SPIN90 associates with mDia1 and the Arp2/3 complex to regulate cortical actin organization

Cell shape is controlled by the submembranous cortex, an actomyosin network mainly generated by two actin nucleators: the Arp2/3 complex and the formin mDia1. Changes in relative nucleator activity may alter cortical organization, mechanics and cell shape. Here we investigate how nucleation-promoting factors mediate interactions between nucleators. In vitro, the nucleation-promoting factor SPIN90 promotes formation of unbranched filaments by Arp2/3, a process thought to provide the initial filament for generation of dendritic networks. Paradoxically, in cells, SPIN90 appears to favour a formin-dominated cortex. Our in vitro experiments reveal that this feature stems mainly from two mechanisms: efficient recruitment of mDia1 to SPIN90–Arp2/3 nucleated filaments and formation of a ternary SPIN90–Arp2/3–mDia1 complex that greatly enhances filament nucleation. Both mechanisms yield rapidly elongating filaments with mDia1 at their barbed ends and SPIN90–Arp2/3 at their pointed ends. Thus, in networks, SPIN90 lowers branching densities and increases the proportion of long filaments elongated by mDia1. Cao et al. show that SPIN90 enhances formation of Arp2/3-mediated unbranched filaments and promotes a formin-based cortex by recruiting mDia1 or forming a ternary SPIN90–Arp2/3–mDia1 complex.

L iving cells have a remarkable ability to change shape during physiological processes such as division, migration and differentiation. These shape changes are governed by mechanical changes in the cortex, a thin network of actomyosin below the membrane 1 . Changes in cortical mechanics can originate from changes in myosin activity 2 or cortex architecture, which arise from changes in actin filament length 3 or network organization 4 . One potential mechanism to control cortex architecture involves regulation of actin nucleators. Indeed, in vitro-in the presence of profilin-formins generate longer filaments than those created by Arp2/3-mediated branching 5,6 and in cells, single-molecule experiments suggest a similar trend 7 . Furthermore, actin nucleators generate varied network topologies, ranging from highly branched networks generated by the Arp2/3 complex to linear arrays generated by formins and Ena/VASP proteins. Thus, a switch in the dominant F-actin nucleator might also alter network organization. However, little is known about how nucleator activity is controlled to change network topology and mechanics.
Many cellular actin-based structures such as the cortex, the leading edge of migrating cells, phagocytic cups and intercellular junctions require both formins and the Arp2/3 complex for their formation [8][9][10][11][12] . This is surprising, because the actin networks they generate differ extensively in their topology, protein interactors, dynamics and force generation 4,11,13,14 . Some reports have shown synergistic action of pointed end nucleators (such as Spire or adenomatous polyposis coli) with barbed end nucleators (such as formins) [15][16][17] , whereas others have shown sequential action of nucleators 18 . Overall, these observations underscore the importance of nucleator crosstalk for the generation of functionally optimal actin structures in cells.
In addition to RhoGTPases, nucleation-promoting factors (NPFs) are involved in activating nucleators or maintaining their activity 19 . The best studied NPF is probably the wave regulatory complex (WRC), which consists of five subunits (SRA1, NAP1, ABI1, BRK1 and WAVE2) and activates the Arp2/3 complex to generate branched actin networks 19 . After the WAVE complex detaches from Arp2/3, another NPF, cortactin, protects Arp2/3 against debranching 19 . Some NPFs can interact with multiple nucleators, making them prime candidates as mediators of interplay. IQGAP1 can maintain the activity of mDia1 via its C-terminal Dia1-binding region (DBR) (Fig. 1a) and promote Arp2/3 activity by interacting with N-WASP and the WAVE complex via its N-terminal calponin homology domain [20][21][22] . Another NPF, SPIN90 (also known as DIP, WISH or NCKIPSD), has been reported to interact with Arp2/3 in some studies and with formins in others. SPIN90 forms a complex with Arp2/3 to stimulate formation of unbranched filaments 23 .
In addition to providing the initial filament necessary for generation of dendritic networks by the WRC and Arp2/3 24 , recent work suggests that SPIN90 competes with the WRC to modulate the degree of branching in networks 25 . Furthermore, SPIN90 can interact with the diaphanous related formins (DRF) mDia1 and mDia2 via its leucine-rich repeat (LRR) and/or SRC homology 3 (SH3) domains 26,27 (Fig. 1a) and, surprisingly, it inhibits actin filament elongation by mDia2 but not mDia1 26 .
In this study, we examine how NPFs regulate nucleator activity in the actin cortex to control its organization, assembly kinetics and mechanics. We show that IQGAP1 controls the activity of the formin mDia1, WRC regulates Arp2/3 branching activity and SPIN90 mediates an unexpected synergistic action between Arp2/3 and mDia1.

Results
Several NPFs localize to the actin cortex. Previous proteomic analyses revealed the presence of two actin nucleators in the cortex of M2 blebbing melanoma cells, the Arp2/3 complex and the formin mDia1 9 . Several NPFs that could regulate these nucleators were also present (Supplementary Table 1): the WRC, IQGAP1, cortactin, SPIN90 and flightless-I homologue (Fli-1), an NPF that prevents formin autoinhibition 28 . Published proteomic datasets indicate that these NPFs are also expressed in HeLa cells (Supplementary Table 2), suggesting they may have a general role in controlling cortical nucleator activity [29][30][31] .
We examined NPF localization in M2 melanoma cell blebs and metaphase HeLa cells. Blebs provide a snapshot into the cortex life cycle 32 , whereas mitosis represents a key physiological function of the cortex. All NPFs identified in proteomics localized to the cortex of mitotic HeLa cells and retracting blebs, in which a cortex reforms de novo by nucleation 9 (Fig. 1b,c and Extended Data Fig. 1a-d).
Thus, these NPFs are promising candidates for controlling the activity of cortical nucleators.
We concentrated on WRC and SPIN90 because they mediate the transition from branched to unbranched actin networks nucleated by Arp2/3, and IQGAP1 and SPIN90 because they may coordinate the activity of Arp2/3 and mDia1.

SPIN90 depletion mimics mDia1 depletion in blebbing cells.
To determine how NPFs modulate nucleator activity, we examined their effect on bleb size in M2 cells, knowing that Arp2/3 depletion results in small blebs, whereas mDia1 depletion results in large blebs 9 . As expected, IQGAP1 depletion resulted in more cells with large blebs ( Fig. 1d and Supplementary Fig. 1h,j), consistent with a role in maintaining mDia1 activity 20 , and depletion of WRC subunits resulted in small blebs (Fig. 1e,f and Supplementary Fig. 1h,k,l), consistent with its known role in regulating Arp2/3. Surprisingly, SPIN90 depletion resulted in large blebs (Fig. 1d,f and Supplementary Fig. 1h,i), suggesting that SPIN90 cooperates with mDia1 directly or indirectly rather than inhibiting DRFs as previously reported 26 . These results (summarized in Supplementary Table 3) led us to focus on SPIN90.
SPIN90 depletion perturbs cell proliferation. Previous work showed that mDia1 depletion increased cell death, whereas Arp2/3 depletion did not 9 . When we examined how NPF depletion affected proliferation in HeLa cells, we found that depletion of WRC subunits did not increase cell death, similar to Arp2/3 depletion, but IQGAP1 depletion led to a twofold increase in cell death (Fig. 1g,h and Supplementary Fig. 1a-g). This latter effect was probably mediated by interaction with mDia1, because expression of the mouse IQGAP1 DBR domain in IQGAP1-depleted cells decreased cell death to near baseline levels (Fig. 1a,i). Similar to mDia1 depletion, SPIN90 depletion increased cell death threefold ( Fig. 1h and Supplementary Fig. 1a-g). We confirmed the specificity of depletion by expressing full-length mouse SPIN90 in SPIN90-depleted cells (Fig. 1i). Examination of changes in mRNA transcript abundance by quantitative PCR confirmed that the effect of SPIN90 and IQGAP1 depletion was not due to indirect regulation of NPFs or nucleators at the transcriptional level ( Supplementary Fig. 2). Thus, mDia1 activity regulated by IQGAP1 and SPIN90 is necessary for proliferation (Supplementary Table 3).

SPIN90 depletion decreases cortical mesh size in blebs.
We then examined how NPF depletion affects the organization of F-actin in the cortex of M2 cell blebs. Previous work showed that depletion of mDia1 led to large gaps in the cortex, whereas Arp2/3 depletion led to no clear change in actin filament density 9 . IQGAP1 depletion led to a twofold increase in the proportion of gaps with diameters larger than 140 nm (Fig. 2a,b and Supplementary Fig. 3), similar to mDia1 depletion. NAP1 depletion did not change gap diameter distribution, similar to ACTR2 depletion (Fig. 2a,b and Supplementary Fig. 3). SPIN90 depletion led to a visibly denser cortex (Fig. 2a) and a 20% increase in the proportion of gaps less than 30 nm in diameter ( Fig. 2b and Supplementary Fig. 3). This was unexpected, because previous cell-scale assays showed that SPIN90 depletion phenocopied mDia1 depletion (Fig. 1). However, SPIN90 has been proposed to compete with WRC for Arp2/3 to regulate the degree of branching 25 . Depletion of SPIN90 might therefore lead to a denser cortical network because of increased branching. Thus, SPIN90 may mediate crosstalk between mDia1 and Arp2/3 (Supplementary Table 3).
SPIN90 governs cortical actin density and thickness in mitotic cells. We next determined whether NPF depletion affected cortical thickness and density during mitosis in HeLa cells by analysing the fluorescence profile of an F-actin fluorescent reporter with respect to the plasma membrane 33 (Fig. 2c). Previous work showed that depletion of mDia1 leads to a decrease in cortical thickness, whereas depletion of Arp2/3 does not affect thickness 3 . Depletion of IQGAP1 and NAP1 had no impact on cortical thickness or density (Fig. 2d,e). Similar to mDia1 depletion 3 , SPIN90 depletion substantially decreased cortical thickness by about 25% (Fig. 2d). However, SPIN90 depletion also led to an increase of about 50% in cortex density (Fig. 2e), something not observed with mDia1 depletion 3 , but consistent with the smaller cortical gap sizes measured by scanning electron microscopy (SEM) (Fig. 2a,b).

SPIN90 controls cortical actin accumulation rate in mitotic cells.
To quantitatively examine the role of NPFs in regulating nucleator activity, we measured the cortical F-actin accumulation rate in blebs generated by ablation of the cortex of a metaphase HeLa cell using a pulsed-UV laser 34 (Fig. 3a and Extended Data Fig. 2). Previous work showed that depletion of mDia1 decreases F-actin accumulation rate, whereas depletion of Arp2/3 subunits increases it 9 . Depletion of IQGAP1 and NAP1 mirrored the respective effects of mDia1 and Arp2/3 depletion, consistent with their proposed roles (Fig. 3c,d). SPIN90 depletion led to a notable decrease in actin regrowth rate, providing further evidence for cooperation with mDia1 (Fig. 3b,d). Together, these results indicate that NPFs regulate actin-network growth kinetics in the mitotic cortex (Supplementary Table 3).
SPIN90 depletion stiffens the mitotic cortex. Recent work has shown that modulating nucleator activity affects cell mechanics 2,3,7 . To probe cortical mechanics, we indented rounded metaphase HeLa cells with a blunt atomic force microscopy (AFM) tip, limiting indentation to depths of less than 500 nm; the cortex dominates mechanics in this range 35 (Fig. 3e). The apparent stiffness measured is sensitive to contributions from cortical elasticity and tension.
stiffness to reduction in cortical tension 3 . Inhibition of the Arp2/3 complex with CK666 or depletion with short hairpin RNA (shRNA) led to a more than twofold increase in apparent stiffness that could be reversed by blebbistatin treatment (Fig. 3f and Extended Data Fig. 2e), consistent with observations that inhibition of the Arp2/3 complex increases cell contractility 36 .
Depletion of NAP1 led to an approximately twofold increase in apparent stiffness, consistent with Arp2/3 depletion (Fig. 3f), but contrary to reports of a decrease in tension with SRA1 depletion 37 . Depletion of IQGAP1 did not change apparent stiffness, similar to mDia1, but surprisingly, SPIN90 depletion led to an approximately twofold increase in stiffness (Fig. 3g). Changes in apparent stiffness did not correlate with changes in cortical phosphorylated myosin light chain (pMLC) abundance (Extended Data Fig. 3 and Supplementary Table 3) or cortical density (Fig. 2e), suggesting that these changes stem from complex changes in network organization at the micro scale.
SPIN90 is essential for cell division during embryonic morphogenesis. Having found an unexpected role for SPIN90 in cancer  Supplementary Fig. 3. c, Top: schematic representation of the measurement; cortex thickness and density are extracted from the fluorescence profiles of mCherry-CAAX (plasma membrane, red) and GFP-actin (cortex, green) in prometaphase HeLa cells 33 . Average spatial fluorescence profiles of the plasma membrane and the actin cortex are generated (see box on right). Cortex thickness is calculated from the distance between fluorescence peaks. Bottom: representative confocal image of a prometaphase HeLa cell expressing GFP-actin (green) and mCherry-CAAX (magenta). Scale bar, 7 μm. d, Cortex thickness for prometaphase HeLa cells transfected with non-silencing siRNA or siRNA targeting IQGAP1, NAP1 or SPIN90. e, Cortex density for prometaphase HeLa cells transfected with non-silencing siRNA or siRNA targeting IQGAP1, NAP1 or SPIN90. In d,e, *P < 0.05 and **P < 0.01 compared with the appropriate control using two-sided Welch's t-test. cells, we examined its role in embryonic tissues, where cells must frequently divide and change shape for tissue morphogenesis. Previous work has demonstrated roles for the WRC and IQGAP1 during embryonic morphogenesis 38,39 as well as the presence of mDia1 in the gastrula epithelium 40 . We therefore investigated a role for the SPIN90 orthologue in early Xenopus laevis embryos using morpholino injections (Supplementary Figs. 4 and 5). SPIN90-depleted embryos initially developed normally, but by stage 9, they displayed mutinucleated epidermal cells many times larger than epidermal cells in control embryos (Fig. 4a,b)-consistent with the role for SPIN90 in cell cycle progression observed in cancer cells (Fig. 1h,i). This phenotype eventually led to epidermis rupture and embryonic death at gastrulation. The late onset of this phenotype is perhaps due to reliance on maternal protein before mid-blastula transition or to a change in the nucleation pathway of cortical actin after ectoderm specification around stage 7 (ref. 41 ). Cell enlargement and multinucleation could both be rescued by co-injection of full-length mouse SPIN90 (Fig. 4c,d). Junctional F-actin in SPIN90-depleted embryos at stage 9 appeared separated from the cell membrane (Fig. 4e, arrows), a phenotype that was also rescued by co-injection of full-length mouse SPIN90. Thus, SPIN90 is necessary for embryonic morphogenesis, probably via a role in cell division.  SPIN90 mediates crosstalk between nucleators to control network organization. Thus far, our results indicate that IQGAP1 acts as an NPF for mDia1 and that the WRC acts as an NPF for the Arp2/3 complex (Supplementary Table 3). Many of the phenotypes linked to SPIN90 depletion suggest that it acts cooperatively with mDia1 (such as bleb size, cell cycle progression, thickness and actin accumulation rate) and stiffness measurements suggest that it acts as an NPF for Arp2/3, while it presents a distinct phenotype in other assays (mesh size and cortical density) (Supplementary Table 3). As SPIN90 can interact with both nucleators, we investigated a possible role in mediating their crosstalk using purified proteins. After confirming that SPIN90 interacts with each nucleator separately without enhancing mDia1 activity (Extended Data Fig. 4 and Supplementary Fig. 6), we examined how SPIN90 altered the organization of F-actin networks in a minimal system of purified proteins containing actin, profilin, Arp2/3, VCA (a domain of WAVE2 that activates Arp2/3), and mDia1. In the absence of SPIN90, nucleation of mother filaments was slow, generating a network consisting of few very densely branched regions (Fig. 5a). When SPIN90 was present, filaments were rapidly nucleated, giving rise to a more homogenous network with longer filaments. SPIN90 greatly increased the nucleation of mother filaments and decreased the density of branches on mother filaments (Fig. 5b-e and Supplementary Fig. 7), consistent with recent results 25 . Overall, adding SPIN90 leads to the formation of more barbed ends, which appear more evenly distributed over the surface (Fig. 5a-c). Of note, the fraction of mDia1-bearing barbed ends (identified by their rapid elongation in the presence of profilin) also grew with increasing concentrations of SPIN90 (Fig. 5f), suggesting that SPIN90 somehow favours the recruitment of mDia1 to barbed ends.
Thus, SPIN90 modulates network organization by shifting the balance of branched to unbranched filaments generated by the Arp2/3 complex and by facilitating the addition of mDia1 to barbed ends to generate long filaments. Importantly, these effects were noticeable at NPF ratios close to those present in cells (Supplementary Table 2).
Filaments nucleated by SPIN90-Arp2/3 efficiently recruit mDia1. Next, we investigated whether mDia1 might have different propensities to bind barbed ends generated with and without SPIN90 (Extended Data Fig. 5, Supplementary Results and Supplementary Fig. 8). We found that mDia1 was equally recruited by the barbed ends of mother filaments and branches in dendritic networks nucleated by Arp2/3 and WAVE (Extended Data Fig. 5a). When mDia1 was added to a mix of preformed filaments nucleated either spontaneously or by SPIN90-Arp2/3, we found that SPIN90-Arp2/3-nucleated filaments recruited mDia1 more efficiently to their barbed ends (Extended Data Fig. 5b). Further experiments using microfluidics revealed that mDia1 was recruited twofold faster to barbed ends nucleated by SPIN90-Arp2/3 than to those nucleated by spectrin-actin seeds ( Fig. 6a and Extended Data Fig. 9a,b). Finally, we found that recruitment of mDia1 to SPIN90-Arp2/3-nucleated filaments was further enhanced when mDia1 was present during nucleation (Extended Data Fig. 5c and Supplementary Fig. 8a). Together, these data suggested that SPIN90-Arp2/3 may recruit mDia1 to nucleate a rapidly growing filament.
SPIN90 forms a ternary complex with Arp2/3 and mDia1 to nucleate fast-growing filaments. To probe the existence of a ternary complex, we performed pull-down experiments, which showed that mDia1 binds to the SPIN90-Arp2/3 complex with a higher affinity than to SPIN90 alone, and that the SH3 domain of SPIN90 is not necessary for mDia1 to bind the SPIN90-Arp2/3 complex (Extended Data Fig. 6a, Supplementary Results and Supplementary  Fig. 8c). Next, we performed negative-staining electron microscopy experiments using mixtures of Arp2/3 with and without SPIN90 and mDia1 ( Fig. 6b and Extended Data Figs. 7 and 8). When mDia1 was added to the Arp2/3 complex, three-dimensional (3D) reconstructions of Arp2/3 could be docked with published structures 42 , with no additional electron density. When SPIN90 was added to the Arp2/3 complex, an additional electron density was present in the region where SPIN90 has been shown to interact with Arp2/3 43 . Finally, when both SPIN90 and mDia1 were included, a further electron density that could be docked with a dimer of FH2 was present close to the Arp2 and Arp3 subunits (Fig. 6b). These results confirm the formation of a ternary SPIN90-Arp2/3-mDia1 complex, and suggest that mDia1 may interact with the Arp2/3 complex rather than SPIN90 in the ternary complex.
To characterize filament nucleation by the ternary SPIN90-Arp2/3-mDia1 complex, we used a microfluidics assay in which we exposed a SPIN90-decorated surface sequentially to Arp2/3 followed by mDia1, before flowing in profilin-actin to enable filament nucleation and growth (Fig. 6c-e). We observed three populations of filaments: (1) slow-growing filaments, (2) filaments that elongated slowly and suddenly switched to rapid elongation, and (3) filaments elongating rapidly from the start. Since rapid elongation indicates the presence of mDia1 at the barbed end 44 , we interpreted population (3) as being nucleated by the ternary complex.
The nucleation rate of population (1) was identical to that of the filaments in a region of the same microchamber that we did not expose to mDia1 (Fig. 6e and Extended Data Fig. 9c), and to that measured in an independent experiment with no mDia1 (Extended Data Fig. 4a) indicating that population (1) corresponds to filaments nucleated by SPIN90-Arp2/3 that did not bind mDia1. The nucleation rate of population (2) was similar to that of population (1). We thus hypothesized that these slow-then-fast growing filaments were also nucleated by SPIN90-Arp2/3 and later captured an mDia1 adsorbed on the surface. Control experiments (Extended Data Fig. 6b) indicated that such events could indeed account for a substantial fraction of population (2). The filaments in population (3) were nucleated fivefold faster than the filaments nucleated by SPIN90-Arp2/3 (Fig. 6e), consistent with the notion that they are nucleated by the ternary SPIN90-Arp2/3-mDia1 complex. The control experiment in Extended Data Fig. 6b also confirmed that population (3) does not result from the capture of mDia1 after nucleation. Together, our results indicate that the ternary SPIN90-Arp2/3-mDia1 complex is a much more potent nucleator than SPIN90-Arp2/3, and that the resulting filaments have SPIN90-Arp2/3 bound to their pointed end and mDia1 bound to their barbed end.
To directly observe nucleation of filaments by this ternary complex, we performed single-molecule experiments using fluorescently labelled mDia1 and SPIN90. When mixing these proteins with Arp2/3, profilin and actin (15% of which was labelled with Alexa488), we observed rapidly growing filaments bearing mDia1 at their barbed end. Even though labelled SPIN90 was less active than unlabelled SPIN90, it was present at the pointed end of these rapidly growing filaments (Fig. 7a). This observation illustrates that, as in the absence of mDia1 (Extended Data Fig. 4a), SPIN90 and Arp2/3 remain at the pointed end of filaments after nucleation (Fig. 6c-e). It was challenging to observe filament nucleation, but colocalization of SPIN90 with mDia1 could sometimes be observed prior to the appearance of fluorescently labelled actin, which then separated the peaks of SPIN90 and mDia1 fluorescence, as expected, since each protein occupies a different end of the growing filament (Fig. 7b).
Overall, our in vitro observations show that SPIN90 activates Arp2/3 to nucleate linear filaments at the expense of Arp2/3 branching, and that these filaments have an increased probability of bearing mDia1 at their barbed ends due to (at least) two key features: the efficient recruitment of mDia1 to SPIN90-Arp2/3 nucleated filaments and the formation of a ternary SPIN90-Arp2/3-mDia1 complex that greatly enhances filament nucleation. Both mechanisms result in rapidly elongating filaments with mDia1 at their barbed ends and SPIN90-Arp2/3 at their pointed ends (Fig. 7c).

Discussion
In this study, we identify an NPF, SPIN90, that synergizes the action of Arp2/3 and mDia1 by forming a ternary complex and has wide-ranging effects in governing actin-network architecture.
The organization of F-actin networks is central to specifying their physiological function and mechanics. Understanding what mediates the passage from one network organization to another is a key unresolved question in cell biology. Competition between nucleators can help specify structures. Indeed, the passage from the 2D-branched network topologies found in lamellipodia to the 1D topologies present in filopodia appears to be mediated by competition between actin nucleators for G-actin monomers, regulated by profilin 45,46 . We report here that SPIN90, which thus far had only been considered as a mechanism to nucleate the mother filaments for the formation of branched networks and as a means to modulate branching density (refs. 23,25 and Fig. 5), is actually a potent enhancer of formin mDia1. SPIN90 competes with the WRC to tune the degree of Arp2/3 branching and co-opts formins to generate long (indicated on the right side of each row). In graphs, nucleation (green bars) is quantified by counting the number of mother filaments over 56,063 μm 2 after 400 s (b) or 300 s (c). The branch density (grey points) is quantified as the number of branches per μm of mother filament (counted on n = 20 mother filaments for each condition gathered in one experiment, except n = 10 for 0 nM SPIN90 in c). Branches were counted after mother filaments had elongated for 900 s, typically becoming 5.5 µm long (b) or after mother filaments had elongated with mDia1 for 400 s, typically becoming 18 µm long (c). Data are mean ± 95% confidence interval. d, TIRFM images of preformed actin filaments (15% Alexa568-labelled, red) mixed with the same protein solution as in a, with or without 0.2 nM mDia1 and with or without 125 nM SPIN90. Yellow arrowhead indicates growing barbed end of mother filament, and white arrowheads indicate growing branches. e, Quantification of the branching rate in d (mean ± s.e.m.) is determined from the number of branches appearing on the preformed mother filaments over time ( Supplementary Fig. 7) as a function of SPIN90 concentration, in the absence (dark grey bars) or presence (light grey bars) of 0.2 nM mDia1. f, Fraction of rapidly growing barbed ends (that is, bearing a formin), observed after 400 s in the experiment shown in c. Data are mean ± 95% confidence interval. The number of filaments (N) is indicated below each bar. In d-f, measurements were obtained from two independent experiments. Experiments were repeated twice independently in b-f and three times independently in a, with similar results. Statistical source data can be found at Source data Fig. 5.  Fig. 6 | mDia1 binds preferentially to SPIN90-Arp2/3 filaments and forms a ternary complex with SPIN90-Arp2/3 to generate fast-elongating filaments. a, In a microfluidics experiment, filaments were nucleated by surface-anchored spectrin-actin seeds or SPIN90-Arp2/3 complexes, and identified as such (Extended Data Fig. 9a,b) before flowing in a solution containing 0.5 μM G-actin (15% Alexa488-labelled), 3.5 μM profilin and 0.4 nM mDia1. Acceleration of elongation indicated the binding of mDia1. The graph shows measured mDia1-elongating filament fractions versus time and exponential fits (black lines) for filaments nucleated by spectrin-actin seeds (pink, n = 39 filaments; on rate, k on = (2.8 ± 0.011) × 10 −3 s −1 ) and by SPIN90-Arp2/3 (grey, n = 40 filaments; k on = (5.5 ± 0.035) × 10 −3 s −1 ). Two-sided log-rank test, P = 0.01. Shaded regions represent 95% confidence intervals. b, Three-dimensional structures of protein complexes reconstructed from electron microscopy negative stains at 27 Å resolution, from Arp2/3 incubated with mDia1 (left), SPIN90 (middle) or both (right). Electron microscopy densities were fitted using UCSF Chimera with crystal structures of Arp2/3 (PDB: 4XF2) (left) or Arp2/3-SPIN90 complex (PDB: 6DEC) (middle), or an atomic model of an mDia1 FH2 dimer (PDB: 1Y64) bound to the Arp2/3-SPIN90 complex (PDB: 6DEC) (right). Further information is presented in Extended Data Figs. 7 and 8. c, Microfluidics experiment in which a SPIN90-decorated surface is sequentially exposed to Arp2/3 then mDia1, before flowing in profilin-actin. Filaments are observed with their pointed ends attached to the surface and their barbed ends growing either slowly (bare barbed ends) or rapidly (mDia1-bearing barbed ends). d, Proportion of filaments elongating slow ((1), light blue), slow-then-fast ((2), blue) and fast ((3), dark blue) for different mDia1 concentrations, illustrated by representative kymographs. The SPIN90-decorated surface was exposed to 40 nM Arp2/3 (for 120 s), followed by 10 or 50 nM mDia1 (30 s), buffer (120 s), and then filaments were observed in the presence of 0.5 µM actin and 0.5 µM profilin. Scale bar, 5 µm. e, Nucleation of the three filament populations observed in d with 50 nM mDia1, compared to the nucleation of filaments in a chamber region unexposed to mDia1 (grey) (Extended Data Fig. 9c). Exponential fits (lines) yield nucleation rates of 8.8 × 10 −4 s −1 for (1), 1.2 × 10 −3 s −1 for (2) (P = 0.1, two-sided log-rank test) and 4.6 × 10 −3 s −1 for (3) (P = 2 × 10 −16 , two-sided log-rank test). In d,e, n indicates the total number of filaments analysed in one experiment. Experiments were repeated twice independently in a, and three times independently in c-e, with similar results. Statistical source data can be found at Source data Fig. 6.
rapidly growing filaments to generate a rich variety of actin filament networks ( Supplementary Fig. 13). Notably, such transitions were observed for ratios of SPIN90 to WRC close to those present in cells (Supplementary Table 2). Overall, our results indicate that SPIN90 regulates branching, filament growth rate and the resulting filament length, which are key parameters of the architecture of the network (Supplementary Fig. 13).
Our data reveal that SPIN90 enables synergistic action between mDia1 and Arp2/3 through at least two mechanisms, both of which result in increased nucleation of fast-growing actin filaments with SPIN90-Arp2/3 at their pointed ends and mDia1 at their barbed ends (Fig. 7c).
First, mDia1 displays an enhanced apparent on rate for the barbed ends of filaments generated by SPIN90-Arp2/3. This surprising Experiments were repeated twice independently with similar results. AU, arbitrary units. c, Interplay at the molecular level. SPIN90 activates the Arp2/3 complex (1) to nucleate linear filaments, in competition with Arp2/3 branching (left) following activation by the WAVE complex (not sketched) and binding to the sides of existing filaments. SPIN90 remains bound to the activated Arp2/3 complex at the pointed end of the nucleated filament. Formin mDia1, maintained in an active conformation by IQGAP1 (not sketched), can bind to the filament barbed end and accelerate its elongation using profilinactin. Barbed ends nucleated by SPIN90-Arp2/3 are particularly prone to binding mDia1 (3). The formin mDia1 can also bind to the SPIN90-Arp2/3 complex, forming a ternary complex (2), which efficiently nucleates rapidly growing filaments. Both routes lead to the formation of filaments with mDia1 at their barbed ends and SPIN90-Arp2/3 at their pointed ends. Note that active WAVE complex and active mDia1 are probably anchored to the plasma membrane.
observation may result either from microstructural differences in the barbed ends induced at nucleation 13 or from recruitment of mDia1 to the pointed end by binding to SPIN90-Arp2/3 followed by diffusion of mDia1 along the filament towards the barbed end, as already reported for mDia1 in a different context 47 . Second, we showed that SPIN90-Arp2/3 can form a ternary complex with mDia1 that increases filament nucleation fivefold compared with SPIN90-Arp2/3 alone. Our electron microscopy data (Fig. 6b) and single-filament observations (Fig. 6c-e) suggest that, following activation of Arp2/3 by SPIN90, a dimer of the FH2 domain of mDia1 binds to Arp2 and Arp3-which mimic a barbed end-to enhance nucleation and catalyse addition of monomers to the new filament. Such a 'rocket-launching' complex has been described for collaborative action of adenomatous polyposis coli and mDia1 48 , but not for Arp2/3 and mDia1, arguably the two most abundant actin nucleators. After elongation starts, our in vitro data indicate that mDia1 remains associated with the barbed end of the filament, whereas SPIN90-Arp2/3 remains at the pointed end (Figs. 6c-e and  7a,b), as already reported in the absence of mDia1 24 . This formation mechanism takes advantage of the efficient nucleation activity of the Arp2/3 complex and rapid filament elongation by mDia1.
The total amounts of SPIN90 and other NPFs in cells can be estimated on the basis of proteomics data and measurements of cellular actin concentration 31,49 (Supplementary Table 2). While determining the concentrations of available, active proteins in the cortical region is more challenging (for example, the WRC and mDia1 need to be activated and are probably bound to the membrane), we can estimate the relevant SPIN90 concentration to be in the 50-200 nM range (Supplementary Table 2).
In addition, in cells, our data indicate that IQGAP1 also modulates mDia1 activity (Figs. 1-3). IQGAP1 may bind active mDia1 following release from the ternary complex (Fig. 7c), allowing it to maintain elongation activity through its C-terminal DBR domain (Fig. 1a) and perhaps to scaffold mDia1 to the cortex through its N-terminal calponin homology domains.
Overall, our results show that the activity of the two main cortical actin nucleators, mDia1 and Arp2/3, is modulated by the interplay of three NPFs to finely tune cortical actin organization, kinetics and mechanics.

online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41556-020-0531-y.

Methods
Cell culture and generation of cell lines. M2 melanoma cells were a gift from T. Stossel (Harvard Medical School) and were described in ref. 50 . Cell lines were cultured in MEM with Earle's salts and l-glutamine (Gibco) with penicillinstreptomycin, and 10% 80:20 mix of newborn calf serum:fetal bovine serum. M2 cells stably expressing the F-actin reporter LifeAct-Ruby have been previously described in ref. 9 .
Wild-type HeLa cells were a gift from the MPI-CBG Technology Development Studio. The stable HeLa GFP-actin line and the HeLa LifeAct-Ruby line were described in ref. 9 . These cells were cultured in DMEM (Gibco) with penicillinstreptomycin, l-glutamine, 10% fetal bovine serum and 750 μg ml −1 G418. Cells were arrested in prometaphase with 100 nM nocodazole (Merck Biosciences) for 16 h for the localization studies. For metaphase arrest in laser-ablation and AFM studies, cells were treated for at least 1 h with 10 μM MG132 (Sigma).
To obtain stable protein knockdown, M2 or HeLa cells (wild type or LifeAct-Ruby) were transduced with lentiviruses encoding shRNA constructs targeting the genes of interest (see Supplementary Information) or transfected with linearized cDNA (see Supplementary Information). The cells were then selected with 250 ng ml −1 puromycin for 2 weeks. Flow cytometry was performed to obtain homogenous levels of expression (see Supplementary Information).
None of the cell lines in this study were found in the database of commonly misidentified cell lines maintained by International Cell Line Authentication Committee and National Center for Biotechnology Information Biosample. All cell lines were cultured at 37 °C with 5% CO 2 . All lines were routinely screened for the presence of mycoplasma using the mycoALERT kit (Lonza).
All imaging was done in Leibovitz L-15 medium (Gibco) supplemented with 10% fetal calf serum or in phenol red-free DMEM (Gibco) supplemented with 10% fetal bovine serum, l-glutamine and penicillin-streptomycin.

Plasmid construction and transfection. Full-length human and mouse SPIN90
(also known as NCKIPSD) were obtained from the Mammalian Gene collection or the I.M.A.G.E. library and cloned into EGFP-C1, EGFP-N1 or EBFP2-C1 vectors using restrictions sites inserted by PCR.
SPIN90, SPIN90-NT, and SPIN90-CT were described in ref. 26 . IQGAP1-GFP 51 was a gift from D. Sacks (National Institutes of Health), the DBR domain of IQGAP1 tagged with GFP 20 was a gift from R. Grosse (University of Marburg, Germany), and Fli-I-GFP 52 was a gift from R. Tombes (Virginia Commonwealth University). WAVE2-GFP with a truncated CMV promoter was a gift from O. Weiner (University of California San Francisco). eBFP2 53 was obtained from Addgene (plasmid 14893) and substituted for eGFP in some constructs. All gene products were verified by sequencing.
cDNA was purified from bacteria using the Qiagen Spin Miniprep Kit. Transfections were carried out using Lipofectamine 2000 according to the manufacturer's instructions.
Confocal microscopy. All fluorescence imaging (except for cortex-ablation experiments) was performed using a ×100 oil-immersion objective on an inverted microscope (IX-81, Olympus) fitted with a spinning-disk head (Yokogawa, CSU22). Images were acquired with an Andor iXon camera and analysed using ImageJ (http://rsbweb.nih.gov/ij/). Excitation with a 488 nm laser was utilized for GFP-tagged proteins, with a 543 nm laser for mRFP-, Ruby-and mCherry-tagged proteins as well as Alexa568-labelled antibodies, with a 405 nm laser for eBFP2-tagged proteins, and with a 647 nm laser cells labelled with CellTracker Deep Red (Life Technologies).

SEM of the cortex of blebs.
Sample preparation for SEM was performed as described 54 with minor modifications as described 9 . Two hours before sample preparation, whole cells were plated onto 12 mm glass coverslips. Immediately before fixation, the coverslips were washed three times with L-15 without serum and transferred to cytoskeleton buffer (50 mM imidazole, 50 mM KCl, 0.5 mM MgCl 2 , 0.1 mM EDTA, 1 mM EGTA, pH 6.8) containing 0.5% Triton-X and 0.25% glutaraldehyde for 5 min. This was followed by a second extraction with 2% Triton-X and 1% CHAPS in cytoskeleton buffer for 5 min before washing the coverslips in cytoskeleton buffer three times. The remainder of the protocol was identical to the originally described procedure 54 . The cells were dehydrated with serial ethanol dilutions, dried in a critical point dryer, coated with 5-6 nm platinum-palladium and imaged using the in-lens detector of a JEOL7401 Field Emission Scanning Electron Microscope (JEOL). All samples were prepared in duplicate and images from two independent experiments were acquired. Similar phenotypes were observed in each independent experiment and images of at least eight different cells were acquired for each experimental condition.
Cortical thickness measurement. Measurement of the thickness of the actin cortex in metaphase HeLa cells was performed as described 3 . In brief, two-colour image stacks (30-70 z-slices, acquired at 100 nm intervals) were acquired around the equatorial plane of rounded cells using a ×60 colour-corrected objective (1.40 numerical aperture, OSC2 PlanApoN) mounted on an Olympus FV1200 microscope. After correcting for chromatic shift and magnification using custom software written in MATLAB, a single equatorial plane was selected for each image using Fiji image analysis software. Cortex thickness and density were extracted as described 33 . For membrane width measurements, we measured the full width at half-maximum of the plasma membrane intensity peak by interpolating the x position on the line scan on either side of the peak by linear interpolation of the two closest points. The half-maximum was defined as half of the difference between the peak intensity and intracellular background.
Actin regrowth speed analysis. The rate of actin accumulation during cortex regrowth was measured as described 34,55 . Blebs were induced by exposure of a small region of the cortex of metaphase HeLa cells expressing LifeAct-Ruby to UV pulses. Following induction, blebs grow rapidly before stopping and eventually retracting. Actin is initially absent from below the bleb membrane but progressively accumulates as growth slows and retraction starts. To allow for reliable segmentation of the cell contour for subsequent image analysis, we added 2.5 μM Alexa647 dye to the extracellular medium. Images were then processed, automatically analysed and visualized using KoreTechs. This allowed segmentation of the cortex from the cytoplasm and segmentation of the induced bleb from the rest of the cell body. Cortical and cytoplasmic fluorescence intensities of actin could then be monitored in the bleb and compared with the cell body control value. This analysis yielded curves relating the evolution of the actin fluorescence intensity in the bleb normalized to the cortical fluorescence intensity in the cell body. Actin accumulation in the cortex displayed two markedly different phases: the first started immediately after laser ablation and ended shortly after the cessation of growth; the second started after growth had finished and ended after bleb retraction (Extended Data Fig. 2b). Actin accumulation was approximately linear in both regimes and slopes relating the percentage actin accumulation per second could be measured by fitting straight lines to each interval.
AFM and data analysis. Indentations of cells by AFM were performed using a JPK NanoWizard-1 AFM (JPK) mounted on an inverted microscope (IX-81, Olympus). The day before the experiment, cells were plated onto 35 mm glass-bottom Petri dishes. Cells were incubated in MG132 (10 µM, Sigma) for 2 h before measurement to arrest cells in metaphase. Experiments were performed at room temperature and cells were maintained in Leibovitz L-15 medium (Life Technologies) supplemented with 10% FBS (Sigma-Aldrich) and MG132 (10 µM). Before each experiment, the spring constant of the cantilever was calibrated using the thermal noise method implemented in the AFM software (JPK SPM). The sensitivity of the cantilever was measured from the slope of force-distance curves acquired on glass. For apparent stiffness measurements, we used soft cantilevers with V-shaped tips (BioLever OBL-10, Bruker; nominal spring constant of 0.006 N m −1 ).
For each measurement, the cantilever was first aligned above a metaphase cell using the optical microscope. Then, force-distance curves were acquired over the centre of the cell at the 4 vertices of a square with a 2 µm side. At each of these four positions, up to 10 curves were acquired with an approach speed of 2.5 µm s −1 and a target force of 2.5 nN. Force-distance curves were then post-processed to compute an apparent stiffness. First, we determined the contact point between the cantilever tip and the cell using the method described 56 , implemented in MATLAB (MathWorks). The indentation depth was then calculated by subtracting the cantilever deflection d from the piezo displacement beyond the contact point z (δ = z − d). The resultant force-indentation curves were then averaged over each position and fitted with the Sneddon model to calculate each location's apparent elasticity 57 . Curve fitting was restricted to indentation depths shallower than 500 nm to maximize contributions of the cortex to the restoring force and minimize contributions from the cytoplasm 35 .
Electron microscopy and single-particle analysis. Concentrated mixtures of either Arp2/3-mDia1, Arp2/3-SPIN90, or Arp2/3-SPIN90-mDia1 were diluted and applied to freshly glow-discharged 300 mesh carbon-coated copper grids at final concentrations 22 nM Arp2/3, 100 nM SPIN90 and 50 nM mDia1. The samples were negatively stained with uranyl formate 1%. Data were collected with a FEI tecnai G2 transmission electron microscope equipped with a LaB 6 emission filament operating at 200 kV. Images were captured on a TVIPS F416 CMOS camera at ×50,000 magnification and 1.5-2.5 µm underfocus. The pixel size used was 2.13 Å per pixel. Using the contrast transfer function estimations from CTFFIND 58 , micrographs were phase flipped.
Particles were hand-picked using XMIPP3 software from the image processing framework SCIPION 59 . Particles whose size would correspond to that of a potential complex including Arp2/3 were selected. A total of 7,572 particles were selected for the mixture of Arp2/3-mDia1; 6,568 for the mixture of Arp2/3-SPIN90; 10,044 particles for the mixture of Arp2/3-SPIN90-mDia1. Two-dimensional class averages were obtained in Relion 3.0 60 . Particles belonging to blurred averages were excluded from further analysis.
3D classification was performed to sort out three classes using Relion 3.0. Low-pass-filtered crystal structures were used as references (low-pass filter: 40 Å). The crystal structure of Arp2/3-SPIN90 complex 43 (Protein Data Bank (PDB): 6DEC) was used for Arp2/3-SPIN90 data analysis. Different models generated from the Arp2/3-SPIN90 complex (PDB: 6DEC) and FH2 domains from Bni1pactin complex 61 (PDB: 1Y64) were used for Arp2/3-SPIN90-mDia1 data analysis. Arp2/3 crystal structure 42 (PDB: 4XF2), with and without FH2 domains from Bni1p-actin complex (PDB: 1Y64), were used for Arp2/3-mDia1 data analysis. The particles belonging to the 3D class displaying the best fit to the model complex used as reference were selected for further processing, while the divergent ones were left out.
These sorted particles were then used to reconstruct a 3D model with Relion 3.0 using strongly low-pass filtered references to prevent any bias (low-pass filter: 50-60 Å). The resulting 3D reconstructions shown in Fig. 6b and Extended Data Fig. 7 have a resolution of 27 Å for Arp2/3-mDia1 (5,456 particles used), Arp2/3-SPIN90 (4,690 particles used) and Arp2/3-SPIN90-mDia1 (2,006 particles used). To interpret the resulting 3D reconstructions, the non-filtered crystal structures used as references were fitted in the corresponding 3D maps using UCSF Chimera 62 . Spurious noise from electron microscopy densities was hidden with the 'hide dust' command in UCSF Chimera to facilitate readability.
Image acquisition for in vitro experiments. The microfluidic devices or open flow chambers were placed on a Nikon TiE inverted microscope, equipped with a ×60 oil-immersion objective. An objective heater (Okolab) maintained the temperature at 25 °C on the coverslip. The total internal reflection microscopy (TIRFM) setup was controlled with Metamorph, illuminated by 100 mW tunable lasers (iLAS2, Roper Scientific, now Gataca Systems). Images were acquired on an Evolve EMCCD camera (Photometrics). ImageJ software was used to analyse images.
For single-filament assays (Fig. 6a,c-e, Extended Data Figs. 4a,b, 6b, 9 and Supplementary Fig. 6f), microfluidics experiments were carried out with polydimethylsiloxane (PDMS) (Sylgard 184, Dow Corning) devices with dimensions of 60 μm × 800 μm x 1 cm, based on the original protocols 63 . The devices are cross-shaped, consisting of channels with three inlets and one outlet. Measuring filament nucleation and elongation by SPIN90-Arp2/3. To specifically anchor glutathione S-transferase (GST)-SPIN90 by its GST tag to the coverslip surfaces, clean surfaces were first incubated with a mixture of BSA and biotinylated BSA, then incubated with neutravidin (20 μg ml −1 ) for 5 min and rinsed. Surfaces were exposed to 2 µg ml −1 biotinylated GST antibody (Rockland) and rinsed, then exposed to 250 nM SPIN90 for 5 min and rinsed.
To measure the SPIN90-Arp2/3 nucleation rate (Extended Data Fig. 4a), a GST-SPIN90-decorated surface was exposed to 30 nM Arp2/3 complex for 5 min and rinsed, then exposed to 0.5 μM 15% Alexa488-actin and 0.5 μM profilin. The number of filaments generated is counted as a function of time.
Measuring filament nucleation, elongation and processivity by SPIN90-mDia1. For measurement of filament nucleation, elongation and processivity by SPIN90-mDia1 (Extended Data Fig. 4b), when the anti-GST anchored surface was exposed to 50 nM mDia1 for 5 min and rinsed before being exposed to 1 μM 15% Alexa488-actin for nucleation, only one filament was observed per field of view in 30 min of observation. This shows that nonspecific binding of functional mDia1 to the surface is very low. When the same surface was exposed to 250 nM SPIN90 for 5 minutes at first, then exposed to 50 nM mDia1 for 5 minutes and rinsed, and finally to 1 μM 15% Alexa488-actin, about 126 filaments were observed per field of view during 30 minutes observation. Afterwards, those filaments were incubated with 0.5 μM unlabelled actin and 3.5 μM profilin, 17 of these barbed end anchored filaments were chosen randomly and their elongation rate was measured. The elongation rate of filaments by mDia1 in the absence of SPIN90 was measured on the same day and with the same actin and profilin concentrations. In these experiments, mDia1 was anchored to the surface as described previously 44 .
The processivity of mDia1 ( Supplementary Fig. 6f) was measured on filaments anchored by their pointed ends to the coverslip surface, using spectrin-actin seeds. The presence of mDia1 at the barbed ends of the filaments was assessed by their rate of elongation from profilin-actin. Thus, the departure of mDia1 from the barbed end was detected by an abrupt reduction of the elongation rate 44 . The imaging solution contained 0.5 μM 15% Alexa488-actin, 3.5 μM profilin and 0 or 100 nM SPIN90.
To measure mDia1 binding to actin filaments of different lengths (Extended Data Fig. 9b), spectrin seeds were attached to the coverslip surface. After the surface was passivated with 3% BSA, 0.7 μM 15% labelled Alexa488-actin was flowed into the chamber to generate actin filaments. Then, the surface was briefly exposed to 100% laser power, which may sever filaments. This yielded a surface with filaments with a broad length distribution anchored to it. These filaments were then exposed to 0.4 nM mDia1, 0.5 μM 15% labelled Alexa488-actin and 3.5 μM profilin simultaneously to compare affinity of mDia for long and short filaments.
Measuring filament nucleation and elongation by SPIN90-Arp2/3-mDia1. To grow filaments from the ternary complex of SPIN90-Arp2/3-mDia1 (Fig. 6c-e), a GST-SPIN90 (full length or C-terminal construct)-decorated surface in a microfluidic chamber was first exposed to 40 nM Arp2/3 complex for 2 min and rinsed, followed by exposure to 10 nM (or 50 nM) mDia1 for 30 s. The surface was then rinsed with buffer for 2 min before being exposed to profilin-actin (0.5 μM 15% labelled Alexa488-actin and 0.5 μM profilin). The nucleation and the elongation rates of the observed filaments were monitored.
To compare the nucleation rates of SPIN90-Arp2/3 complex in regions with and without mDia1, a GST-SPIN90-decorated surface in a microfluidic chamber was exposed to 40 nM Arp2/3 complex for 2 min and rinsed; we then exposed only half of the chamber to 50 nM mDia1 by manipulating the input pressures in our microfluidic device. The surface was then rinsed with buffer for 2 min before being exposed to 0.5 μM 15% labelled Alexa488-actin and 0.5 μM profilin (Extended Data Fig. 9c). The nucleation and elongation rates of the observed filaments were monitored (Fig. 6e). The fact that the nucleation rate of slow filaments is the same as in independent control experiments indicates that mDia1 dissociation from the ternary complex is negligible over the course of the experiment (otherwise we would have a substantial portion of delayed nucleation of slow filaments). We are thus confident that the nucleation rate we measure for population (3) is an accurate estimate of the actual nucleation rate of the ternary complex. Fig. 6b, to estimate the contribution of potential traces of mDia1 on the surface (either due to nonspecific binding or to interaction with anchored SPIN90) to the rapid elongation of filaments, the same surface conditions were generated by exposing the GST-SPIN90-decorated surface to Arp2/3 and mDia1. Filaments were then nucleated by exposing the surface to profilin-actin for 20 s, and the chamber was then rinsed for 2 min. During these 2 min, the laser intensity was increased in order to sever the filaments, and these were rinsed out by the flow. Then, the chamber was again exposed to the profilin-actin solution to generate new barbed ends from the same, bare SPIN90-Arp2/3 complexes. Their elongation rate was monitored, and the rapidly growing population was greatly reduced compared to the experiment (Extended Data Fig.  6b), indicating that these filaments did not result from the capture of mDia1 by the growing filament barbed ends. By contrast, the slow-then-fast growing population was not substantially reduced, indicating that a proportion may have resulted from the capture of mDia1 by the growing filament barbed ends.

Nucleation and elongation by SPIN90-Arp2/3-mDia1 control experiments. For control experiments in Extended Data
To determine if mDia1 (either adsorbed on the surface, or bound to SPIN90 or SPIN90-Arp2/3) was damaged by the strong exposure to light during the severing of the filaments in the control, the experiment (in which filaments were not severed) was also performed with exposure to the same amount of light. To do so, we exposed the GST-SPIN90-decorated surface sequentially to 40 nM Arp2/3 for 2 min and to 50 nM mDia1 for 30 s. The chamber was then rinsed with buffer for 2 min; during this time, the laser intensity was set to 100% (same conditions as in the control experiment, except that no filaments have elongated yet). Then, the chamber was exposed to 0.5 μM 15% labelled Alexa488-actin and 0.5 μM profilin. The elongation rates of the filaments generated in the first 5 min were monitored (Extended Data Fig. 6b, 'complex'). The proportion of fast-growing filaments was similar to that in the experiment without the strong exposure to light, indicating that the light did not damage the formins.
GST pull-down assay to detect protein-protein interactions. For pull-down assays to detect protein-protein interactions (Extended Data Figs. 4c and 6a and Supplementary Fig. 8c), first, glutathione beads (Glutathione Sepharose 4B, GE Healthcare) were incubated with 5% BSA for 5 min and washed 3 times with 500 μl GST pull-down buffer containing 50 mM Tris-HCl pH 7.5, 1 mM DTT, 50 mM KCl and 5% glycerol.
To investigate direct binding between SPIN90 and mDia1 (Extended Data Fig. 4c), we carried out a pull-down assay with mDia1 anchored to glutathione beads. Fifty microlitres of 7.7 μM beads decorated with GST or GST-mDia1 was mixed with 2.5-20 μM His-tagged SPIN90 (full-length) for 1 h at 4 °C. The beads were washed by adding 500 μl GST pull-down buffer. After centrifugation at 500g for 1 min, the supernatant was discarded. After the washing steps were repeated three times, proteins attached to beads were eluted with 50 μl 20 mM reduced glutathione. The sample was separated by SDS-PAGE for analysis by western blot (Extended Data Fig. 4c and Supplementary Fig. 12). Pulled-down SPIN90 was detected by anti-His antibody (QIAGEN).
To investigate formation of a ternary complex between SPIN90, the Arp2/3 complex and mDia1, we carried out a pull-down assay with SPIN90 anchored to glutathione beads. Fifty microlitres of 12 μM Glutathione Sepharose 4B decorated with full-length GST-SPIN90 was mixed with 0-540 nM Arp2/3 for 1 h at 4 °C. After washing 3 times with 300 μl GST pull-down buffer, these beads were further incubated with 700 nM His-tagged mDia1 dimer for 1 h at 4 °C. After washing the beads 3 times with 300 μl GST pull-down buffer, proteins attached to beads were eluted with 50 μl 20 mM GSH. The sample was separated by SDS-PAGE for western blot analysis. His (QIAGEN) and ArpC2 (Sigma) antibodies were used to detect His-tagged mDia1 and Arp2/3, respectively. Membranes were imaged with ImageQuant LAS-4000 Mini Imaging System. Pulled-down mDia1 was quantified using ImageJ (Extended Data Fig. 6a and Supplementary Fig. 12).
The control experiment ( Supplementary Figs. 8c and 12) was done in the same way, except in the first step, 50 μl of 12 μM Glutathione Sepharose 4B decorated with GST or GST-SPIN90 C-terminal construct was mixed with or without 540 nM Arp2/3 for 1 h at 4 °C.
Statistics and reproducibility. Phenotype distribution after gene depletion was compared with cells stably expressing non-silencing shRNA using a χ 2 -test (Supplementary Table 3). Cells were imaged from at least two independent experiments; P < 0.01 was deemed statistically significant. The amount of cell death was compared using Welch's two-sided t-test with unequal s.d.; P < 0.05 was deemed statistically significant. Changes in the proportion of gaps of a given size across conditions was compared with a two-sided Wilcoxon's rank test; P < 0.01 was deemed statistically significant. Changes in cortical thickness, cortical density and cell area were examined using two-sided Welch's t-test; P < 0.05 was deemed statistically significant. Changes in actin accumulation rates and cortical stiffness across conditions were examined using two-sided Student's t-tests; P < 0.01 was deemed statistically significant. Changes in mean cortical pMLC were compared across conditions using one-way ANOVA on ranks. P < 0.01 was deemed significant. Changes in mDia1 binding to filaments that nucleated differently or with different lengths, measured with the microfluidic device, were compared with two-sided log-rank test; P < 0.05 was deemed significant. Changes in mDia1 binding to mother and daughter filaments, or to filaments nucleated differently, measured with an open chamber, were compared with Pearson's χ 2 -test; if one of the populations was too small (<5 individual samples), then the changes were compared with a one-sided Fisher's exact test; P < 0.05 was deemed significant. Changes in actin nucleation rates of different filament populations were compared with a two-sided log-rank test. P < 0.05 was deemed significant. All experiments were repeated at least twice independently with similar results. The exact number of repeats for each experiment is indicated in the relevant figure legend.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Mass spectrometry data have been deposited in ProteomeXchange with the primary accession code PXD018318. All data supporting the conclusions of this paper are available from the authors upon reasonable request.

Code availability
The code used to analyse data in this study is available from the authors upon request. Fig. 2 | Control experiments for actin accumulation rate and cortical stiffness. a, Representative laser ablation experiment. All images are a single confocal section. Top row: Alexa647 is added to the medium for robust segmentation of the cell. Bottom row: LifeAct Ruby. Left column: cell before ablation, middle column: after ablation, right column: segmented image. Timings are indicated on the top row images. Scale bar=10 µm. b, Representative actin regrowth curve in a bleb induced by laser ablation as in A. Pink curve: evolution of mean cortical actin fluorescence in the bleb normalized to the mean cortical intensity in the cell body (pink). Blue curve: evolution of bleb area normalized to cell body area. Initial regrowth rates are linear with time (initial accumulation rate). t = 0 s, ablation onset. (c-e) Data plotted as box-whisker plots. The distributions' medians, first and third quartiles, and ranges are represented by the central red bars, bounding boxes and whiskers, respectively. Statistics are derived from the total number (n, indicated above each box) of cells examined in three independent experiments. Each dot represents one cell measurement. Statistical outliers are indicated by red dots. **p < 0.01 compared to the appropriate control. c, Actin accumulation rate in metaphase cells at 37 C and Room Temperature. Two-sided Student t-test: 37 C vs RT: p = 0.19. d, Apparent elastic modulus for cells expressing Non-Silencing shRNA (NS sh), transfected with Non-Silencing siRNA (NS si), and treated with DMSO. One-way ANOVA on ranks: NS siRNA vs NS shRNA: p = 0.001, DMSO vs NS shRNA: p = 0.001. DMSO vs NS siRNA: p = 0.99. e, Apparent elastic modulus of WT and ACTR2 shRNA cells treated with DMSO and blebbistatin (Bb). One-way ANOVA on ranks: WT DMSO vs WT blebbistatin: p = 2 10 -9 ; WT DMSO vs ACTR2 shRNA DMSO: p = 4 10 -5 ; ACTR2 shRNA DMSO vs ACTR2 shRNA blebbistatin: p = 5 10 -13 ; WT blebbistatin vs ACTR2 shRNA blebbistatin: p = 0.71. Statistical source data can be found at Source data figure ED2. Fig. 3 | Regulation of myosin localization and phosphorylation in cells with nucleator or NPF depletion. a, Representative pMLC distribution in metaphase HeLa cells visualized by immuno-staining for different protein depletions and for non-silencing (NS) siRNA. Each image is a single section of a confocal microscopy stack and is shown in inverted contrast. b, Mean cortical pMLC fluorescence intensity for different treatments normalized to the mean cortical pMLC fluorescence intensity for non-silencing (NS) siRNA or shRNA. The distributions' medians, first and third quartiles and ranges are represented by the central red bars, bounding boxes and whiskers, respectively. Statistics are derived from the total number (n, indicated above each box) of cells examined in three independent experiments. Each dot represents one cell measurement. Statistical outliers are indicated by red dots. Statistical comparisons of the means were performed using one-way ANOVA on ranks compared to NS siRNA/shRNA. ACTR2 siRNA: p = 7 10 -5 , NAP1: p = 0.03, mDia1: p = 0.85, IQGAP1: p = 0.08, SPIN90: p = 0.74. **p < 0.01 compared to the appropriate control. See Supplementary Figure 9 for controls. c, Change in cortical myosin regulatory light chain fluorescence intensity upon treatment with DMSO (left panel) or CK666 (right panel). In each panel, the top row shows the fluorescence intensity before treatment and the bottom row after treatment for the same cell. The left most column shows myosin regulatory light chain fluorescence (MRLC-GFP), the middle column shows LifeAct-Ruby, and the right column shows the overlay with MRLC in green and LifeAct in Magenta. Experiments were repeated twice independently with similar results. (a,c) Scale bars=10 µm. Statistical source data can be found at Source data figure ED3. Fig. 6 | mDia1 binds to the SPIN90-Arp2/3 complex. a, mDia1 is dose-dependently eluted with Arp2/3 bound to GST-SPIN90 decorated beads. Left: Anti-His and Anti-ArpC2 western blots of GST pull down assay. GST beads were incubated with 12 μM GST-SPIN90 and with the indicated amounts of Arp2/3. The beads were washed and incubated with 750 nM mDia1. Right: quantification of pulled-down mDia1, normalized to GST-SPIN90 without Arp2/3 (mean ± SD, n = 3 independent experiments). Black dots show the data points. For each concentration, the red dot shows the mean and the red bars the standard deviation. Uncropped Western blot can be found in Source data figure ED6. b, Left: schematic diagram of a control experiment (related to Fig. 6C-D). Control experiments consist in observing the elongation of new, bare barbed ends growing in a microfluidics chamber with a SPIN90-decorated surface exposed to Arp2/3 and mDia1. This is achieved by first following the sequence shown in Fig. 6C, and then photo-severing the filaments with a strong laser illumination (see Methods). We then observed the regrowth of filaments from SPIN90-Arp2/3. The photo-severing ensured that the newly elongating barbed ends were initially without mDia1. The distribution of slow (light blue), slow then fast (blue), and fast (dark blue) filaments was compared to experiments like the ones presented in Fig. 6C-D (bar charts, on the right, n=number of filaments). "Complex" denotes the experiments performed as in Fig. 6C-D (with an additional exposure to light, to have the same conditions as in the control, see Methods) and "control" denotes the control experiments. Comparison using one-tailed Fisher's exact test: p = 0.31 for the slow-then-fast population, and p = 0.0013 for the fast population. n is the number of filaments randomly picked and investigated in an independent experiment. The experiment was repeated twice independently with similar results. Statistical source data can be found at Source data figure ED6. Fig. 7 | Visualisation of different orientations of the protein complexes formed by Arp2/3, SPIN90, and mDia1. Different combinations of complexes (binary or ternary) involving the Arp2/3 complex, SPIN90, and mDia1 were analyzed by electron microscopy and single particle analysis. For each protein complex, three-dimensional reconstructions were obtained from 3D classifications and compared to existing or generated crystal structures (see methods). a, Incubation of mDia1 and Arp2/3 resulted in a 3D structure which only accommodates Arp2/3, suggesting that mDia1 and Arp2/3 do not interact when SPIN90 is absent. b, Conversely, mixing the Arp2/3 complex and SPIN90 results in a complex for 76 % of the particles. Within the complex, SPIN90 (green) clearly appears as an additional density when compared with the truncated docked crystal structure. c, When compared with the Arp2/3-SPIN90 complex in B, the 3D envelope (resolution of 27 Å) resulting from the ternary complex (SPIN90-Arp2/3-mDia1) exhibits an additional density accommodating a dimer of FH2 domains for 22 % of the particles. (a-c) Each protein complex is subjected to a variety of rotations to visualize its full structure.