Selective Binding of Collagen Subtypes by Integrin α1I, α2I, and α10I Domains*

Four integrins, namely α1β1, α2β1, α10β1, and α11β1, form a special subclass of cell adhesion receptors. They are all collagen receptors, and they recognize their ligands with an inserted domain (I domain) in their α subunit. We have produced the human integrin α10I domain as a recombinant protein to reveal its ligand binding specificity. In general, α10I did recognize collagen types I–VI and laminin-1 in a Mg2+-dependent manner, whereas its binding to tenascin was only slightly better than to albumin. When α10I was tested together with the α1I and α2I domains, all three I domains seemed to have their own collagen binding preferences. The integrin α2I domain bound much better to fibrillar collagens (I–III) than to basement membrane type IV collagen or to beaded filament-forming type VI collagen. Integrin α1I had the opposite binding pattern. The integrin α10I domain was similar to the α1I domain in that it bound very well to collagen types IV and VI. Based on the previously published atomic structures of the α1I and α2I domains, we modeled the structure of the α10I domain. The comparison of the three I domains revealed similarities and differences that could potentially explain their functional differences. Mutations were introduced into the αI domains, and their binding to types I, IV, and VI collagen was tested. In the α2I domain, Asp-219 is one of the amino acids previously suggested to interact directly with type I collagen. The corresponding amino acid in both the α1I and α10I domains is oppositely charged (Arg-218). The mutation D219R in the α2I domain changed the ligand binding pattern to resemble that of the α1I and α10I domains and, vice versa, the R218D mutation in the α1I and α10I domains created an α2I domain-like ligand binding pattern. Thus, all three collagen receptors appear to differ in their ability to recognize distinct collagen subtypes. The relatively small structural differences on their collagen binding surfaces may explain the functional specifics.

Collagens are abundant structural proteins in the extracellular matrix. So far, 19 different triple helical protein trimers have been classified as a collagen subtype (1). The collagens can be grouped into subclasses according to their structural details. Many collagen subtypes (namely types I, II, III, V, and XI) have long continuous triple helices, and they can form large fibrils. In other collagens the triple helix has interruptions. Some collagens form networks (types IV, VIII, and X) or beaded filaments (type VI). Other collagen subclasses include fibrilassociated collagen with short interruptions in the triple helices (collagen types IX, XII, XIV, XVI, and XIX), anchoring fibril-forming collagen (type VII), and transmembrane collagen (types XIII and XVII). Collagen types XV and XVIII are found in association with basement membranes (the multiplexins; see Ref. 2).
The integrins form a large family of heterodimeric cell surface receptors involved in cell-extracellular matrix as well as in cell-cell adhesion and communication. Of the 24 different integrin receptors presently known, 5 function as a collagen receptor (3). The collagen receptors are composed of the ␤ 1 subunit in complex with either an ␣ 1 , ␣ 2 , ␣ 3 , ␣ 10 , or an ␣ 11 subunit. Integrin ␣ 1 ␤ 1 is the collagen receptor of many mesenchymal cells. ␣ 1 null mice are viable (4), but they seem to have defects in the feedback regulation of collagen synthesis (5), in the regulation of matrix metalloproteinase expression (5,6), in tumor-related angiogenesis (6), and in lymphocyte function (7). Integrin ␣ 2 ␤ 1 is an important collagen receptor on platelets and epithelial cells (8). Little is known about the biology of the recently identified ␣ 10 ␤ 1 and ␣ 11 ␤ 1 integrins. ␣ 10 ␤ 1 is one of the collagen receptors on chondrocytes (9), and ␣ 11 ␤ 1 was originally found in fetal muscle (10). Integrin ␣ 3 ␤ 1 can bind some collagens, but it may function as an assisting collagen receptor rather than as a primary receptor (11).
One notable feature common to the primary collagen receptors is the existence of an I ("inserted") domain at the N terminus of the ␣ subunit. No other extracellular matrix receptor contains this independently folding domain, but five integrintype cell-cell adhesion receptors do have it. The I domain is built up of ␤ sheets surrounded by amphipathic ␣ helices (12). I domains are homologous to the A domains found in von Willebrand factor and in cartilage matrix protein, in some collagen subtypes, and in components of the complement system. The A and I domains are commonly involved in molecular interactions, and they are responsible for the collagen binding activity of von Willebrand factor and collagen receptor integrins (13)(14)(15).
At the "top" of the I domain, where ligands bind, resides a metal ion in a conserved coordination site called MIDAS 1 (metal ion-dependent adhesion site; see Ref. 16). Five amino acid side chains bind the magnesium ion, directly or through water molecules, and a water molecule or glutamate from a collagenous ligand may complete the coordination to the metal ion. The MIDAS site is centered on a groove restricted on one side by a helix called the ␣C helix. This ␣C helix can be found only in collagen binding integrin ␣I domains. Ligand binding induces a conformational change in the ␣ 2 I domain resulting in the unwinding of the ␣C helix and opening of the binding site (17). Recombinant ␣ 2 I domain missing the ␣C helix has been shown to have altered kinetics in binding type I collagen (18). A prominent feature in the ␣I domain binding surface is the presence of several charged amino acids surrounding the MIDAS site. The binding surface of ␣ 2 I is more negatively charged than the binding surfaces of ␣ 1 I and ␣ 10 I (19).
Despite the structural similarity of the integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 , their binding specificity differs from each other. Integrin ␣ 1 ␤ 1 prefers network-forming collagen type IV over fibrillar collagen type I, whereas ␣ 2 ␤ 1 binds type I collagen more strongly than collagen type IV (20). Moreover, ␣ 1 ␤ 1 is a much better receptor for type XIII collagen than ␣ 2 ␤ 1 (21). The same binding pattern can also be seen with the corresponding recombinant I domains. Here we have tested the binding of ␣ 1 I and ␣ 2 I domains to various collagen subtypes. We have also unveiled the ligands recognized by the ␣ 10 I domain. The similarities and differences in the ligand binding patterns were correlated with the atomic structures, and based on experiments with mutated ␣I domains, we suggest that Asp-219 in the ␣ 2 I domain, and the corresponding amino acid Arg-218 in the ␣ 1 I and ␣ 10 I domains, is a critical residue in the determination of collagen binding specificity.

MATERIALS AND METHODS
Cloning and Mutagenesis of the Human Integrin ␣ 10 I Domain-␣ 10 I domain cDNA was generated by reverse transcription-PCR from RNA isolated from KHOS-240 cells (human Caucasian osteosarcoma). Total cellular RNA was isolated by using an RNeasy Mini Kit (Qiagen). Reverse transcription-PCR was done using the Gene Amp PCR kit (PerkinElmer Life Sciences). The ␣ 10 I forward primer (5Ј-CAG GGA TCC CCA ACA TAC ATG GAT GTT GTC-3Ј) contained a BamHI restriction site at the 5Ј-end. The reverse primer (5Ј-GGC TGA ATT CCC CTT CAA GGC CAA AAA TCC G-3Ј) also contained an EcoRI restriction site at the 5Ј-end. The primers were delivered by CyberGene, Sweden. Forty cycles of PCR amplification were done in 2 mM MgCl 2 using the following protocol: denaturation for 1 min at 94°C, annealing for 1 min at 67°C, and extension for 2 min at 72°C. The amplified ␣ 10 I domain cDNA was digested along with the pMAL-c expression vector (New England Biolabs) or pGEX-2T expression vector (Amersham Pharmacia Biotech) using the BamHI and EcoRI restriction enzymes (Promega). The cDNA was ligated to the pMAL-c vector with T4 DNA ligase (Promega). To the pGEX-2T vector the ␣ 10 I domain cDNA was ligated with the SureClone ligation kit (Amersham Pharmacia Biotech). The constructs were transformed into the Escherichia coli BL21 strain for production. The DNA sequences of the constructs were checked with DNA sequencing and compared with the published ␣ 10 DNA sequence (10) (data not shown). Site-directed mutation of the ␣ 10 I domain cDNA in a pGEX-2T vector was made using PCR according to Stratagene's QuikChange mutagenesis kit instructions. Briefly, mutagenesis was based on primers that contained the planned mutations (primers were delivered by CyberGene). At 68°C the proofreading Pfu DNA polymerase (Promega) makes only one copy of the plasmid without replacing the primers. The mutated strands were then selected from parental ones by digesting the template with DpnI endonuclease (Promega), which digests only methylated and hemimethylated DNA. By that means only mutated strands remained. The presence of mutations was checked by DNA sequencing. Mutant constructs were then transformed into E. coli strain BL21 for production of recombinant protein.
Cloning and Mutagenesis of the Human Integrin ␣ 1 I Domain-The ␣ 1 I domain was cloned into the pGEX-4T vector (Amersham Pharmacia Biotech) as described earlier by Nykvist et al. (21). The full-length integrin ␣ 1 cDNA was a gift from Dr. E. Marcantonio (Columbia University, New York). The ␣ 1 I domain cDNA sequence differed from the originally published sequence (23) at one position, and as a result the amino acid lysine was replaced with glutamate at position 170 (numbered from the beginning of the mature peptide). Site-directed mutations to ␣ 1 I domain cDNA were made as stated above for the ␣ 10 I domain.
Cloning and Mutagenesis of the Human Integrin ␣ 2 I Domain-Human integrin ␣ 2 I domain was generated as described earlier by Ivaska et al. (24). The integrin ␣ 2 cDNA was a gift from Dr. M. Hemler, Dana Farber, Boston. The ␣ 2 I domain was cloned into the pGEX-2T vector (Amersham Pharmacia Biotech). The site-directed mutations were made as stated above for the ␣ 10 I domain.
Production and Purification of MBP Fusion Protein-To produce the wild type ␣ 10 I domain as a fusion with the maltose-binding protein (MBP), 500 ml of LB medium containing 50 g/ml ampicillin was inoculated with an overnight culture of 10 ml of E. coli BL21 cells transformed with the pMAL-c␣ 10 I vector. Cells were grown at 37°C for about 4 h. Protein production was induced with isopropyl-␤-D-thiogalactoside (Amersham Pharmacia Biotech) and allowed to continue for 2 h. The cells were pelleted by centrifugation (4,000 ϫ g, 20 min, 4°C). Cell pellets were resuspended in column buffer (20 mM Tris-HCl, pH 7.4, 200 mM NaCl, 1 mM EDTA, 10 mM ␤-mercaptoethanol) and stored at Ϫ20°C until purification. The cell suspension was thawed on ice and sonicated in 15-s bursts for 3 min. Cell debris was removed by centrifugation (9,000 ϫ g, 30 min, 4°C). The supernatant was incubated with amylose resin (New England BioLabs) in an end-over-end rotor at 4°C overnight. Amylose resin with bound MBP-␣ 10 I fusion was transferred to 10-ml columns (Bio-Rad) and washed three times with 10 ml of column buffer. The bound fusion protein was eluted with 5 mM maltose in column buffer after a 2-3-h incubation at room temperature. Maltose was removed using hydroxyapatite, first washed with column buffer. One ml of washed hydroxyapatite was mixed with eluted protein and incubated in an end-over-end rotor for 1 h at 4°C, after which the fusion protein was attached to the hydroxyapatite. Soluble maltose was washed away with column buffer, repeated three times. The fusion protein was eluted from the hydroxyapatite with sodium phosphate buffer after a 30-min incubation at 4°C. The protein concentrations were measured using Bradford's method (22). The purity and folding of the proteins were checked by SDS-and native polyacrylamide gel electrophoresis (PhastSystem, Amersham Pharmacia Biotech; data not shown).
Production and Purification of GST Fusion Proteins-The wild type and mutated ␣ 1 I, ␣ 2 I, and ␣ 10 I domains were produced as glutathione S-transferase (GST)-␣I domain fusion proteins. 500 ml of LB medium (with 50 g/ml ampicillin) was inoculated with an overnight culture of E. coli BL21 cells transformed with either wild type or mutant plasmid. The cultures were grown at 37°C until the A 600 was in the range of 0.5-1. Next, an inducer of the tac promoter, isopropyl ␤-D-thiogalactoside, was added to a final concentration of 0.4 mM, and the temperature was lowered to room temperature. The bacteria were allowed to produce the recombinant protein for 3-4 h after which the cultures were centrifuged (5,550 rpm, 10 min, 4°C) to collect the cells. Cells were stored until purification at Ϫ20°C overnight and at Ϫ70°C for longer times.
The bacterial cells were resuspended in PBS for purification. The cell suspension was sonicated and detergent (Triton X-100) was added to a final concentration of 1%. To dissolve the recombinant protein, the disrupted cells were incubated with agitation and Triton X-100 for 1 h.
Cell debris was removed by centrifugation (17,000 rpm, 10 min, 4°C). Glutathione-Sepharose 4B (Amersham Pharmacia Biotech) was added to the supernatant, and the mixture was incubated for 1-1.5 h in gentle agitation. The glutathione-Sepharose 4B was then transferred to columns (Bio-Rad), and unbound proteins were washed away with PBS. The bound fusion protein was eluted with reduced glutathione (Sigma). Protein concentrations were measured with Bradford's method (22). The purity and folding of the fusion proteins were checked with SDSand native polyacrylamide gel electrophoresis (data not shown).
Solid Phase Binding Assays-The binding of wild type or mutated I domains to collagen was studied using solid phase binding assays. Ninety-six-well microtiter plates (Wallac) were coated with each collagen type: I (rat tail, Sigma), II (bovine, Chemicon), III (human, Chemicon), IV (Engelbreth-Holm-Swarm mouse sarcoma basement membrane, Sigma), V (human, Chemicon), or VI (human, Biodesign International) or with laminin-1 (Engelbreth-Holm-Swarm mouse sarcoma basement membrane, Sigma), tenascin (chicken, Chemicon) 5 g/cm 2 in PBS and control wells with Diluent II (containing BSA, Wallac) 1:2 in PBS at 4°C overnight. Before carrying out the binding assays, wells were blocked with Diluent II for 1 h at 37°C. Dilutions of the I domains were made with assay buffer (Wallac) and added to the wells. A concentration series of 1-500 nM was used for the I domains. The I domains were allowed to attach to the wells for 1 h at 37°C in the presence of 2 mM MgCl 2 , unless stated otherwise. The unattached I domain was washed away with PBS containing 2 mM MgCl 2 , repeated three times. Detection of the I domain was based on the existence of the GST or MBP fusion partner. Goat anti-GST antibody (Amersham Pharmacia Biotech) or anti-MBP (New England Biolabs) was added to the wells in a 1:8,000 or 1:1,000 dilution, respectively, in assay buffer and incubated for 1 h at 37°C. The wells were then washed three times with PBS containing 2 mM MgCl 2 . Europium 3ϩ -labeled protein G (Wallac) (1:100 in assay buffer) was then added to the wells and incubated for 1 h at 37°C. Wells were washed three times with PBS containing MgCl 2 . Enhancement solution (Wallac) was used to dissociate the highly fluorescent Europium 3ϩ label. The fluorescence was measured with a timeresolved fluorometer (Victor 2 , Wallac). All assays were performed at least in triplicate.
Structures modeling package 2 using a structure-based sequence comparison matrix (29). MALIGN constructs a multiple sequence alignment from pairwise alignments according to a tree relating the sequences being matched. The alignment between the three I domains contains no gaps and for ␣ 1 I and ␣ 2 I matches their structural alignment. The sequence identity between the ␣ 2 I domain and the ␣ 1 I domain is 52% and between the ␣ 2 I domain and the ␣ 10 I domain is 46%, thus leading to high quality reliable models. The program MODELLER 4.0 (30) was used to construct three-dimensional model structures of the holo form of the ␣ 1 I domain and both the apo and holo forms of the ␣ 10 I domain.

RESULTS
␣ 10 I Domain Binds Type I-VI Collagens and Laminin-1 in a Magnesium-dependent Manner-␣ 10 I domain was produced in E. coli as a recombinant protein fused to MBP or GST. The recombinant proteins were purified with affinity chromatography and characterized with SDS-and native polyacrylamide gel electrophoresis (data not shown). The proteins appeared as one band in native polyacrylamide gel electrophoresis, suggesting the presence of a single molecular form. The binding of purified recombinant ␣ 10 I domain to collagen was studied using a solid phase binding assay. Wells were coated with different collagens, laminin-1, or tenascin (5 g/cm 2 ). ␣ 10 I fusion protein (300 nM) was allowed to bind to the coated wells. The ␣ 10 I domain was found to bind to cartilage-derived type II collagen in a magnesium-dependent manner (Fig. 1A) because 5 mM EDTA inhibited binding completely. Thus, the basic binding mechanism of ␣ 10 I is cation-dependent as are the other integrin ␣I domains (16,17). To investigate whether other fibrillar collagens, namely subtypes I, II, III, and V, also can function as ligands for ␣ 10 I, a binding assay was carried out (Fig. 1B). The ␣ 10 I domain bound to the collagen subtypes I-III nearly equally, but in repeated experiments the binding to type V collagen was weaker. Collagen receptors ␣ 1 ␤ 1 and ␣ 2 ␤ 1 can both bind to laminin-1 (31,32). ␣ 2 ␤ 1 is also able to bind tenascin (33). The binding of ␣ 10 I to these extracellular matrix proteins was also studied. The ␣ 10 I domain was found to bind to laminin-1, but binding to tenascin was only slightly better than to BSA (Fig. 1C). The ligand recognition patterns of ␣ 10 I-GST and ␣ 10 I-MBP were identical. In some experiments the ␣ 10 I-MBP showed higher unspecific binding to BSA than ␣ 10 I-GST (data not shown).
The Binding Specificities of ␣ 1 I and ␣ 2 I Domains Differ from Each Other-Integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 and their corresponding ␣I domains differ in collagen binding specificity even though they are structurally highly similar. Overall about half of their ␣I domain sequence is identical, and the ␣I domain binding surfaces resemble each other. The ␣ 1 I domain prefers type IV collagen over type I collagen, whereas the preference of ␣ 2 I is reversed (20). To study the binding of these ␣I domains to collagen type VI in addition to collagen types I and IV, binding assays with an ␣I domain concentration series of 1-500 nM were carried out (Fig. 2). The results were fit to the Michaelis-Menten equation to obtain estimates of the K d of binding. In accordance with previous studies, the integrin ␣ 1 I domain bound better to type IV collagen (K d Ϸ 65 Ϯ 20 nM; panel A) than type I collagen (K d Ϸ 160 Ϯ 40 nM). The integrin ␣ 2 I domain, in contrast, associated considerably better with collagen type I (K d Ϸ 20 Ϯ 5 nM) than with collagen type IV (K d Ϸ 140 Ϯ 30 nM; panel B).
Type VI collagen forms beaded filaments, and it has been reported to be a ligand of both ␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrins (34). However, in chondrocytes ␣ 1 ␤ 1 may be the preferred receptor for collagen type VI (35). Here, the binding of the ␣ 1 I and ␣ 2 I domains to type VI collagen was tested (Fig. 2). Integrin ␣ 1 I domain showed the strongest binding (K d Ϸ 200 Ϯ 80 nM; Fig.  2C), whereas the binding of the ␣ 2 I domain was so weak that no reasonable estimate of the half-maximal value for binding could be obtained (Fig. 2D).
Integrin ␣ 10 I Domain Binds Collagen Types I, IV, and VI with Nearly Equal Affinity-To test whether the ␣ 10 I domain can also bind collagens other than the fibrillar subtypes and to get quantitative information, a binding assay with collagen types I, IV, and VI was carried out. In the solid phase assay the range of ␣ 10 I domain concentrations spanned 1-500 nM. The ␣ 10 I domain was found to bind type IV collagen slightly better (K d Ϸ 300 Ϯ 100 nM) than collagen types I and VI (Fig. 3). The estimated K d for binding to fibrillar collagen type I was 350 Ϯ 150 nM, similar to the value for collagen type VI binding (K d Ϸ 350 Ϯ 200 nM). In terms of binding properties, these results place the ␣ 10 I domain closer to the ␣ 1 I domain, which prefers collagen type IV over collagen type I or VI. Moreover, the fact that the ␣ 10 I domain binds well to collagen type VI, in contrast to the ␣ 2 I domain, suggests that the binding preferences of the ␣ 1 I and ␣ 10 I domains are similar.
The Sequence Surrounding the MIDAS Site Differs in the ␣ 1 I, ␣ 2 I, and ␣ 10 I Domains-Sequence differences in the vicinity of the binding site were identified by using the sequence alignment of the ␣ 1 I, ␣ 2 I, and ␣ 10 I domains, and the x-ray structures of the apo form of the ␣ 1 I domain (25) and both the apo and holo forms (1aox, Ref. 26; 1dzi, Ref. 17) of the ␣ 2 I domain. To make it easier to visualize the differences among the surface regions surrounding the MIDAS site, both the ␣ 1 I (holo form) and ␣ 10 I (apo and holo forms) domain structures were modeled using MODELLER 4.0. The electrostatic potentials were mapped to the solvent accessible surfaces of the apo and holo forms (Fig. 4) for each of the ␣I domains, and the sequence differences are listed in Table I. The ␣ 1 I, ␣ 2 I, and ␣ 10 I domains have among them 11 amino acid positions (Table I) tively and 2 negatively charged residues); but ␣ 10 I has, in addition to glycine, a total of 10 polar residues, and 7 of them are charged residues: 4 positively and 3 negatively.
Differences between ␣ 10 I and ␣ 2 I-Asp-219 in ␣ 2 I may participate directly in ligand binding (17). The corresponding amino acid in ␣ 10 I is Arg-218. Of the 11 positions that exhibit chemical and size changes among the three integrins, ␣ 2 I and ␣ 10 I have several residues in common: Gly-259 (␣ 10 I) and Gly-260 (␣ 2 I); Asp-291 and Asp-292; Arg-297 and Lys-298. Of the 4 positively charged residues in ␣ 10 I, 2 (Arg-288 (Asn-289 in ␣ 2 I) and Arg-290 (Leu-291 in ␣ 2 I)) are not found in ␣ 2 I but are located in ␣ 10 I within a region that undergoes a conformational change on collagen binding in ␣ 2 I (residues 283-293, ␣ 2 numbering). The model of the ␣ 10 I holo form places these residues away from the MIDAS site. The structure of ␣ 2 I with a bound collagen fragment, a GFOGER hexapeptide complex (17), shows that this region is not directly involved in collagen binding, suggesting that if binding occurs in a similar manner in ␣ 10 I then these 2 residues may not be important for collagen binding in ␣ 10 I either.
Differences between ␣ 10 I and ␣ 1 I-Arginine is found at position 218 in both of these ␣I domains. Thus, the differences in their collagen binding patterns must be the result of interactions with other residues. Sequence differences in the vicinity  (17); Ala-290 (␣ 2 ) is located far from the collagen binding area in both the unbound and bound ␣ 2 I crystal structures.
Mutation Analysis Indicates a Critical Role for Asp-219 in ␣ 2 I and Arg-218 in ␣ 1 I/␣ 10 I in the Discrimination between Different Collagen Subtypes-Based on sequence comparisons and molecular modeling, mutations were introduced into the ␣ 1 I, ␣ 2 I, and ␣ 10 I domains. To test the idea that Arg-218 in ␣ 1 I is important for the collagen binding specificity, this residue was swapped using site-directed mutagenesis to the corresponding amino acid of ␣ 2 I, an aspartate, thus reversing the charge. The ␣I domain was produced as a GST fusion protein, and binding assays for collagen types I, IV, and VI were performed ( Fig. 5A and Table II). The ligand binding profile of ␣ 1 I R218D resembles that of ␣ 2 I. The binding to collagen types IV and VI weakened considerably compared with type I collagen binding, which remained nearly the same (Fig. 5A and Table  II). These results indicate that Arg-218 in the ␣ 1 I domain is one of the key amino acids functioning to discriminate among collagen subtypes. The equal weakening in the binding to type IV and type VI collagen suggests that the binding mechanisms of these collagen subtypes may be similar. A corresponding mutation in the ␣ 10 I domain had very similar effects. The R218D ␣ 10 I domain bound to type I collagen in the same way as the   (Table II), whereas its binding to type IV collagen was considerably weaker (Table II).
To investigate the importance of the ␣ 1 I and ␣ 10 I Arg-218 equivalent amino acid Asp-219 in ␣ 2 I, mutant I domains were produced. Asp-219 was mutated to arginine, changing the negatively charged residue to a positively charged residue. Binding assays using collagen types I, IV, and VI were performed. The D219R mutation was found to retain the collagen type IV binding level better, whereas collagen type I binding weakened significantly ( Fig. 5B and Table II). The effect of mutation to collagen type VI binding is more difficult to evaluate because the binding of the ␣ 2 I domain wild type is also so weak (Table  II). These data stress the importance of Asp-219 in the ␣ 2 I domain for collagen type I recognition, although it is obvious that one single amino acid cannot explain all of the differences seen in collagen binding among the ␣I domains.

DISCUSSION
Integrins are found in all metazoans from marine sponges to man (36). However, the I domains containing collagen-binding integrins are not present in the published genomes of Drosophila melanogaster or Caenorhabditis elegans, suggesting that they may have evolved significantly later (37). The four collagen receptors are structurally very similar, and their tissue distribution is partially overlapping. Especially in bone-and cartilage-derived cells, all four receptors may be present at the same time. 3 Because the short intracellular domains of the collagen-binding ␣ subunits are very different, it has been assumed that the four receptors have distinct signaling functions after binding ligand. Opposite effects on cells have been observed for the ␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrins. Cell adhesion to collagen, when mediated by ␣ 1 ␤ 1 integrin, promotes cell growth and suppresses collagen synthesis (3, 38 -40). In contrast, ␣ 2 ␤ 1 integrin inhibits the cell cycle and promotes matrix remodeling by increasing both collagen synthesis and degradation (3,38,39,41,42). It is not known whether ␣ 10 ␤ 1 or ␣ 11 ␤ 1 binding to collagen leads to alterations in cell behavior.
Another reason for the simultaneous existence of multiple collagen receptors may be the fact that the collagens, despite their common basic structure, have significant structural and functional differences. The collagen receptor integrins recognize their ligands by means of a special ␣I domain. Integrin ␣ 1 I and ␣ 2 I domains have been reported to bind to various collagen subtypes (20), but the ligand pattern of ␣ 10 I has been unknown. We have produced a human ␣ 10 I domain as a recombinant protein and tested it together with ␣ 1 I and ␣ 2 I domains. In general, ␣ 10 I binds to collagen types I-VI with a K d that is within the range typical of ␣ 1 I and ␣ 2 I domains. Collagen binding required the presence of magnesium. Integrin ␣ 10 I could also recognize laminin-1, but its ability to bind to tenascin was weaker. A more detailed analysis with different collagen subtypes revealed interesting similarities and differences among the three ␣I domains investigated here. The ␣ 1 I and ␣ 10 I domains resemble each other in the sense that they both bind to non-fibrillar collagen types IV and VI, whereas ␣ 2 I prefers fibrillar collagens. The difference between ␣ 1 I and ␣ 2 I domains in the recognition of type I and IV collagen had been noted previously (20). A recent study suggested that in chondrocytes, ␣ 1 ␤ 1 is the predominant type VI collagen receptor (35). In support of this observation, our results demonstrate 3  that ␣ 1 I binds type VI collagen significantly better than does the ␣ 2 I domain. Type VI collagen is expressed in various tissues, including cartilage. The fact that the ␣ 10 I domain also binds type VI collagen suggests that chondrocytes may have several receptors mediating interactions with the collagen in beaded filaments. A crystal structure of a collagen-like triple helix in complex with ␣ 2 I identified amino acids in ␣ 2 I which can make a direct contact with the ligand (17). Among others, Asp-219 was found to form a salt bridge with an arginine residue in the collagenous peptide. Accordingly, earlier modeling and mutational studies had suggested that Asp-219 is important for ␣ 2 I binding to type I collagen (18) and to peptides derived from snake venom (19). No crystal structure of the ␣ 1 I domain in contact with any ligand is yet available. However, it is obvious that there must be differences in the ligand binding mechanism because in the ␣ 1 I domain the amino acid corresponding to Asp-219 (␣ 2 I) is Arg-218 and therefore has an opposite charge. Here, the mutation D219R in the ␣ 2 I domain dramatically decreased the ratio of type I collagen binding to type IV and to type VI collagen binding. Correspondingly, the mutation R218D in the ␣ 1 I and ␣ 10 I domains increased this ratio. Thus, we suggest that Asp-219 (␣ 2 I) and Arg-218 (␣ 1 I/␣ 10 I) are critical amino acids in the determination of binding preferences. It is obvious that in addition to this single amino acid position other differences in the structure of the three ␣I domains must contribute to their different functions. We have no direct evidence (complex structures) that Arg-218 in the ␣ 1 I domain or the ␣ 10 I domain interacts directly with their ligands. The fact that both types I and IV collagen have multiple integrin binding sites (43,44) suggests that a single ␣I domain can bind to different collagens in different ways.
In conclusion, we have shown that the ␣ 10 I domain can recognize other collagen subtypes, in addition to type II collagen (9), including fibrillar and non-fibrillar collagens. ␣ 10 I may also function as a receptor for laminin-1. We have identified an essential determinant of ligand binding specificity of the integrin ␣I domains. The amino acid Arg-218 in the ␣ 1 I and the ␣ 10 I domain seems to promote binding to non-fibrillar collagen types IV and VI, whereas an opposite charge at the same position shifts the preference of the ␣I domain toward fibrillar collagens, partially explaining the function of the ␣ 2 I domain.