A Direct Interaction between Actin and Vimentin Filaments Mediated by the Tail Domain of Vimentin*

The assembly and organization of the three major eukaryotic cytoskeleton proteins, actin, microtubules, and intermediate filaments, are highly interdependent. Through evolution, cells have developed specialized multifunctional proteins that mediate the cross-linking of these cytoskeleton filament networks. Here we test the hypothesis that two of these filamentous proteins, F-actin and vimentin filament, can interact directly, i.e. in the absence of auxiliary proteins. Through quantitative rheological studies, we find that a mixture of vimentin/actin filament network features a significantly higher stiffness than that of networks containing only actin filaments or only vimentin filaments. Maximum inter-filament interaction occurs at a vimentin/actin molar ratio of 3 to 1. Mixed networks of actin and tailless vimentin filaments show low mechanical stiffness and much weaker inter-filament interactions. Together with the fact that cells featuring prominent vimentin and actin networks are much stiffer than their counterparts lacking an organized actin or vimentin network, these results suggest that actin and vimentin filaments can interact directly through the tail domain of vimentin and that these inter-filament interactions may contribute to the overall mechanical integrity of cells and mediate cytoskeletal cross-talk.

Growing evidence suggests that cells have evolved specialized multibinding proteins that mediate the cross-linking of two or all three cytoskeleton filament networks. These proteins include fimbrin (11), the motor proteins kinesin (12), dynactin and dynein (13,14), as well as members of the plakin family of large coiled-coil proteins (15,16), including plectin (17) and BPAG1 (18,19). Despite indirect evidence, it remains unclear whether major cytoskeleton filamentous proteins can interact directly, i.e. without auxiliary proteins.
IFs play a central role in helping cells resist mechanical stresses, maintaining the mechanical integrity of cells (20 -22). Vimentin (57 KDa) is a type III IF found predominantly in cells of mesenchymal origin and a number of epithelial cell lines (23)(24)(25)(26). Vimentin plays key roles in cell migration (27), proliferation (28), contractility (29,30), as well as the gene expression profile of the cell when cells are subjected to fluid shear stress (31). Immunofluorescence microscopy indicates that vimentin structures colocalize with microfilaments and actin-containing structures (10,32,33). It has been speculated that this apparent interaction may be mediated through the tail domain of vimentin (10), but no evidence supports a direct interaction between actin and vimentin filaments. This may be in part due to the fact that traditional biochemical assays as well as light and electron microscopy are ill-suited to functionally test filament-filament interactions.
The tail domain of vimentin alone cannot form filaments (10,34), but it interacts with its rod domain (35). The role of the IF tail domain in organizing vimentin and potentially F-actin networks is somewhat unclear. In vitro, the assembly and network formation of tailless (⌬T) vimentin filaments only appear compromised when there are mutations in the rod domain (9), whereas the ability of transfected vimentin to enter the nucleus is affected by its tail (36,37). Transfected tailless Xenopus vimentin (36,37) and human tailless keratin (38) have been shown to migrate into the nucleus of some cell lines, although this is not the case for human tailless vimentin, which does not migrate into the nucleus under similar conditions (36). These different cellular processes seem to rely on either direct or indirect interactions between vimentin and actin filament networks (10,33).
Here, using purified proteins, we test the hypothesis that vimentin and actin filaments interact directly with each other to produce structures that are stiffer than the two individual networks they are made of. Like actin, vimentin can be assem-bled into filamentous structures upon addition of a polymerizing salt (39). Exploiting this property, actin and vimentin were polymerized simultaneously in the same buffer to form a mixed filamentous network.
Using quantitative rheometry, we measured the elasticity and viscosity of the resulting mixed networks of actin and vimentin filaments. The mechanical response of vimentin filament networks in the presence and absence of F-actin was assessed as a function of the protein composition of the mixed network, as well as the frequency and the amplitude of applied shear deformations (40). Our results show that a network containing both vimentin and actin filaments at a molar ratio of 3 to 1 displays an elasticity that is significantly higher and less frequency-dependent than the elasticity of networks containing only F-actin or vimentin filaments. ⌬T vimentin, which assembles into filamentous structures of morphology similar to that formed by full-length (FL) vimentin, completely abrogates this synergistic effect. This suggests that the tail domain mediates a direct inter-filament interaction between vimentin IF and F-actin. Together with the fact that cells featuring prominent vimentin and F-actin networks are much stiffer than their counterparts lacking an organized F-actin network or vimentin network (41,42), our results in vitro suggest that the mechanical synergy between actin and vimentin filaments is mediated by the tail domain of vimentin and may contribute to the overall mechanical response of cells and cytoskeletal cross-talk.

MATERIALS AND METHODS
Protein Purification-The purifications of human vimentin, ⌬T vimentin, and the tail domain of vimentin are based on published protocols (10,37,39). Human vimentin, ⌬T vimentin, and vimentin tail cDNAs were generous gifts from R. D. Goldman (Northwestern University Medical School), H. Herrmann (Institute of Cell and Tumor Biology, German Cancer Research Center, Heidelberg), and M. W. Klymkowsky (University of Colorado, Boulder), respectively. Vimentin was dialyzed in a stepwise manner out of urea-containing buffer into 5 mM Tris-HCl (pH 8.4), 1 mM EDTA, 0.1 mM dithiothreitol. Actin was prepared from chicken breast (43) with an extra step of gel filtration by Sephacryl S-300 (Sigma) (44). Purified actin was stored as Ca 2ϩ -actin in continuous dialysis at 4°C against buffer G (0.2 mM ATP, 0.5 mM dithiothreitol, 0.2 mM CaCl 2 , 1 mM sodium azide, and 2 mM Tris-HCl, pH 8.0).
Rheology-The mechanical properties of networks of actin and vimentin filaments were measured using a strain-controlled, 50-mm diameter cone-and-plate rheometer (ARES-100, TA Instruments, New Castle, DE) (45,46). Shear deformations of controlled amplitude and frequency were applied by precise dynamic rotations of the lower plate of the rheometer. The upper cone of the rheometer is coupled to a computercontrolled motor, which applies either steady or oscillatory shear deformations of controlled frequency and amplitude. The cone is connected to a torque transducer, which measures the stress induced in the protein solutions by the applied shear deformations. The gelation of the filament networks was monitored by measuring the time-dependent elasticity. Elasticity was measured by applying two oscillatory deformations of 1% amplitude at a frequency of 1 rad/s every 120 s until a steady state was attained (ϳ1-3 h). At steady state, the frequency-dependent viscoelastic moduli of the filament networks, GЈ() and GЉ(), were computed by dividing the in-phase and out-ofphase components of the measured oscillatory stress induced within the network by the applied strain amplitude. The phase angle, ␦ ϭ tan Ϫ1 (GЉ/GЈ) was also calculated.
Next, oscillatory deformations of small amplitude (1%) and frequency between 0.01 and 100 rad/s were applied to measure simultaneously the frequency-dependent elastic modulus, GЈ(), and frequency-dependent viscous modulus, GЉ(), of the filament networks. Finally, deformations of fixed frequency (1 rad/s) and amplitude between 0.1 and 1000% were applied to measure the elastic and viscous moduli as a function of the amplitude of the input shear deformation, g 0 . The statistical significance between moduli of networks of different protein composition was verified by pairwise comparison using the Student's t test, which yielded values of p Ͻ 0.05.
Electron Microscopy-The ultrastructure of negatively stained vimentin and actin filaments was examined by electron microscopy. Polymerizing vimentin and actin solutions were incubated in assembly buffer (25 mM Tris buffer, pH 7.4) obtained by mixing 1 volume of polymerizing buffer to 9 volumes of storage buffer. 10 l of protein solution was placed on each collodion-coated electron microscopic grid. Grids were washed with assembly buffer and stained with 2% uranyl acetate solution (47). Electron microscopy was performed at the Integrated Imaging Center in Johns Hopkins University with a Philips 410 transmission electron microscope at magnifications between 65,000ϫ and 105,000ϫ, as indicated in the figure captions.

Actin and Vimentin Filaments
Interact Directly-We hypothesized that vimentin and actin filaments interact directly, i.e. without auxiliary proteins, forming structures that are stiffer than the two networks they are made of. Exploiting the unique assembly properties of vimentin among IFs, we co-polymerized vimentin and actin filaments in the same buffer, and monitored the viscoelastic properties of the mixed networks. Using a cone-and-plate rheometer, the elastic modulus, GЈ (which measures the propensity of the polymers to rebound after shear deformation), and the viscous modulus, GЉ (which measures how much the specimen can flow under stress), of the protein solutions were measured upon onset of filament assembly.
The relative concentration of vimentin and actin was varied for a total protein concentration of 24 M. The steady-state elastic modulus for the mixed filament networks reached a distinct maximum in stiffness of GЈ ϳ 40 dyn/cm 2 for [actin] ϭ 6 M and [vimentin] ϭ 18 M, compared with 6 dyn/cm 2 for a 24 M F-actin network (no vimentin present) and 15 dyn/cm 2 for a and B). Therefore, a mixed filament network containing onefourth actin and three-fourths vimentin was stiffer than the networks containing only F-actin or only vimentin filaments.
Vimentin filament networks in the absence of actin displayed a low phase angle, ␦ ϳ 5° (Fig. 1C), which corresponds to a solid-like rheological behavior, close to that displayed by a stiff rubber. The phase angle describes the delay between the imposed (input) deformation and the measured (output) mechanical stress induced by that deformation in the network. This delay is minimum (0°) for an elastic solid-like rubber and maximum (90°) for a viscous liquid-like glycerol. As the actinto-vimentin molar ratio increased, mixed vimentin/actin networks became significantly more liquid-like, reaching a phase angle of ϳ30°for an F-actin network containing no vimentin (Fig. 1C).
Through electron micrographs of negatively stained suspensions of either pure F-actin or pure vimentin filament network, we verified that the present buffer conditions did not affect their normal morphology. EM was also used to examine the structures formed by the copolymerization of actin and vimentin (Fig. 1D). Analysis of micrographs of individual filaments in mixed vimentin/actin filament networks resulted in a distribution of filament diameters with two narrow peaks confirming the presence of separate actin and vimentin filaments within the network. The mean Ϯ S.D. values of the distributions were 7.7 Ϯ 1.4 nm and 10.8 Ϯ 2.2 nm, which correspond to the known diameters of actin and vimentin filaments, respectively. Therefore, each filament in the mixed filament networks contained only one type of subunit.
Dynamic Response of Vimentin/Actin Filament Networks-For a fixed polymer length and polymer concentration, the mobility of filaments in a network depends mostly on their potential to interact through cross-linking interactions (48). To assess the extent of mobility of the filaments within entangled vimentin/actin filament networks, these networks were subjected to shear deformations of increasing frequency. The elastic modulus of F-actin networks (no vimentin present) increased steadily with increasing rates of shear, i.e. with increasing shear frequency ( Fig. 2A). This result indicates that actin filaments had progressively less time to relax the mechanical stress imposed by the rheometer for increasing rates of shear rates. However, by reducing actin concentration and increasing vimentin concentration in mixed networks, we observed a significantly reduced frequency dependence of the elastic modulus ( Fig. 2A). This indicates that filaments in mixed vimentin/actin filament networks became less mobile than in networks containing only F-actin. Therefore, these mixed filament networks were less capable of relaxing mechanical   (closed diamonds). B, concentration-dependent exponent a, which was extracted from power law fits, GЈ ϳ a , of profiles such as shown in A. The exponent a increased as the molar ratio of actin to vimentin increased, which suggests that the filamentous networks became more dynamic, i.e. there was more relative sliding among filaments, for increasing actin concentration in mixed vimentin/actin filament networks. OCTOBER 13, 2006 • VOLUME 281 • NUMBER 41 JOURNAL OF BIOLOGICAL CHEMISTRY 30395 stresses, presumably due to enhanced interactions between actin and vimentin filaments (Fig. 2A).

Interactions between Actin and Vimentin
These frequency-dependent elasticity profiles ( Fig. 2A) could be fit approximately to power laws of the type GЈ() ϳ a , with an exponent a that describes the steepness of the frequency dependence of GЈ(). The exponent a increased for increasing actin content in mixed networks, which suggests that there was more relative movements among filaments in actin-rich networks than in vimentin-rich networks (Fig. 2B).
The Tail Domain of Vimentin Mediates Interactions between Vimentin and Actin Filaments-We hypothesized that the tail domain of vimentin mediates the synergistic enhancement of elasticity observed in mixed vimentin/actin filament networks. Therefore we repeated our rheological assays for mixed suspensions of actin and ⌬T vimentin filaments. The steady-state elastic modulus of networks containing both actin and ⌬T vimentin filaments decreased as the molar ratio of actin to vimentin in the mixture increased. However, unlike their FL vimentin/actin filament network counterparts, the relative elastic modulus (GЈ/GЈ vim ϭ 24M ) of ⌬T vimentin/actin filament networks displayed no distinct maximum for an intermediate protein composition (Fig. 3A). Moreover, the higher phase angle value of the networks suggested that actin and ⌬T vimentin filaments formed more liquid-like networks compared with actin and FL vimentin filaments over the same concentration range (Fig. 3B).
We further tested whether the absence of mechanical synergy between ⌬T vimentin and actin filaments correlated with diminished inter-filament interactions by analyzing the frequency-dependent elasticity, GЈ(). Although the exponent a from power-law fits of GЈ() also increased with actin concentration, it was higher in the presence of ⌬T vimentin than in the presence of FL vimentin (Fig. 3C). This suggests that ⌬T vimentin allowed significantly more inter-filament sliding in mixed networks than their FL vimentin counterparts. EM studies did not show ultrastructural differences between FL vimentin/actin and ⌬T vimentin/actin filament networks (Fig. 3D). Hence the observed difference in the mechanical response of mixed filament protein networks is probably due to direct interactions between vimentin and actin filaments (Fig. 3).
Binding studies showed an interaction between F-actin and the tail domain of vimentin (supplemental Fig. S1). We evaluated the binding kinetics of vimentin tails to F-actin using a pelletting assay (49). Solutions of actin monomers of fixed concentration were mixed with various concentrations of vimentin tails before initiating actin filament assembly. The pellet recovered from a high speed centrifugation of the polymerized proteins was analyzed on SDS-gel electrophoresis, and the concentration of free (unbound) tails in the supernatant was measured. Analysis of the Coomassie-stained protein gels showed the presence of actin and vimentin tails in the pellet. A fit of the binding curve using nonlinear regression (supplemental  (34,37), hence we hypothesized that it would have negligible effects on the mechanical properties of vimentin networks. We compared the mechanical properties of FL and ⌬T vimentin filament networks. EM studies showed that both FL vimentin and ⌬T vimentin formed filaments of similar morphology (Fig. 4) (34). There was a negligible difference in elastic moduli of both FL and ⌬T vimentin filament networks (no actin present) over the tested concentration range (Fig. 4A). The elastic moduli of both networks increased weakly with protein concentration, C, scaling as GЈ(C) ϳ C 0.37 , and GЈ(C) ϳ C 0.35 (estimated at a frequency of 1 rad/s), for FL and ⌬T vimentin, respectively. Polymer theory suggests that this concentration dependence is consistent with cross-linked flexible filaments (50). The value of the phase angle also suggested that both networks harbored a strong solid-like character, with FL vimentin being slightly stiffer that ⌬T vimentin (Fig. 4B). As the networks gelled, we tested the mobility of FL and ⌬T vimentin filaments in networks, by exposing the networks to shear deformations of increasing frequency before a steady-state elasticity was reached (Fig. 5, A and B). The frequency-dependent gelation profiles were fit to power laws, GЈ() ϳ a , and an exponent a was extracted from the plots shown in Fig. 5 (A and B). The measured value of the exponent a indicated that inter-filament interactions were similar for both FL and ⌬T vimentin networks. Although a was slightly lower for FL vimentin (0.12 Ϯ 0.01) than for ⌬T vimentin (0.16 Ϯ 0.01) filament networks (mean Ϯ S.D.), they had a parallel downwards trend, which indicated that inter-filament interactions grew stronger with time. Together these results suggest that networks of ⌬T vimentin filaments have mechanical properties similar to those of networks of FL vimentin filaments.
By subjecting vimentin filament networks to increasing strain amplitudes, we found that the amplitude of the deformation at which vimentin networks began to yield and break down (point where the elastic modulus dropped by 10% of its original value) was independent of vimentin concentration (data not shown). Unlike keratin IF 5/14 network, whose mechanical resilience is significantly affected when keratin-14 tail domain is truncated (51), the truncation of the tail domain of vimentin did not affect the mechanical resilience of its network. More-over, FL and ⌬T vimentin filaments formed networks of similar ultrastructures, as assessed by EM and negative staining (Fig. 5C).

DISCUSSION
The three major cytoskeleton proteins (F-actin, microtubule, and IF) form networks that are mutually affected by the re-organization or disassembly of any one of them. Immunofluorescence microscopy shows that FL vimentin and the vimentin tail localize near actin networks in cells, suggesting an interaction between vimentin and actin filaments mediated by the tail domain of vimentin (10). However, this interaction could be indirect, i.e. mediated by auxiliary proteins.
We report the strongest evidence yet of a direct interaction between two of the three major cytoskeletal networks, vimentin and actin, which has remarkable effects on their network mechanical features. Actin and vimentin are copolymerized simultaneously and the viscoelasticity of the resulting filament networks is assessed by quantitative rheology. Unlike mixed IF proteins (e.g. desmin and vimentin (52)), which can form filaments containing both subunits, mixed actin and vimentin networks contain filaments with only one type of subunit. The mixed vimentin/actin filament network displays an elasticity that does not uniformly interpolate between the elasticity of networks of pure F-actin and that of pure vimentin IF. Analysis of the rheology measurements shows that a maximum in network stiffness occurs at an actin:vimentin ratio of 1:3 and that the tail domain of vimentin mediates inter-filament interactions between F-actin and filamentous vimentin.
A binding assay showed that the vimentin tail binds F-actin with a K d of ϳ14 M. This equilibrium dissociation constant is relatively high, but similar to dissociation constants of the cross-linking protein ␣-actinin for F-actin (44). Like other F-actin cross-linking proteins, vimentin does not require a high affinity for F-actin. Indeed, the interactions between F-actin and vimentin tails are multiplied by the high number of entanglements formed between each vimentin filament and the actin filaments in a dense network. These entanglements, reinforced by direct interactions between vimentin tails and actin filaments, prevent easy sliding of vimentin filaments when the network is under mechanical stress. This inhibited filament movement induces a high network elasticity, higher than in networks of actin and tailless vimentin filaments, which do not interact directly with each other.
In the absence of mutations in the rod domain, ⌬T vimentin assembles into filaments that appear identical to those formed by FL vimentin (9), as assessed by conventional electron microscopy. However, there are also reports that tailless vimentin filaments are slightly thicker than FL vimentin filaments (34). In vivo, the presence of the tail domain of vimentin appears to affect its ability of transfected Xenopus vimentin to localize inside the nucleus of cells (36,37). This suggests that, within the cell, the tail domain could affect the organization of vimentin. In many cell types, IFs appear to align more preferably with microtubules than with actin structures, presumably due to stronger interactions between microtubules and IFs (10). However, isolated vimentin tails colocalize with actin networks and not with microtubule networks, which suggests that vimentin tails have a higher affinity for F-actin structures and that microtubule-IF interaction is not mediated by the IF tail domain.
Scanning transmission EM of unstained intermediate filaments (including keratin, vimentin, and desmin) reveals that IFs have filament widths of ϳ16 nm, which is larger than the 10 nm filament width measured using conventional heavy metal stain EM (53,54). Cary et al. (10) propose that the excess 6 nm is possibly occupied by the head and tail domain, which projects out of the central axis of its filament. The hypothesized projection of nonhelical IF tails is similar to the observed projection of neurofilaments tails of their NF-H and NF-M subunits (55). Antibodies against NF-H and NF-M subunits localize to the peripheral as well as cross-bridges of NFs (56 -58). Bousquet et al. (51) also showed that the nonhelical tail domain of keratin K14 facilitates lateral filament association with keratin filaments forming bundles in vivo and gives the structures high mechanical attributes. These putative outward projecting vimentin tails may display the proper orientation to promote strong interactions with F-actin. Another mechanism of synergistic interaction between actin and vimentin filaments could be that the tails mediate electrostatic interactions that would influence the overall mechanical properties of mixed actin and vimentin filament networks. These observations together with our results suggest a direct interaction between F-actin and vimentin filaments mediated by the tail domain of vimentin, which results in an enhanced elasticity of mixed vimentin/actin filament in networks.
For a constant protein concentration, polymer theory predicts that the elasticity of a polymer network depends weakly on the intrinsic rigidity of the filaments (i.e. their persistence length) and depends mostly on inter-filament interactions between filaments. Increasing the rate of shear on a network interrogates the ability of filaments within the network to slide past one another. Actin filaments slide past each other much more readily than vimentin filaments. In further support of strong inter-filament interactions, vimentin structures also exhibit a low phase angle (e.g. the phase angle of water is 90°, whereas that of rubber is 0°) compared with F-actin. On the other hand, actin forms filament networks with a higher phase angle that decreases as the ratio of vimentin within the network increases.
Vimentin assemble into filaments with or without their tail domain, which suggests that vimentin assembly is not mediated by its tail. The elastic modulus of FL vimentin filament networks is not significantly higher in the absence of its tail, which implies that vimentin network elasticity does not stem from the interaction of its tail. This indicates that our observed enhanced stiffness of FL vimentin/actin filament networks does not simply stem from enhanced content in FL vimentin filaments in mixed networks.
Our study does not identify the vimentin-binding domain of F-actin, but our data strongly suggest that there is a direct interaction between the tail domain of vimentin and F-actin, which mediates a significant enhancement of the mechanical properties of mixed actin/vimentin filament networks compared with networks containing only F-actin or only vimentin.