Cofilin cooperates with fascin to disassemble filopodial actin filaments

The dynamic rearrangement of actin filaments is fundamental for cell motility, and cells use a plethora of proteins to precisely regulate and coordinate the polymerization and depolymerization of actin filaments (Pollard and Cooper, 2009; Bugyi and Carlier, 2010). Although much has been learnt about the cellular players that mediate actin filament nucleation and elongation in protrusive structures such as lamellipodia and filopodia (Chesarone and Goode, 2009; Insall and Machesky, 2009), little is known about the mechanisms of filament disassembly.

One of the key players in filament depolymerization and actin monomer recycling is the actin depolymerization factor cofilin (Bernstein and Bamburg, 2010). Cofilin preferentially binds to ADP–F-actin and mediates severing and depolymerization by altering the mechanical properties of the filament: structural and biochemical analyses revealed that cofilin binding changes the subunit tilt and increases the helical twist of the filament (McGough et al., 1997; Galkin et al., 2001), thereby weakening lateral contacts between actin monomers (McGough and Chiu, 1999; Bobkov et al., 2004; Paavilainen et al., 2008) and considerably increasing the elasticity of the filament (Prochniewicz et al., 2005; McCullough et al., 2008).

In cells, cofilin is associated with the entire lamellipodium (Lai et al., 2008) and in one model promotes actin filament treadmilling in the lamellipodium and in Listeria comet tails by mediating filament disassembly of aged actin filaments (Loisel et al., 1999; Iwasa and Mullins, 2007; Kiuchi et al., 2007; Lai et al., 2008). Consistently, cofilin is an essential component of reconstituted motility medium (Loisel et al., 1999). However, in contrast to its established role in the turnover of lamellipodial actin filaments, a potential function in the dynamics of filopodia has so far been elusive. Filopodia are spiky, actin-rich protrusions that frequently emerge from the leading edge of the cell, and serve as sensory organelles, e.g. in axon guidance or endothelial zippering (Dent et al., 2007). Importantly, the fast growing, barbed ends of the filaments packed together in a filopodium point towards the membrane, producing enough force to push the membrane outwards. There is much debate about the mechanisms of filopodia initiation, but it is commonly believed that protrusion and retraction of these structures are effected by modulating the insertional assembly of actin monomers at their tips, and by inducing filament disassembly, potentially mediated by cofilin, at their base (Gehler et al., 2004; Mattila and Lappalainen, 2008; Faix et al., 2009; Hotulainen and Hoogenraad, 2010). In addition, actin filaments in filopodia are compacted into dense bundles by specialized proteins such as fascin, which localizes along their shafts (Otto et al., 1979; DeRosier and Edds, 1980), enhancing their rigidity and thus promoting their protrusion from the cell body (Vignjevic et al., 2006; Lieleg et al., 2007). Notably, fascin appears to undergo frequent cycles of association and dissociation with actin bundles both in vivo and in vitro instead of stably binding to filaments, which might be an important feature in regulating the dynamic growth and shrinkage of filopodia (Vignjevic et al., 2006; Aratyn et al., 2007). Moreover, fascin was shown to induce overtwisting of actin filaments within bundles, resulting in bundles with a hexagonal geometry consisting of ~20 actin filaments at maximum (Claessens et al., 2008).

We investigated the molecular details of the interplay of fascin and cofilin during actin assembly and disassembly with a combined approach, using multi-color total internal reflection fluorescence (TIRF) microscopy with purified proteins, fluorimetric assays and correlated live-cell imaging and electron microscopy (EM) tomography. We found that cofilin-mediated filament severing is enhanced in fascin-bundled actin filaments, and that cofilin can efficiently disassemble fascin-crosslinked actin bundles in vitro. Furthermore, we show that cofilin has an unexpected role in disassembly of filopodia, as it massively accumulates in tip and shaft regions of these protrusions, resulting in their rapid retraction and disassembly.

Initially, we employed in vitro TIRF microscopy of polymerizing, freely diffusing actin filaments to visualize the dynamic process of fascin-mediated actin bundle formation. This setup made it possible to discriminate between the fast growing barbed end and the slowly growing pointed end of the filaments and therefore to unambiguously determine the polarity of the formed bundles, as well as to determine the number of filaments per bundle and to analyze the kinetics of bundle formation. Single actin filaments that grew in the presence of fascin frequently fused with other filaments, resulting in a network of actin filament bundles (Fig. 1A; supplementary material Movie 1). The elongation rate of actin filaments was not altered by fascin and was ~11 subunits/second for TIRF experiments using 1.3 μM actin (30% Alexa-Fluor-488 labeled). The fluorescence intensities of the actin bundles correlated directly with the number of actin filaments within the bundle, which could readily be determined by following the time-resolved fusion of single actin filaments with each other and with other bundles (Fig. 1B). The number of filaments per bundle was largely independent of the fascin concentration used in our assay, as a concentration range from 10 nM to 10 μM fascin produced bundles consisting of approximately three to six actin filaments at steady state, with a maximal average number of filaments per bundle at 500 nM fascin (Fig. 1C; supplementary material Fig. S1). Although fascin is known to restrict bundle thickness to ~20 filaments (Claessens et al., 2008), the small effect of fascin concentration on bundle thickness observed here is probably caused by the low total amount of actin filaments in these assays, limiting the number of actin filaments available for bundle formation. Nevertheless, the total amount of bundled actin increased with fascin concentration, but reached a maximum at 50–100 nM fascin in the reaction mixture. This was true for both TIRF microscopy and low-speed sedimentation assays of actin bundles followed by densitometry after SDS-PAGE (supplementary material Fig. S1).

The analyses of the dynamics of fascin-mediated bundle formation showed some striking effects: first, the bundles formed became very rigid after fusion of two to three filaments, resulting in the appearance of numerous ‘kinks’ (supplementary material Movie 1), and the force produced by fascin-mediated stiffening of the bundle was apparently high enough to occasionally break existing bundles (supplementary material Movie 2). Second, analysis of the directionality of single growing filaments within the bundles revealed that all bundles formed by fascin were exclusively composed of parallel actin filaments (Fig. 1D; supplementary material Movie 2). Third, single filaments were rapidly fused to the actin bundle after coming in close proximity, a process we will refer to herein as ‘zippering’ (Fig. 1E; supplementary material Movie 3). This process was remarkably rapid, exceeding velocities of 700 subunits per second (Fig. 1E), which corresponds to the fusion of more than 100 μm actin filament/minute. Thus, the rate of fascin-mediated bundle formation is considerably faster than both actin filament elongation rates in vitro and protrusion rates of 2–30 μm/minute of filopodia and microspikes seen in different cell types (Sheetz et al., 1992; Costantino et al., 2008; Geraldo and Gordon-Weeks, 2009), and thus likely not rate-limiting during filopodia formation.

A previous study showed that spontaneously formed actin bundles were more resistant to cofilin-mediated severing than single filaments in vitro (Michelot et al., 2007). However, the effects of filopodial bundling proteins such as fascin on cofilin-mediated filament severing were not investigated. Therefore, we analyzed the dynamics of actin networks formed in the presence of both fascin and cofilin. By using single filament TIRF microscopy, we show that cofilin alone efficiently severs growing actin filaments with a robust activity at 300 nM in our assay (Fig. 2A; supplementary material Movie 4), whereas concentrations below 200 nM did not induce substantial filament severing (data not shown). Monitoring cofilin-mediated severing of polymerizing filaments instead of filaments at steady state has the advantage of allowing the unambiguous identification of severing events resulting from growth of newly formed barbed ends. We next directly visualized cofilin-binding to growing actin filaments using two-color TIRF microscopy with Alexa-Fluor-633-labeled actin and EGFP-tagged cofilin: the latter having been previously shown to have similar activities in vitro to the untagged protein (Lai et al., 2008). Comparison of the severing efficacies revealed that both untagged and EGFP-tagged cofilin are equally effective (Fig. 2B). In order to further confirm the functionality of EGFP-tagged cofilin, we performed fluorescence titrations with pyrene-labeled F- and G-actin and found that both cofilin variants exhibited comparable affinities for both actin species (supplementary material Fig. S2). Thus, EGFP-tagged cofilin is suitable to directly visualize actin-filament binding and severing. Two-color TIRF microscopy revealed that decoration of actin filaments with EGFP–cofilin was barely detectable in a region up to ~2.8 μm behind the growing barbed end, whereas it markedly decorated further proximal, aged actin filaments (Fig. 2B; supplementary material Fig. S3A and Movie 5), consistent with recent work using an Alexa-Fluor-488-labeled yeast cofilin mutant (Suarez et al., 2011). Interestingly, filament decoration with EGFP–cofilin occurred in a discontinuous fashion, indicating that side binding of cofilin to F-actin is a cooperative process, as previously proposed (supplementary material Fig. S3B) (Cao et al., 2006; De La Cruz, 2005; Ressad et al., 1998). Moreover, the simultaneous imaging of growing filaments and EGFP–cofilin allowed us to analyze the time-course of cofilin binding. Because actin filaments grew at 11 subunits/second in our assay, we calculated that EGFP–cofilin preferably binds actin filaments older than ~2 minutes, which is in line with a slow rate of phosphate release from ADP+P_i actin in the filament (Carlier et al., 1988), as well as with the selectivity of cofilin for binding to ADP–F-actin (Maciver et al., 1991; Michelot et al., 2007). However, sporadic EGFP signals could also be detected within the distal 2.8 μm of growing filaments (supplementary material Fig. S3). By analyzing the fragment lengths of cofilin- and EGFP–cofilin-severed filaments, we found that severing was most effective in a region of ~4.5–5 μm behind the growing barbed end. Importantly, the lengths of severed barbed-end fragments in the presence of 300 nM EGFP-tagged and untagged cofilin were virtually identical, affirming that the EGFP tag does not noticeably interfere with F-actin binding or concomitant severing. The majority of severed fragments were 3–10 μm in length, and several EGFP–cofilin decorated filaments where not severed at all during the experiments (Fig. 2D; supplementary material Movie 5). These findings are in good agreement with recent observations (Suarez et al., 2011) and corroborate a model predicting that cofilin severing activity is maximal when only parts of the filament are bound by cofilin, whereas a complete decoration results in decreased filament severing (De La Cruz and Sept, 2010).

To test whether the selectivity of cofilin for aged actin filaments persisted in polymerizing, fascin-crosslinked bundles and whether and how these bundles can be efficiently severed by cofilin, we first initiated the polymerization of actin in the presence of fascin and allowed the reaction to proceed until several growing bundles were formed and their barbed ends identified. Subsequently, cofilin was added to the reaction mixture, resulting in rapid disassembly of aged bundles, whereas short, bundled barbed-end fragments persisted (Fig. 2C,D; supplementary material Movies 6 and 7). In addition, the majority of bundled barbed-end fragments formed immediately after cofilin addition was only ~2 μm in length, and hence significantly shorter on average than the barbed end fragments produced by stochastic cofilin severing of individual actin filaments in the absence of fascin (Fig. 2D). The lengths of the severed barbed-end fragments in the presence of fascin correlated reasonably well with the lengths of filament fragments not decorated by cofilin in the absence of fascin (Fig. 2B,D), suggesting that filament severing is highly efficient in fascincrosslinked bundles and that it occurs immediately after cofilin binding. Most importantly, the generated bundled barbed ends continuously increased in fluorescence intensity, indicative of severing and continuous growth of barbed ends formed in these conditions (Fig. 2C,E; supplementary material Movies 6 and 7). However, the fact that the polarity of these bundles still persisted suggests that fascin crosslinking is not directly affected by cofilin. Notably, the bundles did not show a uniform increase in fluorescence, but displayed a varying but characteristic fluorescence distribution along their length: although the first 1–2 μm at the barbed end of the growing bundle contained only few filaments, upon further polymerization, the number of filaments increased substantially in a region 2–8 μm behind the bundle tips. Furthermore, the fluorescence intensity in the proximal ends of the bundles slightly decreased again with increasing amounts of bound cofilin (supplementary material Movie 6). Colocalization of EGFP–cofilin colocalization of EGFP–cofilin and Alexa-Fluor-633–actin confirmed that cofilin preferentially binds to aged F-actin within the bundle, as is the case for individual filaments (Fig. 2B), because it starts to accumulate in a region ~2–3 μm behind the growing bundle tip (supplementary material Fig. S4 and Movie 8). Most notably, the maximal fluorescence of actin and cofilin within fascincrosslinked bundles did not correlate directly, but instead the peak of cofilin fluorescence lagged behind that of actin by ~3 μm (supplementary material Fig. S4 and Movie 8). Thus, the data indicate that low amounts of cofilin rapidly sever actin filaments upon initial binding, resulting in formation of new, growing barbed ends within the bundle in these conditions, and that increasing amounts of cofilin in the rear of the bundle eventually promote its disassembly (supplementary material Movies 7 and 8). Subsequent analyses revealed that the prominent bundle thickening depended on the cofilin concentration and that robust thickening of fascin-crosslinked bundles even occurred at concentrations as low as 50–200 nM, which is not sufficient to efficiently sever single actin filaments in the absence of fascin in our assays (Fig. 2F; supplementary material Fig. S5A; and data not shown). Addition of 500 nM cofilin enhanced the calculated maximal number of filaments per bundle from between five and ten filaments to more than 100 as estimated by fluorescence intensity measurements (Fig. 2F). We subsequently analyzed the time course of cofilin binding to pyrene-labeled F-actin crosslinked by different amounts of fascin to analyze whether cofilin binding to crosslinked filaments might be impaired. Binding of cofilin to F-actin causes a decrease in pyrene fluorescence, providing a means of analysing the binding kinetics (Blanchoin and Pollard, 1999). The rate of cofilin-binding indeed decreased with increasing amounts of fascin in the bundling reaction (Fig. 2G). However, the overall change in fluorescence was only modestly lowered at increasing fascin concentrations compared with non-crosslinked filaments, suggesting that the slower binding kinetics result from impaired diffusion into the actin bundles or lower association rates of cofilin to fascin-decorated filaments rather than direct competition of fascin and cofilin for F-actin binding sites. Consistently, low-speed sedimentation assays revealed that the amount of fascin in F-actin bundles is not reduced by cofilin (supplementary material Fig. S5B). Taken together, these results suggest that cofilin severing activity is greatly enhanced in fascin-crosslinked bundles despite lower binding kinetics of cofilin when compared with individual filaments in the absence of fascin.

We reasoned that the increased actin fluorescence within growing bundles in the TIRF polymerization assays resulted from the formation of new barbed ends generated by cofilin severing, and that severing might be enhanced in fascin-crosslinked bundles when compared with single filaments. To test for barbed-end formation in an alternative, more quantitative fashion, pyrene assays were employed. Cofilin was previously shown to enhance actin polymerization in these assays (Carlier et al., 1997), possibly by filament severing, thereby creating new barbed ends for subsequent elongation (Carlier et al., 1997; Du and Frieden, 1998; Gandhi et al., 2009). Cofilin increased the rate of actin polymerization in a concentration-dependent manner, and 600 nM cofilin enhanced the rate of the assembly of 4 μM G-actin into F-actin approximately threefold (Fig. 3A,C). Fascin alone had virtually no effect on spontaneous actin assembly (Fig. 3B). However, the presence of fascin markedly enhanced the rate of cofilin-mediated actin polymerization, by as much as sevenfold, supporting our previous hypothesis of enhanced severing and barbed-end formation by cofilin in fascin-crosslinked bundles (Fig. 3B,C).

To evaluate filament depolymerization by cofilin, we monitored the spontaneous disassembly of 100 nM F-actin in KMEI buffer by measuring the decrease in pyrene fluorescence upon dilution. We found that the depolymerization rate was enhanced by cofilin alone in a concentration-dependent manner, by as much as fivefold (Fig. 3D–F). These effects of cofilin on spontaneous filament disassembly are in good agreement with results obtained in previous studies (Moseley et al., 2006; Gandhi et al., 2009). In contrast to non-bundled actin filaments, fascincrosslinked filaments depolymerized much slower (Fig. 3D–F). The fascin-dependent decrease in depolymerization rates is most probably caused by crosslinking of actin subunits and thus impaired monomer release from filament ends (Fig. 3D,E,G) (Schmoller et al., 2011). However, cofilin enhanced the depolymerization rate of fascin-bundled filaments in a concentration-dependent manner, by as much as 35-fold when compared with the respective controls without cofilin (Fig. 3F). This suggests that cofilin partially counteracted the inhibitory effect of fascin on actin disassembly by creating new filament ends in fascin-crosslinked bundles more efficiently than in isolated filaments (Fig. 3G). Because the rates of depolymerization, in contrast to polymerization, depends on both the number of barbed ends and the amount of crosslinking protein within the bundle, the enhancement of depolymerization rates by cofilin did not simply increase with higher fascin concentrations. Instead, it was maximal at a concentration of 20 nM fascin in the reaction, but decreased again at higher fascin concentrations (Fig. 3F). Notwithstanding this, the enhancement of depolymerization of fascin-bundled actin filaments by cofilin was higher when compared with non-bundled filaments for the entire range of fascin concentrations tested (Fig. 3F). Remarkably therefore, in spite of reduced cofilin binding kinetics in the presence of fascin (Fig. 2E), the latter was capable of enhancing cofilin activity to effect actin filament severing and disassembly in this assay.

Previous in vivo studies in B16-F1 and MTLn3 cells have shown that cofilin readily associates with the entire lamellipodium (Lai et al., 2008), and it probably modulates lamellipodium architecture and treadmilling by severing and depolymerizing aged actin filaments (Svitkina and Borisy, 1999; Iwasa and Mullins, 2007; Bugyi and Carlier, 2010). Another model suggests a cooperation of cofilin-induced barbed-end generation and Arp2/3-complex-mediated nucleation to drive lamellipodial actin assembly (Oser and Condeelis, 2009). Irrespective of the precise molecular mechanism, localization and functional interference studies commonly suggest a positive regulatory function for cofilin in driving lamellipodium protrusion and membrane ruffling (Mouneimne et al., 2004; Hotulainen et al., 2005; Iwasa and Mullins, 2007; Lai et al., 2008).

Inspired by our in vitro observations, we decided to reinvestigate the localization and dynamics of cofilin in migrating melanoma cells and fibroblasts. A polyclonal antibody was employed to determine the subcellular localization of endogenous cofilin in B16-F1 cells by indirect immunofluorescence. Consistent with previous findings, cofilin was found to accumulate in the lamellipodial actin network of migrating cells, indicating specificity of the antibody (Fig. 4A). Notably, however, cofilin was also found to be markedly enriched at the tips of some filopodia, while it was clearly absent from others (Fig. 4A). The localization of endogenous cofilin to filopodial tips was also confirmed in fixed T10_1/2 fibroblasts (Fig. 4B), suggesting that cofilin could affect filopodia dynamics in different cell types. In order to explore in more detail the puzzling finding that cofilin accumulates only in a subpopulation of filopodia, we reinvestigated the dynamics of fluorescently tagged cofilin in B16-F1 mouse melanoma cells. Interestingly, we found that mCherry–cofilin was weakly associated with the base of protruding filopodia (Fig. 5A), which is in line with proposed models in which filopodium formation is regulated by actin assembly by proteins of the filopodium tip complex (FTP) and disassembly by cofilin in the rear (Mattila and Lappalainen, 2008; Faix et al., 2009; Hotulainen and Hoogenraad, 2010). However, mCherry–cofilin changed its localization pattern significantly upon cessation of filopodium protrusion, now more prominently localizing along the entire shaft and/or massively accumulating in the filopodium tip (Fig. 5A; supplementary material Movie 9). Most notably, the localization of cofilin to the filopodium tip was exclusively detected in retracting filopodia, whereas the tips of protruding filopodia were virtually devoid of cofilin (Fig. 5B). Identical results were obtained with EGFP-tagged cofilin used previously (data not shown) (Lai et al., 2008). Since our in vitro results indicated a functional cooperation between cofilin and fascin, a well-established filopodial regulator, we also compared the dynamics of both proteins in vivo. To do this, we coexpressed EGFP-tagged fascin and mCherry–cofilin to analyze their relative dynamics in protruding and retracting filopodia. As expected, fascin was highly enriched in the entire shaft of protruding filopodia, whereas cofilin was, at best, weakly associated with the rear of these filopodia (Fig. 6; supplementary material Movie 10). However, upon cessation of protrusion, cofilin and fascin briefly colocalized in the shaft, followed by a decrease in fascin and overall increase in cofilin intensities, frequently culminating in strong accumulation of the latter in retracting filopodia tips (Fig. 6; supplementary material Movie 11). These data are consistent with the absence of cofilin from freshly polymerized barbed ends of filaments bundled by fascin in vitro (supplementary material Movie 8), and identify cofilin as a potential decisive factor in actin filament disassembly causing retraction of filopodial bundles in vivo. Consistent with this view, comparison of actin (mCherry) and cofilin (EGFP) dynamics in the same cells revealed that cofilin accumulation in filopodia upon cessation of protrusion and/or retraction directly correlates with a reduction of actin intensities, as expected if cofilin was to operate in actin filament disassembly at these sites (Fig. 7; supplementary material Movie 12). To study this directly, we used correlated live cell imaging and electron tomography to analyze the structure of filopodia captured by fixation in various phases of protrusion or withdrawal. Frequently, filopodia and microspikes in B16-F1 melanoma cells tagged with EGFP–fascin, performed translational and folding movements, including fragmentation and entry into the lamella, as previously described (Nemethova et al., 2008: supplementary material Movie 13). Additionally, however, filopodia labeled with cofilin showed characteristic retraction and kinking motions (Fig. 8A). Electron tomography revealed a correlation of the cofilin label with a marked fragmentation of actin filaments within the filopodium shaft, including the tip region (Fig. 8C,D). By contrast, neighboring filopodia in phases of protrusion that typically lacked the cofilin label showed intact filaments, arranged in parallel from base to tip (Fig. 8B).

Cofilin was previously shown to cooperatively bind ADP–actin filaments, thereby enhancing their flexibility and the twist of the filament that eventually leads to filament breaking (McGough et al., 1997; Ressad et al., 1998; Blanchoin and Pollard, 1999; Pavlov et al., 2007; McCullough et al., 2008; Suarez ez al., 2011). However, the effects of actin crosslinking proteins on cofilin activity have not been investigated.

By employing in vitro TIRF microscopy, we were able to show, for the first time, the dynamics of fascin-meditated fusion of single filaments into bundles, and directly demonstrate that fascin-mediated actin filament bundling is rapid, zippering together actin filaments at rates of up to 1000 subunits/second. We also confirmed that bundles formed by fascin are parallel, as reported previously (Ishikawa et al., 2003). Because the rate of spontaneous actin assembly into filaments is approximately two orders of magnitude slower than fascin-mediated bundling, we propose that filaments that grow in fascin–actin bundles are crosslinked immediately in vitro and probably also in vivo. We also used this in vitro assay to analyze the effects of cofilin on polymerizing, fascin-crosslinked actin bundles. Surprisingly, although we expected to observe reduced activities of cofilin on bundled filaments, as previously reported for spontaneously formed bundles (Michelot et al., 2007), we found that filament severing was greatly enhanced within these bundles. This was true in spite of slower binding kinetics of cofilin to fascin-bundled actin filaments as compared with isolated actin filaments, resulting in a dramatic increase in the number of barbed ends upon initial binding of cofilin. The enhancement of filament severing was proportional to the amount of both fascin and cofilin in the reaction mixture. Most notably, cofilin already efficiently severed filaments in bundles at concentrations that were not sufficient to sever individual actin filaments. How can these results be explained? It was proposed that the severing mechanism of cofilin relies on local changes of the helical twist of the filament upon cofilin binding. The F-actin twist was shown to be variable within a given filament: while the largest part of the filament twists by 167°, some segments twist only by 162° (Galkin et al., 2001; Galkin et al., 2003). Binding of cofilin to the latter shifts this equilibrium and induces 162° twists also in adjacent segments, which in turn promotes cooperative binding of additional cofilin molecules. However, efficient severing occurs predominantly at the borders of decorated and non-decorated segments of the actin filament (McGough et al., 1997; Prochniewicz et al., 2005; McCullough et al., 2008; De La Cruz and Sept, 2010, Suarez et al., 2011), so rapid decoration of the filament with cofilin will reduce severing frequency.

Based on these previous findings, two, not necessarily exclusive, mechanisms could account for the observed enhancement of cofilin severing by fascin. Crosslinking probably reduces the flexibility of the actin filament within the bundle. As a consequence, an actin filament segment cannot compensate for cofilin-induced twisting by relaxation of the entire filament (Fig. 9A). Thus, the mechanical stress generated upon cofilin binding would be restricted to the region between the crosslinkers, leading to rapid filament breaking (Fig. 9B). Alternatively, or in addition, fascin-mediated crosslinking could also counteract cooperative binding by cofilin (Fig. 9B). In this scenario, fascin crosslinking would prevent twisting of adjacent segments of the filament upon cofilin binding, leading to non-uniform decoration of the filament with cofilin, accompanied by more efficient severing (De La Cruz and Sept, 2010). The latter scenario is in line with our observation that cofilin binds to fascin-crosslinked bundles with slower kinetics than to isolated filaments. Moreover, fascin was recently shown to induce overtwisting of actin filaments upon bundling (Claessens et al., 2008). This effect might additionally account for an enhanced severing activity of cofilin due to larger differences in the twists of cofilin-bound compared with fascin-bound actin filament segments. Irrespective of this, the efficiency of cofilin severing was enhanced by increasing concentrations of fascin in the reaction, despite reduced cofilin binding kinetics, and filament severing was already observable at concentrations of cofilin too low to induce marked severing of isolated filaments, which is consistent with both models.

It is well established that cofilin regulates the protrusion of lamellipodia, in which it is strongly enriched (Lai et al., 2008; Oser and Condeelis, 2009). Also, its knockdown was observed to reduce the actin filament turnover in lamellipodia and other actin structures such as stress fibers (Hotulainen et al., 2005). This effect was interpreted as F-actin stabilization due to reduced depolymerization, and hence depletion of the actin monomer pool (Kiuchi et al., 2007).

We show here, for the first time, that cofilin is also highly enriched in the tip region of retracting filopodia and that specific localization patterns of fascin and cofilin are directly correlated with filopodial protrusion, stalling and retraction. Interestingly, the transient association of cofilin with retracting rather than protruding filopodia suggests it specifically operates in promoting the severing and disassembly of actin filaments in these structures, as in lamellipodia (Hotulainen et al., 2005; Lai et al., 2008). This notion is supported by our analyses of actin filament organization in filopodia at different stages of protrusion and retraction using a combination of live cell imaging and electron tomography. Our experiments revealed compact bundles of actin filaments in filopodia during protrusion, as expected. However, filopodia at different stages of retraction had regions of fragmented actin filaments that precisely correlated with the accumulation of cofilin, established by preceding live-cell imaging. This criss-cross arrangement of short filaments is reminiscent of the actin organization reported for Dictyostelium filopodia, presumably fixed during their retraction (Medalia et al., 2007), indicating that this mechanism of filopodium retraction might have been conserved throughout evolution. Furthermore, enhancement of cofilin-induced severing by fascin, as observed in vitro, would be consistent with a transient colocalization of both proteins in filopodia (Figs 6 and 8C). However, the frequently observed strong accumulation of cofilin in the tip region of the filopodium–a region assumed to largely contain filament barbed ends–is striking. The occurrence of rearward moving cofilin puncta within filopodia (supplementary material Movies 9 and 10) indicates continuous actin polymerization in the filopodium tip even when net elongation stalls, suggesting that cofilin might bind to newly polymerized actin filaments. Taking into account that cofilin preferentially binds to ADP–F-actin, it is possible that phosphate release is enhanced by components of the filopodium tip complex. Importantly, proteins of the formin and Ena/VASP family, which are crucial components of the filopodium tip complex (Schirenbeck et al., 2005, Kwiatkowski et al., 2007; Breitsprecher et al., 2008 ) were suggested to mediate actin filament elongation by coupling ATP hydrolysis and phosphate release (Dickinson and Purich, 2002; Dickinson et al., 2004; Romero et al., 2004; Romero et al., 2007). Thus, these proteins might not only be important players in the initiation and elongation of filopodia but could also operate in the regulation of filopodium length and lifetime by providing ADP–F-actin as a depolymerizable substrate for cofilin; a hypothesis deserving future investigation. However, it should be emphasized that proper, subcellular cofilin positioning must rely on additional components, because in these cells cofilin is actively kept away from actin filaments in structures other than lamellipodia and retracting filopodia, so the preference for ADP–actin filaments might be one of many aspects contributing to cofilin localization and dynamics in vivo.

In contrast to promoting actin filament disassembly, cofilin has also been proposed to drive the assembly of actin filaments, at high concentrations both in vitro (Andrianantoandro and Pollard, 2006) and in vivo (Oser et al., 2009), although our own results comparing actin and cofilin dynamics in the lamellipodia of different cell types were not confirmatory of this view (Lai et al., 2008). However, such a function of cofilin cannot be excluded for other subcellular locations, e.g. invadopodia and matrix-degrading adhesive structures employed by cancer cells during metastasis. Interestingly, both fascin and cofilin are key components of invadopodia and are crucial for matrix invasion. Both proteins localize to the actin-rich region of these structures, and the depletion of either fascin or cofilin results in decreased invadopodia lifetimes and lower actin levels (Yamaguchi et al., 2005; Oser et al., 2009; Li et al., 2010). Considering the cooperation between cofilin and fascin in filament severing observed in vitro, it is tempting to speculate that their combined activities might result in distinct outcomes in different conditions or subcellular structures, e.g. driving actin polymerization in invadopodia while disassembling actin filaments in filopodia. Our data add to the list of functions established for cofilin in live cells, by demonstrating its transient engagement in filopodia retraction.

Murine non-muscle cofilin-1 (also known as n-cofilin) and human EGFP-tagged n-cofilin were purified as described previously (Lai et al., 2008). mCherry–actin (Koestler et al., 2008), EGFP–fascin (Adams and Schwartz, 2000) and EGFP–cofilin (Mannherz et al., 2005) were described previously. mCherry–cofilin was produced by excision from pEGFP–cofilin with NheI and BsrGI and ligation into the mCherry vector at the corresponding sites. Human fascin was amplified from a human macrophage cDNA library (Invitrogen) and inserted into the BamHI and XhoI sites of pGEX-6P1. Briefly, GST-tagged cofilins and GST-tagged fascin were expressed in Escherichia coli strain Rosetta (Promega) and purified from bacterial extracts on glutathione-conjugated agarose (Sigma-Aldrich, Germany) using standard procedures. The GST tags were cleaved by incubating the purified fusion proteins with PreScission protease (GE Heathcare) in phosphate-buffered saline (PBS), pH 7.3, supplemented with 1 mM dithiothreitol (DTT) and 1 mM EDTA overnight at 4°C. After cleavage, the GST tags were removed by gel filtration on a 26/60 Superdex G75 column (GE Healthcare). Cofilin- or fascin-containing fractions were pooled, dialyzed against 200 mM KCl, 1 mM DTT, 60% glycerol and 20 mM imidazole (pH 7.4) and stored at −23°C. Ca²⁺-ATP–actin was purified from rabbit skeletal muscle according to the method of Spudich and Watt (Spudich and Watt, 1971), stored in G-buffer (5 mM Tris-HCl pH 8.0 and 0.2 mM CaCl₂, 0.5 mM DTT, 0.2 mM ATP) and prepared for time-lapse total internal reflection fluorescence microscopy (TIRFM) and pyrene assays as described previously (Kuhn and Pollard, 2005; Schirenbeck et al., 2006). For two-color TIRF experiments, rabbit muscle actin was labeled on Cys374 with Alexa-Fluor-633–maleimide (Invitrogen). Protein concentrations were determined by absorption spectroscopy using extinction coefficients predicted from the amino acid sequences using Vector NTI software (Invitrogen).

TIRFM with Alexa-Fluor-488-labeled actin was performed essentially as described previously (Breitsprecher et al., 2011). Images from an Olympus IX-81 inverted microscope were captured every 5 or 10 seconds using a Hamamatsu Orca-R2 CCD camera (Hamamatsu Corp., Bridgewater, NJ). For two-color TIRFM experiments, Alexa-Fluor-633-labeled actin was excited at 632 nm with a 30 mW HeNe laser (DS Uniphase Corp., Milpitas, CA) with exposure times of 300 mseconds, and EGFP–cofilin was excited at 488 nm with a 30 mW DPSS laser (Novalux, Sunnyvale, CA) with exposure times of 150 mseconds. The specimens were excited successively with time intervals <20 mseconds between the excitations to minimize time-shifts. The pixel size corresponded to 0.11 μm.

Reactions in TIRF assays contained 1.3 μM actin (labeled with 30% Alexa-Fluor-488 or -633). Before the experiments, Ca²⁺-ATP–actin was converted to Mg²⁺-ATP–actin by addition of 10× magnesium-exchange buffer (1 mM MgCl₂, 10 mM EGTA, pH 7.4). Polymerization experiments were performed in TIRF buffer [10 mM imidazole, 50 mM KCl, 1 mM MgCl₂, 1 mM EGTA, 0.2 mM ATP, 10 mM DTT, 15 mM glucose, 20 μg/ml catalase, 100 μg/ml glucose oxidase, and 0.5% methylcellulose (4000 cP), pH 7.4]. The fluorescence intensity of bundles was analyzed using the line-scan tool in ImageJ software (http://rsbweb.nih.gov/ij/), and subsequently normalized to the fluorescence intensity of a single actin filament from the same experiment to calculate the number of filaments per bundle. Small filaments transiently interacting with the surface before their rapid disappearance were not included in the analyses.

Before experiments, Ca²⁺-ATP–actin was converted to Mg²⁺-ATP–actin by addition of 10× Mg-exchange buffer. For assembly assays, dilutions of proteins to be assayed were prepared in protein storage buffer (200 mM KCl, 20 mM Hepes, 1 mM DTT, pH 7.3) and 10× low salt KMEI buffer (250 mM KCl, 10 mM MgCl₂, 10 mM EGTA, 100 mM imidazole, pH 7.3) was added. Anti-foam 204 (Sigma) was added to the mixture to reach a final concentration of 0.005%. 180 μl aliquots were placed in an 8-well microtiter assembly strip (Thermo Scientific). 18 μl of a 40 μM solution of 10% pyrene-labeled G-actin (in 2 mM Tris-HCl, pH 8.0, 0.2 mM ATP, 0.1 mM CaCl₂, 0.5 mM DTT) were placed in another 8-well microtiter assembly strip. The assembly reaction was initiated by transferring 162 μl of the protein solution to 18 μl of pyrene-labeled actin. The polymerization of actin was followed by measuring the fluorescence increase of pyrene-actin (excitation at 364 nm and emission at 407 nm) in a fluorescence plate reader for at least 1500 seconds.

Spontaneous depolymerization of F-actin was analyzed by diluting 10 μl of a 1 μM F-actin solution in 1× KMEI buffer (50 mM KCl, 1 mM MgCl₂, 1 mM EGTA, 10 mM imidazole, pH 7.3) crosslinked by 500 nM fascin into 90 μl 1× KMEI buffer containing the cofilin concentration as indicated. The decrease of pyrene fluorescence was monitored with a Jasco FP-6500 fluorimeter (excitation at 364 nm and emission at 407 nm), and the relative depolymerization rate was derived from the initial slope of the depolymerization reaction. Cofilin-binding to bundled F-actin was analyzed by adding 10 μl of a solution of 10 μM F-actin crosslinked with 0.15–10 μM fascin to 90 μl of a solution of 1.11 μM cofilin in 1× KMEI buffer. After rapid mixing, the decrease in pyrene fluorescence was monitored with a Jasco FP-6500 fluorimeter.

G-actin (5 μM) were polymerized in 1× KMEI buffer in the presence of fascin and cofilin concentrations as indicated. Bundles were sedimented by centrifugation for 30 minutes at 11,000 g, and supernatants and pellets were analyzed by SDS-PAGE and Coomassie-Blue-staining. Densitometric analysis of band intensities was performed with ImageJ software (http://rsbweb.nih.gov/ij/).

Mouse embryonic Ch3T10_1/2 (T10_1/2) fibroblasts and B16-F1 mouse melanoma cells (ATTC CRL-6323) were grown in high glucose DMEM (Gibco) supplemented with 10% FCS (PAA Laboratories, Pasching, Austria) and 2 mM glutamine. Transfections were performed overnight using Superfect (Quiagen GmbH, Hilden, Germany) and FuGene 6 (Roche, Basel, Switzerland) according to manufacturers' instructions, respectively.

For indirect immunofluorescence microscopy, T10_1/2 fibroblasts were plated on uncoated glass coverslips, and B16-F1 cells were plated on acid-washed coverslips coated with 25 μg/ml laminin (Sigma). After fixation with 4% paraformaldehyde in PBS containing 2.7 mM KCl, 1.8 mM KH₂PO₄, 10 mM Na₂HPO₄, 140 mM NaCl, pH 7.3 for 15 minutes, the specimens were treated with PBS containing 100 mM glycine and 0.1% Triton X-100 for 15 minutes. After washing five times with PBS containing 0.05% (v/v) cold fish gelatin (Sigma) and 0.5% (w/v) bovine serum albumin (Sigma), the cells were incubated with a polyclonal anti-n-cofilin antibody (Dianova, Pinole, CA) at a dilution of 1:1000 in the same buffer overnight and then incubated with Alexa-Fluor-488-conjugated goat anti-rabbit IgG (Molecular Probes). F-actin was visualized with tetramethylrhodamine B isothiocyanate (TRITC)-conjugated phalloidin (Sigma). Confocal fluorescence images were recorded in multi-track mode with a Zeiss LSM 510 confocal microscope, equipped with a 63/1.3 Plan-Neofluar objective. Data were processed with Adobe Photoshop and CorelDraw software.

For live-cell imaging, B16-F1 cells were plated on acid-washed coverslips coated with 25 μg/ml laminin (Sigma), mounted in an open heating chamber (Warner Instruments, Reading, UK), and observed at 37°C with a Zeiss Axiovert S100TV inverted microscope (Carl Zeiss, Jena, Germany) equipped with a 63×/1.4 NA plan-apochromatic objective, and a cooled CCD camera (Coolsnap HQ2, Photometrics, Roper Scientific, Germany) with a filter wheel and shutters controlled by Metamorph software (Molecular Devices).

Correlated live-cell imaging and electron tomography were performed essentially as described previously (Auinger and Small, 2008; Urban et al., 2010). Briefly, B16 melanoma cells expressing mCherry–fascin and EGFP–cofilin were plated onto coverslips coated with a Formvar film embossed with grid patterns for cell relocation. Selected cells were imaged at 37°C in an open, temperature-controlled chamber using a 100×, 1.4 NA phase-contrast objective on an inverted Zeiss Observer fluorescence microscope equipped with an LED light source (CoolLed, Andover, UK) and a rear-illuminated cooled CCD Micromax camera (Roper Scientific). After recording paired images in two fluorescent channels every 15 seconds, the growth medium in the chamber was rapidly aspirated and replaced with the fixative mixture containing 0.5% Triton X-100 and 0.25% glutaraldehyde in cytoskeleton buffer (10 mM MES, 150 mM NaCl, 5 mM EGTA, 5 mM glucose, 5 mM MgCl₂, pH 6.1). After 1 minute, the fixative mixture was exchanged for 2% glutaraldehyde in the same buffer supplemented with 10 μg/ml phalloidin and stored at 4°C in this solution until use. For electron microscopy, the Formvar film was peeled from the coverslip under cytoskeleton buffer, inverted onto the buffer surface and a 50 mesh EM grid positioned onto the region of the film containing the imaged cell using forceps mounted in a micromanipulator. The grids were stained with 7% sodium silicotungstate containing a gold sol, as described previously (Urban et al., 2010) and a tomogram tilt series recorded in an FEI Polara electron microscope at 300 kV using primary magnifications from 23–31,000× at a defocus of 5 μm. Tomograms were generated using IMOD software (Kremer et al., 1996).

We thank Annette Breskott and Brigitte Denker for excellent technical assistance. We also thank Josephine Adams (Cleveland, USA), Hans Georg Mannherz (Bochum, Germany) and Elena Korenbaum for kindly providing expression constructs and T10_1/2 cells.