Self-Assembly of Filopodia-Like Structures on Supported Lipid Bilayers

See allHide authors and affiliations

Science  10 Sep 2010:
Vol. 329, Issue 5997, pp. 1341-1345
DOI: 10.1126/science.1191710


Filopodia are finger-like protrusive structures, containing actin bundles. By incubating frog egg extracts with supported lipid bilayers containing phosphatidylinositol 4,5 bisphosphate, we have reconstituted the assembly of filopodia-like structures (FLSs). The actin assembles into parallel bundles, and known filopodial components localize to the tip and shaft. The filopodia tip complexes self-organize—they are not templated by preexisting membrane microdomains. The F-BAR domain protein toca-1 recruits N-WASP, followed by the Arp2/3 complex and actin. Elongation proteins, Diaphanous-related formin, VASP, and fascin are recruited subsequently. Although the Arp2/3 complex is required for FLS initiation, it is not essential for elongation, which involves formins. We propose that filopodia form via clustering of Arp2/3 complex activators, self-assembly of filopodial tip complexes on the membrane, and outgrowth of actin bundles.

Actin assembly is largely responsible for cell shape and movement in eukaryotic cells (1). Actin filaments underlie a range of morphological structures; among these, filopodia, which contain closely apposed bundled parallel actin filaments, are of particular interest (2, 3). They have numerous biological roles, but their mechanism of formation is poorly understood. The persistence of actin growth in a parallel bundle in filopodia suggests roles for proteins with anti-capping and/or processive elongation activity [such as vasodilator-stimulated phosphoprotein (VASP) and formins] (48). But formins are also important in lamellipodia (9). The role of the Neural Wiskott-Aldrich syndrome protein (N-WASP)/Arp2/3 complex pathway in filopodia formation is also contentious (8, 1014). Although there are in vitro models of actin bundling (1517), no current experimental system recapitulates two physiological aspects of filopodia: (i) spontaneous formation of actin bundles in the presence of capping activity and (ii) the assembly of a membrane-localized tip complex.

To investigate actin polymerization at membranes and to exploit advanced microscopic techniques like total internal reflection fluorescence (TIRF) microscopy, we have replaced liposomes as the source of lipids with supported lipid bilayers (18). Purified Cdc42.GTPγS, N-WASP-WIP, toca-1, Arp2/3 complex, and actin comprise a minimal set of proteins for stimulation of actin nucleation by PI(4,5)P2-containing liposomes in vitro (19). When these proteins are supplied to the supported bilayer, a thin actin layer is formed at the membrane surface (Fig. 1, A and B). When we substituted concentrated frog egg extracts for the purified proteins, unusual dense, focal, and long actin structures with a diameter of 0.3 to 1.5 μm rose from the surface of the bilayer (Fig. 1, C to E, fig. S1A, and movies S1 to S4). We term these filopodia-like structures (FLSs).

Fig. 1

Reconstitution of filopodia-like structures on supported lipid bilayers containing 45% PC, 45% PI, or 10% PI(4,5)P2. (A) Confocal microscopy of supported bilayers with Cdc42, N-WASP-WIP, toca-1, Arp2/3 complex, and actin generates uniform, short polymerized actin. (B) Z-stack reconstruction of (A) showing growth of actin in the z axis. (C) Confocal microscopy of supported lipid bilayers with Xenopus egg extracts and Alexa 647-actin (seen along the x-y plane) shows growth of focal actin structures from the bilayer. (D) Z-stack reconstruction looking at the x-z plane shows the height of the actin structures. (E) Three-dimensional reconstruction. Scale bars, 5 μm. (F to G) Negative-stain EM of the phalloidin-stabilized actin structures shows that these are made of bundled, unbranched actin filaments. Scale bars, 100 nm. (H) Time-lapse sequence of FLS formation seen. Scale bars, 2 μm. (I and J) Pulse-chase experiments show that actin polymerization occurs at membrane. The reaction is started with Alexa 647-actin (red) and chased by Alexa 488-actin (green). Scale bars, 2 μm. (I) 1 min timepoint. (J) 2 min 40 s timepoint. (K) Immunostained FLSs from the side: green, fascin immunostain; red, Alexa 568-phalloidin. (L) 60° tilt from the x-y plane: green, Drf1 immunostain; red, Alexa 568-phalloidin. (M) 60° tilt from the x-y plane: green, GFP-VASP; red, Alexa 647-actin. (N) 60° tilt from the x-y plane: green, profilin immunostain; red, Alexa 568-phalloidin. (O) Top view: red, mCherry-Cdc42; green, Alexa 488-actin. (P) 60° tilt from the x-y plane: green, N-WASP immunostain; red, Alexa 568-phalloidin. (Q) 60° tilt from the x-y plane: green, GFP-toca-1; red, Alexa 647-actin. (R) Side view: red, Alexa 568–Arp2/3 complex; green, Alexa 647-actin. Scale bars, 2 μm.

By means of electron microscopy (EM) using negative stain, we observe bundled actin filaments in the FLSs, which are characteristic of filopodia and distinct from the dendritic networks made by Listeria or ActA beads (Fig. 1, F and G). The parallel alignment of actin filaments is revealed when smaller FLSs spread out two-dimensionally (Fig. 1G). We observed actin bundles of the complete size range seen by light microscopy (fig. S1B). Prefixing the FLSs produces similar actin bundles (fig. S1C).

If FLSs recapitulate filopodia, we would predict that new actin monomers would be added at the membrane-localized tip with dynamics similar to filopodial growth in vivo (20, 21). Time-lapse confocal imaging and z-stack reconstructions show that the typical initial rate of FLS growth is 2.5 μm/min, which is within the range of filopodia (Fig. 1H and movies S1 to S4). A pulse-chase experiment starting actin growth with Alexa 647-actin (Invitrogen, Carlsbad, CA) and adding Alexa 488-actin after 20 min shows that actin polymerization occurs at the membrane-localized tip (Fig. 1, I and J, and fig. S1, D to G). Actin monomers are added at 2.8 μm/min (fig. S1H). Growth of the bundles into the membrane is prevented by the underlying glass support; hence, assembly is constrained to occur away from the membrane. With this difference, the assembly of bundled and parallel actin filaments at the membrane recapitulates filopodial formation and growth.

For filopodia in vivo, a distinct collection of proteins was observed by means of EM and termed the tip complex (5, 20). To determine whether such a structure is present in the FLSs, we immunostained for known filopodial components or added fluorescently tagged proteins. Like filopodia, FLSs immunostain for fascin along the length of the structure (Fig. 1K, and antibody specificity is shown in fig. S1I) (22). The tip of the FLSs contains proteins that localize to the tips of filopodia—diaphanous-related formins, VASP, profilin, and N-WASP (Fig. 1, L to N and P) (5, 8, 2326). Cdc42 and toca-1 also localize to the tip (Fig. 1, O and Q). Alexa 568–labeled Arp2/3 complex is enriched at the tip and also decorates the actin shaft (Fig. 1R). In some filopodia, Arp2/3 complex is excluded from the shaft, whereas in others it is present (5, 27).

Tip complex assembly requires negatively charged membranes because no FLSs form on glass or phosphatidylcholine (PC) bilayers alone. A direct comparision of PC/ phosphatidylserine (PS) and PC/PS/phosphatidylinositol 4,5-bisphosphate [PI(4,5)P2] membranes of equivalent net charge shows that there is specificity for PI(4,5)P2, which supports a 2.5-fold higher density of FLSs at steady state (20 min); their initial rate of appearance is also fivefold faster than with just PS (Fig. 2A, blue and orange). Substitution of PS for PI (phosphatidylinositol) leads to a twofold decrease in the rate of appearance, which may reflect the role of PS in binding some F-BAR proteins (Fig. 2A) (28). All phosphoinositides (PIPs) nucleate actin spots on the membrane (bis/tris-PIPs are shown in Fig. 2B, and monophosphorylated PIPs are shown in fig. S2A), although few of the actin spots nucleated by monophosphorylated PIPs elongate. Among bisphosphorylated PIPs and phosphatidylinositol 3,4,5-trisphosphate [PI(3,4,5)P3], PI(4,5)P2 gives the highest number and elongation rate (Fig. 2, B and C), which is consistent with filopodia forming on the plasma membrane. One caveat in comparing the different PIPs is that the extracts may convert one form into another.

Fig. 2

Membrane requirements for FLS formation. (A) Time course of FLS appearance shows that negatively charged lipids are essential for FLS formation and there is specificity for PI(4,5)P2 and PS. Blue, 60% PC/30% PS/10% PI(4,5)P2; purple, 60% PC/30% PI/10% PI(4,5)P2; orange, 40% PC/60% PS; red, 55% PC/45% PS; dark green, 70% PC/30% PS. (B) Time course of FLS appearance. All PIPs can nucleate FLSs, but there is preference for PI(4,5)P2. Compositions are 60% PC/30% PS/10% PIP, where lime indicates PI(4,5)P2, gray indicates PI(3,4)P2, purple indicates PI(3,5)P2, dark blue indicates PI(3,4,5)P3, and brown indicates PI. Data are the mean of three time courses normalized to the average number of structures from the three experiments from five or more pictures over each supported bilayer. (C) Rate of actin addition (using the pulse-chase approach) for the different PIPs shows PI(4,5)P2 specificity. Error bars are SD; n = 18, 19, 16, 20 FLSs, from left to right. (Four-way analysis of variance, P < 0.001). (D) Confocal microscopy of supported bilayer surface. GFP-PH domain (green) addition to supported bilayers including rhodamine-PE (red) shows membrane domains. Scale bar, 2 μm. The lipid composition was 45% PC, 45% PI, and 10% PI(4,5)P2, and similar data was obtained with 60% PC, 30% PS, and 10% PI(4,5)P2. (E) FLSs grow preferentially from rhodamine-PE–depleted (PH domain binding) regions. Alexa 647-actin is in green, and rhodamine-PE is in red. Scale bar, 2 μm. (F) Contour plot of the number of FLSs per field of view at steady state (20 min) from fluid membranes in response to PI(4,5)P2 and extract concentrations. The lipid composition was 45% PC, 45% PI, and 10% PI(4,5)P2. For comparison, overlaid single points show the number of FLSs formed from the gel phase. Data are plotted logarithmically. Example pictures and control data are in fig. S4.

The tendency of lipid mixtures to segregate into domains suggests two different models for tip assembly and FLS nucleation. In the first, the FLSs would be templated by small domains enriched in PI(4,5)P2. In this model, the size of the tips and the girth of the FLSs should reflect the size of the domains. In the second, tip complexes would self-assemble on the lipid bilayer in relatively homogeneous lipid domains. We found no evidence for preformed domains that template FLS formation because green fluorescent protein (GFP)–phospholipase C–δ pleckstrin homology domain (GFP-PH domain), which has specificity for PI(4,5)P2, binds evenly over large interconnected domains, and there is no enrichment of GFP-PH domains at sites of FLS growth (Fig. 2D and fig. S2, B to D). Furthermore, the kinetics of PH domain binding to the bilayer does not reveal hotspots of PI(4,5)P2 (fig. S2, E and F). Rhodamine–phosphatidylethanolamine (PE), which labels the liquid-disordered phase of membranes (29), has the inverse distribution to GFP-PH domain, confirming that lipid is present in regions depleted of the GFP-PH domain. This also indicates that either PI(4,5)P2 partitions differently than rhodamine-PE, or that the GFP-PH domain cannot bind PI(4,5)P2 within these domains (Fig. 2D). Domains of rhodamine-PE enrichment vary in size, with 53% less than 10 μm2, 38% 10 to 100 μm2, 7% 100 to 1000 μm2, and 2% >1000 μm2. Rhodamine-PE–enriched regions nucleate 95% fewer FLSs than the rhodamine-PE depleted regions (Fig. 2E). DiI, another fluorescent membrane marker that labels the disordered phase, also localizes to areas with fewer FLSs (fig. S3A). Fluorescence recovery after photobleaching (FRAP) experiments confirm that the rhodamine-PE–positive regions are fluid (fig. S3, B and C). The irregular boundaries between rhodamine-PE regions and GFP-PH domain regions also suggest the coexistence of the liquid-disordered (rhodamine-PE) and gel phases (GFP-PH domain) (30). FRAP of GFP-PH domain confirms the low fluidity of these regions (fig. S3D and E). Thus, FLS formation occurs preferentially but not exclusively in the gel phase. The tendency of FLS not to form from the liquid-disordered phase can be overcome by increasing the mole fraction of PI(4,5)P2 or by increasing the extract concentration (Fig. 2F and fig. S4), and therefore the gel phase is not an absolute prerequisite for FLS formation.

Thus, the assembly of FLSs does not occur by direct templating at preformed domains in the membrane but instead occurs by a process of self-assembly driven by proteins at a permissive membrane surface. Several mechanisms could contribute to FLS formation: (i) clustering of oligomeric activation proteins [such as Bin–Amphiphysin–Rvs (BAR) domain superfamily proteins] (31); (ii) positive feedback on small Arp2/3 complex–catalyzed clusters of polymerized actin through recruitment of more nucleation factors (32), such as occurs during symmetry-breaking by N-WASP on lipid-coated glass beads; (iii) lattices generated by cooperative protein-protein interactions between signaling molecules; (iv) cooperative association of proteins with PI(4,5)P2, as has previously been shown for N-WASP (33). Any small local fluctuations in PI(4,5)P2 could be magnified in such a mechanism.

To investigate the kinetics of the self-assembly process of the FLSs, we added fluorescently labeled filopodial proteins to the extracts and followed their recruitment to sites of FLS formation by using TIRF microscopy (Fig. 3 and fig. S5). The recruitment times are normalized to the recruitment time of labeled actin, with the time differences plotted as a histogram (Fig. 3). The membrane-binding, F-BAR domain protein toca-1 is recruited earliest, at sites that later go on to form FLSs (Fig. 3, A and G). After a variable time, N-WASP is recruited, again before actin (Fig. 3, B and G). The Arp2/3 complex is recruited concomitantly with actin (Fig. 3, C and G). VASP and mDia2 proteins, which are implicated in the formation of long, unbranched actin filaments, are recruited to the tip complex after the first appearance of actin (Fig. 3, D, E, and H). The bundling protein fascin is recruited last (Fig. 3, F and H).

Fig. 3

Kinetics of signaling protein recruitment to filopodia-like structures; Arp2/3 complex–signaling proteins are recruited before actin and formin, and VASP and fascin are recruited later. (A to F) Total internal reflection fluorescence images of FLS tips with fluorescently labeled proteins (GFP-toca-1, GFP-N-WASP, Alexa 568–Arp2/3 complex, GFP-mDia2, GFP-VASP, and GFP-fascin) are in green, and Alexa 647-actin is in red. Time intervals of 20 s are shown at the time of first recruitment of the signaling protein or actin. 5 nM–labeled proteins were added to the extracts. (G) Histogram showing the relative time of first recruitment of Arp2/3 complex (n = 48 FLSs), toca-1 (n = 26 FLSs), and N-WASP (n = 34 FLSs) compared with the first appearance of actin. (H) Histogram showing the relative time of first recruitment for filopodia elongation and bundling proteins, VASP (n = 34 FLSs), mDia2 (n = 35 FLSs), and fascin (n = 34 FLSs) compared with actin. Using the KS-test, P = 0.000 for toca-1-N-WASP, N-WASP-Arp2/3 complex, and Arp2/3 complex-VASP; P = 0.005 for Arp2/3 complex-mDia2; P = 0.003 for mDia2-fascin; and P = 0.004 for VASP-fascin and no significant difference for mDia2-VASP. The lipid composition was 60% PC, 30% PS, and 10% PI(4,5)P2.

These data suggest that an Arp2/3 complex–driven initiation step nucleates an initial branched actin structure in a small patch, which stimulates the recruitment of filament elongation and bundling factors. To test this mechanism further, we looked at the recruitment of Arp2/3 complex and mDia2 to toca-1 spots in the presence of the actin monomer–sequestering drug latrunculin B. The number of toca-1 spots that recruit Arp2/3 complex are unaffected by latrunculin (Fig. 4A), whereas latrunculin completely blocks toca-1–mDia2 colocalization (Fig. 4A).

Fig. 4

Initiation and elongation of FLSs occur by means of separable molecular mechanisms. (A) Latrunculin B does not affect the co-localization of Arp2/3 complex with toca-1 but completely inhibits mDia2 recruitment to toca-1 sites (n = 4 independent experiments). (B) Dose-response of FLS initial elongation rate with GST-CA preincubated in the extract before addition to the supported bilayer (n = 20 FLSs). The dotted line indicates the minimum elongation rate (0.1 μm/min) attributable to the axial resolution limit of confocal microscope (~0.7 μm). (C to F) Pulse-chase experiment starting with Alexa-647 actin and Alexa 568-labeled Arp2/3 complex in the extract, with later addition of 40 μM GST-CA and Alexa-488 actin. Shown are Alexa 647-actin [(C), blue]; Alexa 488-actin [(D), green]; Alexa-568-Arp2/3 complex [(E), red]; and a three-color overlay (F). Addition of new actin monomers continues in the absence of the Arp2/3 complex recruitment into the FLS. Scale bars, 2 μm. (G) The time course of FLS elongation (measured by the second color of actin) at increasing GST-CA concentrations (n = 20 FLSs). (H) Dependence of rate of FLS elongation on the concentration of GST-CA. Maximal FLS elongation rate is unchanged by addition of GST-CA. (I to M) Similar pulse-chase experiment explained in (C) to (H), with the additional use of GST-LRR to inhibit formin activity. [(I) and (K)] Control addition of GST-CA plus the second color of actin: Alexa-647 (red) was first, and Alexa-488 (green) was second. (I) Side view. (K) Actin on the bilayer. (J and L) Inclusion of 5 μM GST-LRR in the extract, then addition of GST-CA and Alexa-488 actin. GST-LRR leads to the detachment of the FLSs so that fewer punctae are present at the membrane surface. (J) Side view. (L) Actin on bilayer. Scale bars, 2 μm [(I) and (J)] and 5 μm [(K) and (L)]. (M) Quantification of GST-DIP-LRR addition (n = 6 fields of view). *P < 0.001. All error bars are SD. The lipid composition was 45% PC, 45% PI, and 10% PI(4,5)P2.

To probe the role of the Arp2/3 complex in FLS formation, we added glutathione S-transferase cofilin homology and acidic (GST-CA) domain, which inhibits N-WASP activation of the Arp2/3 complex; this reduces the number of nucleation sites and abolishes FLS elongation (Fig. 4B and fig. S6, A and B). Immunodepletion of N-WASP significantly decreases but does not completely inhibit FLS formation and elongation (fig. S6, D and H to I), which is consistent with studies in cultured cells that suggest the involvement of other Arp2/3 complex nucleation–promoting factors (11, 12). Immunodepletion of toca-1 has only a minor effect on elongation (fig. S6, E, H, and I); however, there are more than a dozen candidate BAR domain superfamily proteins that could compensate for loss of toca-1. We conclude that signaling through the Arp2/3 complex plays a vital role in the initiation of FLS formation, although these proteins alone are not sufficient to generate FLSs (Fig. 1, A and B).

The usual product of Arp2/3 complex activation is an array of disorganized or branched actin structures rather than organized parallel actin bundles. The kinetics of protein recruitment to the nascent FLSs suggests that a formin-driven elongation process occurs after the formation of the first actin nucleus. To test whether the elongation phase is dependent on the Arp2/3 complex, we started the reaction with Alexa 647-actin and Alexa 568–Arp2/3 complex and after 20 min added GST-CA to block further Arp2/3 complex function and Alexa 488-actin to record any further actin polymerization (Fig. 4, C to H). We found that new actin monomers are still added at the FLS tip after Arp2/3 complex inhibition (Fig. 4, C to F). In the presence of GST-CA, there is no further incorporation of Arp2/3 complex into the FLS. The region of newly assembled actin in the shaft lacks Arp2/3 complexes (Fig. 4, E and F, and fig. S7A). After ~5 min, addition of new actin to the FLS slows, indicating a requirement for Arp2/3 complex to maintain FLS growth over the long term (Fig. 4G). At high concentrations of GST-CA, there is a noticeable lag in the incorporation of new actin monomers (Fig. 4G), probably because the occupancy of GST-CA on N-WASP–binding sites of the Arp2/3 complex stimulates a reorganization of the tip complex. Even at high doses of GST-CA, an Arp2/3 complex–independent component of elongation is maintained (Fig. 4H).

Diaphanous-related formins are the obvious engine for filopodial growth in the absence of the Arp2/3 complex because formins are thought to drive filopodial elongation in different cell types. Because there are many formin proteins, we used a dominant-negative approach known to inhibit filopodia formation (34, 35). The leucine-rich region (LRR) of diaphanous-interacting protein (DIP) binds and inhibits Drf1 and Drf3 (35). When GST-DIP-LRR is added to extracts, FLS formation still occurs with no reduction in elongation rate or the number of structures. However, when Arp2/3 complex function is inhibited by GST-CA in the presence of GST-DIP-LRR, the number of structures that incorporates new actin monomers is significantly reduced (Fig. 4, I to M). This is accompanied by the detachment of many FLSs from the lipid bilayer, suggesting that the tip complex undergoes conformational changes and that continuing reorganization of the tip complex cannot occur in the presence of GST-LRR (Fig. 4M). Furthermore, because diaphanous-related formins are activated by RhoA we also tested dominant-negative RhoA, GST-RhoA-N19. This produced similar results to GST-DIP-LRR (fig. S7, B to F). Thus, there are important functions contributed by both the Arp2/3 complex and formins. In the presence of capping activities, the Arp2/3 complex may be required for generating new actin nuclei (7, 36).

Currently, there are two principal models for filopodial assembly: convergent elongation (5) and de novo nucleation (37). In the former, Arp2/3 complex–driven actin filaments continually coalesce into parallel arrays through the continuous action of bundling proteins, such as fascin. In the latter, filopodia are formed through the establishment of a tip complex of formins, which nucleates long actin filaments that are then bundled. In the model we propose (fig. S8), signaling by a negatively charged membrane leads first to the recruitment and clustering of BAR domain superfamily proteins and N-WASP or other nucleation promoting factors at the membrane, leading to subsequent recruitment of the Arp2/3 complex and the formation of a small patch of short actin filaments. This early clustering step represents the key difference from the convergent elongation model because symmetry-breaking occurs through the focal recruitment of activators of the Arp2/3 complex rather than by a coalescence of preexisting actin-barbed ends. In the proposed clustering-outgrowth model, local assembly of actin initiated by the Arp2/3 complex is converted into a filopodial tip complex by the recruitment of formins and VASP. This recruitment enables linear outgrowth of the filopodium, with short actin filaments elongated by formins and/or VASP and bundled by fascin (15, 16). Actin filaments generated by the Arp2/3 complex appear to be continually required to feed the elongation process, although other actin nucleators could fulfill a similar role. The observation of short actin filaments at the tips of filopodia in Dictyostelium by means of electron tomography is consistent with this model (38). The clustering of actin assembly proteins and outgrowth by elongation factors may be served by different components in different circumstances. In this view, we should expect that although the overall mechanism would be conserved, there could be flexibility in filopodial composition (39).

Supporting Online Material

Materials and Methods

Figs. S1 to S8


Movies S1 to S4

References and Notes

  1. K.L. was the recipient of a Korea Science and Engineering Foundation (KOSEF) Scholarship and Harvard-MIT Health Sciences and Technology Fellowship. J.L.G was the recipient of an European Molecular Biology Organization Long-Term Fellowship. This work was funded by NIH grant GM26875. We thank G. Bokoch for the gift of the GST-RhoA-N19 plasmid, S. Field for the gift of the pEGFP GFP-PLCδ PH domain plasmid, O. Weiner for the RhoGDI plasmid, R. Tsien for the mCherry plasmid, A. Alberts for the mDia2 plasmid, and D. Vignjevic for the fascin plasmid. We thank A. Lebensohn for helping extract preparation and valuable discussions and M. Coughlin, T. Walz, and D. Kelly for help with EM. We thank J. Waters, W. Salmon, and Nikon Imaging Center at Harvard Medical School.
View Abstract

Stay Connected to Science

Navigate This Article