority of the ␣1I sequence. Larger decreases in exchange rates were observed with Mg 2ϩ and Mn 2ϩ , consistent with the stabilizing effects measured using DSC and tryptophan fluorescence. Ca 2ϩ led to smaller but still significant effects on exchange rates, further demonstrating that Ca 2ϩ is competent to bind the MIDAS site and partially stabilize ␣1I.
Generally, the exchange rates of ␣1I in Mg 2ϩ and Mn 2ϩ were very similar (Fig. 5). This was confirmed by generating difference maps comparing these two conditions over a wide range of exchange times (1-300 s) (Fig. 6A). One set of peptides covering the ␣C helix region, however, exchanged much faster in Mn 2ϩ than in Mg 2ϩ . This was exemplified by the 283-290-residue peptide, which incorporated notably higher levels of deuterium at exchange times Ͻ100 s in the presence of MnCl 2 than it did in MgCl 2 (Fig. 6B). This result was observed even when the protein was first treated with EDTA and reconstituted with excess MgCl 2 or MnCl 2 , indicating that the result was not due to the effect of metals introduced during protein production or purification (data not shown). The observed differences in exchange rates in this region were surprising, because Mg 2ϩ and Mn 2ϩ were shown to bind integrin I domains with similar geometries (11, 62, 64) and showed essentially identical abilities, at saturating concentrations, to stabilize integrin structure ( Fig. 3) and promote ligand binding (42,60). The faster exchange rate in the presence of Mn 2ϩ has led us to generate a new model to explain the differences in the relative abilities of various metal ions to activate integrin I domains, in which  Fig. 1), and the positions of I domain structural elements are annotated above (red twist, ␣-helix; blue arrow, ␤-strand). Each bar below the sequence is divided into rows corresponding to each time point from 1 to 300 s (top to bottom). Color coding indicates the percentage of deuterium incorporation into exchangeable amide protons after a given incubation time. Vertical arrows indicate residues Thr-141 and Phe-332, which are the N-and C-terminal residues, respectively, within which the ␣1 I domain crystal structure (Protein Data Bank code 1CK4) showed measurable electron density. P indicates the positions of proline residues, which have no exchangeable protons. Other areas in gray did not have sufficient coverage in the map to allow accurate quantitation of deuteration levels. B, annotated list of the primary structural features of ␣1I and the corresponding positions within the sequence (26). C, exchange rates at 100 s were mapped onto ␣1I crystal structure 1MHP. Color coding indicates the percentage of deuterium incorporation into exchangeable amide protons (identical scale as in A). The positions of prominent structural elements are indicated, and the metal ion in the MIDAS is represented by a gray sphere. SEPTEMBER 21, 2012 • VOLUME 287 • NUMBER 39  Mn 2ϩ preferentially unfolds the ␣C helix by assuming a pentacoordinated geometry (discussed below). In any case, the ability of Mn 2ϩ and Mg 2ϩ to induce detectably different conformational changes in ␣1I indicates that, unlike previous suggestions (65), the choice of divalent cation impacts the local structure around the ligand-binding site.

Dynamics of Ligand Binding to ␣1 Integrin I Domain
DXMS Profile of the Complex between ␣1I and Full-length Collagen-A crystal structure of ␣2I in complex with a triplehelical collagen-based peptide, but not a corresponding ␣1Icollagen peptide structure, has been reported (6). Because intact fibrillar collagen cannot be crystallized, no structure has been reported for full-length collagen bound to an integrin. To explore the nature of these interactions under native conditions, we determined the DXMS exchange pattern of the ␣1I protein bound to full-length type IV collagen. For these studies, a complex was formed by incubating the GST-␣1I fusion protein with collagen IV that was immobilized on a bead-based support in the presence of 10 mM MgCl 2 . Deuterium exchange was then performed on the immobilized GST-␣1I. The DXMS exchange rates in the presence of collagen were subtracted from those determined for GST-␣1I alone to generate a series of difference maps (Fig. 7A). The exchange data at early (1 s) and late (10,000 s) time points were superimposed on a model of ␣1I bound to collagen IV (Fig. 7B), generated by docking ␣1I, in an open conformation, with a collagen THP as described previously (31).
Regions at the interface of protein-protein binding surfaces are typically protected from amide exchange with water (66). Using DXMS, only residues in MIDAS loop 2 of ␣1I, exemplified by peptide 207-221, showed significant protection against deuterium exchange when bound to collagen (Fig. 7A). The structure of the ␣2I-collagen peptide complex suggested interactions between all three MIDAS loops (151-156, 212-220, and 253-260) and the collagen peptide (6). Similar interactions were proposed for the ␣1I-collagen IV structure, including stabilizing interactions with collagen involving Arg-218 in loop 2 and Glu-255 in loop 3 (31). The DXMS results indicate that little change in solvent accessibility occurs in either loop 1 or loop 3 upon binding to collagen. A closer examination of the model (Fig. 7B) suggested an explanation for this result. Loops 1 and 2 both make significant contacts with the collagen. Loop 2 has exposed backbone amides, particularly in residues 215-219, which are apparently protected upon collagen binding. Contacts of loop 1 with collagen are mediated mainly by the side chains of Asn-153 and Tyr-156, but the backbone amides of loop 1 are largely buried by folding within ␣1I itself. As a result, exchange is substantially slower in loop 1 even in the absence of collagen (see Fig. 4). Consequently, it is not surprising that collagen binding has a negligible effect on the loop 1 amides. In contrast, loop 3 has exposed backbone amides, but the model suggests it has very limited contact with collagen, which explains the limited protection observed in loop 3. This result does not support the proposal that Glu-255 forms a salt bridge with collagen (31) but is instead consistent with data showing that the corresponding glutamic acid in ␣2 (Glu-256) is not critical for ligand binding (67).
Several regions of ␣1I displayed significantly increased rates of exchange when complexed with collagen (Fig. 7). These increased rates of exchange reflect greater solvent exposure of the affected amide protons and suggest an increase in flexibility in these regions. The changes were clustered in the C-terminal region of the molecule. Residues 283-290, which include the ␣C helix, displayed very rapid exchange, with essentially complete deuterium incorporation within 1 s of solvent exposure, suggestive of extensive structural changes in this region. Residues 318 -332, in the C-terminal ␣7 helix, and other intervening structural elements between ␣C and ␣7 (residues 298 -317) also showed notable increases in the rates of deuterium incorporation, although exchange was not as fast as in the ␣C helix. These results provide direct evidence that ␣1I undergoes an allosteric conformational change in the C-terminal helix upon ligand binding, as has been postulated for other I domain-containing integrins on the basis of static crystal structures (3,4,6,12), In addition, the striking increase in conformational flexibility of the ␣7 helix in the presence of ligand provides direct support for the proposal that the I domain C-terminal helix is unusually mobile (5), although this mobility appears to be induced by ligand binding rather than an intrinsic property of the unliganded I domain structure (discussed below).
Effect of the "Pseudo-ligand Mimetic" Antibody AQC2 on ␣1I Exchange Rates-The ␣1 I domain was previously crystallized with a Fab fragment of the anti-␣1␤1 function-blocking antibody AQC2 (30). AQC2 can be classified as a pseudo-ligand mimetic antibody, because it binds to ␣1I in a ligand-mimetic fashion via direct interaction of an antibody aspartate residue with the MIDAS metal ion, but it does not induce a change to the open conformation (30,68). The AQC2-␣1I complex was generated in the presence of Mg 2ϩ , and DXMS studies were carried out in a similar manner as described above for the collagen-␣1I complex (Fig. 8). A difference map is shown in Fig.  8A, with blue and red regions indicating decreased and increased exchange, respectively, upon antibody binding. The data at 1 and 1000 s were mapped onto the co-crystal structure of the AQC2-␣1I complex (Fig. 8B) (30).
The most dramatic decreases in exchange rates were in residues 214 -226, corresponding to MIDAS loop 2. This is consistent with the inability of AQC2 to bind to wild-type rat ␣1I (41), which has four amino acid differences in the 213-218residue region, and with direct evidence from the crystal structure showing Arg-218 located inside of a cleft within the antibody (30). As in the collagen-␣1I complex, no significant effects on MIDAS loops 1 or 3 were observed. The correlation between the collagen-␣1I and AQC2-␣1I exchange patterns within the MIDAS region indicates that the local interactions of AQC2 with ␣1I at the MIDAS site are very similar to those of the native ligand. This result is in striking contrast to the crystal structure of AQC2-␣1I, which showed interactions of the antibody with all three loops, and suggests that the loop 2 region of ␣1I is predominantly responsible for the interaction of ␣1I with the antibody.
Similar to the collagen complex, the exchange rate in the ␣C helix (residues 283-290) increased upon AQC2 binding, indicating that the antibody induces a significant loss of secondary structure in this region. However, unlike collagen binding, the structural changes in the ␣7 C-terminal helix (residues 318 -332) were modest. The crystal structure of ␣1I bound to AQC2 (30) adopted a closed conformation very similar to that of the unbound ␣1I (26,28), with no evidence for a shift in the C-terminal helix. Thus, even though AQC2 binding to ␣1I leads to similar effects as the native ligand within the local region surrounding the MIDAS, it does not induce allosteric changes that would indicate a shift from the closed to the open conformation, characteristic of ligand binding.

DISCUSSION
We have successfully used hydrogen-deuterium exchange mass spectrometry (DXMS) to explore the conformational dynamics of cation, ligand, and antibody binding to the I domain of integrin ␣1 in aqueous solution. Most detailed structural information on integrin I domains is based on crystallographic data, so DXMS provided a unique opportunity to study the solution dynamics of the ␣1 I domain. The DXMS mapping method provided coverage of 93% of the ␣1I sequence with over 200 exchangeable hydrogens. The average peptide length was 15 amino acids. The observed H/D exchange rates for unliganded ␣1I were remarkably consistent with the structural elements observed in crystallographic studies of this domain. Regions at the N and C termini of the protein that did not show measurable density in the crystal structures incorporated deuterium rapidly, establishing the limits of the globular domain as comprising residues Asp-144 to Phe-332. Slow exchanging regions throughout the sequence largely correlated with defined secondary structural elements, although the unstructured loop regions, including the MIDAS loops involved in ligand binding, exchanged more rapidly.
In I domain-containing integrins, the I domain MIDAS coordinates a divalent cation such as Mg 2ϩ , which interacts directly with an acidic residue (Asp or Glu) in the ligand. Some nonphysiological cations, such as Mn 2ϩ , are capable of activating integrins much more strongly than Mg 2ϩ , whereas Ca 2ϩ is incapable of mediating ligand binding, but the structural basis for this phenomenon is not well understood. Previous studies established that Mg 2ϩ and Mn 2ϩ at millimolar concentrations stabilize the I domain from ␣1 (41). Using DSC, we observed up to 8 -10°C increases in the T m values for ␣1I at saturating concentrations of either Mn 2ϩ or Mg 2ϩ . This overall stabilization by Mn 2ϩ or Mg 2ϩ was further established using DXMS, which revealed decreased H/D exchange rates throughout the ␣1I sequence in the presence of these cations.
Interpretation of the role of Ca 2ϩ in integrin/ligand interactions is complicated (69). In ␣ I domains, Ca 2ϩ affinity for MIDAS, and its ability to regulate ligand binding, is both concentration-and activation-state dependent (61,70,71). The affinity of the ␣M I domain for Ca 2ϩ was shown to be low (millimolar) (61)(62)(63), and quantum calculations on the ␣L I domain predicted that the Ca 2ϩ -bound state would have lower affinity than either the Mg 2ϩ -or Mn 2ϩ -bound state for ligand (71). However, some integrin I domains have shown differences in their relative abilities to bind calcium (70,72), suggesting that the affinity of each I domain MIDAS site for Ca 2ϩ may be different. Interestingly, even though we observed no effect of millimolar levels of Ca 2ϩ by DSC, several other measures provided clear evidence for a stabilizing effect of Ca 2ϩ on ␣1I. DXMS showed decreases in exchange rates throughout the ␣1I sequence. The concentration of urea required to denature ␣1I was increased by 2 M in the presence of 10 mM Ca 2ϩ , which was comparable in magnitude to the effect of Mn 2ϩ . Finally, Ca 2ϩ at an equimolar concentration reversed the stabilizing effect of 10 mM Mg 2ϩ ions. Thus, our results demonstrate that Ca 2ϩ is capable of binding to the ␣1I MIDAS and that even though Ca 2ϩ binding does not stabilize the I domain to thermal denaturation or support ligand binding, it significantly impacts the solution dynamics of the protein.
The DXMS patterns of ␣1I in Mg 2ϩ , Ca 2ϩ , and Mn 2ϩ were very similar, except for a striking difference in the ␣C helix, a structural element unique to collagen-binding integrins that is believed to be critical for binding and collagen subtype selectivity (6,59,73). Relative to apo-␣1I, saturating concentrations of Mg 2ϩ stabilized the ␣C helix toward deuterium exchange to a much greater extent than either Mn 2ϩ or Ca 2ϩ . This result was surprising, because both Mg 2ϩ and Mn 2ϩ support binding of collagen IV to ␣1␤1 at these concentrations (42,58), and crystallographic data on integrin I domains support the concept that Mg 2ϩ -and Mn 2ϩ -bound I domains form essentially identical structures (65). These data, in conjunction with the recently reported crystal structure of ␣1I with the activating mutation E317A (32), led us to propose a model for collagenbinding integrin I domain activation that provides a structural basis for the enhanced activating ability of Mn 2ϩ and the inability of Ca 2ϩ to mediate ligand binding (Fig. 9). The ␣1I(E317A) structure, which was the first to display an unwound ␣C helix in an I domain in the absence of a bound collagen peptide, contained a unique penta-coordinated Mg 2ϩ in the MIDAS, suggesting that a change from penta-to hexa-coordinated metal ion upon ligand binding drives allosteric conformational changes within ␣1I. In our model, the closed conformation, with a hexa-coordinate metal ion and tightly wound ␣C helix, is in equilibrium with the activated conformation, containing a penta-coordinate metal and unwound ␣C helix. Collagen ligand binding leads to conversion back to the more stable hexa-coordinate geometry and induces allosteric changes in the C-terminal helices. Importantly, Mg 2ϩ ions pay a significant energetic penalty for deviating from a hexacoordinate (octahedral) geometry, whereas Mn 2ϩ is capable of adopting lower coordination states (32). Thus, Mn 2ϩ , with a lower energetic barrier to adopting the activated conformation, is more effective at driving the opening of the ␣C helix and enabling ligand binding. Ca 2ϩ , which is unable to adopt a pentavalent conformation, is incapable of mediating ligand binding.  (at top of figure). Color coding on ␣1I is identical to that shown in A.
The observation of such striking metal ion-mediated differences in the ␣C helix conformation demonstrates the importance of studying the dynamics of integrins in solution as a complement to crystallographic studies. Because there are no crystal structures of free (not bound to ligand or antibody) ␣1 or ␣2 I domains in Mn 2ϩ , we cannot rule out the possibility that Mg 2ϩ and Mn 2ϩ would produce different conformations of ␣C even in a crystalline state. More likely, however, this difference is only detectable in solution, where crystal packing effects are eliminated. The concept that Mg 2ϩ and Mn 2ϩ induce different conformations of ␣C in solution is supported by tryptophan fluorescence studies on ␣1I, which demonstrated that binding of Mg 2ϩ led to conformational changes in the vicinity of Trp-158, which is located at the end of MIDAS loop 1, very near the ␣C helix in the three-dimensional structure of ␣1I, and that these changes differed from the effects of added Mn 2ϩ (72).
DXMS provided an ideal tool to characterize the conformational changes in ␣1I induced upon collagen binding and to test the existing models for integrin I domain/ligand interactions. The DXMS pattern of ␣1I bound to collagen showed significantly reduced exchange only in MIDAS loop 2, suggesting that this region of ␣1 integrin is a critical element mediating the interaction with collagen. Collagen also induced faster exchange in the ␣C, ␣6, and ␣7 helices, consistent with models in which ligand binding causes a downward shift of the C-terminal ␣7 helix of the I domain, leading to engagement of Glu-335 (in the case of ␣1 integrin) with the MIDAS site of the ␤ subunit I-like domain (1). It has been proposed that the ␣7 helix is intrinsically mobile, allowing it to trigger this process with minimal energetic penalty (5, 10). Our DXMS data on ␣1I indicate that the flexibility of ␣7 is greatly increased upon ligand binding. We therefore propose a modification to this model in which ligand binding is responsible for relaxing the ␣7 helical structure and facilitating the conformational changes necessary for effective intersubunit communication, e.g. in outside-in signaling.
One of the key differences between the studies reported here and those previously described is the use of full-length collagen instead of a THP fragment. Solution binding studies on ␣2 I domain binding to a collagen THP have been reported using NMR (74). A number of residues located near the ␣7 helix showed changes in chemical shift pattern, although only one residue (Gly-329, equivalent to Gly-328 in ␣1I) within ␣7 itself was noted. The relatively small number of chemical shift changes observed in this region compared with our observations using DXMS on ␣1I-collagen could be due to limitations of the NMR method, such as the inherently reduced peak intensity observed upon peptide binding. Alternatively, the use of a THP instead of a full-length collagen ligand may have led to less robust conformational changes in the integrin. It is important to note that a single molecule of collagen contains multiple integrin-binding sites with differing affinities (75). This has the  potential to lead to dispersion in the DXMS signal. It is unclear whether different residues on ␣1I are involved in binding to different collagen GXXGER motifs. If so, one could imagine that DXMS studies using ␣1I bound to individual THP units might be informative.
In contrast to collagen, binding of the function-blocking ␣1 antibody AQC2 led to much more limited changes in the DXMS pattern of ␣1I. In the crystal structure of the AQC2-␣1I complex, Asp-101 from the antibody heavy chain bound directly to the metal ion in the ␣1I MIDAS, but the remainder of the I domain, including the C terminus, retained the closed conformation (30). We found that AQC2 binding selectively protected MIDAS loop 2, the same region protected from exchange by collagen. The similarity of AQC2 and collagen in their interactions with the metal ion and MIDAS loop 2, coupled with the stark differences in their abilities to transmit allosteric changes to the C terminus of ␣1I, is remarkable.
Two other examples of ligand-mimetic I domain antibodies have been reported, both of which also present Asp residues that interact with the MIDAS metal ion. One of these is the ␣L antibody AL-57 (76). Unlike AQC2, AL-57 binds to the open form of ␣L (68). It was suggested that a key salt bridge formed with Glu-241 of ␣L by either the native ligand ICAM-1 or AL-57 is necessary for the allosteric rearrangement to occur and that the reason AQC2 binds to the open form of ␣1I is that it does not form a similar bridge with the corresponding residue in ␣1I, Glu-255 (68). Unlike in the ␣L-ICAM interaction, this conserved glutamic acid (Glu-255 in ␣1 or Glu-256 in ␣2) is not critical for collagen binding by integrins (6,67). Nonetheless, the finding that collagen but not AQC2 can induce long range conformational changes in ␣1I is consistent with the conclusion that AL-57 is a true ligand mimetic, whereas AQC2 coopts a key element of the ligand-binding site (i.e. the MIDAS cation) without engaging other regions necessary for induction of signaling. One would predict that solution DXMS studies using AL-57 and the ␣L I domain should demonstrate pronounced antibody-induced conformational changes in the ␣7 helix. A true ligand-mimetic antibody would be less desirable as a therapeutic, however, because it could lead to agonistic effects through outside-in integrin signaling. Another related antibody, mAb107, preferentially binds to the ␣M I domain in Ca 2ϩ via a bidentate interaction with Asp-107 (61,77). Similarly to AQC2, ligation of mAb107 did not induce conversion to the open conformation. Interestingly, the bidentate mode of binding led to a heptavalent Ca 2ϩ ion in the mAb107-I domain complex, highlighting the range of different geometries accessible by the MIDAS.
In summary, we have explored the dynamic conformational changes in the ␣1 integrin I domain induced by divalent cations and by ligand binding, and we tested several elements of the existing models for collagen-induced conformational changes and allosteric regulation, using DXMS. The data validated some elements of the models derived from crystallographic studies but importantly identified novel features of ligand and metal ion binding that were not detected using static methods. The application of DXMS to other integrin/ligand interactions should be very powerful, as it provides a method to explore the interaction of isolated I domains or even complete integrin het-erodimers with native ligands, and it does not require the use of either conformationally restricted integrins (e.g. locked-open I domains) or engineered fragments of ligands (e.g. collagen THPs or RGD peptides).