Structure-Function Analysis of Arg-Gly-Asp Helix Motifs in αvβ6 Integrin Ligands*

Data relating to the structural basis of ligand recognition by integrins are limited. Here we describe the physical requirements for high affinity binding of ligands to αvβ6. By combining a series of structural analyses with functional testing, we show that 20-mer peptide ligands, derived from high affinity ligands of αvβ6 (foot-and-mouth-disease virus, latency associated peptide), have a common structure comprising an Arg-Gly-Asp motif at the tip of a hairpin turn followed immediately by a C-terminal helix. This arrangement allows two conserved Leu/Ile residues at Asp+1 and Asp+4 to be presented on the outside face of the helix enabling a potential hydrophobic interaction with the αvβ6 integrin, in addition to the Arg-Gly-Asp interaction. The extent of the helix determines peptide affinity for αvβ6 and potency as an αvβ6 antagonist. A major role of this C-terminal helix is likely to be the correct positioning of the Asp+1 and Asp+4 residues. These data suggest an explanation for several biological functions of αvβ6 and provide a structural platform for design of αvβ6 antagonists.

Understanding of the molecular basis of the binding interface of integrins with their ligands still is relatively poor. The most detailed information available comes from x-ray crystallography of RGD 4 peptide binding to ␣v␤3, where the RGD motif bridged the ␣ and ␤ subunits, the arginine associating with the ␣v subunit and the aspartate coordinating with the bivalent metal ion on the ␤3 subunit (1). However, the RGD motif occurs in many extracellular matrix ligands so specificity is modified by other residues, often flanking the RGD site (2), although distant residues also can affect ligand binding (3). Detailed comparison of the different ligands of one integrin could illuminate the essential elements that determine specificity and affinity, improve biological understanding of integrinligand interactions, and allow rational design of targeting peptides and peptidomimetics (4 -6); we have made just such a comparison for ␣v␤6.
The integrin ␣v␤6 is an epithelial specific integrin that is expressed at low or undetectable levels in adult tissues but can be up-regulated during tissue remodeling. Thus increased ␣v␤6 expression occurs during wound healing, development and inflammation (7), and in more severe pathologies, including chronic skin wounds (8) and cancer (9). The ability of ␣v␤6 to promote migration and invasion, in part through protease upregulation (10 -12), may explain why ␣v␤6 expression is an independent indicator of colon cancer aggressiveness (13). These data suggest that reagents designed to specifically antagonize ␣v␤6 could have clinical utility in colon cancer and possibly other diseases.
␣v␤6 binds to the arginine-glycine-aspartate (RGD) motif in its ligands, which include fibronectin, tenascin, the latency associated peptides (LAP) of TGF␤1 (14) and TGF␤3 (15), and the VP1 coat protein of foot-and-mouth disease virus (FMDV) (16). In addition, the motif DLXXL was identified by phage display of 7-and 12-residue peptides (17) as an ␣v␤6-specific motif and was shown to be crucial to the ability of a 15-mer peptide to inhibit infection of baby hamster kidney cells by FMDV (18).
We have characterized specific peptide antagonists to ␣v␤6 and have discovered the basis of high affinity binding of natural ligands for this integrin. Such ligands (including the GH-loop of the VP1 protein of foot-and-mouth-disease virus and latency associated peptide of TGF␤1) possess a C-terminal post-RGD helix that positions leucine or isoleucine residues in the Asp ϩ1 and Asp ϩ4 positions as adjacent amino acids on the outer face of the helix, allowing for a potential hydrophobic interaction with ␣v␤6. Our data present a structural explanation for certain biological functions of ␣v␤6 and provide a structural platform for design of ␣v␤6 antagonists.

EXPERIMENTAL PROCEDURES
Cell Lines and Antibodies-Retroviral transduction generated ␣v␤6-positive, from ␣v␤6-negative, mouse NIH 3T3 fibro-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. □ S The on-line version of this article ( blasts (3T3␤6.19) as described previously (10). Recombinant soluble ␣v␤6 (rs␣v␤6) was purified from CHO␤6 cells, a kind gift from Dr. D. Sheppard (University of California, San Francisco), as described previously (19). VB6 is an ␣v␤6-expressing oral squamous carcinoma cell line (10). Antibody 10D5 to ␣v␤6 was from Chemicon International, (Harrow, UK). 6.3G9 (anti-␣v␤6) (20) was supplied by Biogen Idec. All other reagents were from Sigma unless stated otherwise. A375Ppuro and A375P␤6puro cell lines were generated from the human melanoma cell line A375P, which was infected with pBabe retroviruses encoding puromycin resistance or, in addition, cDNA for human ␤6, as described previously (10). Cells were selected in puromycin (1.25 g/ml) and then ␣v␤6-expressing cells were selected by magnetic bead sorting, using 10D5 (anti-␣v␤6; Chemicon International), according to the manufacturer's instructions (Invitrogen).
Cell Adhesion Assays-Adhesion of 51 Cr-labeled cells to 96-well flexible plates coated with extracellular matrix ligands has been described previously (10).
Enzyme-linked Immunosorbent Assay-80 ng/well rs␣v␤6, in 100 l, was immobilized on 96-well plates (18 h at 4°C). Plates were washed with Wash Buffer (20 mM Tris, 150 mM NaCl, 0.5 mM MgCl 2 , 1 mM CaCl 2 ) before blocking for ϳ2 h with Conjugate Buffer (1% casein (w/v) in Wash Buffer). All subsequent washes were with Wash Buffer. After washing, 100 l of biotinylated peptide was added to triplicate wells in Conjugate Buffer, bound for 1 h before washing and the addition of 50 l of ExtrAvidin horseradish peroxidase, diluted 1:500 in Conjugate Buffer. After 1 h, plates were washed and developed with TMBϩ (tetramethylbenzidine; DAKO; according to manufacturer's instructions). IC 50 values were generated using Graph-Pad Prism software; data shown represent mean Ϯ S.D. of three independent experiments.
Peptide Synthesis-20-mer peptides derived from the latency associated peptide of TGF␤1 (A20LAP, GFTTGRRGDLATIH-GMNRPF) and the GH-loop of VP1 protein of FMDV serotypes C-S8c1 (A20FMDV1, YTASARGDLAHLTTTHARHL; Ref. 18) and 0 1 BFS (A20FMDV2, NAVPNLRGDLQVLAQKVART (21)) were generated on an automatic synthesizer (model 431A updated and 433A solid phase synthesizer, Applied Biosystems) by the Cancer Research UK Peptide Synthesis Laboratory using standard procedures. Where stated peptides were biotinylated in situ on resin support using standard procedures.
Circular Dichroism-CD spectra were recorded (Jasco J-600 spectropolarimeter) at room temperature using 0.4 mM peptide in NMR buffers containing TFE between 0 and 50% (v/v). Each solution was loaded onto 5-mm path length quartz cuvettes, and spectra were obtained from an average of four scans (range between 190 and 260 nm), recorded 20 nm/min, 1 nm bandwidth with a 2-s response and a resolution of 0.2 nm. Spectra shown have no base-line correction. Spectropolarimeter OD values were converted into ellipticity and adjusted to the relative peptide concentrations by J-700 windows standard analysis (version 1.50.01) software. The maximal helicity ( max [] 222 ) was calculated by methods as described (22).
NMR Sample Preparation-All NMR samples were prepared to a final volume of 300 l for use in a Shigemi BMS005V NMR tube by dissolving purified freeze-dried peptide in 2 mM phos-phate-buffered saline, phosphate concentration of 25 mM and saline concentration of 100 mM. For structural studies, TFE was added to a final concentration of 30% (v/v). Saturation transfer difference NMR (STDNMR) samples were prepared with additional components as follows: 28 M integrin ␣v␤6, 0.5 mM Mg 2ϩ (MgCl 2 ), and 1.0 mM Ca 2ϩ (CaCl 2 ) but without TFE.
Structural and Saturation Transfer Difference NMR Spectroscopy-Experiments were recorded (Varian Unity INOVA 600 MHz NMR spectrometer) with a z-shielded gradient triple resonance probe. Structural experiments, at 10°C, included two-dimensional nuclear Overhauser effect spectroscopy (NOESY), total correlation spectroscopy (TOCSY), rotating frame Overhauser effect spectroscopy (ROESY), and double quantum-filtered correlated spectroscopy. Mixing times were 25.0, 70.0, and 100 ms for NOESY, TOCSY, and ROESY, respectively. Experiments were collected with 512/1024 and 2048 complex points with acquisition times of 64/128 and 256 ms in the indirectly and directly acquired 1 H dimensions. Slow exchanging amide protons were detected from the fingerprint region of a 50-ms mixing time two-dimensional NOESY experiment. Data processing analysis used Sun Silicon Graphics and Linux work stations with NMRPipe (23) to process and NMRView (24) to analyze spectra. Spin systems were identified using standard sequential assignment methods (25).
STDNMR experiments were run as described (26,27) but incorporating a 30 ms Hahn-echo filter (28). STD control and difference data were obtained at 25°C with 600-l sample volumes as follows: (a) A20-FMDV-2 (2 mM) ϩ ␣v␤6 (28 M); (b) A20-FMDV-2 (2 mM) ϩ mouse IgG (10 M); and (c) peptide GYQQGDYQQYNPD ϩ ␣v␤6 (28 M). All STD difference spectra had a spectral width of 6000 Hz and data points and transients of 8192 and 16,384, respectively. On resonance irradiation was set to Ϫ2.5 ppm and off resonance irradiation was set to Ϫ70.0 ppm with irradiation applied using a train of 9.4 ms Gaussian pulses, applied for 2.0 s, with each pulse having a 100 Hz bandwidth and separated by a 1.7-ms delay. STDNMR transfer data were assigned using peptide NMR assignments obtained with no TFE at 25°C. Resonance volume integrals were obtained using VNMR software (Varian Inc., Palo Alto, CA) Data were analyzed as outlined (27) to obtain the STD amplification factor using a ligand excess of 71.4. Individual amplification factors for each amino acid residue, from a sum of amplification factors from each 1 H resonance for each residue, were converted to percentage STD amplification factors to compare with the highest residue factor (assigned 100%).
Structural Calculations and Analysis-All structural calculations were obtained using Crystallography and NMR System (CNS) version 1.1 running on Silicon Graphics Octane2 and Transtec X2100 SuSE 8.1 Linux work stations (29). NOE and ROE contacts were classified between 2.5 and 5.0 Å with final structures calculated from extended coordinates using the standard CNS NMR anneal protocol with sum averaging for dynamic annealing with restraints from both extended and folded precursors. A final structural ensemble of 40 structures per peptide was produced with all structures used to produce statistical energy and r.m.s. deviation structural information. Backbone and heavy atom r.m.s. deviation values were obtained using MOLMOL version 2k.2 (30) on a PC running Microsoft Windows 2000. The structural integrity of each ensemble was evaluated using PROCHECK-NMR (31) run on a Transtec X2100 Linux work station. Energy comparisons between structure ensembles created in CNS were made using GROMOS96 43Bl parameter set (32) within DEEPVIEW version 3.7 (33).

RGDLXX(L/I) 20-mer Peptides Can Have Significantly Different Potencies
-In initial studies we tested a panel of 7-12-mer RGD peptides, derived from high affinity natural ligands of ␣v␤6 (foot-and-mouth-disease virus and latency associated peptide), for their ability to inhibit ␣v␤6-dependent cell adhesion. Results (not shown) revealed that the most potent peptides tended to be longer and all had an RGDLXXL or RGDLXXI motif. We therefore analyzed further three 20-mer peptides as follows: A20FMDV1 (YTASARGDLAHLTT-THARHL), A20FMDV2 (NAVPNLRGDLQVLAQKVART), and A20LAP (GFTTGRRGDLATIHGMNRPF). Each of these peptides exhibited a dose-dependent inhibition of ␣v␤6dependent cell adhesion using two different cell models. Thus, the IC 50 value for inhibiting ␣v␤6-dependent adhesion of 3T3␤6.19 for A20FMDV2 was 1.2 Ϯ 0.2 M, for A20LAP was 13.8 Ϯ 3.3 M, and for A20FMDV1 was 86.5 Ϯ 49.9 M (Fig. 1a). We obtained a similar response using VB6 cells (supplemental Fig. 1). We also observed that there was a dose-dependent adhesion of biotinylated peptides to immobilized recombinant ␣v␤6 that followed the same order of activity; thus, 50% maximal binding to rs␣v␤6 occurred at 0.5 Ϯ 0.15 nM for A20FMDV2, 1.4 Ϯ 0.93 nM for A20LAP, and 6.8 Ϯ 3.49 nM for A20FMDV1. These data indicate that A20FMDV2 has the highest affinity of binding to ␣v␤6 followed by A20LAP and then A20FMDV1 (Fig. 1b), the same order of activity as shown for their behavior as antagonists.

RGDLXX(L/I) Peptide Potency Correlates with Helical
Propensity-Although all three peptides possessed the previously identified ␣v␤6-specific RGDLXX(L/I) motif (17), it was not apparent why there should be such large differences in potency. Logan et al. (21) had noted that if a disulfide bond was reduced in the GH-loop of the VP1 protein of FMDV O 1 serotype, from which A20FMDV2 was derived, it was then able to form a helix in the post-RGD region. Therefore, we analyzed the sequences of our peptides in the helix-predictive software Agadir and found that A20FMDV2 and A20LAP were predicted to form helices and that the probability of this was greatest for A20FMDV2; A20FMDV1 was not predicted to form a helix (Fig. 2a). To confirm the ability of these peptides to form helices, we performed far-ultraviolet CD analysis in increasing concentrations of the helix-promoting solvent TFE (34). Fig. 2b confirms, by the presence of ellipticity at 222 nm, that there is a strong propensity for A20FMDV2 to form helices followed by A20LAP; A20FMDV1 formed a helix but only weakly. Thus experimental results confirmed the in silico predictions by Agadir software. To determine the extent and position of the helix in each of the peptides, we conducted structural analysis of these peptides using NMR spectroscopy.
NMR Analysis of ␣v␤6-binding 20-mer Peptides-Using the lowest concentration of TFE required to support helix formation (30% TFE), we performed structural NMR on all three pep-tides. NMR solution structures were determined using 1 H chemical shift assignments for each peptide (supplemental Table 1). Through-space assignment data were from two-dimensional NOESY and ROESY spectra in 30% TFE (v/v). Specific NMR experiments highlight 1 H nuclei proximal in space (typically separated by less than 6 Å) by showing an NOE or ROE contact between these nuclei. These contacts provide the major restraints used to calculate particular structural models. Additional restraint data from the assignment of hydrogen bond donors and dihedral angles were obtained using NMR NOESY data and high resolution double quantum-filtered correlation spectra, respectively (supplemental Table 2). Distributions of NOE and ROE contacts were more numerous between nuclei in residues on the C-terminal side of the RGD for each peptide. Fig. 3 summarizes contact types and additional restraints that indicate standard helix formation directly C-terminal to the RGD motif. Fig. 4 highlights the main helical contact regions of the NOESY spectra for all three peptides (the number of contacts and resonance dispersion are the greatest with A20FMDV2 and the least with A20FMDV1). FIGURE 1. a, ␣v␤6-dependent adhesion is inhibited with 20-mer peptides derived from ␣v␤6 ligands. Inset shows that binding of 3T3␤6.19 cell to LAP is abrogated entirely by 6.3G9 (␣v␤6-blocking antibody) versus control antibody W6/32 (anti-major histocompatibility complex class 1). Radiolabeled 51 Cr-3T3␤6.19 in various concentrations of peptides A20LAP, A20FMDV1, or A20FMDV2 was added to 96-well plates coated with 50 l (0.25 g/ml) LAP. b, 20-mer peptides bind with varying affinities to rs␣v␤6. Biotinylated peptides were added to plates coated with rs␣v␤6. After 1 h, unbound peptide was removed by washing, and bound peptide was detected with ExtrAvidinhorseradish peroxidase. Peptides did not bind significantly to control (casein)-coated wells (data not shown). Data show the mean Ϯ S.D. of triplicate wells and are representative of three separate assays with similar results. MARCH 30, 2007 • VOLUME 282 • NUMBER 13

JOURNAL OF BIOLOGICAL CHEMISTRY 9659
Structure Calculations and Analysis-All structures were determined using CNS. None gave violations greater than 0.2 Å or bond angle violations greater than 5°. Structural restraint data were incorporated to generate the structural ensembles ( Fig. 5, a, d, and g) and average structures (Fig. 5, b, e, and h) and defined the structural limits of the ␣-helices. NMR and structural calculations confirmed the observations from the CD data. Residue i-j contacts, as shown for NH-NH in Fig. 3, identify restraints that make all three peptides adopt a turn conformation that enables the RGD motif to be presented at the turn of a hairpin structure. NOE contacts used to confirm the turnhelix arrangement were completely unambiguous. We thus excluded many potentially ambiguous NH-␣H turn contacts from i-i ϩ 2 residues that could artificially influence calculated structures. However, the turn was still capable of being defined from a combination of short (i-i ϩ 2 and i-i ϩ 3) and long range NOE side chain contacts as can be seen from NOE contact maps (supplemental Fig. 7). This turn is well defined and can be simulated from short range NOEs (i-i ϩ 3 or less) alone, excluding the long range NOEs from the calculation. These data provide assurance that this structural characteristic is not forced by the analysis. Helix-associated residues for each peptide were Leu 10 -Val 17 for A20FMDV2, Leu 10 -Gly 15 for A20LAP, and Ala 10 -Thr 14 for A20FMDV1. 3 J HNHa values less than 5 Hz were used to restrain for that residue to Ϫ60 Ϯ 30°because 3 J HNHa values obtained by double quantum-filtered correlation spectroscopy always are larger than those obtained by more accurate heteronuclear NMR methods (25). Structural energy statistics and backbone r.m.s. deviations are shown in supplemental Table 3. Backbone r.m.s. deviations are quoted over residues DLXX(L/I)XX (supplemental Table 3). PROCHECK-NMR analysis of the 40-structure ensembles identified 94.3, 94.8, and 93.6% of residues that fell in the allowed regions of the Ramachandran plot for A20FMDV1, A20FMDV2, and A20LAP, respectively. Residues outside the allowed regions were from the first four amino acids in each ensemble. Helix limits and the turn definitions for each peptide structure were identified from a combination of dihedral angles, hydrogen bond geometries, and NOE contacts. This approach proved to be invaluable for cases where overlap from chemical shift degeneracy prevented complete unambiguous NOE assignment that was particularly apparent in the NOE data from the A20FMDV peptide. Thus both the length and definition of the helix increases in the order A20FMDV1 Ͻ A20LAP Ͻ A20FMDV2, which correlates with peptide potency and further supports the CD and Agadir data. The data also predict that the residues at Asp ϩ1 and Asp ϩ4 are exposed almost in apposition on one face of the helix (represented in Fig. 5, c, f, and i), particularly in A20FMDV2 and A20LAP; these residues are implicated strongly in the efficacy of peptide inhibitors of ␣v␤6-dependent functions (17,18).
A Post-RGD Helix Increases Affinity and Potency of RGDLXX(L/I) Peptides-To confirm the role that a post-RGD helix had for binding to ␣v␤6, we generated A20DV1217, a derivative of A20FMDV2 that replaced the two L-valines at positions 12 and 17 with D-valines. This substitution was designed to maintain the biochemical nature and charge distribution over the length of the peptide while completely disrupting the helix. Analysis by CD (supplemental Fig. 2) and solution NMR (supplemental Fig. 3, a and b) confirmed that A20DV1217 was unable to form a stable helix. In flow cytometry assays with biotinylated peptides, more than 1000-fold concentrations of A20DV1217 were required to obtain similar fluorescence signals on A375P␤6puro cells compared with A20FMDV2 (Fig. 6a). Thus the affinity of binding of the A20FMDV2 to cellular ␣v␤6 is significantly higher than the helix-deficient A20DV1217. In inhibition of cell adhesion assays, the concentration of A20DV1217 required to inhibit 50% ␣v␤6-dependent adhesion by 3T3␤6.19 cells, compared with A20FMDV2, was 48.5 Ϯ 37.0 M versus 1.2 Ϯ 0.2 M, respectively, a 40-fold reduction in activity (Fig. 6b). A similar result was seen with ␣v␤6-dependent adhesion of VB6 cells (supplemental Fig. 4). These data confirm that a post-RGD helix promotes the affinity of peptides for ␣v␤6 and their potency as ␣v␤6 antagonists.
STDNMR Suggests Peptide Ligands for ␣v␤6 Bind in Helical Form-Our data indicate that the presence of an RGDLXX(L/I) motif is not sufficient for high affinity binding to ␣v␤6 (compare A20FMDV1 with A20FMDV2; see Fig. 1, a and b). The studies of Mateu et al. (18) and Kraft et al. (17) identified, in addition to the RGD motif, the Asp ϩ1 and Asp ϩ4 residues as critical in peptide efficacy as ␣v␤6 antagonists. In contrast single-residue substitution studies did not identify any essential residues C-terminal to the LXXL motif (18). Thus we speculated that the role of the extended helix in A20FMDV2 and A20LAP is to stabilize the RGD-proximal helix region, ensuring the appropriate presentation of the leucines/isoleucines in the Asp ϩ1 and Asp ϩ4 positions. Structural NMR had confirmed that these residues are appropriately located on one external face of the post-RGD helix for such an interaction to occur. Thus, if this hypothesis is correct, residues at Asp ϩ1 and Asp ϩ4 would necessarily be physically close to the integrin to allow for direct interaction with ␣v␤6. To investigate this question, we performed saturation transfer determination NMR (26,27). This technique combines receptor and peptide ligand and ana-lyzes the NMR signal transfer from the receptor to residues in the ligand. Transfer signals generally imply a separation distance from the receptor of Յ6 Å, and the higher the transfer signal, the smaller the separation distance compared with other residues in the ligand. In addition, STDNMR has the significant advantage of being conducted in physiological buffer unlike solution NMR, described above, which required nonphysiological TFE. In the histograms shown in Fig. 7 (derived from spectral data shown in supplemental Figs. 5 and 6), the RGD residues have been highlighted in yellow and the Asp ϩ1 and Asp ϩ4 residues have been highlighted in red. It can be seen that the Asp ϩ1 and Asp ϩ4 residues usually gave the highest signal of any of the residues in the RGDLXXL motif indicating very close proximity to the surface of ␣v␤6. These data suggest that the Asp ϩ1 and Asp ϩ4 residues actually bind to ␣v␤6, presumably through a hydrophobic interaction. Because the Asp ϩ1 and Asp ϩ4 residues are not adjacent, these data are also consistent with the likelihood that the peptide is in a helical form when bound to ␣v␤6 in a physiological setting. Moreover, as shown in the space-fill models (Fig. 5, d-f) that highlight the RGD (yellow) and Asp ϩ1 and Asp ϩ4 (red) residues, there is a common binding interface for all three peptides; the A20FMDV2 and A20LAP peptides form a similar stereochemical arrangement of the key residues, whereas A20FMDV1 is less well structured, again correlating with affinity and potency.
The histograms in Fig. 7 show also that there are additional residues, outside of the RGDLXXL region, exhibiting significant transfer signals. Most of these additional signals are C-terminal for A20FMDV2 and A20LAP. These data suggest that additional interactions are occurring along the length of the post-RGD helix. Leu 6 is highlighted in the STDNMR results for A20FMDV2 in Fig. 7a. This is because of 1 H chemical shift overlap making data from this residue merge with that from Leu 10 and Leu 13 . Because previous analyses have not revealed that a leucine preceding the RGD motif is critical either in ␣v␤6 specificity (17) or FMDV infection (18), it is unlikely that Leu 6 is significantly involved in the interaction with ␣v␤6 as compared with Leu 10 and Leu 13 . The STDNMR data for A20FMDV1 (Fig.  7c) show a relatively even dispersion of transfer signals across the whole of the molecule. These results are consistent with functional and binding data showing that this peptide binds relatively weakly and thus is interacting with ␣v␤6 in multiple conformations.
Note that the RGD motif in all contact maps (Fig. 5) appears in different orientations as these residues occur at the turn in the peptides with few NOEs defining their position. The position of the aspartate residue is well defined as it occurs at the N terminus of the helix, the most structured part of the peptides. However, the structures calculated show that arginine and glycine have a high degree of variation. Additionally, as there are few structural restraints observed for the N termini of the peptides, a similar degree of structural variation also applies to these regions, and such variability must be considered when viewing the structural models. Within the RGD motif, the arginine gave a higher transfer signal than the aspartate residue, presumably because the aspartate is acting to coordinate a cation (1), and thus, compared with arginine, proton contacts will be more distant from the integrin. In fact, the RGD transfer signal is lower than that compared with transfer signals determined for many other peptide residues in accordance with that observed from the STDNMR of cyclic peptides associating with ␣IIb␤3 on platelets (35).
To confirm that the STDNMR signals were derived from a real integrin-ligand interaction, we performed several controls. First, we examined STDNMR signals from various control combinations (A20FMDV2 (2 mM) ϩ IgG (supplemental Fig.  6d), A20FMDV2 (2 mM) ϩ carbonic anhydrase (30 M; data not shown), and an unrelated peptide GYQQGDYQQYNPD ϩ ␣v␤6 (28 M; supplemental Fig. 6e)). In no case were significant transfer signals detected. Second, we repeated the STDNMR signals with A20FMDV2 (supplemental Fig. 6, a--c) but suspended the ␣v␤6 in 20 mM EDTA to chelate the cations that are essential for integrin-ligand binding. Supplemental Fig. 6 shows that EDTA essentially abrogates saturation transfer signals. Because cations are essential for RGD binding, the ability of EDTA to inhibit peptide binding completely indicates that the RGD region of the peptide must bind first. Thus the proposed hydrophobic binding of the post-RGD helix, which would be cation-independent, acts to stabilize this interaction because functional experiments above show that the helix increases affinity and potency of the peptides.

DISCUSSION
The expression of the integrin ␣v␤6 is significantly up-regulated on many carcinomas compared with the corresponding nontransformed tissues where expression is usually undetectable (reviewed in Ref. 9). This differential expression, together with accumulating experimental (10 -12) and clinical data (13), suggests that ␣v␤6 is promoting cancer progression and identifies this integrin as a promising candidate for imaging and therapy of carcinoma. However, there are no clinically useful reagents available for investigating these avenues. We therefore began by developing ␣v␤6-specific peptides designed from high affinity ligands of ␣v␤6, namely latency associated peptide of TGF␤1 (14) and the GH-loop of the VP1 protein of foot-and-mouthdisease virus (16). Our lead peptides, which were all 20-mers (as shorter peptides were much less effective; data not shown), each include the motif RGDLXX(L/I). The sequence DLXXL had previously been identified in peptide phage-display studies as an ␣v␤6-specific motif (17), and in earlier studies, this same motif was required for the ability of peptides to inhibit FMDV infection (18). However, our data showed that there were large differences in potency of the lead 20-mers indicating that the presence of the residues RGDLXX(L/I) was insufficient to predict the behavior of an antagonist. As the peptides were linear, initially we did not consider secondary structure as a possible explanation. In fact, in the context of native FMDV, the RGD-containing GH-loop is unstructured in solution (21,36). However, Logan et al. (21) had reported that if a disulfide bond was reduced in the GH-loop of the VP1 protein of FMDV O 1 serotype, from which our most potent peptide (A20FMDV2) was derived, it was then able to form a helix in the post-RGD region.
We therefore investigated whether our linear peptides possessed secondary structure. Far-UV circular dichroism analysis confirmed that all of the lead peptides had varying degrees of helical propensity and that this propensity correlated with their potential as antagonists of ␣v␤6. Structural NMR revealed the location of the helix within the peptide and showed that all the peptides possessed a common structure, a hairpin with RGD at the tip followed immediately on the C-terminal side by the helix. NMR also showed that the extent of the helix correlated with peptide potency, the longest helix being in the most potent peptide A20FMDV2. The structural data generated here for the A20FMDV2 peptide are supported strongly by earlier x-ray crystallography studies of the same sequence in the reduced GH-loop of FMDV O 1 that identified a post-RGD helix (21). Thus the potential of three linear peptides to form secondary structures that included a post-RGD helix seemed to be an important component of their efficacy. This was confirmed by analysis of A20DV1217, a peptide conservatively designed to be an almost identical biochemical match of A20FMDV2 but which lacked the propensity to form a helix. In flow cytometry 1000-fold higher concentrations of biotinylated A20DV1217 were required to give signals of the same levels as biotinylated A20FMDV2. These data suggest that the affinity of binding of A20FMDV2 to ␣v␤6 is significantly increased by the presence of the post-RGD helix, and this was reflected also in a dramatic drop in efficacy as an antagonist.
Earlier studies (18) identified the leucines at positions Asp ϩ1 and Asp ϩ4 as critical for peptide inhibitors of FMDV infection. We examined the DLXX(L/D)LXXI region of the NMR structures and noted that the presence of a post-RGD helix caused the normally nonadjacent pair of leucines in A20FMDV2 (leucine/isoleucine in A20LAP) to come into juxtaposition on the outer face of the helix (Fig. 5, c and  f). The residues at these positions are highly conserved in FMDV (18,37), and it is likely that they contribute to the binding. As A20FMDV1 is less well structured (Fig. 5h), the positions of the Asp ϩ1 and Asp ϩ4 leucines are not likely to be as close to each other as the corresponding residues in A20FMDV2 and A20LAP, as represented in Fig. 5i. The likelihood that the Asp ϩ1 and Asp ϩ4 residues are involved in binding to ␣v␤6 is supported by our STDNMR analyses. By extrapolation from previous x-ray crystallography studies examining the binding of RGD peptides to ␣v␤3, we can assume that both the arginine and aspartate residues of the RGD motif in our peptides are participating in binding to the integrin. The high energy transfer to Asp ϩ1 and Asp ϩ4 residues indicates a very close proximity Ͼ6 Å to the ␣v␤6 surface, closer in fact than either the arginine or aspartate residues in A20FMDV2 and A20LAP. The STDNMR data therefore also support the likelihood that the peptides bind as RGD-helix structures to ␣v␤6. This conclusion is supported also by previous reports that, when bound to infection-blocking antibodies, fragments of the GH-loop formed a post-RGD helix (38 -40).
Although the STDNMR data indicate significant amounts of energy transfer to residues in the C terminus, it is not possible to conclude from our data whether the peptides form additional contacts with ␣v␤6 along the length of the helix or whether the role of a more stable C-terminal helix is required only for optimal presentation of the Asp ϩ1 and Asp ϩ4 residues in the RGDproximal helix. Certainly the observation that EDTA inhibits both peptide binding and STDNMR signals indicates that the primary binding event is the binding of the RGD motif to the integrin. As the helical propensity enhances affinity and potency of peptides, together these data show that one important function of the helix is to stabilize this RGD interaction,  diagrams (b, e, h, and c, f, i) are of a single structure from the ensemble that is closest to the mean structure calculated for each peptide from the ensemble data. b, e, and h show a side view with the RGD residues in ball-and-stick mode. c, f, and i show the end-on view and the Asp ϩ1 and Asp ϩ4 residues are in ball-and-stick mode. All figures were created in MOLMOL 2k.2 (30).
probably through a hydrophobic interaction of the Asp ϩ1 and Asp ϩ4 residues with the surface of the ␤6 subunit.
Our data, together with the work of others (38 -40), suggest that when the GH-loop binds to its natural receptor, ␣v␤6 (16,41), a helix forms in this loop stabilizing this interaction. We can therefore predict that a post-RGD helix will also form when LAP binds to ␣v␤6, thereby contributing to an important biological function of ␣v␤6. Integrin ␣v␤6 can activate latent TGF␤ by binding to LAP and, through actin-dependent trac-tion forces, expose the TGF␤ molecule to its receptors (14). Our data suggest that a post-RGD helix in LAP would serve to enhance binding affinity and provide the high stability of interaction with ␣v␤6 required for this physical activation of TGF␤.
In summary, we have defined the structural basis of high affinity binding of physiological ligands to ␣v␤6, an RGD motif at the tip of a hairpin loop followed immediately by a C-terminal helix. Structural NMR analyses showed also that the highly conserved leucine or isoleucine residues at Asp ϩ1 and Asp ϩ4 positions in the binding site are presented as adjacent residues on the exterior face of the loop, and STDNMR showed that these residues bind very closely to the ␣v␤6 surface, possibly through a hydrophobic interaction with the integrin. These structurally determined conclusions are supported by the recent findings of Burman et al. (37), who predicted a similar conclusion based on the use of peptides to block virus infection. Furthermore, a combination of STDNMR and the use of a helix-deficient mutant peptide confirmed the strong likelihood that, when bound, ␣v␤6 interacts with an RGD-helix motif in its high affinity ligands. These data provide a structural explanation for the necessity of the DLXXL motif (17) for binding to ␣v␤6 and establish a structural template upon which ␣v␤6targeting antagonists can be designed. In addition, these data raise the possibility that structural motifs flanking integrinbinding sites in other ligands may also participate in determining ligand affinity and potency. . The post-RGD helix is required to improve peptide affinity and potency. Flow cytometry was used to measure binding of biotinylated peptides to A375P␤6puro (a, c, and e) or A375Ppuro (b, d, and f). a and b, unfilled histograms, IgG control; black solid histograms, 10D5 (mouse anti-␣v␤6). c and d, A20FMDV2 at 1 nM (unfilled histograms), 10 nM (gray filled histograms), and 100 nM (black filled histograms). e and f, A20DV1217 at 100 nM (unfilled), 1 M (gray filled), and 10 M (black filled). Note that A20DV1217 requires more than 1000-fold more peptide than A20FMDV2 to achieve the same level of binding to A375P␤6puro. g, radiolabeled 51 Cr-VB6 cells were added to 96-well plates coated with 50 l (0.25 g/ml) LAP in various concentrations of peptides A20FMDV2 or A20DV1217. Binding to bovine serum albumin-coated wells and binding in the presence of 10 g/ml ␣v␤6-blocking antibody 63G9 were not significant (data not shown). Data shown are from one experiment using triplicate samples and are representative of three separate experiments with similar results. Data show a significant reduction in potency as an antagonist to ␣v␤6-specific cell adhesion compared with A20FMDV2. Because the only significant difference between these two peptides is the absence of a helix in A20DV1217, these data highlight the importance of the helix. FIGURE 7. Analysis of peptide binding to rs␣v␤6 using 1 H STDNMR spectroscopy. STDNMR was performed using 70 M excess of A20FMDV2 (a and d), A20LAP (b and e), or A20FMDV1(c and f) mixed with 1 mg of high pressure liquid chromatography-purified (Ͼ95% purity) rs␣v␤6 in phosphate-buffered saline supplemented with 1 mM Ca 2ϩ and 0.5 mM Mg 2ϩ . Histograms a-c show absolute energy transfers, represented as a percentage for each residue, normalized to the residue with the highest transfer energy. The RGD motif is highlighted in yellow and the Asp ϩ1 and Asp ϩ4 residues are highlighted in red. d-f show the corresponding CYK space-fill diagrams of the peptides predicted from solution NMR with the RGD and Asp ϩ1 and Asp ϩ4 residues highlighted. Note that A20FMDV2 and A20LAP have a similar binding face.