Integrin α2I Domain Recognizes Type I and Type IV Collagens by Different Mechanisms

The collagens are recognized by the αI domains of the collagen receptor integrins. A common structural feature in the collagen-binding αI domains is the presence of an extra helix, named helix αC. However, its participation in collagen binding has not been shown. Here, we have deleted the helix αC in the α2I domain and tested the function of the resultant recombinant protein (ΔαCα2I) by using a real-time biosensor. The ΔαCα2I domain had reduced affinity for type I collagen (430 ± 90 nm) when compared with wild-type α2I domain (90 ± 30 nm), indicating both the importance of helix αC in type I collagen binding and that the collagen binding surface in α2I domain is located near the metal ion-dependent adhesion site. Previous studies have suggested that the charged amino acid residues, surrounding the metal ion-dependent adhesion site but not interacting with Mg2+, may play an important role in the recognition of type I collagen. Direct evidence indicating the participation of these residues in collagen recognition has been missing. To test this idea, we produced a set of recombinant α2I domains with mutations, namely D219A, D219N, D219R, E256Q, D259N, D292N, and E299Q. Mutations in amino acids Asp219, Asp259, Asp292, and Glu299 resulted in weakened affinity for type I collagen. When α2 D219N and D292N mutations were introduced separately into α2β1 integrin expressed on Chinese hamster ovary cells, no alterations in the cell spreading on type I collagen were detected. However, Chinese hamster ovary cells expressing double mutated α2 D219N/D292N integrin showed remarkably slower spreading on type I collagen, while spreading on type IV collagen was not affected. The data indicate that α2I domain binds to type I collagen with a different mechanism than to type IV collagen.

The collagens are recognized by the ␣I domains of the collagen receptor integrins. A common structural feature in the collagen-binding ␣I domains is the presence of an extra helix, named helix ␣C. However, its participation in collagen binding has not been shown. Here, we have deleted the helix ␣C in the ␣ 2 I domain and tested the function of the resultant recombinant protein (⌬␣C␣ 2 I) by using a real-time biosensor. The ⌬␣C␣ 2 I domain had reduced affinity for type I collagen (430 ؎ 90 nM) when compared with wild-type ␣ 2 I domain (90 ؎ 30 nM), indicating both the importance of helix ␣C in type I collagen binding and that the collagen binding surface in ␣ 2 I domain is located near the metal ion-dependent adhesion site. Previous studies have suggested that the charged amino acid residues, surrounding the metal iondependent adhesion site but not interacting with Mg 2؉ , may play an important role in the recognition of type I collagen. Direct evidence indicating the participation of these residues in collagen recognition has been missing. To test this idea, we produced a set of recombinant ␣ 2 I domains with mutations, namely D219A, D219N, D219R, E256Q, D259N, D292N, and E299Q. Mutations in amino acids Asp 219 , Asp 259 , Asp 292 , and Glu 299 resulted in weakened affinity for type I collagen. When ␣ 2 D219N and D292N mutations were introduced separately into ␣ 2 ␤ 1 integrin expressed on Chinese hamster ovary cells, no alterations in the cell spreading on type I collagen were detected. However, Chinese hamster ovary cells expressing double mutated ␣ 2 D219N/D292N integrin showed remarkably slower spreading on type I collagen, while spreading on type IV collagen was not affected. The data indicate that ␣ 2 I domain binds to type I collagen with a different mechanism than to type IV collagen.
The integrins are a large family of transmembrane proteins forming ␣/␤ heterodimers, which act as receptors for connective tissue components and also bind to counter receptors on other cells (1,2). Integrin ␣ and ␤ subunits show some similarity but have evolved independently (3). Integrin ␣ subunits can be divided into two different subgroups depending on the presence of an inserted domain (I domain) in their N-terminal part. In general, the ␣ subunits participating in the formation of extracellular matrix receptors do not contain ␣I domain, whereas it is found in all four ␤ 2 integrin-associated ␣ subunits involved in cell-cell adhesion of leukocytes. The four collagen-binding integrin ␣ subunits, namely ␣ 1 , ␣ 2 , ␣ 10 , and ␣ 11 , are an exception; they contain an ␣I domain, and at least ␣ 1 I and ␣ 2 I domains can, as recombinant proteins or proteolytic fragments, bind to collagen (4,5). Thus, the collagen receptor integrins have a unique ligand binding mechanism when compared with the interaction of the other matrix receptor integrins with their ligands. Indeed, whereas many functional motifs in integrin ligands contain aspartic acid in a critical position (6,7), often situated at a protruding loop (8), these motifs are not recognized by ␣I domains (9). Integrin ␣I domains contain a Mg 2ϩ ion that might coordinate the folding of the domain (10). There is some evidence that the Mg 2ϩ ion also forms the actual binding site for the ligands (10). One model of collagen binding to ␣ 2 I domain suggests that a glutamic acid residue on the surface of the tropocollagen molecule binds to Mg 2ϩ in the ␣ 2 I domain (11). Some evidence suggests, however, that the metal ion is displaced from the ␣ 2 I domain during the process of ligand binding (12). The initial recognition of collagen and also its binding might be a complex process involving multiple steps and interactions. Furthermore, it seems to be evident that both the fibrillar collagens, such as type I collagen (13), and the basement membrane-associated network-forming type IV collagen (5,14) have multiple individual sites recognized by integrins.
We have recently described a short cyclic peptide that binds to ␣ 2 I domain and prevents its interaction with matrix molecules, including type I and type IV collagens (15). Three amino acids, RKK, and the proper cyclic conformation seem to be critical for the function of the peptide (15). The structure of the peptide suggests the importance of charged amino acid residues in collagen recognition. Other studies support this as well. For example, in type IV collagen, three amino acid residues, one arginine and two aspartic acids, all from separate collagen ␣ chains, might be necessary for receptor binding (16). Recently, an ␣ 2 ␤ 1 integrin binding type I collagen-derived triple helical peptide was described (17). The importance of a short sequence, GER, and especially the essential role of the two charged amino acids in it, also suggest that interactions between positively and negatively charged amino acid residues are involved in integrin-collagen binding. However, in previous studies collagen binding could be affected only by mutation of the amino acid residues directly interacting with Mg 2ϩ ion (4).
Integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 are known to have some differences in their ligand binding specificity. Integrin ␣ 1 ␤ 1 binds with better affinity to type IV collagen than to type I collagen or to other fibrillar collagens, whereas ␣ 2 ␤ 1 is a better receptor for fibrillar collagen than for type IV collagen (18). A similar difference can be seen also in the function of the corresponding recombinant ␣I domains (5) and a chimeric ␣ 1 ␤ 1 integrin containing ␣ 2 I domain instead of ␣ 1 I has ligand binding functions similar to ␣ 2 ␤ 1 (19). The determination of the atomic structure of ␣ 1 I (20 -22) and ␣ 2 I (11) domains has made it possible to estimate their similarities and differences and to study the structural basis for the distinct function of the two collagen receptors. A common feature of the collagen-binding ␣I domains is the presence of an extra ␣ helix, helix ␣C (11, 20 -22). It is probably present also in the collagen-binding ␣ 10 I and ␣ 11 I domains (23,24) but missing from the other integrin ␣I domains (10,25). Here, we have made mutations to the ␣ 2 I domain and produced the variants as recombinant proteins. Some of the mutations were also introduced into ␣ 2 ␤ 1 integrin expressed on CHO 1 cells. Our results indicate the important role of helix ␣C and several charged amino acids surrounding the Mg 2ϩ binding site in the recognition of type I collagen. Furthermore, ␣ 2 ␤ 1 integrin seems to recognize type IV collagen by a mechanism different from that used in type I collagen binding.

EXPERIMENTAL PROCEDURES
Generation of ␣ 2 I and Mutant I Domains-Recombinant ␣ 2 I domain was produced as described previously (15). Protein concentrations were determined using Bradford's method (26). Protein purity and folding were checked in native and SDS-polyacrylamide gel electrophoresis (Phast-System; Amersham Pharmacia Biotech).
Site-specific mutations in the ␣ 2 I domain were made using the Stratagene QuickChange mutagenesis kit, essentially following the manufacturer's instructions. PCR primers having the desired mutation for both DNA-strands were designed, and then PCR was performed using Pfu polymerase (Stratagene), which makes at 68°C exactly one copy of the whole GEX-2T vector (Amersham Pharmacia Biotech) containing the ␣ 2 I domain sequence. The PCR was digested with DpnI, which cuts only methylated DNA (i.e. only the template is digested). After this, PCR product DNA strands having the desired mutation were paired. The resulting GEX-2T vector having the mutated ␣ 2 I domain was transformed into Escherichia coli strain DH5␣, and the construct was checked by sequencing the entire ␣ 2 I domain. For the production of the recombinant protein, the construction was transformed into E. coli strain BL21.
Binding Assay with Europium-labeled I Domains-The I domain binding assay, based on the use of europium labeling, was used as described previously (15). Labeling of I domains with europium was carried out using 1 ⁄10 volume of 1 M NaHCO 3 (pH 8.5) with I domains, and the labeling reagent (Wallac) was added in the molar ratio 0.3-2:1 (label:protein) and incubated overnight at 4°C. Unbound label was removed by gel filtration on a Sephadex G50 column (Amersham Pharmacia Biotech), and the fractions containing the labeled protein were pooled. Typically, coating of 96-well immunoplates (Maxisorp, Nunc) was performed by exposing the plates to 0.1 ml of PBS containing 5 g/cm 2 (about 16 g/ml) type I collagen (from rat tail; Sigma) or type IV collagen (native, isolated from basement membrane of Engelbreth-Holm-Swarm mouse sarcoma; Sigma), overnight at 4°C. Alternatively, matrix proteins (5 g/cm 2 ) were coated on 96-well amine binding plates (Costar) according to the manufacturer's instructions. Residual protein absorption sites on all wells were blocked with the Dilution II solution (including 7.5% bovine serum albumin (BSA); Wallac) in PBS for 1-2 h at 37°C. Europium-labeled ␣ 2 I was added at a concentration of 1 g/ml in PBS, 2 mM MgCl 2 to the coated wells and incubated for 3 h at 37°C. Wells were then washed three times with PBS, 2 mM MgCl 2 . Finally, 0.1 ml of Delfia enhancement solution (Wallac) was added to each well, and the Europium signal was measured by fluorometry (Victor 2 or model 1232 Delfia; Wallac).
Cell Lines and Construction of the ␣ 2 Integrin Expression Plasmid-CHO cells obtained from the American Type Culture Collection (Manassas, VA) were used as hosts for expression of integrin ␣ 2 subunits. Integrin ␣ 2 cDNA corresponding to nucleotides 1-4559 in the published sequence (27) was kindly provided by Dr. M. Hemler (Dana-Farber Cancer Institute, Boston, MA). cDNA was ligated into the pAWneo2 expression vector (Ref. 28; a kind gift from Dr. A. Weiss; University of California, San Francisco, CA), which carries the spleen focus-forming virus long terminal repeat promoter and a neomysin resistance gene. Stable transfections were carried out using Lipofectin reagent (Life Technologies Inc.) according to the manufacturer's recommendations. Neomysin analogue G418 (400 g/ml) was added to the culture medium. After 2-3 weeks of selection, the nontransfected control cells were dead, and G418-resistant clones were isolated and analyzed for their expression of ␣ 2 integrin by flow cytometry as follows. Cells were grown to early confluence and detached with trypsin-EDTA, and trypsin activity was inhibited by medium supplemented with serum. Cells were washed with PBS (pH 7.4) and then incubated with PBS containing 10 mg/ml BSA, 1 mg/ml glycine, and 0.02% NaN 3 for 20 min at 4°C. Cells were collected by centrifugation, exposed to a saturating concentration of monoclonal antibody against ␣ 2 integrin (12F1) in BSA/PBS (BSA concentration 1 mg/ml) containing NaN 3 for 30 min at 4°C and stained with rabbit anti-mouse IgG coupled to fluorescein (1:20 dilution; Dacopatts, Denmark) for 30 min at 4°C. Cells were washed twice with PBS containing NaN 3 and suspended in the same buffer. In order to measure the amount of ␣ 2 integrin on the cell surfaces, the fluorescent excitation spectra were analyzed by using a FACScan apparatus (Becton Dickinson). Control samples were prepared by treating the cells without primary antibodies.
Cell Spreading Experiments-96-well maxisorp plates (Nunc) or high binding microtitration plates (Nunc) were coated by exposing them to 0.1 ml of PBS containing 1 or 5 g/cm 2 (3 or 15 g/ml) type I or type IV collagen for 12 h at 4°C. Residual protein absorption sites on all wells were blocked with 0.1% heat-inactivated bovine serum albumin in PBS for 1 h at 37°C. CHO cell clones expressing ␣ 2 ␤ 1 integrins were used in spreading studies. Cells (10,000 cells/well) were allowed to spread in serum-free Dulbecco's modified Eagle's medium. Cells were fixed with 4% formaldehyde (Merck) and 5% sucrose (BDH) for 30 min, and at least three parallel wells (three parallel fields each) were examined and photographed by using a phase-contrast microscope. The total number of cells attached to collagen and the percentage of spread cells were counted. A spread cell was characterized as one having a clearly visible ring of cytoplasm around the nucleus.
IASYS Experiments-IASYS experiments were performed on the IASYS Auto plus apparatus (Affinity Sensors, Ltd.). A carboxymethyldextran cuvette was coated with type I collagen according to the manufacturer's instructions. The coupling buffer was sodium acetate, pH 5.5, and in the coupling process the type I collagen concentration was 100 g/ml. Type I collagen was coupled to the cuvette up to the saturating level (based on the reading of the apparatus).
A concentration series from 10 to 300 g/ml in PBS, 2 mM MgCl 2 for wild-type and mutant ␣ 2 I domains was measured. Regeneration of the cuvette was performed with 10 mM or 100 mM EDTA in PBS, and the task was not always easy; for the tightest bound I domain, several successive regeneration steps were needed to remove most of the bound I domain from collagen. Results were analyzed with Fastfit software from IASYS.
Molecular Modeling of ⌬␣C␣ 2 I-A three-dimensional model of the structure of the ⌬␣C␣ 2 I domain was built using MALIGN (29,30) in the BODIL modeling package 2 and MODELLER 4.0 (31). The model was constructed almost entirely from the x-ray structure of the ␣ 2 I domain (11), but the ␣MI (10) and ␣LI (25) domains were used to build the structure in the vicinity of the deleted helix. This was done because the number of residues in that area in ⌬␣C␣ 2 I is exactly the same as in ␣M and ␣L. This makes the modeling more reliable in that region. The x-ray structures were obtained from the Protein Data Bank (32).

Deletion of Helix ␣C in the ␣ 2 I Domain Inhibits Binding to
Type I Collagen-The helix ␣C is a unique structural feature shared by the collagen-binding ␣I domains. Furthermore, one of the major differences between the putative collagen binding surfaces of ␣ 1 I and ␣ 2 I domains (the ␣ 2 I domain is shown in Fig. 1) is in the structure of helix ␣C. In the ␣ 1 I, the groove situated on the metal ion-dependent adhesion site (MIDAS) face of the I domain is wider than in the ␣ 2 I, because ␣ 1 I residues Ser 284 and Gly 288 , oriented toward the groove in the helix ␣C, are replaced in ␣ 2 I by the bulky residues tyrosine and asparagine (20 -22). To study the role of the helix ␣C for the collagen binding function of ␣ 2 I domain, a deletion was produced; residues 284 -288 (GYLNR) were deleted from the ␣ 2 I domain by using PCR with specifically designed primers. We used molecular modeling to estimate the effects of helix ␣C deletion on the ␣ 2 I domain (Fig. 1). The model predicted that the surface of mutant ␣ 2 I domain would become more negatively charged than the wild type and more flattened. In the x-ray structure of the ␣ 2 I domain, there is an arginine (Arg 288 ) at the end of helix ␣C. This arginine forms a salt bridge with glutamate (Glu 318 ) between strand ␤F and helix ␣ 7 . As a result of the deletion of helix ␣C in ␣ 2 I, Glu 318 and the adjacent aspartate (Asp 317 ) are placed in the vicinity of the MIDAS.
The recombinant ␣ 2 I domain lacking the helix ␣C (named as ⌬␣C␣ 2 I) was produced in E. coli as a GST fusion protein. To measure the effect of the helix ␣C deletion on the affinity of type I collagen binding, we used IASYS technology. For this, the ⌬␣C␣ 2 I domain was tested as a fusion protein (GST/␣ 2 I domain) to achieve greater mass and easier detection by IA-SYS. Collagen (100 g/ml) was attached chemically to the cuvette according to the manufacturer's instructions, and different concentrations of I domain (10 -300 g/ml) were added into the cuvette. The binding event was monitored in real time, and examples are shown in Fig. 2A. The ⌬␣C␣ 2 I binding kinetics was different when compared with wild-type ␣ 2 I domain. The association and dissociation parts were clearly faster than with the wild type, and the equilibrium was reached quickly (Fig. 2A). The total effect of these changes was seen on K d values, which were estimated from Req (binding at equilibrium) versus ligand concentration: for wild-type K d was 90 Ϯ 30 nM, and for ⌬␣C␣ 2 I K d was 430 Ϯ 90 nM (Fig. 2B). As a conclusion, ⌬␣C␣ 2 I can bind type I collagen about twice as much as wild-type at high ligand concentrations, but the binding is about 5-fold weaker than for wild type. This result demonstrates, for the first time, the important role of the ␣C helix in type I collagen binding. Furthermore, it confirms the previous suggestions that the collagen binding surface of ␣ 2 I domain is located near the MIDAS site.
After GST was removed, the resultant ␣ 2 I domain was labeled with europium. Binding of the mutant and wild-type ␣ 2 I domains to collagen types I and IV was tested. Wells were coated with different concentrations of collagen type I or IV (1-15 g/cm 2 ), incubated with europium-labeled I domain (1 g/ml), and bound I domain was measured using a fluorescence spectrophotometer. These experiments suggested that the recognition of collagen subtypes by ⌬␣C␣ 2 I may be different when compared with the wild-type ␣ 2 I domain. The mutant preferred type IV collagen over type I collagen (Fig. 3), indicating differences in type IV collagen binding mechanism when compared with type I collagen binding. Interestingly, the collagen binding pattern of ⌬␣C␣ 2 I domain resembled the one previously described for ␣ 1 I domain (5), suggesting that the differences in ␣C structure might partially explain the functional differences between ␣ 1 I and ␣ 2 I domains. In collagen-binding ␣I domains, there is one extra amino acid residue in the loop between helices ␣ 3 and ␣ 4 . However, in ␣ 1 I and ␣ 2 I domains, the extra amino acid has an opposite charge (Asp 219 in ␣ 2 I domain, Arg 218 in ␣ 1 I domain), suggesting another possible mechanism for the different functions of the two integrins. To test this possibility, we produced a mutant ␣ 2 I domain, D219R. This mutation did not, however, have a clear effect on the ratio of  IASYS binding curves for the wild-type ␣ 2 I and the mutant ⌬␣C␣ 2 I GST fusion proteins to type I collagen. Fusion protein concentration was 100 g/ml, and buffer was PBS, 2 mM MgCl 2 . Temperature was 25°C (A). Shown are K d (binding to type I collagen) determination of the wild-type ␣ 2 I (E) and the ⌬␣C␣ 2 I GST fusion protein (q). Binding at equilibrium was determined as a function of fusion protein concentration, and results were fitted to a Michaelis-Menten form equation (solid lines) to obtain K d . Running buffer was PBS, 2 mM MgCl 2 , and temperature was 25°C (B). type I collagen/type IV collagen binding (Fig. 3).
Negatively Charged Amino Acids Surrounding the Mg 2ϩ Binding Site in ␣ 2 I Domain Are Involved in Type I Collagen Binding-Five negatively charged amino acid residues in the ␣ 2 I domain were mutated, and the effects on ligand binding were tested. The ␣ 2 I domain variants were prepared by using PCR with primers having the desired mutation, and the entire DNA of the mutated I domain was checked by sequencing. The mutations were the following: D219A, D219N, D219R, E256Q, D259A, D259N, D292N, and E299Q. Mutant ␣ 2 I domains were expressed as GST fusion proteins, and they were purified as both "plain" I domains and as fusion proteins (GST/I domain). Mutant ␣ 2 I domain D259A could not be expressed in E. coli BL21, and it was not studied any further. Expression levels of ␣ 2 I domain D259N were also quite low. All expressed proteins were routinely checked by SDS and native gel electrophoresis for purity and quality of protein. In native gel electrophoresis, every ␣ 2 I domain protein sample had a small extra band having larger mass than the true I domain band (not shown). The extra band disappeared with a reducing agent (dithiothreitol). This extra band is likely to be a small proportion of misfolded I domain that was able to refold back to correct conformation when dithiothreitol was added. A second, shorter version of wild-type ␣ 2 I domain was prepared, where the amino terminus of the I domain was deleted at the position of the first cysteine (Cys 140 ). This protein, named as wild type (⌬C140), did not show an extra band in native gel electrophoresis (not shown) under nonreducing conditions, but unfortunately the collagen binding ability of wild type (⌬C140) decreased significantly. All tested mutant ␣2I domains were based on the first wild-type version.
Since mutations in the ␣ 2 I domain were made close to the MIDAS, it was necessary to check that none of the mutations affected significantly the metal dependence of the I domain binding. The binding of mutant ␣ 2 I domains to type I collagen was measured at four different magnesium concentrations (1, 2, 5, or 10 mM) and with 10 mM EDTA, which was used as "0 mM." Eu-labeled I domain (1 g/ml) was incubated in collagencoated (5 g/cm 2 ) wells, and bound I domain was measured. The results are shown in Fig. 4. The data indicate that the produced mutations, including ⌬␣C␣ 2 I, have no effects on metal binding site, and the concentration of 2 mM Mg 2ϩ , conditions where all ␣I domains were tested, represent satisfac-tory saturating conditions.
The effects of the mutations on collagen binding were tested by using IASYS technology. The mutant ␣ 2 I domains were produced as GST fusion proteins. The binding event was monitored in real time, and examples are shown in Fig. 5. Concentration series were measured for each ␣ 2 I domain, and the results representing saturating conditions (300 g/ml) are listed in Table I. The wild type (⌬C140) showed only about 20 -40% binding when compared with wild-type ␣ 2 I domain, indicating the importance of the amino terminus of the I domain. ␣ 2 I mutants D219A, D219N, D219R, D259N, D292N, and E299Q bound collagen at 5-60% of the wild-type levels ( Table I). While the other mutants gave satisfactory reproducible results, the binding levels of the E256Q ␣ 2 I domain mutant were more variable, and the measured binding to type I collagen could be at wild-type level (35-100% compared with the wild-type binding levels). The participation of Glu 256 , as well as Asp 219 , in collagen binding was therefore studied further by first removing the GST tag and then analyzing its binding by Biacore technology and solid phase binding assays (not shown). In these experiments, D219N showed constantly lowered binding to type I collagen, as was seen already by using IASYS, whereas the mutation E256Q had no effect. The results indicate that at least four negatively charged amino acids in ␣ 2 I, namely Asp 219 , Asp 259 , Asp 292 , and Glu 299 , have a significant role in collagen I binding.
Double Mutation D219N/D292N in the ␣ 2 Subunit of ␣ 2 ␤ 1 Integrin Affects CHO Cell Spreading on Type I Collagen but Not on Type IV Collagen-Amino acids Asp 219 and Asp 292 were found to be essential for ␣ 2 I domain binding to collagen. To test the importance of these residues for the function in the complete receptor, mutations D219N and D292N were introduced into full-length ␣ 2 cDNA before it was transfected into CHO cells. Wild-type CHO cells have no endogenous collagen receptors and cannot bind to collagens. They produce, however, the ␤ 1 integrin subunit, and therefore ␣ 2 ␤ 1 integrin is expressed on the cell surface after ␣ 2 cDNA transfection (not shown). The expression levels of ␣ 2 ␤ 1 integrin on the cell surface were analyzed by flow cytometry. Two ␣ 2 transfected cell clones (clone 9 had about 10% of ␣ 2 ␤ 1 integrin expression when compared with clone 12) were used in all assays. Cell clones harboring the D219N or D292N mutations or the double mutation D219N/D292N had ␣ 2 ␤ 1 expression levels comparable with the FIG. 3. Helix ␣C deletion (⌬␣C␣ 2 I) changes the type I/type IV collagen binding profile. Shown is a solid phase binding assay of wild-type ␣ 2 I and mutants ⌬␣C␣ 2 I and D219R to type I (q) and type IV (E) collagens. Wells were coated with proper collagen concentrations and were blocked with BSA. Eu-labeled ␣ 2 I (1 ng/l in PBS, 2 mM MgCl 2 ) was added into the wells for 3 h at 37°C. Wells were washed three times, and bound ␣ 2 I was measured with time-resolved fluorescence. BSA coating was used as control, and its signal was subtracted from the signal obtained from the samples. The data shown are the mean values Ϯ S.D. of a representative experiment done in triplicate.
CHO-␣2 clone 12 (Table II). To test the function of ␣ 2 ␤ 1 integrin, we measured the spreading of cells plated on type I collagen in a 2-h spreading assay. Neither of the cell clones harboring a single point mutation showed any significant alterations in the spreading rate (Table II). However, CHO cells expressing doubly mutated ␣ 2 integrin spread on type I collagen remarkably more slowly than the CHO-␣ 2 cells (Table II, Fig. 6). Their spreading was comparable with the clone 9 expressing 10-fold less ␣ 2 ␤ 1 integrin (not shown). The fact that one mutation alone is not enough to cause significant changes in this integrin function supports the idea that several charged amino acids in the ␣ 2 I domain participate at the same time in type I collagen binding. Interestingly, no differences in cell spreading were seen on type IV collagen ( Fig. 6; Table II), indicating that the ␣ 2 I domain binds to type I collagen by using a different mechanism than is used to recognize type IV collagen.

DISCUSSION
The development of collagen-like cell adhesion proteins and their cellular integrin-type receptors has been suggested to be one of the steps critical for the evolution of metazoan organisms (33). In modern vertebrates, collagens are recognized by at least four I domain-containing integrins, namely ␣ 1 ␤ 1 , ␣ 2 ␤ 1 , ␣ 10 ␤ 1 , and ␣ 11 ␤ 1 . Other matrix receptor integrin ␣ subunits do not have the ␣I domain, and, somewhat surprisingly, the collagen-binding ␣ integrins seem to belong to the same evolutionary branch of the integrin family tree as the leukocyte cell-cell adhesion integrin ␣ subunits (3). One cell type, for example chondrocytes, can express three collagen receptors at the same time (23,34). This raises the question of whether the three receptors have, despite their structural similarities, their own specific function. Indeed, ␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrins seem to differ at least in their signaling function (35). Furthermore, ␣ 1 ␤ 1 is a better receptor for type IV collagen than ␣ 2 ␤ 1 , whereas ␣ 2 ␤ 1 binds type I collagen with better affinity (5,18,19). The distinct signaling functions may be due to differences in intracellular (35) and transmembrane domains, while the ligand binding specificity is dictated by the structural properties of the ␣I domains.
The three-dimensional structures of the ␣ 1 I and ␣ 2 I domains have been solved recently by x-ray crystallography (11, 20 -22), but the mechanism of collagen recognition and binding is mostly unknown. One possibility is that Mg 2ϩ , present in all ␣I domains at the MIDAS site (10, 11, 20 -22, 25), directly mediates also collagen binding (11). Although the exact role of the metal is still under discussion, most models have located the FIG. 5. IASYS assays. IASYS binding curves are shown for the wild-type ␣2I, D259N, and D292N GST fusion protein to type I collagen. Fusion protein concentration was 300 g/ml. Buffer was PBS, 2 mM MgCl 2 , and temperature was 25°C.  4. The effect of magnesium concentration on ␣ 2 I binding to type I collagen. Solid phase binding assay of the wild-type ␣ 2 I, ⌬␣C␣ 2 I, D219R, and D292N to type I collagen as a function of magnesium concentration. Wells were coated with 5 g/cm 2 type I collagen and were blocked with BSA. Eu-labeled ␣ 2 I (1 ng/l) in PBS buffer with a proper magnesium concentration was added into wells for 3 h at 37°C. Wells were washed three times, and bound ␣2I (E) was measured with time-resolved fluorescence. BSA coating was used as a control (q). The data shown are the mean values Ϯ S.D. of a representative experiment performed in triplicate.
collagen binding site close to the MIDAS. On the putative collagen binding surface of ␣ 1 I and ␣ 2 I domains, and probably on ␣ 10 I and ␣ 11 I based on the DNA sequence, helix ␣C helps to form a suitable groove for the collagen molecule to bind to (11, 20 -22). However, in previous reports, the mutation of single amino acid residues on that surface, D219A, E256A, or D292A, did not affect CHO cell attachment on collagen (4), unlike mutations to residues directly participating in Mg 2ϩ binding (36).
We have tested the binding of recombinant mutant ␣ 2 I domains to type I collagen by using solid phase binding assays and IASYS technology. IASYS biosensor measures the mass changes during the binding process indirectly. With this technique, the kinetics of binding can be monitored in real time. The experiments indicated that residues Asp 219 , Asp 259 , Asp 292 , and Glu 299 are important for type I collagen binding by ␣ 2 I domain. However, none of the mutations could completely prevent collagen binding, suggesting that several amino acid residues on the binding surface may simultaneously contribute to the phenomenon. The mutations in amino acids Asp 219 and Asp 292 were also introduced to full-length ␣ 2 cDNA, which was then expressed in CHO cells. In place of the cell attachment assay, we used a cell spreading assay that, according to our experience, is more sensitive. In agreement with a previous report (4), we were not able to see differences when compared with CHO cells expressing wild-type ␣ 2 integrin. However, the importance of Asp 219 and Asp 292 for collagen binding could be confirmed in experiments in which both amino acids had been mutated at the same time. The possibility that Asp 219 may participate in collagen binding has been suggested previously based on molecular modeling (11). We have recently published indirect evidence that Asp 259 and Asp 292 may participate in collagen binding by suggesting their interaction with RKK peptides inhibiting collagen binding (37). Here we show the first direct evidence that amino acid residues next to the MI-DAS site, but not essential for Mg 2ϩ binding, can participate in collagen recognition. We also suggest that several residues make concomitant and more or less equal contribution to the phenomenon.
Another interesting observation made with the CHO cells harboring the double mutant ␣ 2 D219N/D292N was that, de-spite the fact that their spreading on type I collagen was affected, they could spread on type IV collagen with similar efficiency as the wild-type ␣ 2 integrin-expressing cells. This finding clearly indicates that type I and type IV collagens are recognized by distinct mechanisms. It is also tempting to speculate that by using the same approach it is possible to reveal the structural reasons for proposed differences in ligand binding by ␣ 1 I and ␣ 2 I domains (5). Neither Asp 219 nor Asp 292 is conserved in ␣ 1 I domain, but the corresponding amino acid residues are arginine and serine, respectively. The idea that the fact that ␣ 1 I binds better to type IV collagen than ␣ 2 I could be due to one single amino acid residue was tested by making the mutation D219R. However, the mutant ␣ 2 I domain did not favor type IV collagen over type I collagen, indicating that the hypothesis was wrong.
A major structural difference between the putative ligand binding surfaces of ␣ 1 I and ␣ 2 I domains is the bulky helix ␣C in ␣ 2 I, which limits the width of the neighboring groove. Despite the fact that helix ␣C is found only in collagen-binding ␣I domains, its participation in collagen recognition has never been shown. Here, the importance of helix ␣C was tested by deleting it from ␣ 2 I. No drastic structural changes were expected, since helix ␣C is a separate "loop" on the surface of the ␣ 2 I domain. It was encouraging to observe that there was no change in metal dependence of ligand binding. IASYS experiments showed that helix ␣C was important for collagen type I binding. K d of wild-type ␣ 2 I binding to type I collagen was close to values reported earlier and measured with Biacore technology in the presence of manganese (5). The K d of ⌬␣C␣ 2 I was about 5-fold higher than the K d of the wild type. The curious kinetics of ⌬␣C␣ 2 I can be explained so that after the deletion more space is available for the binding of type I collagen to the surface of the I domain. In the absence of the helix ␣C, the specificity of the binding is diminished, and the binding and release of the ligand occurs more easily. The experiments measuring recombinant ⌬␣C␣ 2 I binding to type I and IV collagens showed that the binding profile may be different when compared with the profile of the wild-type ␣ 2 I. The ⌬␣C␣ 2 I seemed to bind better to type IV collagen than to type I collagen, and type IV collagen binding was saturating at higher coating concentrations than type I collagen binding. Based on previous  D292N, D219N, or double mutation D219N/D292N (two clones: number 2 and 4) were used. 10,000 cells/well were allowed to spread in serum-free Dulbecco's modified Eagle's medium for 2 h. They were fixed, and at least three parallel wells were examined by using a phase-contrast microscope. The ratio of spread cells:total number of cells attached was counted. A spread cell was characterized as one having a clearly visible ring of cytoplasm around the nucleus. The level of ␣ 2 ␤ 1 integrin on cell surface was measured by flow cytometry (mean fluorescence value). In experiment (Expt.) 4, D219N clone had some ␣ 2 negative cells, and the result was confirmed by staining with anti-␣ 2 antibodies and counting the spread cells under a fluorescence microscope. studies, interaction between ␣ 2 I domain and type IV collagen is too weak to be measured by surface plasmon resonance technique (5) and could not be used to confirm this observation. Interestingly, the collagen binding profile of ⌬␣C␣ 2 I resembled what has been previously suggested for the ␣ 1 I domain (5). However, in our solid phase binding assays, the difference between type IV and type I collagen binding by ␣ 1 I domain has been very small. 3 Still, it could be speculated that the bulky helix ␣C in the ␣ 2 I domain narrows the collagen binding groove and may actually disturb the binding of type IV collagen, which contains interrupted sequences between triple helices and is therefore less compact than type I collagen.
To conclude, our results strongly support the earlier suggestions that the collagen binding surface in integrin ␣ 2 I domain is located around the metal ion-binding MIDAS site. Furthermore, we have shown the important role of helix ␣C and amino acid residues Asp 219 and Asp 292 in type I collagen binding. They may not, however, participate in type IV collagen binding, suggesting that the two collagen types are recognized by different mechanisms.
FIG. 6. The spreading of ␣ 2 integrin-transfected CHO cells on type I and type IV collagen. CHO cells were transfected with the wild-type or the variant ␣ 2 integrin. 96-well plates were coated by exposing them to type I (A) or type IV collagen (B). Residual protein absorption sites on all wells were blocked with heat-inactivated bovine serum albumin. CHO cell clones expressing the wild-type ␣ 2 integrin or ␣ 2 integrin with double mutation D219N/D292N (A; two cell clones, numbers 2 and 4) were used in spreading studies. Cells (10,000 cells/ well) were allowed to spread in serum-free Dulbecco's modified Eagle's medium for 1, 2, or 4 h. They were fixed, and at least three parallel wells (three parallel fields each) were examined and photographed by using a phase-contrast microscope. The number of spread cells was counted and divided by the total number of cells attached to collagen. A spread cell was characterized as one having a clearly visible ring of cytoplasm around the nucleus.