Use of Synthetic Peptides to Locate Novel Integrin α2β1-binding Motifs in Human Collagen III*

A set of 57 synthetic peptides encompassing the entire triplehelical domain of human collagen III was used to locate binding sites for the collagen-binding integrin α2β1. The capacity of the peptides to support Mg2+-dependent binding of several integrin preparations was examined. Wild-type integrins (recombinant α2 I-domain, α2β1 purified from platelet membranes, and recombinant soluble α2β1 expressed as an α2-Fos/β1-Jun heterodimer) bound well to only three peptides, two containing GXX′GER motifs (GROGER and GMOGER, where O is hydroxyproline) and one containing two adjacent GXX′GEN motifs (GLKGEN and GLOGEN). Two mutant α2 I-domains were tested: the inactive T221A mutant, which recognized no peptides, and the constitutively active E318W mutant, which bound a larger subset of peptides. Adhesion of activated human platelets to GER-containing peptides was greater than that of resting platelets, and HT1080 cells bound well to more of the peptides compared with platelets. Binding of cells and recombinant proteins was abolished by anti-α2 monoclonal antibody 6F1 and by chelation of Mg2+. We describe two novel high affinity integrin-binding motifs in human collagen III (GROGER and GLOGEN) and a third motif (GLKGEN) that displays intermediate activity. Each motif was verified using shorter synthetic peptides.

The integrins form an important and widespread group of cell adhesion receptors, interacting either with proteins of the extracellular matrix such as fibronectin and collagen or with counter-receptors expressed on other cells, as occurs, for example, in the binding of the leukocyte integrins to intercellular adhesion molecules (ICAM), vascular cell adhesion molecules (VCAM), and others. Integrins exhibit Mg 2ϩdependent adhesion, the cation being coordinated within an acidic pocket known as the metal ion-dependent adhesion site (MIDAS). 3 The MIDAS is located within a structure known as an I (inserted)-domain, which is closely related to the von Willebrand factor A-domains. Thus, aspartate-containing motifs such as RGD and LDV can bind to some integrin ␤-subunits through their I-domains, and this is the primary mode of interaction of the integrins with fibronectin, fibrinogen, and many other ligands (1).
The collagens in native triple-helical form interact with a subset of the integrins: ␣ 1 ␤ 1 , ␣ 2 ␤ 1 , ␣ 10 ␤ 1 , and ␣ 11 ␤ 1 . These differ from the ␤ 3 and other ␤ 1 integrins in that they recognize their ligands through an I-domain located between blades 2 and 3 of the integrin ␣-subunit ␤-propeller rather than through the I-domain within their ␤-subunit (2). Integrin ␣ 2 ␤ 1 , in common with other members of the integrin family, can exist in at least two affinity states that are believed to correspond to the two conformations of the ␣ 2 I-domain revealed by crystallography (3,4). In these structures, helix 7 moves axially within the I-domain to adopt two extreme positions, up or down, causing rearrangement of the loops that conceal or expose the MIDAS and so regulate its ligation by GER motifs found in the triple-helical domain of the collagens. Platelet integrins, including ␣ 2 ␤ 1 and the fibrinogen receptor ␣ IIb ␤ 3 , can be regulated by cell activation. Platelet agonists that can activate ␣ 2 ␤ 1 through such inside-out signaling include ADP and thrombin. However, monoclonal antibody TS2/16 binds to the integrin ␤ 1 -subunit and induces a high affinity conformation of ␣ 2 ␤ 1 directly. All these agents increase the affinity of ␣ 2 ␤ 1 for integrin-binding motifs in collagen (5,6). This capacity for activation has led to the proposal that more than one active state of ␣ 2 ␤ 1 exists (5,7). Recent models and mutagenesis of the ␤ 2 integrins suggest that their I-domains can adopt three positions determined by the position of helix 7, although such diversity has been discounted for ␣ 2 ␤ 1 (8). Several recognition sites have been identified within collagen for ␣ 1 ␤ 1 , ␣ 2 ␤ 1 , ␣ 10 ␤ 1 , and ␣ 11 ␤ 1 . They include GXXЈGER motifs, with GFOGER (where O is hydroxyproline) (9) and GLOGER and GASGER (10) being located within the collagen ␣1(I) chain and with GMOGER, GLSGER, and GAOGER (all found in collagen III) being identified from the use of model peptides as integrin-binding sequences (7). Like GAS-GER, GAOGER is a low affinity motif, requiring activation, whereas GFOGER will bind ␣ 2 ␤ 1 in the resting platelet. The role of specific residues in stabilizing the binding of collagen to the activated ␣ 2 I-domain can be deduced from the structure of the co-crystal complex of the ␣ 2 I-domain with a GFOGER-containing triple-helical peptide (4). There may be selectivity between the different integrins for different GER sequences, with ␣ 1 ␤ 1 and ␣ 11 ␤ 1 (11) binding to GLOGER almost as well as to GFOGER. GEK motifs can be substituted for GER, binding with somewhat lower affinity (9,11).
Platelet collagen receptors may provide novel targets in thrombosis, being important in the adhesive process itself, the subsequent activation of platelets, or both. We believe that ␣ 2 ␤ 1 exercises a largely adhesive role, capturing platelets on subendothelial collagens that become exposed in the blood vessel wall by injury or disease, because blockade of ␣ 2 ␤ 1 reduces the number of platelets deposited without preventing their activation. In contrast, glycoprotein VI (GPVI), a non-integrin receptor of the immunoglobulin family, is a signaling receptor that is predominantly activatory rather than adhesive because blockade of GPVI does not prevent human platelet deposition on collagen I fibers under arterial shear conditions (12). Collagens I, III, and several others are present in the vascular subendothelium (13) and are enriched in atherosclerotic plaque (14), where they may be instrumental in facilitating platelet adhesion and in promoting platelet activation. Platelet adhesion to collagen III is dependent upon ␣ 2 ␤ 1 under static and shear conditions (15); however, the high affinity ␣ 2 ␤ 1 -binding sequence GFOGER does not occur in collagen III. This raises the question as to how collagen III interacts strongly with the platelet surface. GPVI may play a more prominent role in the adhesion of platelets to collagen III because, unlike collagen I (12), impairment of platelet adhesion to collagen III may occur after GPVI blockade (16). This suggests either direct GPVImediated adhesion to collagen III or GPVI-mediated platelet activation of integrin ␣ 2 ␤ 1 , as has been described previously (5).
To investigate collagen III in detail, we synthesized the entire triplehelical domain of collagen III as a set of overlapping homotrimeric peptides (termed the Collagen III Toolkit), an extension of the strategy we used previously to identify binding sites in cyanogen bromide peptide 3 of collagen ␣1(I) (9,17). The Toolkit approach is comprehensive, allowing us to examine, for example, GXXЈGER motifs (where X need not be hydrophobic) and to include sequences that differ from this prototype motif. Our past focus has been on sequences with a hydrophobic residue (Phe, Leu, Met, or, less favorably, Ala) in the X position, but the Toolkit allows us to examine other sequences without bias. Although we assume that the glutamate residue is an absolute requirement for coordination of the divalent cation in the MIDAS (18), the use of the Toolkit allows this assumption to be tested rigorously for collagen III.
Peptides identified as active by this strategy were used to test the binding of resting and activated human platelets. Furthermore, we used HT1080 human fibrosarcoma cells, which express ␣ 2 ␤ 1 as their major collagen-binding integrin, to verify that these findings are applicable to cells other than platelets.

MATERIALS AND METHODS
Human platelets were isolated from citrate-anticoagulated whole blood, provided by the National Blood Service (Cambridge, UK) (20). Anti-␣ 2 monoclonal antibody 6F1 was a kind gift from Dr. Barry Coller (Mount Sinai Hospital, New York). GR144053F was a gift from Glaxo-SmithKline (Stevenage, UK). Horseradish peroxidase-conjugated antiglutathione S-transferase antibody was from Amersham Biosciences (Buckinghamshire, UK). Unless stated otherwise, all other reagents were from Sigma (Poole, Dorset, UK).
Peptides-The primary sequences of the peptides of the Collagen III Toolkit are shown in Table 1. Other peptides are described below. The host-guest strategy (21,22) was applied as described (9,17), in which the guest sequence of interest is placed between (GPP) 5 hosts, the flanking sequences that impart triple-helical conformation on the whole peptide. Each peptide contains a guest sequence of 27 amino acids, the C-terminal 9 amino acids of which form the first 9 guest amino acids of the next peptide. Thus, the guest sequence of the Toolkit advances 18 amino acids along the triple-helical sequence of collagen III with each successive peptide, and a 9-amino acid overlap is included between adjacent peptides. A subset of peptides was synthesized in which hexameric GXXЈGEXЉ guest sequences were included between (GPP) 5 hosts, as described above, to test the adhesive properties of sequences of interest.
Peptide Synthesis-Peptides were synthesized as C-terminal amides on TentaGel R RAM resin in a PerSeptive Biosystems Pioneer auto-mated synthesizer. In general, Fmoc-amino acids (4 eq) were activated with O-(1H-6-chlorobenzotriazole-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate (4 eq) in the presence of N-ethyldiisopropylamine (8 eq). The cysteine residues were coupled as Fmoc-Cys-(trityl)pentafluorophenyl ester (4 eq) in the presence of 1-hydroxy-7-azabenzotriazole (4 eq). When followed by an aspartate residue, glycine was introduced as Fmoc-(2-Fmoc-oxy-4-methoxybenzyl)Gly-OH to avoid aspartimide formation. Fmoc deprotection (Method A) was with a mixture of 2% (v/v) piperidine and 2% (v/v) 1,8-diazabicyclo [5.4.0]undec-7ene in N,N-dimethylformamide. In the presence of the dipeptidyl unit Asp-Xaa (where Xaa is not Pro, Hyp, or Gly), Fmoc deprotection (Method B) was with a mixture of 20% piperidine and 0.1 M 1-hydroxybenzotriazole in N,N-dimethylformamide to minimize aspartimide formation. Peptides were released from the resin by treatment with a mixture of trifluoroacetic acid, water, triisopropylsilane (92.5:2.5:2.5, v/v), and dithiothreitol (2.5 g) for 3 h at room temperature. After precipitation with diethyl ether, the crude products were purified by reverse phase HPLC on a 10-m Ace Phenyl-300 column using a linear gradient of acetonitrile in water containing 0.1% trifluoroacetic acid. Fractions containing homogeneous product were identified by analytical HPLC on a 5-m Ace Phenyl-300 column, pooled, and freeze-dried. All peptides were characterized by matrix-assisted laser desorption ionization time-of-flight mass spectrometry at the Protein and Nucleic Acid Chemistry Facility of the Cambridge Centre for Molecular Recognition.
Integrin Preparations-The recombinant I-domain-encoding plasmids for ␣ 2 (2) and the ␣ 2 mutant E318W (19) were a generous gift from Dr. Danny Tuckwell (F2G Ltd., Manchester, UK). For protein expression and purification of ␣ 2 , T221A, and E318W I-domains, a 40-ml overnight culture of transformants (BL21) was used to inoculate 400 ml of Luria broth containing 50 g/ml ampicillin. The culture was grown for 1 h at 37°C and then induced for 4 h with 0.1 mM isopropyl ␤-Dthiogalactopyranoside at room temperature. Cells were harvested by centrifugation at 4500 ϫ g for 10 min, and pellets were resuspended in Dulbecco's phosphate-buffered saline (Sigma) without divalent cations. Suspensions were sonicated and centrifuged at 2500 ϫ g for 10 min, and Triton X-100 was added to the supernatants to 1% (v/v). Pellets were resuspended in Dulbecco's phosphate-buffered saline without divalent cations, sonicated, and centrifuged twice more; and the supernatants were pooled. The lysate was passed down a glutathione-agarose column equilibrated in Tris-buffered saline (20 mM Tris-HCl (pH 7.5) and 150 mM NaCl); the column was washed with 10 volumes of Tris-buffered saline; and the glutathione S-transferase-I-domain fusion proteins were eluted with 10 mM glutathione in 50 mM Tris-HCl (pH 8.0). The proteins were then dialyzed against Tris-buffered saline and concentrated using a Microcon-3 microconcentrator (Amicon, Stonehouse, Gloucestershire, UK). The I-domains were checked for purity and degradation by 10% SDS-PAGE and Western blotting. Nitrocellulose blots were probed with horseradish peroxidase-conjugated anti-glutathione S-transferase polyclonal antibody.
Construction of Baculovirus Vectors Encoding Soluble ␣ 2 -Fos and ␤ 1 -Jun Integrin Subunits-Vectors containing the cDNAs of the human ␣ 2 and ␤ 1 integrins and the coding sequence of the human transcription factor Fos were purchased from American Type Culture Collection (Manassas, VA). The ␣ 2 ectodomain fused to the Fos dimerization helix was prepared by two-step PCR. First, a DNA fragment (3439 bp) encoding the ␣ 2 ectodomain, a 7-amino acid spacer (GGSTGGG), and a small segment of the Fos dimerization helix was obtained by PCR using ␣ 2 cDNA as the template and oligonucleotides 5Ј-ATATATAGATCCA-TGGATATGGGGCCAGAACGGACA-3Ј (forward primer with an NcoI site) and 5Ј-GGAGTGTATCAGTTAAACCGCCGCCCGTCG-ACCCTCCTGTTGGTACTTCGGCTTTCTC-3Ј (reverse primer). The Fos dimerization helix sequence was obtained by PCR using Fos cDNA as a template. The forward primer was antiparallel and complementary to the ␣ 2 ectodomain reverse primer defined above, and the reverse primer was AAGCGGCCGCTCTAGACGCGCGGATCCT-CAGT and contained an NotI site. In the second PCR step, both the ␣ 2 ectodomain and Fos dimerization helix fragments were templates, and the ␣ 2 forward and Fos reverse primers were used to obtain the whole ␣ 2 ectodomain fused to the Fos dimerization helix.
cDNA encoding ␤ 1 -Jun (which contains the ␤ 1 ectodomain fused to the transcription factor Jun dimerization helix) was prepared with a cloning strategy similar to that of Eble et al. (23). The Jun dimerization domain was first constructed using four oligonucleotides in overlapping PCR: Primer 1, TAGTTGGTCTGTCTCCGCTTGGAGTGTATCAG-TTAAACCGCCGCC; Primer 2, GCGGAGACAGACCAACTAGAA-GATGAGAAGTCTGCTTTGCAGACC, Primer 3, CTTCAGCAGG-TTGGCAATCTCGGTCTGCAAAGCAGACTTCTCATC; and Primer 4, ATTGCCAACCTGCTGAAGGAGAAGGAAAAACTAG- GPC(GPP) 5 a Fmoc-(2-Fmoc-oxy-4-methoxybenzyl)Gly-OH was introduced in the presence of an Asp-Gly sequence. b Fmoc-PP-(GPP) 4 GPC-TentaGel R RAM was introduced at the C terminus.  (23) using beads coated with GFOGER peptides (23). Platelet Adhesion-Platelet-rich plasma was prepared from fresh whole blood after two spins for 1 min at 1200 ϫ g. 10% (v/v) buffer containing 39 mM citric acid, 75 mM trisodium citrate, and 135 mM D-glucose (pH 4.5) and prostaglandin E 1 (280 nM final concentration) were added, and the platelet-rich plasma was spun for min at 700 ϫ g. The platelet pellet was resuspended in 6 ml of buffer containing 5.5 mM D-glucose, 128 mM NaCl, 4.26 mM Na 2 HPO 4 , 7.46 mM NaH 2 PO 4 , 4.77 mM trisodium citrate, 2.35 mM citric acid, and 0.35% bovine serum albumin (BSA; pH 6.5). Prostaglandin E 1 was added as before, and the platelets were spun for 6 min at 700 ϫ g. Platelets were resuspended to 1.25 ϫ 10 8 platelets/ml in adhesion buffer (0.05 M Tris-HCl, 0.14 M NaCl, and 0.1% BSA (pH 7.4)), and the adhesion of platelets from 100-l portions was determined colorimetrically as described (24). All peptides were coated at 10 g/ml on Immulon 2 HB 96-well plates (Thermo Electron Corp., Basingstoke, UK), at which concentration platelet binding was maximal, as verified for each of the major platelet-binding peptides (peptides 4, 7, and 31). Adhesion was allowed to proceed in the presence of the ␣ IIb ␤ 3 antagonist GR144053F (2 M); where indicated, platelets were preincubated with 2 mM Mg 2ϩ , 2 mM EDTA, or 2 g/ml anti-␣ 2 monoclonal antibody 6F1, which was used to prevent ␣ 2 ␤ 1 -dependent binding. Monoclonal antibody TS2/16 was applied to platelets for 15 min in the presence of 0.1 mM Co 2ϩ where indicated before the addition of platelets to the plate.
HT1080 Cell Adhesion-HT1080 human fibrosarcoma cells (obtained from the European Collection of Animal Cell Cultures, Porton Down, UK) were maintained in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum, 2 mM glutamine, 100 IU/ml penicillin, 100 g/ml streptomycin, and 2.5 g/ml amphotericin. Cells were harvested with trypsin/EDTA, washed, and suspended at 0.3 ϫ 10 6 /ml in Tris-buffered saline supplemented with 5.5 mM glucose and either 2 mM EDTA or 2 mM MgCl 2 . 100 l of the cell suspension was added to peptide-coated wells at 20°C for 20 min in the presence of antibody 6F1 where indicated. Adhesion (percent of the total number of added cells) was determined colorimetrically as described above for platelets. Cells of both low and high passage numbers gave similar results.
I-domain Binding-Adhesion of either purified ␣ 2 ␤ 1 or recombinant ␣ 2 I-domain (0.1 g in 100 l of adhesion buffer) was measured by incubation for 1 h at 20°C on peptide-coated surfaces. All recombinant I-domains were detected using rabbit anti-glutathione S-transferase antibody as described (7). Purified ␣ 2 ␤ 1 and recombinant ␣ 2 ␤ 1 were detected using mouse anti-␣ 2 I-domain monoclonal antibody (Serotec MCA743) at 1:2000 dilution and incubated at room temperature for 45 min. Wells were washed five times before the addition of horseradish peroxidase-conjugated anti-mouse IgG (Dako Corp.) at 1:5000 dilution and incubation at room temperature for 30 min. After washing, color was developed using an ImmunoPure 3,3Ј,5,5Ј-tetramethylbenzidine substrate kit (Pierce) according to the manufacturer's instructions.

RESULTS
Sequences and Quality of the Collagen III Toolkit-Peptides were synthesized as shown in Table 1. Matrix-assisted laser desorption ionization time-of-flight mass spectra confirmed the identity of the major component of all synthetic peptides and provided an assessment of purity. Two peptides (peptides 26 and 43) were estimated at 70 and 80% purity, respectively, with 11 others at ϳ85% and the remaining 45 being Ͼ95% pure. The method used for synthesis of each peptide (Table 1) reflects the difficulty of synthesizing the sequence in question and correlates quite well with the observed peptide purity. 55 of 57 peptides showed triple-helical structure as evidenced by sigmoidal polarimetric melting curves. Examples are shown in Fig. 1A. Two peptides (peptides 6 and 52) showed no melting transition despite having good purity. We believe that this reflects the influence of primary sequence on melting temperature, the subject of a separate study. Two other peptides (peptides 32 and 43) showed small changes in optical rotation (Fig. 1B), indicating limited assembly as a triple helix, this time reflecting the relative impurity of the peptide. The following motifs were synthesized between (GPP) 5 hosts to generate prospective integrin-binding peptides: GKDGES, GROGER, GLKGEN, GLOGEN, and GLOGEA. All of these showed good melting transitions (Fig. 1C).
Binding of the Wild-type ␣ 2 I-domain to the Toolkit- Fig. 2A shows binding of the recombinant ␣ 2 I-domain in the presence of 2 mM Mg 2ϩ or EDTA, which chelates divalent cations. The background response is indicated by wells coated with GPP10 or BSA. Fig. 2B shows Mg 2ϩ -dependent adhesion, which was observed in a few peptides, notably peptides 4, 31, 7, and 8. Peptides 5, 18, 30, 40, and 53 showed anomalous behavior, with adhesion greater in the presence of EDTA compared with Mg 2ϩ . We cannot account for this activity, but because it was caused by the removal of divalent cations, we do not consider that it represents any physiological adhesion process. A peptide containing the GFOGER motif was used as a positive control in these and all other experiments (9).
Binding of Constitutively Active and Inactive ␣ 2 I-domains to the Toolkit- Fig. 4A shows the binding of a constitutively active ␣ 2 I-domain construct, E318W (19), to the Toolkit peptides. E318W bound equally well to peptides 31, 4, and 7 and then in descending order to peptides 32,8,46,25,5,17,19, and 35. Anomalous behavior was observed with peptides 25, 17, and 35, which also supported adhesion in the presence of EDTA. By contrast, in a single experiment, the ␣ 2 I-domain mutant T221A, which binds minimally to collagen and GFOGER (25), showed no Mg 2ϩ -dependent adhesion to any Toolkit peptide (Fig. 4B). Platelet Binding to Toolkit Peptides-The interaction of platelets with collagen-derived peptides is likely to be more complex than that of the recombinant or purified proteins because several different receptors and modes of adhesion are known (26,27). Those mediated by GPVI, for example, are considered to be Mg 2ϩ -independent (28). To simplify this complex behavior for this study, we examined adhesion of human plate-lets to the subset of Toolkit peptides containing the GER triplet. The most prominent Mg 2ϩ -dependent adhesion was observed with several peptides, notably peptides 4, 31, 32, 7, 39, and 8 (Fig. 5A).
To examine the effect of increased integrin affinity while avoiding receptor-mediated stimulation of platelets, we treated human platelets with 0.25 mM Ca 2ϩ in the presence of 2 mM Mg 2ϩ to inhibit the activity FIGURE 3. A, Mg 2؉ -dependent binding of recombinant ␣ 2 ␤ 1 to Toolkit peptides; B, Mg 2؉ -dependent binding of ␣ 2 ␤ 1 purified from human platelet membranes. Experiments were performed as described under "Materials and Methods" and in the legend to Fig. 2. Two experiments were carried out using recombinant ␣ 2 ␤ 1 , and either one or two experiments (the latter denoted by the presence of error bars) were carried out using purified ␣ 2 ␤ 1 . ABS 450 , absorbance at 450 nm. of integrin ␣ 2 ␤ 1 (24) and then stimulated binding using the integrin ␤ 1 -subunit-specific antibody TS2/16 together with 1 mM Co 2ϩ . We observed a marked increase in binding to peptides 31, 4, 7, and 32, with a smaller effect on adhesion to peptides 8, 18, and 35. Adhesion to GFOGER was not enhanced under these conditions (Fig. 5B).
Binding to Prospective Novel Integrin Motifs from Collagen III-To progress from mapping integrin-binding sites in collagen to a precise definition of active motifs, we synthesized new shorter peptides, focusing on sequences of interest. We assumed that hexameric GXXЈGEXЉ sequences represent the minimal sequence involved, with glutamate in the 5th position being essential, based upon past experience in this and other laboratories (7, 9 -11) and especially upon the ␣ 2 I-domain- GFOGER co-crystal structure (4). All integrin-binding Toolkit peptides contain such glutamate residues. We speculated that four novel motifs might be responsible for integrin binding and synthesized peptides containing the candidate sequences (GKDGES, GROGER, GLKGEN, and GLOGEN) within (GPP) 5 flanking sequences to test the binding of integrins and cells as described above. Similar profiles of binding activity were obtained using recombinant ␣ 2 ␤ 1 , the active I-domain, E318W, and human platelets (Fig. 6, A-C), justifying this approach. The alternative, an alanine scan of each of the active Toolkit peptides, would be a less economical and, for peptides 4 and 8, which each contain 2 glutamate residues, an ambiguous approach. GROGER supported as much binding as peptide 4 in most of the assays, but was a less effective substrate compared with GFOGER. Similarly, GLOGEN was a good substrate. In marked contrast, GLKGEN supported limited adhesion, whereas GKDGES was inactive. To examine the role of the asparagine residue, a modified peptide containing the sequence GLOGEA was synthesized. This proved as effective as GLOGEN in supporting the adhesion of integrin preparations. Fig. 7 shows the ability of anti-␣ 2 antibody 6F1 (29) to inhibit purified ␣ 2 ␤ 1 binding to the four most active Toolkit peptides and to shorter derivatives. Similar inhibition was observed for platelet and HT1080 binding (data not shown). Table 2) in the fibrillar collagen ␣1(I), ␣2(I), and ␣1(III) chains. Apart from that, GROGER is found only in the human collagen ␣1(VII), Drosophila collagen ␣1(IV), and chick collagen ␣1(XIV) chains. Fibrillar collagens II, V, XI, and XXVII do not contain GROGER. Two other conserved loci in collagen I contain GKOGER, which may also have integrin-binding activity. Prospective GLXЈGEN sequences are less abundant in the collagens. GLK-GEN occurs at a locus in collagen III conserved across species, but only in human collagen III is it located next to a second and better integrinbinding motif, GLOGEN. GLOGEN itself occurs elsewhere only in the bovine collagen ␣2(I) chain. GLOGER, previously shown to bind the ␣ 1 ␤ 1 and ␣ 2 ␤ 1 integrins (9), occurs in human collagens I, II, and VII, whereas GLOGEXЉ is widely distributed, especially in collagen IV. These data are summarized in Table 2.

DISCUSSION
The peptide Toolkit enables us to produce comprehensive and exhaustive maps of the binding of cells and receptors within the triplehelical COL1 domain of human collagen III. It is important to note that this study does not address potential integrin binding either to the COL2-containing propeptide or to the short non-helical telopeptides of collagen III. A second limitation of our approach is that we have assumed in the design of the Toolkit that integrin-binding motifs will span a 6-amino acid motif, GXXЈGEXЉ, taking the ␣ 2 I-domain-GFOGER co-crystal as a model (4). The overlap between adjacent peptides of the Toolkit allows such binding motifs to be located with a resolution of ϳ9 amino acids within collagen III, or ϳ3 nm of the 300-nm length of its triple-helical COL1 domain. This compares well with the use of rotary shadowing to map receptor binding to the collagens (10,30,31).
The data from the Toolkit raised several analytical issues. First, given the number of independent ligands (57 peptides), the chances of obtaining false positive (Type I) errors was high. Second, because the amount of receptor preparation required to test all the peptides was also considerable, the number of replicates for each experiment was inevitably low, thereby increasing the chances of obtaining false negative (Type II) errors. To identify the Toolkit peptides of most interest, we ranked the binding results for each of the four experiments in which all peptides were tested: i.e. ␣ 2 I-domain, recombinant ␣ 2 ␤ 1 , purified ␣ 2 ␤ 1 and FIGURE 6. Adhesion of recombinant ␣ 2 ␤ 1 , E318W, and human platelets to peptides with the indicated guest sequences presented in (GPP) 5 flanking sequences as for the Toolkit peptides. Mg 2ϩ -dependent adhesion was determined in each experiment by subtracting adhesion in the presence of EDTA from that in its absence. Data were transformed using a power transformation such that the Levene statistic for homogeneity of variances was minimized. One-way analysis of variance with the Waller-Duncan post hoc test was performed in which the K ratio of Type I/Type II errors was set at 100. The Waller groupings are shown by the numbers on each individual chart. Data that fall within the same group are not significantly different from each other (p Ͻ 0.05). ABS 450 , absorbance at 450 nm. E318W mutant binding. The mean Ϯ S.D. of the four ranks of each peptide was calculated and plotted in rank order (Fig. 8). In addition, we generated 100 sets of pseudo data by assigning random ranks to 60 treatments (representing 57 Toolkit peptides and three controls) in four separate experiments and calculated the mean Ϯ S.D. of these. From this, we could estimate where our ranked data would lie if the null hypothesis (i.e. that all the peptides bound integrin equally well) were true. This line is also shown in Fig. 8. The analysis clearly revealed six peptides (GFOGER and peptides 4, 31, 7, 8, and 32) that behaved distinctly from all others. All contain glutamate residues, confirming their requirement for cation-dependent integrin binding, as we assumed at the outset. This analysis does not conclusively demonstrate that there are no biologically significant integrin-binding activities in the other peptides, but suggests that the data sets presented in this study are insufficient to justify any such conclusion.
The overlap region of peptides 31 and 32 (GMOGERGGL) contains the motif GMOGER, which we have shown recently to be a moderately active integrin-binding motif (7). The unique region of peptide 8 con-tains GAOGER, similarly shown to be a weak integrin-binding motif (7). Because the overlap region of peptide 8 lacks any glutamate or aspartate residue that could recognize a MIDAS, we consider that these past observations on GMOGER and GAOGER account largely for the activity of peptides 31, 32, and 8. It is interesting that peptide 29, which also contains the sequence GAOGER but does not bind integrin (discussed further below), is the site where a platelet inhibitory antibody binds to the epitope GLAGAOGLR (32). This antibody may act by blocking the active GMOGER motif, nearby in peptide 31, although explanations involving other receptors such as GPVI cannot as yet be excluded.
The Toolkit contains 14 separate GER motifs, four GEK motifs, and two GEN motifs, although GETGER in peptide 52 and GRNGEK in peptide 6 are most likely not helical. These are indicated in boldface type in Table 1. Integrin adhesion to all but a few of these potential binding sites was very modest, allowing many such motifs to be excluded from future investigations of ␣ 2 ␤ 1 binding.
Role of the X Residue in GXXЈGEXЉ-Wild-type ␣ 2 I-domain and ␣ 2 ␤ 1 preparations bind well to only four peptides, with minor activity for a few more. However, the active E318W I-domain displays significant binding to several more peptides. With the exception of peptide 4, all have a hydrophobic residue in the X position, consistent with our previous suggestion that, as for the Phe residue of GFOGER, the general role of the X residue is to interact with hydrophobic regions on the surface of the I-domain (3,7).
The unique sequence containing the activity of peptide 4 is AGK-DGESGROGROGERGL, within which lie two possible glutamate-containing MIDAS-binding motifs, GKDGES and GROGER. We have noted the presence of GROGER in collagen III, but have not hitherto tested it directly (7). GES motifs occur in three other peptides (peptides 22, 45, and 50), but none of these supports binding of any of the species tested. Furthermore, the sequence GPOGES, located within the collagen ␣2(I) chain at a site corresponding to GFOGER in the ␣1(I) chain, lacked integrin-binding activity when tested as a homotrimer, arguing against a general activity of GES motifs. 4 Another DGE motif (DGEA) located in cyanogen bromide peptide 4 from the collagen ␣1(I) chain has been proposed (33) and is widely used as a linear ␣ 2 ␤ 1 -binding peptide; but in triple-helical form, it lacks integrin-binding activity (9,17). We synthesized the motif GKDGES within (GPP) 5 hosts and were unable to detect adhesion of any cell or recombinant protein. For these reasons, we focused initially on GROGER and were able to demonstrate binding activity for a new triple-helical peptide containing that motif alone comparable with that of the parent peptide. This sequence represents a novel departure from the assumed requirement for a hydrophobic X residue in ␣ 2 ␤ 1 -binding motifs. GKOGER is also widely distributed in the collagens and is under investigation at present.
Role of the XЈ Residue-Our data suggest that Hyp in the XЈ position is favored, with Ser also supporting slight E318W binding in GLSGER (present in peptide 46), in line with our previous report (7). Positively charged side chains in the XЈ position appear to be unfavorable because neither peptide 37 nor 39 consistently binds integrin; however, our only direct test of either hypothesis lies in the comparison of GLKGEN and GLOGEN, where we cannot discriminate between a positive influence of Hyp and any negative effect of Lys. Hydrogen bonding of peptide side chains to the I-domain seems insufficient to support binding because Asn in the XЈ position (peptide 24) and Arg and Lys (peptides 16, 56, and 39) support poor binding activity, whereas GLOGEN is a better substrate compared with GLSGER, the latter binding only to activated integrins to any measurable extent. Previous evidence (4)   Role of the XЉ Residue-GER motifs in GMOGER and GROGER appear to account for much of the integrin-binding activity of collagen III. GMOGER has not been identified until recently because mapping of integrin-binding sites depended hitherto on the use of cyanogen bromide to define collagen fragments, and cyanogen bromide cleaves at methionine, thus destroying the recognition motif. We have previously reported reduced integrin binding to GEK motifs (9,11), but no GXXЈGEK motif tested here (Table 1) showed great activity. Nevertheless, peptide 19 contains GIOGEK, which remains to be tested further.
Peptide 7 is of considerable interest because it contains neither GER nor GEK, but rather tandem GXXЈGEN motifs in the sequence GLK-GENGLOGEN. Neither of these has hitherto been implicated in integrin binding. There is also a GET triplet at the start of the sequence (not repeated in peptide 6, which is non-helical) that could conceivably bind integrin. However, there are three other GET-containing peptides that form good helices in the Toolkit, peptides 25, 35, and 36, none of which has much activity. For this reason, we focused on GEN and synthesized three new peptides (GLKGEN, GLOGEN, and GLOGEA) within (GPP) 5 flanking sequences. GLKGEN had modest binding activity, whereas both GLOGEN and GLOGEA supported good Mg 2ϩ -dependent integrin binding, suggesting that the asparagine side chain is not involved in binding because alanine will serve instead. Thus, in the absence of a stabilizing contribution from electrostatic interactions between the I-domain and either the XЉ arginine or lysine, the binding emphasis shifts toward the first GXXЈ triplet, which may perhaps be able to adopt a position closer to the I-domain surface as the peptide rocks upon its glutamate fulcrum, allowing the XЈ leucine to form a closer interaction with the hydrophobic surfaces of the I-domain. The bending of the peptide revealed in the ␣ 2 I-domain-GFOGER co-crystal (4) is consistent with the idea that its N-and C-terminal interactions are strong enough to distort the rod-like triple helix: this distortion would be relieved in the absence of its C-terminal positive charge when the XЉ asparagine is substituted for arginine or lysine and would constitute a new means of integrin-binding activity. It is of interest that one of us recently reported an Arg/Lys-independent integrin-binding locus in collagen IX (30). This collagen contains the sequences GLOGEL and GLOGEI in its ␣1(IX) and ␣2(IX) chains, respectively, at the site identified. The activities we report here define the interaction of integrin ␣ 2 ␤ 1 with collagen III and may contribute to the understanding of how integrins recognize other collagens, especially collagen IX, as discussed above, and also collagen IV, which contains several other GLOGEXЉ motifs.
Role of Adjacent Sequence-An anomaly occurs in our data: the inability of peptide 29, which contains GAOGER, to support anything but the slightest binding of ␣ 2 ␤ 1 , when peptide 8, which also contains GAOGER, is active. Both peptides have high melting temperatures, suggesting good helix formation, and contain the sequences GDA-GAOGERGPO and GPRGAOGERGRO, respectively. Sequence comparison may suggest that electrostatic interactions occur between either of the nearby Arg residues (shown in boldface) in peptide 8 and the I-domain surface. No such interaction could be revealed in the integrinpeptide co-crystal (4) because the peptide used lacked collagen primary sequence adjacent to the GFOGER motif. Coincidentally, the inactive peptide 46 also contains a GPO motif next to GLSGER. Conceivably, the tighter triple helix engendered by the imino acid-rich environment may render these weak sequences less active. It is therefore important that adjacent sequence in peptides 8 and 46 be tested by substitution in further experiments and that additional co-crystal structures be produced to assess whether local changes in helix conformation that regulate longer range interactions with the I-domain occur.
Role of Integrin Activation-As outlined above, the inactive T221A mutant, in which the ability of the MIDAS to coordinate the cation is impaired, does not bind any Toolkit peptide, whereas the active E318W mutant binds positively to a wider range of peptides compared with the wild-type I-domain. HT1080 cells, which we consider to be constitutively active (7), and E318W bind to a similar peptide set. Our study with platelets supports the same conclusion; but in this case, the data are complicated by the presence of other collagen receptors, direct (such as GPVI and CD36) and indirect (such as ␣ IIb ␤ 3 and the glycoprotein Ib-V-IX complex). Analysis of this activity is beyond the scope of this work. To test the role of activation of platelet ␣ 2 ␤ 1 , we incubated human platelets with the integrin ␤ 1 -subunit-specific antibody TS2/16 in combination with Co 2ϩ . We observed increased binding to these same conserved Toolkit peptides. HT1080 cells, which express a higher constitutive binding activity compared with platelets (7), recognized low affinity sequences and bound a wider range of GER-containing Toolkit peptides, although some peptides remained inactive, providing negative controls.
These data lead us to propose that the collagens present an affinity series of integrin-binding motifs and that the cell will recognize these motifs depending on its activation state. Thus, GFOGER in collagen I can be recognized without cell activation, whereas other sequences (GLOGER and GASGER) will bind integrin according to the activity of the cell. In collagen III, two novel sequences (GROGER and GLOGEN) are presented with activity approaching that of GFOGER and not requiring cell activation. GMOGER is common to both collagens and has been reported previously (7). Thus, the resting platelet, flowing over a damaged vessel wall, may be captured by these high affinity sequences, allowing weaker interactions with GPVI that cause platelet activation (which might also be mediated by either ADP or thromboxane A 2 , both intermediates in important pathways in platelets). Hence, initial platelet adhesion leads through cell activation to increasingly tight association with collagen mediated by ␣ 2 ␤ 1 .
While this manuscript was in preparation, Kim et al. (31) reported GROGER as an integrin-binding motif in collagen III. However, GLK-GEN and GLOGEN were not identified, perhaps because they were too close to GROGER to be resolved by rotary shadowing, and GKDGES was not tested. (Moreover, a previous report (10) from the same authors of GLOGER as an integrin-binding motif in collagen I may similarly have failed to resolve by rotary shadowing this motif from the nearby GROGER in the collagen ␣1(I) and ␣2(I) chains.) The authors proposed that GROGER forms a hydrogen bond between the X arginine and Gln 215 in the ␣ 2 I-domain, which remains to be demonstrated crystallographically. They tentatively suggested GROGER to be a higher affinity substrate compared with GFOGER for the ␣ 2 I-domain. Our data are not consistent with this proposal because GFOGER was able to support better binding of all cellular or protein substrates we tested.
The present work has identified without ambiguity the key sequences in collagen III (GROGER, GLOGEN, and GMOGER) that bind ␣ 2 ␤ 1 , the first two of which represent novel integrin-binding motifs.