WAVE complex forms linear arrays at negative membrane curvature to instruct lamellipodia formation

Plasma membrane shape changes are crucial for many aspects of cell physiology. For example, migratory cells build flat, sheet-like membrane protrusions called lamellipodia to power cell motility (Innocenti, 2018; Krause and Gautreau, 2014), and some cells also build finger-like membrane protrusions called invadopodia to focally degrade the extracellular matrix to support metastasis (García et al., 2012; Gligorijevic et al., 2012; Yu et al., 2012). Actin rearrangements play a key role in the spatial and temporal regulation of these membrane deformations. Nucleation-promoting factors (NPFs) help specify when and where actin is polymerized (Chesarone and Goode, 2009; Pollard and Borisy, 2003; Rottner et al., 2017; Takenawa and Suetsugu, 2007). The neural Wiskott–Aldrich syndrome protein (N-WASP) and the WASP family verprolin homologous protein (WAVE) complex both activate the actin-related protein 2/3 (Arp2/3) complex to seed actin nucleation but result in distinct membrane morphologies. N-WASP orchestrates actin networks that support finger-like podosomes, invadopodia, and filopodia (García et al., 2012; Gligorijevic et al., 2012; Linder et al., 2022; Miki et al., 1998; Nemethova et al., 2008; Ward et al., 2004; Yu et al., 2012), while the WAVE complex orchestrates sheet-like branched actin networks that underlie broad and thin lamellipodial structures (Fritz-Laylin et al., 2017a; Fritz-Laylin et al., 2017b; Leithner et al., 2016; Machesky and Insall, 1998; Machesky et al., 1999; Steffen et al., 2004; Weiner et al., 2007).

How do different NPFs instruct different patterns of membrane protrusion? Biochemical reconstitutions using different distributions of micropatterned surfaces coated with the same NPF have provided some insights into the relation between actin network structures and NPF organization patterns. While a dot-like NPF organization drives finger-like actin networks, a linear pattern of the same NPF drives lamellipodial-like actin networks in vitro (Boujemaa-Paterski et al., 2017; Carlier et al., 2003). This suggests that the spatial organization of NPFs could dictate the resulting morphology of actin assembly and cell protrusion.

For in vitro studies, NPF distribution is typically externally constrained (Boujemaa-Paterski et al., 2017; Carlier et al., 2003). For living cells, NPF distribution is internally specificized through self-organization. For example, N-WASP phase separate to form focal structures on the plasma membrane that direct the extension of invadopodia and filopodia (Banjade and Rosen, 2014; Case et al., 2019; Ditlev et al., 2012; Kim et al., 2019; Ward et al., 2004). In contrast, to build a flat, sheet-like lamellipodial protrusion, the WAVE complex needs to assemble into a linear pattern on the plasma membrane. The multivalent interactions that generate N-WASP phase transition are relatively well understood and have been reconstituted with purified proteins in vitro (Banjade and Rosen, 2014; Case et al., 2019). However, the mechanism of WAVE complex organization into linear distributions is not understood and has not proven amenable to biochemical reconstitutions of lamellipodia formation (Chen et al., 2010; Koronakis et al., 2011; Lebensohn and Kirschner, 2009).

In a migratory cell, the WAVE complex propagates as a “wave” pattern at the tips of lamellipodia, with expanding and contracting zones of WAVE complex accumulation marking the expansion and contraction of the lamellipod (Video 1; Pipathsouk et al., 2021; Steffen et al., 2004; Stradal et al., 2001; Weiner et al., 2006). These wave patterns arise from an excitable feedback network composed of WAVE complex self-recruitment activated by Rac GTPases and delayed inhibition from actin polymerization that strips the WAVE complex off of the plasma membrane (Chen et al., 2017; Millius et al., 2012; Pipathsouk et al., 2021; Weiner et al., 2007). However, neither the upstream activator Rac nor the downstream effector actin exhibits such restricted pattern at the tips of lamellipodia as the WAVE complex: Rac activation (assayed via the PBD [p21-binding domain] of PAK1 [p21-activated kinase 1]) and actin (through phalloidin staining) are broadly distributed at the front of the cell and span a broader permissive zone compared with the WAVE complex (Fig. S1, A–F). Furthermore, the WAVE complex but not its downstream nucleator the Arp2/3 complex is required for formation of lamellipodia (Buracco et al., 2024; Pipathsouk et al., 2021). Notably, the WAVE complex also forms stereotypic nanoscale ring structures in the absence of actin cytoskeleton, a phenomenon not observed for Rac or the Arp2/3 complex (Pipathsouk et al., 2021). Collectively, this evidence suggests a self-organizing propensity of the WAVE complex that is independent of the spatial pattern of upstream permissive inputs and downstream effectors, underscoring a crucial organizational role of the WAVE complex in lamellipodia formation.

Here, we investigate the rules of the WAVE complex organization through in vivo biochemistry. With quantitative fluorescent imaging and analysis, we demonstrate that the WAVE complex forms linear distributions on cell membranes as expected for an NPF that builds sheet-like actin networks. To further investigate the biophysical properties of the WAVE complex in a cellular context, we use single-molecule tracking in combination with fluorescent molecular counting and show that the WAVE complex forms solid-like linear arrays with a stereotypic multilayered organization pattern. Additionally, the WAVE complex organization pattern remains invariant regardless of its expression level, suggesting that the WAVE complex is a core and rate-limiting component of the linear array. We had previously demonstrated a correlation of WAVE complex enrichment to membrane curvature—the WAVE complex linear arrays are always associated with negative membrane curvature in the native cellular context (Drab et al., 2023; Pipathsouk et al., 2021). To functionally test how membrane curvature affects WAVE complex dynamics, we use nanobeads, nanopatterns, and cell compression to manipulate cellular membrane curvature. We find that negative membrane curvature is both sufficient and necessary for WAVE complex assembly on cell membranes in the presence of upstream activators (Rac, Arf1/6, and PIP₃) and the proline-rich domains (PRD) of both WAVE2 and Abi2. Finally, through computational modeling, we demonstrate that the WAVE complex’s linear organization and preference for negative curvature both play important roles in robust lamellipodia formation.

To build a sheet-like lamellipod, we expect the WAVE complex to organize into a linear distribution on cell membranes. To quantify the native accumulation patterns of the WAVE complex on cell membranes, we used fluorescent quantification to measure the relation between the number of subunits and the pattern of the structure. A linear distribution (that grows in one dimension) exhibits a linear relation between the length of the pattern and the number of subunits, while a planar distribution (that grows in two dimensions) exhibits a quadratic relation between the length of the pattern and the number of subunits (Fig. 1 A). To visualize the WAVE complex distribution patterns on cell membranes, we stably expressed eGFP-Sra1 (one of the WAVE complex subunits) in neutrophil-like HL-60 cells and imaged them with total internal reflection fluorescence–structured-illumination microscopy (TIRF-SIM). Quantification of the fluorescent intensities of the WAVE complex versus the lengths of the wave patterns at the tips of lamellipodia revealed a highly linear relation between them (Pearson correlation coefficient r = 0.96; Fig. 1, B–D). To study the intrinsic WAVE complex accumulation patterns freed from the constraints of the actin cytoskeleton, we used latrunculin B (latB) to deplete actin polymers. Under these conditions, the WAVE complex forms nanoscale ring structures (Fig. 1, E and F; and Video 1; Pipathsouk et al., 2021). As for lamellipodia, the fluorescent intensity of the WAVE complex nanorings also scales linearly with the radius (length) of the nanoring structures in cells treated with latB (Pearson correlation coefficient r = 0.86; Fig. 1 G). This linear relation between the fluorescent intensity (that indicates the number of monomers) and the length of the pattern suggests an underlying linear distribution of the WAVE complex. Additionally, the WAVE complex concentration per unit length in nanorings under latB treatment is consistent with that at the tips of lamellipodia (Fig. 1 H), suggesting an invariant pattern of WAVE complex linear assembly over a wide range of cellular conditions. As a comparison with other lamellipodia components, we also examined the distribution of polymerized actin as a function of lamellipodial length (Fig. S1 E). Actin correlates with lamellipodial length (r = 0.69) but exhibits a significantly poorer linear relation compared with WAVE complex in lamellipodia (Fig. S1 F). Lastly as a control for a planar structure (comparable with N-WASP focal droplets), we adapted the SPARK-ON system that drives protein separation in the cytosol through the multivalent interactions between HoTag3 and HoTag6 (Fig. S1 G; Chung et al., 2023). By adding a membrane Fyn tag (Xia and Götz, 2014) to one of the SPARK-ON system components, HoTag3, we found that it can self-oligomerize when targeted to the cell membrane in HEK293T cells even in the absence of HoTag6 (Fig. S1, G and H), possibly due to the high local concentration of HoTag3 upon membrane recruitment. Unlike the WAVE complex that forms linear patterns on cell membranes, HoTag3 exhibits a quadratic relation between the fluorescent intensities and the radius of the pattern and a linear relation to the area (Fig. S1 I), suggesting the formation of a planar structure in contrast to the linear distribution of the WAVE complex.

The WAVE-related NPF N-WASP undergoes liquid–liquid phase separation on cell membranes (Banjade and Rosen, 2014; Case et al., 2019). In contrast, the mechanism by which the WAVE complex forms linear organizations in cells is not known; purified WAVE complex does not assemble on its own (Chen et al., 2010). We envisioned two possibilities for the linear organization of the WAVE complex—either the WAVE complex binds to a pre-existing linear protein scaffold, like microtubule-associated proteins binding to microtubules (Bodakuntla et al., 2019), or the WAVE complex could represent a core component of a linear array, like tubulin subunits in microtubules. To differentiate between these possibilities, we titrated the expression level of the WAVE complex in cells. If the WAVE complex is associated with an independently formed protein scaffold, the WAVE complex concentration per unit length, which we measure by unit length fluorescent intensity on the scaffold, should scale within the WAVE complex expression level (Fig. 2 A). In contrast, if the WAVE complex is a core component of a linear array, the WAVE complex should exhibit invariant concentration per unit length (Fig. 2 B). We titrated the WAVE complex abundance via stable Hem1-eGFP expression in Hem1 KO HL-60 cells (Fig. S2, A–C): disruption of any WAVE complex subunit destabilizes the remaining subunits of the WAVE complex (Innocenti et al., 2004; Mendoza, 2013), resulting in loss of lamellipodia formation in HL-60 cells (Graziano et al., 2019); reintroduction of Hem1-eGFP in Hem1 KO cells titrates the WAVE complex expression level proportional to Hem1-eGFP rescue (Fig. S2 C). Our data show that the WAVE complex linear arrays at the tips of lamellipodia exhibit invariant fluorescent intensity per unit length at all levels of Hem1-eGFP expression (Fig. 2, C and D). A similar behavior was observed for WAVE complex nanorings in the absence of the actin cytoskeleton (Fig. S2, D and E). These data suggest that the WAVE complex is a core and rate-limiting component of the linear array rather than associating with a pre-existing scaffold.

Next, we investigated the modes of WAVE complex subunit organization on cell membranes. These linear arrays could either be solid-like (like actin filaments or microtubules) or diffusive (like N-WASP condensates) (Fig. 2 E). Previous studies examining the WAVE subunit dynamics at the tips of lamellipodia have demonstrated their diffusive behavior on cell membranes with lateral movement driven by actin elongation (Mehidi et al., 2021). To explore the intrinsic modes of WAVE complex subunit organization in the absence of actin-applied forces, we employed single-molecule tracking to examine WAVE complex dynamics in latB-treated cells. Upon treatment with actin inhibitor, the WAVE complex organizes into nanorings as well as linear structures at the cell periphery. We focused our analysis on these linear structures at the cell periphery as they represent a larger substrate for lateral diffusion. We expressed Halo-tagged Sra1 in HL-60 cells and labeled the cells with two HaloTag ligand dyes, a high concentration of JF646 for full labeling of the WAVE complex and a lower concentration of JF549 (sparse labeling) for single-molecule tracking (Fig. 2 F). As a control for a protein that is diffusive in the membrane, we expressed Halo-tagged membrane-binding motif CAAX in HL-60 cells (Fig. 2 G). Cells were treated with an actin inhibitor latB and then imaged with TIRF microscopy at a rate of ∼26 frames per second (Fig. S2 F). The single-molecule traces indicate that the WAVE complex monomers in the linear WAVE complex arrays are mostly stationary and with minimal diffusion, whereas the CAAX-tagged proteins exhibit significant lateral diffusion (Fig. 2, F–H and Video 2). Compared with the WAVE complex subunit at the tips of propagating lamellipodia (D = 0.12 µm²/s; Mehidi et al., 2021), the WAVE complex in the absence of the actin cytoskeleton shows almost an order of magnitude decrease in diffusion (D = 0.014 µm²/s; Fig. 2 H). Particles with such low diffusion coefficient (<0.015 µm²/s) are usually defined as immobile particles (Mehidi et al., 2021; Orré et al., 2018), suggesting that the intrinsic property of WAVE complex linear arrays in the absence of actin assembly is solid-like, similar to protein polymers like actin filaments or microtubules.

The WAVE complex solid-like linear array could either adopt a single-layered or a multilayered arrangement. To investigate the subunit arrangement of the WAVE complex linear array, we performed molecular counting of the WAVE complex linear arrays on cell membranes. We leveraged fluorescent standard candles to calculate the number of fluorescent molecules using the microscopically measured fluorescent intensities to construct a fluorescence-based standard curve (Akamatsu et al., 2020; Hsia et al., 2016). The membrane-tagged fluorescent candles with 12, 60, or 120 copies of eGFP proteins were expressed in HL-60 cells (Fig. 3 A). We used TIRF-SIM to measure the fluorescence intensities of the standard candles in cells that were chemically fixed to limit the movement of the candles on the cell membrane (Fig. 3 A). We used this data to construct a fluorescence-based standard curve that confirms good linearity between the fluorescent intensities and the number of fluorescent molecules in each standard candle (Fig. 3 A).

If the WAVE complex is forming a single-layered array, it would necessitate at least 50–125 copies of the WAVE complex per micron, given the dimension (200 × 110 × 80 Å) of a single WAVE complex (Fig. 3 B; Z. Chen et al., 2010). To determine the number of the WAVE complexes within the linear array, we used the eGFP-Sra1 subunit of the WAVE complex to calculate the absolute abundance of the WAVE complex at the cell membrane in HL-60 cells. The total expression level of the WAVE complex in each cell was not affected by eGFP-Sra1, as any subunit not incorporated into the WAVE complex is degraded (Graziano et al., 2019; Pipathsouk et al., 2021; Weiner et al., 2006). We quantified the fluorescent intensities of the eGFP-Sra1 at the tips of lamellipodia with TIRF-SIM that allows highly consistent and reproducible TIRF image acquisition and quantification (Fig. S2 I). Correcting for fluorescence loss after fixation and the tagged ratio of the Sra1 subunits (Fig. S2, G and H) yielded a mode of 680 copies of the WAVE complex per micron at the tips of lamellipodia (Fig. 3 C). This value of 680 is much greater than the copies of WAVE complex that would be required to make a single-layered array (Fig. 3 B), suggesting that the WAVE complex forms a multilayered array at least 5–14 layers thick (considering potential co-oligomerization with other binding proteins) at the tips of lamellipodia (Fig. 3 D).

To investigate the mechanism of WAVE complex linear array assembly, we used FRAP to study the turnover dynamics of the WAVE complex on cell membranes. The WAVE complex exhibits rapid turnover dynamics at the tips of lamellipodia in migrating cells with a half-life of 6.3 s and complete exchange along the length of the array (Fig. S2 J and Video 3; Lai et al., 2008; Mehidi et al., 2021; Weiner et al., 2007). When treated with the actin inhibitor latB, the turnover dynamics of the WAVE complex are decreased to 22 s, with about one third of the WAVE complex molecules failing to exchange within the linear array (Fig. 3 H). The exchanging molecules exhibit a uniform recovery pattern along the bleached region (Fig. 3, E–G and Video 4), indicating that the molecule exchange occurs between the array and the cytosolic pool with minimal diffusion within the array itself (Fig. 2, F–H). This turnover behavior is notably distinct from actin or microtubule polymers, where subunits typically only exchange at the end of the protein array (Fig. 3 I; Katsuno et al., 2015; Smith et al., 2013; Vorobjev et al., 1999). This uniform but partial recovery pattern suggests that the WAVE complex could either be a single multilayered linear array like a microtubule that permits molecule exchange within the array as microtubules do under certain conditions (Kuo et al., 2022), or the WAVE complex could form a bundled array where molecular exchanges occur at the end of each short polymer within the array (Fig. 3 J).

Some multilayered polymers, like microtubules, can polymerize into a curved parallel array where their lateral extension of polymerization is constrained by the intrinsic curvature of the polymer (Chrétien et al., 1995; Nogales and Wang, 2006). Is curvature playing a similar role for the assembly of the multilayered WAVE complex linear arrays? Consistent with this idea, there is a characteristic membrane curvature seen for WAVE complex recruitment in cells. The WAVE complex localizes to the tips of the lamellipodia that exhibit negative curvature of a radius of 65 nm in the axis of protrusion (x-z plane) and little-to-no curvature along the lamellipod (x-y plane) (Fig. 4 A; Pipathsouk et al., 2021; Schmeiser and Winkler, 2015). For cells treated with actin inhibitor, the WAVE complex forms nanorings localized to the saddle curvature neck of membrane invagination sites, consisting of a protruding negative curvature of a radius of 65 nm perpendicular to the invagination neck (x-z plane) and a positive curvature of a radius of 115 nm around the invagination neck (x-y plane) (Fig. 4 B; Pipathsouk et al., 2021). Despite dramatic differences in the overall membrane morphology, the 65 nm radius of negative curvature is preserved between both of these contexts. These correlative data suggest that this negative membrane curvature could be a key permissive factor in WAVE complex association on cell membranes.

To test the functional role of negative membrane curvature in WAVE complex linear assembly, we first investigated whether negative curvature is sufficient for WAVE complex membrane association. We exposed migrating cells to polystyrene beads of 200 nm that could induce membrane invaginations with negative curvature (Fig. 4 C). As eGFP-Sra1–tagged HL-60 cells migrate over these beads, the WAVE complex forms ring structures at the neck of membrane invagination sites (Fig. 4 C; and Videos 5 and 6). The WAVE complex signal is persistent on beads over time and significantly higher than the membrane dye signal (Fig. 4, D and E), indicating that the increase in WAVE signal is not due to excess membrane accumulation but rather reflects WAVE complex accumulation at the negative membrane curvature induced by the beads. These data suggest that negative curvature is sufficient to induce WAVE complex membrane association in the context of a migrating cell; though, as we show later, this recruitment depends on the permissive inputs of upstream activators (Fig. 5, F–H).

Unlike other negative curvature-sensing proteins, such as I-BAR domain proteins, the WAVE complex does not localize to sites of double-negative membrane curvatures like the tips of filopodia (Pipathsouk et al., 2021). This suggests that the WAVE complex organization may be influenced by another axis of curvature. To investigate whether the membrane curvature perpendicular to the negative curvature axis influences the linear membrane association of the WAVE complex, we exposed migrating cells to polystyrene beads of varying diameters of 200 or 500 nm. The WAVE complex was recruited to beads of both sizes but exhibits different association dynamics depending on the bead curvature: the WAVE complex associates persistently with 200 nm beads while showing only transient association on the 500 nm beads (Fig. S3, A–C). This preference for the 200 nm beads aligns with the stereotypic diameter (230 nm) of the WAVE complex nanorings formed in the absence of actin cytoskeleton (Pipathsouk et al., 2021), indicating a preference for positive curvature around 200 nm in diameter. These findings suggest that while negative curvature is the key geometric permissive feature for WAVE complex membrane association, the positive curvature in the perpendicular axis also plays a role in regulating WAVE complex membrane recruitment.

Next, we tested whether negative membrane curvature is necessary for WAVE complex membrane association by inhibiting negative membrane curvature. The native negative curvature that proved most amenable to manipulation was the membrane invaginations at the ventral surface of latB-treated cells. We overlaid cells with a precast agarose pad to induce compression and iron out the membrane invagination sites. If negative curvature is required for WAVE complex membrane association, disturbing membrane invagination sites should interfere with the WAVE recruitment to nanorings (Fig. 4 F). Indeed, the WAVE complex nanorings disappear following cell compression (Fig. 4, G and H). Because the WAVE complex persists at the rim of the cell where the negative curvature is maintained (Fig. 4 G), it is unlikely that compression exerts its effects by blocking the upstream regulators WAVE complex recruitment. This suggests that the disappearance of the WAVE complex nanorings is driven by membrane curvature changes, highlighting the necessity of negative membrane curvature for WAVE complex membrane association.

Finally, we tested whether WAVE complex membrane association can be rescued during compression by reintroducing negative membrane curvature to cells. For this purpose, we provided latB-treated cells with nanopatterns at the ventral surface that could sustain negative membrane deformations following compression with the agarose pad (Fig. 4 I; Cail et al., 2022). Under these conditions, the WAVE complex forms linear arrays around the regions of micropattern-induced negative curvature (Fig. 4 J and Fig. S3 D). Moreover, the WAVE complex is also recruited to regions of negative curvature around nanopatterns in migratory cells (Fig. S3 E and Video 7). Together, these data suggest that the negative membrane curvature plays a key role in controlling WAVE complex association on cell membranes.

Here, we sought to investigate the molecular mechanism of WAVE complex organization into this linear curvature-sensitive array. Since purified WAVE complex does not oligomerize or sense curvature on its own, the WAVE complex is likely to associate with other membrane-bound partners. One of the most well-known binding partners of the WAVE complex is the I-BAR domain protein family that senses negative membrane curvature and associates with lamellipodia and filopodia (Frost et al., 2009; Miki et al., 2000; Nishimura et al., 2021; Suetsugu et al., 2006; Zhao et al., 2011). Previous studies have shown that one of the I-BAR proteins, iRsp53, requires interactions with the WAVE complex to enrich to lamellipodia but is dispensable for WAVE complex’s linear organization and patterning of lamellipodia (Pipathsouk et al., 2021). Because the BAR domain protein family is known to have redundant functions (Chou et al., 2017; Insinna et al., 2019; Nkosi et al., 2013; Park et al., 2010), we utilized a B16-F1 cell line (I-BAR 4×KO) lacking all four of the I-BAR family proteins (IRSp53, MIM, iRTKS, and ABBA) from Pokrant et al. (2023). We investigated how the I-BAR family proteins affect WAVE complex membrane organization by stably expressing eGFP-Sra1 in WT and I-BAR 4×KO B16-F1 cells. Knockout of the I-BAR family has no effect on the ability of the WAVE complex to form linear arrays at the tips of lamellipodia or nanorings in the absence of actin cytoskeleton (Fig. 5 A). Both the WT and the I-BAR 4×KO cells exhibit similar unit length fluorescent intensities of the WAVE complex at the tips of lamellipodia (Fig. 5 B) as well as similar nanoring fluorescent intensities and sizes (Fig. 5 C). These data indicate that the I-BAR protein family is dispensable for WAVE complex membrane organization and curvature sensation. In addition, we also generated a triple knockout of all the TOCA family proteins (FBP17, Toca-1, and CIP4) in HL-60 cells (TOCA 3×KO). These are the most well-known sensors of positive curvature for actin regulators and have been shown to interact with the WAVE complex (Bai and Grant, 2015; Gligorijevic et al., 2012). Stably expressing mCherry-Sra1 in TOCA 3×KO HL-60 cells shows that the WAVE complex exhibits normal localization to the tips of lamellipodia and forms nanorings in the absence of actin cytoskeleton (Fig. S3 F), indicating that the TOCA protein family is not required for WAVE complex linear organization. To further investigate if the TOCA protein family is involved in WAVE complex curvature sensation, we analyzed TOCA 3×KO HL-60 cells migrating on 200 nm beads. The WAVE complex can still be recruited to the 200 nm beads (Fig. S3, G and H) in the absence of TOCA protein family, suggesting that they are not required for WAVE complex curvature sensation.

The WAVE complex contains two PRDs in the WAVE2 and Abi2 subunits, which bind to Src-homology 3 domains and mediate interactions with other proteins (Joseph et al., 2017; Mendoza, 2013). Previous studies have shown that deletion of both PRDs in B16-F1 cells impairs lamellipodia formation (Buracco et al., 2024). Here, we investigated how the PRD affects WAVE complex linear organization and membrane association in HL-60 cells. We expressed eGFP-tagged WAVE2∆PRD (∆298–407) or Abi2∆PRD (∆174–409) in WT HL-60 cells in the presence of endogenous WAVE2 and Abi2. Both WAVE2∆PRD and Abi2∆PRD can be recruited to the tips of lamellipodia and form linear arrays (Fig. 5 D), although the Abi2∆PRD exhibits reduced membrane signal at the tips of lamellipodia and increased unrecruited WAVE complex (cytosolic localization) (Fig. S3 I). When treated with actin inhibitor, WAVE2∆PRD and Abi2∆PRD can still form nanorings, but both exhibit reduced nanoring formation and increased cytosolic localization (Fig. 5, D and E), suggesting a role for their PRDs in WAVE complex membrane association.

Lastly, WAVE complex is known to be activated by several upstream proteins for membrane association, including Rac, Arf, and PIP₃ (B. Chen et al., 2017; Koronakis et al., 2011; Oikawa et al., 2004; Rottner et al., 2021; Suetsugu et al., 2006). We investigated how those upstream biochemical links could affect WAVE complex membrane recruitment by treating HL-60 cells stably expressing eGFP-Sra1 with inhibitors of these upstream activators (Fig. 5 F): Rac/Rho/Cdc42 inhibitor Clostridium difficile toxin B (Just et al., 1995; Pipathsouk et al., 2021), Rac inhibitor EHT 1864 (Onesto et al., 2008; Shutes et al., 2007), Arf1 inhibitor brefeldin A (Humphreys et al., 2012), Arf6 inhibitor chlortetracycline (Macia et al., 2021), and PI3K inhibitor duvelisib (Belly et al., 2024, Preprint). Cells treated with Rac inhibitor (toxin B or EHT 1864) show severely disrupted lamellipodia and WAVE complex nanoring formation. Arf1/6 or PIP₃ inhibitor also impairs the formation of WAVE complex recruitment to lamellipodia and nanorings, but not as potently as for Rac inhibitors (Fig. 5, F–H). These data suggest that Rac plays an indispensable role in WAVE complex activation and membrane association, while Arf1/6 and PIP₃ are involved but may be redundant with other WAVE complex activators.

Previous work (Drab et al., 2023; Sadhu et al., 2023b; Sadhu et al., 2021) have used computational simulations to show that linking actin nucleation with self-associating proteins that sense negative curvature suffices to induce lamellipodial-like protrusions. In addition to actin nucleation and a preference for negative curvature, our current work suggests a linear organization of the WAVE complex. Here, we explored how the negative curvature sensing and linear organization of actin nucleators influence lamellipodia formation in virtual cells. We compare the behavior of previous models (negative curvature preference + isotropic interactions of nucleators) with that of the WAVE complex (negative curvature preference + anisotropic/linear interactions of nucleators).

Our modeling approach is based on simulating the dynamics of a triangulated surface that represents the cell membrane, where we consider that each surface node to be either a bare membrane or occupied by a curved membrane protein complex (CMC) (Fig. S5 A). These CMCs diffuse on the membrane while also having some finite binding energy between them, such that they can form clusters. The dynamics of the membrane shape and organization of the curved proteins is calculated using the Monte Carlo method, where the system evolves to lower its global energy, while including the effects of thermal fluctuations. The energy of the system contains the bending energy (including the effects of the spontaneous curvature of the CMC), protein–protein binding energy, and adhesion to an external surface. In addition to these components, we include the protrusive forces exerted on the membrane by actin polymerization activity inside the cell, which we assume to be recruited to the membrane by the CMC. We therefore apply a constant protrusive force on the nodes that are occupied by the CMC, directed locally outward. This model was previously explored for isotropic protein–protein binding (Fig. 6 A). Here, we are motivated by our new experimental observations to explore the dynamics when the protein–protein binding is polymer-like (Fig. 6 B). By polymer-like, we mean that the CMC can only bind at most two neighboring CMCs, such that they naturally form linear clusters. This modification gives rise to major changes in the overall CMC organization and membrane shapes (Fig. 6 B).

We investigated the steady-state vesicle shapes from our simulations as a function of the strength of the protrusive forces (F) and the binding energy between the CMC (w) (Fig. 6). The lamellipodia phase (light green) is defined to be when the average number of clusters becomes less than four (Fig. S5, I and J) and the steady-state vesicles are either in a crescent (motile) or two-arc (immotile) shape. When there are more than four clusters, the phase is considered to be disordered (yellow regions in Fig. 6). The budding phase (dark green region in Fig. 6 A) is distinguished from the lamellipodia phase (light green regions in Fig. 6) in the following way: when the shape of the protein clusters becomes hemispherical (below a prescribed shape threshold, see Fig. S5, I and J), it is considered to be a bud, otherwise a lamellipod. For isotropic CMC interactions, when both the binding energy and protrusive force are low, the CMC clusters are small and disorganized (the “disordered” phase, yellow region in Fig. 6 A and Fig. S5 B and left column in Video 8). When the binding energy dominates over the protrusive force (large w small F regime, dark green region in Fig. 6 A and Fig. S5 C; and lower left panel in Video 9), the CMC tends to form spherical buds. Above a critical strength of actin-induced forces, the CMC self-organizes into lamellipodia-like leading-edge clusters, either as a single crescent (motile) shape or a two-arm (immotile) shape (green region in Fig. 6 A and upper left panel in Video 9).

When the CMC interacts with at most two neighbors, forming linear polymer-like chains, we find that the phase diagram changes significantly. The buds phase vanishes, and the lamellipodia-like shapes extend to lower values of both polymerization force and binding energy between CMC interaction (Fig. 6 B; and Fig. S5, D and E, and right column in Videos 8 and 9). This indicates that by having an intrinsic tendency to form linear chain-like clusters, the CMC can form more robust leading-edge clusters that initiate lamellipodia-like protrusions and cell polarity. In Fig. S5, F and G, we show that the linear aggregation tendency allows the lamellipodia-forming region to significantly extend to lower values of actin force and surface adhesion compared with the isotropic aggregation. Note that the spontaneous negative curvature still plays an essential role, and in its absence, the tendency to form chain-like clusters is not sufficient to initiate lamellipodia protrusions (Fig. S5 H and Video 10).

The WAVE complex is an NPF that organizes many different cellular behaviors, including chemotaxis, macropinocytosis, phagocytosis, and cell integrity (King and Kay, 2019; Pipathsouk et al., 2021; Seastone et al., 2001; Williams et al., 2022). These processes all involve flat, sheet-like actin-based protrusions (Fritz-Laylin et al., 2017b; Humphreys et al., 2016; Mylvaganam et al., 2021; Veltman et al., 2016; Yang et al., 2021), but how the WAVE complex instructs this pattern of protrusion is not well understood. Since the WAVE complex does not form linear structures in cell-free biochemical reconstitutions (Chen et al., 2010; Koronakis et al., 2011; Lebensohn and Kirschner, 2009), our work leverages in vivo biochemistry approaches to understand the logic of lamellipodia formation. We reveal a highly consistent linear organization pattern of the WAVE complex at the cell membrane that could suggest a self-organizing template for these flat, sheet-like actin networks (Fig. 1). The organization pattern of the linear array is not dependent on the expression level of the WAVE complex (Fig. 2, C and D; and Fig. S2, D and E), suggesting that the WAVE complex is a core and rate-limiting component of these linear arrays. Unlike N-WASP, which undergoes phase separation to form focal liquid-like protein condensates (Banjade and Rosen, 2014; Case et al., 2019), WAVE complex monomers form ordered linear arrays in which the subunits exhibit a fixed position relative to one another (Fig. 2, F–H). Similar to microtubules that align multiple protofilaments in a naturally curved pattern closing at the other end to form a tube, these linear arrays of the WAVE complex also exhibit a multilayered organization (Fig. 3) whose lateral extent (and thus the width of the polymer) depends on the associated negative membrane curvature, possibly forming a half tube-like structure (Fig. 7 B; Chrétien et al., 1995; Gov and Gopinathan, 2006; Nogales and Wang, 2006). Upstream activators such as Rac, Arf1/6, and PIP₃ are required for membrane curvature to recruit the WAVE complex (Fig. 5, F–H and Fig. 7 A). Through computational simulations, we show that WAVE complex’s linear organization and preference for negative curvature are both essential features for robust lamellipodia formation (Fig. 6).

The linear organization pattern of the WAVE complex plays a central role in lamellipodia formation. For neutrophils and many other cells, depletion of the WAVE complex abolishes cells’ ability to build lamellipodia (Graziano et al., 2019; Law et al., 2013; Leithner et al., 2016; Whitelaw et al., 2020). In some cell types, like Dictyostelium, WASP can compensate for the lack of Scar/WAVE by organizing into a linear pattern that builds lamellipodia (Veltman et al., 2012). However, in many other cell types, including the neutrophil-like cells studied here, neither WASP nor N-WASP can rescue lamellipodia in the absence of the WAVE complex (Buracco et al., 2024; Graziano et al., 2019). In these cells, WAVE plays a central role in templating lamellipodia formation—the WAVE complex stimulates activation of the Arp2/3 complex, but the Arp2/3 complex is not required for WAVE’s ability to generate lamellipodia (Pipathsouk et al., 2021); neither the Arp2/3 complex nor actin assembly is needed for the WAVE complex to form linear arrays at the cell membrane (Buracco et al., 2024; Pipathsouk et al., 2021). Furthermore, our results show that WAVE also appears to be a core component and rate-limiting for this assembly (Fig. 2, C and D; and Fig. S2, D and E). Although these linear arrays seem to be highly regular (Fig. 2) and likely self-organizing, purified WAVE complex does not oligomerize on its own (Chen et al., 2010). In the presence of upstream activators, motility mix, and complete cytosol, the WAVE complex initiates the actin polymerization but fails to form any self-organizing distributions (Chen et al., 2010; Koronakis et al., 2011; Lebensohn and Kirschner, 2009). One possibility is that these previous assays fail to provide the strong negative membrane curvature that we have found to be required for WAVE complex membrane recruitment (Fig. 4); previous assays have used positively curved surfaces like vesicles or lipid-coated beads (Humphreys et al., 2016; Lebensohn and Kirschner, 2009) or planar lipid bilayers (Banjade and Rosen, 2014; Lebensohn and Kirschner, 2009). Another possibility is that the WAVE complex is an obligate component of the linear array but copolymerizes with other essential components, similar to the way alpha- and beta-tubulin need one another to form microtubules. The PRDs of both WAVE2 and Abi2 may play a role in the formation of this array (Fig. 5, D and E; Fig. 7 A; and Fig. S3 I). In Dictyostelium, there could be an independent linear scaffold that recruits either WAVE complex or N-WASP to support lamellipodia formation. Our work suggests a regular, possibly polymer-like arrangement of the WAVE complex at the tip of lamellipodia. Future higher resolution work, such as in situ cryo-electron tomography, will be required to understand the precise arrangement of the WAVE complex during lamellipodial formation. This approach may also reveal the biochemical basis of the proposed array.

In addition to the membrane-bound upstream activators of the WAVE complex, such as Rac, Arf, and phosphoinositide (Fig. 5, F–H and Fig. 7 A; Koronakis et al., 2011; Oikawa et al., 2004; Suetsugu et al., 2006), our work suggests an additional requirement, strong negative curvature, for WAVE complex membrane recruitment. Negative curvature is not the only determinant—positive curvature along the other axis (corresponding for example to the lagging region of a lamellipod) also plays a role, but here a much wider range of curvatures are tolerated (Fig. S3, A–C) compared with the strict requirements for strong negative curvature (Fig. 4). This negative curvature dependence could also relate to and help explain some of the emergent behaviors in cell migration. For example, when cells undergo a collision with an obstacle, they cease migration and change directions (Roycroft and Mayor, 2016; Weiner et al., 2007; Yamada and Sixt, 2019). This barrier avoidance could be achieved through several possible mechanisms, including forces acting at the leading edge (Begemann et al., 2019; Diz-Muñoz et al., 2016; Nakamura, 2024; Prass et al., 2006; Saha et al., 2018; Uray and Uray, 2021) or the nucleus (Lomakin et al., 2020; Long and Lammerding, 2021; Venturini et al., 2020), cell stalling (Brunner et al., 2006), and/or perturbation of cell shapes and curvatures (Fig. 7 C; Begemann et al., 2019; Sadhu et al., 2024; Sadhukhan et al., 2025; Sitarska et al., 2023). If membrane curvature is a primary input to barrier avoidance, then the barriers will not be recognized if they do not change the morphology of the lamellipodia, in particular the strong negative curvature. Consistent with this idea, lamellipodia do not stall at the necks of membrane invaginations or nanopatterns, since the overall negative curvature of the lamellipodia is preserved at these barriers (Fig. 4 E; Fig. S3, A and C; and Videos 5, 6, and 7). Membrane curvature could also enable cell repolarization after collision, as computational modeling shows that negative membrane curvature-sensing proteins that recruit the actin cytoskeleton suffice to guide cell reorientation and spreading by selectively aligning with the substrate (Begemann et al., 2019; Sadhu et al., 2021; Sadhu et al., 2024; Sadhukhan et al., 2025).

How the WAVE complex recognizes negative curvature is not known. The WAVE complex does not have any well-characterized curvature-sensing motifs and may require interactions with other curvature-sensing partners. One family of candidates are the curvature-sensing BAR domain proteins that interact with the WAVE complex at the tips of lamellipodia (Bai and Grant, 2015; Frost et al., 2009; Gligorijevic et al., 2012; Miki et al., 2000; Nishimura et al., 2021; Pipathsouk et al., 2021; Suetsugu et al., 2006; Zhao et al., 2011). Given the possible redundancy of the BAR domain protein family, we removed all four of the negative curvature-sensing I-BAR proteins or all three of the positive curvature-sensing TOCA proteins and show that neither of them are required for lamellipodia formation (Pokrant et al., 2023) or WAVE complex linear organization and curvature sensation at the cell membrane (Fig. 5, A–C and Fig. S3, F–H). These results rule out the strongest candidates for WAVE complex curvature sensation. It is possible that the WAVE complex may interact with additional or unknown curvature-sensitive proteins (Frost et al., 2009; Hanawa-Suetsugu et al., 2019; S. Liu et al., 2015; Pykäläinen et al., 2011). Alternatively, assembly of the WAVE complex into a linear array could present the individual WAVE complexes in a geometry that is amenable to curvature sensation. Other curvature-sensing proteins, like septins or the BAR domain family, have been shown to exhibit dramatically different curvature preferences following oligomerization compared with the preferences of the isolated monomers (Bridges et al., 2016; Nepal et al., 2021).

Lamellipodia are frequently observed protrusive structures for a wealth of migratory and morphogenetic contexts (Fritz-Laylin et al., 2017a; Ibarra et al., 2006; Kunda et al., 2003; Leithner et al., 2016; Pollard and Borisy, 2003; Rakeman and Anderson, 2006; Weiner et al., 2006; Yolland et al., 2019), suggesting a potential value of these sheet-like protrusions in these settings. The broad actin networks that compose lamellipodia can generate greater protrusive forces compared with filopodia and ensure uniform advancing of the cell membrane (Dimchev et al., 2017; Fritz-Laylin et al., 2017b; Gardel et al., 2010; Pipathsouk et al., 2021), while their flat and thin geometry could enable efficient barrier avoidance, spreading along surfaces (Sadhu et al., 2023a), and regulation of membrane reservoirs (Gauthier et al., 2011; Mueller et al., 2017). The flat, sheet-like membrane structures built by the WAVE complex are also involved in other morphogenetic processes. During macropinocytosis/phagocytosis, the WAVE complex assembles at the rim of the macropinocytic/phagocytic cups (King and Kay, 2019; Seastone et al., 2001; Veltman et al., 2016). This discrete distribution depends on upstream biochemical activators like Ras and PIP₃ (Buckley et al., 2020; Yang et al., 2021); our results suggest that this focused accumulation could also be informed by the local membrane curvature of the macropinosome.

Without a defined organization structure, the default protrusive pattern for NPF-based actin assembly at membranes is finger-like membrane protrusions (Liu et al., 2008). To generate different organizations of actin networks, in vitro biochemical reconstitutions have used the same NPF artificially arrayed into different patterns, like points to generate finger-like actin networks or lines to generate sheet-like actin networks (Boujemaa-Paterski et al., 2017; Carlier et al., 2003). In cells, different NPFs are used for different morphological structures, likely because these NPFs have different patterns of self-organization: N-WASP undergoes phase separation and forms focal protein droplet structures that orchestrate the finger-like invadopodia and filopodia formation (Banjade and Rosen, 2014; Case et al., 2019), while the WAVE complex oligomerizes into linear arrays that participate in a wide variety of membrane structures involving the flat, sheet-like membrane geometries (Fig. 7 B; Gov and Gopinathan, 2006; Pipathsouk et al., 2021; Seastone et al., 2001; Veltman et al., 2016; Yang et al., 2021). In future work, it will be important to define how other rules of nucleator (and other actin regulators) self-organization underlie other actin structures, including the cell cortex and stress fibers (Bovellan et al., 2014; Hotulainen and Lappalainen, 2006; Lehtimäki et al., 2021; Svitkina, 2020; Tojkander et al., 2012).

HL-60 cells were cultured in RPMI 1640 media supplemented with l-glutamine and 25 mM HEPES (MT10041CV; Corning) containing 10% (vol/vol) heat-inactivated FBS (16140071; Gibco BRL). Cultures were maintained at a density of 0.2–1.0 million cells/ml at 37°C/5% CO₂. HL-60 cells were differentiated with 1.3% (vol/vol) DMSO (358801; Santa Cruz Biotechnology) in culture media for 5 days before experiments. HEK293T cells were grown in DMEM (SH30243.01; Cytiva) containing 10% (vol/vol) heat-inactivated FBS and maintained at 37°C/5% CO₂.

HEK293T cells were seeded into 6-well plates (2.5 ml per well) and grown until about 70% confluent. For each well, 1.5 µg pHR vector (containing the appropriate transgene), 0.167 µg vesicular stomatitis virus-G vector, and 1.2 µg cytomegalovirus 8.91 vector were mixed and prepared for transfection using TransIT-293 transfection reagent (Mirus Bio) as per the manufacturer’s instructions. After transfection, cells were grown for ∼48 h, after which virus-containing supernatants were harvested and concentrated 30–40-fold using an Lenti-X Concentrator (Takara Bio) according to the manufacturer’s instructions. Lentivirus was used immediately or stored at −80°C. HL-60 cells were transduced by overnight incubation of 0.32 million cells with 4 µg/ml polybrene and 125 μl of concentrated virus. Cells expressing desired transgenes and expression levels were isolated using FACS (FACS Aria2 or FACS Aria3; BD Biosciences). For the cells expressing the fluorescent standard candles, the lowest expression cells above the background were selected to avoid multiple candle aggregation (more than one candle in each spot).

HEK293T cells were seeded into 24-well glass-bottom plates (P24-1.5H-N; Cellvis) and grown until about 70% confluent. For each well, 0.5 µg plasmid containing the appropriate transgene was used for transfection using TransIT-293 transfection reagent (Mirus Bio) according to the manufacturer’s instructions. After transfection, cells were grown for 24 h before imaging.

Plasmids were constructed using standard molecular biology protocols. DNA segments were PCR amplified and cloned into a pHR lentiviral backbone and driven by a promoter from spleen focus-forming virus (SFFV) via standard Gibson assembly. The membrane-tagged Hotag3 construct was modified from Chung et al. (2023), by inserting a membrane-bounded Fyn tag into the N terminus of the original coding sequence. The fluorescent standard candle constructs were modified from Akamatsu et al. (2020), by switching the fluorescent protein from tag2GFP to eGFP, followed by subcloning into the pHR lentiviral backbone. The ∆PRD constructs were made by deleting the PRD region (298–407 for WAVE2 and 174–409 for Abi2) of the eGFP-tagged WAVE complex subunits.

Protein content from one million HL-60 cells was extracted by chilled TCA precipitation and resuspended in 2× Laemmli sample buffer. Protein samples were separated via SDS–PAGE, followed by transfer onto PVDF membranes (#1620177; BIO-RAD). Membranes were blocked at RT for 1 h in a 1:1 solution of TBS (20 mM Tris, 500 mM NaCl, pH 7.4), and Odyssey Blocking Buffer (#927-40000; LI-COR), followed by overnight incubation at 4°C with primary antibodies in a solution of 1:1 TBS + 0.2% wt/vol Tween 20 (TBST) and Odyssey Blocking Buffer. Membranes were then washed three times with TBST and incubated for 45 min at RT with secondary antibodies diluted 1:10,000 in 1:1 solution of Odyssey Blocking Buffer and TBST. Membranes were then washed three times with TBST, one time with TBS, and imaged using an Odyssey Fc (LI-COR). Primary antibodies Sra1 (rabbit; 1:200; #NBP2-16060; Novus) and secondary antibodies IRDye 680RD goat anti-mouse (926-68070; 1:10,000; LI-COR) and HRP-conjugated goat anti-rabbit (1:5,000, 65-6120; Invitrogen) were used for this study.

Membrane-labeling solution was made with CellMask Deep Red (Invitrogen) freshly diluted 1:500 in imaging media (Leibovitz’s L-15 [Gibco] with 2% FBS). 1 ml of differentiated HL-60 cells was spun down at 200 × g for 3 min and resuspended in 500 μl of the labeling solution. Immediately after resuspension, cells were washed twice with imaging media before plating.

To label the Halo-Sra1 cells, 0.05 nM JF549 (GA1110; Promega) dye was mixed with 10 nM JF646 dye (GA1120; Promega). To label the Halo-CAAX cells, 0.001 nM JF549 with 10 nM JF646 was used. Cells were incubated with the dyes diluted in imaging media in a 37°C/5% CO₂ incubator for 20 min followed by two washes with imaging media.

The desired number of wells in a 96-well plate (MGB096-1-2-LG-L; Matrical, Inc.) or an 8-well Lab-Tek II chamber (155409; Thermo Fisher Scientific) were coated with 100 or 200 μl of 40 µg/μl porcine fibronectin (prepared from whole blood) diluted in imaging media for 20 min at RT, followed by two to three times with imaging media. Differentiated cells (∼0.8 million/μl) were resuspended in the imaging media of the same volume. 100 μl of cells per well were used for a 96-well plate, and 200 μl of cells per well were used for an 8-well chamber. For cell migration on beads, 200 μl of cells were mixed with 0.8 μl of 0.2-µm red fluorescent (580/605) carboxylate-modified microspheres (F8887; Invitrogen) or 1.5 μl of 0.5-µm red fluorescent (580/605) carboxylate-modified microspheres (F8887; Invitrogen) or 0.8 μl of 0.2-µm blue fluorescent (365/415) carboxylate-modified microspheres (F8805; Invitrogen; for Sra1-mCherry labeled TOCA 3×KO cells only) before plating. The imaging plate was then transferred to a 37°C/5% CO₂ incubator for 15 min to allow cells to adhere. The plate was next transferred to the microscope, which had been preheated to 37°C for imaging. For chemoattractant stimulation, a 2× stock of 50 nM N-formyl-L-methionyl-L-leucyl-L-phenylalanine (fMLP; Sigma-Aldrich) was added. For F-actin inhibition, a 2× stock of 1 µM latB (Sigma-Aldrich) and 25 nM fMLP was used. All initial stocks were dissolved in 100% dry DMSO and freshly diluted in imaging media before experiments.

Coverslips with nanopatterns were obtained from Cail et al. (2022). The nanopatterned coverslips were held in Attofluor Cell Chambers (A7816; Thermo Fisher Scientific) and coated with 200 μl of 40 µg/μl porcine fibronectin dissolved in imaging buffer for 20 min at RT. The fibronectin solution was then removed and washed twice with the imaging buffer. 500 μl of differentiated HL-60 s (∼0.8 million/ml) labeled with CellMask Deep Red (Invitrogen) were plated onto the coverslip. The chamber with the coverslip was then transferred to a 37°C/5% CO₂ incubator for 15 min to allow cells to adhere. The coverslip was then gently washed twice with imaging buffer to remove unattached cells and finally exchanged for 500 μl of fresh imaging media. For the agarose compression experiment, the imaging media was removed, and a precast agarose pad with 1 µM latB was gently placed on top of the cells on nanopatterns right before imaging. For cell migration on nanopatterns, 20 nM fMLP was added to cells before imaging.

Cells were prepared and plated on a fibronectin-coated 96-well plate as described above. 2× drug solution diluted in imaging buffer was added to cells for 40 min and imaged with TIRF microscope. Drugs used in the experiments are C. difficile toxin B (#SML1153; 1 µg/ml; Sigma-Aldrich), EHT 1864 (#HY-16659; 25 µM; MedChemExpress), Brefeldin A (#SIALB6542; 20 µM; Sigma-Aldrich), Chlortetracycline (#C4881l; 100 µM; Sigma-Aldrich), and Duvelisib (#HY-17044; 5 µM; MedChemExpress).

Cells were fixed by adding 2× fixation buffer (4% glutaraldehyde [Sigma-Aldrich] in PBS) and incubation at RT for 10 min. Cells were washed twice with PBS and then quenched with 0.1% sodium borohydride (Sigma-Aldrich) for 7 min, followed by two or three 10-min PBS washes.

Cells were fixed as described above and stained with Alexa Fluor 647 phalloidin (5 μl/ml; A22287; Invitrogen) in PBS for 30 min and washed twice with PBS buffer.

Cells were fixed as described above and blocked with 3% BSA and 0.1% Triton X-100 diluted in PBS for 1 h at RT. Primary antibody was diluted in blocking solution at 4°C overnight, washed three times with PBS, and then incubated with the secondary antibody and washed three times with PBS. WAVE2 (rabbit; #3659; Cell Signaling Technology) primary antibody was used at 1:50 dilution with secondary antibody goat anti-rabbit IgG conjugated to Alexa Fluor 647 (#A-21245; Invitrogen) at 1:1,000 dilution.

A solution of 4% low-melt agarose (A-204; Gold Biotechnology) was made in imaging media and microwaved in a loosely capped conical tube placed in a water reservoir. Heating was done in short increments to promote melting while preventing the solution from boiling over. Once completely melted, the gel was kept at RT to allow cooling while preventing solidification. 1 µM of latB was then added to the gel solution after cooling. The mold for gel casting was made by cutting the back end (that goes into a pipette) of a 1,000-μl pipette tip to get a cylinder of ∼0.7-cm tall. The mold was then placed in a 35-mm dish with the uncut (smooth) edge facing the bottom. ∼700 μl of gel solution was added to each mold and left at RT for solidification for 10–15 min. The mold was then removed carefully to let the gel keep solidifying for another 30 min at RT. To compress the cells, the imaging solution was removed, and the agarose pad was handled with a tweezer and placed directly on top of the cells.

All TIRF and epifluorescence images (except Fig. 2, F and G) were acquired at 37°C/5% CO₂ with the DeltaVision OMX SR microscope (GE Healthcare) with a 60×/1.42 NA oil Plan Apochromat objective (Olympus) and 1.518 refractive index oil (Cargille). Images were acquired with Acquire SR software and processed with softWoRx. TIRF-SIM images were reconstructed using OMX SI Reconstruction with the default parameters. TIRM imaging was performed using the ring TIRF light path on the DeltaVision OMX SR. Epi-fluorescence imaging was performed with the conventional fluorescent light path on the DeltaVision OMX SR. FRAP experiments were performed using the ring TIRF light path combined with the FRAP/photoactivation module on the DeltaVision OMX SR.

The single-particle tracking experiments (Fig. 2, F and G) were performed on a Nikon Eclipse Ti microscope equipped with a Borealis beam-condition unit (Andor Technology), a 100× Plan Apochromat TIRF 1.49 NA objective (Nikon), and an iXon Ultra EMCCD camera. Environmental control (37°C/5% CO₂; Okolab) was used. Acquisition was controlled with Micro-Manager.

The WAVE complex at the tips of lamellipodia or the other lamellipodial components (PAK-PBD and phalloidin-stained actin) was segmented based on the fluorescence intensity from the desired channels. Segmented regions at the desired location were manually selected and quantified with the Python package scikit-image (Van Der Walt et al., 2014). For identification of the WAVE complex nanorings, each segmented region was fitted to a circle and filtered based on the fitting score (Fig. S4), followed by quantification as described above. For the membrane-bounded droplets, the images were segmented based on the fluorescence intensity from the Fyn-Hotag3 channel. The radius was calculated by taking the average of the long and the short axes of the segmented region. The fluorescent intensities of the segmented regions were binned (data points showing the average and standard deviation) and plotted against the length of the structure (length of the lamellipodia or radius for the nanorings and droplets). The linear correlation was measured by the Pearson correlation coefficient.

To compare the WAVE complex organization pattern between the tips of lamellipodia and nanorings (Fig. 1 H), images for each cell were first acquired during migration under fMLP treatment and then acquired again after latB treatment. For quantification, linescans across the lamellipodia at multiple sites for each cell were averaged and fitted to a Gaussian curve with one peak; linescans across multiple nanorings for each cell were averaged and fitted to a Gaussian curve with two peaks. The heights of the Gaussian curves for the lamellipodia and nanorings (average of two peaks) in each cell were normalized and plotted.

To calculate the WAVE complex unit length intensity, the WAVE complex fluorescent intensity at the tips of lamellipodia or nanorings was quantified as described above and then divided by the length of lamellipodia or the radius of the nanorings. The overall Hem1-eGFP expression level for each cell was determined by the mean Hem1-eGFP fluorescent intensity segmented and calculated through the epi-fluorescence channel. The WAVE complex unit length intensity against the Hem1-eGFP expression level for each cell was then plotted and linearly fitted to determine the Pearson correlation coefficient. The linear expectation curve for a protein binding to a pre-existing polymer was plotted using two data points: the background fluorescence (where the Hem1-eGFP fluorescence level is 0) of the TIRF images and the WAVE complex intensity at the lowest measured Hem1-eGFP expression level.

Hem1 KO cells expressing different levels of Hem1-eGFP were fixed and immunostained for WAVE2 protein. The WAVE complex expression level for each cell was determined by segmenting and calculating the mean WAVE2 immunofluorescence level, plotted against the corresponding Hem1-eGFP fluorescence level. Pearson correlation coefficient was calculated to determine if the WAVE complex expression level scales with Hem1-eGFP fluorescence level.

The diffusion coefficient of the WAVE complex (Halo-Sra1) and the diffusive Halo-CAAX were calculated as described above and plotted.

Fixed cells were imaged five consecutive times and bleach-corrected by fitting into a decaying exponential curve. The fluorescent candles were then segmented based on multi-otsu thresholding. Each fluorescent spot was fitted with a 2D Gaussian, and the background-independent signal was calculated using the equation for volume under the fitted Gaussian curve. The spots were then filtered based on the fitting quality and the amplitude of the Gaussian curve to remove the spots that could contain more than one candle. Finally, the spots were tracked using the Python package Trackpy, and the mean fluorescent intensities of the spots that could be detected for more than three frames were used for histogram plotting. To generate the fluorescent standard curve, the median of the histogram distribution for each fluorescent candle was calculated by kernel density estimation and plotted against the fluorescent molecules contained in each structure and finally fitted through linear regression.

HL-60 cells stably expressing the 60mer and 120mer were used for determining the fluorescence difference after fixation. For fixed cells, cells were first fixed as described above and then imaged with ring-TIRF microscope. For unfixed cells, live cells were plated on fibronectin and then imaged with ring-TIRF microscope. The fixed and unfixed puncta were then quantified as described above, and the median fluorescent intensities for the fixed and unfixed cells were plotted and linearly fitted separately. The fixed versus unfixed fluorescence ratio was determined by the ratio between the slope coefficients of the fixed and unfixed linear curves.

Three groups of HL-60 cells expressing eGFP-Sra1 were imaged on TIRF-SIM on three different days. The unit length fluorescent intensities of eGFP-Sra1 at the tips of lamellipodia were quantified as described above and plotted as a histogram.

The WAVE complex fluorescent signal at the tips of lamellipodia was segmented and quantified as described above, then divided by the length of the lamellipodia. For each lamellipod, the number of the WAVE complex per micron was calculated based on the fluorescent standard curve, the ratio of fluorescently tagged WAVE subunit, as well as fixed versus unfixed fluorescence ratio determined as described above. The numbers of the WAVE complex per micron were then plotted as a histogram, where the mode of the distribution was calculated by kernel density estimation.

For the recovery pattern analysis (Fig. 3, F and G), a 1 × 30-pixel region along the middle region of the bleached array was extracted for each cell in Fiji, and the fluorescent intensities at each time point were then averaged between all the cells. For the recovery kinetics analysis (Fig. 3 H and Fig. S2 J), the normalized fluorescent intensity of the bleached region is determined using the mean fluorescent intensity of the top 10 pixels within the bleached region, normalized by the average fluorescent intensity (top 10 pixels) of a neighboring unbleached region at the corresponding time point. The half-time and amplitude of WAVE complex recovery were determined by plotting the normalized fluorescence within the bleached region as a function of time and fitting the data with an exponential function using the Stowers Institute ImageJ Plugins.

The beads under the cells were identified by finding the overlap between the beads and the cells. Beads were segmented based on the bead channel. And cells were segmented based on the eGFP-Sra1 channel. Tracking of the beads under the cells was performed using the Python package Trackpy. The fluorescent intensity of the WAVE complex/membrane dye on each bead was normalized by the WAVE signal/membrane dye surrounding that bead (not on the bead). The data used for plotting were filtered and aligned based on the peak WAVE complex signal. For tracking the fluorescent intensity of the WAVE complex (eGFP-Sra1 channel) on beads over time, the normalized WAVE complex and membrane dye signal on each bead was plotted as a function of time. For comparison between the WAVE complex and membrane signal, the average of three top fluorescent signals for each channel was calculated and then averaged for each cell.

Linescans of the membrane dye and the WAVE complex (eGFP-Sra1) channel were generated by projection of the fluorescent signal onto the long axis of each nanopillar, followed by 1D Gaussian smoothing of the linescan curve.

For Fig. 1 H and Fig. 4 D, paired two-tailed t tests were performed to compare different conditions/measurements in the same cells. For Fig. 2 H; Fig. 4 H; Fig. 5, B and C; and Fig. S3 I, unpaired two-tailed t tests were performed to compare different conditions/treatments on different groups of cells. For Fig. 5 E, a MANOVA test was performed to compare multiple dependent variables between the FL and the ∆PRD WAVE subunits. For Fig. S2 I, an ANOVA test was performed to compare the TIRF performance between three groups. Normality was assumed under the central limit theorem but was not formally tested.

Our theoretical model consists of a closed vesicle that is described by a triangulated network as shown in Fig. S5 A (Fošnaricˇ et al., 2019; Sadhu et al., 2023a; Sadhu et al., 2021; Sadhu et al., 2022; Sadhu et al., 2024). The vesicle membrane is a 2D surface, and we do not consider its thickness. The triangulated surface is formed by N vertices connected by bonds. The vertices of this triangulated surface are displaced by random Monte Carlo moves, driving the dynamics of the membrane. The bonds are also allowed to flip to allow diffusion of the vertices on the vesicle surface. In the simulation, the displacement of the vertices or the bond flip is accepted if the total energy of the system decreases by the displacement. However, if the energy is increased by the displacement, the move is only accepted with a probability that decreases exponentially with increased energy. We use here a coarse-grained model, such that we deal with length scales where the continuum description of the membrane is valid (i.e., larger than tens of nanometers) and do not include details of the molecular scale.

We consider four energy terms described below.

2. Protein–protein interaction energy: The protein complexes on the lipid membrane can bind with other neighboring complexes and form clusters. We consider two different cases here: An isotropic interaction between proteins that allows them to bind with proteins on neighboring nodes without any directional bias (Fošnaricˇ et al., 2019; Sadhu et al., 2023a; Sadhu et al., 2021; Sadhu et al., 2022; Sadhu et al., 2024), and an anisotropic or polymer-like interaction, where they bind to form linear chain-like aggregates. We added an extra constraint for the polymer-like interactions, such that a protein can only bind with a maximum of two of its neighbors.

For the polymer-like interaction, there is a possibility that a closed loop of proteins can form with three or more proteins. To avoid this, we do not allow all three vertices of a triangle to be simultaneously occupied with proteins. No constraints were used on clusters containing more than three proteins, as the probability that they will form closed loops over a single triangle sharply decreases as their size increases.

We did not introduce a bending energy for the shape of the linear polymer-like protein cluster itself (different from the bending energy of the membrane [Eq. 1]).

The total energy cost ∆E for a particular movement is calculated by adding all these individual energy costs. If this energy cost is negative, the movement is always accepted; otherwise, it is accepted with a probability exp(−∆E/k_BT) where k_B is the Boltzmann constant and T is the absolute temperature at which the system is kept. We express all the energy terms in units of k_BT for simplicity.

We perform Monte Carlo simulations of the vesicle with a total number of vertices N = 1,447 and the number of proteins N_c = 50, such that the density of proteins is ρ$≅$ 3.45%. We choose these values based on our previous study (Sadhu et al., 2021), where these values of parameters give lamellipodia-like structures for a wide range of E_ad and F (Fig. S5 F). We consider both the isotropic and polymer-like interaction cases and compare these results for a range of adhesion strength E_ad and the magnitude of active force F.

When isotropic interactions are allowed between the membrane proteins, a lamellipodia-like structure forms only when the adhesion strength E_ad and the magnitude of active force F are sufficiently large. In Sadhu et al. (2021), we showed this for small protein–protein interaction strength (Fig. S5 F).

In Fig. S5 B, we show a few snapshots for isotropic protein–protein interactions ω = 1k_BT for small E_ad and F, where no lamellipodial structures form. When we increase the interaction strength between proteins to ω = 10.0k_BT (Fig. S5 C) small buds and finger-like protrusions form.

Next, we consider polymer-like interactions between proteins and compare the results with the isotropic case. For small ω = 1k_BT, we note that there is no qualitative difference between the isotropic and polymeric cases (Fig. S5, B and D). However, as w increases, lamellipodia-like structures form, as shown in Fig. S5 E. Thus, polymer-like interactions increase the parameter range where lamellipodia can form compared with the isotropic interactions, as shown in Fig. 6.

For proteins with zero spontaneous curvature and isotropic clustering interactions, no lamellipodia formation was observed in earlier studies (Sadhu et al., 2021) even with large E_ad and F. The lamellipodia formation for isotropic interaction is mainly due to the aggregation of curvature-sensitive proteins close to the adhesive substrate or due to large active protrusive forces. For a protein with zero spontaneous curvature, since there is no such aggregation, there is no lamellipodia formation.

For polymer-like interaction among the proteins, there is a tendency to form linear clusters that might help them form lamellipodia. For small ω, the tendency to form linear aggregates is weak; thus, we choose here only large ω. We note that proteins form circular ring-like aggregation for small F (Fig. S5 H). For large F, these proteins pull the vesicle and form filopodia-like structures, and vesicles often get detached from the substrate if E_ad is not high enough (Fig. S5 H, two left-most snapshots corresponding to F = 2.0, 4.0k_BT/l_min respectively).

Here, we discuss the quantification of the different phases and the transition lines between these phases, as shown in Fig. 6.

To distinguish between a disordered cluster from lamellipodia, we measure the average number of clusters and plot it as a function of F for a given value of ω. We set the threshold to 4 such that if the average number of clusters is smaller (larger) than four, the phase will be considered as a lamellipodia (disordered protein cluster). We show in Fig. S5 I such an example for ω = 3.0k_BT. Other parameters are the same as in Fig. 6.

For large ω with isotropic interaction between proteins, the proteins can form buds for small value of F, while a lamellipodia-like structure is formed for large F. Thus, there is another transition line that separates the lamellipodia phase from the bud-like structures. In this case, even for small values of F, the average number of clusters is never larger than four. To distinguish between these two phases (buds and lamellipodia), we use additional criteria by quantifying the shape of the protein clusters. We measure the ratio between the quantity R_max to the square root of the area of the protein cluster, where R_max is defined as the maximum distance of a protein from the center of mass of the protein cluster to which it belongs. For a hemispherical aggregation, this value is smaller than 1/2, while for a perfect linear aggregation, this value can be >1. We set a threshold of 0.8, above which the shape looks like the elongated cluster at the leading-edge of a lamellipodia, while below this threshold, the proteins can form hemispherical buds (Fig. S5 J).

Fig. S1 shows the WAVE complex adopts a restricted linear pattern compared with WAVE complex activator (Rac) and lamellipodial components (actin). Fig. S2 shows the supplemental data for Hem1 KO, single-molecule tracking, molecular counting, and FRAP experiments. Fig. S3 shows the supplemental data for WAVE complex curvature sensing and molecular mechanism. Fig. S4 shows identifying WAVE complex nanoring structures by circle fitting. Fig. S5 shows the model description and supplementary data for computational simulation. Video 1 shows the WAVE complex organization pattern in HL-60 cells. Video 2 shows the single-molecule tracking on sparse labeled Halo-tagged cell lines. Video 3 shows the FRAP on WAVE complex at the tips of lamellipodia in a migratory HL-60 cell. Video 4 shows the FRAP on WAVE complex in a latB-treated HL-60 cell. Video 5 shows the TIRF imaging of an HL-60 cell migrating on 200 nm beads. Video 6 shows the TIRF-SIM imaging of an HL-60 cell migrating on 200 nm beads. Video 7 shows the HL-60 cells migrating on nanopatterns.Video 8 shows the comparison between isotropic versus polymer-like interaction for small ω and F. Video 9 shows the comparison between isotropic versus polymer-like interaction for large ω. Video 10 shows the vesicle with zero spontaneous curvature proteins with polymer-like interaction, comparing small and large F cases.

All image analysis code and CSV files of all data are available at https://github.com/MuziyueWu/WAVE.

We thank the Woolfson lab and the Weiner lab for helpful discussion. We thank Sue Sim, Dong Li, and Evelyn Strickland for a critical reading of the manuscript. We thank the Drubin lab and the Shu lab for providing the plasmids. We thank the Faix lab for providing the I-BAR 4×KO B16-F1 cell line. We thank Rachel Brunetti for providing the TOCA 3×KO cell line. We thank Kari Herrington and SoYeon Kim of the UCSF Imaging Core for their microscopy expertise.

This work was supported by an American Heart Association Predoctoral Fellowship 23PRE1018810 (M. Wu.), the National Institute of General Medical Sciences GM118167 (O.D. Weiner), National Science Foundation/Biotechnology and Biological Sciences Research Council grant 2019598 (O.D. Weiner and D.N. Woolfson), the National Science Foundation Center for Cellular Construction (DBI-1548297), and the Royal Society Wolfson Visiting Fellowship (N.S. Gov). Data for this study were acquired at the Center for Advanced Light Microscopy at UCSF on an OMX-SR obtained using grants from the NIH (5R35GM118119), the UCSF Program for Breakthrough Biomedical Research funded in part by the Sandler Foundation, the UCSF Research Resource Fund Award, and HHMI. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. N.S. Gov is the incumbent of the Lee and William Abramowitz Professorial Chair of Biophysics, and acknowledges support by the Israel Science Foundation (Grant No. 207/22).

Author contributions: M. Wu: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, software, supervision, validation, visualization, and writing—original draft, review, and editing. R.K. Sadhu: formal analysis, methodology, resources, software, visualization, and writing—review and editing. K. Meyer: methodology. Z. Tang: data curation, formal analysis, software, and writing—review and editing. P. Marchando: resources. D.N. Woolfson: funding acquisition, supervision, and writing—review and editing. N.S. Gov: conceptualization, formal analysis, investigation, methodology, supervision, and writing—review and editing. O.D. Weiner: conceptualization, funding acquisition, project administration, supervision, visualization, and writing—original draft, review, and editing.