Definition of an Unexpected Ligand Recognition Motif for αvβ6 Integrin

Integrin interactions with extracellular matrix proteins are mediated by brief oligopeptide recognition sequences, and synthetic peptides containing such sequences can inhibit integrin binding to the matrix. The RGD peptide motif is recognized by many integrins including αvβ6, a specific receptor for fibronectin thought to support epithelial cell proliferation during wound healing and carcinoma progression. We report here the discovery of an unexpected non-RGD recognition motif for integrin αvβ6. We compared the recognition profiles of recombinant αvβ6 and αvβ3 integrins by using phage display screening employing 7-mer and 12-mer peptide libraries. As predicted, phages binding strongly to αvβ3 contained ubiquitous RGD sequences. However, on αvβ6, in addition to RGD- containing phages, one-quarter of the population from the 12-mer library contained the distinctive consensus motif DLXXL. A synthetic DLXXL peptide, RTDLDSLRTYTL, selected from the phage sequences (clone-1) was a selective inhibitor of RGD-dependent ligand binding to αvβ6 in isolated receptor assays (IC50 = 20 nm), and in cell adhesion assays (IC50 = 50 μm). DLXXL peptides were highly specific inhibitors of αvβ6-fibronectin interaction as synthetic scrambled or reversed DLXXL peptides were inactive. NH2- and COOH-terminal modifications of the flanking amino acids suggested that the preceding two and a single trailing amino acid were also involved in interaction with αvβ6. The DLXXL sequence is present in several matrix components and in the β chain of many integrins. Although there is as yet no precise biological role known for DLXXL, it is clearly a specific inhibitory sequence for integrin αvβ6 which has been unrecognized previously.

Integrins are a family of heterodimeric class I transmembrane receptors involved in numerous cell-matrix and cell-cell adhesion phenomena (1). They can be grouped roughly into three classes: the ␤1 series, which are ubiquitous receptors for extracellular matrix (2); the ␤2 series, which are activatable on leukocytes and are triggered during the inflammatory response (3); and the ␣v series, which bind and mediate the cell response to provisional extracellular matrices found during wound repair and other pathological processes (4).
The integrins ␣5␤1 (5), ␣IIb␤3 (6), ␣8␤1 (7), ␣v␤1 (8), ␣v␤3 (9), and ␣v␤6 (10) all bind the Arg-Gly-Asp-(RGD) peptide sequence in fibronectin, where it is presented in a constrained loop (11). Soluble RGD-containing peptides can inhibit the interaction of each of these integrins with fibronectin. However, to analyze the function of individual fibronectin receptors in a particular cellular environment it is useful to have more specific inhibitors, and a variety of inhibitory antibodies, modified peptides, and non-peptidic substances has been developed. However, as yet no inhibitor has been discovered specific for ␣v␤6. ␣v␤6 is a rare integrin, induced during repair processes in epithelia (10,12). Its only known specificities are for fibronectin (10), where it can be the dominant receptor mediating cell adhesion (13), and for tenascin (14). ␣v␤6 is believed to be involved in supporting the proliferation of epithelia during repair processes (15), and it can promote the proliferation of carcinoma cells (16).
Phage display technology has proved useful for identifying novel specific peptide sequences that act as ligand mimetics (17,18). Accordingly, both constrained cyclic and linear peptide libraries have been used to discover novel peptides that interact with integrins (19 -23). But, with few exceptions, these peptides contain RGD sequences, whereas the non-RGD sequences found have only bound weakly.
Here we have used recombinant ␣v␤6 expressed as a transmembrane truncated soluble receptor to screen phage libraries displaying peptides of 7 or 12 amino acids. This revealed a strong and previously unpredicted recognition motif for ␣v␤6 integrin which we describe here. Similar motifs are displayed in several extracellular matrix molecules.
␣v␤3 was purified in a soluble transmembrane truncated form from a baculovirus expression system as detailed previously (28) with minor modifications using 14D9.F8 antibody affinity chromatography (27). Briefly, the extracellular domains of ␣v and ␤3 human integrin chains were cloned into the pBacPAK expression system, the resulting recombinant baculoviruses containing both chains were used to coinfect High Five insect cells, and the soluble receptor was harvested from the culture supernatant at 48 -72 h of culture by passing the supernatant over 14D9.F8 antibody affinity columns, washing, and eluting at pH 3.1. Peak fractions were neutralized, concentrated, and dialyzed before shock freezing and storage at Ϫ80 C. The soluble human receptor (␣v␤3-⌬TM) had ligand binding specificities indistinguishable from the native receptor isolated from placenta (27).
␣v␤6 was purified in a soluble transmembrane truncated form (13) from a baculovirus expression system as detailed previously for ␣v␤3 (28) using 14D9.F8 antibody affinity chromatography (27). The ␤6 cDNA clone pCDNAneo␤6 was the generous gift of Dr. D. Sheppard (University of California, San Francisco). The procedure and the cloning will be detailed elsewhere. 1 In brief, the transfer vector pAcUW31 (CLONTECH Laboratories, Inc.) allowed simultaneous expression of two different target cDNAs and was used to make recombinant baculovirus-expressing transmembrane truncated ␣v␤6. The preparation of truncated ␣v transfer vectors was as described (28). Transmembrane truncated ␣v was excised from ␣v-⌬TM(pBac9) (28) using EcoRI and XbaI and cloned into the EcoRI site of pAcUW31 downstream of the polyhedrin promoter by blunt end ligation. Transmembrane truncated ␤6 cDNA was excised from pCDNAneo␤6 (13) using EcoRI and XbaI and cloned into the BamHI site of pAcUW31 downstream of the polyhedrin promoter by blunt end ligation. The tandem vectors containing truncated ␣v and truncated ␤6 were used to prepare recombinant baculovirus as described (28). The recombinant baculoviruses were used to infect High Five insect cells, and the soluble receptor was harvested from the culture supernatant at 48 -72 h of culture by passing the supernatant over 14D9.F8 antibody affinity columns, washing, and eluting at pH 3.1. All processes were carried out at room temperature and in the absence of detergents. Peak fractions were neutralized, concentrated, and dialyzed at 4°C before shock freezing and storage at Ϫ80°C. The recombinant soluble human receptor (␣v␤6-⌬TM) is biologically active and retains ligand specificity (13).
Peptides-Peptides were synthesized, purified, and analyzed inhouse as described (30) using an Fmoc (N-(9-fluorenyl)methoxycarbonyl) strategy with acid-labile side chain protection on acid-labile Wang resin using a commercially available continuous flow peptide synthesizer (Milligen 9050 plus). Purification was done by reversed phase HPLC and analysis by HPLC and fast atom bombardment-mass spectrometry. The peptide EMD 66203, cyclic(Arg-Gly-Asp-(D-Phe)-Val), is abbreviated as c(RGDfV).
Selection of Integrin-binding Phages-M13 phage display libraries displaying linear peptides of 7 or 12 amino acids (PHD system from New England Biolabs) were used to select integrin-binding phages. Panning was performed as described in the product manual with the following modifications. Integrins were diluted to 1.5 g/ml in TBSϩϩ (1 mM CaCl 2 , 1 mM MgCl 2 , 0.01 mM MnCl 2 , 150 mM NaCl, 50 mM Tris, pH 7.4) and adsorbed onto Petri dishes for 16 h at 4°C. After blocking for 2 h with 0.5% BSA in TBSϩϩ, aliquots of the phage display libraries (10 11 plaque-forming units) were diluted in TBSϩϩ containing 0.1% Tween 20, and phages were allowed to bind for 1 h at 30°C. Unbound phages were removed by washing with TBSϩϩ containing 0.3% Tween 20, and bound phages were eluted with 0.1 M glycine, pH 2.2, and amplified in Escherichia coli XL-1. Phages were prepared from the culture supernatant by standard polyethylene glycol/NaCl precipitation and used for the next round of panning. After the third round of panning randomly selected phage plaques were amplified, and singlestranded DNA was sequenced using the Amplitaq FS sequencing kit (Applied Biosystems) with the primer 96gIII (New England Biolabs).
Phage Binding Assay-Integrins were adsorbed onto 96-well plates as described above for the Petri dishes. Phages amplified from single plaques and purified by polyethylene glycol/NaCl precipitation were serially diluted in TBSϩϩ containing 0.1% Tween 20 and 0.1% BSA and allowed to bind to integrins for 1 h. After washing, bound phages were detected with an anti-M13 horseradish peroxidase-conjugated antibody (Amersham Pharmacia Biotech) and 3,3Ј,5,5Ј-tetramethylbenzidine substrate.
Integrin Adsorption Control-The adsorption of the integrins to the plates was investigated using indirect ELISA. Integrins were immobilized by adsorption on 96-well plates, and the plates were blocked with BSA as described under "Selection of Integrin-binding Phages" (above). 17E6 antibody, specific for the ␣v chain of all ␣v integrins (27), was biotinylated (28), titrated in phage diluent buffer (0.1% Tween 20, 0.1% BSA, TBSϩϩ), and allowed to bind (1 h, 37°C); after washing (0.3% Tween 20 and TBSϩϩ), bound antibody was detected with anti-biotin alkaline phosphatase conjugate (Bio-Rad) and p-nitrophenyl phosphate substrate.
Integrin-Ligand Competition Assays-Integrin-ligand competition assays were performed as described (30). Integrins were immobilized as described under "Selection of Integrin-binding Phages" (above). Serially diluted peptides in TBSϩϩ containing 0.1% BSA were added in parallel with biotinylated fibronectin, vitronectin, or fibrinogen (to 1 g/ml). After a 3-h incubation at 37°C and washing with TBSϩϩ, bound ligand was detected by incubation with an anti-biotin alkaline phosphataseconjugated antibody (Bio-Rad) followed by development with a p-nitrophenyl phosphate substrate. The reaction was stopped by the addition of NaOH and the color intensity read at 405 nm. The optimal time course for the integrin-ligand interaction was determined in preliminary experiments.
Inverted Integrin-Ligand Competition Assay-␣v␤6-⌬TM was biotinylated as described previously for matrix proteins and antibodies (28). Fibronectin was diluted (to 5 g/ml) in TBSϩϩ and immobilized by adsorption to 96-well plates for 16 h at 4°C as described under "Selection of Integrin-binding Phages" (above). After blocking for 2 h with 0.5% BSA in TBSϩϩ (BTBSϩϩ) and washing in TBS, biotinylated integrin (3 g/ml in BTBSϩϩ) was added in the presence of peptides serially diluted in the same buffer. After incubation (3 h, 37°C) and washing with TBSϩϩ, bound integrin was detected by incubation with anti-biotin alkaline phosphatase-conjugated antibody and color development as described above.
Cell Attachment Assays-Cell culture of lines M21-L and HT-29, and their use in cell attachment assays, has been described in detail elsewhere (27,31). M21-L is a human melanoma cell line selected for null expression of ␣v integrins (32) which uses only ␣5␤1 to attach to fibronectin. HT-29 is a human carcinoma cell line that uses only ␣v␤6 integrin to attach to fibronectin (33). The integrin profile of the cells was monitored routinely by fluorescence-activated cell sorter and showed the expected lack of ␣v integrins on M21-L and the lack of ␣5␤1 and ␣v␤3 on HT-29 (see also Fig. 6). 96-well plates were coated with fibronectin (12.5 g/ml) and blocked with BSA. Serially diluted peptides or antibodies were added followed by cells (2.5 ϫ 10 4 cells/well). After 1.5 h at 37°C nonadherent cells were washed away and attached cells detected using an assay measuring cellular hexosaminidase activity (31).
Recombinant ␣v␤6 expressed as a transmembrane truncated soluble receptor bound two distinct classes of phages in phage display panning experiments from a linear 12-mer library as determined from the sequences of more than 100 clones: those containing RGD sequences (51%) and those containing an XX-DLXXLX motif (27%). Some phages contained RGD and also continued with the motif as RGDLXXL (9%), whereas others displayed the sequence RGDL (38%); the remaining phages that were bound often contained DLXXL-related sequences. A selection of representative displayed sequences is shown (Table I).
In the non-RGD sequences, the amino acid distribution at X within X 1 X 2 DLX 3 X 4 LX 5 appeared nonrandom. Arg was favored at X 1 and X 5 , Thr/Ser/Asp/Gly at X 2 , whereas at X 3 X 4 Ser/Thr were often paired with a charged amino acid. These characteristics were typified by a dominant clone with the sequence RTDLDSLRTYTL (clone-1). Non-RGD, DLXXL-containing peptides were represented in only 5% of clones isolated from a 7-mer phage display library, suggesting that sequences COOHterminal to DLXXL are also involved in integrin binding (Table   I). Indeed, the sequence COOH-terminal to the peptide insertion site in the phage pIII protein continues Gly-Gly-Gly (i.e. the 7-mer library inserts would read XXDLXXLGGG), and related sequences were not isolated from the 12-mer library. Pro at X 3 X 4 was never found in a DLXXL motif, and it was similarly excluded from the 4 amino acids COOH-terminal of the DLXXL sequence, although not from these positions in RGD-containing sequences. One sequence was found where the motif was concatenated as GDLDLLKLRLTR. To investigate whether the presence of DLXXL sequences was a library artifact, we also screened on ␣v␤3. Here mainly RGD-containing phages were bound, the DLXXL sequence was absent, and RGDL was present in fewer than 10% of clones. This distribution was distinct from that of the ␣v␤6 phage display library screen (Table I) and similar to that reported for linear 15-mer library screens (23).
To eliminate the possibility that the differences in phage interactions seen between ␣v␤3 and ␣v␤6 were artifacts caused by differential adsorption of integrins during the phage screen, we estimated the amounts of each integrin adsorbed to the screening plates under the conditions of the screen using an indirect ELISA technique (Fig. 1B). The ELISA titration curves derived using the 17E6 antibody, which recognizes the ␣v chain, were similar in form, in saturation level of antibody binding, and in the amount of antibody needed to titrate 50% of the receptors. This indicated that similar amounts of the ␣v integrins were adsorbed to the plates.
To test the specificity of interaction of phage clones we used a quantitative ELISA to measure the binding of representative phages from the ␣v␤6 screen to immobilized purified integrins (Fig. 2). DLXXL-containing phages bound only to ␣v␤6, whereas RGD phages bound both ␣v␤6 and ␣v␤3 and weakly to ␣v␤5 and to ␣IIb␤3. This indicated that the XXDLXXLX clones isolated in the screen were specific for ␣v␤6.
RGD-containing peptides inhibit the interaction of RGDcontaining ligands with their integrin receptors. To test whether DLXXL peptides were able to inhibit integrin-ligand interaction, we synthesized the clone-1 peptide. We compared the effects of clone-1 peptide and RGD-containing peptides on ligands binding to integrins ␣v␤3, ␣v␤5, and ␣v␤6 (Fig. 3). The clone-1 peptide specifically inhibited fibronectin binding to ␣v␤6 with an IC 50 of 20 nM but did not affect ligand binding to ␣v␤3, ␣v␤5, or ␣IIb␤3 (IC 50 Ͼ 50 M). By contrast, the peptide GRGDSPK inhibited ligand binding to all four integrins with IC 50 Х 1 M. A cyclic peptide inhibitor c(RGDfV) showed specificity for ␣v␤3 (IC 50 ϭ 10 nM) over ␣v␤6 and ␣IIb␤3 (IC 50 Х 1 M). ␣v␤3 also binds to fibronectin. To test whether the effects of DLXXL peptides were a result of the ligand used with ␣v␤6, fibronectin, we also examined the effect of clone-1 peptide on fibronectin binding to ␣v␤3 (Fig. 4). The DLXXL peptide also had no effect on fibronectin binding, whereas the GRGDSPK peptide inhibited (IC 50 ϭ 200 nM). Thus, the results for vitronectin and fibronectin on ␣v␤3 were similar.
We also examined whether the assay geometry was producing an artifactual binding of DLXXL to ␣v␤6. In vivo, ␣v␤6 binds to insoluble fibronectin in the extracellular matrix rather than to the soluble form used in the receptor assay here. In addition, adsorption to plastic might change the conformation of the ␣v␤6 integrin and so alter its specificity. We therefore tested the effect of DLXXL peptides on ␣v␤6 binding to immobilized fibronectin (Fig. 5). Once again, the RTDLDSLRTYTL peptide strongly inhibited the ␣v␤6-fibronectin interaction (IC 50 ϭ 100 nM). Both c(RGDfV) and GRGDSPK peptides were also effective in this assay configuration. Thus, the effect of DLXXL peptide on ␣v␤6 was probably not an artifact of integrin adsorption to plastic. , and ␣v␤5 (solid triangles down) were adsorbed to 96-well plates from 1.5-g/ml solutions and blocked as for phage display screens and ligand competition assays. Biotinylated antibody 17E6 (27), serially diluted as indicated, was added, and after incubation and washing bound antibody was detected with anti-biotin-alkaline phosphatase conjugate. The non-␣v integrin control, ␣IIb␤3, gave background binding of 17E6 (A 405 Ͻ 0.05) (data not shown).
Clone-1 peptides were also tested in cell attachment assays. HT-29 carcinoma cell attachment to fibronectin was strongly inhibited by clone-1 peptide and by GRGDSPK (Fig. 6A). This attachment was also suppressed completely by the 17E6 (anti-␣v) antibody, showing that it was dependent on ␣v integrins. HT-29 attachment to fibronectin has been shown to be mediated by ␣v␤6. M21-L melanoma cells attached to fibronectin, and this was inhibited by GRGDSPK but was little affected by the clone-1 peptide (Fig. 6B). M21-L attachment was suppressed by either P4C10 (anti-␤1) or P1D6 (anti-␣5) antibodies, indicating that it was mediated by the ␣5␤1 integrin. Together, these data indicated that the peptide displayed by clone-1 was an active and specific inhibitor of ␣v␤6 independent of the RGD sequence.
We next investigated which elements of the X 1 X 2 DLX 3 X 4 LX 5 motif were important for its inhibitory activity. NH 2 -and COOH-terminal truncated forms of the peptide were synthesized and tested for their activity to block fibronectin binding to ␣v␤6. The data are summarized in Table II. Truncation of the COOH terminus of DLXXL had little effect on inhibitory activity until the group at X 5 was deleted, when activity diminished by 30-fold. Removal of the groups at X 1 X 2 also abolished the activity, indicating that the core motif was the 8-amino acid   sequence X 1 X 2 DLX 3 X 4 LX 5 .
Although the specific selection of DLXXL sequences from a highly degenerate (2 ϫ 10 9 clones) display library implies specificity of interaction, we examined this more directly by investigating the effect of reversed (TRLSDLDTR) and scrambled (LDTRTRLSD) peptides on ␣v␤6-fibronectin interaction. These peptides were Ͼ 4 orders of magnitude less active than the corresponding DLXXL peptide (Fig. 7).
The high specificity of the clone-1 sequence for ␣v␤6 suggested that it might represent a sequence in a native ␣v␤6 ligand. Indeed, a FASTA search of the GEMBL data bases revealed several extracellular matrix components with related consensus sequences (Table III) including fibrinogen ␥ chain, tenascin, laminin ␥1, ␤3, and ␣1, ␣2, and ␣3 chains. With the exception of tenascin, which has been reported to bind in an RGD-dependent way to ␣v␤6 (14), none of these molecules has been implicated previously as a ligand for ␣v␤6, thus the possible biological relevance of such homologies remains unknown.
Interestingly, the sequence DLYYLMDL is strongly conserved in human integrin ␤ chains and may interact directly with ligands, hinting that the clone-1 sequence might function by disturbing the interaction from the side of the receptor rather than as a ligand mimetic (e.g. like an RGD peptide). To test this possibility, we examined a synthetic DLXXL peptide sequence derived from the ␤6 chain, P 132 VDLYYLMDL 141 , for its effect on fibronectin-␣v␤6 interaction (Fig. 7). Compared with the corresponding 10-mer derived from clone-1, RTDLD-SLRTY, the ␤6 peptide was Ͼ 4 orders of magnitude less active as an inhibitor of this interaction. Once again, this supported the concept that the sequences selected from the phage display library were highly specific. We examined the structural basis   of the inhibition by exchanging the flanking sequences from the clone-1 10-mer with those of the ␤6 DLXXL peptide. Exchange of the DLYYL core gave a highly active inhibitor, as did the core and the trailing 3 amino acids. Thus the 2 amino acids preceding the DLXXL sequence were also important for integrin inhibitory activity. DISCUSSION Here, we describe the discovery by phage display screening of an unexpected binding sequence (DLXXL) for ␣v␤6 integrin, which has previously been thought to interact only with RGD sequences in its target ligands. ␣v␤6 is a rare inducible integrin (12,15), and isolation from native tissue sources has not been reported. ␣v␤6 has been prepared successfully by recombinant methods (13), and we have adapted this technology to produce biochemical amounts of soluble receptor. 1 As with ␣v␤3 (28), we found that soluble, recombinant ␣v␤6 retains the biological activity, specificity, and inhibitor profiles of the native receptor in situ, thus confirming and extending the studies of Sheppard and co-workers (13). Because the receptor is secreted into the serum-free culture supernatant, affinity purification gives rise to a very highly purified molecule with a minimal amount of other proteins 1 and thus is an excellent target for a phage display screen.
It is of interest that RGD-dependent extracellular matrix binding to ␣v␤6 is inhibited strongly by peptides containing DLXXL sequences that bear no close similarity to sequences in fibronectin, although they do resemble a sequence in tenascin. DLXXL sequences have rarely been found in other phage display studies on integrins; the only clone described in the literature bearing an RGDLXXL sequence was a weak inhibitor of ␣v␤3 (23).
We were able to demonstrate directly that synthetic DLXXL peptides, although strong inhibitors of ␣v␤6, had minimal effects on the interaction of integrins ␣v␤3, ␣v␤5, and ␣IIb␤3 with their ligands. Although indirect, cell adhesion assays implied that ␣5␤1 was also not affected. The DLXXL sequences, therefore, represent the first specific inhibitory peptides for ␣v␤6 integrin.
Analysis of the structures of phage-encoded ␣v␤6-binding XXDLXXLX peptides showed that although RXDLXXL was a favored sequence, it was by no means obligatory. Phages binding to ␣v␤6 were found where displayed arginine was replaced by hydroxylated, neutral, or basic residues or was absent. The inhibitory effect of DLXXL peptides is apparently not the result of nonspecific charge effects because neither reversed nor scrambled versions of highly inhibitory DLXXL peptides have significant effects on the ␣v␤6-fibronectin interaction. Further structural constraints on the context of DLXXL were found in experiments to investigate the possible biological significance of the ␣v␤6-DLXXL interaction. DLXXL sequences are found in seven integrin ␤ chains (Table III), in a region directly implicated in ligand binding (34). We hypothesized that the DLXXL motif was involved in the interaction of integrin ␣ and ␤ chains and that the DLXXL peptide might act by interacting with a binding site on the ␣ chain, usually occupied by ␤ chain sequences, so disrupting integrin function. To test this, we examined a 10-mer peptide corresponding to a DLXXL sequence in the ␤6 chain. It was inactive, indicating that the DLXXL peptides were not acting by competing for ␤6 chain binding sites on the ␣v chain. However, the study did reveal that the context of the DLXXL sequence was crucial because appropriate NH 2 -terminal flanking amino acids were critical for inhibitory activity, whereas those in the COOH-terminal flanking region were less crucial.
At present we have no indication where, if at all, the XX-DLXXLX sequences play a biological role with ␣v␤6 integrin. As discussed previously for the CRRETAWAC inhibitory peptide sequence for ␣5␤1 (21), XXDLXXLX may bind a site distinct from the RGD site and act as an allosteric inhibitor of RGD-ligand binding. But in the absence of kinetic data at present we believe that the XXDLXXLX sequences are ligand mimetics. Similar inhibitory mimetic binding sequences with no homology to known ligands are found in the phage display literature (35,36). The unusual strong and specific inhibitory binding sequences that we characterize here provide a unique pharmacological tool from which to investigate ␣v␤6 biological function further.