α11β1 Integrin Recognizes the GFOGER Sequence in Interstitial Collagens*

The integrins α1β1, α2β1, α10β1, and α11β1 are referred to as a collagen receptor subgroup of the integrin family. Recently, both α1β1 and α2β1integrins have been shown to recognize triple-helical GFOGER (where single letter amino acid nomenclature is used, O = hydroxyproline) or GFOGER-like motifs found in collagens, despite their distinct binding specificity for various collagen subtypes. In the present study we have investigated the mechanism whereby the latest member in the integrin family, α11β1, recognizes collagens using C2C12 cells transfected with α11 cDNA and the bacterially expressed recombinant α11 I domain. The ligand binding properties of α11β1 were compared with those of α2β1. Mg2+-dependent α11β1 binding to type I collagen required micromolar Ca2+ but was inhibited by 1 mmCa2+, whereas α2β1-mediated binding was refractory to millimolar concentrations of Ca2+. The bacterially expressed recombinant α11 I domain preference for fibrillar collagens over collagens IV and VI was the same as the α2 I domain. Despite the difference in Ca2+ sensitivity, α11β1-expressing cells and the α11 I domain bound to helical GFOGER sequences in a manner similar to α2β1-expressing cells and the α2 I domain. Modeling of the α I domain-collagen peptide complexes could partially explain the observed preference of different I domains for certain GFOGER sequence variations. In summary, our data indicate that the GFOGER sequence in fibrillar collagens is a common recognition motif used by α1β1, α2β1, and also α11β1 integrins. Although α10and α11 chains show the highest sequence identity, α2 and α11 are more similar with regard to collagen specificity. Future studies will reveal whether α2β1 and α11β1integrins also show overlapping biological functions.

Studies of collagen-binding integrins in various in vitro assays show that they take part in cell adhesion, cell migration, control of collagen synthesis, matrix metalloproteinase synthesis, remodeling of collagen matrices, and influence such complex processes as cell proliferation, cell differentiation, angiogenesis, platelet adhesion/aggregation, and epithelial tubulogenesis (10,11).
Using transfected cells and recombinant I domain, ␣ 1 ␤ 1 has been shown to bind collagens, with a preference for collagens IV and VI over collagen I and II. Collagen XIII in vitro is also a ligand for ␣ 1 ␤ 1 (12). Other identified ligands include laminin-1/-2 (13,14), the cartilage protein matrilin-1 (15), and the C-propeptide of collagen I (16). The affinity of ␣ 1 ␤ 1 for laminin-1 has been reported to be about 10-fold lower than for collagen IV (17). In accordance with this, when the ␣ 1 integrin chain is expressed in K562 cells, ␣ 1 ␤ 1 will bind type IV collagen, but it requires activation to bind laminin-1 (18). In Chinese hamster ovary cells, ␣ 1 ␤ 1 integrin does not mediate spreading on collagen II (12). It has been suggested that in these cells a coreceptor is needed for ␣ 1 ␤ 1 -mediated spreading on collagen II.
The integrin subunit ␣ 10 was originally identified by affinity purification of collagen type II-binding integrins from adult chondrocytes (3). The rather restricted expression of ␣ 10 ␤ 1 to cartilage indicates that the ligands are to be found in the cartilage extracellular matrix. Intriguingly, using recombinant protein, the collagen binding preference of the ␣ 10 I domain is most similar to that of the ␣ 1 I domain, so that the ␣ 10 I domain prefers the basement membrane collagen IV and the beaded filament-forming collagen VI over the interstitial collagens I and II (27). In the same study, mutational analysis of the I domains showed that the amino acid residues Arg-218 in ␣ 1 and ␣ 10 and Asp-219 in ␣ 2 are involved in determining this collagen preference.
␣ 11 was initially detected in differentiating human fetal muscle cells (28). ␣ 11 protein and mRNA expression analysis in human embryos, however, revealed that expression is localized to mesenchymal non-muscle cells in areas of highly organized interstitial collagen networks. No expression was seen in muscle cells in vivo (29). In the developing skeletal system, ␣ 10 ␤ 1 and ␣ 11 ␤ 1 thus show nonoverlapping, complementary expression patterns (11). In accordance with the expression of ␣ 11 ␤ 1 in areas rich in interstitial collagens, ␣ 11 ␤ 1 binds more efficiently to collagen I than to collagen IV (29).
Cyanogen bromide cleavage of collagen chains identified the non-RGD-containing helical CB3 fragment of collagen ␣ 1 I as a cell-binding fragment that could be used to purify ␣ 1 ␤ 1 (30). The ␣ 1 I and ␣ 2 I integrin binding site located within triplehelical ␣ 1 I CB3 has been identified as GFOGER (31,32). Two related sequences, GLOGER 2 and GASGER, were identified elsewhere in collagen I (33), and other GER-containing sequences in the collagen chains can also mediate cell adhesion through ␣ 2 ␤ 1 . 3 The GER motif thus appears to be a major cell adhesion motif used by collagen-binding integrins.
Examination of the crystal structure of an ␣ 2 I domain-GFOGER complex suggested that other hydrophobic residues might replace phenylalanine, which together with the glutamate and arginine residues provided the main side chain interactions between the collagen-like peptide and the integrin (34). Interactions also occurred with the main chain carbonyl group of the hydroxyproline residue, suggesting that hydroxyproline itself may not be required specifically for collagenintegrin interaction.
Arginine interacts with negatively charged Asp-219 on the surface of the ␣ 2 I domain, and although this appears a relatively nonspecific interaction, GEK will not substitute fully in human platelets. The ligand binding groove of the ␣ 2 I domain is relatively deep compared with that of ␣ 1 . Thus, the coordination of Mg 2ϩ in the metal ion dependent adhesion site (MI-DAS) may only be achieved by glutamate residues from the GER motif, aspartate being too short for this purpose. This may not be the case for other integrins, and the abundance of GDR triplets within the collagens suggests the possibility that GDR motifs might serve as well. The present study was designed in part to test the possibility that ␣ 11 , which lacks an acidic residue equivalent to Asp-219, and whose structure is not yet defined, might recognize GEK and GDR triplets, both of which occur frequently in human collagens.
Other modes of collagen binding also exist, and ␣ 1 ␤ 1 binds to collagen type IV using the amino acids arginine and aspartate contributed from different collagen ␣ chains (35,36). The residues recognized by ␣ 1 ␤ 1 in collagen XIII, lacking a GFOGER sequence, have not been identified. Studies with fragments of laminins have shown that the I domain integrin binding sites are present in the short arm of the ␣ chain, but the exact region(s) has not yet been mapped (13,14).
The interaction between integrins and their physiological ligands displays several requirements for divalent cations. First, the GER glutamate residue binds directly to a Mg 2ϩ ion coordinated within the I domain MIDAS. Other divalent cat-ions will serve this purpose, notably Mn 2ϩ and Co 2ϩ , but in nature such ions are unlikely to be sufficiently abundant to contribute significantly to the adhesion process. Integrins possess several other divalent cation binding sites, three or four within the blades of the ␣ subunit ␤ propeller, and perhaps two within the ␤ subunit I-like domain. Some of these sites likely bind Ca 2ϩ and account for the biphasic role of Ca 2ϩ in the competence of the integrins. Adhesion of collagen to ␣ 2 ␤ 1 in human platelets, in common with the binding of ligand by other integrins, has recently been shown to require micromolar Ca 2ϩ (37) and to be inhibited by millimolar Ca 2ϩ . Conceivably the latter effect reflects competition for Ser-123 lying between Mg 2ϩ in the MIDAS and Ca 2ϩ in the ADMIDAS sites of the ␤ subunit I-like domain, such that high levels of Ca 2ϩ render the integrin ␤ subunit incapable of regulating ␣ subunit function properly. The requirement for Ca 2ϩ appears to differ even for the same integrin when expressed in different cells. Thus, although a biphasic effect of Ca 2ϩ on ␣ 2 ␤ 1 competence in platelets can clearly be shown, the sensitivity of ␣ 2 ␤ 1 to either high or low levels of Ca 2ϩ in HT1080 cells is much less obvious. 4 In the present study we set out to study further the mechanism whereby ␣ 2 ␤ 1 and ␣ 11 ␤ 1 integrins bind collagens. Our data suggest that the approximated K d of the ␣ 11 I domain for type I collagen is higher than that of the ␣ 2 I domain and that in C2C12 cells the ␣ 11 ␤ 1 -mediated binding to collagen I, but not that of ␣ 2 ␤ 1 , is sensitive to the presence of Ca 2ϩ . However, both integrins display a similar collagen specificity, and both recognize the helical GFOGER sequence. Modeling of ␣ 2 I and ␣ 11 I domain-ligand complexes could in part explain the observed differences in ␣ 2 I and ␣ 11 I domain binding to different collagen peptides. The results are potentially promising for future attempts to generate reagents effective in blocking multiple collagen-binding integrins simultaneously.

Production of Human Recombinant
Integrin ␣ 1 I, ␣ 2 I, and ␣ 11 I Domains as Fusion Proteins cDNAs encoding ␣ 1 I and ␣ 2 I domains were generated by PCR as described earlier (27) using human integrin ␣ 1 and ␣ 2 cDNAs as templates. Vectors pGEX-4T-3 and pGEX-2T (Amersham Biosciences) were used to generate recombinant glutathione S-transferase (GST) fusion proteins of human ␣ 1 I and ␣ 2 I domains, respectively. Human integrin ␣ 11 cDNA (4) was used as a template when the ␣ 11 I domain was generated by PCR. The PCR product having BamHI and EcoRI sites was cloned to pGEX-KT, and the DNA sequence was checked by sequencing the whole insert. The same vector was used for expression of recombinant GST fusion proteins of the human ␣ 11 I domain. Competent Escherichia coli BL21 cells were transformed with the plasmids for protein production. 500 ml of LB medium (Biokar) containing 100 g/ml ampicillin was innoculated with a 50-ml overnight culture of BL21/p␣ 1 I, BL21/p␣ 2 I, or BL21/p␣ 11 I, and the cultures were grown at 37°C until the A 600 of the suspension reached 0.6 -1.0. Cells were induced with isopropyl-1-thio-␤-D-galactopyranoside and allowed to grow for an additional 4 -6 h before harvesting by centrifugation. Pelleted cells were resuspended in PBS (pH 7.4) and then lysed by sonication followed by the addition of Triton X-100 to a final concentration of 2%. After incubation for 30 min on ice, suspensions were centrifuged, and supernatants were pooled. Glutathione-Sepharose (Amersham Biosciences) was added to the lysate, which was incubated at room temperature for 30 min with gentle agitation. The lysate was then centrifuged, the supernatant was removed, and glutathione-Sepharose with bound fusion protein was transferred into disposable chromatography columns (Bio-Rad). The columns were washed with PBS, and fusion proteins were eluted using 30 mm glutathione. Purified recombinant and glutathione-tagged ␣ 1 I, ␣ 2 I, and ␣ 11 I domains were analyzed by SDS and native PAGE. The recombinant ␣ 1 I domain produced was 227 amino acids in length, corresponding to amino acids 123-338 of the whole ␣ 1 integrin, whereas the ␣ 2 I domain was 223 amino acids long, which corresponded to amino acids 124 -339 of the whole ␣ 2 integrin. The carboxyl termini of the ␣ 1 I and ␣ 2 I domains contained 10 and 6 non-integrin amino acids, respectively. Recombinant ␣ 11 I domain contains a total of 204 amino acids: at the amino terminus there are 2 extra residues (GS) before the ␣ 11 I domain, which starts from CQTY and ends with SLEG (residues 159 -354); at the carboxyl terminus there are 6 extra amino acids (EFIVTD). The recombinant ␣ 11 I domain contains some GST as an impurity caused by endogenous protease activity during expression and purification. Recombinant I domains were used as GST fusion proteins for collagen binding experiments.

Synthesis of Peptides
Peptides were synthesized as carboxyl-terminal amides on TentaGel R RAM resin in a PerSeptive Biosystems 9050 Plus PepSynthesizer exactly as described in our earlier studies (30,31). Peptides were purified by reverse phase high performance liquid chromatography (HPLC) on a column of Vydac 219TP101522 using a linear gradient of 5-45% acetonitrile in water containing 0.1% trifluoroacetic acid. Fractions containing homogeneous product were identified by analytical HPLC on a column of Vydac 219TP54, pooled, and freeze dried. All peptides were found to be of the correct theoretical mass by mass spectrometry. The triple-helical stability of each peptide was assessed by polarimetry as described previously.

Solid Phase Binding Assay for ␣ 1 I, ␣ 2 I, and ␣ 11 I Domains
The coating of a 96-well high binding microtiter plate (Nunc) was done by exposure to 0.1 ml of PBS containing 5 g/cm 2 (15 g/ml) collagens or 20 g/ml synthetic triple-helical collagen peptides overnight at 4°C. Type I rat (rat tail) collagen, type IV mouse (basement membrane of Engelbreth-Holm-Swarm mouse sarcoma) collagen, and type IV human "cut" (human placenta) collagen were purchased from Sigma. Type IV human collagen and type II bovine collagen were purchased from Biodesign International and Chemicon, respectively. Type I bovine (bovine dermal) collagen was from Cellon S. A. Blank wells were coated with a 1:1 solution of 0.1 ml Delfia Diluent II (Wallac) and PBS. Residual protein absorption sites on all wells were blocked with a 1:1 solution of 0.1 ml of Delfia Diluent II and PBS. Recombinant proteins ␣ 1 I-GST, ␣ 2 I-GST, and ␣ 11 I-GST were added to the coated wells at the desired concentration in Delfia assay buffer and incubated for 1 h at room temperature. Europium-labeled anti-GST antibody (Wallac) was then added (typically 1:1,000), and the mixtures were incubated for 1 h at room temperature. All incubations mentioned above were done in the presence of 2 mM MgCl 2 . Delfia enhancement solution (Wallac) was added to each well, and the europium signal was measured by time-resolved fluorometry (Victor2 multilabel counter, Wallac). In every case, at least three parallel wells were analyzed.

Cells
Murine C2C12 myoblast cells from the American Type Culture Collection were provided by A. Starzinski-Powitz. The generation of C2C12 cells stably transfected with integrin ␣ 2 cDNA or integrin ␣ 11 cDNA has been described previously (29). Cells were cultured at 37°C in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and antibiotics (Statens veterinä rmedicinska anstalt, Uppsala). The cells were grown to subconfluence and passaged every 2-3 days.

Antibodies
Rabbit antibodies to the cytoplasmic tail of ␣ 11 integrin have been described previously (4). To immunoprecipitate ␤ 1 integrins, a polyclonal antibody to rat integrin ␤ 1 chain was used (38).

Immunoprecipitation and Electrophoresis
Cell cultures were washed three times in Dulbecco's modified Eagle's medium devoid of cysteine and methionine and metabolically labeled overnight in the presence of 25 Ci/ml [ 35 S]methionine/cysteine (pro-Mix 35 S cell labeling mix; Amersham Biosciences). Proteins were extracted from the tissue culture dishes by the addition of 1 ml of solubilization buffer (1% Triton X-100, 0.15 M NaCl, 20 mM Tris-HCl pH 7.4, 1 mM MgCl 2 , 1 mM CaCl 2 ) containing protease inhibitors (1 mM Pefabloc SC (Roche Molecular Diagnostics), 1% aprotinin, 1 g/ml pepstatin, 1 g/ml leupeptin). Solubilized proteins were centrifuged for 10 min at 15,000 ϫ g. The centrifuged supernatant was precleared by incubating with 100 g/ml preimmune IgG and protein A-Sepharose CL4B (Amersham Biosciences) for 2 h. After centrifugation, immune IgG was incubated with the extract for 2 h. Specifically bound proteins were recovered with protein A-Sepharose. The precipitate was washed three times with buffer A (1% Triton X-100, 0.5 M NaCl, 20 mM Tris-HCl pH 7.4, 1 mM MgCl 2 , 1 mM CaCl 2 ) and three times with buffer B (0.1% Triton X-100, 0.15 M NaCl, 20 mM Tris-HCl pH 7.4, 1 mM MgCl 2 , 1 mM CaCl 2 ) prior to solubilization in electrophoresis sample buffer. Proteins were separated on 6% SDS-polyacrylamide gels and processed for fluorography.

Cell Attachment Assay
General Setup-Ligands and metal ions are described separately for the two experiments. 24-well cell culture plates (Nunc) were coated with ligands (500 l to a 2-cm 2 well) diluted in PBS overnight at 4°C, followed by blocking with 2% BSA in PBS for 2 h at room temperature and then washed in Puck's saline (137 mM NaCl, 5 mM KCl, 4 mM Na 2 CO 3 , 5.5 mM D-glucose, pH 7.0). Transfected cells were trypsinized, washed four times in Puck's saline, deeded into the wells at a concentration of 250,000 cells/well, and were allowed to attach for 45 min at 37°C and 5% CO 2 . Wells were washed three times in Puck's saline, and plates were rapidly frozen at Ϫ20°C for later assay using the hexoseaminidase test as described previously (29). For each cell line used, a cell number standard was made. Each experiment was performed in triplicate. To minimize errors from unequal trypsinization stress between cell lines and handling of plates, for example, data were normalized as follows. For each plate the adhesion to 10 g/ml fibronectin (provided by S. Johansson, Uppsala University) was used as the 100% reference level, and the background found on BSA-only coated wells was used as the base-line (0%) reference level.
Ca 2ϩ Inhibition Setup-Wells were coated with 10 g/ml bovine collagen type I (Vitrogen100, Cohesion) or 10 g/ml human plasma fibronectin. Wells were filled with Puck's saline, and MgCl 2 ϩ EGTA was added to obtain a final concentration (after addition of cells) of 2 mM MgCl 2 and 0.01 mM EGTA. CaCl 2 was added according to Fig. 1.
Cell Attachment to Synthetic Peptides-24-well plates were coated with 10 g/ml synthetic triple-helical collagen peptides at 4°C overnight according to Fig. 4, or 10 g/ml bovine collagen type I (Vitro-gen100), or 10 g/ml fibronectin. MgCl 2 and CaCl 2 were added to a final concentration of 2 mM MgCl 2 and 0.01 mM CaCl 2 .
The sequence alignment was made using the program MALIGN (43) in the BODIL modeling environment 5 (www.abo.fi/fak/mnf/bkf/research/johnson/bodil.html) using a structure-based sequence comparison matrix (44) with a gap penalty of 40.
Homology models were built with HOMODGE in BODIL. The amino acid side chain rotamer library incorporated into BODIL was used to evaluate alternative possibilities for side chain conformations for sequence differences in the alignment of ␣ 1 and ␣ 11 I domain sequences with the template structure.
In the ␣ 2 I domain-peptide complex structure, 1dzi, and the three peptide chains (identical in sequence but having different interactions with the ␣ 2 I domain) of the collagen mimetic tripeptide are labeled B, C, and D. The corresponding chain labels are used for the tripeptides docked to the model structures built for the ␣ 1 I and ␣ 11 I domains.

RESULTS
Influence of Ca 2ϩ on Cell Attachment to Collagen-To compare the mechanism whereby ␣ 11 ␤ 1 and ␣ 2 ␤ 1 recognize collagens, we used the satellite cell line C2C12 transfected with ␣ 2 or ␣ 11 cDNAs (C2C12 ␣ 2 ϩ and C2C12 ␣ 11 ϩ , respectively). The parental cell line C2C12 expresses members of the ␤ 1 subfamily, such as ␣ 5 ␤ 1 and ␣ 7 ␤ 1 (45), but does not adhere to collagen (29). C2C12 cells transfected with either ␣ 2 or ␣ 11 acquire the ability to interact with collagens I and IV, with a preference for collagen I (29). The ␣ 2 ␤ 1 -mediated binding of platelets to collagen has been reported to require micromolar Ca 2ϩ but to be inhibited by millimolar Ca 2ϩ in the presence of Mg 2ϩ (37). To test the effect of Ca 2ϩ on cell adhesion to collagen I, the transfected C2C12 cells were plated on collagen I in the presence of Mg 2ϩ and EGTA with increasing concentrations of Ca 2ϩ added.
In the absence of Ca 2ϩ , adhesion of cells expressing ␣ 11 ␤ 1 was virtually absent, as reported for human platelets (45), and a biphasic response to added Ca 2ϩ was observed with peak adhesion occurring at a free Ca 2ϩ of around 30 M and adhesion being substantially abolished at 4 mM, with an IC 50 of about 1 mM. In marked contrast, C2C12 ␣ 2 ϩ cell adhesion to collagen I was largely refractory to both the removal of Ca 2ϩ using EGTA or the subsequent addition of millimolar Ca 2ϩ (Fig. 1).
␣ 11 and ␣ 2 I Domains Differ in Their Affinity for Collagen I-To estimate the K d of ␣ 11 for collagen I, we produced an ␣ 11 I domain in E. coli. Initial attempts to express the ␣ 11 I domain as a bacterial GST fusion protein yielded low amounts of protein. We therefore expressed the ␣ 11 I domain as a His-tagged fusion protein in Pichia pastoris. After purification on nickel-Sepharose only low binding to collagen I was noted, and background binding was high. Gel filtration revealed that a majority of protein appeared in a high molecular weight fraction, indicating aggregation (data not shown). Production of ␣ 11 I-GST in E. coli was subsequently optimized. Large scale expression of the ␣ 11 I domain yielded enough protein to perform binding studies. Approximated K d values based on solid phase binding assays (26) and the use of the Michaelis-Menten equation suggested a relatively low ␣ 11 I domain avidity to collagen I (750 Ϯ 50 nM) when compared with the ␣ 2 I domain binding to collagen I (20 Ϯ 5 nM (27)) (Fig. 2).
Collagen Preference of ␣ 11 I Domain-Previous studies from several laboratories have shown that ␣ 2 ␤ 1 integrin prefers fibril-forming collagens over network-forming type IV and beaded filament-forming type VI collagen. The same pattern can be seen in the binding of the ␣ 2 I domain. Here ␣ 2 -and ␣ 11 -mediated binding to different collagens was compared. The ␣ 11 I domain was shown to prefer the fibril-forming collagen types I and II (Fig. 3), whereas its binding was weaker to type III (data not shown), a member of the same collagen subgroup. Collagens IV and VI were poor ligands for the ␣ 11 I domain. Thus, in the terms of its binding pattern the ␣ 11 I domain was closer to the ␣ 2 I domain than to either the ␣ 1 or ␣ 10 I domain.
Binding to GER-containing Peptides-Helical GFOGER and GFOGER-like peptides have recently been shown to represent high affinity integrin recognition motifs in collagens (32,33). To determine whether ␣ 11 ␤ 1 also differed from ␣ 2 ␤ 1 with regard to its recognition sequences, C2C12 ␣ 2 ϩ and C2C12 ␣ 11 ϩ cells were tested for their ability to attach to different collagenlike peptides. C2C12 ␣ 2 ϩ cells adhered to GFOGER and GFO-GEK peptides as reported previously. C2C12 ␣ 11 ϩ cells also bound these peptides in a similar pattern, whereas untransfected cells failed to do so (Fig. 4).To confirm that the observed binding occurred via the ␣ I domain, I domains from ␣ 1 , ␣ 2 , and ␣ 11 were compared with regard to binding to the different collagen-derived peptides (Fig. 5). In these studies we also included the GLOGER peptide (33) which bound all I domains. The ␣ 11 I domain preferred the GFOGER peptide followed by the GFOGEK and GLOGER peptides, much like the ␣ 2 I domain. The ␣ 1 I domain bound the GLOGER peptide as efficiently as the GFOGER peptide. The peptide containing the sequence GASGER, reported as a weak binding site for ␣ 2 ␤ 1 and ␣ 1 ␤ 1 showed only low capacity to bind any of the I domains, and substitution of aspartate for glutamate within the GER triplet similarly abolished I domain binding. The sequence GPOGES, from the collagen I ␣ 2 chain where it corresponds to GFOGER in the ␣ 1 I chain, was similarly without significant binding activity. Comparing the overall peptide binding pattern, ␣ 2 I and ␣ 11 I domains appear most similar in their peptide binding preferences.
Modeling of Collagen Peptide Binding to the ␣ 11 I Domain-The basic assumption in modeling was that the collagen mimetic tripeptide GFOGER and its mutants bind to all of the integrin ␣ I domains in a way similar to that seen in the crystal structure of the complex between ␣ 2 I domain and the GFOGER triple-helical peptide (34). The sequence identities of the ␣ 1 and ␣ 11 I domains to the ␣ 2 I domain are 51 and 45% respectively, thus experience dictates that high quality models will be produced. Only one region of the ␣ 11 I domain model is uncertain, where Pro-310 (threonine in the ␣ 1 and ␣ 2 I domains) is located within a region that corresponds to helix 6 of the open fold of the ␣ 1 and ␣ 2 I domains. Proline generally does not promote helix stability, so it is very likely that the helix begins at or after position 310 in the ␣ 11 I domain. In addition, the local alignment of residues 179 and 180 seems peculiar because the charged residue Glu-180 would be buried, and the hydrophobic residue Val-179 would be exposed toward the solvent. If Glu-180 is buried, then the conserved residue Tyr-157 may change its conformation in the ␣ 11 I domain and affect the binding of the collagen mimetic tripeptide. Thus, it is possible that the binding conformation seen in ␣ 2 I domain-tripeptide complex structure may be different in the case of ␣ 11 .
Glutamate in the Collagen Mimetic Tripeptide-In the crystal structure of the integrin ␣ 2 I domain in complex with the collagen mimetic peptide (3 ϫ GFOGER), the side chain of only one of the glutamate residues in the collagen mimetic tripeptide, that of the middle strand, chain C, interacts with the I domain. This glutamate is coordinated to the metal ion of the MIDAS motif and thus, represents a key interaction in tripeptide binding (Fig. 6). Moreover, even a conservative change, mutation to aspartate, lowers the binding dramatically for the ␣ 1 I, ␣ 2 I, and ␣ 11 I domains (GFOGDR; Fig. 4). Aspartate is one methylene group shorter than glutamate and would not reach the metal ion of the MIDAS motif as easily as glutamate can.
In the ␣ 2 I domain there is a leucine residue at position 286, which forms an interaction with phenylalanine of the collagen mimetic tripeptide from chain B, whereas both the ␣ 1 and ␣ 11 I domains have tyrosine at that position. Tyrosine could either form a hydrophobic interaction with a planar face of the Arg-148 side chain, or tyrosine could form a hydrogen bond with glutamate from the collagen mimetic tripeptide chain B. If the latter case is true then the hydrogen bond would not be possible in the collagen mimetic tripeptide containing the Glu 3 Asp mutation.
In the ␣ 1 I domain model, Arg-218 (aspartate in ␣ 2 and threonine in ␣ 11 ) may form an additional interaction with the glutamate of chain D of the collagen mimetic tripeptide. This interaction would still be possible in the tripeptide containing the aspartate mutation.
Arginine in the Collagen Mimetic Tripeptide-In the crystal structure of the ␣ 2 I domain (1dzi), arginine in chain C of the collagen mimetic tripeptide is bound to the area where Asp-219, Asn-189, and Leu-220 are located. The N⑀ and N2 atoms of arginine would interact with side chain carboxylate of Asp-219 but only weakly because the angle is not optimal for forming a strong hydrogen-bond/salt bridge. In addition, the side chain of Leu-220 forms an optimal site for hydrophobic interactions with a planar face of the arginine side chain (Fig. 6A). In both ␣ 1 and ␣ 11 , the arginine of the tripeptide could form a salt bridge with glutamate at the position equivalent to Asn-189 in ␣ 2 (Fig. 6, B and C). Furthermore, in ␣ 11 there is a threonine equivalent to Asp-219 in ␣ 2 whose side chain hydroxyl group can accept a hydrogen bond from the N⑀ of arginine from the tripeptide (Fig. 6C). For ␣ 2 , the Arg 3 Lys mutation in the collagen mimetic tripeptide (GFOGEK) does not affect binding as dramatically as seen for ␣ 1 and ␣ 11 (Fig.  5). In ␣ 2 , the repulsion resulting from the charged amino group of lysine positioned near the Leu-220 side chain would be offset by the formation of a somewhat more optimal hydrogen bond/ salt bridge between lysine and Asp-219 (Fig. 6D). The mutation of arginine to lysine reduces the binding affinity of collagen mimetic tripeptide to the ␣ 1 and ␣ 11 I domains because the salt bridge to glutamate, at the position equivalent to Asn-189 in ␣ 2 , cannot be maintained. When arginine of the collagen mimetic tripeptide is mutated to lysine, lysine cannot reach glutamate because a lysine residue is shorter than an arginine residue. Moreover, in ␣ 11 the lysine residue can form a hydrogen bond with the threonine equivalent to Asp-219 in ␣ 2 , and thus, the effect of the mutation is not as dramatic as for ␣ 1 .
In the ␣ 2 I domain, the arginine from chain B of the collagen mimetic tripeptide interacts mainly with other parts of the tripeptide and not with the I domain. N⑀ is hydrogen-bonded to the main chain oxygen of arginine in chain C, and the planar end of the arginine side chain has a hydrophobic interaction with proline in peptide chain C. In addition, hydrophobic interactions with the hydrophobic part of the Glu-256 side chain and weak electrostatic interactions with the main chain oxygen atom of Ser-257 can be seen. These interactions should be present and identical in each of the I domains in this study. Thus, the effect of the mutation of arginine to lysine, caused by chain C, should be same for all I domains.
In the ␣ 2 structure, the arginine from chain D of the collagen mimetic tripeptide is exposed to the solvent, and thus, the mutation can only have an indirect influence on I domain binding.
Phenylalanine in the Collagen Mimetic Tripeptide-In the ␣ 2 I domain structure, the phenyl ring of phenylalanine in chain B of the collagen mimetic tripeptide is stacked with the phenol ring of the conserved tyrosine (position 157 in ␣ 2 ). This phenylalanine also has hydrophobic interactions with Leu-286 in ␣ 2 . In addition, the phenylalanine of chain B forms an unfavorable interaction with the main chain oxygen atom of Tyr-285 (  Fig. 7B.

7A ). A corresponding view of phenylalanine interactions with the ␣ 1 I domain is shown in
When phenylalanine of the collagen mimetic tripeptide is mutated to leucine, some favorable interactions would be lost, but this loss is offset by the removal of unfavorable interactions with the main chain oxygen atom of the residue at position 285 (tyrosine in ␣ 2 and ␣ 11 ; serine in ␣ 1 ). Thus, there is a small change in the binding affinity of ␣ 1 and ␣ 2 when the Phe 3 Leu mutant is compared with the "wild type" tripeptide (Fig. 5). The effects seen for ␣ 11 are difficult to predict because the model is inaccurate in this region. The binding affinity is lowered dramatically when phenylalanine is replaced with alanine, resulting in the loss of all favorable interactions (Fig. 5).
In the ␣ 2 I domain structure, the phenylalanine in chain C of the collagen mimetic tripeptide leans against the side chain of Asn-154, which is conserved in the ␣ 1 , ␣ 2 , and ␣ 11 I domains. This interaction is not very critical, and thus the mutation of phenylalanine to leucine or alanine would not affect the binding affinity by much. The phenylalanine in chain D is exposed to the solvent, and thus it has practically no role in binding. DISCUSSION In recent years an increasing effort has been spent trying to understand the mechanism whereby cells bind collagen. In vertebrates more than 24 different collagens exist, and the role of some of these is yet unclear. Integrins are major receptors for collagens. A common feature of the collagen-binding integrins is the presence of an ␣ I domain that is directly involved in ligand binding.
The I domain is not found in integrin ␣ chains from the invertebrate Drosophila melanogaster but it is present in 9 of the 18 currently known vertebrate integrin ␣ chains (11) including ␣L, ␣M, ␣X, ␣D, and ␣E, which are all involved in different aspects of leukocyte functions and pair exclusively with the ␤ 2 subunit (10). The overall importance of integrinmediated cell-collagen interactions involving the ␣ 1 , ␣ 2 , ␣ 10 , and ␣ 11 integrin chains is largely unknown because of the limited information available for ␣ 10 ␤ 1 and ␣ 11 ␤ 1 . Based on the appearance of I domain integrin ␣ chains during vertebrate evolution it is possible that these integrin chains play important roles in vertebrate-specific structures of the musculoskeletal system.
Gene knockout experiments and recombinant expression of the ␣ I domains have yielded considerable information about the characteristics and functions of collagen-binding integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 (11,46,47). Phylogenetically, ␣ 1 and ␣ 2 form a subfamily distinct from ␣ 10 and ␣ 11 , which most likely have formed through two distinct gene duplication events.
Both ␣ 1 and ␣ 2 integrin chains are fairly widely expressed throughout the body. Gene knockout experiments of ␣ 1 and ␣ 2 chains have shown that inactivation of the individual collagenbinding integrins does not seem to impair embryonic development (46,47). Rather, mild phenotypes are observed where either fibroblast and leukocyte interactions with collagens or platelet interactions with collagens are affected. Recent analysis of ␣ 10 and ␣ 11 expression (3, 29) reveals a restricted embryonic expression pattern, which is not overlapping but complementary. In the near future it will be important to determine to what extent the collagen-binding integrins show overlapping functions and to what extent different collagenbinding integrins can functionally compensate for each other's absence. Crossing different mice strains lacking certain collagen receptors will shed light on these issues.
As a part of understanding the biological function of collagen-binding integrins it is important to characterize all of the different collagen-binding integrins with regard to collagen affinity, collagen specificity, divalent ion requirements, and ligand recognition motifs. Studies of ␣ 1 I, ␣ 2 I, and ␣ 10 I domains have shown that they bind collagens with different specificity (27). This specificity seems in part to be determined by residues located outside the MIDAS motif in the ␣ I domain. Data from several groups have convincingly shown that ␣ 1 prefers collagens IV and VI over collagen I and that the preferences of ␣ 2 are opposite. More recently the ␣ 10 I domain was shown to display a collagen binding specificity similar to that of ␣ 1 (27).
Prior to this study no binding studies had been performed with the ␣ 11 I domain. It thus appears that although ␣ 1 is, in terms of evolution, more similar to ␣ 2 , and ␣ 10 is more closely related to ␣ 11 , another grouping can be made based on their collagen specificity. The finding that ␣ 11 I prefers interstitial collagens over nonfibrillar collagens supports our previous cell binding data (29), but the difference is even more pronounced at the I domain level. A candidate amino acid that might play a role in determining this preference is Thr-238 found in a position corresponding to Arg-218 in ␣ 1 .
The relatively low avidity for collagens estimated for both ␣ 10 I and ␣ 11 I domains is intriguing. Our experience is that as recombinant GST fusion proteins, these I domains are less soluble than ␣ 1 I and ␣ 2 I domains, and they might have a tendency to form aggregates. This may affect the K d estimates. Furthermore, we have shown that in the length of the produced protein a difference of one amino acid residue might lead to changes in the avidity of collagen binding (48). Thus, the approximated K d values should be used for comparing the binding of a recombinant ␣ I domain with different collagens rather than for comparing the ␣ I domains with each other. Low avidity may indicate that the major function for ␣ 10 and ␣ 11 is not that of firm adhesion but that these integrins engage in dynamic interactions with collagen during events such as cell migration. It is also possible that the true ligands have not yet been identified. For example, for ␣ 10 , a cartilage ligand, possibly a collagen other than collagen II, might be the preferred ligand. In the case of ␣ 11 , a perichondrium ligand other than collagen I might bind this integrin with higher affinity.
The ␣ 2 ␤ 1 -mediated binding of platelets to collagen I is inhibited by mM concentrations of Ca 2ϩ (37). The finding that ␣ 2 ␤ 1 is not inhibited by Ca 2ϩ when expressed in C2C12 cells is intriguing. In the case of platelets and C2C12 cells this difference might be related to the activation status of the integrin. Whereas platelets and leukocytes have a more elaborate system for regulating integrin activation status, integrins in C2C12 cells are expected to be mainly in the activated state, displaying a higher affinity. ␣ 11 ␤ 1 is not expressed on platelets, so a direct comparison with ␣ 2 ␤ 1 is not possible. 6 However, when expressed in C2C12 cells, ␣ 11 ␤ 1 binding to collagen I requires M Ca 2ϩ and is sensitive to mM concentrations of Ca 2ϩ , as ␣ 2 ␤ 1 when expressed on the platelet surface. The recent crystallization of soluble ␣ v ␤ 3 supports a role of Ca 2ϩ ions in allosteric modulation of integrin conformation (49). It is possible that a high affinity interaction with collagen I, such as that mediated by ␣ 2 ␤ 1 , is less affected by Ca 2ϩ -induced allosteric conformational changes. Conversely, a lower affinity interaction with collagen I, such as that mediated via ␣ 11 ␤ 1 might be more sensitive to allosteric changes in other regions of the receptor. This differential sensitivity to Ca 2ϩ might be physiologically important in the formation and turnover of the musculoskeletal system, where the local concentration of Ca 2ϩ varies.
Despite the differences in collagen specificity, helical GFOGER-like sequences are recognized by ␣ 1 ␤ 1 , ␣ 2 ␤ 1 , and as shown in this study, also by ␣ 11 ␤ 1 . Careful analysis of the occurrence of GFOGER-like peptides has shown that in addition to the CB3-derived sequences GFOGER and GFOGEK, which are present in the central part of the collagen chain, an amino-terminal ␣ 2 I and ␣ 1 I domain binding site overlaps with the peptide GLOGER (33). As shown in this study, the binding of the different I domains to different collagen peptides varied somewhat. The glutamate in GFOGER was central for the binding of all I domains, whereas the phenylalanine seemed to be more important for ␣ 11 binding, and the arginine was especially important for ␣ 1 binding. It is possible that in vivo collagen-binding integrins prefer certain sites on the collagen molecules. In a particular cell expressing multiple collagen receptors, a number of factors might determine which region in collagen is bound by a particular integrin. Some of the factors that might affect ligand binding include local Ca 2ϩ concentrations, expression levels of the different receptors, and subcellular localization within the cell. Collagen receptors have been shown to affect collagen and matrix metalloproteinase synthesis. Already now it is possible to envisage how GFOGER peptides have the potential to become universal reagents blocking cell-collagen interactions. It will be important to determine whether ␣ 10 ␤ 1 also binds GFOGER peptides. Triple-helical collagen peptides might be of use in conditions characterized by excessive collagen production such as various fibrotic conditions.