An Autoinhibited Structure of α-Catenin and Its Implications for Vinculin Recruitment to Adherens Junctions*

Background: α-Catenin is an actin-binding protein that recruits vinculin to adherens junctions. Results: An elongated autoinhibited structure of α-catenin indicates structural and functional coupling of its vinculin- and actin-binding mechanisms. Conclusion: The anchoring strength of adherens junctions is dynamically regulated by α-catenin to match the actomyosin-generated tension. Significance: Multistate conformations of α-catenin facilitate the direct and vinculin-assisted linkages between the cadherin-catenin complex and the actin cytoskeleton. α-Catenin is an actin- and vinculin-binding protein that regulates cell-cell adhesion by interacting with cadherin adhesion receptors through β-catenin, but the mechanisms by which it anchors the cadherin-catenin complex to the actin cytoskeleton at adherens junctions remain unclear. Here we determined crystal structures of αE-catenin in the autoinhibited state and the actin-binding domain of αN-catenin. Together with the small-angle x-ray scattering analysis of full-length αN-catenin, we deduced an elongated multidomain assembly of monomeric α-catenin that structurally and functionally couples the vinculin- and actin-binding mechanisms. Cellular and biochemical studies of αE- and αN-catenins show that αE-catenin recruits vinculin to adherens junctions more effectively than αN-catenin, partly because of its higher affinity for actin filaments. We propose a molecular switch mechanism involving multistate conformational changes of α-catenin. This would be driven by actomyosin-generated tension to dynamically regulate the vinculin-assisted linkage between adherens junctions and the actin cytoskeleton.


␣-Catenin is an actin-and vinculin-binding protein that regulates cell-cell adhesion by interacting with cadherin adhesion receptors through ␤-catenin, but the mechanisms by which it anchors the cadherin-catenin complex to the actin cytoskeleton at adherens junctions remain unclear.
Here we determined crystal structures of ␣E-catenin in the autoinhibited state and the actin-binding domain of ␣N-catenin. Together with the small-angle x-ray scattering analysis of full-length ␣N-catenin, we deduced an elongated multidomain assembly of monomeric ␣-catenin that structurally and functionally couples the vinculin-and actin-binding mechanisms. Cellular and biochemical studies of ␣Eand ␣N-catenins show that ␣E-catenin recruits vinculin to adherens junctions more effectively than ␣N-catenin, partly because of its higher affinity for actin filaments. We propose a molecular switch mechanism involving multistate conformational changes of ␣-catenin. This would be driven by actomyosin-generated tension to dynamically regulate the vinculin-assisted linkage between adherens junctions and the actin cytoskeleton.
The cadherin-catenin complex mediates the formation of intercellular adherens junctions (AJs) 3 by having cadherins on adjoining cells participating in the Ca 2ϩ -dependent homotypic interactions, whereas connecting intracellularly to the actin cytoskeleton and various signaling pathways through the catenins (1). Coordination of the extracellular and cytoplasmic events by the cadherin-catenin complex is crucial for AJs to be stable enough to support tissue structure and integrity, as well as to remain sufficiently dynamic to rapidly disassemble obsolete connections and foster new junctions during morphogenesis (2)(3)(4). The cadherin-catenin complex consists of the cadherin adhesion receptor and the catenins, which include ␣-catenin, ␤-catenin, and p120 catenin (supplemental Fig.  S1A). Although ␤-catenin and p120 catenin directly bind to the cytoplasmic tail of cadherin and regulate the stability of cadherin on the cell surface (5), ␣-catenin, an actin-binding protein, indirectly associates with cadherin through ␤-catenin and anchors AJs to the actin cytoskeleton. Three types of ␣-catenin have been identified in mammals, designated as ␣E-catenin, ␣N-catenin, and ␣T-catenin (6). Their critical role in cell adhesion is underscored by pathological consequences of deletion or truncation of ␣-catenin genes in mammals, such as cancer metastasis (7), cerebellar hypoplasia (8), and arrhythmogenic cardiomyopathy (9). As well, the embryonic lethality has been reported for ␣-catenin-null flies (10). Cell biology studies revealed that ␣E-catenin is essential for organization of the apical junctional complex (AJC) (11), containing tight junction and AJ, by forming two types of AJs, punctum adherens and zonula adherens, in polarized epithelial cell monolayer (supplemental Fig. S1B) (12), and ␣N-catenin is required for synapse stability in hippocampal neurons (13).
A traditional model of cell-cell adhesion assumed that ␣-catenin would directly connect the cadherin-catenin com-plex and actin filaments (F-actin), but this model was called into question when previous studies revealed that the interaction between cadherin-catenin complexes and F-actin is much weaker than expected, and that these showed different levels of mobility at AJs (14,15). Currently, at least four potential actinbinding modes of ␣-catenin have been proposed (supplemental Fig. S1C): autoinhibited state (15), allosteric-binding state involving ␣-catenin homodimers binding to F-actin near AJs (14), direct-binding state (16), and indirect-binding state through other actin-binding proteins, such as vinculin (17,18) and EPLIN (epithelial protein lost in neoplasm; also known as Lima-1) (12,19). Nonetheless, recent studies have shown that monomeric ␣-catenin bound to the cadherin⅐␤-catenin complex mediates the physical linkage between AJs and the actin cytoskeleton (20), and that cadherin is constitutively under actomyosin-generated tension (21,22). Moreover, ␣-catenin has been proposed to act as a mechanosensor in AJ formation by facilitating the recruitment of vinculin to AJs in the actomyosin-derived tension-dependent manner (17). However, how ␣-catenin regulates the linkage between the cadherin-catenin complex and the actin cytoskeleton by translating the mechanical forces into biochemical signals remains unclear.
␣-Catenin has a modulatory architecture that consists of three major domains with distinct functionalities (Fig. 1A): an N-terminal (N) domain participates in ␤-catenin binding and homodimerization (23,24); a modulatory (M) domain interacts with several actin-binding proteins, including a closely related actin-binding protein vinculin (17,25,26); and a C-terminal (C) domain directly (16) and/or indirectly (19,27) binds to F-actin. As ␣Eand ␣N-catenins share 82% primary sequence identity (supplemental Fig. S2), these proteins should be structurally related to each other. Also, ␣-catenin and vinculin share ϳ30% sequence identity at three distinct vinculin homology regions, VH1, VH2, and VH3 (Fig. 1A), suggesting some commonalities between these proteins as well. It is, however, unclear how three domains of ␣-catenin are structurally and functionally coupled together in a full-length molecule, as currently available crystallographic details are limited to the individual N-and M-domain fragments of ␣E-catenin (24, 25, 28 -30).
To better understand how ␣-catenin facilitates actomyosindependent tension-induced recruitment of vinculin to AJs, we employed x-ray crystallography, small-angle x-ray scattering (SAXS), and other complementary methodologies to reveal how the assembly of different functional domains within a fulllength mouse ␣-catenin molecule facilitates a unique activation mechanism of ␣-catenin involving multistate conformational changes.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Mouse ␣E-catenin was cloned into the pET28a vector (Novagen) and expressed as an N-terminal His fusion protein. Mouse ␣N-catenin, ␣N-catenin C domain, and human vinculin D1 domain were cloned into the pGEX4T1 vector (GE Healthcare) and expressed as N-terminal glutathione S-transferase (GST) fusion proteins. Site-directed mutagenesis was performed using the QuikChange protocol (Stratagene) to produce all single residue and deletion mutants. Recombinant proteins were expressed in Escherichia coli BL21-CodonPlus (Stratagene) strain for overnight at 288 K. After collection by centrifugation, cells were resuspended in lysis buffer (50 mM Tris-HCl, pH 8.0, 300 mM NaCl, 10 mM ␤-mercaptoethanol, 1 mM TCEP), sonicated on ice, and subjected to centrifugation to separate proteins in the soluble fraction. His fusion proteins were isolated using nickel-nitrilotriacetic acid resin (Qiagen). GST fusion proteins were isolated using glutathione-Sepharose resin (GE Healthcare). His and GST fusion proteins were cleaved by thrombin to remove the affinity tags and the cleaved proteins were further purified by size-exclusion chromatography using either Superdex 75 or Superdex 200 (GE Healthcare). The purified proteins were exchanged into protein storage buffer (50 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM TCEP).
Crystallization and Data Collection-Crystals of full-length ␣E-catenin were obtained by incubating a protein solution containing purified full-length ␣E-catenin from the monomer fraction (10 mg/ml) against a crystallization solution (0.2 M sodium citrate, 20% (w/v) PEG-3350) by vapor diffusion at 293 K. Crystals were often attached to the crystallization plate or to a skin of denatured protein on the surface on the drop. Dimeric fulllength ␣E-catenin and monomeric full-length ␣N-catenin samples were also subjected to crystallization, however, these efforts were fruitless. For the crystallization of the ␣-catenin C domain, a proteolytically stable fragment of the C domain (residues 652-906) was identified when the full-length ␣E-catenin sample was subjected to degradation under nonreducing conditions during purification. The equivalent region of the ␣N-catenin C domain (␣NcatC; residues 651-905) was cloned, expressed, purified, and crystallized by incubating a protein solution (20 mg/ml) against a crystallization solution (0.1 M HEPES, pH 7.0, 1.6 M (NH 4 ) 2 SO 4 , 0.2 M NaCl, and 0.1 M K/Na tartate) by vapor diffusion at 277 K. Selenomethione-substituted ␣NcatC was expressed, purified, and crystallized in the same manner. Crystals were briefly soaked in crystallization solution containing 20% glycerol for data collection at 100 K. Diffraction data were collected at the Advanced Photon Source beamlines 19BM and 19ID (Argonne National Laboratory, Argonne, IL) and processed with HKL2000 (31). Statistics pertaining to the diffraction data are presented in Table 1.
Crystal Structure Determination and Refinement-The crystal structure of ␣EcatNM was determined at 6.5-Å resolution. The structure solution was solved by molecular replacement with PHASER (32) using the ␣E-catenin N domain residues 82-262 (PDB code 1DOV) as a search model for the N domain and a homology model of ␣E-catenin M II-III regions generated from the full-length vinculin structure (PDB code 1ST6) as a search model for the M domain. For refinement at 6.5-Å resolution, the deformable elastic network (DEN) refinement was performed by using CNS 1.3 (33) as previously described (34), except for the sole implementation of overall anisotropic B-factor refinement and the constitutive use of the DEN restraints throughout the process (35). The optimum values of the ␥ and W DEN parameters of DEN refinement were determined by a two-dimensional grid search using the SBGrid DEN refinement portal. The M I region in the "unfurled" conformation was modified according to A -weighted difference electron density maps (2F o Ϫ F c and F o Ϫ F c ) and a simulated annealing omit F o Ϫ F c map of the M I region by manual model building using Coot (36). The crystal structure of ␣NcatC was initially determined at 3.0-Å resolution by the single-wavelength anomalous dispersion method, and further refined at 2.6 Å by using PHENIX (37). Refinement statistics are presented in Table 1. Electrostatic surface potential values of the ␣-catenin and vinculin structures were calculated using APBS (38). Molecular graphics representations were prepared using PyMOL.
Mammalian Expression Plasmid Construction and Transfection-Full-length or deletion clones for ␣Eand ␣N-catenin genes were amplified by PCR, and subcloned into pCA-sal-EGFP-IRES-hygro vector (39). To generate chimeric constructs, 1-662 and 663-906 amino acids fragments of ␣E-catenin were fused with 662-905 and 1-661 amino acids fragments of ␣N-catenin, respectively, by using overlapping PCR method. R2/7 cells were cultured in a 1:1 mixture of DME and Ham's F-12 medium (Iwaki) supplemented with 10% fetal calf serum (FCS). Cells were transfected with expression vectors by using Effectene Transfection Reagent (Qiagen), according to the manufacturer's instructions. Isolation of stable transfectants was performed as described earlier (12).
Immunostaining, Image Acquisition, and Processing-Immunofluorescence staining was carried out as described previously (19). Confocal images were acquired with a Zeiss LSM710 confocal microscope (Carl Zeiss) through a Plan-Apochromat ϫ63/1.40 NA oil-immersion objective lens (Carl Zeiss). Images were processed by Photoshop (Adobe). Projection of Z-stack images (0.3-m sections) was acquired using Zen software (Carl Zeiss).
Small-angle X-ray Scattering Data Collection-Monomeric full-length ␣N-catenin samples at 10 -78 M in a storage buffer (25 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM TCEP) were used for SAXS data collection at McGill University and APS beamline 12. Both the monomeric and dimeric samples of ␣E-catenin were deemed unsuitable for SAXS analysis due to its heterogeneous nature. Home source data were measured on an Anton Paar SAXSess mc2 equipped with a PANalytical PW3830 x-ray generator and a Roper/Princeton CCD detector, with the sample temperature at 277 K. Synchrotron data were collected at the APS beamline 12-ID-C using the x-ray beam at 18 Kev and a MAR CCD detector, with the sample temperature at 277 K.
Small-angle X-ray Scattering Data Processing and Analysis-Data processing and analysis were performed using the ATSAS package (40); PRIMUS (41) was used for calculating the radius of gyration (R g ) and forward scattering (I 0 ) with the Guinier approximation; GNOM (42) was used to calculate the distance distribution function. The I 0 of each concentration was compared with lysozyme and bovine serum albumin datasets to verify the molecular weight and monomeric state, and to eliminate any potential for concentration-dependent effects. PRI-MUS was used to produce a merged home source dataset for further analysis by combining a low-angle part of 15 M data in the range of S from 0.002 to 0.089 nm Ϫ1 and a higher-angle part of 78 M data in the range of S from 0.055 to 0.299 nm Ϫ1 . Similarly, a merged synchrotron dataset was produced for further analysis by combining a low-angle part of the 20 M data in the range of S from 0.017 to 0.088 Å Ϫ1 and a higher-angle part of 30 M data in the range of S from 0.031 to 0.285 Å Ϫ1 . Ab initio shape reconstruction was carried out by generating 20 bead models with DAMMIN (43), and averaging with DAMAVER (44). Theoretical scattering curves were computed with CRYSOL (45). Rigid-body docking of the ␣EcatNM and ␣NcatC crystal structures into the SAXS-derived molecular envelope of ␣N-catenin was implemented using the six-dimensional search procedure in the Situs program package (46). Three-dimensional model of full-length ␣N-catenin was produced by using Modeller (47) with the docked crystal structures of ␣EcatNM and ␣NcatC, as well as full-length vinculin (PDB code 1ST6) as references, and used to calculate the theoretical scattering curve.
Actin-pelleting Assays-Monomeric rabbit skeletal muscle actin (Cytoskeleton) was polymerized in the polymerization buffer (5 mM Tris Supernatant and pellet fractions were analyzed by SDS-PAGE with Coomassie Blue stain. Gel band intensity was measured by using ImageJ (48).
Analytical Ultracentrifugation-Sedimentation equilibrium experiments on ␣Eand ␣N-catenins were carried out at the Ultracentrifugation Service Facility in the Department of Biochemistry at the University of Toronto. 6 -21 M ␣-Catenin protein in a buffer containing 50 mM Tris-HCl, pH 8.0, 300 mM NaCl, 10 mM ␤-mercaptoethanol, and 1 mM TCEP was spun at 7000, 8000, and 9000 rpm, at 4°C, in an An-60 Ti rotor in a Beckman Optima XL-A analytical ultracentrifuge. Absorbance was recorded at 230 and 280 nm. Data analysis was performed using the Origin MicroCal XL-A/CL-I Data Analysis Software Package Version 4.0.
Fluorescence Resonance Energy Transfer Spectroscopy-Multiple rounds of PCR were performed to produce a chimeric ␣N-catenin construct with Cerulean (49) inserted between Pro 633 and Glu 634 . To generate ␣N-catenin FRET constructs, Venus (50) was inserted into the ␣N-catenin-Cerulean construct by introducing a HindIII site at various locations in the C domain of ␣N-catenin (between Arg 650 -Ser 651 , Gly 803 -Gly 804 , or Phe 905 -STOP) by mutagenesis. Emission spectra of purified ␣N-catenin-FRET proteins were acquired with a Shimadzu RF-5301-PC spectrofluorophotometer. Cerulean and Venus emission spectra were collected at 1-nm intervals from 445 to 600 nm with excitation at 433 nm using slit widths of 3 nm for excitation and 5 nm for emission.
Electron Microscopy-␣N-catenin protein samples were diluted to 0.5 M with a buffer containing 20 mM Tris-HCl, pH 8.0, 100 mM NaCl, and 1 mM TCEP, and applied to continuous carbon film-coated copper/rhodium EM grids that were previously subjected to glow-discharge in air. After 2 min of incubation, the samples were washed with distilled water, stained with 0.75% uranyl formate, blotted, and allowed to air-dry. Images acquired with an FEI F20 electron microscope equipped with a field emission gun and operating at an accelerating voltage of 200 kV and a magnification of ϫ19,000 were recorded on TVIPS F114 CCD camera.
Blue Native Polyacrylamide Gel Electrophoresis-The oligomeric states of ␣Eand ␣N-catenin protein samples were examined by performing blue native polyacrylamide gel electrophoresis as previously described (51).

RESULTS
We have crystallized full-length ␣E-catenin and determined a 6.5-Å resolution structure by molecular replacement. The structure of ␣E-catenin (Fig. 1B), designated ␣EcatNM, consists of the N domain residues 85-262 in the homodimer state (Fig. 1C) and M domain residues 290 -631 in an autoinhibited conformation, with no density observed for the C domain and the preceding linker region. The N domain structure consists of two helical bundles sharing a long central ␣-helix (Fig. 1C), and resembles previously determined N-terminal domain structures of ␣E-catenin and vinculin (24, 56) (supplemental Fig.  3A). In the crystal, this domain facilitates homodimerization through two N-terminal helices from each protomer forming an intermolecular four-helix bundle in a similar manner as the previously described dimer domain structure (residues 82-279) (supplemental Fig. 3, B and C) (24). In addition, our ␣EcatNM structure reveals that the C-terminal four-helix bundle of the N domain is adjacently located to another bundle formed by the M II region (residues 396 -506) of the M domain (Fig. 1B). The M domain contains two additional four-helix bundles formed by M I (residues 290 -395) and M III (residues 507-631) regions, and three bundles are organized into a ""shaped configuration (Fig. 1B). The M I region, which contains the vinculin-binding site (VBS; residues 305-352), forms a relatively short four-helix bundle consisting of two short and two long ␣-helices (Fig. 1B, supplemental Fig. S4, A-D), and this bundled conformation of M I region is a stark contrast to the recently described unfurled or activated conformation of M I region bound to the D1 domain of vinculin (29,30) (Fig. 1D,  supplemental Fig. S4E). Based on these crystallographic observations, we propose that the VBS adopts the "autoinhibited" conformation by occluding the vinculin-binding hydrophobic surfaces within the hydrophobic core of the M I helical bundle (Fig. 1E). Unraveling of the helical bundle would allow the VBS to interact with vinculin through the "helix bundle conversion" mechanism (29,30).
Previously, the M III region was implicated in inhibiting the interaction between the VBS of ␣-catenin and vinculin (17). Our ␣EcatNM structure shows that the bundled conformation of the M I region is supported by having its C terminus partially wedged between M II and M III bundles at the center of the M domain (Fig. 1B). Several charged amino acid residues from all three regions gather at the M I -M III and M II -M III interfaces to form salt bridges (Fig. 1F). To examine how the conformational change of M I region could be triggered by disrupting these inter-bundle interfaces, we made single-residue mutations within the M III region of ␣E-catenin based on our structure and performed GST pulldown assays to test the ability of these variants to associate with vinculin. Comparing to the wild type ␣E-catenin, ␣E-catenin mutants R551A, and to lesser effect, E521A that are expected to disrupt the M II -M III and M I -M III interfaces (Fig. 1F), respectively, showed noticeable increases in binding to the D1 domain of vinculin (Fig. 1G). Similar observations were made with the equivalent mutations in ␣N-catenin (supplemental Fig. S5). Interestingly, the arrangement of the M II and M III bundles within ␣EcatNM is markedly different from previously determined M II-III fragment structures in the absence of the N domain and the M I region (Fig. 2, A-D) (25,28), but surprisingly similar to the arrangement of D3-D4 domains in vinculin (Fig. 2E) (56,57). This observation is particularly interesting as the compact arrangement of D3-D4 domains is stabilized through additional interdomain interactions involving the D2 domain, and the D3a bundle, which is equivalent to the M I region of ␣-catenin, is less likely to "unfurl" due to the formation of a longer four-helix bundle (supplemental Fig. S4, C and D) (56). These structural features are missing in ␣-catenin (Fig. 1, A and B), which render ␣-catenin more conformationally "switchable" at the VBS. Together these observations suggest that the bundling of the M I region effectively precludes vinculin from accessing the VBS and that the activation of VBS through unfurling of the M I bundle is coupled to the destabilization of the compact arrangement of three helical bundles within the M domain.
We next asked how the autoinhibition could be released in cells and how multiple domains of ␣Eand ␣N-catenins contribute to the vinculin binding mechanism (17). We monitored the AJC formation and localization of vinculin by expressing ␣-catenin in the colon carcinoma-derived R2/7 cells, which are unable to organize the AJC due to the absence of ␣-catenin (26). We confirmed that typical AJC linearly delineating apical cellcell boundaries are restored by expressing either full-length ␣E-catenin, which contains the C domain necessary to activate the autoinhibited VBS (17), or the NM I fragment, which contains the VBS without the inhibitory M III region, as monitored by co-immunostaining for ZO-1 (a tight junction protein), EGFP-tagged ␣E-catenin constructs and vinculin (Fig. 3A). In contrast, expression of the NM I-III fragment of ␣E-catenin in R2/7 cells resulted in no typical AJC formation (Fig. 3A). In these cells, NM I-III fragments containing the intact ␤-catenin binding site were only irregularly localized to cell-cell boundaries, where vinculin accumulation was less clear. Similar results are obtained when C-terminal truncation mutants of ␣N-catenin are expressed in R2/7 cells (Fig. 3B). However, R2/7 cells expressing full-length ␣N-catenin failed to establish the AJC, although faint vinculin signals were occasionally detecta-ble together with this molecule, indicating that this "neural" ␣-catenin cannot normally function in this carcinoma line. Nevertheless, AJC formation, including vinculin recruitment to cell-cell boundaries, was induced in R2/7 cells expressing an ␣N/␣E-chimera consisting of the ␣N-catenin NM domains fused with the ␣E-catenin C domain (Fig. 3B), but not with an ␣E/␣N-chimera consisting of the ␣E-catenin NM domains fused with the ␣N-catenin C domain (Fig. 3A). Our in vitro actin-pelleting assay revealed that a significantly larger proportion of ␣E-catenin C domain (20 Ϯ 2.7%) cosediments with F-actin compared with the ␣N-catenin C domain (12 Ϯ 2.0%) (Fig. 4, A and B). These results support that the direct interaction between the ␣-catenin C domain and F-actin is critical for localization of vinculin at the AJC (17), and further suggest that the threshold of activation of ␣-catenin VBS is different in ␣Eand ␣N-catenins, with the C domain of ␣E-catenin having higher affinity for F-actin than the C domain of ␣N-catenin.
To examine the structural details responsible for direct F-actin interaction by ␣-catenin, we solved a crystal structure of the ␣N-catenin C domain (␣NcatC), which consists of proteolytically stable residues 651-905 (supplemental Fig. S6), at 2.6-Å resolution (Fig. 4C). The ␣NcatC structure contains a five-helix bundle that closely resembles the actin-binding D5 domain structure of vinculin (58) (supplemental Fig. S7A). It has an additional ␣-helix preceding the bundle that caps the C-terminal end of the third ␣-helix, as well as the N-terminal (residues 651-661) and C-terminal extensions (residues 862-905) that are mostly disordered (Fig. 4C). Structural alignment of the actin-binding domains of ␣N-catenin and vinculin reveals that these structures are quite similar (root mean square deviation of 1.4 Å over 134 C␣ atoms), but the residues proposed to be involved in the "upper" and "lower" F-actin-binding sites of vinculin are not well conserved in ␣N-catenin (59) (Fig. 4, D-F, supplemental Fig. S7, B and C). We then examined the electrostatic surface potential of ␣NcatC, which revealed prominent basic and acidic clusters similar to those located within the actin-binding domain of vinculin (supplemental Fig. S7, D and E). To gain insights into the differences in the actin-binding mechanisms of ␣Nand ␣E-catenins, we built a homology model of the ␣E-catenin C domain (␣EcatC; residues 663-862) (supplemental Fig. S7F). As expected from the highly conserved primary sequences of ␣E and ␣N-catenins (supplemental Fig.  S2), the structure and electrostatic surface potential of ␣EcatC resemble those of ␣NcatC (supplemental Fig. S7, F and G). However, careful inspection of the ␣E-catenin model identified an increased number of hydrophobic residues within the loop region (residues 849 -856) following the ␣-helix 6 (Fig. 4, D-F). The location of this unique hydrophobic patch in the ␣EcatC model coincides with the lower actin-binding site of vinculin (59), suggesting a potential role in the enhanced association of ␣EcatC to F-actin.   Fig. S7B) are indicated. D and E, close-up views of the "upper site" (D) and "lower site" (E) of ␣N-catenin. Side chains of non-identical residues between ␣E-catenin (blue) and ␣N-catenin (green for conserved residues and red for non-conserved residues) are shown as sticks. F, structural alignment of actin-binding domain sequences of ␣N-catenin, ␣E-catenin, and vinculin (PDB code 1ST6). ␣Eand ␣N-catenins share 87% identical and 93% similar residues within their C domain sequences. ␣E-catenin and vinculin residues that are identical to ␣N-catenin residues are indicated by dots. Non-identical residues of ␣E-catenin are shown in blue, whereas ␣N-catenin residues in green and red indicate conserved and non-conserved substitutions found between ␣Eand ␣N-catenins. Vinculin residues that constitute the upper and lower sites (59) are shown as orange and magenta, respectively. Secondary structure information (green bar for ␣-helix and red arrow for ␤-strand) based on the crystal structures of ␣NcatC is shown.
To gain further insights into how functionally distinct N, M, and C domains are physically coupled together in ␣-catenin, we performed the SAXS analysis of full-length ␣N-catenin. We determined that ␣N-catenin is suitable for the SAXS analysis, as analytical ultracentrifugation (Fig. 5A), size exclusion chromatography with multi-angle light scattering (supplemental Fig. S8A), blue native polyacrylamide gel electrophoresis (BN-PAGE) (supplemental Fig. S8B), and electron microscopy (supplemental Fig. S8C) confirmed that it predominantly exists as a monomer in solution (20). In contrast, ␣E-catenin exists as both monomer and dimer in vitro (14,20). We also show that our recombinant full-length ␣N-catenin protein is functionally active, as it binds to F-actin in its monomeric state (Fig. 5B). Based on the scattering curve (Fig. 5C), the ab initio molecular envelope of full-length ␣N-catenin was calculated (Fig. 5D). It has an elongated shape that reflects the empirically determined radius of gyration (R g ) of 51 Å and maximum dimension (D max ) of 175 Å (Fig. 5E). Computational docking of ␣EcatNM and ␣NcatC crystal structures into the molecular envelope of ␣N-catenin revealed that the C domain is located near the M III bundle of the NM domains (Fig. 5D). The relative location of the C domain is further supported by a good agreement of our full-length ␣N-catenin model (supplemental Fig. S9A) with the experimental scattering curves (Fig. 5C, supplemental Fig. S9B) and the shorter D max (154 Å) (Fig. 5F) of the ␣N-catenin mutant devoid of the C domain (␣NcatNM), which yielded a truncated molecular envelope (supplemental Fig. S9C). The orientation of the docked C domain relative to the NM domains in the fulllength model is consistent with our fluorescence resonance energy transfer (FRET) studies (supplemental Fig. S9, D-F), which indicated that the ␤-hairpin motif of the C domain is in close proximity to the C terminus of the M domain. These observations confirm that ␣N-catenin predominantly exists as a monomer with the elongated conformation that exposes a large portion of the actin-binding C domain (Fig. 5D). This structural feature is also consistent with full-length ␣N-catenin binding to F-actin during in vitro co-sedimentation assays (Fig.  5B). Comparison of the SAXS profile of ␣N-catenin with previously reported R g and D max values of related molecules (Fig.  5F) (60,61) suggests that the monomeric ␤-␣E-catenin chimera likely adopts an elongated conformation similar to ␣N-catenin, whereas HMP-1, an ␣-catenin homolog from Caenorhabditis elegans, is more likely to resemble the globular conformation of vinculin (56,57). Superposition of our full-length ␣N-catenin model and a crystal structure of full-length vinculin (PDB code 1ST6) reveal that the assembly of N and M domains of ␣E-catenin is remarkably similar to the pincer-shaped arrangement of D1-D4 domains in vinculin (Fig. 6, A and B), however, unlike the D5 domain in vinculin, the C domain does not fold back toward the N terminus (Fig. 6C). Taken together, we propose that ␣Eand ␣N-catenin monomers in solution exist in the elongated conformation comprising the autoinhibited VBS with the C domain unhindered for F-actin interaction.

DISCUSSION
In this study, we have determined the three-dimensional structure of full-length ␣-catenin in the autoinhibited state, revealing its unique multidomain assembly necessary for ␣-catenin to act as a mechanosensor at AJs. The elongated structure of ␣-catenin consists of the centrally located, conformationally switchable modulatory domain flanked by the N-terminal ␤-catenin-binding domain and the C-terminal actin-binding domain. This domain arrangement and the fully accessible key ligand-binding sites on both N and C termini appear to be crucial for mechanically coupling the individual domain functionality to the conformational states of ␣-catenin. Moreover, it differentiates ␣-catenin from vinculin, which likely evolved from a common ancestral actinbinding protein involved in cell polarity of unicellular organisms (61).
Recently, the dimer structure of human ␣E-catenin mutant lacking the first 81 residues (⌬N-␣E-catenin) was determined (62). It revealed that the N and M domains are similarly arranged as the mouse ␣EcatNM structure (Fig. 1B), but the relative location of the C domain differs from where this domain is positioned in our full-length mouse ␣N-catenin model. Interestingly, the C domain of ⌬N-␣E-catenin is folded back toward the M domain, but the relative orientation of C domains of the two protomers in the ⌬N-␣E-catenin dimer differs by almost 180° (62). Furthermore, the D max of the human ⌬N-␣E-catenin crystal dimer (ϳ130 Å) and the protomers in the dimer (ϳ105 Å) are considerably shorter than previously reported D max values of mouse ␣E-catenin dimer and monomer in solution (Fig. 5F) (60). We believe that variations of the C domain location among our ␣N-catenin model and the two protomers of the human ⌬N-␣E-catenin dimer may have resulted from the highly mobile nature of the C domain in both solution and crystal. In the crystallographic studies of human ⌬N-␣E-catenin (62), a crystal dehydration procedure employed to improve the crystal diffraction quality may have contributed to the observation of the compact conformation and domain rearrangements (63). In essence, these results together strongly suggest that the C domain of ␣-catenin is relatively mobile and not likely to adopt a fixed conformation within a full-length molecule in solution.
In our proposed model (Fig. 7A), unique "open" and "closed" conformations of ␣-catenin at AJs allow this molecule to couple the activation of autoinhibited VBS to the tension-sensitive actin-bound state of the C domain described previously (17). This "switch"-like mechanism provides the adjustable connectivity between cell junctions and the actin cytoskeleton necessary to counterbalance the actomyosin-generated tension by recruiting additional actin-binding protein vinculin to cell junctions (Fig. 3) (17). In the no-tension state, autoinhibited ␣-catenin forms a "closed (off)" conformation with the C domain of ␣-catenin capable of directly interacting with F-ac-  Fig. S3A, S2E, and S7A). A, the elongated structure of full-length ␣N-catenin. B, the full-length crystal structure of vinculin (PDB code 1ST6) in the globular state. The actin-binding D5 domain forms interdomain interactions with the vinculin head region consisting of D1-D4 domains in a "pincer-shape" (56). C, superposition of ␣N-catenin and vinculin (gray). The D2 domain of vinculin is shown as a transparent ribbon to help visualize other structural features. The spatial arrangements of the N and M domains of ␣-catenin highly resemble the "pincer"-shaped arrangement of D1, D3, and D4 domains required for clamping the D5 domain and partially occluding the actin-binding sites (56).
tin, but the VBS remains in the bundled conformation (Fig. 7A). In the high-tension state, ␣-catenin adopts the "open (on)" conformation as the contraction of actomyosin filaments "pulls" the actin-bound C domain and triggers multistate conformational changes of the M domain, involving unclasping of the M I bundle by the inhibitory M III region and the subsequent "unfurling" of the VBS (Fig. 7A). As the activated pool of vinculin is expected to interact with the VBS of ␣-catenin (29,30), it is likely that vinculin localized at AJs also engages in actin binding, hence further reinforcing the connectivity between the cadherin-catenin complex and the F-actin. The strength of anchoring the AJ to the actin cytoskeleton can be dynamically regulated to match the actomyosin-generated tension by adjusting the polarized distribution of the cadherin-catenin complex (64), as well as the number of activated ␣-catenins and vinculins at AJs (Fig. 7B). On the other hand, the dissipation of contractile forces would favor the ␣-catenin-vinculin interaction to dissociate and return ␣-catenin into the no-tension state signified by the "closed" conformation (Fig. 7B). In epithelia, the C domain of ␣E-catenin could directly bind to F-actin and/or indirectly through EPLIN to further stabilize the actin belts at zonula adherens (12). The relatively limited capacity of the C domain of ␣N-catenin to activate the VBS in R2/7 cells suggests that neuronal cells, the major cell types requiring ␣N-catenin (65) for their interactions, could express other regulators for the conformational changes of this catenin, which are missing in this carcinoma line.
Collectively, the present study provides the first bona fide structure of the autoinhibited state of ␣-catenin for its vinculinbinding activity, which is critical for the tension-dependent regulation of the linkages between the cadherin-catenin complex and the actin cytoskeleton. Unlike the globular structure of vinculin, the elongated conformation of ␣-catenin in an autoinhibited state suggests that ␣-catenins at AJs could associate with F-actin, but requires actomyosin-generated mechanical force (17) to trigger the conformational changes within the M domain and activate its vinculin-binding activity. Varied tension sensitivity of ␣-catenin and conformational dynamics displayed by different ␣-catenin isoforms and homologs are likely tailored to their tissue-/organismspecific functions, and could be further adjusted by other binding partners and post-translational modifications. The multidomain assembly of ␣-catenin revealed in this study is crucial for further delineating how ␣-catenin serves as a focal point to integrate various signals into regulating the stability of the AJs.