Vimentin undergoes liquid–liquid phase separation to form droplets which wet and stabilize actin fibers

Significance

Vimentin, a type III intermediate filament (IF) protein, forms filaments which are an integral part of the cell cytoskeleton. We study a mutant vimentin which cannot form filaments. We show that it instead forms liquid droplets which are a result of liquid–liquid phase separation. Wild-type vimentin also transiently forms similar droplets as precursors to filaments, showing that there are at least two distinct states of vimentin. Additionally, these droplets of vimentin wet actin stress fibers and even remain attached as the actin fibers move due to treadmilling. The wetting interaction is sufficiently strong such that the vimentin coats the actin fibers enough to prevent drug-induced actin depolymerization. The wetting droplets may act as nucleation sites to facilitate vimentin IF assembly.

Abstract

The cytoskeleton is composed of F-actin, microtubules, and intermediate filaments (IFs). Vimentin is one of the most ubiquitous and well-studied IFs. It is involved in many activities including wound healing, tissue fibrosis, and cancer metastasis, all of which require rapid vimentin IF assembly. In this paper, we report that vimentin forms liquid condensates which appear to enable rapid filament growth. Given the transient nature of these droplets, we focus on properties of vimentin-Y117L, which has a point mutation that leads to formation of condensates but not IFs, enabling us to study these droplets in detail. The droplets dissolve under 1,6-Hexanediol treatment and under decreasing concentration, confirming that they are liquid, and phase separated. These condensates extensively wet actin stress fibers, rendering them resistant to actin-binding drugs and protecting them from depolymerization. We show similar behavior occurs in wild-type vimentin during its assembly into filaments.

The cytoskeleton is a set of biopolymer networks that is responsible for the motion and mechanical properties of the cell, as well as for intracellular cargo transport (1). In eukaryotic cells, it is primarily made of three interpenetrating networks: actin (2), microtubules (3), and intermediate filaments (IFs) (4). The actin cytoskeleton is formed by actin filaments and their associated proteins such as myosin, α-actinin, and filamin. Filamentous actin is concentrated in the cell cortex, frequently in the form of bundles or stress fibers. Cortical arrays of actin together with their associated proteins play major roles in cell motility, contractility, and mechanotransduction. Microtubules are the primary instrument for intracellular cargo transport. Structurally, they are tube-like polymers of alternating alpha and beta tubulin subunits. Motor proteins like kinesin and dynein transport vesicles and organelles throughout the cell along microtubules. During cell division, microtubules form the mitotic spindle, which is responsible for chromosome separation. The third component of the cytoskeleton, IFs, is more varied; it encompasses a host of proteins, including vimentin (5, 6), neurofilaments (7), and keratins (8). Vimentin, the most well-studied type III IF protein, is involved in the regulation of the mechanical properties of cells and is involved in a wide array of functions including wound healing and the epithelial–mesenchymal transitions that occur in both healthy tissues and in cancer cells undergoing metastasis. In cancer cells, the vimentin IF network provides the unique mechanical properties necessary to squeeze through a dense tumor tissue environment (9).

The known assembly mechanism of vimentin IFs (VIFs) involves multiple steps. The smallest stable subunits are vimentin tetramers which are thought to laterally assemble into unit length filaments (ULFs); these anneal longitudinally to form mature VIFs. In cells this assembly process occurs in multiple stages (10–12). This is best evidenced by MCF7 cells where the concentration of vimentin can be modulated gradually via a cumate inducible switch. When the concentration of vimentin is low, VIFs do not form. Instead, the cells exhibit small vimentin particles, which mature into VIFs with increasing vimentin concentration (13). A similar assembly of VIFs is observed in freshly trypsinized cells spreading over a surface, where vimentin forms dots before developing into VIFs (14). During the mitosis in BHK cells, the entire VIF network breaks down into punctate structures, similar in appearance to those observed in MCF7 cells; these puncta are recycled into VIFs in the daughter cells (15). The vimentin dots are also observed in lamellipodia of mouse embryonic fibroblasts (MEFs) (16). Peripherin, another type III IF protein closely related to vimentin, also forms similar particulate structures prior to the formation of neutrites in PC12 cells exposed to nerve growth factor (NGF) (17). This nonfilamentous state of vimentin, referred to in the literature as dots, particles, or puncta, is a ubiquitous state of vimentin and possibly of other type III IFs. These punctate structures are the precursors of the filamentous form of vimentin. Their formation, structure, and nature appear to be essential in determining their behavior and may influence their evolution from the punctate state to VIFs. However, very little is known about the punctate structures themselves. A better understanding of these structures would not only elucidate their importance in VIF assembly but would also provide insight into any possible function of the punctate structures themselves in the cell.

In this paper, we show that the vimentin puncta are condensates formed through liquid–liquid phase separation (LLPS). We focus on vimentin-Y117L (18–20) which assembles into ULFs but is inhibited from forming mature VIFs. When expressed in cells, this mutant forms punctate structures but cannot assemble into VIFs, allowing us to study the properties of the otherwise transient condensates. These structures have many hallmarks of liquid droplets: They merge, split, undergo shape fluctuations, and dissolve under 1,6-hexanediol treatment. Moreover, their rapid and reversible dissolution with decreasing concentrations is consistent with them being formed via LLPS. Unlike VIFs, which exhibit limited colocalization with the actin network but mostly interact with microtubules (21, 22), these condensates interact extensively with actin stress fibers. A large fraction of these droplets can wet stress fibers and remain attached to them for a long time. Such fiber-attached droplets exhibit specific types of motion: It is random fluctuations at short time scales but directed along the fibers at longer time scales. This directed motion is not driven by motors but originates from the retrograde flow of actin. When the actin stress fibers are wet by the vimentin droplets, they become resistant to actin-binding molecules. This, in turn, protects them from depolymerizing drugs like Cytochalasin B. We also demonstrate that wild-type vimentin, during its assembly, goes through similar phase-separated liquid condensates before forming VIFs. These results suggest that local increases in concentration due to phase separation may lead to more rapid assembly of VIFs.

Results and Discussions

Vimentin Forms Liquid Droplets Through Phase Separation.

The network formed by VIFs is an integral part of the cytoskeleton which determines the mechanical properties of the cell (5, 6). Once formed, the VIF network incorporates virtually all available vimentin subunits, leaving very few of them free (23). As indicated above, vimentin can also exist in a punctate state that is distinct from VIFs. For example, when vimentin expression is induced in vimentin-null MCF7 cells, filamentous networks do not form when its concentration is low. Instead, it forms small spherical puncta which ultimately transform into the VIF network (13). Similarly, when BHK cells undergo mitosis, their vimentin filaments dissolve into small spherical puncta which resemble those in MCF7 cells and transform back into a VIF network structure in each daughter cell (15). However, although these punctate states of vimentin are each precursors to the VIFs, they are transient; thus, making it difficult to study their properties.

To overcome this difficulty, we use vimentin-Y117L, where the substitution of the 117th amino acid of vimentin from Tyrosine (Y) to Leucine (L) inhibits VIF formation; this lets us study the behavior of the puncta in more detail (18, 19). To avoid interference from other IF networks we use MEFs which are devoid of any other type of cytoskeletal IFs (24). Stable expression of vimentin-Y117L fused to fluorescent mEmerald in vimentin null MEFs reveals their cytoplasm is replete with vimentin puncta that are predominantly spherical, as can be seen in a cell with larger than average puncta shown in green in Fig. 1A and Movie S1. Observing this behavior at higher magnification reveals that the puncta are highly dynamic, undergoing constant shape fluctuations. This can be seen from the different shapes of the same punctum marked by the yellow arrows in Fig. 1 B–E, and even more clearly in Movie S2. This behavior is reminiscent of liquid droplets. Even more convincing evidence of such liquid droplet-like behavior comes from observing two spherical puncta that come in contact with one another and merge into a single, larger punctum over 5 s as shown by the magenta circle in Fig. 1 B–E and Movie S2 as well as Movie S1. The merged punctum continues to undergo shape fluctuations and remains a single structure for more than 3 min, which suggests that they are not just randomly overlapping. Thus, these puncta must be liquid droplets and not solid aggregates or membrane-bound organelles.

Fig. 1.

Appearance and dynamics of vimentin-Y117L puncta. (A) MEF cell containing vimentin-Y117L-mEmerald shows fluorescent puncta throughout the cytoplasm. (B–E) Two spherical vimentin-Y117L puncta shown inside the magenta circle, come in contact with one another and merge into a larger punctum over 5 s. (F–I) A vimentin-Y117L punctum undergoes shape fluctuation into an ellipse which eventually splits into two smaller puncta. [Scale bar, 10 µm for (A) and 1 µm for (B–I).]

These liquid droplets also exhibit a very surprising feature: They can attain a highly elliptical shape which can split into two smaller spherical droplets, as illustrated in the magenta circle in Fig. 1 F–I and Movie S3. The highly elliptical shape, which precedes their splitting is unlikely to be due to shape fluctuations alone; it may instead result from shear forces, although these are not directly observed.

To explore whether the vimentin-Y117L droplets are formed by weak, nonspecific attractive interactions, we investigate their response to 1,6-Hexanediol; this is an aliphatic diol which selectively disrupts weak attractive interactions responsible for the formation of membrane-less liquid compartments in cells (25, 26). After 2 min in 0.5% 1,6-Hexanediol followed by fixation, cells lose a majority of the droplets. Instead, the vimentin-Y117L fluorescence appears uniformly spread through the cytoplasm, as shown by the uniform green color of the cell shown in Fig. 2A and SI Appendix, Fig. S1. By contrast, when we fix untreated cells, we observe an abundance of droplets, as shown in Fig. 2B. The 1,6-Hexanediol-induced dissolution of the liquid vimentin-Y117L droplets confirms that they are formed by weak, nonspecific attractive interactions. As a control, we compare the effect of the drug on another cytoskeletal structure, actin stress fibers, which are solid, unlike the liquid droplets. The vimentin-Y117L expressing cells are treated with the actin-binding dye SiR-actin for visualization of filamentous actin network seen mainly as stress fibers. The drug treatment has no obvious effect on the actin network of the cells, which can be seen by comparing actin stress fibers of the same drug-treated and untreated cells shown in magenta in Fig. 2 C and D, respectively. This demonstrates that 1,6-Hexanediol does not affect nonliquid polymeric actin. By contrast, the 1,6-Hexanediol treatment leads to a 10-fold decrease in the average number of droplets per cell, as shown in the boxplot Fig. 2E.

Fig. 2.

Effect of hexane-1,6-diol on MEF cells. (A) Vimentin-Y117L (green) of a MEF cell after 0.5% 1,6-Hexanediol treatment for 2 min followed by fixation showing uniform vimentin fluorescence throughout the cytoplasm. (B) Vimentin-Y117L (green) of a MEF cell without 1,6-Hexanediol treatment exhibiting droplets throughout the cell. (C) Actin stress fibers (magenta) of a MEF cell after 0.5% 1,6-Hexanediol treatment for 2 min followed by fixation. (D) Actin stress fibers (magenta) of a MEF cell without 1,6-Hexanediol treatment. (E) Boxplot of number of droplets in cells before (N = 17) and after (N = 22) 1,6-Hexanediol treatment demonstrating statistically significant decrease in number of droplets from ~450 per cell to ~50 per cell. (F) VIFs (green) in a MEF cell after 0.5% 1,6-Hexanediol treatment for 2 min. (G) VIFs (green) in an untreated MEF cell. (H) Actin stress fibers (magenta) in a MEF cell after 0.5% 1,6-Hexanediol treatment for 2 min. (I) Actin stress fibers (magenta) in an untreated MEF cell. (Scale bar, 20 µm.)

Since fusion of vimentin-Y117L with the fluorescent protein mEmerald could affect its physical properties, including the state of the punctate structures, we study cells with nonfluorescent vimentin-Y117L using immunostaining following fixation. We again observe abundant droplet-like vimentin structures as seen in SI Appendix, Fig. S2A. These droplets again disappear when the cells are treated with 0.5% 1,6-Hexanediol for 2 min, as seen in SI Appendix, Fig. S2B. Thus, the liquid nature of the vimentin-Y117L droplets is an inherent property of the protein and not a result of mEmerald fusion. As a control, we again use actin and visualize it by Phalloidin staining. The actin stress fibers remain unaffected by the 1,6-Hexanediol treatment, as shown in SI Appendix, Fig. S2 C and D. By staining the actin after the 1,6-Hexanediol treatment, we avoid any actin-stabilizing effect of SiR-actin.

To test whether the 1,6-Hexanediol-induced dissolution of droplets can be reversed, we treat the cells with 0.5% 1,6-Hexanediol for 2 min, remove the compound, replace normal culture medium to let the droplets reform and visualize actin and vimentin using immunostaining. Vimentin-Y117L droplets reappear within 2 to 4 h of 1,6-Hexanediol withdrawal, with some cells regaining droplets under 2 h as shown in SI Appendix, Fig. S3. This confirms that the 1,6-Hexanediol-induced dissolution of vimentin-Y117L droplets is reversible.

To explore the effect of the compound on fully polymerized VIFs, we carry out the same 1,6-Hexanediol treatment on WT MEFs which contain vimentin tagged with mEmerald, as shown in Fig. 2F. The treatment has no visible effect on the VIFs of these cells, as can be seen by comparing cells with and without 1,6-Hexanediol treatment, where the VIF networks are shown in green in Fig. 2 F and G, respectively. As a further control, actin stress fibers, visualized using SiR-actin, also remain unaffected by the compound treatment; this can be seen by comparing the same cells with and without 1,6-Hexanediol treatment, where the F-actin network is shown in magenta, in Fig. 2 H and I, respectively.

The fact that VIFs do not dissolve under conditions where the vimentin-Y117L droplets do dissolve confirms that the filamentous state of wild-type vimentin is distinct from the liquid droplet state of vimentin-Y117L. One way for cells to form membrane-less liquid compartments, such as these droplets, is LLPS (27, 28). This process is driven by weak interactions, causing components of a solution to spontaneously demix. The condensates formed through LLPS are typically liquid, owing to the weakness of the underlying interactions responsible for them. Furthermore, intrinsically disordered regions (IDRs) of proteins are often implicated in the weak interactions which give rise to LLPS (29, 30). Vimentin has two terminal IDRs which is consistent with its ability to phase separate (31, 32).

To test whether the droplet formation is indeed driven by LLPS, we explore the phase behavior of vimentin-Y117L. Because phase separation is concentration dependent, we change the concentration of all proteins in the cell by exposing it to culture medium diluted to 20% v/v in water. As the osmotic pressure outside the cell drops, its volume increases due to the influx of water thereby decreasing the concentrations of all cytoplasmic proteins. The extra water in the cell also hydrates the IDRs which alters their interaction, further affecting the structures that are already reliant on weak interactions (33). When cells with abundant vimentin droplets are exposed to diluted medium for 30 s, the droplets begin to dissolve, as can be seen by comparing the magnified images of an area of the cell in Fig. 3 A and B. Within 2 min, all droplets disappear, and a uniform vimentin fluorescence is observed throughout the cytoplasm, as Fig. 3C demonstrates. Since a change in the concentration and interaction strength should only affect phase-separated structures, we also tested the impact of dilution on actin stress fibers, which are not phase separated and have no IDRs, as a control. In cells treated with SiR-actin, actin stress fibers are detected throughout the diluted medium treatment. In the time it takes for the vimentin droplets to dissolve entirely, the actin stress fibers remain unaffected, as can be seen by comparing the actin network of the cell, shown in magenta in Fig. 3 A–C. If this dissolution reflects the phase behavior, it should be reversible. To test this, we remove the diluted medium and add fresh normal medium to the cell and observe both vimentin and actin in the cell. Within 30 s, the uniform fluorescent background becomes nonuniform and within 2 min droplets start reappearing throughout the cell, as shown in green in Fig. 3 D and E, respectively. The droplets are completely recovered by 5 min, as the green in Fig. 3F exemplifies. The actin, however, remains unaffected, as seen in magenta in Fig. 3 D–F. The whole cell image of the cell shown in Fig. 3 is shown in SI Appendix, Fig. S4. These results demonstrate that vimentin droplet formation is due to phase separation. The reappearance of droplets is much faster than it is in the 1,6-Hexanediol experiment. When osmotic pressure is reversed, the water is rapidly driven out of the cell. By contrast, when 1,6-Hexanediol treated cells are exposed to fresh culture medium, the 1,6-Hexanediol molecules must diffuse out of the cell before droplets can reform, which is substantially slower than the osmotic-pressure-driven water efflux. This accounts for the drastically different times required for droplet reappearance in the two experiments.

Fig. 3.

Effect of concentration change on vimentin-Y117L droplets in live MEF cells. (A–C) Vimentin-Y117L droplets (green) and actin stress fibers (magenta) in a cell before (A), 30 s after (B), and 2 min after (C) diluted medium treatment. (D–F) Vimentin-Y117L droplets (green) and actin stress fibers (magenta) in the same cell after 30 s (D), 2 min (E), and 5 min (F) of restoring fresh culture medium. (Scale bar, 5 µm.)

Vimentin Droplets Colocalize with Actin Stress Fibers.

In our observation of droplets in cells expressing vimentin-Y117L, we notice that they frequently appear in linear arrays reminiscent of the distribution of actin stress fibers. To explore whether the vimentin-Y117L droplets interact with actin, we examine how they are spatially distributed in the cell. A large fraction of the droplets overlap with the actin stress fibers, as seen by comparing the actin network shown in magenta with the vimentin-Y117L droplets shown in green in Fig. 4 A and B, respectively. To better visualize this phenomenon, we employ a threshold function to obtain an image, where the areas with overlapping fluorescence signals are white; the extent of colocalization can be seen from the abundance of white areas in Fig. 4C. A smaller, magnified area of the cell which demonstrates the overlap better is shown in SI Appendix, Fig. S5. On average, ~65% of vimentin is colocalized, as seen in Fig. 4G. The droplets which do not colocalize with actin might be free or may interact with microtubules (14, 19), as previous studies show.

Fig. 4.

Colocalization between F-actin and vimentin. (A and B) Actin (magenta) and vimentin (green) of a MEF cell with vimentin-Y117L droplets. (C) Colocalization map of the cell showing colocalized (white) region, un-colocalized actin (magenta), and un-colocalized vimentin (green). (D and E) Actin (magenta) and vimentin (green) of a MEF cell with VIFs. (F) Colocalization map of the cell showing colocalized (white) region, un-colocalized actin (magenta), and un-colocalized vimentin (green). (G) Boxplot of percentage of colocalized vimentin for MEF cells with vimentin-Y117L droplets (N = 31) and VIFs (N = 24). (Scale bar, 20 µm.)

To study how this behavior compares to WT MEFs containing VIFs, we quantify the overlap between actin stress fibers and VIFs. The two filament networks are spatially distinct, as seen by comparing the F-actin network shown in magenta and the VIF network shown in green in Fig. 4 D and E, respectively. The thresholded image for this cell shows mostly independent actin stress fiber and VIF signals, as seen by the absence of white areas in Fig. 4F. The average colocalization of VIFs is less than 30%, in stark contrast to the 65% for the droplets, as seen in Fig. 4G. The 30% overlap between VIFs and F-actin is consistent with recent results which show that actin and vimentin networks are mainly colocalized in the cortical region of the cell (34). In contrast, our results demonstrate that the actin–vimentin colocalization behavior is enhanced in the droplet state of vimentin.

To investigate the nature of the interaction between the vimentin-Y117L droplets and actin stress fibers, we study the motile properties of the droplets. While free droplets move rapidly in every direction, colocalized droplets move along the actin stress fibers they are on, rarely coming off, as seen in Movie S4. In one example, a droplet travels ~10 µm along a single actin stress fiber for ~9 min, as shown by the yellow circles in Fig. 5 A–C. Thus, the interaction between the vimentin droplets and stress fibers must be strong enough to prevent detachment. Both these motions are distinct from the microtubule-dependent motion of such droplets, which is bidirectional (19) with much larger step sizes (14), and has been confirmed in cells where the microtubules have been dissolved using Nocodazole (14, 19). Furthermore, our results demonstrate that the droplets on actin stress fibers are not static but move in a specific manner.

Fig. 5.

Motion of vimentin-Y117L droplets. (A–C) Vimentin-Y117L droplet (green), shown inside the yellow circle traveling ~10 µm along an actin stress fiber (magenta) over ~9 min. (D) Tracks of all the identifiable vimentin-Y117L droplets in the peripheral region of a cell overlaid with actin stress fibers (magenta). (E) Droplet tracks longer than 16 s overlaid with actin stress fibers (magenta). (F) Droplet tracks longer than 30.5 s overlaid with actin stress fibers (magenta). (G) Plot of mean squared displacement, $⟨\mathrm{\Delta }{r}^2(t)⟩$, as a function of time, $\mathrm{\Delta }t$, for a droplet in a cell without (red) and with Nocodazole treatment (blue), both showing two different slopes.

To quantify the motion of the droplets, we determine their trajectories as a function of time. When these are all plotted on the same image, the cell is filled with paths going in all directions; they are jumbled up and cannot be visually distinguished from one another, as shown in Fig. 5D. However, longer trajectories predominantly lie along stress fibers. This can be illustrated by plotting trajectories longer than 16 s which preferentially eliminates the off-actin movement, as shown in Fig. 5E; doing the same for trajectories longer than 30.5 s almost exclusively shows the motile droplets along stress fibers, as shown in Fig. 5F. Being stuck to the actin stress fibers makes the colocalized droplets stay within the imaging volume longer than the free ones, resulting in longer trajectories.

We further investigate the droplet dynamics by measuring the mean squared displacement, <$\mathrm{\Delta }{r}^2(t)$>, as a function of time difference, $\mathrm{\Delta }t$. The longest tracks contain both long time-scale and short time-scale behavior; therefore, we use them for this analysis. The plot of <$\mathrm{\Delta }{r}^2(t)$> has two distinct slopes, corresponding to two distinct modes of motion, as can be seen from the data shown in the solid curve in the logarithmic plot in Fig. 5G. At long time scales, the slope is 1.9 (±0.03), indicating the droplets move directionally with a uniform velocity. The slope drops significantly at shorter time scales, which corresponds to random fluctuating motions of the droplets. The cross-over from fluctuations to the directed movement takes place at $\mathrm{\Delta }t\approx 10$ (±2) s. This transition is replicated by virtually all droplets, as can be seen, for example, from the lines corresponding to other particles, as shown in red SI Appendix, Fig. S6A. The average velocity of this directed motion is 1.3 (±0.05) µm/min. This is too slow to be motor driven: actin-based myosin motors move at speeds of 12 to 3,600 µm/min (35–37) and velocities of microtubule-based motors kinesin and dynein are 50 and 100 µm/min (38, 39), respectively. However, actin filaments also undergo treadmilling: actin filaments at the periphery of the cell incorporate monomers while filaments near the nucleus disassemble, causing the filaments to move from the periphery toward the nucleus at 0.6 to 1.8 µm/min (40, 41). This is termed retrograde flow and is independent of motors. The vimentin droplet velocities are well within this range and the directional, on-actin motion of the droplets is thus attributable to actin treadmilling.

To confirm that the motion we observe depends primarily on actin, we treat the cells with 10 µM Nocodazole for 1 h to depolymerize the microtubules (19). Just as in the cells without drug treatment, the motion of these droplets transition from random fluctuations to directional motion with uniform velocity around $\mathrm{\Delta }t\approx 10$ (±0.25) s, as can be seen from the blue line in Fig. 5G and SI Appendix, Fig. S6B. The average velocity of the directed motion of the droplets in these cells is 1.37 (±0.075) µm/min, which is, again, very similar to the velocity of droplets in cells without Nocodazole treatment. This demonstrates that this motion is not driven by microtubules.

Vimentin Droplets Wet and Coat Actin Stress Fibers and Protects Them.

Unlike spherical free droplets, those colocalized on actin stress fibers are often elongated, as shown by the yellow arrows in Fig. 6 A and B. They can also undergo drastic changes in shape as they move along the actin stress fibers, as seen by comparing the shapes of the same droplet shown in Fig. 6 C–F. The ability of the droplets to change shape suggests that they remain liquid even when bound to actin stress fibers. This suggests that the vimentin droplets wet the stress fibers and hence spread along the fibers. Wetting of a liquid drop on a surface occurs because the interaction energy between the liquid and the surface is less than the surface energy of the liquid and the surrounding fluid. This causes the droplets to spread over the fiber surface resulting in the loss of their spherical shapes (42–44). The elongation of droplets along actin fibers might also explain the elliptical deformation observed in Fig. 1H preceding the droplet fission.

Fig. 6.

Vimentin droplets coat the actin stress fibers in MEF cells. (A) Vimentin-Y117L droplets (green) showing elongated droplets with yellow arrow. (B) Overlay image of the vimentin-Y117L droplets (green) and the actin stress fibers (magenta) of the same cell; yellow arrow showing that the elongated droplets colocalize with the actin stress fibers. (C–F) Expanded images of a vimentin-Y117L droplet traveling along actin stress fibers exhibiting different shapes. (G) MEF cells with VIFs showing their nucleus (blue) and actin (magenta) after 5 µg/mL Cytochalasin B treatment for 1 h demonstrating almost no cells have actin stress fibers left. (H) MEFs with vimentin-Y117L showing their nucleus (blue) and actin (magenta) after 5 µg/mL Cytochalasin B treatment for 1 h demonstrating surviving stress fibers in cells. (I) Boxplot of total actin stress fiber length extracted from the fluorescence images for cells with VIFs and cells with vimentin-Y117L droplets with and without Cytochalasin B treatment.

To further explore the consequences of vimentin droplets wetting actin stress fibers, we treat cells with Cytochalasin B, which binds to the polymerizing end of actin filaments and prevents their elongation (45). However, the depolymerization step of actin treadmilling remains unaffected and thus the actin filaments eventually dissolve. When we treat WT MEFs containing normal networks of VIFs with 5 µg/mL of Cytochalasin B for 1 h, they lose almost all actin stress fibers, as can be seen from the lack of fibers in the actin image (magenta) of cells shown in Fig. 6G. By contrast, cells with vimentin-Y117L droplets retain a substantial amount of actin stress fibers after the same drug treatment, as seen in Fig. 6H. We quantify the amount of stress fibers in the cell from the fluorescent images. We identify the fibers, evaluate lengths of individual fibers, and use the sum of the length of all the identified actin stress fibers in the cell to compare the amount of actin stress fibers in cells. Without drug treatment cells containing VIFs have an average of 731 (±23) µm of actin stress fibers and cells containing vimentin-Y117L droplets have an average of 783 (±21) µm. In drug-treated, VIF-containing cells, the total fiber length is almost fivefold lower than the control population. In contrast, vimentin-Y117L cells have half the length of actin stress fibers even after drug treatment. This can be seen by comparing the sharp decrease between the first and second boxes with the less drastic decrease between the third and fourth boxes in Fig. 6I. To verify whether the wetting of actin stress fibers by vimentin-Y117L droplets prevents their dissolution under Cytochalasin B treatment, we evaluate the fraction of actin fibers coated by vimentin in the mutant cells. Using the colocalization threshold analysis, we find that ~55% of the actin stress fibers are coated with vimentin-Y117L droplets, which is consistent with the survival of half the fibers in the mutant cells.

Microscope images show that the actin stress fibers are partially coated by distinct droplets, as can be seen by the fluorescence image of vimentin-Y117L droplets (green) and actin stress fibers (magenta) shown in Fig. 6B. Therefore, we expect only those coated sections of the fibers to be protected after Cytochalasin B treatment. Surprisingly, however, some drug-treated cells containing vimentin-Y117L exhibit continuous stress fibers, as can be seen in the image of actin shown in magenta in Fig. 6H. One possibility to account for this behavior is that a thin layer of vimentin, which cannot be imaged by with a fluorescence microscope, coats the entirety of these fibers.

Wild-Type Vimentin Forms Liquid Droplets Which Colocalize with Actin Stress Fibers.

The results regarding LLPS and wetting, though conclusive, are from vimentin-Y117L which cannot form VIFs. Therefore, we explore the properties of wild-type vimentin, capable of forming filaments, during its assembly. We treat MEF cells expressing mEmerald tagged VIFs with 20% (v/v) diluted culture medium which reversibly dissolves VIFs (46, 47), as demonstrated in Fig. 7A and in Movie S5. When we replace the diluted medium with fresh, complete medium, the vimentin reassembles into VIFs in a series of steps, as shown in Movie S6. First, the dissolved vimentin reassembles into nonfilamentous punctate structures, as shown in Fig. 7B. When two of these puncta come into contact with one another, they can merge, as seen inside the yellow circle in Fig. 7 D–F, confirming that the puncta are liquid droplets likely formed through LLPS. These droplets, consisting of wild-type vimentin, eventually form VIFs, as shown in Fig. 7C. This confirms that even wild-type vimentin can undergo phase separation to form droplets and that LLPS appears to play a role in the assembly of VIFs.

Fig. 7.

Vimentin filament precursors are liquid droplets and colocalize with actin stress fibers in wild-type MEF cells. (A) Dissolved vimentin (green) after VIF dissolution using 20% diluted medium. (B) Wild-type vimentin droplets (green) formed as precursors to VIFs after 38 s of complete medium replacement. (C) Reassembled VIF network (green) after 200 s of complete medium replacement. (D–F) Two spherical wild-type vimentin droplets (green) merging into a single droplet inside the yellow oval. (G and H) Actin stress fibers (magenta) and VIF precursor droplets (green) of an area in a cell, respectively. (I) Overlay image showing the droplets colocalize with actin stress fibers. (J and K) Actin stress fibers (magenta) and VIFs (green) of a cell, respectively. (L) Overlay image showing very little colocalization between the two networks.

To explore the behavior of the precursor droplets during VIF assembly after water treatment, we observe how they interact with actin stress fibers. These droplets colocalize with stress fibers, which can be seen from the image of the actin network (magenta) and the vimentin droplets (green), as well as the overlaid image shown in Fig. 7 G–I, respectively, and the whole cell image in SI Appendix, Fig. S7. In the same cells, when vimentin exists as VIFs, they show very little overlap with actin stress fibers, as can be seen by comparing the actin network (magenta) and the vimentin network (green), as well as the overlaid image shown in Fig. 7 J–L, respectively. The droplets differ from the VIFs only in their physical state, thus the enhanced overlap between vimentin droplets and actin stress fibers indicates that actin–vimentin interaction is governed by vimentin’s state.

Conclusions

In this paper, we show that vimentin-Y117L forms liquid condensates through LLPS in MEF cells. Although the exact composition of these droplets is unknown, the vimentin-Y117L mutant cannot elongate beyond tetramers and ULFs, and therefore these subunits must be sufficient to form the condensates (18, 32). The phase separation of proteins is strongly linked to IDRs in their structure, which can lead to weak, nonspecific attractions that drive the phase separation (29, 30). Vimentin-Y117L has IDRs on both its N- and C- terminals, and thus its phase separation is not surprising. Interestingly, wild-type vimentin has the same IDRs as vimentin-Y117L and therefore should exhibit similar phase behavior. Indeed, we observe that wild-type vimentin forms liquid droplets. This suggests that these droplets may be phase-separated condensates. The ability of vimentin to reversibly form biomolecular condensate under oxidative stress further supports its ability to phase separate (48).

In this paper, we show that wild-type vimentin droplets transform into VIFs during their assembly. This implies that these condensates are important for VIF assembly. Moreover, wild-type vimentin cannot form filaments when its N-terminal IDR is removed (49); this suggests that weak interactions driven by IDRs may also contribute to VIF assembly (50).

The VIF network is one of three filamentous protein networks in MEFs and virtually all mesenchymal cells, filamentous actin, and microtubules being the other two. The growth of each of these cytoskeletal filaments requires addition of subunits with both the correct position and correct orientation, which slows the extension of the filaments. Increasing the local concentration at the growing tip of the filaments increases the growth rate. Interestingly, this seems to be accomplished by phase-separated condensates in each case. For actin filaments, polymerization occurs 14 times faster inside N-WASP, Nephrin, and Nck condensates (51). For microtubules, assembly occurs inside centrosomes, which are phase-separated condensates formed by numerous auxiliary proteins (52). Our results suggest that a similar condensate-mediated process may be involved in VIF assembly; vimentin subunits are concentrated inside droplets where VIFs might assemble. Interestingly, no other protein is known to be required for VIF assembly, thus, suggesting that vimentin might be able to form these condensates by itself, unlike actin or microtubules. However, we do not know exactly which vimentin subunits make up the wild-type condensates. The attractive interaction due to IDRs are essential for phase separation to form condensates and in turn increase the concentration during the VIF growth process; these condensates are therefore implicated in the formation of the VIFs themselves. It is therefore conceivable that the formation of VIFs may depend on the same attractive interactions due to the IDRs.

In this paper, we show that vimentin-Y117L condensates colocalize extensively with actin stress fibers and remain attached to them. This interaction is driven by wetting. It is likely strong enough to coat the fibers continuously and prevent their disassembly under Cytochalasin B treatment; such a strong wetting interaction would be rare in proteins. We similarly observe wild-type droplets colocalizing with stress fibers before assembling into filaments. This suggests that actin plays a role in the assembly of VIFs. Actin stress fibers may act as a surface on which vimentin droplets nucleate. Additionally, being colocalized on actin may help to orient the vimentin subunits to better incorporate into the growing filaments. The VIFs growing out of these droplets may use actin stress fibers as their template; as these VIFs elongate, they eventually detach from the stress fibers. This might be due to the mismatch in their persistence lengths (53–55). The VIFs have a persistence length of 1 µm, which is about 15 times lower than that of single actin filaments; stress fibers, being bundles of cross-linked actin filaments, have an even higher persistence length. Due to the lower persistence length of VIFs, they would have to be straightened to remain attached to the stress fibers. This becomes energetically unfavorable with increasing length of the VIF, which would cause them to detach at longer length scales as they grow. This demonstrates how components of the cytoskeletal network may influence one another’s assembly through phase separation and wetting. By contrast, VIFs do colocalize well with microtubules which is likely due to interactions with microtubule-based motor proteins. Further, the wetting interaction may continue to influence the behavior and organization of VIFs after their assembly. Previous studies have demonstrated that VIFs and individual F-actin components of stress fibers are located within 11 nm of each other in the cortical region of the cell (34). Though plectin is instrumental in cross-linking VIFs to actin in cells, it has a size of ~200 nm; thus, it is unlikely that the 11 nm space, seen in this case, would accommodate any isoform of plectin. However, wetting, being a direct interaction between the two proteins might allow for the extreme proximity of VIFs and F-actin in the cell cortex and thereby provide further insight into their overall organization in the cell.

Materials and Methods

Cell Culture and Live Actin Staining.

MEF vimentin-mEmerald (MEF vim-mEmerald), MEF vimentin-Y117L (MEF vim-Y117L) and MEF vimentin-Y117L-mEmerald (MEF vim-Y117L-mEmerald) cell lines were provided by the labs of Robert Goldman and Stephen Adam of Northwestern University. MEF cells are cultured in Dulbecco’s modified Eagle’s Medium (DMEM, Corning, Catalog No. 10-013-CV) supplemented with 10% fetal bovine serum (FBS, Avantor, Catalog No. 97068-085) and 1% Penicillin/Streptomycin (Corning, Catalog No. 30-002-CI). For live actin staining, cells are cultured for at least 24 h and then treated with 100 nM SiR-actin (Cytoskeleton Inc., Catalog No. CY-SC001) for 12 h. After treatment, the cells are washed with 1X Phosphate-buffered saline (PBS, Corning, Catalog No. 21-040-CV) and treated with fresh complete culture medium before imaging.

Immunostaining.

Cells are fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton-X-100 (Thermo Fisher Scientific, Catalog No. A16046.AE), blocked with blocking buffer,^* treated with antivimentin primary antibody (Abcam, ab45939) overnight at 4 °C and treated with secondary antibody (Thermo Fisher Scientific, A21244) and phalloidin (Abcam, ab176753) for 1 h at room temperature. Next, they were treated with DAPI at room temperature for 10 min, washed, and imaged.

1,6-Hexanediol Treatment.

MEF cells are cultured for at least 24 h and stained with SiR-actin. They are treated with 0.5% (w/v) 1,6-Hexanediol (Sigma-Aldrich, 240117) in complete culture medium for 2 min, fixed with 4% paraformaldehyde and imaged. For immunostaining, MEF cells are cultured for at least 24 h, treated with 0.5% (w/v) 1,6-Hexanediol in complete culture medium for 2 min, fixed with 4% paraformaldehyde and stained for vimentin, paxillin, and actin before imaging.

Diluted Media Treatment.

MEF cells are grown for at least 24 h and treated with SiR-actin. While the cells are on the microscope, the media is aspirated leaving some liquid behind to prevent cells from drying out. Immediately, the 20% (v/v) diluted culture media in water is added and imaging is continued. For recovery, the diluted media is aspirated, and fresh complete culture media is added to the cells.

Droplet Counting.

The droplets are identified by using the “Analyze Particle” function in Fiji. We optimize the threshold for every image separately to ensure that the noise outside the cell is not detected as particles. Each result is verified visually to avoid spurious particle detection.

Colocalization Threshold Calculation.

Colocalization between actin and vimentin images is calculated by using the “colocalization threshold” plugin in Fiji. It follows the Costes method for determining thresholds; if the intensities in the two channels are above the threshold values of the respective channels for a particular pixel, then it is considered colocalized. The two thresholds are determined iteratively to ensure that the Pearson’s correlation coefficient for all non-colocalized pixels is zero and the coefficient is greater than zero for all colocalized pixels.

Cytochalasin B Treatment.

MEF cells are grown for at least 24 h and treated with 5 µg/mL of Cytochalasin B (Millipore-Sigma, Catalog No. 250225) in complete culture medium for 1 h and fixed with 4% paraformaldehyde. Cells are stained with Phalloidin for actin and DAPI (Thermo Fisher Scientific, Catalog No. 62247) for the nucleus before imaging.

Nocodazole Treatment.

MEF cells are grown for at least 24 h and treated with 10 mM Nocodazole in complete culture medium for 1 h at 37 °C before imaging.

Droplet Tracking and Motion Analysis.

Droplets are tracked from fluorescence microscopy images using the “trackmate” plugin in Fiji. First, the particles are detected using a Laplacian of Gaussian (LoG) detector for every single frame and quality threshold is adjusted to ensure reliable detection. The tracking of particles is done by “advanced Kalman tracking.” The X, Y, and time coordinates of the particles are exported as .xml files and the following analyses are performed using Matlab.

We calculate the mean squared displacement (<$\mathrm{\Delta }{r}^2$>) for all possible time differences ($\mathrm{\Delta }t$) in the trajectories and plot them on a log–log scale. For ballistic, or directional, motion the governing equation of motion is $\mathrm{\Delta }r=v\ast \mathrm{\Delta }t$, where v is the velocity of the particle. So, the mean squared displacement is related to time as <$\mathrm{\Delta }{r}^2>=v^2\ast \mathrm{\Delta }{t}^2$. On a log-scale, the equation becomes $\begin{array}{c}log\left(<\mathrm{\Delta }{r}^2>\right)=2\ast log\left(v\right)+2\ast log\left(\mathrm{\Delta }t\right)\hfill \end{array}$, which results in a slope of 2. By fitting the ballistic part of the data to a linear regression model, we can estimate the intercept of the plot from which the directional velocity of the droplets can be calculated. The codes for the Matlab analyses are provided on GitHub.

Actin Fiber Detection.

The microscopic images are manually segmented into images of single cells. We adjust the brightness and contrast of these images for better visibility, as can be seen by comparing the same actin network image shown at different brightness and contrast levels in SI Appendix, Fig. S8 A–C. The subsequent detection of the fibers is done by an original Python based algorithm using the computer vision library OpenCV2 (56).

A Gaussian blur is applied on the image to remove high frequency noise. Then a Canny edge detection identifies the filament boundaries; this works by examining the gradient of the image at each pixel in all directions. Pixels that are local gradient maxima along their gradient direction are edge candidates. Then we set an upper and a lower threshold for gradients. All edge candidates above the upper threshold are considered true edges; all edge candidates below the lower threshold are rejected. Whether edge candidates between the two thresholds are rejected or not depends on their spatial position. If they are contiguously connected to other true edges, then they are considered true edges themselves, if not they are rejected. The detected edges are skeletonized whenever necessary to a width of 1 pixel.

The gradient direction of an edge is always perpendicular to the filament itself, and for a filament, there are two edge pixels corresponding to the opposite sides of the filament. We detect the filaments by the existence of two such edges that are within 0.7 µm along the gradient direction of the original, unblurred image; this threshold is set for individual images. A “filament pixel” is placed on the position between two edge pixels. The filament pixels are connected together by drawing a circle of fixed radius around them and then skeletonizing. Filaments with an in-image area smaller than a specific threshold are eliminated. We estimate the length of individual filaments by contourizing them with OpenCV2, which is done by calculating their perimeter and dividing by two. This also reduces the length impact of spurious pixels that might emerge from skeletonization. From individual filament lengths, we obtain the total filament length in cells. All the relevant codes are provided on GitHub. We verify the accuracy of the detection by comparing the original actin network image with an image of the detected fibers overlaid on it, as is shown in SI Appendix, Fig. S8 D and E, respectively.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Video of a live MEF cell with vimentin-Y117L-mEmerald droplets showing the dynamic nature of the droplets.

Movie S2.

Magnified video of two vimentin-Y117L droplet merging and other droplets undergoing shape fluctuations in a live MEF cell.

Movie S3.

Magnified video of a vimentin-Y117L droplet undergoing shape fluctuation and eventually splitting into two droplets in a live MEF cell.

Movie S4.

Vimentin-Y117L droplets moving along treadmilling actin stress fibers at the periphery of a live MEF cell.

Movie S5.

Wild type VIFs dissolving under 20% diluted medium treatment in a live wild type MEF cell.

Movie S6.

Reassembly of the wild type VIF network under complete medium in a live wild type MEF cell.

Data, Materials, and Software Availability

The code used for analyzing the droplet motion is available at the GitHub repository (https://github.com/benfdup/r_squared_analysis/blob/main/rsquared_3.m) (57). The code used for filament extraction and filament length analysis is available at the GitHub repository (https://github.com/benfdup/filament_code/blob/main/filament_detect.py) (58). The images of cells used in this work is available at the Zenodo repository (https://zenodo.org/records/14624919) (59).