Identification of an α3β1 Integrin Recognition Sequence in Thrombospondin-1*

A synthetic peptide containing amino acid residues 190–201 of thrombospondin-1 (TSP1) promoted adhesion of MDA-MB-435 breast carcinoma cells when immobilized and inhibited adhesion of the same cells to TSP1 when added in solution. Adhesion to this peptide was enhanced by a β1 integrin-activating antibody, Mn2+, and insulin-like growth factor I and was inhibited by an α3β1 integrin function-blocking antibody. The soluble peptide inhibited adhesion of cells to the immobilized TSP1 peptide or spreading on intact TSP1 but at the same concentrations did not inhibit attachment or spreading on type IV collagen or fibronectin. Substitution of several residues in the TSP1 peptide with Ala residues abolished or diminished the inhibitory activity of the peptide in solution, but only substitution of Arg-198 completely inactivated the adhesive activity of the immobilized peptide. The essential residues for activity of the peptide as a soluble inhibitor are Asn-196, Val-197, and Arg-198, but flanking residues enhance the inhibitory activity of this core sequence, either by altering the conformation of the active sequence or by interacting with the integrin. This functional sequence is conserved in all known mammalian TSP1 sequences and in TSP1 from Xenopus laevis. The TSP1 peptide also inhibited adhesion of MDA-MB-435 cells to the laminin-1 peptide GD6, which contains a potential integrin-recognition sequence Asn-Leu-Arg and is derived from a similar position in a pentraxin module. Adhesion studies using recombinant TSP1 fragments also localized β1 integrin-dependent adhesion to residues 175–242 of this region, which contain the active sequence.

Expression of the ␣ 3 ␤ 1 integrin is essential for normal development in the kidney and lungs (1). Targeted mutation of the murine ␣ 3 integrin gene resulted in abnormal branching morphogenesis of kidney capillary loops and lung bronchi. Based on antibody inhibition, this integrin may also be important for branching morphogenesis in mammary epithelia (2). In addition to its essential roles in normal development, the ␣ 3 ␤ 1 integrin may play important roles in disease processes, such as cancer. Loss of integrin ␣ 3 subunit expression is a negative prognostic factor in lung adenocarcinoma (3). Conversely, overexpression of ␣ 3 ␤ 1 integrin in a human rhabdomyosarcoma line suppressed tumor formation in mouse xenografts (4).
The ␣ 3 ␤ 1 integrin has been reported to recognize several extracellular matrix ligands, including some laminins, type IV collagen, fibronectin, thrombospondin-1, and entactin/nidogen (5)(6)(7)(8). Although short peptide recognition motifs have been identified in ligands for some integrins (reviewed in Ref. 9), previous attempts to define recognition sequences for binding of matrix ligands to the ␣ 3 ␤ 1 integrin have produced conflicting results. High affinity binding of recombinant soluble ␣ 3 ␤ 1 could be detected only to laminin-5 (10), so binding to other matrix ligands may be of relatively low affinity. Under specific conditions, this integrin can recognize the common integrin binding sequence RGD in fibronectin (6). However, recombinant entactin with the RGD sequence deleted (11) and synthetic peptides from laminin-1 and type IV collagen that lack the RGD motif (12,13) also bound specifically to the ␣ 3 ␤ 1 integrin. Laminin peptide GD6 (KQNCLSSRASFRGCVRNLRLSR) and the type IV collagen peptide affinity purified ␣ 3 ␤ 1 integrin from cell extracts when immobilized on agarose beads (12,13), but the active peptides from these two proteins share no apparent sequence homology. These data, combined with the evidence that RGD-dependent and RGD-independent adhesion are differentially regulated in ␣ 3 ␤ 1 integrin (6), have led to the proposal that the ␣ 3 ␤ 1 integrin uses distinct mechanisms to interact with each of its ligands and that no conserved binding motif may exist (6).
We recently found that ␣ 3 ␤ 1 is the major TSP1 1 -binding integrin on several human breast carcinoma cell lines (14). We have further examined this interaction and report the identification of a peptide sequence from TSP1 that supports ␣ 3 ␤ 1dependent adhesion and chemotaxis and is a potent inhibitor of adhesion to TSP1.

EXPERIMENTAL PROCEDURES
Proteins and Peptides-Calcium replete TSP1 was purified from human platelets (15). Synthetic peptides containing TSP1 sequences were prepared as described previously (16 -21). Recombinant fragments (provided by Dr. Tikva Vogel) and GST fusion proteins expressing fragments of TSP1 (provided by Dr. Jack Lawler, Harvard University) were prepared as described previously (22,23). Bovine type I collagen and murine Type IV collagen were obtained from Becton Dickinson Labware division, and human vitronectin was from Sigma. Fibronectin was purified from human plasma (National Institutes of Health Blood Bank) as described (24). Murine laminin-1 purified from the EHS tumor was provided by Dr. Sadie Aznavoorian (NCI, National Institutes of Health). Recombinant human insulin-like growth factor-1 (IGF1) was from Bachem.
Motility Assays-Chemotaxis of MDA-MB-435 cells to TSP1 peptides was measured in modified Boyden chambers using polylysine-coated 8 m polycarbonate filters as described previously for intact TSP1 (14).
Multiple Sequence Alignment-Protein sequences were compared using MACAW software (National Center for Biotechnology Information, National Library of Medicine, version 2.0.5) by the segment pair overlap and Gibbs sampler methods (27,28).

RESULTS
In initial attempts to localize the region of TSP1 recognized by the ␣ 3 ␤ 1 integrin, we tested approximately 85% of the TSP1 sequence in the form of synthetic peptides or GST or T7 fusion proteins for promotion of ␤ 1 integrin-dependent adhesion of MDA-MB-435 cells (Fig. 1). Among the recombinant fragments tested, only an 18-kDa fragment of the N-terminal heparinbinding domain had significant adhesive activity, although the recombinant type I repeats had adhesive activity for MDA-MB-435 cells in some experiments (results not shown). A recombinant GST fusion of the type 3 repeats of TSP1 including the RGD sequence had minimal adhesive activity for MDA-MB-435 cells (Fig. 1), in contrast to human melanoma cells, which avidly attached on substrates coated with the same concentrations of this fragment (26). The ␤ 1 integrin-activating antibody TS2/16 did not enhance cell attachment to any of these recombinant fragments but reproducibly stimulated attachment on intact TSP1 (Fig. 1). Synthetic heparin-binding peptides from the type 1 repeats (peptide 246) (16) and the CD47-binding peptide 4N1K (21) also promoted adhesion, but TS2/16 did not enhance adhesion of MDA-MB-435 cells to these peptides. CD36-binding peptides from the procollagen domain (peptide 500) or the type 1 repeats (Mal-II) (29) had weaker adhesive activities and were also insensitive to TS2/16. The focal adhesion disrupting peptide Hep1 from the N-terminal domain of TSP1 (20) did not promote MDA-MB-435 cell adhesion. Although these experiments did not detect a ␤ 1 integrin-dependent adhesive sequence in TSP1, the possibility remains that these regions of TSP1 contain a conformation-dependent recognition motif that is inactive in the recombinant fusion proteins due to misfolding.
A multiple alignment search using MACAW software was used to identify TSP1 sequences that might be related to the ␣ 3 ␤ 1 integrin-binding GD6 peptide derived from the A chain of murine laminin-1 (12), which strongly promoted MDA-MB-435 cell adhesion ( Fig. 2A). This search identified four TSP1 sequences related to the laminin peptide (Table I). The single peptide identified by the Gibbs sampler method, derived from the C-terminal domain of TSP1 (residues 1059 -1077), did not support adhesion or inhibit adhesion of MDA-MB-435 cells to TSP1 or other ␣ 3 ␤ 1 integrin ligands ( Fig. 2A and results not shown). Because a synthetic peptide containing the last 12 residues of peptide GD6 (peptide 679, Table I) had similar activity to the intact peptide (see below), we did not test the two peptides identified by segment pair overlap that aligned outside this sequence. Both of these peptides were derived from regions of the type 1 (residues 392-405) or type 2 (residues 598 -608) repeat sequences that did not support ␣ 3 ␤ 1 -dependent adhesion when expressed as GST fusion proteins (Fig. 1). The remaining sequence is from a region of the N-terminal domain of TSP1 (residues 188 -199) that was not covered by the recombinant fragments tested in Fig. 1 and conserves most of the hydrophilic residues in the laminin-1 GD6 peptide that could mediate protein-protein interactions. This sequence also overlaps with a region identified in a screen of N-terminal TSP1 peptides as having heparin-independent adhesive activity (30). A synthetic peptide containing this TSP1 sequence (peptide 678) had strong adhesive activity for MDA-MB-435 cells ( Fig. 2A). Spreading of two breast carcinoma cell lines on this peptide, laminin peptide GD6, and TSP1 was enhanced in the presence of the ␤ 1 integrin-activating antibody TS2/16 ( Fig.  2B). We previously found that IGF1 strongly stimulated ␤ 1 integrin-mediated adhesion to TSP1 (14). IGF1 similarly stimulated spreading of MDA-MB-435 cells on the TSP1 peptide 678 and to the laminin peptide GD6 (Fig. 2B).
The TSP1 peptide 678 strongly inhibited spreading of MDA-MB-435 cells on TSP1 and murine EHS tumor-derived laminin-1/entactin but did not inhibit spreading of the same cells on the ␣ 5 ␤ 1 integrin ligand fibronectin or on type IV collagen (Fig. 3A). The TSP1 peptide in solution strongly inhibited MDA-MB-435 cell attachment to itself and to GD6 (Fig. 3B), a known ␣ 3 ␤ 1 integrin-binding peptide from murine laminin-1 (12). In contrast, the laminin peptide was a relatively weak inhibitor of adhesion to either peptide or TSP1 when tested in solution (IC 50 ϭ 700 M, data not shown).
The TSP1 peptide 678 sequence was not in the recombinant N-terminal fragment tested in Fig. 1, but the previously reported 28-kDa N-terminal fragment of TSP1 contains this sequence (22). Adhesion assays using MDA-MB-231 (Fig. 3C) and MDA-MB-435 breast carcinoma cell lines (data not shown) verified that the larger fragment, expressing residues 1-242, contains a ␤ 1 integrin-dependent adhesion sequence that is not present in residues 1-174. Adhesion to the longer fragment was stimulated by the ␤ 1 -activating antibody TS2/16 and inhibited by peptide 678 (Fig. 3C). Therefore, a ␤ 1 integrinbinding site is present in residues 175-242 of TSP1 and is functional when expressed as a recombinant protein.
To verify that the TSP1 peptide 678 contains an ␣ 3 ␤ 1 integrin recognition sequence, integrin ␣-subunit antibodies were tested for blocking adhesion to the peptide (Fig. 4). The ␣ 3specific blocking antibody P1B5, which we have shown to inhibit adhesion of the same cells to intact TSP1 (14), partially inhibited adhesion of MDA-MB-435 cells on peptide 678 and completely reversed the enhancement of MDA-MB-435 cell adhesion to the same peptide stimulated by the ␤ 1 integrinactivating antibody TS2/16. In a further control experiment, the ␣ 2 ␤ 1 -blocking antibody 6D7 inhibited adhesion of MDA-MB-435 cells to type I collagen but not to peptide 678 (Fig. 4B). Function-blocking antibodies for ␣ 4 ␤ 1 and ␣ 5 ␤ 1 integrins also had no effect on adhesion to peptide 678 (data not shown). Therefore, the peptide does not support adhesion mediated by ␣ 4 ␤ 1 or ␣ 5 ␤ 1 integrins or inhibit adhesion to other integrin ligands.
Divalent cation dependence is also characteristic for binding of integrin ligands. Although Mn 2ϩ but not Ca 2ϩ induced the expected increase in MDA-MB-435 cell spreading on TSP1 peptide 678 and intact TSP1 (Fig. 4C), addition of EDTA only minimally inhibited basal spreading on peptide 678. EDTA completely inhibited the spreading on TSP1 observed in medium containing Mg 2ϩ as the sole divalent cation, although it did not inhibit cell attachment on TSP1 ( Fig. 4C and results not shown). This residual adhesion probably results from the significant contribution of proteoglycans to adhesion of MDA-MB-435 cells on TSP1 (14). Spreading on peptide 678 with Mg 2ϩ as the divalent cation became partially sensitive to EDTA, however, in the presence of the ␤ 1 -activating antibody TS2/16. Addition of Mn 2ϩ further stimulated spreading on peptide 678  and intact TSP1 in the presence of TS2/16, but addition of Ca 2ϩ produced a dose-dependent inhibition of spreading on both substrates. Specific inhibition by Ca 2ϩ is consistent with previous data for the ␣ 3 ␤ 1 integrin (31). These results suggest that integrin binding to peptide 678 is partially independent of divalent cations, but MDA-MB-435 cell spreading on this peptide may involve both ␣ 3 ␤ 1 integrin binding and divalent cationindependent interactions with another cell surface molecule.
Truncated peptides that contained portions of peptide 678 were synthesized to identify essential residues (Fig. 5). Truncation of the N-terminal Phe or the C-terminal Val-Phe only moderately decreased adhesive activity, but further truncations from either end of the peptide greatly diminished its activity. Inhibition assays confirmed that the loss of adhesive activity reflected loss of integrin binding rather than loss of ability to adsorb on the substrate (Table II). As found in the direct adhesion assays, peptides without the N-terminal Phe or the C-terminal Val-Phe retained significant inhibitory activities, but all shorter peptides were weak inhibitors or inactive. These results imply that the integrin recognizes an extended sequence, but this approach could not discriminate conformational effects of flanking sequences from a direct contribution to integrin binding.
To better define those residues involved in ␣ 3 ␤ 1 integrin binding, we systematically substituted Ala residues into the peptide 678 and tested each for adhesive activity (Fig. 6). Based on the complete loss of adhesion activity for MDA-MB-435 cells following its substitution, only Arg-198 was essential for adhesive activity of peptide 678 (Fig. 6). Replacement of Arg-198 with a His also dramatically reduced adhesive activity. Ala substitutions at several other positions significantly decreased adhesive activity, except for the two N-terminal residues, which only slightly decreased adhesive activity.
Although only the Arg residue was essential for direct adhesion, substitution of several additional residues with Ala markedly decreased or abolished inhibitory activity of the respective soluble peptides in solution to block ␣ 3 ␤ 1 -dependent adhesion to immobilized peptide 678 (Table III). These experiments showed that Arg-198, Val-197, and Asn-196 are essential for inhibitory activity of the peptides in solution. Substitution of Phe-199 and Phe-201 decreased the inhibitory activities of the respective peptides 5-8-fold, indicating that these flanking residues also contribute to activity of the peptides in solution. In contrast, peptides with Ala substitutions at four of the six N-terminal residues in this sequence had inhibitory activities equivalent to that of the native TSP1 sequence. Therefore, NVR is the essential sequence for binding to the ␣ 3 ␤ 1 integrin, but flanking residues may be necessary for inducing the proper conformation of this minimal sequence in peptide 678.
The specificity for an Arg residue at position 198 was further examined using conservative amino acid substitutions (Table  III) side chain with hydrogen bonding ability is required at this position.
The active peptides strongly promoted formation of filopodia in MDA-MB-435 cells (Fig. 7A) similar to those induced by attachment on intact TSP1 (14). Addition of IGF1 enhanced spreading and increased formation of lamellipodia on the same peptide (Fig. 7B). Phallacidin staining demonstrated organization of F-actin at the cell periphery but no organization of stress fibers across the cell body (Fig. 7C). Using antibodies recognizing vinculin (Fig. 7D) and focal adhesion kinase (data not shown) as markers of focal adhesion formation, we could not detect any induction of focal adhesions in MDA-MB-435 cells attaching on these peptides, although the same markers showed typical focal adhesion staining patterns in the cells when attaching on vitronectin or fibronectin substrates (results not shown). Staining for the ␣ 3 ␤ 1 integrin was punctate and prominently localized in filopodia extended by MDA-MB-435 cells on immobilized peptide 678 (Fig. 7F), whereas total ␤ 1 integrin staining was more diffuse and concentrated over the cell body.
TSP1 stimulates chemotaxis of MDA-MB-435 cells, and this response is inhibited by the ␣ 3 ␤ 1 -blocking antibody P1B5 (14). Peptide 678 also stimulated chemotaxis of MDA-MB-435 cells (Fig. 8). Chemotaxis to peptide 678 was dose-dependent with a maximal response at 10 M (Fig. 8A). This response was specific in that peptide 690 was inactive. In agreement with the observations that IGF1 stimulated ␤ 1 integrin-dependent chemotaxis of MDA-MB-435 cells to TSP1 (14) and adhesion of the same cells to peptide 678 (Figs. 2 and 7), the chemotactic response of MDA-MB-435 cells to peptide 678, but not to peptide 690, was increased in the presence of IGF1 (Fig. 8B). DISCUSSION Based on examination of synthetic peptides and recombinant fragments representing approximately 90% of the TSP1 sequence, only the sequence FQGVLQNVRFVF from the N-terminal domain exhibited activities that are expected for an ␣ 3 ␤ 1 integrin binding sequence in TSP1. A recombinant fragment of TSP1 containing this sequence also promoted ␤ 1 integrin-dependent adhesion. In solution, this peptide specifically inhibited adhesion to TSP1 but not to ligands recognized by other integrins. Adhesion to this peptide and to TSP1 was stimulated by IGF1 receptor ligands that stimulate integrin-dependent    adhesion to intact TSP1, by Mn 2ϩ , and by a ␤ 1 integrin-activating antibody and partially inhibited by an ␣ 3 ␤ 1 functionblocking antibody. Based on systematic amino acid substitutions in the active sequence, NVR appears to be the essential core sequence in this TSP1 peptide for recognition by the ␣ 3 ␤ 1 integrin.
Adhesive activities of the immobilized peptides imply that only Arg-198 may directly participate in this interaction, although the partial resistance to inhibition by an ␣ 3 ␤ 1 integrin antibody and EDTA suggest that the peptides with Arg may also support adhesion independent of integrin binding. The context surrounding the Arg is important, however, because other peptides with similar sequences (such as peptide 701, with a QVRT sequence) had no activity, and Ala substitutions of the flanking residues in peptide 678 eliminated or markedly decreased its inhibitory activity in solution. The essential amino acid residues are completely conserved in human, murine, bovine, and Xenopus TSP1, but in chicken TSP1, a His replaces the Arg. A similar motif is found in murine and human TSP2, with a His residue replacing the Arg. As a free peptide, the TSP1 sequence with a His substitution was much less active, so it is not clear whether the TSP2 sequence could be recognized by ␣ 3 ␤ 1 integrin. Activity of the latter sequence may be increased in an environment that increases protonation of the imidazole in His.
Previous publications have not identified a consensus ␣ 3 ␤ 1 integrin recognition sequence in its ligands. One hypothesis is that different ligands have unrelated binding sequences, which is supported by a recent mutagenesis study (32). However, other recent data have raised questions about whether all of the proteins reported to mediate ␣ 3 ␤ 1 -dependent adhesion are true ␣ 3 ␤ 1 ligands (33). LamA2 and LamA3 were verified to bind ␣ 3 ␤ 1 integrin. These have potential binding motifs based on our data, but human LamA1, which was found not to bind ␣ 3 ␤ 1 with high avidity, has an Ala in the position occupied by the essential Arg in the TSP1 sequence. Substitution of Ala for the Arg in the TSP1 sequence abolished all activity of the synthetic TSP1 peptide. Among the five G domain modules of LamA3, G2 has a better consensus sequence based on our results (NLK) than does G4 (NFQ) or G5 (NIH). Expressed as recombinant proteins, only the G2 module promoted ␣ 3 ␤ 1 -dependent adhesion (34). Although RGD was reported to be an ␣ 3 ␤ 1 ligand, the RGD in entactin is not required for recognition, and the RGD in the type 3 repeats of TSP1 is not recognized by this integrin. A binding site for the ␣ 3 ␤ 1 integrin in entactin was mapped to the G2 domain (residues 301-647) (11). Multiple alignment of this region of entactin against the TSP1 sequence and the murine laminin-1 peptide GD6 identified a related sequence, FSGIDE-HGHLTI, but this sequence lacks all of the essential residues in the TSP1 sequence. This domain of entactin also contains two NXR sequences: NNRH and NGRQ. It remains to be determined whether either of these can function as an ␣ 3 ␤ 1 integrin recognition sequence.
The absence of an Asp residue in peptide 678 may account for its partial independence of divalent cations. An Asp residue is usually considered an essential element for integrin peptide ligands (35,36). According to one model for integrin ligand binding, the divalent cation participates directly in binding an Asp-containing peptide ligand (reviewed in Ref. 37). Thus an integrin peptide ligand without a carboxyl side chain cannot coordinate with a bound divalent cation and therefore may not have a divalent cation requirement for binding to the integrin. The alternate model, proposing an indirect role of divalent cations in integrin activation (37), would be consistent with the observed stimulation of cell spreading on peptide 678 by Mn 2ϩ but not Ca 2ϩ and the partial inhibition following chelation of divalent cations.
Another interpretation of the partial divalent cation independence for the adhesive activity of peptide 678 is that ionic interactions of the Arg side chain in the TSP1 peptide with the negatively charged cell surface contribute to the adhesive activity of this peptide. Weak ionic interactions could promote adhesion to the immobilized peptide through multivalent interactions with negatively charged glycoproteins and proteoglycans on the cell but would not significantly contribute to binding of the same monovalent peptide to the cell in solution. This hypothesis would explain why the Arg-containing peptides 686 and 691, in which the essential Val or Asn residues were substituted with Ala, lacked activity in solution to inhibit adhesion to ␣ 3 ␤ 1 ligands but retained some adhesive activity when immobilized. Thus, inhibitory activities in solution may provide a more reliable assessment of integrin binding specificity for Arg-containing peptides.
Spreading of MDA-MB-435 breast carcinoma cells on intact TSP1 (14) or the ␣ 3 ␤ 1 integrin-binding peptide 678 induces formation of filopodia. In cells plated on peptide 678, these structures are enriched in the ␣ 3 integrin subunit, suggesting that engagement of this integrin by TSP1 triggers formation of filopodia. Formation of filopodia or microspikes has been noted during attachment of other cell types on TSP1 (38). This response may be mediated by the ␣ 3 ␤ 1 integrin, because lamellar spreading rather than formation of filopodia was typically observed on melanoma cells that predominantly use the ␣ v ␤ 3 integrin receptor for spreading on TSP1 (26). Using multiple sequence alignment, the N-terminal domains of thrombospondins were recently shown to contain a module related to pentraxins and to the G domain modules of laminins (39). Based on this alignment, both the ␣ 3 ␤ 1 integrin-binding sequence from TSP1 identified here and the GD6 sequence of laminin are located at the C terminus of a pentraxin module. The known three-dimensional structures of other members of the same superfamily (40,41) lead to the prediction that both potential integrin binding sequences are located in the last ␤-strand of a pentraxin module and therefore may be presented with similar topologies on the laminin G domain and the Nterminal domain of TSP1. This observation suggests an evolutionary relationship between the thrombospondin N-terminal domains and laminin G domains that is consistent with their proposed common function as recognition sites for a ␤ 1 integrin receptor.