Dimerization and Actin-bundling Properties of Villin and Its Role in the Assembly of Epithelial Cell Brush Borders*

Villin is a major actin-bundling protein in the brush border of epithelial cells. In this study we demonstrate for the first time that villin can bundle actin filaments using a single F-actin binding site, because it has the ability to self-associate. Using fluorescence resonance energy transfer, we demonstrate villin self-association in living cells in microvilli and in growth factor-stimulated cells in membrane ruffles and lamellipodia. Using sucrose density gradient, size-exclusion chromatography, and matrix-assisted laser desorption ionization time-of-flight, the majority of villin was identified as a monomer or dimer. Villin dimers were also identified in Caco-2 cells, which endogenously express villin and Madin-Darby canine kidney cells that ectopically express villin. Using truncation mutants of villin, site-directed mutagenesis, and fluorescence resonance energy transfer, an amino-terminal dimerization site was identified that regulated villin self-association in parallel conformation as well as actin bundling by villin. This detailed analysis describes for the first time microvillus assembly by villin, redefines the actin-bundling function of villin, and provides a molecular mechanism for actin bundling by villin, which could have wider implications for other actin cross-linking proteins that share a villin-like headpiece domain. Our study also provides a molecular basis to separate the morphologically distinct actin-severing and actin-bundling properties of villin.

Villin is an epithelial cell-specific actin-binding protein that regulates cell migration, cell death, and epithelial-to-mesenchymal transition underscoring the significance of this protein to epithelial cell function (1)(2)(3)(4)(5)(6)(7). Villin is expressed in differentiated epithelial cells with a brush border such as intestinal villi, proximal renal tubules, oviduct, and seminiferous ducts (8 -10). Microvilli are unique apical cell surface structures of epithelial cells and are extensions of uniform diameter with a core of cross-linked actin filaments that extend from the electron-dense material at the tip into the cell cortex as rootlet. It may be noted that, although most of the microvillar structural proteins have been isolated, the identification of the in vivo functions of these proteins, the role they play in the assembly of the microvilli as well as the molecular mechanism of cytoskeletal-protein interactions, has been a challenge. The microvilli of renal and intestinal epithelial cells contain a compact core of ϳ20 highly ordered parallel microfilaments that are believed to increase the cell surface thus regulating the absorptive functions of these tissues. Villin is the major protein associated with actin filaments in the microvillus core. Simple mixtures of villin and F-actin form uniformly polarized bundles similar to those seen in the ultrastructural analysis of brush borders (11). Overexpression or microinjection of villin in fibroblasts results in the growth of microspikes on the dorsal surface of these cells containing bundled actin filaments to which villin is associated (12,13). Further, down-regulation of the endogenous villin mRNA using antisense RNA reversibly inhibits brush-border assembly (14). These studies suggest that villin facilitates formation of organized actin structures such as those seen in the microvilli. Changes in the microvillar structure have been noted at the ultrastructural level in villin knock-out mice. The actin bundles in the microvillar core are not as well organized or as tightly packed as those in the wild-type littermates (15). Likewise, the terminal web appears thicker and less well organized in the villin null mice (15). The absence of more significant changes in the brush-border morphology in the villin-null mice may be due to compensation by other brush-border actin-bundling proteins that share a villin-like headpiece domain such as advillin or other actin-bundling proteins such as espin or fimbrin. The genetic deletion of villin accompanied by significant changes in the ultrastructure of the microvilli has been noted in pediatric cholestasis (16). This study points to the physiological significance of villin in the structure and function of the microvilli in the bile canaliculi. Although the role of villin as an actin-bundling protein is well established, the molecular mechanisms regulating this function of villin are not well characterized like the regulation of the microvilli, which itself remains poorly understood.
Villin is a 92.5-kDa protein with two tandem homologous halves (segments S1-S3 and segments S4 -S6 that form the villin core) and a carboxyl-terminal headpiece (segment S7). Villin shares structural and functional homology with two conserved families of proteins. One family of proteins, which includes gelsolin and adseverin, shares the conserved domain(s) found in the villin core. The second family of proteins, which includes advillin, supervillin, Drosophila quail, and protovillin from Dictyostelium, shares a villin-like headpiece. Villin is unique among the actin-regulatory proteins in that it can sever, cap, nucleate, and bundle actin filaments. It has been speculated that villin must use two F-actin binding sites, one located in the core and the second in the headpiece domain, to bundle F-actin (17,18). However, while characterizing the ligand-binding functions of villin, we made the serendipitous observation that a villin mutant that lacks the F-actin binding site in the core could cross-link actin filaments in vitro. Similar observations have been noted in other studies (13,18,19), although nothing was reported about the relevance of these findings to the actin-bundling activity of villin. To resolve this paradox, we hypothesized that villin must self-associate to exhibit actin bundling using a single F-actin binding site. In this study we test this hypothesis and demonstrate for the first time the self-association of villin in vitro and in living cells. Further, we identify the dimerization site in villin and demonstrate that deletion of this site prevents actin bundling by villin. Based on these studies a model for actin cross-linking by villin dimers is proposed, providing new insights about villin function in cells.

EXPERIMENTAL PROCEDURES
Materials-Monoclonal antibodies for villin, gelsolin, and ezrin were obtained from BD Transduction Laboratories. HA 2 monoclonal antibody was purchased from Roche Applied Science. pmCerulean-C1 was a kind gift from Dr. David Piston (Vanderbilt University, Nashville, TN). DsRed2-N1 was purchased from Clontech. pBudCE4.1 was purchased from Invitrogen. GelCode Blue was from Pierce. An actin-binding kit was purchased from Cytoskeleton (Denver, CO).
Transfection of MDCK Tet-Off Cells with Wild-type and Mutant Villin Proteins-MDCK Tet-Off cells were stably transfected with HA-, SEYFP-, cerulean-, or DsRed2-tagged villin as described previously (23). Transfected cells were cultured in Dulbecco's modified Eagle's medium containing 100 g/ml G418 sulfate, 100 g/ml hygromycin B, and 10% fetal calf serum. To repress the expression of the villin gene in MDCK Tet-Off cells transfected with HA-tagged villin, cells were cultured in the presence of 10 ng/ml doxycycline.
Cell Motility Assay-Cell migration was measured as described previously with slight modification (4). Confluent monolayers were scraped with a sharp blade across the diameter of the well to produce wounds of ϳ1 mm. Both villin-null as well as villin-expressing cells were treated with EGF (10 ng/ml). Data are expressed as distance moved (in microns) from the original wound margin 8 h post-wounding. Comparisons between mean values were made using one-way repeated-mea- sures analysis of variance and Tukey's modified t test (Bonferroni criteria) with a p Ͻ 0.05 considered significant. Chemical Cross-linking of Villin Protein-In vitro cross-linking of purified recombinant human villin protein (20 nM) with chemical cross-linkers was performed at different concentrations of cross-linker (0-to 100-fold molar excess) in phosphatebuffered saline at room temperature for different time intervals (0 -60 min). Reactions involving DTNB were performed in Tris-Cl buffer at pH 8. Cross-linking reactions were stopped by the addition of 20 mM lysine or cysteine (for DTNB), pH 7.4. Proteins were cross-linked in Caco-2 and MDCK Tet-Off cells by incubating the cells at room temperature in various concentrations (0 -2.5 mM) of cross-linkers for different time intervals (0 -60 min).
Sucrose Density Gradient Sedimentation-A 10-ml linear 10 -30% (w/v) sucrose gradient in phosphate-buffered saline was prepared by carefully layering 2 ml each of 10, 15, 20, 25, and 30% sucrose solutions prepared in phosphate-buffered saline, pH 7.2. The gradient was incubated at room temperature for 2 h to allow the uniform diffusion and the formation of the continuous gradient. The villin samples (200 l of 20 M villin cross-linked with DFDNB, 50-fold molar excess for 30 min at room temperature), and calibration proteins were run in parallel. Samples were loaded onto the gradient and centrifuged at 200,000 ϫ g for 16 h in an SW-41-Ti rotor (Beckman Instruments) at 4°C. 500-l samples were collected and analyzed by Western analysis. Protein standards run simultaneously were detected by using GelCode Blue stain.
SEC-HPLC-Cross-linked recombinant villin proteins were separated by size-exclusion chromatography-high performance liquid chromatography (SEC-HPLC) on a Waters 600 HPLC system equipped with a 996-photodiode array detector. Separation was performed on a TSK-GEL G3000SWXL column (Tosoh Biosep, 7.8 mm ϫ 30 cm, 5 m) using 0.05 M ammonium acetate buffer (pH 7.3) as mobile phase at a flow rate of 0.5 ml/min. 70 l of the reaction mixture was injected per run. Runs were monitored by absorbance at 280 nm, fractions were collected manually, and monomeric, dimeric, and oligomeric fractions were identified by MALDI-TOF. Protein standards were used for correlation of retention time and molecular mass.
MALDI-TOF Analysis of Cross-linked Villin Proteins-MALDI-TOF spectra were acquired on a Voyager DE Biospectrometry Workstation (PerSeptive Biosystems, Framingham, MA) in the linear mode using a nitrogen laser (337 nm), in the positive ion mode. All experiments utilized an acceleration voltage of 25 kV, a grid voltage of 93%, a guide wire of 0.2%, and a delay time of 225 ns. Each spectrum obtained was the sum of 100 laser shots. Crosslinked samples were prepared using sinapinic acid as the MALDI matrix, and each sample was applied using the sandwich method on stainless steel sample holders. The mass spectrometer was calibrated with bovine serum albumin. Spectra were recorded, and the data were processed with Data Explorer 3.5.0.0.
In Vitro Interaction of Recombinant Villin Proteins-GSTtagged recombinant villin as well as recombinant villin protein with the GST tag cleaved by thrombin digestion (1 unit/100 g of protein at room temperature, 16 h) were used to examine villin self-association in the absence of cross-linkers. The GST-tagged proteins were incubated with glutathione-Sepharose 4B beads (20 l of a 50% (v/v) slurry for 1 h at 4°C. After centrifugation the resin was washed, and the GST-fused recombinant villin protein immobilized on glutathione-Sepharose beads was incubated with untagged villin protein for 1 h at 4°C (pulldown assay). The beads were centrifuged, washed (in buffer containing 0.1% Triton X-100 and 150 mM NaCl), and the proteins bound to the beads were separated by 8% SDS-PAGE and visualized by using GelCode Blue staining.
FRET Analysis in Living Cells-FRET analysis was done in MDCK Tet-Off cells co-expressing SEYFP-and ceruleantagged or DsRed2-and cerulean-tagged villin proteins. To study the localization of the FRET signal in migrating cells, cells were serum-starved overnight and treated with EGF (100 ng/ml). FRET calculations were made using the method of sensitized FRET. FRET was calculated from at least 15 cells in 5 independent experiments. Cerulean and SEYFP were excited with the 448 nm or 514 nm line of an argon ion laser, respectively, and fluorescence was recorded with a 475 nm or a 530 nm long pass filter, respectively. DsRed2 and cerulean were excited with the 543 nm or 448 nm lines of a helium neon and argon ion laser, respectively, and fluorescence was recorded with a 560 nm and 475 nm long pass filter, respectively. Calculation of corrected FRET (FRET c ) was carried out on a pixel-bypixel basis for the entire image. The bleedthrough of cerulean and SEYFP through the FRET filter channel was corrected by applying the equation: FRET c ϭ raw FRET Ϫ (A* acceptor) Ϫ (B* donor), where A* and B* ϭ coefficient A and coefficient B. Similar measurements were made for DsRed2-and ceruleanco-expressing cell lines, where cerulean was the donor and DsRed2 the acceptor fluorophore. The Förster distance, R o , which is the donor-acceptor distance at which FRET efficiency is 50%, for DsRed2/cerulean is 50.9 Å and for cerulean/SEYFP is 49.2 Å (24, 25). The fluorescence was analyzed by confocal laser scanning microscopy (LSM 5 Pascal, Carl Zeiss, Thornwood, NY). The overall intensity of FRET was calculated using Metamorph Software. Images are displayed in pseudocolor mode, where white and black areas display high and low values of FRET in the range of 0 -200 relative light units, respectively.
Actin-bundling Activity of Villin-Actin bundles were examined by electron microscopy as described before (26). VIL/HP was incubated without or with DTNB (equimolar concentration) for 1 h at room temperature, and the chemical cross-linking reaction was terminated by incubation with excess cysteine. A low speed actin bundling assay was performed by incubating F-actin with recombinant full-length villin (VIL/WT), mutant villin proteins, uncross-linked or cross-linked villin headpiece (VIL/HP), or GST protein alone in bundling buffer (25 mM HEPES buffer, pH 7.4, 100 mM KCl, 2 mM EGTA for 60 min). The samples were centrifuged at low speed (10,000 ϫ g for 15 min), the supernatant and pellet fractions were separated by SDS-PAGE, and distribution of F-actin in these two fractions was analyzed by using GelCode Blue staining.

Villin Can Bundle Actin Filament Using a Single F-actin
Binding Site-Previous in vitro studies using truncation mutants of villin or proteolytic fragments of the villin protein have reported that neither the core nor the headpiece alone can bundle actin filaments in vitro, leading to the hypothesis that both actin binding sites in villin are required for this function (13,17,18). In this study, we determined that villin does not require both F-actin binding sites to cross-link actin filaments in vitro. We examined the effects of villin deletion mutant lacking the F-actin binding site in the core (⌬PB2) on actin filament morphology. F-actin in the absence of villin appear as single, long filaments (Fig. 1A, panel a), while in the presence of saturating amounts of wild type villin show long, straight, closely aligned, well-organized, tight bundles (Fig. 1A, panel b). Deletion of the F-actin binding site in the villin core (⌬PB2) had no effect on the actin-bundling activity of villin (Fig. 1A, panel c). The low speed actin bundling assay demonstrates the ability of villin mutant ⌬PB2 to bundle actin similar to full-length villin protein (Fig.  1B). In the absence of villin, the actin filaments were not bundled and sedimented in the supernatant fraction (control). We have previously shown that deletion of the F-actin binding site in the headpiece eliminates actin cross-linking by villin (21). Taken together, these data demonstrate that the F-actin binding site in the villin headpiece is obligatory, however the F-actin binding site in villin core is not necessary for the actin-bundling function of villin in vitro.
Formation of Villin Dimers-We hypothesized that the actinbundling function of villin in the absence of the second F-actin binding site could be determined by villin self-association. To evaluate villin self-association in vitro, recombinant villin protein (20 nM) was cross-linked in vitro with chemical cross-linkers of varying spacer length (supplemental Table S1). We elected to use multiple cross-linkers, including the zero-length cross-linking reagent, DTNB. Disulfide bonds can be reversibly formed by oxidation using DTNB (Ellman's reagent) and has been used to examine protein-protein interactions, including oligomerization of cofilin (27,28). We were able to demonstrate villin oligomerization with all cross-linkers under non-reducing conditions ( Fig. 2A). Under reducing conditions villin oligomers were observed with non-cleavable cross-linkers, EGS and DSS (supplemental Fig. S1A). Because DFDNB can also react with imidazolyl and phenolate groups of amino acid side chains, we could detect villin oligomers cross-linked with DFDNB in the presence of ␤-mercaptoethanol (29). Cross-linkers of varying spacer lengths, including DTNB, resulted in the oligomerization of villin, suggesting that in villin dimers two thiol groups must come as close as 2 Å (the length of a disulfide bond). To determine the ability of purified villin proteins to self-associate in the absence of cross-linkers we examined the association of GST-tagged and untagged villin (GST tag cleaved by thrombin digestion) proteins in vitro. For these studies GST-tagged fulllength recombinant villin immobilized on glutathione-Sepharose beads was incubated with untagged villin protein for 1 h at 4°C. The beads were washed, and the proteins bound to the beads were separated by 8% SDS-PAGE and visualized by using GelCode Blue staining. As shown in Fig.  2B, GST-tagged villin protein associates with untagged villin protein even in the absence of cross-linkers. Bovine serum albumin does not associate with GST-tagged villin protein.
Untagged villin protein does not associate with glutathione Sepharose beads, GST protein, or bovine serum albumin (data not shown). Together these data provide convincing evidence of inter-molecular interaction between villin molecules and of the close proximity of villin proteins in this complex.
To characterize the oligomeric species of villin, we separated cross-linked villin proteins using sucrose gradient centrifugation. Recombinant villin protein (20 M) was cross-linked with DFDNB (50-fold molar excess) and separated on a 10 -30% continuous sucrose density gradient. Protein standards were fractionated in parallel and used to estimate the molecular mass of villin oligomers. The molecular mass standards like the monomeric and oligomeric villin fractions were spread over multiple fractions (data not shown). Based on the sedimentation properties of molecular weight standards we estimated that cross-linked villin protein consisted of monomers, dimers, trimers, and possibly tetramers. The mobility of the villin dimers on 8% SDS-PAGE was in excess of their predicted masses (Fig. 2C). Overestimation of molecular masses of proteins following chemical cross-linking on SDS-PAGE has been documented in other studies (30 -33). However, we determined that, if cross-linked villin samples were separated by 5% SDS-PAGE and if the 150-kDa molecular mass marker was run close to the gel front, the majority of the oligomeric villin fraction separated as a dimer with the expected molecular mass (supplemental Fig. S1B). However, to examine both the monomeric as well as dimeric villin fraction, 8% gels were used throughout this study. From these studies we inferred that the majority of the villin protein exists as a monomer or dimer.
To validate these observations, we performed a combination of SEC and mass spectrometry. Cross-linked recombinant villin protein was separated by SEC-HPLC. Based on the retention time of protein standards, it was estimated that the majority of the villin protein was recovered as monomer and dimer and a  very small fraction consisting of oligomers (Fig. 2D). Each of these fractions was further analyzed by MALDI-TOF. The mass/charge ratios (m/z) of each of these fractions allowed us to confirm the presence of mostly villin monomers and dimers with very minor amounts of trimers and tetramers (Fig. 2E).
Villin Self-associates in Actin Bundle-rich Structures in Living Cells-Oligomerization of villin in epithelial cells was examined in the human colon adenocarcinoma cell line, Caco-2, which expresses villin endogenously, as well as MDCK Tet-Off cells transfected with human villin cDNA. Under non-reducing conditions, MDCK cell lysates from chemically cross-linked cells showed the presence of both monomers and oligomers of villin (Fig. 3A). Oligomers of villin were also seen in Caco-2 cells (Fig. 3B). The cross-linked villin protein from Caco-2 cell extracts separated as a dimer when examined by 5% SDS-PAGE (supplemental Fig. S4). Thus, both cross-linked recombinant villin as well as villin from cell extracts consists primarily of villin monomers and dimers. Because the majority of the selfassociating villin protein exists in the dimeric form, self-associating villin fractions will be referred to as dimers henceforth. Western analysis of cross-linked Caco-2 cell extracts with actin monoclonal antibody demonstrated that the villin dimers did not include any bound actin (data not shown).
We also examined the ability of the homologous protein, gelsolin, to self-associate under similar experimental conditions. Caco-2 cells treated with different cross-linkers did not show the presence of gelsolin oligomers (supplemental Fig.  S2A). These studies suggest that, even though villin and gelsolin share significant structural and functional homology, the two proteins differ in their ability to self-associate. In contrast, ezrin, a protein that is known to self-associate, demonstrated the presence of oligomers under similar conditions (supplemental Fig. S2B) (33). Together these studies demonstrate the ability of villin to self-associate and further suggest that this property of villin may not be shared by all proteins of its family.
To evaluate villin self-association in living cells, a FRETbased analysis using fusion proteins of villin that carried the cerulean or SEYFP tags at the amino terminus of villin protein was established. To verify that cerulean-villin and SEYFP-villin were fully functional, we examined their expression (supplemental Fig. S3A), growth factor-dependent distribution to the developing lamellipodia (supplemental video 1) in MDCK Tet-Off cells treated with EGF (100 ng/ml) as well as in an in vitro wound assay in which villin-induced increase in cell migration by these proteins was compared with the HA-tagged villin protein (supplemental Fig. S3B). The data shown in Fig. 3C (panels a-c) demonstrate no FRET signal in control cells co-expressing only cerulean and SEYFP proteins. In contrast, a strong FRET signal was observed in cells co-expressing SEYFP-villin and cerulean-villin, on the dorsal cell surface in structures that have been characterized previously as actin bundle-rich microvilli (12,13) (Fig. 3C, panels aЈ-cЈ). These data suggest, first, that villin self-associates in living cells and, second, that self-association of villin may be related to the actin-bundling function of villin.
When MDCK cells co-expressing SEYFP-villin and ceruleanvillin were treated with EGF (100 ng/ml), zones of short lifetimes could be detected at the cell periphery ( Fig. 3D and supplemental video 2). The FRET activity was distributed to membrane ruffles and lamellipodia, which was accompanied by an increase in the size of the cell surface protrusion and forward movement of the cell (Fig. 3D and supplemental video 2). To quantify these observations, we compared the average pixel intensity of FRET throughout the cell (30%, p Ͻ 0.01, n ϭ 15). The formation of villin oligomers was both spatially and temporally regulated in the motile cell. The first cell (labeled with the number 1, Fig. 3D), which was already in the process of developing the lamellipodia, shows the presence of villin in these developing cell surface structures. The second cell (labeled with the number 2, Fig. 3D) initially shows villin oligomers in microspikes and filopodia-like structures, which eventually redistribute into the developing lamellipodia. These data demonstrate that villin-villin complexes are localized to cell surface structures undergoing rapid remodeling as the cell migrates and further that villin-villin complexes are associated with cell surface structures that contain crosslinked actin filaments.
Identification of the Villin Dimerization Site-To evaluate the dimerization status of villin in living cells, a FRET-based analysis using fusion proteins of villin that carried the DsRed2 tag at the carboxyl terminus and the cerulean or SEYFP tags at the amino terminus of villin protein was established (Fig. 4A). The Förster distance, R o , which is the donor-acceptor distance at which FRET is 50% efficient for DsRed/CFP, is 50.9 Å, which is comparable to that reported for CFP/YFP, which is 49.2 Å (24,25). The cerulean-villin, SEYFP-villin, and DsRed2-villin were functional and behaved similarly to HA-tagged villin proteins (supplemental Fig. S3 and video 1). These data also demonstrated that the majority of DsRed2-tagged villin does not aggregate nonspecifically. Fig. 4A shows expression of the three proteins in MDCK cells. MDCK Tet-Off cells expressing cerulean-villin (amino-terminal tag) and DsRed2-villin (carboxylterminal tag) were examined for villin-villin interactions, should they form anti-parallel dimers. Using these two proteins and the sensitized emission method no signal in the FRET channel could be observed (Fig. 4B). We suggest that these data are consistent with the idea that villin dimers are formed in a parallel rather than an anti-parallel conformation. To confirm this observation, MDCK Tet-Off cells co-expressing SEYFPvillin and cerulean-villin (both amino-terminal tags) were used to measure FRET activity. The data shown in Fig. 4C demonstrate a strong FRET signal, suggesting that villin dimers are preferentially formed in a parallel fashion. A stable, specific interaction between villin molecules was indicated by the high FRET efficiency (30%, p Ͻ 0.01, n ϭ 5).
To identify the villin dimerization site, we used amino-and carboxyl-terminal truncation mutants of villin that were crosslinked in vitro (with EGS and DFDNB). The amino-terminal truncation mutant, CT-1 (aa 373-827) failed to self-associate in the presence of DFDNB or EGS (Fig. 4D). In contrast, the carboxyl-terminal deletion mutant of villin, VT-3 (aa 1-261) formed villin dimers like wild-type villin (Fig. 4E). Interestingly, VT-3 formed stable dimers even in the absence (control) of chemical cross-linkers. We speculate that the dimer site may be more accessible in VT-3 compared with recombinant fulllength villin protein. These data also support villin self-association in a parallel conformation and suggest that the villin dimerization site lay in the amino-terminal domain. To further characterize the villin self-associating residues, we elected to use several shorter truncation mutants of the amino-terminal recombinant villin protein. As shown in Fig. 5A a villin truncation mutant expressing only residues 1-136 formed stable villin dimers in the absence or presence of chemical cross-linkers. We further characterized this protein in the absence of chemical cross-linkers using sucrose density gradients. Using this approach, the majority of the villin truncation protein 1-136, like full-length villin, was found to migrate as a dimer (Fig. 5B). Based on this observation, we concluded that the villin dimerization site lay in the amino-terminal first 136 amino acids. To confirm the dimerization site, we used several smaller deletion mutants made in the region aa 1-136 of full-length villin protein. Two of these proteins, VIL/⌬21-67 and VIL/⌬112-119 resulted in significant loss of villin dimers, although neither deletion individually completely abolished villin self-association (Fig. 5C). However, mutation of both sites together (VIL/ ⌬21-67/112-119) completely abolished villin self-association in vitro (Fig. 5C). Based on these data we concluded that the villin dimerization site lies within aa 21-67 and aa 112-119 in human villin protein. These data were confirmed in living cells. FRET analysis using villin mutants that lack these dimerization sites confirmed that the amino-terminal residues 21-67 and 112-119 are involved in villin self-association in living cells (Fig. 5D). These data also demonstrate that villin self-association in living cells is not due to local enrichment of the protein with cell surface structures. The villin mutant VIL/⌬21-67/ 112-119 is distributed like VIL/WT in MDCK cells, however it fails to give a positive FRET signal because it lacks the dimerization site.
Villin Dimerization Is Required for Actin Bundling by Villin-To determine if villin self-association is required for the actin-bundling function of villin, actin bundling was measured by low speed centrifugation and sedimentation of bundled actin filaments as well as by electron microscopy. As shown in Fig. 6A, full-length recombinant villin protein (VIL/WT) bundled actin, and the majority of the actin filaments appeared in the pellet (P) fraction. Consistent with our previous results, deletion of the F-actin binding site in the villin headpiece (⌬PB5) resulted in complete loss of actin bundling, and the majority of the actin filaments now appeared in the supernatant fraction (S) (21). Villin truncation mutant, VT-3 (aa 1-261), which also lacks the F-actin binding site in the headpiece, likewise failed to bundle actin filaments. Villin truncation mutant CT-1 (aa 373-827), which lacks the dimerization site, did not bundle actin filaments. Villin mutants VIL/⌬21-67 and VIL/ ⌬112-119 showed some decrease in actin bundling, however villin mutant lacking both the dimerization sites but containing both the F-actin binding sites (VIL/⌬21-67/ 112-119) showed complete loss of actin bundling by villin (Fig. 6B). These observations are summarized in Fig. 6C and demonstrate that actin bundling by villin requires self-association of villin. To further confirm the effects of villin mutant proteins on the actin cross-linking activity of villin, F-actin filaments were incubated with wild-type and mutant villin proteins and also examined by electron microscopy. Wild-type villin bundled F-actin as expected (Fig. 6D, panel a). F-actin filaments incubated with the dimerization site mutant (VIL/⌬21-67/112-119) failed to bundle actin filaments despite the presence of both the F-actin binding sites (Fig. 6D, panel b). It may be noted that this loss of function is not due to misfolding of the mutant protein, because VIL/⌬21-67/112-119 retains the biological activity associated with wild-type villin such as actin nucleation (data not shown). These data confirm our observation that dimerization together with the F-actin binding site in the headpiece are required for effective actin bundling by villin in vitro.
Next we elected to determine if the actin binding site in villin headpiece is sufficient to bundle actin filaments. For these studies, we used a recombinant villin protein expressing the villin headpiece domain only (VIL/HP). The headpiece domain contains no cysteine residues and to avoid cross-linking lysine residues, some of which could potentially be involved in actin binding, the GST-tagged headpiece protein was induced to form dimers because of the GST moiety using a cysteine cross-linker (DTNB). Unreacted DTNB was removed by the addition of excess cysteine. GST-tagged headpiece protein when cross-linked formed dimers while headpiece alone without the GST tag did not (data not shown). Uncross-linked and cross-linked headpiece proteins were used to measure actin bundling by lowspeed centrifugation and sedimentation of bundled actin filaments. Full-length villin protein (VIL/WT) as well as chemically cross-linked (ϩDTNB) recombinant headpiece protein (VIL/HP) bundled F-actin (Fig. 6E). In contrast, untagged and uncross-linked headpiece protein (ϪDTNB) as well as the GST protein alone did not bundle actin filaments.
Electron micrographs showed that actin bundles formed with the cross-linked headpiece of villin were indistinguishable from actin filaments bundled by fulllength villin protein (Fig. 6F). These data lend further support to our hypothesis that the F-actin binding site in villin headpiece is sufficient to cross-link actin filaments provided villin self-associates.
To examine the effects of villin dimerization on actin bundling in living cells, porcine kidney proximal tubule LLC-PK1 cells were transfected with wild-type and mutant villin proteins. As reported earlier, overexpression of full-length human villin in this cell line increased the length and number of microvilli on the dorsal cell surface compared with cells transfected with EGFP-actin (Fig. 7) (34). Overexpression of villin mutants lacking either the F-actin binding site (⌬PB5) in the headpiece or the dimerization site (VIL/⌬21-67/112-119) resulted in the reduction or loss of microvilli from LLC-PK1 cells. In contrast there was no significant change in the length or density of the microvilli in cells expressing villin mutant ⌬PB2, which lacks the F-actin binding site in the core. These data then confirm our observations that villin bundles actin filaments by self-associating and engaging the single F-actin binding site in the headpiece both in vitro as well as in living cells.

DISCUSSION
The microvilli of intestinal epithelial cells are dynamic structures that are shortened by starvation (35), by treatment with cycloheximide (36) or by lectins (37), and can reform once the stimuli have been removed. Similar restructuring of the microvilli has been reported for proximal tubule brush border in response to anoxia and ischemia (38,39). Actin-bundling proteins can influence the initiation, organization, and/or stabilization of actin bundles; however, how these proteins regulate these functions is poorly understood. Villin is the major actin-bundling protein that accumulates predominantly in the microvilli of mammalian intestine and kidney epithelial cells (40) and together with fimbrin is responsible for the initiation, organization, and formation of the rigid structure in the microvilli core (41,42). The two actin-bundling proteins localize to actin bundles sequentially suggesting that two bundling proteins may have non-redundant functions. Studies of developing enterocytes in chicken and mice indicate that villin is the first major actin-bundling protein to concentrate at the apical surface of enterocytes (8,43). Consistent with this observation, undifferentiated intestinal epithelial cells as well as intestinal cell lines such as HT-29 express low levels of villin and do not form brush borders (44). Increase in villin expression in HT-29 cells for instance correlates with differentiation and microvillus assembly (45). This is similar to previous reports that have demonstrated that overexpression of villin at high levels is required in fibroblasts and other non-epithelial cells to form cell surface microvilli and microspikes (19). Expression of human villin in fibroblasts and other cells that do not normally produce villin results in the growth of surface microvilli as well as reorganization of the microfilament network, including loss of stress fibers (4,13,46). These and similar studies done in other epithelial cell lines such as Caco-2 and LLC-PK1 suggest that villin levels contribute to actin reorganization and microvillus density and length regulation (34,47). Similar observations have been made with other actin cross-linking proteins such as espin, where the length of the microvilli has been correlated positively with espin protein expression levels (34). We suggest that a better understanding of the actin-bundling function of villin would allow us to extend our understanding to the microvillus structure and assembly, to understand how the cell cortex is shaped, and to understand how the enterocyte brush-border membrane is remodeled after damage (48).
Villin contains two F-actin binding sites, one in the core between domains S1 and S2 and a second F-actin binding site in the headpiece (49,50). The F-actin binding site in the core regulates the actin-severing function of villin (51). In vitro, neither the headpiece nor the villin core alone can bundle actin filaments, which has led to the existing model in which both the F-actin binding sites are believed to be required to cross-link actin (Fig. 8A) (17). However, based on several previous studies, it is not entirely clear whether the F-actin binding site in villin core is required for filament cross-linking by villin. Villin mutants lacking the F-actin binding site in the core did not prevent actin bundling by villin in vitro or in cells transfected with these mutants (13,18,19). In this report, we demonstrate that the actin binding site in the villin core is not required for actin bundling by villin in vitro or in living cells. Similar to  earlier reports, we also demonstrate no actin-cross-linking by the purified headpiece protein alone (46,52). However, we can demonstrate that the headpiece domain of villin is sufficient to cross-link actin filaments provided villin dimers are formed. In this study we demonstrate for the first time that villin forms dimers both in vitro as well as in living cells. We speculate that chemical cross-linking reactions generate villin dimers, because these exist in dynamic equilibrium with the monomer, although it is possible that these exist at a low concentration and/or are not stable enough to be purified by standard protein purification methods (53). Because villin self-associates in living cells in the absence of chemical cross-linkers, it supports this hypothesis. Structural mapping by site-specific mutagenesis allowed us to identify a region in the amino-terminal domain of villin that is involved in dimer formation. Loss of this domain, despite the fact that both F-actin binding sites are present, results not only in the loss of villin self-association but also loss of actin bundling by villin both in vitro as well as in living cells. The presence of villin dimers in microvilli supports this model, where villin self-association regulates the actin-bundling function of villin. Further, deletion of the dimerization site in villin results in a significant loss of microvillus length and density in LLC-PK1 cells. Thus, disruption of villin dimerization interferes with the actin-bundling function of villin and microvillus assembly and structure in epithelial cells. In light of these findings, we suggest that actin bundling by villin does not require both the F-actin binding sites and can be duplicated by a single F-actin binding site in the headpiece because of villin self-association (Fig. 8B). Our FRET studies as well as our in vitro crosslinking studies with villin truncation mutants demonstrate parallel arrangement of villin dimers, which would facilitate unidirectional F-actin bundle formation, similar to the uniformly polarized bundles observed in intestinal and renal microvilli. Taken together, our detailed functional analysis reveals that dimerization activity of villin and the actin crosslinking function of villin are inseparable events, and both the dimerization site in the amino-terminal domain as well as the F-actin binding site in the headpiece are indispensable for actin bundling by villin.
Interestingly, previous studies done with a villin mutant, ⌬NT, which lacks aa 17-133, failed to induce the formation of cell surface microvilli and F-actin-rich structures (19). Although the authors suggested that this loss may be related to the loss of actin nucleating function of villin, it may be noted that actin nucleation by villin is regulated by both amino-and carboxyl-terminal residues, and some of this function should be maintained in this mutant. Further, the observation that this mutant retained its ability to reorganize the cortical cytoskeleton and resulted in disruption of stress fibers also strongly supports the notion that actin nucleation or capping functions may not be disrupted in this mutant (19). To this effect, later studies by Friederich et al. support the idea that F-actin bundling rather than actin nucleation is required for the formation of microvilli and microspikes by overexpressed villin protein in fibroblasts (18,54). Based on these two studies, we speculate that ⌬NT failed to bundle actin filaments, because it lacked the dimerization site. Villin mutant VIL/⌬21-67/112-119 retains the amino-terminal actin monomer binding site (based on crystal structure of gelsolin-actin this would include residues: Ile-80, Gln-84, Asp-86, Asp-87, Gln-95, and Arg-97 in human villin), nucleates actin like wild-type villin protein (data not shown) and yet failed to bundle actin. These data provide additional support for the role of self-association and bundling rather than nucleation by villin in microvillus assembly.
Villin is both an actin-bundling as well as actin-severing protein. It has been suggested that, at high calcium concentration, villin severs actin and at physiological calcium concentrations it bundles actin filaments, thus explaining the two morphologically distinct functions of this protein. Villin exists in an autoinhibited conformation, which is released either by high calcium (200 M) or tyrosine phosphorylation of villin, thus exposing the F-actin binding site in the villin core and regulating the actin severing activity of villin (55,56). It may be noted that actin cross-linking by villin (which is done in vitro in the absence of calcium and presence of 2 mM EGTA) is thus unlikely to be mediated by the F-actin binding site in the core. Further, based on these previous studies, it is difficult to interpret how villin could engage the F-actin binding site in the core and only bundle and not sever actin filaments. This apparent paradox can be explained by our data, which clearly demonstrate that the F-actin binding site in the core is not required for actin bundling by villin and suggest that this site may exclusively regulate the actin-severing function and, further, that the actin-bundling function is entirely regulated by the F-actin binding site in the headpiece. To further support this contention, it may be noted that we have previously demonstrated that tyrosine phosphorylation not only modifies the secondary structure of villin, allowing villin to sever actin at physiological calcium concentrations (55), but also inhibits the actin-bundling function of villin (26). These data suggest that, under conditions where the F-actin binding site in the core associates with actin filaments, the predominant effect is actin severing and not actin bundling by villin. Thus, our functional analysis of the actin bundling site in this study together with our previous studies also allows us to distinguish between the actin-severing and actin-bundling functions of villin, providing a more physiologically relevant model for these functions of villin in cells.
Actin bundles are necessary to push the cell membrane efficiently during cell locomotion. Our FRET data show that the ratio of villin dimers to villin protein in EGF-stimulated cells was highest close to the plasma membrane. Dimerization of villin may allow for actin bundle formation at the leading edge thus providing the mechanical force necessary to generate protrusion. Alternatively, it is conceivable that distribution of villin dimers at the cell surface could result in actin filaments that are uncapped, bundled, and attached to the plasma membrane resulting in F-actin-barbed ends that are locally focused and protected from capping within a few microns of the lamellipodia, thus regulating cell migration. Villin forms a key link between signaling molecules and the actin cytoskeletal dynamics by regulating spatially confined actin filament assembly (2,4,23,57). Villin self-association could also allow spatial clustering of ligands at the cell surface in the vicinity of fast growing filaments by stabilizing the interaction of villin with these binding partners.
Our data also suggest that the ability of villin to self-associate may not be shared by all proteins of the villin family. We could not demonstrate self-association of gelsolin in Caco-2 cells, however under comparable experimental conditions we could show dimers of an unrelated protein, ezrin, which is known to self-associate. None of the proteins of the villin family, which share structural homology only with the villin core and lack a headpiece, are known to self-associate. However, some proteins that contain a villin-like headpiece domain such as dematin are known to oligomerize (58). It may be noted that a number of proteins have now been cloned that share a villin-like carboxylterminal headpiece and include supervillin, p92 (advillin), dematin, abLIM, a talin homologue, and a putative villin homologue identified in chromosome 3p22-p21.3 (59 -62). Our findings may also help understand the structural and functional differences between different actin-bundling proteins that are all associated with parallel actin bundle containing structures such as fascin, fimbrin, and espin as well as help identify if there is a difference between parallel actin bundles formed by these proteins in filopodia versus microvilli. Taken together, our study provides a new level of understanding in the functioning of villin and may help elucidate the functional properties of such structurally and functionally related proteins.