Vascular Endothelial Growth Factor A (VEGF-A) Induces Endothelial and Cancer Cell Migration through Direct Binding to Integrin α9β1

Integrin α9β1 mediates accelerated cell adhesion and migration through interactions with a number of diverse extracellular ligands. We have shown previously that it directly binds the vascular endothelial growth factors (VEGF) A, C, and D and contributes to VEGF-induced angiogenesis and lymphangiogenesis. Until now, the α9β1 binding site in VEGF has not been identified. Here, we report that the three-amino acid sequence, EYP, encoded by exon 3 of VEGF-A is essential for binding of VEGF to integrin α9β1 and induces adhesion and migration of endothelial and cancer cells. EYP is specific for α9β1 binding and neither requires nor activates VEGFR-2, the cognate receptor for VEGF-A. Following binding to EYP, integrin α9β1 transduces cell migration through direct activation of the integrin signaling intermediates Src and focal adhesion kinase. This interaction is biologically important because it mediates in vitro endothelial cell tube formation, wound healing, and cancer cell invasion. These novel findings identify EYP as a potential site for directed pharmacotherapy.

The integrins are a diverse family of transmembrane heterodimeric adhesion receptors. They mediate outside-in and inside-out signaling through cell interactions with the extracellular matrix and in turn facilitate numerous processes such as cell adhesion, migration, and proliferation (1,2). ␣9␤1 forms a small subfamily of integrins with ␣4␤1 and plays a specialized role in accelerated cell migration, for example more robust compared with ␣3␤1, ␣V␤3, and ␣5␤1 (3). Its biologic role was unclear until genetic deletion of the integrin in mice revealed a novel phenotype manifest by malnutrition and chylothoraces resulting in death 10 -12 days after birth (4). Ultimately, this phenotype was found to result in part from abnormal development of lymphatic valves (5) but also through disrupted direct ␣9␤1-VEGF-C/D interactions (6).
The VEGF family of growth factors mediates many separate but overlapping functions, including inflammation and vascular permeability (7,8), angiogenesis (VEGF-A), lymphangiogenesis (VEGF-C and D) (9), and carcinogenesis (10). VEGF-A under-goes alternative gene splicing resulting in multiple mature forms that share a VEGF homology domain and which interact with a wide array of common and unique partner proteins. VEGF-A165, the most common and well studied VEGF-A isoform, transduces its effects through binding of its canonical receptor, VEGFR-2 (9,11). Following binding of VEGF-A, VEGFR-2 is autophosphorylated and subsequently internalized to endosomes and is either recycled to the cell surface in a functional state or trafficked to late endosomes and degraded (12).
Although VEGF-A signals primarily through the VEGF receptors, it is also able to mediate its effects through interactions with other receptors, including integrins (3,6,13). Of particular note, we have reported that integrin ␣9␤1 is able to bind VEGF-C and D directly (6), which appears to explain partly the abnormal lymphatic development in the ␣9-null mouse. Subsequently, we showed that VEGF-A also binds integrin ␣9␤1, a biologically important interaction because VEGF-induced angiogenesis could be blocked when activation of ␣9␤1 was inhibited (3). Prototypically, integrin activation and function require the presence of cations and ligand binding using an RGD binding motif (14). However, integrin ␣9␤1 can bind its ligands through non-RGD sequences (15,16), which we presumed would explain its binding to VEGF-A, C, and D. Here, we report the identification of the ␣9␤1-specific binding site in VEGF-A: EYP, a three-amino acid sequence encoded by exon 3.
Cells and Cell Culture-Primary adult human umbilical vein endothelial cells (HUVECs) and human dermal microvascular endothelial cells (HMVECs) were from Lonza (Walkersville, MD) and were grown in their cell-specific growth factor-supplemented nutrient medium. All experiments on primary cell cultures were performed on cells between passages 3 and 5. Integrin ␣9 or mock-transfected SW480 colon carcinoma cells were cultured as described previously (6,19). For experiments requiring serum-free conditions, endothelial and SW480 cells were incubated in their respective basal medium for 4 -5 h and 24 h, respectively.
Flow Cytometry-Cultured cells were trypsinized, washed with phosphate-buffered saline (PBS), and incubated with isotype or appropriate primary antibodies for 20 min on ice and subsequently for 10 min with phycoerythrin-conjugated secondary antibody if indicated. Phycoerythrin-conjugated primary VEGFR-2 antibody was used to detect VEGFR-2 expression. Fluorescence of labeled cells was determined with a flow cytometer (FACScan; Becton Dickinson) and analyzed with CellQuest software (Becton Dickinson).
Cell Adhesion Assay-Assays were performed as described previously with some modifications (3). After coating 96-well microtiter plates (ICN, Linbro/Titertek, OH) with 3 g/ml various ligands at 4°C overnight, wells were blocked with 2% bovine serum albumin (BSA) for 1 h at room temperature. Trypsinized cells were washed in PBS and incubated in the absence or presence of ␣9␤1 blocking antibody (15 g/ml, Y9A2) for 25 min on ice prior to seeding 5 ϫ 10 4 cells/well. After a 2-3-h incubation at 37°C, adherent cells were fixed and stained with 1% formaldehyde, 0.5% crystal violet, and 20% methanol and evaluated by measuring absorbance at 595 nm in a microplate reader (SpectraMax 190; Molecular Devices, Sunnyvale, CA). Adherent cells were also assessed by light microscopy (Axiovert; Zeiss) at magnification of ϫ25 and photographed (Axiovert).
Cell Migration Assay-Haptotaxis or chemotaxis assays were performed as described previously (3,6,19) by using 8-m pore Transwell plates (Corning Costar, Cambridge, MA). Transwell membranes were coated at 4°C overnight with 1% BSA or relevant ligand (3 g/ml RAA, 1 g/ml VEGF-A, or 4 g/ml synthesized peptides), then blocked with 2% BSA at 37°C for 1 h. Serum-starved cells were trypsinized, washed in PBS and 5 ϫ 10 5 cells/ml, incubated with or without ␣9 blocking antibody Y9A2 (15 g/ml), for 30 min on ice, and seeded into the top chamber of the Transwell. Basal me-dium containing 1% FBS (20 ng/ml VEGF-A or 4 g/ml synthesized peptide for chemotaxis assays) was added to the bottom well, to serve as a chemoattractant, and the plates were incubated at 37°C for 6 -12 h. Migrated cells were fixed, stained (Hema 3 kit; Fisher Scientific), and counted in at least four randomly selected high power fields (ϫ25) for each condition and photographed (Axiostar; Zeiss).
siRNA Transfection-HMVECs were grown to 50 -70% confluence in full growth medium without antibiotics in Costar 6-well plates (Corning, New York). After washing twice with serum-free basal medium, transfection was performed with 5 l of Lipofectamine 2000 (Invitrogen) and 5 l of 10 M siRNA specific to integrin ␣9-ITGN9 (Ambion, Austin, TX) or nontargeting siRNA negative control in 2 ml of Opti-MEM (Invitrogen). Cells were then incubated for 72 h at 37°C. The extent of ␣9␤1 knockdown was assessed by Western blotting.
Immunoprecipitation and Western Blot Analysis-For immunoprecipitation (IP) of VEGF-R2, HMVECs or HUVECs were grown in 6-well plates with full growth medium to 80% confluence and subsequently in growth medium with 0.1% FBS for 4 h at 37°C. Cells were then treated with 50 ng/ml VEGF-A or 3 g/ml synthesized peptide for 20 min, washed with cold PBS, and lysed with buffer containing 20 mM Tris, pH 7.4, 150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 0.5% sodium deoxycholate, 10% glycerol, 25 mM NaF, 1 mM PMSF, 1 mM Na 3 VO 4 , and phosphatase (Halt Phosphatase inhibitor; Thermo Scientific, Rockford, IL) and protease inhibitors (protease inhibitor mixture; Sigma). Precleared protein samples (150 g of total protein) were immunoprecipitated with 2 g of VEGFR-2 antibody bound to protein A-Sepharose beads (Amersham Biosciences). The beads were centrifuged, washed three times with lysis buffer, resuspended in 2ϫ Laemmli sample buffer, and boiled at 95°C for 5 min. Proteins were resolved on SDS-PAGE under reducing conditions, transferred to polyvinylidene difluoride (PVDF) membranes (Immobilon-P; Millipore), blocked in 5% nonfat dry milk, and then probed with anti-phosphotyrosine or VEGFR-2 antibodies.
For immunodetection of phosphorylated or total VEGFR-2, Src, or FAK and integrin subunit ␣9 or ␤1, proteins were resolved on SDS-PAGE, transferred to PVDF membranes, blocked in 5% nonfat dry milk, and probed with specific primary antibodies followed by HRP-conjugated secondary antibodies and developed using chemiluminescence (ECL; Amersham Biosciences).
Intracellular Ca 2ϩ Release Assay-The assay was performed as described previously (20). Briefly, HMVECs (passage 4) were serum-starved for 4 h, incubated with 5 M Fura-2 AM (Molecular Probes) for 45 min at 37°C. The Fura-2-loaded cells were washed and resuspended in Krebs-Ringers medium, and cells were transferred to a cuvette and stimulated with 10 ng/ml VEGF-A or 4 g/ml peptide P1 or P2 for 15 min. Intracellular Ca 2ϩ release was measured during this time with the Deltascan illumination system (Photon Technology International).
Endothelial Tube Formation Assay-The assay was performed using an endothelial tube formation assay kit (Cell Biolabs, San Diego, CA) according to the manufacturer's instructions. Briefly, 96-well plates coated with extracellular matrix gel were seeded with HMVECs (2 ϫ 10 4 /well) in 3% FBS growth medium containing either VEGF-A (50 ng/ml) or P1 or EYPD (3 g/ml) and incubated for 12 h at 37°C in the presence or absence of ␣9␤1 blocking antibody. Wells were then washed and incubated in staining solution for 30 min at 37°C, and the formation of capillary-like structures was evaluated using a fluorescence microscope (Apo Tome; Zeiss).
In Vitro Scratch Wounding Assay-The assay was performed using the Cytoselect wound healing kit (Cell Biolabs) according to the manufacturer's instructions. Briefly, 2.5 ϫ 10 5 mock-or ␣9-transfected SW480 cells were added to each well and incubated overnight to form a monolayer. Well inserts were then removed to create a wound field of 0.9-mm diameter, and after washing cells were incubated at 37°C for 12 h in the presence of various ligands: 100 ng/ml VEGF-A or 3 g/ml P1, EYPD, or AYPD in 5% FBS growth medium. The extent of wound closure was determined with a phase-contrast microscope (total internal reflection fluorescence microscope; Olympus, Melville, NY).
Cell Invasion Assay-The assay was performed using BD BioCoat (8-m pore size) Matrigel invasion chambers (Becton Dickinson) according to the manufacturer's instructions. Briefly, 2.5 ϫ 10 4 mock-or ␣9-transfected SW480 cells stimulated by either VEGF-A (100 ng/ml) or synthesized peptides P1, EYPD, or AYPD (3 g/ml) were added to the top chamber and 10% FBS to the bottom chamber of the Transwell invasion plates. Cells were incubated for 36 h at 37°C in the presence or absence of ␣9␤1 inhibiting antibody. Invaded cells were fixed and stained (Hema 3 kit) and counted in four separate, random view fields under a light microscope (Axiostar).
Statistical Analysis-Unless otherwise indicated, data are presented as mean Ϯ S.E. from three or more experiments; p values were determined using paired Student's t tests (Prism 4; GraphPad, San Diego, CA) and considered significant if Յ 0.05. Western blotting was performed at least three times, and a representative example is presented.

A 26-Amino Acid Peptide (P1) from the Exon 3-encoded Region of VEGF-A Specifically Binds
Integrin ␣9␤1-We have shown previously that integrin ␣9␤1 is able to bind VEGF-A, C, and D directly. Although these proteins are expressed by different genes, they share a common central region or VEGF homology domain (Fig. 1A) which we reasoned should contain the ␣9␤1 binding site. Because we have shown previously that ␣9␤1 not only binds VEGF-A165 but also VEGF-A121 which lacks exons 6 and 7, it appeared that the binding site would reside within the protein sequence encoded by exons 1-5, which is common to both isoforms (Fig. 1B). VEGF sequence analysis revealed the greatest homology between VEGF-A, C, and D lay within exons 3 and 4, which were also rich in the necessary negative amino acid residues for integrin binding (Fig. 1C). To test for potential ␣9␤1 binding sites we assessed differential cell adhesion and migration in the presence of three large synthesized peptides (P1, P2, and P3), de-rived from regions in VEGF-A165 encoded by exons 3-5. The characteristics of these peptides are shown in Fig. 1D.
We initially tested these three peptides in mock-or ␣9transfected SW480 cells (Fig. 2), which do not express VEGF receptors (6). As we have shown previously, both mock-and ␣9-transfected cells adhered to VEGF-A (Fig. 2B). In ␣9-SW480 cells, the VEGF-A peptide, P1, and not peptides P2 and P3, supported cell adhesion to a degree similar to that of VEGF-A and the ␣9␤1-specific ligand, RAA (Fig. 2B). In contrast, mock-transfected cells did not support cell adhesion to RAA or any of the three peptides. Fig. 2C shows that similar to adhesion, P1 and not the peptides P2 or P3 was able to support migration of ␣9-SW480 cells.
To support the ␣9␤1 specificity of P1 further, we tested the extent of cell adhesion in the presence of blocking antibodies to the ␣3␤1 and ␣v␤3 integrins which are expressed in SW480 cells (Fig. 2D, left) and are known to bind to VEGF-A. Fig. 2D (right) shows that all three blocking antibodies inhibited ␣9-SW480 adhesion to VEGF-A, with the greatest effect following inhibition of ␣9␤1. In contrast, cell adhesion to P1 was only blocked by ␣9␤1 inhibitor and not inhibitors of integrins ␣3␤1 or ␣v␤3. Taken together, these findings suggest that P1 specifically binds integrin ␣9␤1 and transduces cell adhesion and migration independently of other integrins and that VEGF receptors may not be required for this interaction. Peptide 1 (P1) Binds Integrin ␣9␤1 in a VEGFR-2-independent Fashion-To test further the nature of the interaction of P1 with integrin ␣9␤1, we determined whether VEGFR-2, the cognate receptor for VEGF-A, modified binding of P1. To accomplish this, we performed experiments using endothelial cells, HMVECs and HUVECs, which express ␣9␤1 differentially. Although both cell types expressed VEGFR-2 (Fig. 3A,  bottom), only HMVECs expressed integrin ␣9␤1 (Fig. 3A, top). Similar to our findings using ␣9-SW480 cells, HMVECs showed increased adhesion to P1 but not P2 or P3 (Fig. 3B), and the extent of binding was similar to that of both VEGF-A and the ␣9␤1-specific ligand, RAA. Furthermore, cell adhesion to P1, VEGF-A, and RAA was inhibited in the presence of ␣9␤1-blocking antibody (Y9A2). Next, we determined the extent of migration on various doses of P1, and Fig. 3C shows that cell migration induced by P1 is dose-dependent with maximal migration achieved at a dose of 4 g/ml. There was no further increase in cell migration with increasing doses of peptide.
Next, we compared the extent of ␣9␤1-mediated migration in HMVECs or HUVECs, which express ␣9␤1 differentially. Fig. 3D (left) shows that migration of HMVECs, expressing both VEGFR-2 and ␣9␤1, on P1, VEGF-A, or RAA is inhibited by ␣9␤1-blocking antibody. Neither P2 nor P3 induced an increase in HMVEC migration. As expected, migration of HUVECs, expressing VEGFR-2 but not ␣9␤1, was increased in the presence of VEGF-A (Fig. 3D, right) but was not inhibited by ␣9␤1-blocking antibody. In addition, HUVEC migration was not increased in the presence of the ␣9␤1-specific ligand, RAA, or any of the synthesized VEGF-A peptides. Taken together, these results strongly suggest that there is an integrin ␣9␤1 binding site in VEGF-A that is unique and does not appear to interact with VEGFR-2. As with SW480 cells, we wished to determine the integrin specificity of the ␣9␤1specific VEGF-A peptide. Fig. 3E shows that inhibitors of integrins ␣3␤1 and ␣v␤3 partially blocked HMVEC adhesion (left) and migration (right) but to a lesser degree than inhibition of ␣9␤1. In contrast, only inhibition of ␣9␤1 blocked cell adhesion and migration on P1. These findings along with those in SW480 cells strongly support the ␣9␤1 specificity of this VEGF-A peptide.
To confirm further the VEGFR-2-independent interaction of P1 with integrin ␣9␤1 we used an siRNA approach to inhibit expression of the integrin ␣9. Fig. 4A shows an immunoblot of lysates from transfected HMVECs demonstrating that, compared with nontargeted control siRNA, ␣9-targeted siRNA successfully inhibited expression of the ␣9 subunit. We next determined the effect of ␣9 knockdown on HMVEC adhesion and migration. Similar to our findings using ␣9␤1specific blocking antibody, knockdown of ␣9 also inhibited adhesion to P1, VEGF-A, and the ␣9␤1-specific ligand, RAA (Fig. 4B). In a similar fashion, knockdown of ␣9 inhibited HMVEC migration on P1, VEGF-A, and RAA (Fig. 4C).
Glutamic Acid-Tyrosine-Proline (EYP), a Three-amino Acid Sequence within Peptide 1 (P1), Is Essential for ␣9␤1-specific Binding of VEGF-A-To determine the essential amino acids in VEGF-A required for binding to ␣9␤1, we tested the ability of HMVECs to adhere and migrate on numerous smaller overlapping peptides derived from P1 (Fig. 5). Integrins utilize the negatively charged amino acids in their ligands to facilitate binding (1, 2), including glutamic and aspartic acid residues (16). We therefore expected that those regions in VEGF-A rich in Glu and Asp amino acid residues would best support ␣9␤1-specific binding. Fig. 5A shows that only those peptides (P1-A, P1-B, P1-G, P1-H) containing the EYPD sequence (underlined) were able to facilitate HMVEC migra-  tion. Using peptides P1-G and P1-H (Fig. 5B) and P1-A and P1-B (data not shown), we found that similar to full-length P1, HMVEC adhesion (left) to these peptides was inhibited in the presence of ␣9␤1-specific blocking antibody. Similarly, HMVEC migration in the presence of P1-G and P1-H was inhibited by blocking antibody to ␣9␤1 (Fig. 5B, right). Because P1-G did not contain the aspartic acid residue of EYPD and still showed similar adhesion and migration properties to P1-A, B, and H, we hypothesized that aspartic acid was not necessary for binding to ␣9␤1. To confirm this, we synthesized an EYPD peptide and associated EYPD-mutant peptides in which the glutamic and aspartic acid residues were substituted for alanine (EYPA, AYPD, AYPA). Both EYPD and EYPA supported HMVEC migration (Fig. 5C). In contrast, HMVECs demonstrated no increase in cell migration on either AYPD or AYPA, suggesting that within EYPD the glutamic acid residue is necessary for ␣9␤1 binding of VEGF-A. We next performed competitive inhibition assays to define better the ␣9␤1 specificity of the small VEGF-A peptides. Preincubation of cells with peptides P1, P1-G, P1-H, and EYPD significantly inhibited ␣9-SW480 (Fig. 5D) and HMVEC (Fig.  5E, left) migration stimulated by VEGF-A. Compared with SW480 cells, the inhibitory effect of the peptides on HMVEC migration was significantly less, reflecting the uninhibited migratory effect of VEGFR-2 activated by VEGF-A. In contrast to SW480 and HMVECs, preincubation of HUVECs with peptides did not inhibit VEGF-A-induced cell migration (Fig. 5E, right). Taken together, these results suggest that EYP is the essential VEGF sequence needed to support ␣9␤1-specific binding.
Peptides P1 and EYPD Specifically Bind and Activate Integrin ␣9␤1 and Signal Cell Migration through Activation of Src-We next wished to determine the signaling pathways induced by the binding of EYP to ␣9␤1 and, in particular, whether VEGFR-2, the cognate receptor for VEGF-A, was also activated. Fig. 6A (top) shows as expected that in HMVECs and HUVECs, which both express VEGFR-2, VEGF-A induced robust phosphorylation of the receptor. In addition, in both HMVECs, which express ␣9␤1, and HUVECs, which do not express ␣9␤1, there was no VEGFR-2 phosphorylation in response to P1 stimulation.
Because binding of VEGF-A to VEGFR-2 induces calciumdependent receptor dimerization, autophosphorylation, and internalization, we used a calcium release assay (20) to test further whether VEGFR-2 was activated by the VEGF-A peptides. Fig. 6A (bottom) shows that in both HMVECs and HUVECs, VEGF-A induced a robust release of intracellular calcium, indicating activation and internalization of VEGFR-2. In contrast, neither peptide P1 (␣9␤1-activating), nor P2 (non-␣9␤1-activating) induced a calcium release response in HMVECs or HUVECs. Taken together, these results support that there is a specific ␣9␤1 binding site in VEGF-A which is unique from VEGFR-2 binding and activation.
We next wished to investigate the activation and downstream signaling of ␣9␤1 following binding of the ␣9␤1-specific VEGF peptides. As expected, VEGF-A activated the ␤1 integrin subunit in both HMVECs and HUVECs (Fig. 6B). Similarly, both P1 and EYPD activated integrin ␤1 subunit in HMVECs (Fig. 6B, left). In contrast, there was no ␤1 activation over base-line conditions in HUVECs exposed to the same peptides (Fig. 6B, right). These results provide support for the activation of ␣9␤1 by the ␣9␤1-specific VEGF-A peptides.
Following VEGF and other ligand-induced activation, ␣9␤1 transduces its biologic effects through a number of signaling intermediates, including Src and FAK (3,6,19,21). Src is the most proximal ␣9␤1 signaling protein and essential for mediating ␣9␤1 function (21). Fig. 6C (top) shows that in HMVECs (left), which express ␣9␤1, both VEGF-A and P1 activated Src to a similar degree as the ␣9␤1-specific ligand, RAA. In contrast, in HUVECs (right) only VEGF-A and not P1 or RAA induced phosphorylation of Src over base-line conditions. Furthermore, Fig. 6C (bottom) shows that Src is activated by P1 in both a time-(0 -20 min) and dose-(0.5-4 g/ml) de- pendent fashion, and even at the highest dose tested, Src phosphorylation is blocked by Src inhibitor, PP1.
Next, we assessed HMVEC and HUVEC migration in the absence or presence of the Src inhibitor, PP1 (Fig. 6D). As expected, VEGF-A-induced migration in both HMVECs (left) and HUVECs (right), which was significantly inhibited by PP1. In HUVECs, which do not express ␣9␤1 integrin, neither P1 nor the ␣9␤1-specific ligand, RAA, induced significant cell migration. In contrast, robust migration of HMVECs was induced by both P1 and RAA. Inhibition of Src in HMVECs resulted in decreased RAA-, VEGF-A-, and P1-induced cell migration, providing further support for the specific ligation and activation of ␣9␤1 integrin by P1. Fig. 6E shows that like P1, the peptide EYPD did not activate VEGFR-2 in either HMVECs or HUVECs, as determined by phosphospecific antibodies to VEGFR-2 or VEGFR-2 immunoprecipitation (data not shown). In con-trast to HUVECs, EYPD did activate the ␣9␤1-signaling intermediates, Src and FAK, in HMVECs. The specificity of EYP for ␣9␤1 was further supported by the finding that, the alanine-substituted peptide AYPD did not activate any of these signaling proteins.
Second, because cell migration is also essential in VEGFassociated wound healing, we performed in vitro scratch as-says using mock-or ␣9-transfected SW480 cells that were exposed to various ligands (Fig. 7B). Compared with ␣9-SW480, mock cells did not migrate significantly into the scratched area under any of the stimuli studied. Migration of FIGURE 6. The VEGF-A peptide EYP specifically binds and activates ␣9␤1 but not VEGFR-2. A, top, Western blotting to assess for VEGFR-2 phosphorylation in HMVECs (left) or HUVECs (right) following exposure to VEGF-A or P1 for 20 min; lysates were immunoprecipitated with VEGFR-2 antibody and immunoblotted with anti-phosphotyrosine antibody. Total VEGFR-2 was used as loading control. A, bottom, calcium release assays performed in either HMVEC (left) or HUVECs (right) in the presence of VEGF-A or synthesized VEGF-A peptides P1 (binds integrin ␣9␤1) or P2 (does not bind integrin ␣9␤1). B, immunoblots to assess for ␤1 integrin subunit activation in lysates from HMVECs (left) and HUVECs (right) exposed to various ligands for 20 min. GAPDH was used as loading control. C, top, immunoblots to assess for phosphorylated Src in lysates from HMVECs (left) or HUVECs (right) exposed to various ligands for 20 min. Total Src was used as loading control. C, bottom, immunoblots to assess for phosphorylated Src in lysates from ␣9-SW480 at increasing time points following stimulation with 4 g/ml P1 (left) or after 20 min following stimulation with increasing doses of P1. Src inhibitor, PP1, was used in separate lysates from cells exposed to 4 g/ml P1 (right). Total Src was used as loading control for all experiments. D, haptotaxis assays using HMVECs (left) or HUVECs (right) pretreated with ligands as indicated and in the absence or presence of Src inhibitor (PP1). E, immunoblots to assess for phosphorylated VEGFR-2 (pY996 or pY951), Src or FAK (pY925 or pY397) in lysates from HMVECs (left) or HUVECs (right) exposed to the VEGF-A peptides, P1, EYPD, or AYPD. Total VEGFR-2, Src or FAK, respectively, was used as loading control. FIGURE 7. EYP mediates ␣9␤1-dependent endothelial cell tube formation, wound healing and Matrigel invasion. A, endothelial cell tube-forming assays using HMVECs stimulated by VEGF-A or the VEGF-A peptides, P1 or EYPD, and conducted in the presence or absence of integrin ␣9␤1 inhibitor. B, in vitro scratch assays using mock-or ␣9-transfected SW480 cells exposed to VEGF-A or ␣9␤1-specific (P1 and EYPD) VEGF-A peptides or ␣9␤1 nonspecific VEGF-A peptide, AYPD; in the absence or presence of ␣9␤1 inhibitor. C, Matrigel invasion assays using ␣9-transfected SW480 cells exposed to VEGF-A or VEGF-A peptides as indicated, in the absence or presence of ␣9␤1 inhibitor.
␣9-SW480 cells into the scratch site was stimulated by both P1 and EYPD peptides, to a similar degree as VEGF-A. Cell migration into the scratch site induced by either P1 or EYPD was significantly decreased when cells were treated with ␣9␤1 inhibitor, suggesting that the migratory response was ␣9␤1specific. This was further supported by our finding that cell migration was not induced by AYPA, a non-␣9␤1-activating VEGF-A peptide.
Third, because cell migration is necessary for cancer cell movement through tissues and metastasis, we performed Matrigel invasion assays using either mock-or ␣9-expressing SW480 colon carcinoma cells (Fig. 7C). Compared with ␣9-SW480, mock cells did not invade Matrigel when exposed to any of the VEGF stimuli (data not shown). Similar to the results from wound healing assays, ␣9-SW480 cells treated with VEGF-A, P1, or EYPD showed a comparable extent of Matrigel invasion (Fig. 7C). The invasive stimulus induced by P1 and EYPD was inhibited by blocking ␣9␤1 and was not apparent when cells were stimulated by AYPD, a non-␣9␤1-activating VEGF-A peptide. Taken together, these findings provide strong evidence for the biological significance of the direct interaction of VEGF-A with integrin ␣9␤1 through the specific binding site, EYP.

DISCUSSION
The process of ligand binding to integrins is context-specific, depending not only on the availability and affinity of ligand and receptor but also the structural conformation and flanking residues of the ligand binding site (14). The prototypical integrin-ligand binding interaction utilizes the triamino acid sequence, arginine-glycine-aspartic acid (RGD), for example integrin ␣5␤1-binding fibronectin. Furthermore, many integrins share a conserved amino acid sequence, the insert (I) domain, with an associated metal-ion-dependent-adhesion (MIDAS) domain that mediates ligand binding in a cation-dependent manner (14,22). However, not all integrins (including ␣9␤1) have an I domain (23), and in fewer than half of cases is the RGD sequence required for binding of ligand. Taken together with the known large repertoire of integrinbinding partners, it is clear that ligands contain unique amino acid sequences to ensure binding, integrin activation, and signal transduction. Growth factors represent one such unique integrin ligand (24).
Binding of growth factors to their cognate receptors mediates numerous biological functions. In the case of VEGF-A and its cognate receptor VEGFR-2, this includes vascular permeability, cell proliferation, cancer cell survival, and angiogenesis (11). However, numerous reports have demonstrated that growth factors and their cognate receptors may also interact with nonfamily member growth factors and nongrowth factor receptors such as integrins (25,26). Growth factorintegrin interactions play an important role in processes that include development, tumor cell proliferation, angiogenesis, and lymphangiogenesis (4,24,27,28). This cross-talk between growth factors, their receptors, and integrins is complex and in the case of VEGF/VEGFR-2 signaling may modulate integrin function and vice versa (29,30). Both ␤1 and ␤3 integrins appear to be important in this cross-talk interaction (26,30).
Integrin ␣v␤3 has been studied extensively and shown to bind fibroblast growth factor-1 (31) and play a role in both VEGFassociated angiogenesis and carcinogenesis. This cross-talk is transduced through a number of mechanisms including modulation of VEGF expression in the tumor microenvironment and interactions with cancer and normal tissue matrix (32).
Similarly, a number of ␤1 integrins have been shown to interact with VEGF. Through its association with tetraspanin, the ␣5␤1 integrin modulates VEGF/VEGFR function to facilitate VEGF-induced tumor angiogenesis (33). Integrin ␣4␤1, which forms a subfamily with ␣9␤1 sharing 40% amino acid sequence homology, also appears to interact with VEGF to modulate lymphangiogenesis (34). But in neither of these cases were the integrins shown to interact directly with VEGF.
Hutchings et al. identified ␣v␤3 and ␣3␤1 integrins as being able to interact with VEGF-A (13) and mediating at least in part, the adhesion, migration, and survival of human umbilical artery endothelial cells. Although the investigators showed direct VEGF-A binding to ␣v␤3, they did not demonstrate direct binding to ␣3␤1 or specific integrin binding regions within VEGF-A. Of the ␤1 integrins, only integrin ␣9␤1 has been shown by direct protein-protein binding assays to bind with VEGF, and this interaction was found to mediate endothelial cell adhesion and migration and, in a synergistic fashion with VEGFR-2, modulate VEGF-A-induced angiogenesis (3).
Because a number of integrins appear to play a role in VEGF-induced vasculogenesis and carcinogenesis, one might argue that our current findings are not novel. However, we have provided evidence that ␣9␤1 might mediate unique effects compared with other integrins. First, compared with other integrins the relative contribution of ␣9␤1 to cell migration (Fig. 2D) and by association, presumably metastasis, is more robust (3). Second, ␣9␤1 is able to bind not only the VEGF-A165 isoform but also VEGF-A121 (3), suggesting that its inhibition might provide a more broad biologic effect. Finally, the identification of a specific ␣9␤1 binding site provides a focused therapeutic target that may result in a more specific and effective clinical treatment.
Similar to our study, alanine mutagenesis was used to help identify the binding site for VEGFR-2, the VEGF-A cognate receptor. It was found to reside within a positively charged hairpin loop of VEGF-A; the essential amino acids mediating binding within this protein loop were Arg 82 , Lys 84 , His 86 (35). In contrast, the EYP residues reside at positions 64 -66. Based on the three-dimensional structure of VEGF-A (36,37), it appears that EYP is located in an exposed loop on the surface of the molecule and located opposite the VEGFR-2 binding site. Whether blocking ␣9␤1 binding to EYP may also modulate or even disrupt VEGF/VEGFR-2 interactions remains to be determined.
It appears that the ability of ␣9␤1 to interact with growth factors is not limited to VEGF. It can mediate cell proliferation through direct binding of nerve growth factor (38) and granulopoiesis (39) and lymphatic cell migration (27) through interactions with granulocyte-colony-stimulating factor and hepatic growth factor, respectively. Whether the amino acids necessary for interactions between ␣9␤1 and these growth factors are common to VEGF remains to be determined, but based on the known diverse ligand binding sequences used by ␣9␤1, it seems unlikely. The ␤1 cytoplasmic tail is considered the key heterodimeric subunit mediating the activation of integrins and subsequent cell signaling. Whereas the ␣ subunit, through the conserved GFFKR sequence, contributes to ␤1 tail conformation and the cell specificity of integrin activation (40). In the case of ␣9␤1, we have shown previously that the ␣9 cytoplasmic domain is crucial for the initiation of intracellular signaling and subsequent transduction of the integrins biological effect (19,21,41). We would anticipate that this is also true for the transduction of VEGF-induced ␣9␤1 activation and signaling.
In summary, we have used synthesized peptides to map the integrin ␣9␤1 binding site in VEGF-A and found that it lies within the amino acids encoded by exon 3 of VEGF-A, specifically, residues EYP (Glu 64 , Tyr 65 , and Pro 66 ). EYP specifically activated ␣9␤1 in a dose-and time-dependent fashion, but not VEGFR-2, and mediated biologically relevant functions such as endothelial cell tube formation and cancer cell Matrigel invasion. These findings not only support our previous report showing that integrin ␣9␤1 directly binds VEGF-A (3) but also specifically show that this binding interaction requires amino acids expressed in the VEGF homology domain that are common to other ␣9␤1 VEGF ligands, VEGF-C and D (6). Taken together, these findings identify a novel potential pharmacotherapeutic site that could be targeted to block ␣9␤1-mediated actions of VEGF such as cancer-related angiogenesis and metastasis. But also, as the role of ␣9␤1 in human biology expands to now include granulopoiesis (39) and hematopoietic stem and progenitor cell function (42), targeted therapy may not only have a role in pathogenic entities but also in facilitating reparative processes.