Actin Activates a Cryptic Dimerization Potential of the Vinculin Tail Domain*

The tail domain of vinculin (Vt) is an actin binding module containing two regions that interact with F-actin. Although intact Vtpurified from a bacterial expression system is a globular monomer, each actin binding region dimerizes when expressed individually, suggesting the presence of cryptic self-association sites whose exposure is regulated. We show that actin modulates Vt self-association by inducing or stabilizing a conformational change in Vtthat allows dimerization. Chemical cross-linking studies implicate one of the actin binding regions in mediating dimerization in the presence of actin. Actin-induced Vt dimers may play a role in the filament cross-linking activity of this protein. The Vtdimers induced by actin are biochemically distinct from the Vt dimers and higher oligomers induced by acidic phospholipids such as phosphatidylinositol 4,5-bisphosphate, suggesting structural differences in Vt bound to these two ligands that may provide a mechanistic basis for inhibition of F-actin binding by phosphatidylinositol 4,5-bisphosphate. The ability of actin to regulate the dimerization state of an actin binding protein suggests that, rather than serving a passive structural role, actin filaments may directly participate in signal transduction and other cellular events that are known to depend on cytoskeletal integrity.

Adherens junctions are complex and dynamic cytoskeletal microenvironments formed by the interplay of actin, actinbinding proteins, and cellular signaling pathways (1,2). Assembly of one type of adherens junction, the integrin-mediated focal adhesion, is regulated by members of the Rho family of small GTPases (3), through mechanisms that may lead to conformational changes in specific junctional components allowing their associations with one another to occur (4,5). The availability of ligand binding sites in the adherens junction protein vinculin is regulated by an intramolecular interaction between the head (V h ) and tail (V t ) domains of the protein, which is thought to provide a mechanism for regulated recruitment of vinculin to junctions (4,6). Members of the ERM protein family are other examples of junctional proteins in which masking of ligand binding sites in the native conformation is thought to control localization and function (7,8). Whatever the specific mechanisms regulating recuitment of junctional components, once coassembled these proteins act in concert to create a dynamic, mechanical linkage to the extracellular substratum (9,10).
Adherens junctions also provide a conduit for information transfer between the cell and its surroundings. A variety of protein kinases, phosphatases, and other cell signaling molecules localize at least transiently in adherens junctions (1,11) and mediate adhesion-dependent control over such processes as cell division and apoptosis (2). Although the traditional view of the functional organization of an adherens junction is one of tenant signaling molecules residing dynamically within a framework of cytoskeletal elements (12), the extent to which components traditionally considered structural, such as actin, are in fact direct participants in junctional signaling events remains an unexplored question. In this report, we provide evidence that actin induces structural and functional changes in the V t domain of vinculin.
V t is an actin-binding module whose activity is regulated by the intramolecular interaction with the V h domain (4). Several observations suggest that, if relieved of this regulatory intramolecular interaction in vivo, the V t domain would mediate association of vinculin with F-actin. In vitro, free V t domain binds F-actin and cross-links actin filaments into bundles and gels (4,13), and a peptide ligand specific for the open conformation of vinculin can activate F-actin binding of intact vinculin when present at high molar excess (14). Expression of V t in vinculin-deficient F9 cells results in a decrease in actin-dependent cell motility, which can be reversed by coexpression of V h (15). In fibroblasts, V t localizes to actin stress fibers and to focal adhesions when microinjected (13) or when expressed as a fusion with green fluorescent protein, in contrast to both native vinculin and green fluorescent protein-vinculin, which localize exclusively to focal adhesions. 1 Conformational regulation of V t binding to actin in intact vinculin suggests that this interaction is significant to vinculin function in vivo.
In studies seeking to identify the sequences in V t responsible for actin binding, we found two regions of the tail, corresponding to amino acids 940 -1012 and 1012-1066 of the chicken vinculin sequence (16), that individually appeared to constitute an actin binding motif when expressed as a fusion with glutathione S-transferase (17). This observation was confirmed by another group using similar tail constructs produced as fusions with the maltose-binding protein (18). The presence of two actin binding regions in V t suggested that this domain crosslinks F-actin by a mechanism involving simultaneous engagement of two actin filaments by a single V t molecule. However, the existence of multiple actin binding regions does not rule out the possibility that V t self-association is important for actin cross-linking. Electron microscopic evidence suggests that vinculin molecules in which the V h -V t interaction has been disrupted may be capable of oligomerization through their V t domains (19,20). Furthermore, maltose-binding protein fusions containing the actin binding regions of V t are reported to self-associate (18).
We report here that purified V t domain is a monomer but that F-actin induces its dimerization through a mechanism involving conformational changes in V t that result in exposure of cryptic self-association sites in at least one of the actin binding regions. This insight allows us to begin to address the role actin itself plays in controlling the activity of an actinbinding protein. The finding that actin modulates the conformation and possibly the activity of an actin binding protein implies that rather than serving as a passive stage on which signaling events occur, actin filaments can instead play an active role in such cellular processes.

EXPERIMENTAL PROCEDURES
Reagents-Vinculin 30-kDa tail fragment was produced by cleavage of purified chicken gizzard vinculin with V8 protease from S. aureus as described (6). Actin was purified from chicken skeletal muscle according to standard procedures (21,22). PI 2 was purchased from Avanti Polar Lipids (Alabaster, AL), and PIP 2 was purchased from Sigma. Sephadex G75 and G150 superfine gel filtration media were from Sigma. The chemical cross-linkers DSS, DSP, EDC, and MBS were purchased from Pierce. 5,5Ј-Dithiobis(2-nitrobenzoic acid) (Ellman's reagent) was from Pierce and was used according to standard procedures (23). Horseradish peroxidase-conjugated goat anti-rabbit and goat anti-mouse polyclonal antibodies were purchased from Roche Molecular Biochemicals. C4 anti-actin monoclonal antibody was the generous gift of Dr. James Lessard (Children's Hospital Medical Center; Cincinnati, OH). Polyclonal rabbit antibody recognizing the V t domain was raised by us using the proteolytically derived 30-kDa tail fragment of vinculin as an immunogen.
Fusion Proteins-Expression plasmids encoding His-tagged fusion proteins were constructed by polymerase chain reaction amplification of the appropriate region of the chicken vinculin cDNA (16) and subcloning of the amplified region into the pET15b vector (Novagen; Madison, WI); all fusion proteins were constructed with a translation stop codon following the final vinculin residue. Construction of His 6 /V884 -1066 and His 6 /V916 -970 has been described previously (24). Fusion proteins were produced in Escherichia coli BL21(DE3) cells and purified by Ni 2ϩ affinity chromatography essentially according to vendor instructions. Briefly, fusion protein expression was induced by the addition of 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside (Roche Molecular Biochemicals) to bacteria grown at 37°C in 2ϫ YT medium to A 600 ϭ 0.8 -1.2. After 3-4 h, cells were pelleted (5000 ϫ g, 10 min, 4°C), resuspended in 20 mM Tris-HCl, 5 mM imidazole, 0.5 M NaCl, pH 7.9 (binding buffer) containing 0.1% Triton X-100 and sonicated. Sonicates were spun (20,000 ϫ g, 10 min, 4°C) to pellet insoluble material and applied to a Ni 2ϩ -charged His-Bind resin (Novagen). Purified fusion proteins were eluted with 1 M imidazole, 20 mM Tris-HCl, 0.5 M NaCl, pH 7.9, and dialyzed extensively against 10 mM EDTA in 10 mM Tris-HCl, 150 mM NaCl, 0.02% NaN 3 , pH 7.5, to remove residual Ni 2ϩ before further use. His tags were removed from fusion proteins by proteolytic cleavage (16 -36 h, 4°C) with biotinylated thrombin (Novagen) at 0.33 units of thrombin/mg of fusion protein, with subsequent removal of the protease on streptavidin-agarose beads. Cleaved fusion proteins were stored at 4°C in 10 mM Tris-HCl, 100 mM NaCl, 0.1 mM EDTA, 0.1 mM EGTA, 1 mM DTT, 0.02% NaN 3 , pH 7.5. Following thrombin cleavage, four nonvinculin residues (GSHM; single letter code) remain at the amino terminus of the fusion proteins. Fusion proteins used in this study will be referred to using the His 6 /V nomenclature if they retained their His tag (e.g. His 6 /V985-1066); for proteins that were thrombin-cleaved before use, only the vinculin residue numbers are indicated.
His 6 /V985-1066, representing the carboxyl-terminal region of the tail domain, was insoluble after expression in bacteria. Following initial lysis of bacteria expressing His 6 /V985-1066 under the nondenaturing conditions, the insoluble material containing the fusion protein was pelleted as described above and solubilized in binding buffer containing 7 M guanidine HCl. All subsequent purification steps were performed as described, except that buffers contained 7 M guanidine HCl. The purified protein was dialyzed against 10 mM Tris-HCl, 1 mM DTT, 3 M guanidine HCl, pH 7.5, and stored at 4°C. Denatured His 6 /V985-1066 was refolded by rapidly diluting 10 -20-fold into 50 mM Tris-HCl, 10 mM DTT, 0.1% CHAPS, pH 7.5, leaving a final guanidine concentration of 150 -300 mM. Refolded His 6 /V985-1066 could be cleaved with biotinylated thrombin, but cleavage required prolonged incubation (ϳ1 week, 4°C) to ensure complete proteolytic removal of the His tag. Thrombincleaved V985-1066 was stored in 10 mM Tris-HCl, 100 mM guanidine HCl, 0.1% CHAPS, 0.1 mM EDTA, 0.1 mM EGTA, 1 mM DTT, 0.02% NaN 3 , pH 7.5, to maintain solubility. Refolded His 6 /V985-1066 and thrombin-cleaved V985-1066 were soluble and well behaved for several weeks unless the residual concentration of guanidine was lowered below ϳ75 mM (e.g. by dialysis), at which point significant aggregation occurred within 1-2 days. Thus, 100 mM guanidine HCl was included in assay buffers in experiments using V985-1066 to maintain protein solubility.
Phospholipids-PI from a stock solution in chloroform was deposited on the bottom of a glass tube by solvent evaporation under a stream of N 2 and resuspended in H 2 O. A stock solution of PIP 2 was prepared by resuspension of the supplied powder in H 2 O. When first prepared, aqueous stock solutions of PI or PIP 2 were incubated at 37°C and sonicated extensively in a Branson bath sonicator at maximum power to yield small unilamellar vesicles of PI and micelles of PIP 2 . Lipid stocks were stored under N 2 and at Ϫ80°C. Immediately before use, lipid stocks were warmed to 37°C and resonicated at maximum power for 15 min.
Cosedimentation/Supernatant Depletion Assay-Actin in buffer A (2 mM Tris-HCl, 0.2 mM CaCl 2 , 0.2 mM ATP, 0.02% NaN 3 , pH 8.0) was polymerized (30 -60 min, 25°C) by the addition of 100 mM KCl and 2 mM MgCl 2 from a concentrated stock. One-tenth volume of the appropriate V t fusion protein in storage buffer was added to yield a final reaction volume of 50 l. For V985-1066, actin was instead polymerized in 30 mM KCl and 2 mM MgCl 2 , and 100 mM guanidine HCl and 0.1% CHAPS were added immediately before the addition of V985-1066. Following incubation with F-actin (60 min, 25°C), reactions were spun for 20 min at 95,000 ϫ g in a Beckman Airfuge to sediment actin filaments and bound fusion protein. Pellet and supernatant fractions were recovered and analyzed by SDS-PAGE followed by Coomassie Blue staining or immunoblotting. Actin binding was assessed by recovery of V t fusion protein in the F-actin pellet (cosedimentation) and/or by depletion of fusion protein from the supernatant (supernatant depletion).
Gel Filtration-Gel filtration analyses were performed at 4°C on 0.75 ϫ 50 cm columns composed of either Sephadex G75 or G150 superfine resin. With either resin, the total included volume (based on the elution volume of ATP) was approximately 20 ml, and the void volume (based on the elution volume of blue dextran) was approximately 7 ml, yielding an included volume of approximately 13 ml. The column buffer was 2 mM Tris-HCl, 100 mM KCl, 2 mM MgCl 2 , 0.2 mM CaCl 2 , 0.02% NaN 3 , 0.5 mM DTT, pH 8.0, for all fusion proteins except V985-1066, which was chromatographed in 2 mM Tris-HCl, 30 mM KCl, 100 mM guanidine HCl, 0.1% CHAPS, 2 mM MgCl 2 , 0.2 mM CaCl 2 , 0.02% NaN 3 , 0.5 mM DTT, pH 8.0, to maintain protein solubility. Although globular protein standards eluted essentially identically when chromatographed in either buffer, the apparent molecular weight of a test protein was always calculated based on the elution of globular standards under the same buffer conditions. Protein samples (50 g) were applied in a 200-l volume with 10% glycerol added to maintain a sharp sample boundary during loading. Fractions of ϳ330 l were collected and analyzed for protein content using the Bio-Rad protein assay reagent. Eluting V985-1066 was instead detected by SDS-PAGE and silver staining because of interference of buffer components (100 mM guanidine HCl, 0.1% CHAPS) with the Bio-Rad assay.
Chemical Cross-linking-For chemical cross-linking experiments, fusion proteins stocks were dialyzed extensively against CL buffer (10 mM sodium phosphate, 100 mM NaCl, 0.02% NaN 3 , pH 7.5). G-actin was dialyzed against buffer P (2 mM NaHCO 3 , 0.2 mM CaCl 2 , 0.2 mM ATP, 0.02% NaN 3 , pH 7.6). Cross-linking experiments were performed in CL buffer or (for experiments involving F-actin) in buffer P supplemented with 100 mM NaCl and 2 mM MgCl 2 from a concentrated stock to induce actin polymerization.
Immunoblotting-Following SDS-PAGE, proteins were transferred electrophoretically to nitrocellulose filters (25). Filters were blocked in phosphate-buffered saline containing 5% nonfat dry milk. Primary antibody was either a rabbit polyclonal antibody to the vinculin tail domain, which was used at a 1:3000 dilution of immune serum, or 1 g/ml C4 monoclonal anti-actin antibody. Secondary antibody was either goat anti-mouse IgG polyclonal antibody conjugated to horseradish peroxidase for detection of C4 anti-actin antibody or peroxidase-conju-gated goat anti-rabbit IgG polyclonal antibody for detection of polyclonal anti-tail antibody. Primary and secondary antibody steps (1-2 h each, 25°C) were performed in TBST (10 mM Tris-HCl, 150 mM NaCl, 0.2% Tween 20, pH 8.0) containing 1% bovine serum albumin. Filters were washed (3 ϫ 10 min, 25°C) in TBST after each antibody step. Immunoreactive bands were detected using the Renaissance chemiluminescence kit (NEN Life Science Products).
Sequence Analysis-The sequence of the chicken vinculin tail domain (V884 -1066) was analyzed for possible coiled-coil-forming regions using the MacStripe program written by Dr. A. E. Knight (University of York, York, United Kingdom), which is based on the Coils algorithm of Lupas (26) (MacStripe is available for download on the World Wide Web). The secondary structure content of V884 -1066 was analyzed using the PHD program (27,28) (available on the World Wide Web).

RESULTS
Two Actin-binding Subdomains of Monomeric V t Form Homodimers-We examined the oligomeric state of the tail domain of vinculin, isolated either by expression in a bacterial fusion system or by proteolysis of intact vinculin (6,29). V t and regions of V t were expressed in bacteria as His-tagged fusions because proteolytic removal of the His tags proved very efficient, with little loss of the fused V t portion. V t fusion proteins will be referred to by the corresponding chicken vinculin residue numbers (16), since the His tags were removed from fusions before use in most of the experiments described. The fusion protein V884 -1066 comprises most of the tail domain (16) and retains helical secondary structure (24) and F-actin binding activity (14). On gel filtration columns, V884 -1066 migrates as a globular, monomeric protein ( Fig. 1c and Table I).
The concentration of V884 -1066 eluting in this monomer peak was determined to be approximately 1 M, comparable with the K d for V t binding to F-actin and in the range of concentrations at which V t bundles and gels F-actin (4). Depending on the gel filtration medium used, V884 -1066 eluted with a K av corresponding to that of a globular protein of ϳ18 -21 kDa, essentially identical to the predicted monomeric size for this protein (20,942 Da) based on its primary structure.
The 30-kDa tail fragment, derived from cleavage of native vinculin with S. aureus V8 protease and corresponding to vinculin residues 858 -1066, eluted at the position of a globular protein of ϳ34.2 kDa. This value, although higher than the predicted molecular mass of this fragment (ϳ23.3 kDa), is significantly lower than the expected size of a dimer of this species (ϳ47 kDa), suggesting that the additional proline-rich segment composed of residues 858 -884 produces a more extended, asymmetric shape to the 30-kDa fragment. It is noteworthy that, as its name implies, the 30-kDa tail fragment also migrates aberrantly on SDS-PAGE gels in comparison with its predicted mass.
Two regions of the tail domain, V907-984 and V985-1066, corresponding roughly to the amino-and carboxyl-terminal halves of V t , were also produced as fusion proteins in bacteria. These sequences were chosen because they overlap with the regions (V940 -1012 and V1012-1066) to which we had previously mapped F-actin binding sites (17) and because attempts to express and purify His-tagged fusions of V940 -1012 and V1012-1066 were unsuccessful due either to very poor expression or to extensive proteolysis of the proteins in bacteria (data not shown). While the amino-terminal V907-984 construct was well behaved, the carboxyl-terminal region (V985-1066) proved difficult to work with because of insolubility in the bacterial expression system, but it could be purified in a denatured state and refolded (see "Experimental Procedures"). As expected, both of the fusions retain the ability to cosediment with F-actin (Fig. 1b). The cosedimentation assays using His 6 / V985-1066 were performed under somewhat stringent buffer conditions, including 100 mM guanidine HCl and 0.1% CHAPS, to maintain protein solubility. Although both the His-tagged and the thrombin-cleaved V985-1066 aggregate over time in more physiological buffers (see "Experimental Procedures"), no significant sedimentation of His 6 /V985-1066 is observed in the absence of F-actin under the assay conditions used for this construct in Fig. 1. Binding of V907-984 to actin filaments is observed under standard actin cosedimentation assay conditions, where V907-984 does not pellet significantly on its own. Surprisingly, both of these subdomains of monomeric V t eluted from gel filtration columns as dimers (Fig. 1c). Gel filtration was performed in buffer conditions corresponding to those used in the actin cosedimentation assays above for each protein. V907-984 eluted approximately at the position of a dimer (21.5 kDa), based on its calculated monomeric size (9006 Da). Refolded V985-1066 eluted from gel filtration columns with a major peak at an apparent molecular mass of ϳ24.5 kDa and a smaller peak at ϳ14.7 kDa. In SDS-polyacrylamide gels, the material in the second V985-1066 peak was identical in migration to that in the first peak (data not shown), indicating a difference in oligomerization state rather than mass. Since the expected size of a V985-1066 monomer is 9757 Da, these data suggest that the V985-1066 region folds into an asymmetric monomer that self-associates almost quantitatively into a dimer. The His-tagged V985-1066 used in the actin binding study in Fig. 1b eluted as a single peak of a size comparable with the dimer of thrombin-cleaved V985-1066 (data not shown). Thus, both V907-984 and V985-1066 fusion proteins exist as dimers under conditions where both are functional with regard to F-actin binding.
Because the V985-1066 constructs aggregated over time in the absence of Ն75 mM guanidine HCl, we focused on the V907-984 region. We excluded the trivial possibility that dimerization of V907-984 was the result of aberrant disulfide bond formation occurring during expression in bacteria or subsequent storage by direct quantification of cysteine sulfhydryl content using the Ellman reagent. V907-984 displays the number of free sulfhydryl groups anticipated from its amino acid sequence (Table I). The sulfhydryl content per V907-984 molecule was stable over prolonged storage (Ͼ2 months) in buffer lacking reducing agents (data not shown), suggesting that the cysteine residues in this region are relatively insensitive to oxidation.
Chemical Cross-linking Confirms Distinct Oligomeric States of V t and Its Amino-terminal Half-We employed chemical cross-linking to generate independent evidence for the dimeric state of V907-984. The homobifunctional compound DSS quantitatively cross-links V907-985 into a species migrating in SDS-PAGE at the expected position of a cross-linked dimer (Fig. 2a). The dimeric state of the V985-1066 region could not be confirmed by chemical cross-linking, since this fusion protein does not react with DSS or a variety of other cross-linkers to yield a dimer (data not shown). In contrast to the V907-984 region, the full-length V t construct V884 -1066 is not crosslinked significantly by DSS to form a dimer-sized species, although small amounts of dimer-sized products of V884 -1066 can be observed with longer reaction times or higher concentrations of cross-linking reagent (Fig. 2a) or at higher concentrations of V884 -1066 (Fig. 3). Although little dimer-sized product of V884 -1066 is detectable following cross-linking, this protein reacts with DSS to form a species migrating slightly faster in SDS-PAGE gels than the unreacted monomer (Fig. 2a).
We hypothesized that the faster migrating V884 -1066 species observed after DSS cross-linking represented an intramolecularly cross-linked monomer of V884 -1066. To test this hypothesis, DSS cross-linking reactions were performed over a 30-fold range of V884 -1066 concentrations (Fig. 3). The rela-tive amounts of unreacted V884 -1066 and of the faster migrating species remained constant over this range, indicating a zero-order dependence on V884 -1066 concentration, which is indicative of an intramolecular reaction. In contrast, the low level of V884 -1066 dimers detectable is steeply dependent on V884 -1066 concentration, with little to no dimer seen below ϳ5 M V884 -1066. These data are consistent with the interpretation that the faster migrating species represents the product of intramolecular cross-linking of a V884 -1066 monomer and that V884 -1066 is predominantly monomeric in this range of protein concentrations. Cross-linking of V907-984 at various concentrations demonstrates that this region is dimerized essentially quantitatively throughout this range (Fig. 3).
Homodimers of V907-984 can also be detected by crosslinking with either MBS or EDC, although the reactivity with EDC is low and requires relatively high concentrations of crosslinker and prolonged incubation (Fig. 2b). A smaller aminoterminal region of the vinculin tail, V916 -970, also migrates as a dimer on gel filtration columns and is efficiently cross-linked to a dimer in the presence of DSS and MBS (data not shown). This region probably contains the same dimerization motif as the overlapping V907-984 region. Neither 30 -100 M MBS or 1-10 mM EDC yield significant amounts of dimers of full-length V884 -1066, although reaction with MBS does yield an intramolecularly cross-linked monomer species similar to that generated by DSS (data not shown). Taken together, the gel filtration and chemical cross-linking analyses indicate that V907-984 (and probably V985-1066) contains a site that mediates homodimerization of the isolated subdomain but not of intact V t under the assay conditions. Intact V t may have a weak dimerization potential that permits a low level of dimer formation at relatively high V t concentrations. These data suggest the presence of self-association sites in V t that are either masked or of low affinity in the native structure of this domain.
Actin Induces Dimerization of Vinculin Tail Domain-The coincidence of cryptic dimerization sites in V t with the F-actin binding sites suggested the possibility that F-actin binding induces dimerization of the monomeric V t domain via exposure of these self-association sites. Because oligomers of V985-1066 could not be detected by chemical cross-linking, we focused on the possibility that accessibility of a dimerization site in the V907-984 region is actin-dependent. If actin binding leads to the self-association of V t through a mechanism involving exposure of the cryptic dimerization site in V907-984, such V t dimers should be detectable under the same chemical crosslinking conditions at which dimers of isolated V907-984 are observed.
Although a relatively low concentration of the cross-linker DSS (30 M) is effective for cross-linking the V907-984 dimer (Fig. 2a), little or no detectable dimer species of intact V t is produced under these conditions in the absence of F-actin ( Fig.  2a and Fig. 4, lane 2). Following incubation with actin filaments, however, cross-linking with 30 M DSS reveals a significant increase in the level of V t dimers (Fig. 4, lane 3). This was true for both the V884 -1066 fusion protein and the proteolytically derived 30-kDa fragment (data not shown). We used immunoblotting to detect the actin-induced dimer of V884 -1066, because its expected size (ϳ42 kDa) is essentially identical to the size of monomeric actin. The homodimer of V t cosediments with actin filaments (Fig. 4, lanes 8 and 9), indicating that this dimeric species is functionally active. This finding is consistent with the proposition that actin binding induces V t dimerization.
DSS cross-linking of V t in the presence of F-actin produces another novel species migrating at ϳ64 kDa, a mass corresponding to the expected size of an actin-V t heterodimer (Fig. 4, lane 3). Formation of this product depends on the presence of actin (Fig. 4, lane 2), consistent with its being the result of a cross-link between one V t molecule and one molecule of actin. Immunoblotting confirms that the 64-kDa species represents a 1:1 complex of V884 -1066 and actin, since this species also reacts with antibodies to actin (Fig. 4, lane 3, lower panel). The amount of cross-linkable V t homodimer formed in the presence of actin depends on the concentration of F-actin in a biphasic manner over a range of V t concentrations, suggesting a dependence on the ratio of V t to F-actin (Fig. 5). These data indicate that actin-induced dimerization is not simply the result of spurious cross-linking of "crowded" V t monomers that have become effectively concentrated on the surface of the actin filaments through binding. In such a scenario, at any specific V t concentration in this range, the amount of cross-linked dimer should be maximal at the lowest F-actin concentration and should decrease with increasing actin concentration, as the amount of actin surface area available per bound V t molecule increases.
Intramolecular Flexibility within V t Is Important for Actin Binding-In the absence of actin, the principle cross-linked V t species produced by DSS is an intramolecularly cross-linked monomer that migrates slightly faster on SDS gels than does the uncross-linked monomer. In the presence of actin, a new species corresponding to V t homodimers appears, suggesting that actin binding is associated with, or induces, a conformational change in V t that leads to exposure of the cryptic dimerization sites lying within the actin binding regions. In several experiments, the presence of actin appeared to reduce the amount of intramolecularly cross-linked V t species produced by DSS (e.g. Fig. 5a). This observation suggested that the conformation that V t adopts in the absence of F-actin may be distinct from, and possibly incompatible with, its conformation when bound to actin filaments. Reproducible demonstration of this result was difficult under our assay conditions, however, because at concentrations of DSS that are fairly selective for detecting actin-induced V t dimers, the efficiency of intramolecular V t cross-linking is low, and the appearance of this species is not reliable. Moreover, disappearance of the intramolecularly cross-linked V t product in the presence of actin would not necessarily indicate a conformational change resulting from actin binding. Therefore, we sought another strategy to test the possibility that, when bound to actin, V t adopts a different structure from that detected by intramolecular DSS cross-linking in the absence of actin.
We utilized the cross-linking reagent DSP, which is essentially identical to DSS in its reactivity and solubility properties but has a disulfide bridge in its spacer arm to allow for reversible cross-linking by thiol-mediated cleavage. Intact V t (V884 -1066) was first reacted with DSP under conditions at which the tail construct is almost quantitatively converted to the intramolecularly cross-linked species, and the ability of this species to bind actin was tested in a cosedimentation assay before and after cleavage of the cross-link with DTT (Fig. 6). In the absence of DTT, most untreated V t binds F-actin, while DSPcross-linked V884 -1066 is significantly reduced in its ability to bind actin filaments. Reversal of the intramolecular cross-link  Reactions were analyzed for cross-linking by SDS-PAGE and Coomassie Blue staining. Even at the highest DSS concentration and the longest reaction time, V884 -1066 is not significantly cross-linked into dimers (d), although a faster-migrating monomer species (m i ) is apparent even at short times and lower cross-linker concentration. Crosslinking of isolated V907-984 region into a dimeric species is readily detectable at low cross-linker levels and short times and is essentially quantitative at the highest DSS concentration. b, differential reactivity of V907-984 region; dimers of V907-984 are readily detected by chemical cross-linking with low concentrations of either DSS or MBS, whereas even 10 -100-fold higher concentrations of EDC yield little dimer species. with DTT, however, leads to significant recovery of actin binding activity (Fig. 6). Intramolecular cross-linking of V884 -1066 with the irreversible DSS also led to loss of F-actin binding in cosedimentation assays (data not shown), but it was unclear whether ablation of F-actin binding in the DSS-treated monomer was due to introduction of the intramolecular cross-link per se or to chemical modification of the targeted lysine residues. In contrast, the result using DSP-cross-linked V t demonstrates that inhibition of actin binding is due to intramolecular cross-linking and not to side chain modification; DTT cleaves the spacer arm of DSP but does not reverse modification of the targeted lysine residues. Because an intramolecular cross-link in V t would be expected to hinder the mobility of different subdomains of this protein relative to one another, this result suggests that conformational flexibility within the tail domain is important both for high affinity actin binding and for actininduced dimerization.
Actin and Phosphoinositides Induce Biochemically Distinct Oligomers of V t -In addition to F-actin binding sites, the V907-984 and V985-1066 regions of the tail domain also contain sites FIG. 3. Stable dimerization of V907-984 but not of V884 -1066. Varying concentrations of intact V884 -1066 and of isolated V907-984 region were crosslinked with 100 M DSS for 20 min. Aliquots from each reaction containing 250 ng of protein were subjected to SDS-PAGE and visualized by silver staining. Detectable species include monomer (m) and dimer (d) of both proteins as well as intramolecularly cross-linked monomer (m i ) of V884 -1066. Dimers of V884 -1066 monomer are detectable at high protein concentration but are dilution-sensitive, whereas intramolecular cross-linking of V884 -1066 is not concentration-dependent. Cross-linking of the free V907-984 region into dimers is both quantitative and insensitive to dilution over this range, indicating that this polypeptide forms a stable dimer species.  6 -11). Reaction products were analyzed by SDS-PAGE followed by Western transfer and immunoblotting with antibodies specific for either the vinculin tail domain (upper panel) or actin (lower panel). Homodimers of V884 -1066 (ϳ42 kDa) are detectable in the presence, but not in the absence, of F-actin. A heterodimeric species (ϳ64 kDa) containing one actin monomer and one V884 -1066 monomer is also apparent following cross-linking. mediating association with acidic phospholipids such as PIP 2 (24,30). The binding sites for actin and for phospholipid may overlap, since Steimle et al. (14) have reported that PIP 2 inhibits binding of V t to actin filaments. Like F-actin, phospholipids may thus modulate V t oligomerization. Recently, Rudiger and colleagues (30) observed that the acidic phospholipids PIP 2 and phosphatidylserine induce formation of V t oligomers. We reproduced these observations in order to compare lipid-induced V t oligomers with the actin-induced dimer we have described.
We observed dimers and trimers of V884 -1066 following DSS cross-linking in the presence of pure PIP 2 micelles or of small unilamellar vesicles of pure PI (Fig. 7a). The level of phospholipid-induced V t dimerization depended on the concentration of lipid in a biphasic manner, with the maximal amount of dimer and trimer forming at a lipid:protein molar ratio of approximately 3-10:1 in the case of PIP 2 micelles and approximately 10 -30:1 in the case of PI vesicles (Fig. 7b), suggesting a 3-fold difference in efficacy between these lipids. Essentially all of the PIP 2 molecules should be accessible at the surface of a PIP 2 micelle, whereas only about half of the PI in a small unilamellar vesicle would be exposed. If the data in Fig. 7b are corrected to account for this expected difference in surface exposure, the difference in efficacy between PIP 2 and PI is only about 2-fold. Similar results were obtained when EDC was used as the cross-linking reagent (data not shown). With either cross-linker, the optimal ratio of PIP 2 we observed is compara-ble with that found by Rudiger and co-workers (30) using a combination of EDC and N-hydroxysuccinimide for cross-linking. We did not observe significant amounts of lipid-induced V t tetramer or pentamer species following cross-linking with DSS or EDC, although trimers were apparent in many but not all experiments (data not shown). In contrast, F-actin consistently induces only dimers of V884 -1066.
The addition of 0.2% Triton X-100 substantially reduces the amount of cross-linkable V t species formed in the presence of acidic phospholipids, whereas the amount of actin-induced V t dimer is largely unaffected by the addition of detergent (Fig.  8a). The effect of Triton X-100 is not likely to be due to disruption of PIP 2 binding to V884 -1066, since under the same conditions PIP 2 inhibits V t binding to F-actin (Fig. 8b), as reported recently (14). Very low concentrations of PIP 2 (10 -25 M) in 0.2% Triton X-100 are sufficient to inhibit actin binding by V t completely, whereas even at PIP 2 concentrations 100-fold higher than the concentration of V884 -1066, the presence of Triton X-100 largely eliminated PIP 2 -induced oligomers of V t (data not shown). These data suggest that the inhibitory effect of Triton X-100 is mediated by disruption of phospholipid aggregates and dispersion of the phosphoinositide molecules into comicelles with Triton rather than by inhibition of V t binding to the phospholipid.
The actin-induced V t dimer differs from the phospholipidinduced dimers in its reactivity with different cross-linking reagents. Actin-induced dimers of V884 -1066 can be detected readily by DSS or MBS cross-linking but only poorly by EDC DTT cleaves the spacer arm of DSP to release the internal cross-link but does not reverse modification of the protein primary amines that reacted with DSP. Compared with untreated V884 -1066 (ϪDSP), DSPcross-linked monomer (ϩDSP) binds actin very poorly; actin binding is restored substantially following cleavage of the intramolecular crosslink by reducing agent (ϩDSP, ϩDTT).  (Fig. 8c), a pattern of reactivity comparable with that of dimers of the isolated V907-984 region (Fig. 2b). EDC cross-linking does, however, produce a V t -actin heterodimer of ϳ64 kDa, which is also produced by DSS treatment but formed only poorly by MBS cross-linking (Fig. 8c). The V t homodimer induced by actin is thus specifically unreactive to EDC crosslinking. In contrast, both PIP 2 and PI induce V t dimers that can be cross-linked effectively by DSS, MBS, or EDC (Fig. 8c). These results suggest that at least two biochemically distinct species of V t dimers are possible.

Conformational Changes Underlie V t Dimerization and Factin
Binding-Evidence for self-association of comparable actin-binding V t subregions has been reported previously (18). In this study, we have extended this observation and elucidated further its significance by demonstrating that V t is a monomeric protein with cryptic oligomerization potential that is activated by binding to actin. Gel filtration and chemical crosslinking studies show that both the amino-terminal (V907-984) and the carboxyl-terminal (V985-1066) halves of the vinculin tail domain form stable homodimers, whereas little or no selfassociation of intact V t can be detected in the same assays. The presence of cryptic dimerization sites suggested that self-association of V t is regulated through exposure of these sites in response to specific events. In addition to self-association sites, these regions of V t also contain determinants mediating interaction with F-actin (17,18), implicating actin in regulated exposure of the self-association sites. Chemical cross-linking assays demonstrate directly that F-actin binding is one type of event that can induce dimerization of V t . The ability of this actin-induced dimer to react with different cross-linking reagents is similar to the range of reactivity of the homodimer formed by isolated V907-984, implicating this region of the tail in mediating dimerization of the intact domain in the presence of actin. Dimerization of the carboxyl-terminal region may also play a role in formation of the actin-induced dimer, but its contribution cannot be inferred from the present data.
Subdomain motion within V t is critical both for F-actin binding and for exposure of cryptic dimerization sites. Chemical cross-linking of the V t monomer with DSS or DSP in the absence of F-actin leads to formation of an intramolecular crosslink that results in faster migration through SDS-PAGE gels, suggesting that this species retains a more compact structure under denaturing conditions. This interpretation is supported by the observation that DSS-cross-linked monomer is significantly more resistant to papain digestion than untreated monomer (data not shown). F-actin binding reduces the efficiency of intramolecular cross-linking of V t , while the prior introduction of an internal cross-link greatly reduces V t affinity for F-actin. The use of the reversible cross-linker DSP confirms that crosslinking per se, not chemical modification of protein side chains, is responsible for inhibition of actin binding. These data suggest that DSS and DSP target residues on two secondary structural elements that must move away from one another for V t to adopt a conformation competent to bind tightly to F-actin and to dimerize. Identification of the sites of intramolecular crosslinking in V t should provide insight into the structural basis of this conformational change.
V t Oligomerization May Be Induced by More than One Mechanism-Another type of event that can induce V t oligomerization is binding to aggregates of acidic phospholipids such as PIP 2 (30). This finding is of interest, because PIP 2 is thought to modulate recruitment of vinculin to adherens junctions by associating with phospholipid binding sites on the tail domain (24,30,31) and thereby disrupting the intramolecular V h -V t interaction to allow binding to F-actin and talin (32,33). We compared PIP 2 with PI, which binds well to V t (31) but has not been implicated in regulation of vinculin structure under physiological conditions (14,32). PI induces at least dimerization and trimerization of V t in a manner qualitatively and quantitatively similar to PIP 2 . Thus, it is unclear how lipid-induced oligomerization of V t relates to the specific activation of intact vinculin by polyphosphoinositides observed in vitro (14,32).
Two pieces of evidence indicate that the oligomers of V t induced by acidic phospholipids are biochemically distinct from V t dimers induced by F-actin. In the presence of actin, only dimers of V t were evident; in contrast, the major oligomeric species that we find in the presence of either PI or PIP 2 are dimers and trimers, while Hü ttelmaier et al. (30) also detected tetramers and pentamers induced by phosphatidylserine or PIP 2 . The V t oligomers induced by actin and by phospholipids also differ in reactivity with different cross-linking reagents. Although the data on this point are preliminary in nature, it is tempting to speculate that these biochemical differences reflect an underlying structural difference resulting from distinct conformational changes in the V t monomer induced by these ligands. This possibility is particularly appealing in light of recent evidence showing that PIP 2 inhibits the interaction of V t with F-actin (14).
Dispersal of acidic phospholipids into micelles of the nonionic detergent Triton X-100 significantly reduces their ability to induce formation of V t oligomers, despite the fact that when presented under such conditions (0.3-8 PIP 2 molecules/micelle) PIP 2 is capable of binding V t and inhibiting its interaction with F-actin (14). It is noteworthy that Hü ttelmaier et al. (30) have reported that the addition of phosphatidylcholine, a neutral lipid that does not bind V t , to even 10% (w/w), completely inhibits formation of V t oligomers in the presence of PIP 2 . These results suggest that lipid-induced V t oligomerization is the result of binding to an aggregate of pure acidic phospholipid rather than of binding to small clusters or individual molecules of acidic phospholipid. Highly concentrated foci of PIP 2 may exist in vivo and play a role in regulating vinculin conformation (34). However, it should be noted that the sensitivity of phospholipid-induced oligomerization to surface dilution, as well as the broad range of oligomers sometimes observed (30), are also consistent with the possibility that such oligomers represent cross-linking events resulting from forced clustering of V t monomers binding to the same limited surface area of a phospholipid aggregate.
A Model for Ligand-induced V t Dimerization-Examination of the sequence of V t suggests that dimerization is mediated by the V907-984 and V985-1066 regions through homophilic coiled-coil interactions that form subsequent to an activating conformational change. The MacStripe program, which is based on the Coils algorithm of Lupas (26,35), detected significant coiled-coil propensity in residues 944 -971 and residues 1014 -1045 when the V t sequence was analyzed using the MTIDK matrix (35) and a window size of 14 -28 residues (Fig.  9a). Analysis using the MTK matrix (26) also yielded a high coiled-coil score (ϳ0.8) for residues 1014 -1045, but a much lower score (ϳ0.1) for residues 944 -971. These data suggest that the V985-1066 region and possibly the V907-984 region contain coiled-coil-forming motifs.
The high ␣-helical content of V907-984 and V985-1066 is consistent with the possibility of coiled-coil forming segments in these regions. Secondary structure prediction using the PHD program (28) suggests that V t is composed of five or six ␣-helices, connected by short loop regions, with essentially no ␤-strand content (Fig. 9a). V907-984 and V985-1066 are both predicted to be composed largely of two long ␣-helices. Based on the amino acid sequence, these helices would be highly amphipathic in nature (24,36), consistent with the predicted coiled-coil forming propensity. Circular dichroism analysis demonstrates directly a highly ␣-helical secondary structure content for both V884 -1066 and V916 -970, a phospholipidbinding fragment contained within V907-984 (24) that retains the dimerization site.
Together with the results reported above, these observations suggest a model for ligand-induced conformational change and dimerization of V t (Fig. 9b). Based on our hydrodynamic data, V t must adopt an overall compact structure, presumably con-sisting of at least two subdomains in which the hydrophobic faces of the amphipathic helices are sequestered in the interior. A conformational change associated with binding of F-actin or acidic phospholipid would result in movement of at least two subdomains relative to one another, exposing the hydrophobic faces of some of the helices. Specific, exposed hydrophobic regions could form coiled-coil interactions with corresponding regions on another activated V t monomer, leading to formation of V t dimers and potentially of higher order oligomers. The "opening" of the tail domain could also alter the nature of the interaction with the specific activating ligand. In the case of phospholipid binding, exposure of the hydrophobic surface of the V916 -970 region probably leads to insertion of this face into the acyl region of the lipid bilayer, as we have previously proposed (24), while in the case of actin binding, such newly exposed surfaces may mediate high affinity interactions through van der Waals contacts with the filament surface.
Up to this point, we have referred to the ability of F-actin and of acidic phospholipids to "induce" conformational changes in V t , but it should be clarified that a model in which these ligands act by perturbing an equilibrium between different conformational states of V t is also plausible. In this view, a low level of activated V t would be present normally as a result of confor- FIG. 9. Actin-induced dimerization of V t may be mediated by exposure of cryptic coiled-coil motifs. a, the sequence of V884 -1066 was analyzed using the PHD program (28) to identify predicted secondary structure content and the MacStripe program to find predicted coiled-coil motifs. MacStripe is based on the Coils algorithm of Lupas (26). For each residue, the probability of its being in a helix or a loop based on the PHD analysis and the probability of its participating in a coiled-coil interaction based on the MacStripe analysis are plotted. The V907-984 and V985-1066 regions we have used both contain predicted helical elements with high coiled-coil forming potential. b, a model of actin-induced V t dimerization. Actin binding induces or stabilizes a conformational change in which compact V t monomers (i) unfold to expose cryptic self-association sites (ii). Subsequent V t dimerization (iii) may underlie F-actin bundling and gelation by V t (4). mational breathing of the monomer. F-actin or acidic phospholipids would bind preferentially to V t molecules, which are in an activated, less compact conformation, thus stabilizing them and blocking transition back to the compact state. Such a model might better explain the low levels of V884 -1066 dimers occasionally observed following chemical cross-linking in the absence of actin or phospholipid as well as the inhibitory effect of intramolecular cross-linking of V t on its binding to F-actin. The latter result implies that F-actin binds poorly to the compact conformation of V t , and it is difficult to reconcile with the notion that opening of V t occurs subsequent to actin binding.
If our model of actin-and phospholipid-induced opening of V t is correct, an interesting implication is that an allosteric mechanism may underlie intramolecular regulation of V t by V h in native vinculin. Binding of V t to V h diminishes the affinity of V t for these ligands under physiological conditions (4, 24, 31) but does not induce V t oligomerization, since native vinculin exists as a monomer in solution (6,37). It is possible that V h only binds to V t monomers that are in a compact conformation incompatible with V t oligomerization and high affinity actin binding, thus providing a mechanistic basis for inhibition of actin binding of V t by V h (4, 6). As a corollary, V t oligomerization or the associated change in V t conformation may play a role in stabilizing an activated form of vinculin. The balloonon-a-string structure observed in electron micrographs of intact vinculin, in which V t is highly extended (19,38), may represent an activated form of vinculin where the V h -V t interaction is disrupted as a result of unfolding of V t .
Relation of Actin-induced Dimerization to V t Function-Although earlier studies mapping two F-actin binding regions in V t (17) suggested a mechanism of actin cross-linking involving simultaneous engagement of two actin filaments by the same V t monomer, the present results suggest the possibility of a different mechanism. In this view, V t monomer is not itself competent to cross-link F-actin, but exposure of cryptic selfassociation sites associated with actin binding allows for dimerization of actin-bound monomer to form a cross-linking species. Thus, actin filaments might regulate their own assembly into bundles and gels by inducing certain actin-binding proteins to form cross-linking-competent species. If this hypothesis is correct, it is intriguing that each monomer of V t has two actin binding regions, since logically only one actin binding site would be necessary to form a cross-linking species out of a protein that dimerizes. It is possible that the presence of multiple actin binding sites is mechanistically important; simultaneous engagement of both sites by F-actin may provoke or stabilize the conformational change that exposes masked dimerization sites. Further elucidation of the connection between actin cross-linking activity and actin-induced dimerization will be important.
The vinculin tail domain is the only F-actin-binding protein for which direct evidence of actin-induced oligomerization has been reported, although self-association induced by actin binding has been proposed to explain the cross-linking activity of scruin (39). Scruin is a monomeric protein with two actin binding domains. Three-dimensional reconstruction of scruin-decorated F-actin reveals that the two actin binding domains of a single molecule of scruin engage the same actin filament, while filament cross-linking is effected by self-association of scruin monomers bound to two different filaments (40). This suggests that binding of scruin to F-actin leads to its self-association. The oligomeric state of other monomeric actin cross-linking proteins, such as villin, fimbrin, and the Dictyostelium 30-kDa protein (41,42), may likewise be modulated by F-actin in ways that affect their cross-linking activities.
The potential for actin to regulate the activity of its own cross-linking proteins is intriguing. With the large number of actin-binding proteins that also cross-link F-actin (43), it is unclear under what circumstances it would be advantageous to the cell for actin to potentiate its own cross-linking. One role might be in the control of actin bundle formation, which can be quite complex in vivo, involving several stages during which small, disorganized bundles coalesce and reorganize into larger and more ordered assemblies through the sequential action of multiple actin cross-linking proteins (44 -46). Actin-binding proteins that undergo actin-induced self-association might be well suited for participating in critical stages of such processes. The ability of F-actin to modulate the conformation and oligomeric state of its binding proteins suggests an active role for actin filaments in cell signaling. Rather than simply providing a framework for recruitment of signaling proteins, actin may also regulate the activities of some of the many signal transducers that localize to adherens junctions or to other specialized compartments of the actin cytoskeleton (1,47). Such a role would help to explain the importance of cytoskeletal integrity to adhesion-dependent signaling events. Actin-binding proteins traditionally thought to play largely structural roles, such as vinculin, may acquire novel functionality when associated with actin filaments as a result of actin-induced changes in conformation or oligomeric state. An important goal of future studies will be to elucidate what role actin-induced dimers of vinculin may play in the assembly and function of cell adhesion sites.