Prolyl Hydroxylation of Collagen Type I Is Required for Efficient Binding to Integrin α1β1 and Platelet Glycoprotein VI but Not to α2β1

Collagen is a potent adhesive substrate for cells, an event essentially mediated by the integrins α1β1 and α2β1. Collagen fibrils also bind to the integrin α2β1 and the platelet receptor glycoprotein VI to activate and aggregate platelets. The distinct triple helical recognition motifs for these receptors, GXOGER and (GPO)n, respectively, all contain hydroxyproline. Using unhydroxylated collagen I produced in transgenic plants, we investigated the role of hydroxyproline in the receptorbinding properties of collagen. We show that α2β1 but not α1β1 mediates cell adhesion to unhydroxylated collagen. Soluble recombinant α1β1 binding to unhydroxylated collagen is considerably reduced compared with bovine collagens, but binding can be restored by prolyl hydroxylation of recombinant collagen. We also show that platelets use α2β1 to adhere to the unhydroxylated recombinant molecules, but the adhesion is weaker than on fully hydroxylated collagen, and the unhydroxylated collagen fibrils fail to aggregate platelets. Prolyl hydroxylation is thus required for binding of collagen to platelet glycoprotein VI and to cells by α1β1. These observations give new insights into the molecular basis of collagen-receptor interactions and offer new selective applications for the recombinant unhydroxylated collagen I.

The collagens include the most abundant proteins in mammalian tissues providing a scaffold and framework for extracellular matrix assembly. In addition, they can modulate cell behavior through interactions with cellular receptors. All members of the collagen family are built up of three chains and contain at least one triple helix domain formed by repeating GXY sequences (where X is often proline (P) and Y is often hydroxyproline (O)). Such molecules associate to form complex structures, of which the fibril-forming collagens constitute the most abundant matrix component (1). The higher order orga-nization of the collagens, collagen I being the best example, proved crucial to the triggering of specific cell responses; native triple helical structure is essential for cell and platelet adhesion, and the fibrillar structure is required for platelet activation and aggregation (2).
The cell surface receptors ␣ 1 ␤ 1 and ␣ 2 ␤ 1 of the integrin family are principally collagen receptors, although they can interact with other extracellular matrix components (e.g. laminins). The specificity of the more recently discovered collagen binding integrins, ␣ 10 ␤ 1 and ␣ 11 ␤ 1 , remains to be determined. Different receptor families have also been identified as containing collagen receptors, such as the membrane-spanning proteoglycans, syndecans (3), and the recently described tyrosine kinase receptors DDR1 and DDR2 (4). In platelets, the recently described glycoprotein VI is a key signaling collagen receptor and is an immunoglobulinn superfamily receptor (5).
Several important clues to the understanding of the interaction of ␣ 2 ␤ 1 integrin with collagen arose from the development of triple helical synthetic peptides containing designated recognition sequences (6). From these studies, a GFOGER sequence was identified as a crucial, high affinity ␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrin binding site in collagen I (7). Other GXOGER motifs in the collagen I primary sequence were further acknowledged as common collagen binding sites for both ␣ 1 and ␣ 2 integrin subunits (8). The interaction of integrins with collagen occurs via a 200-amino acid inserted domain (I-domain) homologous with the collagen-binding A-domain of von Willebrand factor in the N-terminal region of the integrin ␣-subunit (9), and it is divalent cation-dependent, requiring ions such as Mg 2ϩ for high affinity. The I-domain provides a general means of integrin-collagen recognition, common also to the ␣ 10 and ␣ 11 subunits (10 -12). Co-crystallization of the ␣2-I domain with the GFOGER triple helix motif indicated a crucial role for the glutamate residue, directly coordinating the metal ion, as well as the existence of ligand-induced conformational changes, which probably underlie either the affinity regulation or the signaling capacity of integrins (13).
Although ␣1and ␣2-I-domains are structurally very similar and recognize the same binding sequences, their relative affinities for collagen types differ; for example, ␣ 1 ␤ 1 binds collagen IV with a higher affinity than collagen I, for which ␣ 2 ␤ 1 is a better receptor (14,15), indicating subtle differences in the binding specificities. Distinct binding sites might also exist as only ␣ 1 ␤ 1 was shown to recognize the transmembrane collagen XIII (16). Furthermore, biological functions such as signal transduction of the two integrins are distinct. ␣ 1 ␤ 1 exerts Rasmediated negative control over collagen synthesis (17), whereas ␣ 2 ␤ 1 activates collagen synthesis and metalloproteinase expression via a different pathway (for a review, see Ref. 18). The two integrins can be differentially expressed; e.g. ␣ 2 ␤ 1 , but not ␣ 1 ␤ 1 , is expressed in platelets, where it is essential for platelet adhesion to collagens of the vascular subendothelium, facilitating subsequent interactions with another platelet-collagen receptor, glycoprotein VI (GPVI), 1 resulting in thrombus formation via platelet activation.
Collagen is a major platelet agonist. Although many collagen receptors have been identified in platelets, ␣ 2 ␤ 1 integrin and GPVI are the most important so far (19). Studies on platelets from integrin-deficient mice (20,21) together with the GPVIimpaired platelets from FcR␥ chain knock-out mice (20,22) have provided new insights into the role of these two receptors in platelet-collagen interactions. Together, the results indicate a crucial role for GPVI in the activation of platelets by fibrillar collagen and suggest a requirement for GPVI recognition that precedes the full adhesive interaction with ␣ 2 ␤ 1 . The ability of GPVI to up-regulate the affinity of ␣ 2 ␤ 1 was shown by Jung and Moroi (23). Thus, an interactive model for collagen receptor function in platelets has been proposed, in which collagen binds GPVI, leading to subsequent activation of the integrin to a higher affinity state that strengthens the initial binding.
Collagen-related peptide (CRP), composed of 10 repeated GPO triplets, was shown to be a selective ligand for GPVI (24). CRP, when cross-linked, is a powerful GPVI-dependent platelet agonist, which mimics the GPVI signaling action of the native collagen I (25). Despite its ability to form a triple helix, a corresponding GPP polymer is inactive, suggesting a specific role of hydroxyproline in collagen-GPVI recognition (26). However, other ligands, such as the snake venom toxin convulxin, can react with GPVI using a distinct mechanism (27), suggesting that other modes of binding GPVI may lead to platelet activation. Since the ␣ 1 ␤ 1 and ␣ 2 ␤ 1 binding motif, GXOGER, like the GPO motif for GPVI, contains a hydroxyproline residue, we aimed to define the role of hydroxyproline in collagen recognition by these major collagen receptors. Using unhydroxylated triple helical collagen I produced in transgenic tobacco (28), we have recently shown that hydroxyproline plays an important role in triple helix folding and fibril formation (29). This recombinant unhydroxylated collagen I offers a direct and potent means of gaining insights into the role of hydroxyproline of the whole collagen molecule in collagen receptor recognition, which complements the use of collagen-like peptides. Although the two integrins are structurally similar (8,30), our data showed that ␣ 2 ␤ 1 but not ␣ 1 ␤ 1 was capable of recognizing the unhydroxylated collagen. Moreover, GPVI-dependent platelet aggregation was shown to be strictly dependent on prolyl hydroxylation. Collagens contain a high percentage of hydroxyproline, a very rare residue in other proteins. This molecular signature indicates specific roles for hydroxyproline in collagen; diverse events such as molecule folding, thermal stability and fibril formation require prolyl hydroxylation. To this list we add an essential role in cell adhesion and platelet aggregation, two crucial biological functions of collagen.

MATERIALS AND METHODS
Collagen Preparation-Collagens were extracted from fetal calf bones by pepsin digestion and the two collagen I isoforms, namely the heterotrimer [␣1(I)] 2 ␣2(I) and the homotrimer [␣1(I)] 3 , were separated by repeated salt fractionation as previously described (2). Recombinant collagen I homotrimer (rColl I) was extracted and purified from fieldgrown tobacco plants transformed with human pro␣1(I) chain cDNA lacking the N-propeptide coding sequence, referred to as PRS-⌬Npro␣1(I) (29). Hydroxylated recombinant collagen was obtained by co-transformation of tobacco leaves with the collagen construct PRS-⌬Npro␣1(I) and cDNA coding the two subunits of the prolyl-4-hydroxylase, the C. elegans ␣-subunit and the mouse ␤-subunit, as described (31). Collagen concentrations were determined by amino acid analysis. Samples were hydrolyzed under vacuum (6 N HCl, 115°C, 24 h) in the presence of 2␤-mercaptoethanol in a Pico Tag system (Waters) and then analyzed with a Beckman amino acid analyzer.
Collagen fibrils were obtained by dialyzing collagen solution (400 g/ml in 0.5 M acetic acid) against 10 mM phosphate buffer at 4°C as previously described (29). Standard collagen fibril suspension was a gift from Ethicon Inc. (Somerville, NJ) and was dialyzed against 10 mM acetic acid before use.
96-Well plates (Costar) were coated overnight at 4°C with collagen substrates at a concentration range of 0 -10 g/ml (0.1 ml/well) or other concentrations as specified. After blocking the wells with 1% BSA, cells freshly suspended in serum-free Dulbecco's modified Eagle's medium (6 ϫ 10 5 cells/ml) were plated onto wells (0.1 ml/well) and allowed to attach at 37°C for 20 -45 min, depending on the cell type used. Then the wells were washed with PBS, fixed with 1% glutaraldehyde, and stained with 0.1% crystal violet in water. After extensive washing, the dye adsorbed to the cells was solubilized with 0.2% Triton X-100, and the absorbance was read at 570 nm with an ELISA plate reader (SLT Lab Instrument). Each assay point was carried out in triplicate. For inhibition assays, wells were routinely coated at a collagen concentration of 0.1 g/well; only rColl I was coated at 1 g/well with HBL100. Cells freshly suspended in serum-free Dulbecco's modified Eagle's medium were mixed with the antibodies at a given concentration for 10 min at 37°C and then seeded onto the coated wells. The assays were then continued as described above. Adhesion is normalized to the cell attachment signal on bovine collagen I heterotrimer.
Solid-phase Assays-Soluble human ␣ 2 ␤ 1 integrin, consisting of the two integrin ectodomains joined together by the dimerizing motifs of the transcription factors Fos and Jun, were generated as described previously (34). Similarly, soluble ␣ 1 ␤ 1 integrin was generated, which will be described elsewhere. 2 To measure ␣ 2 ␤ 1 integrin binding, bovine type I collagen and rColl I were coated at 40 g/ml. To determine ␣ 1 ␤ 1 integrin binding, all collagen substrates were coated at 20 g/ml to a microtiter plate at 4°C overnight. Nonspecific binding sites were blocked with 1% heat-denatured BSA in TBS, pH 7.4, supplemented with 2 mM MgCl 2 (TBS/Mgbuffer). Soluble ␣ 2 ␤ 1 or ␣ 1 ␤ 1 integrin at the indicated concentrations was added in 1% BSA in TBS/Mg-buffer, containing 1 mM MnCl 2 , to the wells for 2 h at room temperature. Binding of soluble ␣ 2 ␤ 1 was performed in the presence of 400 nM 9EG7, a ␤ 1 integrin-activating antibody (kindly provided by Dr. D. Vestweber, University of Muenster, Germany). After removing nonbound integrins by washing twice with 50 mM HEPES, pH 7.4, 150 mM NaCl, 2 mM MgCl 2 , 1 mM MnCl 2 , either without or with 10 mM EDTA, bound integrin was fixed to the plate by 2.5% glutaraldehyde in the same buffer for 10 min and detected by an rabbit antiserum directed against the human integrin ␤ 1 subunit, di- 1 The abbreviations used are: GPVI, glycoprotein VI; CRP, collagenrelated peptide; rColl I, recombinant collagen I homotrimer; HUVEC, human umbilical vein endothelial cell; BSA, bovine serum albumin; ELISA, enzyme-linked immunosorbent assay; OHrColl I, hydroxylated recombinant collagen I. 2 S. Niland and S. A. Eble, manuscript in preparation. luted 1:400 in 1% BSA in TBS/Mg-buffer, followed by alkaline phosphatase-conjugated antibodies directed against rabbit immunglobulins (Sigma), diluted 1:600 in 1% BSA in TBS/Mg-buffer. Between the antibody incubations, the wells were washed three times with TBS/Mgbuffer. Eventually, the ELISA was developed by conversion of paranitrophenylphosphate (Sigma), which was measured in an ELISA reader at 405 versus 595 nm. Nonspecific background values, measured in the presence of 10 mM EDTA, were subtracted from the OD values to obtain specific integrin binding signals. Platelet Adhesion and Aggregation-Fresh blood was provided by the National Blood Service (Cambridge, UK) or by the "Centre de Transfusion Sanguine" (Lyon, France).
Platelet adhesion was measured colorimetrically at 20°C using washed platelets essentially as described (35). Washed platelets were adjusted to a platelet concentration of 10 8 /ml with 10 mM Hepes, 145 mM NaCl, 5 mM KCl, 1 mM MgSO 4 , 10 mM glucose, pH 7.4. Immulon2 96-well plates (Dynex Technologies, Ashford, Middlesex, UK) were coated with collagen I monomers or CRP at a concentration of 10 g/ml or with collagen fibrils at 50 g/ml overnight at 4°C and were blocked with 5% BSA for 30 min prior to adhesion. 2 mM MgCl 2 was added 5 min before seeding platelets onto coated wells in triplicate. For inhibitory assays, the platelet suspension was preincubated with 2 g/ml 6F1 or 0.5% anti-GPVI or control rabbit serum or with 2 M GR 144053F compound for 5 min after the addition of MgCl 2 .
Aggregation was measured turbidimetrically at 20°C after the addition of collagen fibrils as indicated, using human citrated PRP obtained by centrifugation of fresh blood 12 min at 200 ϫ g. The aggregation test was carried out for at most 12 min with magnetic stirring at 900 rpm (Icare).
Electron Microscopy-Platelet suspension and aggregates were collected after platelet aggregation measurement, rinsed with PBS, and subsequently fixed in 2% glutaraldehyde in PBS for 30 min at room temperature. The samples were then postfixed in 1% OsO4 in PBS for 30 min, dehydrated in a graded ethanol series, and embedded in epoxy resin. Ultrathin sections were contrasted with uranyl acetate and lead citrate, and observations were performed with a Philips CM120 microscope at the Centre Technologique des Microstructures (Université Claude Bernard, Lyon I, France).

RESULTS
Cell Adhesion-promoting Activity of the Unhydroxylated Collagen-Very many cells can adhere to and spread on collagen type I. Because the adhesive properties of the recombinant collagen are important for its development as a biomaterial, we assessed the ability of primary cell types, including human skin primary fibroblasts and endothelial cells (HUVECs), to bind to rColl I. Both of these cell type express the major collagen receptors, the integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 . Our results show that adhesion and spreading of both the fibroblasts and HUVECs were fully supported by rColl I (Fig. 1).
To investigate whether ␣ 1 ␤ 1 and ␣ 2 ␤ 1 could mediate cell adhesion and spreading on rColl I, we selected three different cell lines with differential integrin expression. HBL100 (a human mammary epithelial cell line) contains significant amounts of both ␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrins; HT1080 (a human fibrosarcoma cell line) and RuGli (a rat gliosarcoma) mainly express ␣ 2 ␤ 1 and ␣ 1 ␤ 1 , respectively. HBL100 cells adhered less efficiently to rColl I than to homotrimeric and heterotrimeric native collagen I. The plateau levels of adhesion were always about 20 -30% lower in the different experiments ( Fig. 2A). HT1080 cells showed comparable levels of adhesion to rColl I and to the bovine collagens, having similar plateau values to each substrate (Fig. 2B). The morphology of the adherent cells under phase-contrast microscopy showed that both cell types were capable of fully spreading on rColl I (data not shown). In contrast, whereas RuGli cells attached well to homo-and heterotrimeric collagens, no adhesion was observed to rColl I (Fig.  2C), suggesting that interaction of cells with rColl I was ␣ 2 ␤ 1 -specific.
Adhesion to Unhydroxylated Collagen I Is ␣ 2 ␤ 1 -but not ␣ 1 ␤ 1dependent-To verify the role of ␣ 2 ␤ 1 , but not ␣ 1 ␤ 1 in cell adhesion to rColl I, function-blocking monoclonal antibodies specific for ␣ 1 , ␣ 2 , and ␤ 1 were introduced. Complete inhibition of HBL100 cell adhesion to bovine homotrimeric and heterotrimeric collagens was achieved only by the simultaneous addition of antibodies against ␣ 1 and ␣ 2 integrin subunits, whereas antibody to ␤ 1 subunit abolished HBL100 cell adhesion to all substrates. In contrast, HBL100 cell adhesion to the recombinant collagen was inhibited by antibodies against ␣ 2 but not ␣ 1 subunit (Fig. 3A). Antibodies against ␣ 2 and ␤ 1 integrin subunits completely abolished HT1080 cell adhesion to all substrates (Fig. 3B). Complete inhibition of RuGli cell adhesion to homo-and heterotrimeric collagens was achieved by the addition of rat ␣ 1 antibodies (Ha31/8) (Fig. 3C), confirming that adhesion was solely mediated by ␣ 1 ␤ 1 . Altogether, the data further support the concept that ␣ 2 ␤ 1 integrin is the dominant adhesion receptor for the unhydroxylated collagen.
Prolyl Hydroxylation of the Recombinant Collagen Improves the ␣ 1 ␤ 1 Binding Activity-The ␣ 1 ␤ 1 and ␣ 2 ␤ 1 binding sites in the collagen I identified so far are identical, and each contains a GXOGER motif (7,8). To assess the dependence of the integrin affinity on proline hydroxylation of collagen I, solid phase binding assays with soluble recombinant integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 were performed. Binding to collagen substrates was measured over a range of integrin concentrations in the presence of 2 mM Mg 2ϩ or of 1 mM EDTA. Cation-dependent binding of ␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrins was confirmed for all substrates. Whereas ␣ 2 ␤ 1 binding to bovine collagen I and to unhydroxylated homotrimer was similar (Fig. 4A), the apparent affinity of ␣ 1 ␤ 1 for the fully hydroxylated bovine collagens was 1 order of magnitude greater than for the unhydroxylated homotrimer (Fig. 4B). The importance of prolyl hydroxylation for ␣ 1 ␤ 1 binding to collagen was further assessed by analyzing ␣ 1 ␤ 1 adhesion to hydroxylated recombinant collagens (OHrColl I), obtained by co-transfection with collagen and prolyl-4-hydroxylase genes as described (31). The samples tested, OHrColl I a and b, were purified from two different stably transfected plants and contained 7.2 and 8.9% hydroxyproline, respectively, whereas native bovine collagen I contained 10.2% hydroxyproline. Recombinant ␣ 1 ␤ 1 at all concentrations tested bound significantly better to OHrColl I than to rColl I, and binding increased depending on the increasing hydroxylation level (Fig. 4B). These data confirmed that reduction of collagen I prolyl hydroxylation significantly impaired ␣ 1 ␤ 1 binding activity, consistent with the finding that cells adhered to the unhydroxylated rColl I solely via integrin ␣ 2 ␤ 1 .
Platelets Adhere to the Recombinant Collagen in an ␣ 2 ␤ 1 Ϫdependent Manner-Platelet-collagen interaction is considered to be mediated essentially by the two collagen receptors, ␣ 2 ␤ 1 and GPVI. According to the original two-site two-step receptor model, platelets first bind via ␣ 2 ␤ 1 and subsequently become activated by GPVI. The present observation that ␣ 2 ␤ 1 mediates the adhesion of nucleated cells to rColl I suggested that the recombinant collagen might also support platelet adhesion. Whereas both rColl monomers and fibrils did support some platelet adhesion, the level of attachment was lower than to the control substrates (i.e. native bovine collagen I fibrils, pepsindigested collagen I monomers, and the GPVI-specific CRP) (Fig. 5), suggesting that the presence of hydroxyproline enhances platelet binding to collagen, via either GPVI or the GPVI-mediated activation of ␣ 2 ␤ 1 . The roles of ␣ 2 ␤ 1 and GPVI in platelet adhesion to rColl I, were assessed by using inhibitory antibodies. To measure primary adhesion rather than processes secondary to platelet activation caused by anti-GPVI antibodies, the fibrinogen receptor ␣ IIb ␤ 3 was blocked by GR144053F, an RGD mimetic that prevents fibrinogen binding. Platelet adhesion to the rColl I monomers and fibrils was completely inhibited only by anti-␣ 2 monoclonal antibody, whereas the addition of anti-GPVI serum did not affect platelet adhesion, as was the case with CRP (Fig. 5).
Defective Platelet Aggregation in Response to Unhydroxylated Fibrillar Collagen I-GPVI has been reported to be essential for collagen-induced platelet activation. To test the requirement for proline hydroxylation on platelet aggregation, we generated rColl I fibrils as previously described (29). The formation of banded fibrils with rColl I and with bovine collagen I was verified by negative staining in transmission electron microscopy (not shown). The conditions used for fibril formation proved suitable to obtain rColl I and bovine collagen I heterotrimer banded fibrils but, as previously observed (29), gave less organized fibrils with the bovine collagen I homotrimer. Although it was previously shown that formation of fibrils increased the melting temperature of the recombinant collagen from 30 to 36°C (29), platelet aggregation was performed at 20°C to avoid microunfolding of the recombinant collagen. Whereas 50 g/ml bovine heterotrimeric and 100 g/ml homotrimeric collagen I fibrils induced platelet shape change and aggregation to the same extent, no change in light transmission was observed after the addition of up to 100 g/ml rColl I fibrils (Fig. 6A). We used electron microscopy to further characterize the interaction of platelets with rColl I fibrils and confirmed that platelets stirred in presence of rColl I fibrils did not become activated, since they changed their shape only poorly, produced few short pseudopods, and seemed to secrete no ␣ and dense granules (Fig. 6B). rColl I fibrils were observed in contact with platelet membranes only occasionally, further supporting the concept that, even if interaction between fibrils and platelets occurred, these fibrils were not able to induce platelet activation (Fig. 6C). In contrast, bovine collagen I fibrils were completely trapped in the platelet aggregates they had induced (Fig. 6D). Altogether, the data indicate that rColl I fibrils can interact with platelets via ␣ 2 ␤ 1 integrin under static conditions as shown by the adhesion assay. Under stirring conditions, this interaction is poor, since platelets do not become activated because the GPVI recognition sequence GPO is replaced by a GPP sequence in the rColl I, and further, GPVI cannot induce platelet (and integrin) activation. In conclusion, lacking hydroxyl-prolyl residues, recombinant collagen I produced in plants is able to interact with ␣ 2 ␤ 1 integrin only and does not support ␣ 1 ␤ 1 integrin-mediated cell adhesion or GPVI-induced platelet aggregation and activation. FIG. 2. Dose-response profiles of HBL100 (A), HT1080 (B), and RuGli (C) cells on bovine collagen I heterotrimer (square), bovine collagen I homotrimer (triangle), and recombinant collagen I (circle). Adhesion was determined by a microcolorimetric assay as described under "Materials and Methods."

DISCUSSION
Collagen I, a major component of extracellular matrix of all mammalian connective tissues, provides structural support and modulates cell behavior through cell surface receptors. Among the receptors described for collagen I, the integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 and the platelet receptor GPVI are by far the most important in terms of cell surface distribution and biological function.
␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrins share common structural features and collagen binding mechanisms. Both integrins interact with the triple helix of collagens through their I-domains, and high affinity binding requires the presence of divalent cations. Most importantly, several collagen motifs have been identified as common binding sites for ␣ 1 ␤ 1 and ␣ 2 ␤ 1 . Despite these similarities, we show several lines of evidence that cell adhesion to unhydroxylated rColl I was entirely dependent on the ␣ 2 ␤ 1 integrin. RuGli cells, which express solely ␣ 1 ␤ 1 integrin, did not adhere to the rColl I, whereas hydroxylated collagen I promoted substantial ␣ 1 ␤ 1 -dependent cell adhesion. This finding was supported by differential binding of recombinant ␣ 2 ␤ 1 and ␣ 1 ␤ 1 to surfaces coated with unhydroxylated rColl I; ␣ 1 ␤ 1 showed substantially decreased affinity to the rColl I compared with bovine collagen I. These binding assays, performed at 20°C, are in good agreement with the cell adhesion assays and exclude the possibility that the lack of ␣ 1 ␤ 1 binding could be due to rColl I unfolding during the cell adhesion assays (37°C for 30 -40 min). Moreover, ␣ 2 ␤ 1 and ␣ 1 ␤ 1 -mediated cell adhesion is known to be strictly dependent on triple helical conformation, and if rColl I unfolding occurred during cell adhesion assays, this would have affected adhesion via both integrins. As reported earlier, ␣ 1 ␤ 1 binding to collagen IV involves individual residues located on the two different chains, ␣ 1 (IV) and ␣ 2 (IV) (36). However, the possibility that ␣ 1 ␤ 1 did not recognize the rColl I because of its homotrimeric form can be excluded, since RuGli expressing ␣ 1 ␤ 1 adhered to homotrimeric collagen I, [␣1(I)] 3 , as well as to the [␣1(I)] 2 ␣2(I) heterotrimeric collagen I (Fig. 2). Previously, ␣ 1 ␤ 1 was also found to bind to the homotrimeric GFOGER-containing peptide (7). Thus, the major difference between rColl I produced in tobacco and the bovine collagen I is the lack of prolyl hydroxylation (29). The requirement of hydroxyproline for ␣ 1 ␤ 1 binding was clearly demonstrated by the rescue of integrin binding by the use of hydroxylated recombinant collagen obtained by co-transfection of tobacco by collagen and prolyl-4-hydroxylase genes. Interestingly, the high affinity binding sites recognized so far for both ␣ 2 ␤ 1 and ␣ 1 ␤ 1 integrins all contain a GXOGER motif (7,8). One might expect that the conserved residues in the motif, including hydroxyproline, would play a role in the interaction. The GFOGER-containing collagen peptide was co-crystallized with the recombinant ␣2-I domain of the ␣ 2 ␤ 1 integrin, the presence of a hydrophobic amino acid at the X position increasing the affinity (13). Structural analyses of the complex revealed the crucial role of the Glu residue that coordinates with the metal Wells were coated with various collagen substrates: rColl, native collagen I molecules (PD), and CRP at 10 g/ml, and recombinant collagen fibrils (rColl F) and native collagen I fibrils (Ethicon) at 50 g/ml. Adhesion was measured as described under "Materials and Methods." B, inhibition of GR144053F-treated platelet adhesion (100% value, open bars) to immobilized fibrillar and monomeric recombinant collagen I and CRP by a monoclonal antibody against ␣ 2 integrin subunit (6F1, gray bars), a rabbit antiserum against GPVI (black bars), or a control rabbit serum (hatched bars). Data are presented as means Ϯ S.D. and was normalized between experiments (n ϭ 4) using adhesion to PD collagen as 1. Some PD inhibition data were obtained from a different set of similar experiments (n ϭ 3). ion, the Arg residue forming a salt bridge to an Asp residue in the integrin subunit. No special importance was attached to the Hyp residue, which made a main-chain hydrogen bond to Asn 154 of the ␣ 2 subunit I domain, a residue that is conserved in the I-␣1 domain sequence (30). Here, we proved experimentally that the hydroxyl group in the collagen binding sequence is only essential for ␣ 1 ␤ 1 binding. The recognition sequence shown to interact with collagen is very similar among the ␣-I domains, leading to the conclusion that the molecular basis of the interaction between the collagen hexapeptide and the ␣2-I domain may represent a general mechanism for integrin-ligand recognition (12,13). However, the structural differences between the two I-domains, a larger trench for ␣ 1 -I compared with that of ␣2-I, suggest a different ligand binding mechanism (37). The affinity of ␣ 1 ␤ 1 for the hydroxylated collagen may have its molecular basis in bonds formed between hydroxyl group of hydroxyprolines and residues from the ␣1-I-domain (so far not characterized). However, we cannot exclude the possibility that the absence of hydroxyproline, known to stabilize the collagen molecule, leads to the formation of a less compact triple helix, although we would expect this to have a larger impact on ␣ 2 ␤ 1 binding, given the shape of the ligandbinding surface of the ␣2-I-domain. We have previously shown that although rColl I formed a stable triple helix, it was more flexible than the bovine collagen I, measured by dynamic light scattering (29). Since the x-ray structure of the complex shows residues from two strands of the triple helix interacting with ␣2-I domain, the increased distance between the collagen ␣ chain strands in the flexible unhydroxylated molecule could cause a decrease in I-domain binding. Co-crystallization of collagen-like peptides with the ␣ 1 -I domain may suggest why this might destabilize the ␣ 1 -I-domain interaction specifically.
Collagen I provides not only a potent adhesive substrate for cells in general, but in its fibrillar form it also activates platelets through the GPVI receptor, leading to platelet aggregation, whereas ␣ 2 ␤ 1 mediates adhesion to both monomeric and fibrillar collagen. However, the role assigned to each receptor in platelet activation is not clear. Loss of function of GPVI in both human and mouse platelets results in loss of platelet aggregation, but deficiencies in human (38,39) and mouse (20, 21) ␣ 2 ␤ 1 provided conflicting data. The two-step model proposes that collagen responses mediated by GPVI require the integrin ␣ 2 ␤ 1 as a co-receptor (40).
The role of hydroxyproline in GPVI recognition was already suggested using collagen mimetic peptides comprising repeated triplets of GPO versus GPP (26). However, the use of short peptides excluded the possible contribution of additional motifs present in the whole molecule in GPVI binding. Here, we made the demonstration of the crucial role of hydroxyproline by testing directly the ability of fibrils formed with either hydroxylated or unhydroxylated collagen I homotrimer to elicit platelet aggregation. Moreover, the use of function-blocking antibodies to GPVI and ␣ 2 ␤ 1 showed that platelet adhesion to the rColl I monomers and fibers was entirely ␣ 2 ␤ 1 -dependent, without GPVI interaction. Furthermore, we observed that, in the presence of 2 mM Mg 2ϩ , relative platelet adhesion to the rColl I monomers and fibrils was surprisingly low compared with the native collagen I fibrils or monomers, whereas ␣ 2 ␤ 1 -containing nucleated cells (Figs. 2 and 3) and soluble recombinant ␣ 2 ␤ 1 bound as well to the rColl I as to the bovine collagen I homotrimer and heterotrimer (Fig. 4). Integrin ␣ 2 ␤ 1 can undergo conformational change from low to high affinity state following GPVI-platelet activation (23). Our data suggest that impaired FIG. 7. Distinct responses to the unhydroxylated collagen of the major collagen receptors. Efficient cell adhesion is ensured by ␣ 2 ␤ 1 recognition of GXPGER-like motifs of the unhydroxylated recombinant, whereas, since the major platelet-collagen receptor GPVI cannot recognize unhydroxylated collagen motifs, platelet aggregation is abrogated. In the absence of GPVI binding to the unhydroxylated collagen, ␣ 2 ␤ 1 is not activated and binds moderately to the recombinant collagen. This scheme illustrates the possible use of the recombinant collagen produced in transgenic plants as a suitable alternative for collagen-based vascular graft by promoting endothelialization of the graft surface while preventing thrombus formation. GPVI binding to collagen prevents ␣ 2 ␤ 1 from becoming activated to a high affinity state and, consequently, platelets adhere more poorly. In mice, GPVI-deficient platelets completely failed to adhere to both fibrillar and soluble collagen (20). Although species-specific differences cannot be excluded, another possible explanation for this discrepancy is that the experiments were carried out in the presence of a low concentration of Mg 2ϩ (1 mM) but at an ␣ 2 ␤ 1 -inhibiting concentration of Ca 2ϩ . Indeed, the specificity and concentration of divalent cations is crucial for ␣ 2 ␤ 1 binding to collagen (13,41). Moreover, electron microscopy of citrated platelet-rich plasma, where cation levels are low, stirred with recombinant collagen fibrils showed few if any platelets interacting with collagen fibrils (Fig. 6) and thus confirmed the requirement for either high divalent Mg 2ϩ concentration (42) or GPVI-mediated platelet activation for efficient ␣ 2 ␤ 1 -dependent platelet adhesion. Therefore, by using modified recombinant collagen I, our overall results, illustrated in Fig. 7, support previous observations that GPVI is essential for platelet activation leading to aggregation and thus enables ␣ 2 ␤ 1 integrin to engage in high affinity binding to collagen (20,23). In conclusion, we provide entirely novel evidence to verify that the GPO motif is a sufficient and highly specific GPVI-recognition site in collagen, critical for platelet activation and, more importantly, show the unexpected crucial role of proline hydroxylation for ␣ 1 ␤ 1 but not ␣ 2 ␤ 1 binding to collagen.
The physiological consequence of platelet interaction with the subendothelial collagen is the formation of a thrombus that prevents blood flow. Such occlusion is the major cause of failure of collagen-based vascular devices and necessitates their replacement. Acting at the level of receptor-specific collagen motifs may also help to generate substitutes for collagen-based materials. The present work showed that the unhydroxylated rColl I promotes efficient ␣ 2 ␤ 1 -mediated cell adhesion of several cell lines including endothelial cells but will not support platelet aggregation into a growing thrombus due to lack of GPVI-dependent activation. Critical for its use as a biomaterial, the thermal stability of recombinant collagen is paramount. The melting temperature of rColl I can be increased to 36°C by forming fibrils (29), which might easily be crosslinked, ensuring higher thermal stability of the final product. Thus, recombinant collagen represents an interesting substrate for the development of collagen-based vascular grafts that might allow fine-tuning of the delicate balance between endothelialization of the graft surface and avoidance of thrombus formation.