The MIG-2/Integrin Interaction Strengthens Cell-Matrix Adhesion and Modulates Cell Motility*

Integrin-mediated cell-matrix adhesion plays an important role in control of cell behavior. We report here that MIG-2, a widely expressed focal adhesion protein, interacts with β1 and β3 integrin cytoplasmic domains. Integrin binding is mediated by a single site within the MIG-2 FERM domain. Functionally, the MIG-2/integrin interaction recruits MIG-2 to focal adhesions. Furthermore, using αIIbβ3 integrin-expressing Chinese hamster ovary cells, a well described model system for integrin activation, we show that MIG-2 promotes integrin activation and enhances cell-extracellular matrix adhesion. Although MIG-2 is expressed in many cell types, it is deficient in certain colon cancer cells. Expression of MIG-2, but not of an integrin binding-defective MIG-2 mutant, in MIG-2-null colon cancer cells strengthened cell-matrix adhesion, promoted focal adhesion formation, and reduced cell motility. These results suggest that the MIG-2/integrin interaction is an important element in the cellular control of integrin-mediated cell-matrix adhesion and that loss of this interaction likely contributes to high motility of colon cancer cells.

Cell-extracellular matrix (ECM) 3 adhesion is a fundamental process that is mediated by transmembrane receptors such as integrins (1)(2)(3)(4)(5)(6). The interactions of integrins with ECM ligands can be controlled by integrin activation via "inside-out" signaling. Talin, a FERM (Band 4.1 (four point one)/ezrin/radixin/ moesin) domain-containing focal adhesion (FA) protein, can play a key role in this process (for recent reviews, see Refs. [7][8][9][10]. Binding of the talin FERM domain to the ␤ integrin cytoplasmic domains results in separation of the ␣ and ␤ integrin cytoplasmic tails and consequently in an increase in integrin extracellular ligand-binding affinity (i.e. integrin activa-tion) (11)(12)(13). Integrin extracellular ligand-binding affinity plays an important role in control of initial cell-ECM adhesion. Additionally, integrin-mediated cell-ECM adhesion can be enhanced through interactions with cytoskeletal proteins, a process that has been termed cytoskeletal strengthening (14 -16). The physical basis underlying the cytoskeletal strengthening of cell-ECM adhesion has been well described (16). However, the molecular interactions that mediate this process remain to be defined.
MIG-2 (mitogen-inducible gene-2, also known as kindlin-2) is a widely expressed and evolutionarily conserved cytoplasmic protein (17)(18)(19)(20)(21). Genetic studies have shown that Caenorhabditis elegans UNC-112, a homolog of MIG-2, is required for attachment of body-wall muscle cells to the hypodermis (17,19). Loss of UNC-112 in C. elegans results in an embryonic lethal Pat (paralyzed, arrested elongation at two-fold) phenotype resembling that of ␣ or ␤ integrin loss (17,19). In mammalian organisms, MIG-2 has been detected in many cell types, including fibroblasts, muscle cells, endothelial cells, and epithelial cells (20,22). In these cells, it concentrates at FAs. MIG-2 interacts with migfilin (20), a filamin-and VASP (vasodilatorstimulated phosphoprotein)-binding protein (20,21,23). Through this interaction, it recruits migfilin to FAs and provides a link from FAs to the actin cytoskeleton (20). Although MIG-2 is crucial for recruiting migfilin to FAs, how MIG-2 is recruited to FAs was not known. Structurally, MIG-2 contains an N-terminal region that exhibits no obvious structural motif and a C-terminal FERM domain. Notably, the MIG-2 FERM domain contains a region that shares considerable sequence similarity with the integrin-binding site of the talin FERM domain. Furthermore, kindlin, a protein that shares significant (62%) sequence identity with MIG-2, interacts with ␤1 and ␤3 integrin cytoplasmic domains (24). However, the functions of the kindlin/integrin interaction remain unknown.
Cell-ECM adhesion is intimately involved in the regulation of cell behavior such as cell motility. Theoretical consideration and experimental studies have shown that the relationship between cell-ECM adhesion and motility is biphasic (25,26). Thus, the regulation of cell-ECM adhesion strength is important in the control of cell motility. Consistent with this, alterations of proteins that are pertinent to the control of cell-ECM adhesion are frequently associated with human diseases such as cancer. Determining the molecular basis underlying the regulation of cell-ECM adhesion and migration is therefore of not only general biological importance but also considerable clinical significance. In this work, we show that MIG-2 interacts with ␤1 and ␤3 integrin cytoplasmic domains and functions as an important regulator of integrin activation, cell-ECM adhesion, and migration.
Mutagenesis, DNA Constructs, and Transfection-The vector encoding green fluorescent protein (GFP)-tagged MIG-2 has been described (20). Point mutations were introduced into MIG-2 coding sequence as specified in each experiment using a QuikChange TM site-directed mutagenesis system (Stratagene). DNA fragments encoding MIG-2 deletion mutants were generated by PCR. DNA fragments encoding MIG-2 mutants were inserted into the pEGFP-C2 vector (Clontech). All mutations were confirmed by DNA sequencing. Cells were transfected with vectors encoding GFP-tagged wildtype or mutant MIG-2 or GFP alone as a control using Lipofectamine 2000 (Invitrogen).
GST Fusion Protein Pulldown Assays-The vectors encoding GST fusion proteins containing the ␤1A integrin cytoplasmic domain (residues 775-786), the ␤3 integrin cytoplasmic domain (residues 716 -762), or a mutant form of the ␤3 integrin cytoplasmic domain in which Tyr 747 was substituted with Ala were generated as described previously (27,28). The vectors were used to transform Escherichia coli DH5␣ cells. Expression of GST and GST fusion proteins was induced with isopropyl ␤-D-thiogalactopyranoside. The bacterial cells were lysed with 150 mM NaCl, 10 mM Tris (pH 8.0), and 1 mM EDTA (STE buffer) containing 100 g/ml lysozyme for 15 min (on ice), followed by sonication in STE buffer containing 5 mM dithiothre-itol, 1.5% Sarkosyl, and protease inhibitors. After the cell debris was removed by centrifugation, the lysates were incubated with glutathione-Sepharose beads (Pierce) at 4°C for 1 h. The glutathione-Sepharose beads were precipitated by centrifugation, washed three times with STE buffer containing 0.1% Triton X-100, and used for GST pulldown assays as we described previously (20). Brief, Chinese hamster ovary (CHO) cells, SK-LMS-1 cells, or SK-LMS-1 cells transfected with vectors encoding GFP or GFP-tagged wild-type or mutant MIG-2 were lysed with 1% Triton X-100 in 20 mM Tris (pH 7.1) containing 150 mM NaCl, 10 mM Na 4 P 2 O 7 , 2 mM Na 3 VO 4 , 100 mM NaF, 10 mM EDTA, and protease inhibitors (lysis buffer). The lysates were incubated with glutathione-Sepharose beads containing GST or GST-integrin cytoplasmic domain fusion proteins for 4 h or longer at 4°C. The glutathione-Sepharose beads were precipitated by centrifugation, washed four times with lysis buffer, and then analyzed by Western blotting and Coomassie Blue staining as specified in each experiment.
Integrin Activation-The effect of MIG-2 on integrin activation was analyzed using CHO cells stably expressing ␣IIb␤3 integrin and activation-specific anti-␣IIb␤3 integrin mAb PAC1 as described (28). Briefly, CHO cells expressing ␣IIb␤3 integrin were transfected with GFP-tagged MIG-2, the GFPtagged talin head domain (talin-H; residues 1-429), or GFP. Twenty-four hours after transfection, the cells were harvested; suspended in Hanks' balanced salt solution/bovine serum albumin (BSA); and stained with mAb PAC1 (20 g/ml) for 30 min at 22°C, followed by incubation with Alexa Fluor 633-conjugated goat anti-mouse IgM Ab for 30 min on ice. After washing, the cells were fixed and analyzed using a FACSCalibur flow cytometer. mAb PAC1 binding was analyzed only on a gated subset of cells positive for GFP expression (i.e. GFP-or GFP fusion protein-expressing cells). The mean fluorescence intensity (MFI; generated with CellQuest software) of mAb PAC1 bound to the GFP-MIG-2-or GFP-talin-H-expressing cells was divided by the MFI of mAb PAC1 bound to the control GFPexpressing cells in the same experiment to obtain a relative MFI value. Seven independent experiments were performed, and the relative MFI in each experimental set, GFP-MIG-2 or GFPtalin-H, were compared with the MFI of the GFP control by a paired t test to determine statistical significance. p values Ͻ0.05 were considered to be statistically significant.
Adenoviral Vectors and Infection-Adenoviral vectors encoding wild-type or mutant MIG-2 were generated using the AdEasy system following a previously described protocol (29,30). Briefly, MIG-2 coding sequences were cloned into the NotI/XbaI sites of the pAdTrack-CMV shuttle vector. The resultant plasmids were linearized with PmeI, purified, and mixed with supercoiled pAdEasy-1. The vectors were transferred into E. coli BJ5183 by electroporation using a Bio-Rad Gene Pulser electroporator. Recombinants that were resistant to kanamycin were selected, and recombination was confirmed by PacI digestion. The positive plasmids were then transformed into DH5␣ by heat shock for large-scale amplification. The plasmid DNAs were digested with PacI, ethanol-precipitated, and used to transfect 293 cells with Lipofectamine PLUS (Invitrogen). The transfectants were harvested ϳ10 days after transfection. The cells were lysed by three cycles of freezing in a

MIG-2/Integrin Interaction in Cell Adhesion and Migration
methanol/dry ice bath and rapid thawing at 37°C, and the lysates containing the recombinant adenovirus were collected. The control adenoviral expression vector encoding ␤-galactosidase was kindly provided by Drs. Tong-Chuan He and Bert Vogelstein (Howard Hughes Medical Institute, Johns Hopkins Oncology Center, Baltimore, MD). Adenoviral vectors were purified by CsCl gradient centrifugation as described (31) and then used to infect cells. The infection efficiency was monitored by the expression of GFP encoded by the viral vectors. The percentage of GFP-positive cells typically reached 80 -90% within 1 day after infection. Overexpression of wild-type or mutant MIG-2 in the infected cells was confirmed by Western blotting.
FA Localization of GFP-tagged MIG-2-SK-LMS-1 cells were transfected with GFP-tagged wild-type or mutant MIG-2. One day after transfection, the cells were trypsinized, replated on fibronectin-coated coverslips, and cultured for 24 h. The cells were fixed with 4% paraformaldehyde, stained with anti-vinculin mAb and Rhodamine Red TM -conjugated anti-mouse IgG Ab, and observed under a Leica DMR fluorescence microscope equipped with GFP and rhodamine filters.
Cell-ECM Adhesion-Cell-ECM adhesion was assessed by centrifugation assays either at 4°C (to measure initial cell-ECM adhesion, which is controlled primarily by integrin ligand-binding activity) or after incubation of the cells at 37°C (to allow cytoskeletal strengthening) as described previously (16). For CHO cells, the cells were transfected with expression vectors encoding GFP-MIG-2, GFP-talin-H (as a positive control), or GFP (as a negative control) using Lipofectamine PLUS. Thirtysix hours after DNA transfection, the transfectants (1 ϫ 10 5 cells/well) were seeded in fibrinogen-coated 96-well plates (Greiner Bio-One). GFP-positive cells were quantified by measuring fluorescence at excitation and emission wavelengths of 485 and 535 nm, respectively, with a GENios Pro fluorescence microplate reader (Tecan). The plates were tightly sealed with sealing films (USA Scientific, Inc.) and centrifuged with a Sorvall RT7 Plus centrifuge equipped with a microplate carrier at 600 rpm for 3 min at 4°C to facilitate cell settlement. The plates were then inverted and centrifuged at 1000 rpm for 3 min at 4°C. The transfectants that remained adhered were quantified by measuring fluorescence at excitation and emission wavelengths of 485 and 535 nm, respectively. Cell adhesion was calculated as the fluorescence reading at excitation and emission wavelengths of 485 and 535 nm, respectively, after inverted centrifugation divided by the fluorescence reading at excitation and emission wavelengths of 485 and 535 nm, respectively, before inverted centrifugation. The adhesion of the cells expressing wild-type or mutant MIG-2 was compared with that of the control GFP cells (normalized to 1). Student's t test was used for statistical analyses. p values Ͻ0.05 were considered statistically significant.
For SK-LMS-1 cells, the cells were infected with the control ␤-galactosidase adenovirus or adenoviruses encoding wildtype MIG-2 or mutants. One day after infection, the cells (5 ϫ 10 4 /well) were seeded in collagen I-coated 96-well plates (Greiner Bio-One). The GFP-positive adenovirus-infected cells were quantified by measuring fluorescence at excitation and emission wavelengths of 485 and 535 nm, respectively, with the GENios Pro fluorescence microplate reader. The plates were tightly sealed and centrifuged with the Sorvall RT7 Plus centrifuge equipped with a microplate carrier at 600 rpm for 5 min at 4°C to facilitate cell settlement. To analyze initial cell adhesion, the plates were inverted and centrifuged at 600 rpm for 1 min at 4°C. To analyze cytoskeletal strengthened cell adhesion, the plates were incubated at 37°C for 60 min and then inverted and centrifuged at 600 rpm for 12 min at 37°C. The adenovirusinfected cells that remained adhered were quantified by measuring fluorescence at excitation and emission wavelengths of 485 and 535 nm, respectively. Cell adhesion was calculated as described above. The adhesion of the cells expressing wild-type or mutant MIG-2 was compared with that of the control ␤-galactosidase cells (normalized to 1). Student's t test was used for statistical analyses. p values Ͻ0.05 were considered statistically significant.
For RKO cells, the cells were infected with ␤-galactosidase adenovirus or adenoviruses encoding wild-type or mutant MIG-2. Two days after the infection, the cells (1 ϫ 10 5 /well) were seeded in collagen I-coated 96-well plates, and the GFPpositive adenovirus-infected cells were quantified as described above. The plates were sealed and centrifuged with a Sorvall RTH750 centrifuge at 600 rpm for 5 min at 4°C. To analyze initial cell adhesion, the plates were inverted and centrifuged at 600 rpm for 1 min at 4°C. To analyze cytoskeletal strengthened cell adhesion, the plates were incubated at 37°C for 10 min and then inverted and centrifuged at 600 rpm for 3 min at 37°C. The amount of the adenovirus-infected cells that remained adhered was quantified by measuring fluorescence at excitation and emission wavelengths of 485 and 535 nm, respectively. Cell adhesion and statistics were calculated as described above.
To assess FA formation, SK-LMS-1 or RKO cells infected with the control ␤-galactosidase adenovirus or adenoviruses encoding wild-type or mutant MIG-2 were plated on coverslips and cultured as specified in each experiment. The cells were fixed with 4% paraformaldehyde, stained with mouse antivinculin mAb and Rhodamine Red TM -conjugated antimouse IgG Ab, and observed under the Leica DMR fluorescence microscope.
Cell Migration-Cell migration was analyzed using Transwell cell motility chambers as described (32,33). Briefly, the undersurfaces of the 8-mm pore diameter cell motility chambers (BD Biosciences) were coated with 10 g/ml fibronectin. The cells were trypsinized 24 h after adenoviral infection and washed twice with phosphate-buffered saline containing 1 mg/ml BSA. The cells were suspended in 0.2 ml Dulbecco's modified Eagle's medium containing 1 mg/ml BSA and added to the upper chambers. After incubation at 37°C in a 5% CO 2 and 95% air atmosphere for the periods of time specified in each experiment, the cells on the upper surface of the membrane were removed. The membranes were fixed, and the cells on the undersurface were stained with Gill's hematoxylin III. The cells JULY 13, 2007 • VOLUME 282 • NUMBER 28 from five randomly selected microscopic fields were counted. The number of MIG-2-or Q614A/W615A mutant-overexpressing cells that migrated through the membrane pores was compared with the number of control ␤-galactosidase cells (normalized to 100%).

MIG-2/Integrin Interaction in Cell Adhesion and Migration
Molecular Modeling-The primary sequences of MIG-2 and talin were aligned using the ClustalW program, which predicted a considerable MIG-2 FERM F3 sequence. A homology model of the MIG-2 FERM F3-free form was constructed using the SWISS-MODEL modeling server (34), which adapted the atomic coordinates of the talin FERM domain (Protein Data Bank code 1Y19) as a template. The ␤1A integrin peptide model was generated using the coordinates of ␤3 integrin bound to talin F3 (Protein Data Bank code 1MK9), the side chains of which were replaced with the ␤1A integrin sequences, and then refined using the program CNS (35). The docking model of the complex between MIG-2 F3 and ␤1A or ␤3 integrin was constructed according to the atomic coordinates of the talin F3/␤3 integrin interaction observed in the crystallographic asymmetric unit. Model inspection and analysis of the MIG-2 F3 subdomain⅐␤1A integrin complex were performed using the TURBO-FRODO program (36). The figure was generated using the PyMOL program (37).

MIG-2 Interacts with ␤1 and ␤3 Integrin Cytoplasmic
Domains-To test whether MIG-2 interacts with the ␤1 integrin cytoplasmic domain, we incubated lysates of MIG-2-expressing mammalian cells with GST or GST fusion protein containing the ␤1〈 integrin cytoplasmic domain (residues 775-786; referred to as GST-␤1). GST, GST-␤1 fusion protein, and associated proteins were precipitated with glutathione-Sepharose beads. Western blot analyses showed that MIG-2 was readily coprecipitated with GST-␤1 (Fig. 1A, lane 3) but not with GST (lane 2). The presence of GST and GST-␤1 in the precipitates was confirmed by staining the membrane with Coomassie Blue (Fig. 1B). In additional control experiments, no protein bands in the GST-␤1 fusion protein preparation were recognized by anti-MIG-2 Ab in the absence of cell lysates (Fig.  1A, lane 4), confirming the specificity of the Western blotting.
The ␤3 integrin tail shares considerable sequence similarity with the ␤1 integrin tail. To test whether MIG-2 recognizes the ␤3 integrin tail, we incubated the cell lysates with GST fusion protein containing the ␤3 integrin tail (residues 716 -762; referred to as GST-␤3) and analyzed MIG-2 binding by the GST pulldown assay. The results show that MIG-2 was readily coprecipitated with GST-␤3 (Fig. 1, C and D, lanes 3) but not with GST (lanes 2). In control experiments, no protein bands in the GST-␤3 fusion protein preparation were recognized by anti-MIG-2 Ab in the absence of cell lysates (Fig. 1, C and D,  lanes 4). These results suggest that MIG-2 interacts with both ␤1 and ␤3 integrin cytoplasmic domains.
Structure-based Mutagenesis Identifies a Single Integrinbinding Site within the MIG-2 FERM Domain-MIG-2 contains multiple protein-binding motifs, including a C-terminal FERM domain. To facilitate studies aimed at determining the functions of the MIG-2/integrin interaction, we sought to identify the MIG-2 site(s) that are involved in integrin binding. The F3 subdomain within the MIG-2 FERM domain shares considerable sequence similarity with the F3 subdomain of the talin FERM domain, including Trp 359 (Trp 615 in MIG-2) and several other residues that are at the integrin-binding interface ( Fig.  2A) (38). To better understand the structural basis of the MIG-2/integrin interaction, we constructed a homology model of the MIG-2 F3 subdomain bound to the ␤1 or ␤3 integrin cytoplasmic tail using the atomic coordinates of the talin F3 subdomain⅐␤3 integrin complex as a template (Fig. 2B). Consistent with the ␤ integrin tail pulldown experiments, structural modeling suggested that the MIG-2 C-terminal region folds into a canonical FERM phosphotyrosine-binding domain capable of recognizing ␤ integrin tails. The integrin-binding surface mostly lies on its ␤5 strand through polar and hydrophobic interactions as well as intermolecular backbone hydrogen interaction. In particular, Gln 614 and Trp 615 of the MIG-2 F3 subdomain appear to play a significant role (Fig. 2B) by interacting with Trp 775 and Asp 776 in the ␤1 integrin tail or Trp 739 and Asp 740 in the ␤3 integrin tail via hydrophobic and hydrophilic interactions, resembling those in the talin F3 subdomain⅐␤3 integrin tail complex (38).
The MIG-2-and Talin-binding Sites in the ␤3 Integrin Cytoplasmic Domain Are Not Identical-Although the F3 subdomain of MIG-2 shares considerable sequence homology with that of talin (Fig. 2, A  and B), overlaying the MIG-2 F3 model structure with the talin F3 structure revealed that the S1-S2 loop, which is crucial for talin interaction with the membrane proximal region of the ␤ integrin cytoplasmic domains (39), is absent in the MIG-2 F3 subdomain (Fig. 2C). There are also other differences between MIG-2 and talin binding. The interaction of the talin FERM domain with the ␤ integrin tails has been extensively characterized (reviewed in Ref. [7][8][9][10]. Structurally, the talin-binding site encompasses both the membrane proximal and central regions of the ␤ integrin cytoplasmic tails (11,38,39) The first NPXY (Tyr 747 in the ␤3 integrin tail) site located in the central region of the ␤ integrin tails is crucial for the formation of the talin⅐␤ integrin complexes. It has been shown, for example, that substitution of Tyr 747 with Ala abrogates the talin-binding activity (40,41). To test whether this site is required for the formation of the MIG-2⅐␤3 integrin complex, we incubated cell lysates with a GST-␤3 protein bearing the Y747A substitution mutation and precipitated it with glutathione-Sepharose beads. Western blotting of the GST-Y747A precipitates showed that MIG-2 was readily pulled down by GST-Y747A (Fig. 1, C and D, lanes 5), suggesting that the first NPXY site is not essential for MIG-2 binding. Collectively, these results suggest that, although the MIG-2/ integrin and talin/integrin interactions may share certain structural similarities, they are not identical. The key residues of MIG-2 at the integrin-binding interface are highlighted in red and blue as described for A. The residues colored in blue (Gln 614 and Trp 615 ) were used for the mutational studies (see "Results"). C, overlay of the MIG-2 F3 subdomain model structure (blue) and the talin F3 subdomain structure (yellow) (39). The dotted circle highlights the difference in the S1-S2 loop between the MIG-2 and talin F3 subdomains. C-term, C terminus. D and E, the MIG-2 F3 subdomain mediates integrin binding. SK-LMS-1 cells were transfected with vectors encoding GFP-tagged wild-type MIG-2 (lanes 3, 7, and 8), the MIG-2 FERM deletion mutant (residues 1-486) (lanes 2, 5, and 6), and the MIG-2 Q614A/W615A substitution mutant (QW; lanes 1, 9, and 10). A GST pulldown experiment was performed using cells expressing GFP-tagged wild-type MIG-2 or mutants as indicated. The cell lysates (10 g of protein/lane; lanes 1-3), GST precipitates (lanes 6, 8, and 10), GST-␤1 precipitates (lanes 5, 7, and 9), and GST-␤1 in the absence of cell lysates (lanes 4) were analyzed by Western blotting with anti-MIG-2 mAb 3A3 (D) or by Coomassie blue staining (E). Note that GFP-MIG-2 (lanes 7), but neither GFP-⌬FERM (lanes 5) nor GFP-Q614A/W615A (lanes 9), coprecipitated with GST-␤1. JULY 13, 2007 • VOLUME 282 • NUMBER 28

JOURNAL OF BIOLOGICAL CHEMISTRY 20459
Integrin Binding Is Essential for MIG-2 Localization to FAs-We showed previously that MIG-2 is a component of FAs (20). However, the interaction that mediates MIG-2 localization to FAs was not known. To test whether integrin binding plays a role in this process, we transfected cells with vectors encoding GFP-tagged wild-type MIG-2 or integrin binding-defective mutant ⌬FERM or Q614A/W615A. The transfectants were stained with antibodies that specifically recognize components of FAs (e.g. vinculin, paxillin, and ␣-parvin). GFP-MIG-2 ( Fig.  3A) was readily recruited to FAs where abundant vinculin (Fig.  3B) and other FA proteins such as paxillin and ␣-parvin (data not shown) were detected. By contrast, the integrin bindingdefective ⌬FERM mutant (Fig. 3C) or the Q614A/W615A substitution mutant (Fig. 3E) failed to localize to FAs where clusters of vinculin were detected (Fig. 3, D and F). These results suggest that integrin binding is essential for MIG-2 localization to FAs.
MIG-2 Can Promote Integrin Activation and Cell-ECM Adhesion-Binding of the talin FERM domain to the ␤ integrin cytoplasmic domains is a key step in integrin activation. Overexpression of FERM domain-containing talin-H in ␣IIb␤3 integrin-expressing CHO cells, a well described model system for integrin activation, promotes ␣IIb␤3 integrin activation (as measured by activation-specific anti-␣IIb␤3 integrin mAb PAC1) (28,40,42). We used the same cell system to assess whether MIG-2 can play a role in integrin activation. Like human MIG-2, hamster MIG-2 bound readily to the ␤3 integrin cytoplasmic domain (Fig. 4, A and B, lanes 3). MIG-2 binding was not abolished by the Y747A mutation in the ␤3 integrin tail (Fig. 4, A and B, lanes 5). To test whether MIG-2 can promote ␤3 integrin activation, we overexpressed GFP-MIG-2, GFPtalin-H as a positive control, and GFP as a negative control in ␣IIb␤3 integrin-expressing CHO cells. In seven transfection experiments that we performed, the expression level of GFP-MIG-2, as indicated by the MFI, was only 15-20% of that of GFP-talin-H. Despite the relatively low expression level, flow cytometry analyses of the cells showed that, in all seven experiments, overexpression of MIG-2 increased mAb PAC1 binding (Fig. 4C). Although modest, this increase induced by MIG-2 was significant (p Ͻ 0.05 compared with the GFP control by paired t test analysis) but was substantially less than that induced by talin-H (Fig. 4D), which itself gives only partial activation (28). Thus, MIG-2 can promote integrin activation, albeit weakly. This weak activation may be due to the low MIG-2 expression level and/or to additional regulatory requirements for optimal effects.
To confirm that MIG-2 promotes integrin activation, we analyzed the adhesion of ␣IIb␤3 integrin-expressing CHO cells to fibrinogen at 4°C. Under this experimental condition, the strength of cell-ECM adhesion is controlled primarily by integrin ligand-binding activity (16). As expected, overexpression of talin-H markedly increased CHO cell adhesion to fibrinogen (Fig. 4E). Notably, overexpression of MIG-2 also significantly increased CHO cell adhesion to fibrinogen (Fig. 4E). These results are highly consistent with the results of the integrin activation assay (Fig. 4, C and D). Collectively, these results reveal a role of MIG-2 in promoting integrin activation and cell-ECM adhesion.

Many Colon Cancer Cell Lines Express No or Very Low Levels of MIG-2 Proteins-Because alteration of integrinmediated cell-ECM adhesion is intimately associated with
human diseases such as cancer, we next investigated the roles of MIG-2 in cancer cells. As an initial step, we analyzed the levels of MIG-2 in various cancer cells. Previous RNA profiling studies have shown that the MIG-2 mRNA level is reduced in colon carcinoma cells (seeharvester.embl.de/ harvester/Q96A/Q96AC1.htm and genome-www.stanford. edu/nci60) (43,44). Thus, we focused our analyses on these cells. SK-LMS-1 leiomyosarcoma cells, which express MIG-2, were used as a positive control. Fig. 5A shows that, although MIG-2 was readily detected in SK-LMS-1 cells (lane 7), no or very low levels of MIG-2 were detected in many colon carcinoma cell lines, including RKO, HT-29, DLD-1, LoVo, and HCT-116 (lanes 2-6). These results are consistent with the RNA profiling studies, suggesting that the level of MIG-2 protein, like that of MIG-2 mRNA, is diminished or lost in many colon cancer cells. One notable exception is Caco-2 cells (Fig.  5A, lane 8), a relatively well differentiated colon carcinoma cell line derived from primary colon tumor. Although Caco-2 cells expressed the highest level of MIG-2 among the colon carcinoma cell lines that were tested, their MIG-2 protein level was lower than that of SK-LMS-1 cells (Fig. 5A, compare lanes 7 and  8). HT-1080 cells, a highly metastatic fibrosarcoma cell line, also expressed a relatively low level of MIG-2 (Fig. 5A, lane 1). Interestingly, the order of abundance was reversed when the same samples were probed with anti-kindlin mAb (Fig. 5B). In control experiments, similar levels of actin were detected in all cell lines (Fig. 5C). To facilitate the functional studies, we generated adenoviral expression vectors encoding wild-type MIG-2 and the integrin binding-defective Q614A/W615A mutant as described under "Experimental Procedures." SK-LMS-1 cells were infected with the adenoviral vectors encoding MIG-2, the Q614A/W615A mutant, or ␤-galactosidase (as a negative control). The infection efficiency was monitored by the expression of GFP encoded by adenoviruses. Fluorescence microscopic analyses showed that 80 -90% of the cells expressed GFP 1 day after infection. Overexpression of MIG-2 and the Q614A/W615A mutant in the corresponding infectants was confirmed by Western blotting (Fig. 6A, compare lanes 2 and 3 with lane 1). As expected, a similar amount of GFP was detected by Western blotting in all three transfectants (Fig. 6A).
We next assessed the effect of MIG-2 and the integrin binding-defective Q614A/W615A mutant on SK-LMS-1 cell-ECM adhesion using the centrifugation method (16). To do this, we plated cells on a collage I-coated surface and analyzed cell adhesion either at 4°C, a condition that permits integrin-ECM ligand interaction but not cytoskeletal strengthening (15,16), or at 37°C, a condition that permits cytoskeletal strengthening (15,16). When cell adhesion was analyzed under the condition that does not permit cytoskeletal strengthening (i.e. at 4°C), no significant differences between the control cells and cells overexpressing MIG-2 or the Q614A/W615A mutant were observed (Fig. 6B). In control experiments, we treated MIG-2overexpressing cells with function-blocking antibodies to ␣2␤1 integrin, a major collagen-binding receptor. Treatment of the cells with either function-blocking anti-␣2 or anti-␤1 integrin Ab nearly completely eliminated the cell-collagen adhesion (Fig. 6C), confirming that the adhesion of these cells to collagen was mediated by ␣2␤1 integrin.
When the integrin-mediated cell adhesion was analyzed under the condition that permits cytoskeletal strengthening (i.e. at 37°C), a significant increase in cell adhesion was observed in response to overexpression of MIG-2 (Fig. 6D). Notably, the increase in the cytoskeletal strengthened cell-ECM adhesion was eliminated by the Q614A/W615A mutation, which disrupted integrin binding (Fig. 6D), suggesting that integrin binding is essential for MIG-2-mediated cytoskeletal strengthening of cell-ECM adhesion.
To further study the influence of MIG-2 on cell-ECM adhesion, we analyzed the effect of the MIG-2/integrin interaction on FA formation. To do this, we plated SK-LMS-1 cells on collagen I and immunofluorescently stained them with mAb to vinculin, a marker of FAs. Under the conditions used, the control cells formed relatively weak vinculin-rich clusters (Fig. 6E). Overexpression of MIG-2 (Fig. 6F), but not of the integrin binding-defective Q614A/W615A mutant (Fig. 6G), induced the formation of much more evident vinculin clusters. These results are highly consistent with the results of the centrifugation cell-ECM adhesion experiments. Collectively, these results suggest that the MIG-2/ integrin interaction contributes to cytoskeletal strengthening of SK-LMS-1 cell-ECM adhesion.
Next, we analyzed cell-ECM adhesion and FA formation in MIG-2-null RKO colon carcinoma cells. MIG-2-null RKO colon carcinoma cells were able to adhere to ECM (Fig. 7, C

MIG-2/Integrin Interaction in Cell Adhesion and Migration
and F-H), suggesting that MIG-2 is not absolutely required for integrin/ECM interactions. Immunofluorescence staining with anti-vinculin mAb showed, however, that the MIG-2-null RKO colon carcinoma cells were devoid of large FAs (Fig. 7C). To test whether this is caused by the lack of MIG-2, we expressed MIG-2 in the MIG-2-null cells by infecting them with adenoviral vectors encoding MIG-2 or ␤-galactosidase (as a control). Expression of MIG-2 in the MIG-2 infectants (Fig. 7A, lane 2), but not in the ␤-galactosidase infectants (lane 1), was confirmed by Western blotting. The level of MIG-2 in the MIG-2 adenovirus-infected RKO cells (Fig. 7B, lane 2) was ϳ2-3-fold above that of endogenous MIG-2 in SK-LMS-1 cells (lanes 3 and 4) but substantially less than the MIG-2 level in SK-LMS-1 cells infected with the MIG-2 adenovirus (lane 5). Expression of MIG-2 prompted the formation of FAs that could be readily detected by anti-vinculin mAb (Fig. 7D). Thus, the deficiency of large FAs in colon cancer cells is indeed caused by the lack of MIG-2 protein.
To test whether the MIG-2-induced FA formation depends on integrin binding, we infected RKO cells with adenovirus encoding the integrin binding-defective Q614A/W615A mutant. Expression of the Q614A/W615A mutant was confirmed by Western blotting (Fig. 7A, lane 3). Expression of the integrin binding-defective mutant, unlike that of wild-type MIG-2, failed to induce the formation of large FAs (Fig. 7, compare D and E). These results confirm the specificity of the MIG-2-induced FA formation. Furthermore, they suggest that MIG-2 promotes FA formation in colon cancer cells through its interaction with the integrins.
Next, we employed the centrifugation assay to quantify the effect of the MIG-2/integrin interaction on colon cancer cell-ECM adhesion. The results show that MIG-2, but not the inte-  grin binding-defective Q614A/W615A mutant, significantly enhanced the cytoskeletal strengthening phase of RKO cell adhesion to collagen (Fig. 7F). No significant increase in the initial cell-collagen adhesion was observed in response to the expression of MIG-2 or the Q614A/W615A mutant (Fig. 7G).
In control experiments, treatment of the cells with function-blocking anti-␣2 or anti-␤1 integrin Ab nearly completely eliminated the cell-collagen adhesion, confirming that ␣2␤1 integrin mediates the cell-collagen adhesion. Collectively, these results suggest that the MIG-2/integrin interaction functions primarily in the cytoskeletal strengthening of cell-ECM adhesion in these colon cancer cells.
The MIG-2/Integrin Interaction Regulates Cancer Cell Motility-Cell-ECM adhesion is intimately involved in the regulation of cell behavior such as cell motility. The findings that the MIG-2/integrin interaction promotes cell-ECM adhesion raised an interesting possibility that it may play a role in the regulation of cell motility. The MIG-2-null RKO colon carcinoma cells were highly motile (Fig. 8A). Consistent with the strengthening of colon cancer cell-ECM adhesion, the expression of MIG-2 significantly reduced colon cancer cell migration (Fig. 8, A, B, and D). No significant inhibition of cell migration was observed in RKO cells expressing the integrin binding-defective Q614A/W615A mutant (Fig. 8, C and D), suggesting that, through integrin binding, MIG-2 suppresses colon cancer cell migration.
To test whether MIG-2 plays a role in the regulation of SK-LMS-1 cell motility, we compared the motility of SK-LMS-1 cells overexpressing MIG-2 or the integrin binding-defective Q614A/W615A mutant with that of the control cells. The results show that the MIG-2-overexpressing SK-LMS-1 cells migrated substantially slower than the control cells (Fig. 9, A, B,  and D). No alteration of cell migration was observed in cells overexpressing the integrin binding-defective Q614A/W615A mutant (Fig. 9, A, C, and D). Thus, consistent with the results  obtained with the colon carcinoma cells, the MIG-2/integrin interaction suppresses leiomyosarcoma cell motility.

DISCUSSION
MIG-2 is a widely expressed and evolutionarily conserved FA protein (17)(18)(19)(20)(21). The experiments presented in this study have demonstrated a specific interaction between MIG-2 and the ␤1 and ␤3 integrin cytoplasmic domains. Using a structure-based mutagenesis approach, we have characterized the sites that mediate the MIG-2/integrin interaction. Functionally, we have shown that the MIG-2/integrin interaction is essential for recruiting MIG-2 to FAs. In addition, using CHO cells expressing ␣IIb␤3 integrin, a well described model system for integrin activation, we found that MIG-2 can weakly promote integrin activation. Consistent with a positive role in promoting ␣IIb␤3 integrin activation, MIG-2 enhances the adhesion of ␣IIb␤3 integrin-expressing CHO cells to fibrinogen. Finally, we have investigated the functions of MIG-2 in cancer cells. Using two different types of cancer cells (MIG-2-expressing leiomyosarcoma cells and MIG-2-null colon carcinoma cells), we have demonstrated that the MIG-2/integrin interaction promotes cell-ECM adhesion and FA formation and reduces cell motility. These results provide new insights into the molecular basis underlying the cellular control of cell-ECM adhesion. Furthermore, our findings suggest that alterations of certain components (e.g. MIG-2) that participate in the cellular control of cell-ECM adhesion may contribute to high motility of certain types of cancer cells (e.g. colon carcinoma cells).
Structurally, the MIG-2/integrin interaction shares certain but not all features with the talin/integrin interaction. One of the most obvious common features of MIG-2 and talin is that both contain FERM domains. In particular, the F3 subdomains within the MIG-2 and talin FERM domains share considerable sequence homology. On the basis of the sequence homology, we constructed a model of the MIG-2 F3 subdomain bound to the ␤1 integrin tail, which suggests a canonical FERM phosphotyrosine-binding domain fold capable of recognizing the ␤1 integrin tail. On the basis of the structural modeling, we predicted that several residues, including Gln 614 -Trp 615 , within the MIG-2 F3 subdomain constitute the integrin-binding interface. We predicted, for example, that Gln 614 and Trp 615 in the MIG-2 F3 subdomain directly interact with Trp 775 and Asp 776 in the ␤1 integrin tail via hydrophobic and hydrophilic interactions. Consistent with this model, substitution of Gln 614 -Trp 615 with alanine residues abrogated integrin binding. The fact that the Gln 614 -Trp 615 substitution mutation eliminated integrin binding also suggests that other regions of MIG-2 are incapable of interacting with the ␤ integrin tails with high affinity.
Although it is clear that the MIG-2 FERM F3 subdomain folds into a three-dimensional integrin-binding structure that resembles, at least globally, the integrin-binding structure of talin, our study has revealed that the first NPXY (Tyr 747 in the ␤3 integrin tail) site in the central region of the ␤ integrin cytoplasmic domains, which is crucial for the formation of the talin⅐␤ integrin complex, is not required for the formation of the MIG-2⅐␤ integrin complex. Thus, although it is likely that the MIG-2-and talin-binding interfaces share certain common residues, there must be a structural difference in the precise integrin-binding modes between MIG-2 and talin. For example, the ␤6-␤7 loop in the talin F3 subdomain, which is responsible for recognizing integrin Tyr 747 (38), appears to differ from that in the MIG-2 FERM F3 subdomain. The latter appears to make significantly less contact with the NPXY region of the integrin tail in our model.
At the cellular level, one important function of the MIG-2/ integrin interaction revealed by this study is that this interaction is essential for recruiting MIG-2 to FAs. It is interesting to compare the MIG-2/integrin interaction with the MIG-2/migfilin interaction. Migfilin is another MIG-2-binding protein that is also found at FAs (20 -22). However, unlike integrin binding, migfilin binding is not required for MIG-2 localization to FAs (20). Instead, the interaction of MIG-2 with migfilin recruits migfilin to FAs (20), which in turn facilitates the localization of certain migfilin-binding FA proteins such as VASP to the adhesion sites (23). Thus, integrins, MIG-2, migfilin, and VASP appear to localize to FAs in a sequential fashion. This sequential localization mechanism is probably utilized by some other, but clearly not all, FA components during the assembly of FAs. Certain FA proteins, notably PINCH, integrin-linked kinase, and ␣-parvin, are first assembled into a multicomponent protein complex and then simultaneously localize to FAs (29,45). Through both mechanisms, a large number of components are recruited to FAs, where they regulate a variety of cellular processes.
In general, integrin-mediated cell-ECM adhesion can be regulated through two distinct mechanisms. The first one is through regulation of integrin activation. Recent studies have demonstrated that binding of talin-H to ␤ integrin tails plays a key role in integrin activation (for recent reviews, see Refs. 7-10). Since Horwitz et al. (46) discovered that talin interacts with integrins more than 2 decades ago, Ͼ20 ␤ integrin tail- Interestingly, among all the cytoplasmic ␤ integrin-binding proteins that had been tested, only talin had been shown to be able to promote integrin activation (24,41,48). The results presented in this study demonstrate that MIG-2 can also play a role in this process. Although this is clearly an exciting finding, several lines of evidence suggest that the role of MIG-2 in integrin activation is not identical to that of talin. First, the extent to which MIG-2 promotes integrin activation appears to be smaller compared with talin-H. This could be due to the relatively low expression level of GFP-MIG-2 (compared with GFPtalin-H) in our experiments. However, despite our efforts (we performed seven transfection experiments), we have not been able to further increase the expression level of GFP-MIG-2 in CHO cells. Alternatively (but not necessarily mutually exclusively), the relatively weak effect of MIG-2 on integrin activation could be an intrinsic property of MIG-2.
A second and perhaps even more fundamental difference is that the MIG-2-binding site in the ␤3 integrin tail is not identical to the talin-binding site. The first NPXY (Tyr 747 in the ␤3 integrin tail) site located in the central region of the ␤ integrin tails is crucial for talin binding. However, this site is not required for MIG-2 binding (Figs. 1C and 4A, lane 5). Although the precise mode by which MIG-2 binds to ␤ integrin tails remains to be determined, it is important to note that MIG-2 lacks several key residues in the S1-S2 loop that are crucial for interacting with the integrin membrane proximal region. For example, Leu 325 in the S1-S2 loop of talin-H is vital for interacting with the ␤3 integrin membrane proximal region (39). However, this residue is not present in MIG-2 (Fig. 2, A and C). Previous studies have shown that a proline substitution mutation at Ser 752 , which is found in a variant of Glanzmann thrombasthenia, suppresses ␣IIb␤3 integrin activation (49,50). The S752P point mutation does not impair talin-H binding (28). Interestingly, preliminary studies suggest that the S752P point mutation impairs the binding of MIG-2 to the ␤3 integrin tail. 4 Thus, MIG-2 appears to interact with the membrane distal region rather than the membrane proximal region in the ␤3 integrin tail. It is widely accepted that interactions of talin-H with the membrane proximal region and consequently separation of the membrane proximal "clasp" are final intracellular steps in integrin activation (7)(8)(9)(11)(12)(13). On the other hand, the membrane distal region appears to regulate integrin activation indirectly (28,51). On the basis of these considerations, we propose that MIG-2 promotes integrin activation through an indirect mechanism. We are currently investigating the threedimensional structure of the MIG-2⅐␤ integrin tail complexes, which should help to shed light on the mechanism through which MIG-2 promotes integrin activation.
A third difference between talin and MIG-2 is that, although talin has been proposed to serve as a common activator of integrins in the final intracellular step of integrin activation (41), the role of MIG-2 in integrin activation appears to be contextdependent. For example, MIG-2-null colon cancer cells can adhere to the ECM, suggesting that MIG-2 is not absolutely required for integrin activation in these cells. Consistent with this, the expression of MIG-2 in the MIG-2-null colon cancer cells fails to significantly increase the initial cell-ECM adhesion, a process that is controlled primarily by integrin ligand-binding activity (15,16).
The second mechanism by which cell-ECM adhesion can be regulated is through cytoskeletal strengthening (14 -16). The results presented in this work suggest that the MIG-2/integrin interaction plays a prominent role in this process. This is based on two lines of evidence. First, quantitative centrifugation cell adhesion assays showed that overexpression of MIG-2, but not of the integrin binding-defective MIG-2 mutant, in SK-LMS-1 leiomyosarcoma and RKO colon cancer cells substantially enhanced the cytoskeletal strengthening phase, but not the cytoskeletal-independent initial phase, of cell-collagen adhesion. Second, immunofluorescence microscopic analyses showed that overexpression of MIG-2, but not of the integrin binding-defective MIG-2 mutant, enhanced the formation of FAs. Thus, the MIG-2/integrin interaction likely represents a key element in the cellular machinery that controls the cytoskeletal strengthening of cell-ECM adhesion.
It has been well established, by both theoretic considerations and experimental studies, that cell motility is favored by an intermediate level of cell-ECM adhesion (25,26). Consistent with this model, we have found that overexpression of MIG-2, which enhances cell-ECM adhesion, reduces the motility of leiomyosarcoma and colon cancer cells. These results suggest that the MIG-2-mediated cytoskeletal strengthening of cell-ECM adhesion likely serves as an "adhesive brake" for the movement of these cells. Loss of this adhesive brake could contribute to high motile behavior of certain diseased cells. Interestingly, large-scale RNA profiling studies have shown that the MIG-2 mRNA level is reduced in colon cancer cells (see harvester.embl.de/harvester/ Q96A/Q96AC1.htm and genome-www.stanford.edu/nci60) (43,44). We have confirmed in this study that the MIG-2 protein level is indeed diminished in a number of colon cancer cell lines. Notably, introducing this adhesive brake (i.e. MIG-2) into MIG-2-null colon cancer cells results in an increase in cell-ECM adhesion and a reduction of cell motility. Because expression of the integrin binding-defective MIG-2 mutant (Q614A/ W615A) in MIG-2-null colon cancer cells did not significantly alter colon cell-ECM adhesion and motility, integrin binding appears to be essential for MIG-2-mediated regulation of colon cell-ECM adhesion and motility.