α-Catenin–mediated cadherin clustering couples cadherin and actin dynamics

Classical cadherins are the principal transmembrane receptors of the polymorphic cell–cell adhesive structures generally known as adherens junctions (AJs). AJs establish tight but highly dynamic contacts between cells in virtually all multicellular tissues. One of the key unanswered questions is whether and how intracellular proteins regulate extracellular cadherin adhesive activity. Although it has been proposed that such inside-out cadherin signaling is based on cadherin oligomerization by the cytoskeleton (Yap et al., 1997; Adams and Nelson, 1998; Kusumi et al., 1999; Gumbiner, 2005), no direct evidence that such a mechanism controls a cell adhesion in vivo has yet been demonstrated.

In recent years, significant progress has been made in understanding the extracellular cell–cell adhesive interface of AJs, which is organized by trans- and cis-intercadherin interactions (Wu et al., 2010; Brasch et al., 2012; Troyanovsky, 2012). Together, these two interactions produce an ordered adhesive structure that interconnects adjacent cells (Harrison et al., 2011). However, these structures, assembled exclusively through cadherin ectodomains, are quite unstable with respect to their lifetimes, morphologies, and mobility of their components. The stability of such ectodomain-based junctions is significantly increased upon their anchorage to the actin cytoskeleton (Hong et al., 2013). Furthermore, cadherin molecules defective for cis-interactions and, therefore, unable to form clusters via their extracellular regions, gain the ability to do so if they interact with actin through a covalently attached actin-binding domain (Hong et al., 2013). These observations suggest that actin filaments can collaborate with extracellular interactions in the formation of AJs. This prompted us to study actin-based mechanisms relevant to cadherin clustering.

Intracellular components of AJs recruit dozens of actin-binding proteins that could in principle participate in cadherin–actin interactions (Kobielak and Fuchs, 2004; Green et al., 2010; Niessen et al., 2011; Ivanov and Naydenov, 2013). Of these, only α-catenin seems to be indispensable. The N-terminal head domain of this protein interacts with cadherin through β-catenin, its C-terminal actin-binding domain (αABD) binds actin filaments, whereas its middle domains (M1, M2, and M3) are thought to control the vinculin-binding site located in the M1 domain (Gomez et al., 2011; Yonemura, 2011). In addition to vinculin, several other actin-binding proteins including EPLIN, ZO1, afadin, α-actinin, spectrin, merlin, and ajuba have been shown to interact with α-catenin (Kobielak and Fuchs, 2004) providing alternative indirect ways for α-catenin to bind F-actin. At least two of these proteins, EPLIN and ZO1, which directly interact with αABD, might constitute alternative linkers between αABD and actin filaments (Imamura et al., 1999; Abe and Takeichi, 2008).

Whether direct αABD binding to actin or alternative indirect mechanisms couple cadherin to actin in AJs is unclear. In vitro binding assays clearly show that β-catenin binding to α-catenin reduces the actin-binding potential of α-catenin (Drees et al., 2005; Yamada et al., 2005), presumably by inhibiting direct αABD–actin interactions. This suggests that αABD, in the context of the cadherin–catenin complex, can interact with actin only if its actin-binding activity is derepressed. On the other hand, there is strong evidence that direct or indirect αABD interaction with actin is an essential step in AJ formation (Pappas and Rimm, 2006; Desai et al., 2013; Thomas et al., 2013) and in initiating a mechanotransduction pathway, resulting in the recruitment of vinculin (Yonemura et al., 2010).

In this study, we explore the role of direct αABD–actin interactions on AJ structure and dynamics. To this end, we identified a set of αABD point mutants unable to interact with actin in vitro. These point mutants allowed us to show that a direct αABD–actin interaction stabilizes AJs, links them to actin filaments, and initiates vinculin recruitment by α-catenin. We also present evidence that AJs connected to actin via αABD, in contrast to those connected to the cytoskeleton through vinculin, are highly dynamic. Exploring the underlying mechanism of these differences in dynamics, we found that αABD binds only to actin filaments located in the cell cortex. This binding, which is transient and cooperative, generates short-lived αABD clusters whose lifetimes are controlled by the turnover of actin filaments. We propose that these transient αABD clusters, formed on cortical actin filaments, facilitate clustering of cadherin molecules and mediate AJ dynamics.

αABD consists of a five-helix bundle (residues 671–841) and a C-terminal extension (residues 842–906), which has been seen in two conformations (Ishiyama et al., 2013; Rangarajan and Izard, 2013). In one, the closed conformation, the side chain of the conserved W859 (Fig. S1) inserts into the bundle and the C-terminal extension forms an interface with the bundle. In the open conformation the extension is largely unstructured (Fig. 1 a). It is not known whether either of these is the conformation that binds actin. Previous actin cosedimentation experiments with αABD identified an 18–amino acid–long stretch (residues 865–883) as an actin-binding region (Pappas and Rimm, 2006). Using two recombinant GST fusion proteins, GST-α(671–883), which bound actin, and GST-α(671–864), which failed to bind actin, we confirmed the importance of the 865–883 region in F-actin binding (Fig. 1 b).

A series of three–amino acid alanine substitutions was introduced within the C-terminal extension, and the resulting mutants were tested for interaction with F-actin. These experiments revealed several triple-alanine mutations that significantly decreased this interaction (Fig. 1 c). Four of these mutations were selected for alanine-scanning mutagenesis, which ultimately identified four residues—K842, L852, K866, and L869—that contribute to αABD binding to F-actin (Fig. 1 c). Of these, K866 and L869 are fully exposed in one of the two crystal structures and unstructured in the other. K842 and L852 are buried to a different extent in each structure and, in both cases, link the C-terminal extension to the five-helix bundle. The bundle itself is unlikely to be affected by any of these mutations. Consistently, circular dichroism experiments suggest that the presence or absence of the 866–906 C-terminal region does not impact the overall structure of αABD (Pappas and Rimm, 2006). Collectively, the results suggest that the four residues we have identified are critical to actin binding either through direct contacts or through an indirect structural role in the binding region.

To further characterize a subset of these mutants, we performed F-actin cosedimentation assays with increasing concentrations of αABD and a fixed concentration of F-actin (Fig. 1 d). Curve fitting yielded an apparent K_D for the intact protein of ∼1 µM (Fig. 1 e). This value is close to that previously reported, ∼0.5 µM (Pappas and Rimm, 2006). The K842A mutant showed some evidence of saturation only at its highest concentration, suggesting that its affinity for F-actin is ∼30 µM (Fig. 1 e). The binding affinities of K866A and the deletion mutant GST-α(671–864) were too low to be measured.

Recently, a point mutation, I997A, was identified in the actin binding domain of vinculin that decreases actin binding but has little effect on vinculin folding (Thievessen et al., 2013; Thompson et al., 2014). Because I997 is conserved between vinculin and α-catenin (Fig. S1), we tested the analogous mutant in αABD, I792A. Cosedimentation assays showed that this mutation also significantly decreased binding of αABD to actin (Fig. 1 c). I792 is fairly distant (over 25 Å) from the C-terminal extension; it is buried in an interdomain interface in full-length α-catenin but is exposed on the surface of αABD in isolation (Fig. 1 a).

Having developed mutants that interfere with αABD actin binding, we are now able to assess the role of this interaction in living cells. We chose three of the characterized mutations (K842A, K866A, and I792A) for further experiments.

We first studied the cadherin–α-catenin chimera EcΔ-Dn-α(280–906). The E-cadherin portion of this chimera harbors an extensive cytoplasmic deletion that removes all known intracellular protein binding sites, including the site for p120 (Fig. 2 a). The absence of the latter site, which had not been removed from the cadherin–catenin chimeras used in previous studies (Nagafuchi et al., 1994; Imamura et al., 1999; Desai et al., 2013), excluded an involvement of p120 in any chimera-associated effects. The deleted region was replaced with the photoswitchable fluorescent protein Dendra2 and the C-terminal part of α-catenin (aa 280–906), which includes αABD and all three α-catenin middle domains (M1–M3). The lack of the head domain of α-catenin (aa 1–279) in this chimera simplified data interpretation because it excluded the possibility that this chimera interacted with endogenous β-catenin or underwent homodimerization.

The EcΔ-Dn-α(280–906) chimera produced well-defined intercellular junctions in cadherin-deficient A431D cells (Fig. 2 a), which were very similar in morphology to AJs in wild-type (wt) A431 cells (Indra et al., 2013). Consistent with previous data (Imamura et al., 1999), we found that these junctions recruited vinculin and were attached to radial actin bundles, which were integrated with a network of actin bundles connected to other junctions or to focal adhesions (Fig. 2 a). However, when the chimera contained either the K866A or I792A mutations, these junctions were completely abolished (Fig. 2 b). The K842A mutation, which only partially weakened αABD binding to actin in the in vitro assay (see also experiments with Dn-αABD below), did not significantly change the junction-forming properties of the chimera (Fig. 2 b). The correlation between the mutant’s abilities to form junctions and their actin-binding affinities suggests that the actin binding site of αABD is essential for the assembly of the chimera’s junctions.

The presence of vinculin in the chimera’s junctions made it unclear whether αABD continuously anchors the junctions to actin filaments or whether this interaction is transient and is needed only for vinculin recruitment, which then would play the major actin-anchoring function. To clarify this issue, we mutated five key residues of the chimera’s vinculin binding site (ΔVin mutation). The resulting chimera still formed actin-associated junctions, which, however, were completely devoid of vinculin (Fig. 2 c). The majority of these junctions were also devoid of ZO-1 and EPLIN (Fig. S2, arrows), thus excluding the possibility that the binding of either of these two proteins participates in indirect interaction of αABD with the cytoskeleton in our ΔVin chimera. Collectively, our data strongly suggest that αABD mediates the direct binding of the EcΔ-Dn-α(280–906) chimera to actin and that this interaction drives junction formation.

To explore the role of αABD binding to actin in real AJs, we used α-catenin–deficient MDA-MB-468 (468) cells (Figs. 3, 4, 5, 6 and S3). In these cells, E-cadherin forms a complex with β-catenin and p120 that could potentially interact with several actin-binding proteins including vinculin (Hazan et al., 1997; Peng et al., 2010; Ray et al., 2013). Nonetheless, 468 cells cannot form cohesive epithelial sheets (Troyanovsky et al., 2011) or interconnect their numerous E-cadherin lateral clusters with the cytoskeleton (Fig. 3 a). These clusters are likely formed by cis- and trans-extracellular interactions (Harrison et al., 2011) and by p120- or β-catenin–dependent intracellular interactions. The majority of these clusters were devoid of vinculin, EPLIN, and ZO1 (Fig. S3). The lack of detectable association between these clusters and actin allowed us to use 468 cells to investigate the role of αABD in AJs.

Expression of Dendra2-tagged α-catenin (Dn-αCat) in 468 cells restored epithelial organization of their cell–cell contacts including the apicolateral belts of tight junctions (TJs) and AJs (Fig. 3, b and c; and Fig. 4). In addition, the Dn-αCat–reconstituted cells produced two other types of AJs (Fig. 3, b and c) that are also typical of epithelial cells (Takeichi, 2014). Numerous spot-like junctions, variable in their sizes and shapes, formed on their lateral surfaces. These lateral junctions, in contrast to the apicolateral junctions, were devoid of vinculin and were not attached to the phalloidin-positive actin structures (Fig. 3 b). These features are similar to those observed for the lateral cadherin clusters of the parental cells. The third type of junctions, vinculin-containing “basolateral” junctions, was located at the base of the cell–cell contacts. These junctions were symmetrically attached to the actin bundles, which eventually merged with the dense network of stress fibers (Fig. 3, b and c).

We first sought to confirm published data that α-catenin lacking the vinculin binding site is able to form actin-associated AJs (Huveneers et al., 2012; Twiss et al., 2012; Desai et al., 2013). Indeed, 468 cells expressing the ΔVin mutant of α-catenin produced vinculin-deficient junctions, the majority of which were also deficient for ZO1 and EPLIN (Fig. S3). These junctions, still aligned with the robust actin-enriched structures (Fig. 5, a–c), exhibited noticeable defects: the apicolateral AJs as well TJs were fragmented (Fig. 4) and the actin filaments associated with the basolateral junctions were not integrated with the network of stress fibers (Fig. 5 a, basal focus plane). Thus, although α-catenin binding to vinculin is required for normal organization of the AJs, vinculin-deficient AJs, similar to the vinculin-deficient junctions produced by the EcΔ-Dn-α(280–906)-ΔVin chimera, still interact with actin.

Next, we introduced the K866A or I792A point mutations into the α-catenin ΔVin mutant. As expected, the resulting mutants Dn-αCat-ΔVin+K866A (Fig. 5 c) or Dn-αCat-ΔVin+I792A (Fig. 4) associated with the endogenous lateral cadherin clusters because their β-catenin–binding region remained intact. Importantly, morphologically these clusters were indistinguishable from the cadherin clusters of the wt 468 cells (compare Fig. 5 c with Fig. 3 a). Also similar to the parental cells, no actin-enriched structures were detected in association with these clusters (Fig. 5 c) and no TJs were found between these cells (Fig. 4). Basolateral AJs were also undetectable. Such a dramatic effect of the actin-uncoupled αABD point mutations on the ability of the α-catenin ΔVin mutant to assemble AJs provides strong evidence that this protein uses its αABD to directly interact with actin filaments in AJs.

The experiments described in this paper suggest that α-catenin interacts with actin in AJs in two ways: a direct interaction through its αABD or an indirect interaction mediated by vinculin. The α-catenin ΔVin mutant, in contrast, can use only its αABD. To reveal specific properties of AJs linked to actin only through vinculin, we constructed the α-catenin deletion mutant, Dn-αCat(1–505) (Fig. 6 a). Its deletion, (aa 506–906) encompassing αABD and the M3 domain, permanently activates the vinculin-binding site located in the α-catenin M1 region (Yonemura et al., 2010; Thomas et al., 2013). Consequently, the Dn-αCat(1–505) mutant can interact with actin only or predominantly through association with vinculin. Indeed, this mutant produced vinculin-enriched AJs (Fig. 6 b). Although these junctions clearly associated with actin, their morphology was abnormal: the apicolateral junctions were mostly fragmented (Figs. 4 and 6 a, apical focus plane), and the basolateral junctions, despite forming actin-enriched structures, were dislocated from the actin bundles (Fig. 6 a, basal focus plane). Thus, each of these actin-associated AJs, apicolateral and basolateral, require both vinculin and αABD to achieve their normal morphology.

Using a Dendra photoconversion assay, we compared the dynamics of the apicolateral junctions formed in 468 cells by intact α-catenin with those formed by Dn-αCat-ΔVin and Dn-αCat(1–505). The results presented in Fig. 6 c reveal remarkable differences. The Dn-αCat-ΔVin–based junctions, connected to actin via αABD only, were very dynamic, losing nearly 30% of their photoconverted fluorescence within 3 min. The junctions formed by the wt α-catenin were slightly more stable. Surprisingly, the Dn-αCat(1–505)–based junctions that were linked to the cytoskeleton through vinculin were extremely stable: their photoconverted fluorescence showed no changes during the same time period. These observations suggest that αABD makes junctions more dynamics, whereas addition of vinculin enhances its strength. In addition, it appears that wt AJs are controlled primarily by αABD-mediated dynamics.

To better understand the role of αABD in AJs, we studied the binding of the Dendra2-tagged αABD (Dn-αABD) to the cytoskeleton in A431 cells. Phalloidin staining showed that this fusion protein decorated cortical actin but not stress fibers or bundles associated with AJs (Fig. 7 a, arrows). This finding was further validated in experiments with Latrunculin A. Its high concentration (4 µM) completely depolymerized the actin cytoskeleton and concomitantly abolished Dn-αABD filamentous organization (Fig. 7 b). At a low concentration (0.4 µM), Latrunculin A left intact some of the bundles. Yet, these bundles remained completely devoid of Dn-αABD, which colocalized with the residual clumps of the cortical filaments (Fig. 7 c). To confirm αABD localization along the actin cortex, we selectively removed actin bundles while maintaining filaments by blocking myosin II activity with Blebbistatin. This treatment did not affect αABD distribution (Fig. 7 d). In the absence of bundles, it became obvious that cortical filaments were not evenly decorated by Dn-αABD, suggesting that some filaments have specific preferences for interactions with αABD.

The K866A or I792A point mutations of αABD completely abolished its association with the cortex (Fig. 7 e). Another mutation, K842A, in contrast, had no obvious effects on Dn-αABD localization. The clearest differences between these proteins were revealed by confocal microscopy: although the intact Dn-αABD was exclusively cortical, its K866A or I792A mutants were cytosolic. The K842A mutant exhibited an intermediate phenotype (Fig. 7 f). Interestingly, the inability of αABD to interact with actin bundles was evident even when recombinant αABD was added to the permeabilized cells (Fig. 7 g).

To increase spatial resolution, we examined the distribution of Dn-αABD using single molecular localization microscopy (SMLM). This technique, which is able to locate individual Dendra2 molecules at 30-nm resolution, confirmed that Dn-αABD does not uniformly bind to the cortex filaments (Fig. 8). Instead, Dn-αABD was localized in numerous clusters. The longest axis of the clusters reached 400 nm and the apparent density of molecules in the cluster reached ∼5 × 10⁴ molecules/µm² (Fig. 8), consistent with the results of Hansen et al. (2013) who found that actin filaments can be completely covered by αABD. We also imaged cells expressing the Dn-αABD-I792A mutant or Latrunculin A–pretreated cells expressing Dn-αABD. Uncoupling of αABD from actin using either of these techniques dramatically changed cluster appearance: the size of the clusters increased concomitantly with a significant decrease of their molecular density (Fig. 8). The latter value for the majority of the Dn-αABD-I792A clusters was ∼2 × 10³/µm², a ∼25-fold reduction from wt values. After latrunculin treatment, the individual clusters were impossible to demarcate.

Dn-αABD and its mutants were photoconverted in a small area, and the resulting red fluorescence of the spot was monitored over time. Fig. 9 a shows that the area of its red fluorescence remained constant in size but rapidly decreased in intensity (t_1/2 = ∼10 s). In contrast, the photoconverted spot of its K842A mutant rapidly expanded in size. A drop in red fluorescence of the other two mutants (K866A, I792A) was too fast to be accurately measured. These results suggest that wt αABD remains bound to the same location, whereas the three mutants diffuse away at rates related to their binding properties (Fig. 1 c). The fast decay of wt αABD fluorescence despite its constant spatial localization is consistent with a high turnover of the actin filaments themselves. This explanation requires that a pool of αABD that dissociates from a filament due to actin depolymerization must have a higher probability of diffusing away than of rebinding to a neighboring actin filament.

To verify the role of actin filament turnover in αABD dynamics, we globally blocked active intracellular processes by ATP depletion or specifically arrested actin dynamics by a triple-drug cocktail containing Jasplakinolide, Latrunculin B, and Y27632 (Peng et al., 2011). Both these approaches significantly slowed down dissipation of the red fluorescence in case of the intact protein but not of its K842A mutant (Fig. 9 b).

In a complementary approach, we photoconverted Dn-αABD at one part of the cell and tracked the photoconverted molecules in the nonconverted “dark” part of the same cell using high temporal resolution total internal reflection fluorescence (TIRF) microscopy. We expected that this imaging technique, known as fluorescent speckle microscopy (FSM; Danuser and Waterman-Storer, 2003), would reveal dynamics of the Dn-αABD clusters. Indeed, the photoconverted Dn-αABD was quickly recruited into the numerous speckles in the dark cell areas (Fig. 9, c and d; and Video 1). The modal lifespan of the speckle was ∼200–400 ms (Fig. 9 e), and new speckles were produced at the rate of 0.2 speckles/s/1 µm². The majority of the speckles showed virtually no displacement (Fig. 9 f). As expected, Dn-αABD, uncoupled from actin either by the point mutation I792A or by actin depolymerization by Latrunculin A, was unable to form the speckles (Figs. 8 and 9 c and Videos 2 and 3).

It is possible that each speckle originates from the fluorescence of a single Dn-αABD molecule. In this case, speckle dynamics represents a behavior of a single molecule in the cluster. Alternatively, a speckle could be formed as a result of simultaneous incorporation of many photoconverted molecules into the same cluster. In the latter case, speckle dynamics would reflect the dynamics of the αABD clusters. To distinguish between these possibilities, we fixed the cells 1 min after photoconversion, and the speckles formed around the photoconverted area were gradually photobleached with a 561-nm laser. Fig. 9 g shows that the speckle photobleaching kinetics were not stepwise, indicating that each speckle fluorescence was generated by numerous molecules, thus favoring the second possibility—that speckle dynamics represent dynamics of the αABD clusters. Collectively, this live-imaging study shows that αABD clusters are continuously assembled and disassembled along the actin cortex with a fast turnover rate.

In the aforementioned experiments, we used isolated αABD. To validate that this domain is functional in full-length α-catenin, we expressed Dn-αCat and its mutant Dn-αCat-I792A in cadherin-deficient A431D cells at levels comparable to those of endogenous α-catenin (Fig. S5 a). Both proteins were cytosolic and their photoconverted fluorescence dissipated rapidly (Fig. S5 b). These observations suggested that Dn-αCat is a freely diffusing cytosolic protein in cadherin-deficient A431D cells. FSM, however, revealed that the Dn-αCat forms numerous speckles (Fig. S5, c and d), the mean lifetime of which was ∼400 ms shorter than speckle’s lifetime of isolated αABD (Fig. S5 d, legend). Speckle formation was undetectable in cells expressing the Dn-αCat-I792A mutant. The predominant cytosolic localization of Dn-αCat in absence of cadherin and the short lifetime of its actin-bound clusters may be caused by differences in cooperativity between the binding of αABD and full-length α-catenin to actin filaments (Hansen et al., 2013).

We then studied whether αABD could cluster the cadherin–catenin complex. We constructed a Dendra-tagged tail domain of E-cadherin linked to the plasma membrane through a myristoylation signal (Ms-Dn-EcTail; Fig. S5), which is known to form a complex with the endogenous β- and α-catenins (Nieman et al., 1999). The Ms-Dn-EcTail was found to be concentrated in the areas enriched with cortical actin (Fig. S5 f), and FSM revealed the formation of numerous very sharp speckles. Latrunculin A abolished formation of such speckles, whereas some irregularities in the distribution of the Ms-Dn-EcTail remained (Fig. S5 g). It is possible that these irregularities, which were undetectable in the Latrunculin A–treated Dn-αABD– or Dn-αCat–expressing cells, were caused by some β-catenin–based interactions. Speckle stability was approximately the same as that of free α-catenin (compare Fig. S5, d and h). To show that speckle formation was driven by αABD, we expressed Ms-Dn-EcTail in cells in which α-catenin had been stably depleted by α-catenin–specific shRNA or in which α-catenin or its I792A mutant was reexpressed in the α-catenin–depleted cells. In agreement with the role of αABD in E-cadherin tail clustering, α-catenin–depleted cells as well as those expressing the I792A mutant exhibited the same pattern of Ms-Dn-EcTail distribution as control Ms-Dn-EcTail–expressing cells after Latrunculin A treatment (Fig. S5 g).

The cadherin ectodomain alone can form junction-like structures through trans- and cis-intercadherin interactions (Wu et al., 2010; Hong et al., 2010, 2013; Harrison et al., 2011). These junctions, which likely resemble initial points of cell–cell attachment, are too dynamic and apparently too weak to maintain stable cell–cell adhesion unless they are reinforced by interactions with the actin cytoskeleton (Hong et al., 2013). As a step toward elucidating the interplay between intracellular, actin-binding, and extracellular, adhesive, processes in AJs, we sought to determine how α-catenin interacts with the actin cytoskeleton.

Although in vitro experiments have demonstrated that αABD binds directly to F-actin (Rimm et al., 1995; Pokutta et al., 2002; Pappas and Rimm, 2006), the role of this binding in vivo has been unclear. Indeed, αABD can also bind to actin indirectly through the actin-interacting proteins ZO1 and EPLIN (Imamura et al., 1999; Abe and Takeichi, 2008), whereas α-catenin in a cadherin–catenin complex in solution cannot bind actin (Drees et al., 2005; Yamada et al., 2005). Observations showing that αABD deletion attenuates AJ formation (Ozawa, 1998; Imamura et al., 1999; Yonemura et al., 2010; Twiss et al., 2012; Desai et al., 2013; Maiers et al., 2013; Thomas et al., 2013) could be interpreted to imply that deletion of αABD abolishes the binding of α-catenin with vinculin, EPLIN, or ZO1 by removing possible modes of indirect interactions.

To clarify whether the in vivo interaction of αABD with actin in AJs is direct, indirect, or both, we exploited two new αABD point mutations, which dramatically decrease αABD actin-binding activity in vitro. These mutations were inserted into the chimeric protein EcΔ-Dn-α(280–906), which forms actin-associated AJ-like structures in A431D cadherin-deficient cells. No such structures were observed when the mutant chimeras were expressed. In contrast, the EcΔ-Dn-α(280–906)-ΔVin chimera, missing only the vinculin binding site, formed actin-associated junctions. ZO1 and EPLIN were also absent from these junctions. Together, these results strongly suggest that, in living cells, α-catenin can associate with actin through direct binding of αABD. However, these results, obtained using chimeric proteins, do not address the possibility that β-catenin–α-catenin interactions inhibit αABD in wt AJs.

To address this issue, we tested the α-catenin mutants in the context of full-length α-catenin transfected into α-catenin–deficient 468 cells. Consistent with results obtained with other vinculin-uncoupled α-catenin mutants (Huveneers et al., 2012; Twiss et al., 2012; Desai et al., 2013) and with vinculin-depleted cells (Watabe-Uchida et al., 1998; Taguchi et al., 2011), the α-catenin ΔVin mutant formed AJs, which, despite being devoid of vinculin, were still associated with an actin-enriched scaffold. Insertion of point mutations that inhibit αABD binding to actin completely abolished this association as well as AJ formation. Collectively, our results obtained using two cell models, A431D and 468 cells, strongly suggest that αABD directly links the cadherin–catenin complex to actin and that this interaction is sufficient for AJ assembly. This implies that the very weak actin-binding activity of α-catenin in the cadherin–catenin complex detected in solution (Drees et al., 2005; Yamada et al., 2005) is enhanced in a cellular context. Whether this activation is based on force-dependent strengthening of αABD binding to actin (Buckley et al., 2014) and/or on cis-intercadherin interactions that could potentially support αABD–actin binding via a cooperative mechanism remains unclear.

A striking feature of the vinculin-deficient junctions is the very fast turnover of the α-catenin mutant Dn-αCat-ΔVin (Fig. 6 c). In this respect, these junctions are similar to AJs that incorporate wt α-catenin, which are also highly dynamic (Fig. 6 c). In contrast, the α-catenin mutant Dn-αCat(1–505), which assembles junctions through vinculin, is significantly less mobile. These findings suggest that the direct binding of αABD to the cytoskeleton is compatible with continuous reassembly of AJs, a process that has been previously reported (Adams et al., 1998; Yamada et al., 2005; Lambert et al., 2007; Cavey et al., 2008; de Beco et al., 2009; Canel et al., 2010; Hong et al., 2010). The fact that indirect, vinculin-based binding of α-catenin to actin is associated with much slower dynamics may suggest that vinculin, once it interacts with actin, suppresses AJ turnover. It is important to note that our staining for TJs (Fig. 4), as well as previous studies of force-dependent stabilization of AJs (Yonemura et al., 2010; Huveneers et al., 2012) or AJ-mediated adhesion strength (Thomas et al., 2013), suggested that neither αABD nor vinculin alone could produce fully functional AJs.

To understand how αABD binding to actin assembles AJs while also playing a role in AJ disassembly, we analyzed this binding in living cells. Our experiments reveal three previously unknown features of αABD–actin interactions. First, αABD binds exclusively to the actin cortex, a network of actin filaments attached to the inner face of the plasma membrane (Heuser and Kirschner, 1980; Svitkina et al., 2003; Morone et al., 2006). This observation suggests that AJs form their own actin bundles but do not interact with preexisting bundles, consistent with previous data showing the rapid reorganization of the actin cytoskeleton upon AJ formation (Adams et al., 1998; Vasioukhin et al., 2000; Mège et al., 2006). We speculate that some bundle-associated proteins, such as tropomyosin, might interfere with αABD binding either by masking binding sites along the filaments or by stabilizing a filament structure that is incompatible with αABD binding.

The second feature is that αABD, instead of being randomly bound to the cortex, exhibits a highly nonuniform distribution on actin filaments. This can be seen even with conventional immunofluorescence microscopy (Fig. 7 d), whereas superresolution SMLM shows the formation of discrete dense αABD clusters consisting of hundreds of molecules (Fig. 8). The formation of such clusters is consistent with in vitro experiments showing that αABD–actin binding is a highly cooperative process: actin filaments have a tendency to be either fully decorated by αABD or not decorated at all (Hansen et al., 2013), an observation that we have reproduced (unpublished data).

The third feature is that cortical αABD clusters are transient and dissociate in a time on the order of seconds (Fig. 9). Experiments with inhibitors of actin dynamics suggest that, at least in part, such short cluster lifetimes result from actin filament depolymerization. Consistently, a previous study found that the lifetime of individual actin filaments in a cell–cell contact area is ∼10 s (Yamada et al., 2005), whereas a more recent study suggested that a subpopulation of cortical filaments might have a lifetime significantly less than 5 s (Fritzsche et al., 2013).

Each cell in a multicellular tissue must continuously readjust its AJs according to changes imposed by its own motility and by the motility of adjacent cells. At least in part, this AJ plasticity is mediated by the intrinsic flexibility of the AJ extracellular cadherin-mediated adhesive interface. This interface in vertebrate classical cadherins is mediated by a strand-swap trans-interaction and a cis-interaction, which together define an ordered structure that is similar for the type I cadherins (Harrison et al., 2011). The trans-adhesive interface exhibits fast rebinding kinetics as a result of binding intermediates called X-dimers (Harrison et al., 2010; Hong et al., 2011). Structural (Harrison et al., 2011), computational (Wu et al., 2010), cell biological (Hong et al., 2013; Strale et al., 2015), and biochemical (Klingelhöfer et al., 2002; Troyanovsky et al., 2006, 2007) studies suggest that the AJ adhesive interface in vertebrates is organized in numerous small adhesive clusters, whose individual instability and fast reassembly maintain both integrity and flexibility of the cell–cell interface.

Interactions between cadherin and the cytoskeleton, however, also appear to be unstable, allowing cadherin molecules to move in and out of AJs, with the lifetime of a cadherin molecule bound in an AJ estimated at less than a minute (de Beco et al., 2009; Hong et al., 2010). Our observation that αABD forms transient actin cortex-bound clusters (Figs. 8 and 9) suggests a source for this instability. Furthermore, it shows that both the adhesive interface and the actin-binding interface of AJs consist of numerous reassembling clusters. This concordance implies that the two types of clusters are functionally interconnected. Experimental observations on basal-to-apical flow of cadherin clusters also support this suggestion (Kametani and Takeichi, 2007). This work observed that basal–apical flow was disrupted upon αABD deletion, indicating that movement of cadherin clusters depends on αABD–actin clustering. Interestingly, although fluorescent speckle microscopy showed that αABD is able to cluster full-length α-catenin as well as the myristoylated E-cadherin tail (Fig. S5), we were unable to detect clustering of full-length E-cadherin (unpublished data). This may be caused by slow E-cadherin diffusion and a low concentration of this protein at the ventral membrane.

A plausible hypothesis for the mechanism of cadherin–αABD interplay is that a build-up of cadherins in cell–cell contact regions as a result of a diffusion trap mechanism and the subsequent formation of small ordered clusters involving trans- and cis-interactions between cadherin ectodomains, nucleates formation of αABD-based clusters inside the cell. The αABD-based clusters could, in turn, reinforce the intercadherin interactions in the ectodomain clusters, thereby enhancing their adhesive function. The resulting cadherin/αABD-based clusters—transient, as a result of actin depolymerization, and highly adhesive, as a result of anchorage to cortical actin filaments—could provide a partial basis for AJ plasticity. In this way, both the intracellular and extracellular regions of cadherins could collaborate to maximize cadherin concentration at intercellular junctions and to enhance the general stability of the junctions. Although additional studies are required to firmly establish this hypothesis, an interplay between these two types of clustering mechanisms would explain the highly dynamic nature of the αABD-only junctions (Fig. 6 c), the ATP dependence of cadherin turnover in AJs (Troyanovsky et al., 2006), and why polymerization of new actin filaments is required for AJ integrity (Vasioukhin et al., 2000; Zhang et al., 2005).

The plasmids (all in pRcCMV; Invitrogen) encoding chimeric protein EcΔ-Dn-α(280–906) presented in Fig. 2 were based on EcΔDn (Hong et al., 2010). The general maps of the EcΔ-Dn-α(280–906) chimera, Dn-α-catenin, Dn-αABD, and their mutants (also in pRcCMV) are presented in Figs. 2, 3, 4, and 5. The α-catenin mutation inactivating its vinculin binding site was constructed based on the structural studies (Choi et al., 2012; Rangarajan and Izard, 2012). It includes five amino acid substitutions to alanine: R329A, R330A, L347A, L348A, and Y351A. The shRNA-target region of α-catenin (5′-CCTGTTCCATCTCAAATAA-3′) in the plasmids used in α-catenin–silenced A431 cells was modified using PCR-directed mutagenesis. The original plasmid encoding human αE-catenin was published (Troyanovsky et al., 2011). All plasmid inserts were verified by sequencing.

Transfection and growth of A431D (provided by J.K. Wahl, The University of Nebraska, Lincoln, NE), A431, and 468 cells were performed as previously described (Troyanovsky et al., 2011). After antibiotic selection, the cells were sorted for moderate transgene expression by FACS. Lentiviral knockdown (shRNA) plasmids (V2LHS-262377; GE Healthcare) were obtained from C. Gottardi and A. Yemelyanov (Northwestern University, Chicago, IL). Before use, the GFP-encoding region of this plasmid was deleted. The infected cells were selected with 5 µg/ml puromycin.

For immunofluorescence, cells were fixed and permeabilized either with methanol-acetone or, in case of phalloidin staining, with 3% formaldehyde–1% Triton X-100, or, in case of anti-vinculin staining, with BM[PEO]3 (see Indra et al., 2013 for details). Wide-field images were taken using a microscope (Plan Apochromat 100×/1.40 NA objective lens; Eclipse 80i; Nikon) and a digital camera (CoolSNAP EZ; Photometrics). The images were then processed using NIS-Elements software (Nikon). The following antibodies were used: mouse anti–E-cadherin and anti-occludin (Invitrogen), rabbit anti-Dendra2 (Evrogen), mouse anti–β-catenin and ZO1 (BD), mouse anti-vinculin and rabbit anti-EPLIN (Sigma-Aldrich), goat anti–α-catenin (Santa Cruz Biotechnology, Inc.), rabbit anti–α-catenin N-terminal domain, EP1993Y (Abcam), and guinea pig anti-cingulin (provided by I. Hofmann, German Cancer Research Center, Heidelberg, Germany). Alexa Fluor 555 phalloidin and Latrunculin A were purchased from Invitrogen.

These experiments were performed essentially as described previously (Hong et al., 2010, 2013). In brief, cells were imaged (in L-15 plus 10% FBS) by a microscope (Eclipse Ti-E; Nikon) at 37°C controlled with NIS-Elements software. The microscope was equipped with an incubator chamber, a camera (CoolSNAP HQ2; Photometrics), Plan Apochromat 60×/1.40 NA and Plan Apochromat VC 100×/1.40 NA lenses, and halogen and mercury light sources. Time-lapse images were taken in both FITC and mCherry filter sets using halogen light that minimized phototoxicity and photobleaching. To analyze cadherin junctional turnover, we used a junctional Dendra photoconversion assay (Hong et al., 2013) in which the point of interest (φ = 2.5 µm) was photoconverted by a 100-ms-long exposure to the 402-nm wavelength laser. Time-lapse images were then taken in a red channel in a stream mode with 1 s (in Fig. 7 h) or in 20-s intervals with 1 s (Fig. 6 c) of image acquisition time. In some cases, the cells immediately before photoconversion were cultured for 5 min in ATP depletion media (Hong et al., 2010).

All images were saved as TIFF files and processed using ImageJ software (National Institutes of Health). In the Dendra photoconversion assay, the red fluorescent intensity was normalized in such a way that 0 and 1 corresponded to the background and the initial (immediately after activation) values. The background value was obtained from the image taken right before the photoconversion. The time course of the intensity change was produced from 10 sets of independent experiments. Mean values were calculated for each time point.

For SMLM, the cells were cultured on glass-bottom dish (P35G-1.5; MatTek Corporation) overnight and then fixed with 3% paraformaldehyde and 0.1% glutaraldehyde for 10 min at room temperature. After washing three times, the cells were immersed in the freshly prepared image buffer containing 50 mM Tris-HCl, pH 8.0, 10 mM NaCl, 10% glucose, 5 mg/ml glucose oxidase, 0.4 µg/ml catalase, and 0.1 M β-mercaptoethylamine. Samples were sealed immediately after adding image buffer. The N-STORM system (Nikon) with a camera (DU897; Andor Technology) was used for superresolution image acquisition. 20,000–40,000 images were acquired at 29-ms/frame exposure time via TIRF illumination, using a Plan Apochromatic TIRF 100× 1.49 NA objective lens. Dendra-fused proteins were activated with a 405-nm laser and acquired with 561-nm laser illumination. The images were rendered using the built-in N-STORM single molecule localization analysis algorithms.

For the spinning-disk confocal microscopy and FSM, the inverted microscope (Nikon-Ti) equipped with a TIRF illuminator module (Nikon) and spinning-disk unit (CSU-X1; Yokogawa Electric Corporation) was used. The cells were imaged in CO₂ stage incubator (37°C) during all live-cell imaging process. To identify the spatial distribution of Dn-αABD and its mutants, 0.3-µm-thick optical sections were taken through whole cells using confocal mode with 488-nm illumination. Optical sections were acquired using NIS-Elements software. For FSM, Dn-tagged molecules were photoconverted using a 405-nm laser within the area of ∼4 µm in diameter, and the converted molecules were imaged for 40 s in the adjacent nonphotoconverted area (20 × 20 µm²) using the 561-nm laser illumination (30% of intensity was used to minimize photobleaching). Images were acquired using TIRF microscopy with Plan Apochromatic TIRF 100×, 1.49 NA objective lens and camera (iXon3; Andor Technology) at 5 Hz frame rate. The spatial and temporal positions of the clusters were analyzed using Imaris 7.3 (Bitplane). Background subtraction and fluorescent speckle selection were performed with Imaris built-in algorithms. Image stacks were recorded for each protein (n > 5), and >1,000 speckles were tracked in each image stack. For the cluster photobleach experiment (Fig. 8 f), the cells were fixed with 3% paraformaldehyde on a microscope stage 1 min after photoconversion. Time series images were acquired under same condition as FSM but with 60% 561-nm laser intensity to photobleach αABD clusters.

For recombinant GST fusion protein production, the indicated α-catenin DNA fragments were subcloned into the bacterial expression vector pGEX-4T-1, which places GST in front of the αABD. The resulting plasmids were transformed into BL21(DE3) cells. Protein purification was performed using GST SpinTrap columns with no deviation from the manufacturer’s protocol (GE Healthcare). The cosedimentation assay and determination of binding affinities were performed according to the published protocol (Pappas and Rimm, 2006). In brief, the prepolymerized actin filaments (rabbit skeletal muscle actin was purchased from the Cytoskeleton, Inc.) in F buffer (2 mM Tris, pH 8.0, 50 mM KCl, 2 mM MgCl₂, 1 mM ATP, 1 mM EGTA, and 1 mM DTT) were incubated with precleared (100,000 g for 30 min) recombinant proteins for 30 min at room temperature. The preclearing step completely removed nonspecific sedimentation of the recombinant proteins in the absence of actin. The reaction was centrifuged at 100,000 g for 30 min. Equivalent volumes of pellet and supernatant fractions were analyzed by SDS-PAGE and densitometry. To assess binding affinities, three independent experiments each of which included αABD and three its mutants were performed, one of which is shown in Fig. 1 e. Although binding curves for the mutants varied between the experiments, the K842A mutant always exhibited slightly better binding than other mutants.

To test αABD binding with specific actin filament structures (Fig. 7 g), subconfluent cultures of A431 cells were first incubated for 3 min in permeabilization buffer (140 mM KCl, 10 mM Hepes, pH 7.0, 3 mM EGTA, 4 mM MgCl₂, 1% BSA, and 0.05% saponin) and then for 5 min in the same buffer with 1.2 µg/ml of the recombinant polyhistidine/mCherry-tagged αABD. Cells then were briefly washed in the same buffer, fixed, and processed for anti-actin staining.

Fig. S1 shows multiple structure-based sequence alignment of orthologous actin binding domains of α-catenin and vinculin. Fig. S2 shows junctions made of the EcΔ-Dn-α(280–906)-ΔVin chimera in A431D cells are devoid of the actin-binding proteins ZO1 and EPLIN. Fig. S3 shows AJs of α-catenin–deficient 468 cells and the same cells expressing the Dn-αCat-ΔVin mutant. Fig. S4 shows Western blotting assays using the anti–α-catenin antibody (EP1993Y) to probe expression levels of Dn-αCat and its mutants. Fig. S5 shows dynamic properties of clusters formed by the full-length α-catenin and myristoylated E-cadherin tail. Video 1 shows FSM of Dn-αABD clusters in A431 cells. Video 2 shows FSM of Dn-αABD in Latrunculin A–treated A431 cells. Video 3 shows FSM of Dn-αABD-I792A mutant in A431 cells.

We are grateful to Dr. I. Hofmann (German Cancer Research Center), Dr. J.K. Wahl (The University of Nebraska Medical Center, Lincoln, NE), and Drs. C. Gottardi and A. Yemelyanov (Feinberg School of Medicine, Northwestern University, Chicago, IL) for providing the anti-cingulin antibody, A431D cells, and α-catenin shRNA lentivirus, correspondingly. Sequencing, flow cytometry, and SMLM were performed at the Northwestern University Genetic, Flow Cytometry, and Advanced Microscopy Centers.

The work was supported by grants from the National Institutes of Health, AR44016 and AR057992 (to S.M. Troyanovsky) and GM062270 (to L. Shapiro), and a grant from a National Science Foundation, MCB-1412472 (to B. Honig).

The authors declare no competing financial interests.