Actomyosin dynamics drive local membrane component organization in an in vitro active composite layer

Significance

This manuscript addresses the role of active processes in the spatial organization and dynamics of cell surface components. Using a reconstituted minimal system, we provide experimental evidence for a proposed clustering mechanism that relies on the intrinsic, active mechanics of actin filaments and myosin motors expected to be present at the cell cortex. The coupling between the actomyosin and the lipid bilayer gives rise to an emergent active composite with properties that resemble those observed in live cells. This clustering mechanism is a key feature of the active composite cell surface model and furthers our understanding of the multiple ways in which the cell surface might regulate its composition.

Abstract

The surface of a living cell provides a platform for receptor signaling, protein sorting, transport, and endocytosis, whose regulation requires the local control of membrane organization. Previous work has revealed a role for dynamic actomyosin in membrane protein and lipid organization, suggesting that the cell surface behaves as an active composite composed of a fluid bilayer and a thin film of active actomyosin. We reconstitute an analogous system in vitro that consists of a fluid lipid bilayer coupled via membrane-associated actin-binding proteins to dynamic actin filaments and myosin motors. Upon complete consumption of ATP, this system settles into distinct phases of actin organization, namely bundled filaments, linked apolar asters, and a lattice of polar asters. These depend on actin concentration, filament length, and actin/myosin ratio. During formation of the polar aster phase, advection of the self-organizing actomyosin network drives transient clustering of actin-associated membrane components. Regeneration of ATP supports a constitutively remodeling actomyosin state, which in turn drives active fluctuations of coupled membrane components, resembling those observed at the cell surface. In a multicomponent membrane bilayer, this remodeling actomyosin layer contributes to changes in the extent and dynamics of phase-segregating domains. These results show how local membrane composition can be driven by active processes arising from actomyosin, highlighting the fundamental basis of the active composite model of the cell surface, and indicate its relevance to the study of membrane organization.

The cell surface mediates interactions between the cell and the outside world by serving as the site for signal transduction. It also facilitates the uptake and release of cargo and supports adhesion to substrates. These diverse roles require that the cell surface components involved in each function are spatially and temporally organized into domains spanning a few nanometers (nanoclusters) to several micrometers (microdomains). The cell surface itself may be considered as a fluid–lipid bilayer wherein proteins are embedded (1). In the living cell, this multicomponent system is supported by an actin cortex, composed of a branched network of actin and a collection of filaments (2–4).

Current models of membrane organization fall into three categories: those invoking lipid–lipid and lipid–protein interactions in the plasma membrane [e.g., the fluid mosaic model (1, 5) and the lipid raft hypothesis (6)], or those that appeal to the membrane-associated actin cortex (e.g., the picket fence model) (7), or a combination of these (8, 9). Although these models based on thermodynamic equilibrium principles have successfully explained the organization and dynamics of a range of membrane components and molecules, there is a growing class of phenomena that appears inconsistent with chemical and thermal equilibrium, which might warrant a different explanation. These include aspects of the organization and dynamics of outer leaflet glycosyl-phosphatidylinositol-anchored proteins (GPI-anchored proteins) (10–13), inner leaflet Ras proteins (14), and actin-binding transmembrane proteins (13, 15, 16).

Recent experimental and theoretical work has shown that these features can be explained by taking into account that many cortical and membrane proteins are driven by ATP-consuming processes that drive the system out of equilibrium (13, 15, 17). The membrane models mentioned above have by-and-large neglected this active nature of the actin cortex where actin filaments are being continuously polymerized and depolymerized (18–21), in addition to being persistently acted upon by a variety of myosin motors (22–24) that consume ATP and exert contractile stresses on cortical actin filaments, continually remodeling the architecture of the cortex (4, 21, 25). These active processes in turn can generate tangential stresses and currents on the cell surface, which could drive the dynamics and local composition of membrane components at different scales (22, 26–29).

Actin polymerization is proposed to be driven at the membrane by two nucleators, the Arp2/3 complex, which creates a densely branched network, as well as formins that nucleate filaments (18, 21, 30). A number of myosin motors are also associated with the juxtamembranous actin cortex, of which nonmuscle myosin II is the major component in remodeling the cortex and creating actin flows (4, 23, 25, 26, 31, 32). Based on our observations that the clustering of cell surface components that couple directly or indirectly to cortical actin [e.g., GPI-anchored proteins, proteins of the Ezrin, Radaxin, or Moesin (ERM) family (13, 15)] depends on myosin activity, we proposed that this clustering arises from the coupling to contractile actomyosin platforms (called “actin asters”) produced at the cortex (15, 33).

A coarse-grained theory describing this idea has been put forward and corroborated by the verification of its key predictions in live cells (15, 33), but a systematic identification of the underlying microscopic processes is lacking. Given the complexity of numerous processes acting at the membrane of a living cell, we use an in vitro approach to study the effect of an energy-consuming actomyosin network on the dynamics of membrane molecules that directly interact with filamentous actin.

A series of in vitro studies have explored the organization of confined, dynamic filaments (both actin and microtubules) (34–39) or the role of actin architecture on membrane organization (40–46). Indeed, these studies have yielded insights into the nontrivial emergent configurations that mixtures of polar filaments and motors can adopt when fueled by ATP (34–37), in particular constitutively remodeling steady states that display characteristics of active mechanics (38, 39, 47). However, the effect of linking these mechanics to the confining lipid bilayer and its organization has not been studied.

The consequences of actin polymerization on membrane organization, in particular on giant unilamellar vesicles (GUVs), have been addressed in a number of studies on the propulsion of GUVs by an actin comet tail (40, 45, 46). In those experiments, the apparent advection of membrane bound ActA or WASP toward the site of actin polymerization is mainly due to the change in binding affinity of WASP to actin through Arp2/3 (44) and the spherical geometry resulting in the drag of actin to one pole of the vesicle after symmetry break of the actin shell. That this dynamic process changes the bulk properties of the bilayer, namely the critical temperature of a phase-separating lipid bilayer, was shown by Liu and Fletcher (40) when the actin nucleator N-WASP was connected to a lipid species (PIP₂) that was capable of partitioning into one of the two phases.

Besides these pioneering studies on the effects of active processes on membrane organization, little was done to directly test the effect of active lateral stresses as well as actomyosin remodeling at the membrane, particularly on the dynamics and organization of membrane-associated components.

To this end, we build an active composite in vitro by stepwise addition of components: a supported lipid bilayer with an actin-binding component, actin filaments, and myosin motors. By systematically varying the concentrations of actin and myosin as well as the average actin filament length, we find distinct states of actomyosin organization at the membrane surface upon complete ATP consumption. More importantly, we find that the ATP-fueled contractile actomyosin currents induce the transient accumulation of actin-binding membrane components. As predicted, the active mechanics of actin and myosin at physiologically relevant ATP concentrations drives the system into a nonequilibrium steady state with anomalous density fluctuations and the transient clustering of actin-binding components of the lipid bilayer (15, 33). Finally, connection of this active layer of actomyosin to a phase-segregating bilayer, influences its phase behavior and coarsening dynamics.

Results

Dynamic Association of a Lipid Bilayer Probe with Actin.

We created supported lipid bilayers (SLBs) composed of 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) doped with 2% Ni²⁺-chelated lipids [DGS-NTA(Ni²⁺)]. The Ni²⁺-chelated lipids recruit histidine-rich proteins, including an engineered membrane–actin linker (HYE) constructed from the actin-binding domain of Ezrin (EzrABD), yellow fluorescent protein (YFP), and 10 tandem histidine residues allowing stable SLB binding during the time of an experiment (Fig. S1A). We chose the EzrABD as a membrane–actin linker because we previously found that it is sufficient to confer actin-dependent clustering to fusion proteins in vivo (15). We also created variants of HYE: one with a point mutation in the EzrABD (R579A) that abolishes actin binding (48) and ablates the ability to cluster transmembrane proteins (15), and another, nonfluorescent derivative, HKE, with the tripeptide KCK in place of YFP (Fig. 1 A and B) (49). As expected, bilayers containing either HYE or HKE recruited actin filaments (Fig. S1B), resulting in a thin film adjacent to the bilayer (Fig. 1C and Fig. S1C), whereas HYE(R579A) was unable to localize actin to the membrane. Actin concentrations were restricted to the range of 100–1,000 nM to limit the actin layer thickness to less than 10 actin filaments (see Table S1 for details of the experimental conditions). We controlled the length of these filaments by adding stoichiometric amounts of capping protein (Fig. S1D). To induce contractile stresses and flows, we added the motor protein myosin II and ATP.

Fig. S1.

(A) Graph showing the normalized intensity of SLB-bound HYE after washout at experimental conditions. (B) Area fraction covered by F-actin binding to SLB with increasing concentrations of HYE (black) or HKE (red); plotted are mean and SD, n = 2. (C) Probability distribution of the thickness of a SLB-bound F-actin layer; actin intensity values obtained from line scans were normalized by the average single filament intensity and binned to integer values in a histogram to get an estimation of the actin layer thickness (n = 791). (D) F-actin length probability distribution after actin polymerization with various capping protein concentrations (n = 158–749); Rhodamine-labeled F-actin was added to HYE-containing bilayers at low density, imaged with TIRF microscopy, and length measurements were performed semiautomatic with the ImageJ plugin NeuronJ. (E) Averaged traces from FRAP experiments on SLB-bound His₁₀-GFP (black), HYE alone (blue), or HYE together with 1,000 nM F-actin (light blue); Insets show representative snapshots for recovery of HYE alone (Top) or HYE in the presence of 1,000 nM F-actin (Bottom); each trace shows an average of five experiments; whiskers depict SD; red lines are fitted single-exponential function [derived time constants: b_GFP = 0.142 (±0.002) s⁻¹; b_HYE = 0.113 (±0.002) s⁻¹; b_HYE+actin = 0.076 (±0.001) s⁻¹; the immobile fractions are d_GFP = 7(±1)%, d_HYE = 12(±1)%, d_HYE+actin = 16(±1)%]. (F and G) Immobile fractions (F) and effective diffusion coefficients (G) of membrane-bound proteins derived from individual FRAP experiments. (H) Individual comparison of FCS autocorrelation curves on HYE bound to a SLB under free conditions (blue), in the presence of 1,000 nM F-actin (light blue) or with 1,000 nM short, capping protein bound F-actin (mauve); traces are averages from different scan positions on three to five individual experiments; Inset shows distribution of diffusion time constant obtained by fitting to a model of 2D diffusion. (I) FCS-derived diffusion coefficients of HYE without actin (HYE, n = 22), with bound long (F-actin, n = 23) and short (CP-actin, n = 16) actin. (J) FCS-derived diffusion coefficients of His10-GFP alone (n = 5) or in combination with HKE and no actin (n = 5) or with bound long F-actin (n = 13). (K) Diffusion coefficients of the lipid probe RhoPE derived by FRAP without actin (n = 9) or with 1,000 nM long F-actin bound to the lipid bilayer via HYE (n = 16); diffusion coefficients of HYE in the presence of actin were obtained on the same samples (n = 11). For all box plot diagrams: small squares depict mean values; box heights, the SD; middle lines, the median; and whiskers, the 5–95% range.

Fig. 1.

In vitro setup. (A) Schematic of the in vitro system. (B) Schematic of the actin–membrane linker constructs. (C) Representative images of SLB-bound F-actin and plot of the diffusion coefficient HYE (n = 59–147), HYE(R579A) (n = 15–24), and a combination of HKE and HYE(R579A) (n = 10–39) at various F-actin concentrations. (D) Representative images of SLB-bound rhodamine-labeled F-actin with capping protein (CP) and plot of the diffusion constant of HYE (n = 50–80) and a combination of HKE and HYE(R579A) (n = 10–20) with F-actin of decreasing average length (n = 158–749). Small squares depict mean values; box heights, SDs; middle lines, medians; whiskers, 5–95% ranges. (Scale bars: 10 µm.)

Table S1.

Summary of the experimental conditions of the experiments that are depicted in Figs. 1–5

Figure	[Actin], nM	${\text{L}}_{\mathrm{initial}}^{\text{F}-\text{actin}}$, µm	[Myosin II], nM	[ATP]initial, µM
Fig. 2A, Left	300	9 (±3)	10	100
Fig. 2A, Center	600	3 (±2)	90	100
Fig. 2A, Right	600	5 (±3)	15	100
Fig. 2B	See Table S2
Fig. 3 A–E	600	4 (±1)	60	100
Fig. 4A	700	8 (±2)	50	1,000
Fig. 4B	600	4 (±1)	60	1,000
Fig. 4E, Fig. S4 E–H	See Table S3
Fig. 5B	1,000	N.A.	0	100
Fig. 5C, Left	0	N.A.	0	0
Fig. 5C, Center	1,000	N.A.	0	100
Fig. 5C, Left	1,000	N.A.	100	1,000
Fig. S2B	Like Fig. 2A, Left and Center
Fig. S2E	800	2.5 (±0.8)	30	100
Fig. S2F	750	3 (±1)	100	100
Fig. S2F	800	3 (±2)	50	100
Fig. S2G	750	4 (±2)	100	100
Fig. S3 A and C	800	3 (±1)	30	100
Fig. S3D	1,000	4 (±2)	150	100
Fig. S4A, Top	600	4 (±1)	100	1,000
Fig. S4A, Bottom	500	4 (±1)	100	500 (constant)
Fig. S4C	500	4 (±1)	100	500 (constant)
Fig. S5A	1,000	N.A.	0	100
Fig. S5B	1,000	N.A.	200	500

Membrane-Tethered Actin Filaments Alter the Mobility of HYE Constructs That Are Competent to Bind Actin as Well as Those That Cannot Bind Actin.

We first characterized the lateral mobility of actin-binding EzrABD constructs in the absence and presence of filamentous actin using fluorescence recovery after photobleaching (FRAP) and fluorescence correlation spectroscopy (FCS). The mobility of SLB-bound proteins was probed by observing the diffusion of His10-tagged fluorescent proteins in FRAP experiments. FRAP curves could be well fit to a single exponential function, from which we computed an effective diffusion coefficient; His10-GFP and HYE showed a similar diffusion coefficient in the absence of actin (Fig. S1 E–G).

Adding increasing concentrations of filamentous actin decreased the effective diffusion coefficient of “wild-type” HYE but had no effect on the R579A mutant (Fig. 1C and Fig. S1G), and the immobile fraction of HYE was slightly increased from 12% to 16% in the presence of 1,000 nM actin (Fig. S1F). Decreasing the average length of actin filaments by using capping protein (CP) (Fig. S1D) increased HYE mobility as judged by FRAP (Fig. 1D) and FCS measurements (Fig. S1 H and I). Importantly, the recovery of HYE in regions covered by actin filaments reflects the transient nature of the interaction of EzrABD with actin. Interestingly, when we tethered actin filaments to bilayers with the nonfluorescent Ezrin construct, HKE, and then measured diffusion of HYE(R579A) in the same bilayers, we observed a significant reduction in mobility. The mobility of HYE(R579A) decreased monotonically with increasing actin concentrations, but this effect was less pronounced than that observed with wild-type HYE (Fig. 1C). However, His10-GFP did not show a significant reduction in its mobility in the presence of actin filaments tethered via HKE to SLBs as detected by FCS (Fig. S1J). In addition, the fluidity of the lipid bilayer itself was not affected by the membrane tethered actin network, as the recovery of the fluorescently labeled lipid 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(lissamine rhodamine B sulfonyl) (RhoPE) was unchanged (Fig. S1K). These results are consistent with a previous report on the effect of a dense actin network on the diffusion of other membrane-associated proteins in vitro (41) and show that the association of actin with the membrane bilayer provides an impedance for the mobility of membrane components that can sterically and biochemically interact with actin filaments, evoking the idea of a membrane fence (7, 50). It is to be noted that this effect is diminished when the actin filaments are too short to form an entangled meshwork as observed when we reduced actin-filament length with titrated amounts of capping protein (Fig. 1D and Fig. S1I).

Myosin Motors and Membrane-Bound Actin Filaments Self-Organize into Distinct Configurations.

When we added bipolar myosin filaments—composed of ∼100–500 proteins (51) with lengths between 0.5 and 1.2 µm (Fig. S2A)—to a bilayer-bound thin film of actin filaments in the presence of a fixed amount of ATP and allowed the system to evolve until the ATP was consumed, we reproducibly observed three distinct organizational states: (i) filament bundles, (ii) linked apolar asters, and (iii) disordered lattices of polar asters (Fig. 2A). The orientation of actomyosin networks in the bundled state was correlated over distances >5 µm (Fig. S2 B and C), whereas the polar aster state was characterized by islands of actin reflected in a local minimum of the spatial density correlation (Fig. S2D). The linked apolar aster phase was characterized by a low correlation in filament orientation (Fig. S2 B and C) and by the absence of oscillations in the spatial density correlation (Fig. S2D). These three states (after consumption of the ATP pool) were obtained by varying (i) actin concentration, (ii) filament length, and (iii) filament-to-motor ratios (Fig. 2B and Table S2). Because the polar asters most closely reflect the organization of the contractile platforms that were proposed by our theoretical model, and that are supposed to be responsible for the clustering of membrane particles (15), we focus in the following on these structures and conducted further experiments in conditions geared toward aster formation.

Fig. S2.

(A) Myosin II filament length (in meters) probability distribution (Top) obtained from fluorescence images of Atto-633–labeled myosin II; example image shown below; Inset is a transmission electron microscopy image. [Scale bars: 2 µm and 200 nm (Inset).] Lengths were measured manually with ImageJ. (B) Orientation analysis of bundled filaments and connected apolar asters, and (C) corresponding radial orientation correlations; plotted are mean and SD (n = 10). (D) Density cross-correlation of actin intensities in a disordered lattice of polar asters and connected apolar asters; plotted are mean and SD (n = 10). (E) Exemplar fluorescence image of F-actin asters (Top) and corresponding interaster distance (l) distribution (Bottom) computed from 10 images of the same sample (n = 3,007). (F) Overlay of exemplar STED images of actin (red) and myosin (green) organization in F-actin asters. (Scale bar: 2 µm.) (G) Overlay of actin (red) and capping protein (CP) (cyan) organization in F-actin asters. (Scale bar: 10 µm.) (H) Examples of actin and capping protein organization in connected apolar asters. (Scale bar: 2 µm.) (I) Anecdotal example of a large aster with labeled capping protein showing a second ring of F-actin pointing inward with their barbed ends; plot shows radial intensity of capping protein (cyan) and actin (red); Inset shows the actual fluorescence image. (Scale bar: 10 µm.) (J) Extra-large field view (440 × 420 µm²) of actin asters in a single experiment; images were taken from random positions across the sample and put together in this montage to show homogeneity of aster formation in a standard experiment; image is shown in fire LUT to enhance contrast. (Scale bar: 50 µm.)

Fig. 2.

Myosin-induced formation of polar actin asters. (A) Images of F-actin after myosin II action show three distinct configurations. (Scale bars: 10 µm, 2 µm.) (B) Phase diagram of actomyosin organization obtained by manual classification. (C) Diameter of polar actin asters as a function of mean filament length (n > 20 for each data point) and linear fit showing a linear dependence of aster size to F-actin length (slope, 0.5 ± 0.2); Insets show average projections of image stacks that were created by cropped single actin aster images (n = 12–20) for each indicated F-actin length. (D) Representative images depicting the organization of polar actin asters by imaging rhodamine-actin and Atto-647-myosin II (i), or rhodamine-actin and Alexa 647 capping protein (ii) to obtain a schematic description of a typical actin aster organization (iii). (Scale bar: 2 µm.) (E) X, Y, Z scans of an actomyosin network stabilized with 0.04% glutaraldehyde and imaged with multipoint structural illumination microscopy (SIM) showing the thin membrane-confined actin layer and the local contraction into asters. (Scale bars: in YZ and XZ scans, 2 µm; in XY scan, 5 µm.)

Table S2.

Initial experimental conditions of individual experiments used for the diagram shown in Fig. 2B

State	[Actin], nM	${\text{L}}_{\mathrm{initial}}^{\text{F}-\text{actin}}$, µm	[Myosin II], nM	[ATP]initial, µM
Filament bundles	300	14 (±4)	100	100
Filament bundles	1,000	10 (±3)	40	100
Filament bundles	500	12 (±4)	40	100
Filament bundles	700	5 (±2)	20	100
Filament bundles	700	5 (±3)	20	100
Filament bundles	300	9 (±3)	10	100
Filament bundles	1,000	8 (±3)	10	100
Linked, apolar asters	700	8 (±2)	50	100
Linked, apolar asters	600	5 (±3)	60	100
Linked, apolar asters	600	4 (±1)	60	100
Linked, apolar asters	600	9 (±3)	90	100
Linked, apolar asters	600	3 (±2)	90	100
Linked, apolar asters	600	3 (±1)	90	100
Linked, apolar asters	600	7 (±3)	90	100
Linked, apolar asters	600	4 (±1)	120	100
Linked, apolar asters	1,000	4 (±2)	150	100
Linked, apolar asters	800	3 (±2)	50	100
Linked, apolar asters	500	4 (±1)	70	100
Linked, apolar asters	400	5 (±2)	100	100
Polar asters	600	4 (±1)	60	100
Polar asters	600	4 (±2)	60	100
Polar asters	600	5 (±3)	15	100
Polar asters	200	6 (±2)	15	100
Polar asters	800	3 (±1)	30	100
Polar asters	800	2.5 (±0.8)	30	100
Polar asters	750	2 (±2)	100	100
Polar asters	600	3 (±2)	90	100
Polar asters	500	2 (±0.8)	70	100
Polar asters	800	3 (±2)	100	100
Polar asters	800	4 (±1.5)	50	100
Polar asters	800	2 (±1.2)	100	100
Polar asters	800	1.7 (±0.6)	100	100
Polar asters	800	3 (±1.3)	100	100
Polar asters	800	3 (±2)	50	100
Polar asters	750	4 (±2)	100	100
Polar asters	600	2 (±0.7)	100	100
Polar asters	700	2 (±0.6)	50	100
Polar asters	500	3 (±1.3)	50	100

The state of the system was determined after 90 min of incubation when it had reached a static state.

The Polar Aster Phase.

We next characterized the architecture and dynamics of the polar aster phase in multiple ways. First, we noted that the size of a single aster depends on the average actin filament length (Fig. 2C), in which myosin motors have contracted multiple filaments into a configuration that blocks further rearrangement, indicative of a jammed state. Second, analyzing their spatial distribution revealed that the asters formed a disordered lattice with a characteristic spacing on the order of microns (Fig. S2E and Materials and Methods). Third, the localization of fluorescently labeled capping protein demonstrated the polar nature of the actin asters, in which all filaments are oriented with their (capped) barbed ends facing inward, toward the core of the aster while pointed ends face outward (Fig. 2D and Fig. S2G). Consistent with this interpretation, we also observed that fluorescently labeled myosin II—a barbed-end–directed motor—lies in the cores of asters (Fig. 2D). Similar observations of actin and myosin organization were obtained in samples imaged by stimulated emission depletion (STED) microscopy (Fig. S2F). Scanning the sample with multipoint structural illumination microscopy (SIM) in all three spatial dimensions allowed the clear visualization of the contraction of single actin filaments into asters (Fig. 2D) and of the thin film of membrane bound F-actin (Fig. 2E). In the case of linked apolar asters, however, the distribution of capping protein was more spread out, as some filaments are linked by multiple myosin filaments on both ends (Fig. S2H). Finally, in the central region of sporadically forming larger asters, we observed distinct rings of capping protein (Fig. S2I), another hallmark of myosin-driven polarity sorting and filament organization (52). It is to be noted that the formation of actin asters (as well as the other configurations obtained under different initial conditions) occurred over the major area of the sample, which was tested by the imaging of multiple regions at random positions (Fig. S2J).

Bilayer Components Are Advected by Contractile Actomyosin Flows.

We next asked whether contractile actomyosin flows advect coupled bilayer components. We followed the dynamics of the contractile system and its associated membrane components as the system approached the steady-state polar aster configuration described above. Particle image velocimetry (PIV) of movies of labeled actin filaments revealed regions where the net divergence of the velocity was negative at the sites of aster formation, and with myosin-driven contractile flows of ∼0.5–1 µm/min (Fig. S3 A–C) as observed in other in vitro studies (53). A key prediction of the active composite model is that contractile actin currents will advect actin-binding membrane proteins into clusters. Consistent with this prediction, we observed that the HYE construct forms clusters that colocalize with forming, myosin-induced actin asters (Fig. 3 A and B and Movie S1). Detailed analysis of the regions of aster formation confirmed that the clustering of HYE occurred only when and where actin flows were contractile (i.e., at positions and times where the divergence of actin velocity is negative; Fig. 3C). Whenever the local contractile currents ceased, the associated HYE clusters dissolved, indicating that cluster formation was due to myosin-generated flows and not simply increased actin density in the aster (Fig. 3 B and C).

Fig. S3.

(A) F-actin after addition of myosin II, and a color-coded time projection of F-actin. (Scale bar: 10 µm.) (B) Probability distribution of myosin-induced F-actin contraction velocities (v) obtained from PIV on actin movies. (C) Kymograph of the density–density correlation of the actomyosin contraction shown in A; the arrow indicates the change in the correlation indicating the transition from a rather uniform F-actin distribution to a clustered one, like it is expected for aster formation. (D) Plot of all measured pairs of HYE-actin cross-correlation and variance of actin density corresponding to Fig. 3 D and E. (E) Color-coded time projection of F-actin distribution and HYE-actin cross-correlation of the actin-binding mutant HYE(R579A) during actomyosin contraction bound to the SLB via the nonfluorescent HKE. (Scale bar: 10 µm; duration: 10 min.) (F) Plot of HYE-actin cross-correlation versus the variance of actin density; mean and SD are from data within the x-axis whisker's range. (G) Color-coded time projection of F-actin distribution, HYE-actin and RhoPE-actin cross-correlation during actomyosin contraction. (Scale bar: 10 µm; duration: 20 min.) (H) Plot of HYE-actin (green) and RhoPE-actin (orange) cross-correlation versus the variance of actin density; mean and SD are from data within the x-axis whisker's range.

Fig. 3.

Actomyosin-induced HYE accumulation during actin aster formation. (A) Snapshots of actin (Left), myosin II (Middle), and HYE (Right) during aster formation. (Scale bar: 10 µm.) (B, Top) View of the outlined region in A. (Scale bar: 2 µm.) (Bottom) Intensity profiles averaged over 40 circular regions cantered on actin asters in A, normalized to [0; 1]; control is computed from 10 circular nonaster regions; images were corrected for photobleaching by a single exponential function. (C) Graph of contractile actin flows computed from PIV data (black; Materials and Methods) and HYE intensity (green) averaged over the same regions as in B at indicated condition. (D) Color-coded time projection of actin (Top) and corresponding HYE-actin cross-correlation (Bottom) of SLBs before (Left) and after addition of myosin II, during contraction (Center) and after its halt (Right). (Scale bar: 10 µm.) (E) Plots of HYE-actin cross-correlation versus the variance of actin density from images in D; mean and SD computed from data within the x-axis whisker’s range.

We further characterized the myosin dependence of density fluctuations in membrane-associated proteins by a cross-correlation analysis. We collected time-lapse movies of labeled actin and HYE and, for every 5 × 5 pixel region, computed (i) the temporal cross-correlation coefficient between the intensities of HYE and labeled actin and (ii) the temporal variance of the local actin density (Materials and Methods). The latter distinguishes between regions of static and dynamic actin. We found that HYE and actin dynamics were correlated only in regions where the actin network was dynamic (Fig. 3 D and E and Fig. S3D). These zones of strong correlation appeared only during the contractile period of aster formation and were absent before myosin addition as well as in the jammed steady state (Fig. 3 D and E and Fig. S3D).

Advection of membrane-associated proteins was not the result of bulk hydrodynamic flow in the membrane, but required specific interaction between the membrane component and filamentous actin, because HYE(R579A) did not cluster in bilayers to which actin was linked via HKE (Fig. S3 E and F). Similarly, no effect of actomyosin-driven HYE on bulk lipid dynamics could be detected by the use of the fluorescently labeled lipid RhoPE and the cross-correlation analysis (Fig. S3 G and H). Both the PIV data of actin flows (Fig. 3C) and the cross-correlation analysis (Fig. 3 D and E) indicated that advection drives clustering of actin-associated components during myosin-dependent aster formation.

Emergence of an Active Actin–Membrane Composite at High Concentrations of ATP.

Upon complete ATP consumption (initial ATP, <0.01 mM), our actomyosin networks eventually reached a static steady state consistent with earlier reports (54). The cortex of a living cell, however, is continuously driven out of equilibrium by high concentrations (2–5 mM) of ATP (55). To mimic this situation, we added ATP (1 mM final concentration) to preformed, membrane-associated actomyosin asters. Addition of ATP induced a period of reorganization during which large, static asters dissolved (Fig. 4A and Movie S2), and the system moved to a new steady state, characterized by highly dynamic actin filaments (Fig. S4A) with a statistically uniform, time-averaged density at length scales of >5 µm (Fig. S4B). Time-lapse movies revealed transient, localized accumulations of actin filaments that accreted and dispersed over tens of seconds (Fig. S4A, Upper, and Movie S3). We observed the same result when we reconstituted the system with an ATP-regenerating system (Fig. S4A, Lower). These dynamic density fluctuations persisted for several minutes after addition of ATP (and much longer with the ATP regeneration system) (Fig. S4 B and C), suggesting a nonequilibrium steady state in which continuous remodeling was driven by persistent myosin activity.

Fig. 4.

The remodeling actomyosin–membrane system shows features of an active composite. (A) Aster disassembly after addition of ATP to polar actin asters. (Scale bar: 10 µm.) (B) Color-coded time projection of actin (Left) and corresponding HYE-actin cross-correlation (Right) of a remodeling actomyosin network. (Scale bar: 2 µm; duration, 15 min.) (C) Plot of HYE-actin cross-correlation versus the variance of actin density from images in B; mean and SD computed from data within the x-axis whisker's range. (D) Individual HYE intensity probability distributions for SLBs containing HYE alone, jammed (actin plus myosin II), or remodeling actomyosin (high ATP); dashed lines depict Gaussian fits of data from each condition. (E) Number fluctuations of HYE ($\mathrm{\Delta }N\sim b^{\alpha }$) under the indicated conditions; squares depict mean values; box heights, SDs; middle lines, medians; dots, individual experiments (n = 7–12). (F) Number fluctuations of HYE as a function of interrogation box size (A_box) obtained from one sample at conditions as indicated.

Fig. S4.

(A) Time series of constitutively remodeling actin networks at high ATP concentrations (1 mM) and with an ATP regeneration system ([ATP] = 0.5 mM). (Scale bar: 2 µm.) (B) Kymograph of the density cross-correlation of a disordered lattice of polar asters after addition of ATP; the dissociation of asters is marked in the slower decay of correlation, and steady state is reached after 2 min. (C) Color-coded time projection of actomyosin remodeling in the presence of constant ATP concentrations and the corresponding HYE-actin density cross-correlation. (Scale bar: 10 µm; duration: 20 min.) (D) Plot of all measured pairs of HYE-actin cross-correlation and variance of actin density corresponding to Fig. 4 B and C. (E) SD of the HYE intensity distributions at various conditions; small squares depict mean values; box heights, SDs; middle lines, medians; and dots, individual experiments (n = 7–12). (F) L-skewness of HYE intensity distributions at various conditions (n = 7–12); small squares depict mean values; box heights, the SD; middle lines, the median; and whiskers, the 5–95% range. (G) Plot of obtained from calculating the slopes of the number fluctuation data shown in Fig. 4F; data depict mean values and SDs. (H) Individual traces for number fluctuations analysis (shown in Fig. 4E) of HYE intensities (Top) and slope values (Bottom) calculated from the above traces that correspond to the value (α − 1), where α is the exponent in the relation <N²>^1/2 ∼ <N>^α.

We next asked whether the dynamics of HYE were affected by the continuous remodeling of the actomyosin network. At high ATP concentrations, the cross-correlation analysis of HYE and actin densities had a trend similar to that observed during contractile aster formation. That is, HYE and actin dynamics were strongly correlated in regions of highly dynamic actin (Fig. 4 B and C and Fig. S4D). The active composite framework (33) predicts large, myosin-dependent spatial and temporal fluctuations in HYE density, similar to those observed for GPI-anchored proteins in living cells (15) and for other active systems in both theory and experiment (47, 56). Compared with other phases of the actomyosin system, the pixel intensity distribution of HYE in the driven steady state showed a broader, non-Gaussian distribution, with a significant rightward skew indicative of HYE clustering (Fig. 4D and Fig. S4 E and F), similar to that observed for GPI-anchored proteins in vivo, consistent with theoretical predictions (15).

A key statistical signature of actively driven systems is the appearance of “giant number fluctuations” (56), which we previously observed for GPI-anchored proteins in live cells (15). We tested for the presence of giant number fluctuations in our reconstituted system by measuring variations in HYE intensity over different spatial scales. Briefly, we varied the size of our observation window and measured how fluctuations in fluorescence intensity change with window size (Materials and Methods). We found that, for systems at equilibrium (i.e., no actin, actin alone, or static actomyosin cases), HYE intensity fluctuations scaled with the window size b as b^α with α = 0.49 (±0.02; n = 27), in good agreement with the expected behavior of number fluctuations in an equilibrium system away from criticality (α = 0.5) (56). In contrast, the constitutively remodeling steady state exhibited α = 0.86 (±0.09; n = 7), when evaluated over window sizes larger than the mean F-actin length (≥5 µm) (Fig. 4 E and F, Fig. S4 G and H, and Table S3). We observed these giant number fluctuations in our reconstituted system only at high concentrations of ATP (Fig. 4E) (30). Note that the average size of dynamic clusters produced in this driven steady state was significantly smaller than our smallest interrogation window, and, therefore, we were analyzing fluctuations that occur on a length scale significantly larger than the cluster size (57). This indicates that the large number fluctuations are a consequence of active driving of HYE by the myosin-propelled actin filaments upon ATP consumption, and that this actomyosin–membrane system can be considered as an active composite system.

Table S3.

Initial experimental conditions of individual experiments used for the detection of number fluctuations as depicted in Fig. 4E

State	[Actin], nM	${\text{L}}_{\mathrm{initial}}^{\text{F}-\text{actin}}$, µm	[Myosin II], nM	[ATP]initial, µM
HYE only (all)	0	N.A.	0	0
Actin	1,000	10 (±4)	0	100
Actin	200	6 (±3)	0	100
Actin	700	5 (±2)	0	100
Actin	300	8 (±3)	0	100
Actin	1,000	9 (±2)	0	100
Actin	600	9 (±3)	0	100
Actin	500	8 (±4)	0	100
Actin	600	7 (±2)	0	100
Actin + myosin II (jammed)	600	3	90	100
Actin + myosin II (jammed)	500	N.A.	40	100
Actin + myosin II (jammed)	500	5	40	100
Actin + myosin II (jammed)	600	7	80	100
Actin + myosin II (jammed)	1,000	10	40	100
Actin + myosin II (jammed)	200	6	15	100
Actin + myosin II (jammed)	200	6	15	100
Actin + myosin II (jammed)	800	3	30	100
Actin + myosin II (jammed)	700	5	20	100
Actin + myosin II (jammed)	300	8	10	100
Actin + myosin II (jammed)	1,000	9	10	100
Actin + myosin II (jammed)	600	7	90	100
Actin + myosin II + ATP	300	14 (±4)	100	1,000
Actin + myosin II + ATP	600	3 (±1)	100	1,000
Actin + myosin II + ATP	600	3 (±2)	90	1,000
Actin + myosin II + ATP	600	4 (±2)	60	1,000
Actin + myosin II + ATP	600	5 (±3)	60	1,000
Actin + myosin II + ATP	1,000	10 (±3)	40	1,000
Actin + myosin II + ATP	700	8 (±2)	200	1,000

Consequences of an Actomyosin Layer on a Phase-Separating Lipid System.

After establishing the effect of the active actomyosin layer on the organization of a single passive membrane component (HYE) in a homogenous bilayer, we asked whether this activity could have effects on the configuration of a multicomponent bilayer, namely the phase segregation behavior of lipids in the SLB. For this, we formed a standard ternary lipid mixture of DOPC, 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC), and cholesterol, doped with RhoPE as a marker for liquid disordered domains and DGS-NTA(Ni²⁺) on mica sheets (42), and imaged the system on a confocal microscope with a low magnification (20×, N.A. 0.75) (Fig. 5A and Materials and Methods). After formation of bilayers and subsequent addition of HYE and actin filaments in the homogenous mixed lipid phase at 37 °C, we decreased the temperature to 28 °C to induce phase separation, mounted the chamber on the microscope, and began imaging (with a minimal dead time of ∼60 s). The transition temperature for this lipid mix was about T_c = 33 °C in agreement with previous reports (58), which we only could roughly estimate by warming the sample up from T = 28 °C to T = 37 °C as our setup did not allow controlled, repeated heating and cooling. Although HYE and actin were homogeneously distributed in regions devoid of visible lipid domains (Fig. 5B, Top), HYE was excluded from liquid ordered (lo) domains and located in the liquid disordered (ld) phase as expected for the DGS-NTA(Ni²⁺) lipids (Fig. 5B, Bottom). We note that actin filaments followed HYE and were excluded from lo domains as well, eventually resulting in a more bundled organization in the ld phase irrespective of the F-actin length (Fig. 5B, Fig. S5A, and Movie S4).

Fig. 5.

Consequences of the remodeling actomyosin on a phase-separating membrane. (A) Schematic showing the setup for SLBs with ternary lipid mixture. (B) Example fluorescent images of SLBs containing RhoPE, HYE, and binding actin with regions devoid of domains (Top) or showing lipid segregation (dark regions) (Bottom). (Scale bar: 10 µm.) T = 28 °C. (C) Examples of lo domains (dark regions) in SLBs containing HYE only (Left), bound F-actin (Center), or a remodeling actomyosin (Right). (Scale bar: 10 µm.) (D) Averaged lo domain sizes at different conditions; lines represent averages for each condition (n = 4–6, with 15–158 domains per experiment), shaded area is SD, and t₀ indicates start of imaging. (E) Relative change in net lo domain area between t₀ and 750 s later for indicated conditions; squares depict mean values; box heights, SDs; dots, individual experiments. (F) Color-coded time projection of actomyosin remodeling (from Fig. S5B) and the corresponding HYE-actin density cross-correlation. (Scale bar: 20 µm; duration, 2 min.) (G) Plot of HYE-actin cross-correlation versus the variance of actin density; mean and SD are from data within the x-axis whisker's range.

Fig. S5.

(A) Snapshots of a time series showing the effect of lo domains (dark regions) on actin filament (cyan) and HYE (magenta) distribution. (Scale bar: 20 µm.) (B) Snapshots of a time series showing the effect of a remodeling actomyosin network on a phase-separating SLB with HYE as linker protein. (Scale bar: 10 µm.) (C) Graphs depicting averaged lo domain radii of individual experiments at different conditions (with 15–158 domains in each experiment; t₀ indicates start of imaging). (D) Plot of total lo domain area at the start of imaging (A₀) of individual experiments used in Fig. 5E.

Following the behavior of lo domains over time, a steady growth was detected in the HYE-only bilayers (Fig. 5 C and D). This growth of the average size of lo domains was slowed down by membrane-bound actin filaments, whereat longer, entangling actin filaments acted as barriers for lo domains (Fig. 5 C and D), but shorter actin filaments (shorter than the domain diameter) succumbed to their displacement (Fig. S5A).

In contrast, the presence of active myosin motors together with longer actin filaments stalled the increase of average domain size even though at the outset of observation the phase segregation was comparable to that obtained with or without bound actin filaments (Fig. 5 C and D and Fig. S5 B and C). We next compared net lo-domain area fraction in a given frame from the beginning of imaging (t₀) to later time points. In free SLBs and those with F-actin alone, the net area of lo domains (ΣA_lo) remained roughly constant over this time window but decreased in the presence of the remodeling actomyosin network (Fig. 5E, Fig. S5D, and Table S4 for experimental details). This is likely to be due to a loss of small domains (compared with the actin filament length) by the local remodeling of the actomyosin network in combination with a cessation in growth of larger domains (Fig. S5B and Movie S5). The cross-correlation between actin filaments and HYE in the presence of myosin contraction still showed the signature of actomyosin-induced advection, indicating that it is indeed the actomyosin activity causing the loss of lo domains (Fig. 5 F and G). Interestingly, we observed this lo domain reduction reliably only in the presence of long, entangled actin filament networks, possibly a result of the size-dependent interplay between actin filaments and lo domains.

Table S4.

Initial experimental conditions of individual experiments to check for the size evolution of lo lipid domains in the absence or presence of a remodeling actomyosin network depicted in Fig. 5 D and E and Fig. S5C

State	[Actin], nM	${\text{L}}_{\mathrm{initial}}^{\text{F}-\text{actin}}$, µm	[Myosin II], nM	[ATP]initial, µM
HYE only (all)	0	N.A.	0	0
Actin	1,000	N.A.	0	100
Actin	670	N.A.	0	100
Actin	1,000	N.A.	0	100
Actin	1,000	N.A.	0	100
Actin	1,000	N.A.	0	100
Actin	670	N.A.	0	100
Actin + myosin II + ATP	1,000	N.A.	200	1,000
Actin + myosin II + ATP	1,000	N.A.	200	500
Actin + myosin II + ATP	1,000	N.A.	100	500 (ATP reg.)
Actin + myosin II + ATP	1,000	N.A.	100	500 (ATP reg.)
Actin + myosin II + ATP	1,000	N.A.	100	500 (ATP reg.)
Actin + myosin II + ATP	1,000	N.A.	100	500 (ATP reg.)

Taken together, these experiments indicate that the engagement of the actomyosin layer on a phase-segregating bilayer influences its phase segregation characteristics in a length scale-dependent manner. The actin filament length scale sets a limit on the domain sizes, the actomyosin network can act on. The extent of this interplay will also depend on the strength of actin–membrane link, and to which component of the lipid bilayer actin is linked to, underscoring the complex nature of active actomyosin membrane composite systems that remain to be studied in more detail in future.

Discussion

In the present study, we show that myosin-driven actin networks affect the organization of membrane proteins, transforming the actin–myosin–membrane system under continuous ATP consumption to an active composite.

Examining first the patterns formed by actin filaments and myosin motors, we found that filament length is an important determinant in actomyosin structure formation. As observed in other in vitro works, long actin filaments form bundled structures (37, 54) or contracted networks (35, 37, 59) under the action of myosin II, whereas at the shorter actin filament lengths used here, they gave rise to compact, isolated asters. We find that the following parameters characterize the different static states obtained: (i) actin filament length, (ii) myosin-to-actin filament ratio, and (iii) F-actin concentration. Taking into consideration the complexity of the cell cortex composition (25), the Arp2/3 complex and the Formin, mDia1, are major factors controlling actin polymerization and giving rise to branched and straight filaments, respectively (21). The action of actin filament severing agent cofilin could serve to further differentiate the population of filaments, giving rise to a hierarchy of length scales in which smaller filaments may form structures such as the asters observed here, whereas longer filaments would assemble into a dense meshwork or stress fibers (2). Consistent with the system exhibiting distinct states, similar starting conditions result for certain parameters (e.g., [actin] = 600 nM, [myoII] = 90, L_f-actin = 3 µm), in different configurations, which most likely reflects the lack of precise control of actin filament length in our system and the proximity to a phase boundary. Finally and importantly, a continuous supply of ATP and the transient binding of actin filaments to the bilayer drives the system into a mode of continuous remodeling.

Focusing on the effect of the actomyosin system on the dynamics of membrane components, we found that the process of aster formation transiently accumulated a membrane component coupled to actin (HYE). The PIV and cross-correlation analysis indicate that the transient accumulation was driven by the contractile flows induced by myosin and was not merely a result of the increased actin density at the aster. Furthermore, the advection of HYE by actomyosin did not induce hydrodynamic flows in the bilayer, as evidenced by the lack of clustering of bilayer components that were not capable of binding actin.

At high ATP concentrations (1 mM; or in the presence of an ATP-regenerating system), persistent motor activity continually exerted stresses on the filaments driving only the actin-associated membrane components out of equilibrium. We observed hallmarks of active dynamics on the bilayer, such as persistent advection, clustering, and anomalous density fluctuations. To our knowledge, this is the first time that such a reconstituted active composite system has been described. Our results suggest that the actin cortex may influence plasma membrane organization not only through the interaction of lipids and proteins with a stable actin meshwork (9, 42) or polymerizing actin (40, 44), but also through the flows of actin filaments generated by myosin-induced stresses. Importantly, the flows of myosin-driven short actin filaments influence only those membrane components (proteins and lipids) that can bind to actin such as HYE, whereas the mutant HYE(R579A) and the inert lipid RhoPE do not show a change in their dynamics. We note that this behavior is entirely consistent with our earlier observations in cells and theoretical predictions, where actin-associated components exhibit nonequilibrium behavior, such as giant fluctuations and temperature-independent diffusion behavior whereas unconnected components remain inert (13, 15). Transmembrane proteins with actin-binding domains in their cytoplasmic tails directly associate with actin at the inner leaflet, and we have recently shown that long acyl chain-containing outer-leaflet GPI-anchored proteins interact with actin-associated phosphatidylserine (PS) at the cytoplasmic leaflet, in the presence of adequate cholesterol (60), indicating that the actin-dependent clustering of membrane components requires a direct link to actin filaments and actomyosin dynamics. These may be understood in terms of the active hydrodynamics of actin filaments and myosin (56), which does not necessarily induce measurable hydrodynamic flows in the plasma membrane of unconnected components. We emphasize that these observations do not exclude either the picket fence or phase segregation mechanisms that have been proposed earlier. However, the role of actin and myosin-induced stresses, i.e., processes described by active mechanics, should be included to arrive at a more complete description of the membrane in a living cell (33, 56).

In the context of a phase-segregating membrane bilayer, we found that the remodeling actomyosin layer contributed to a change in the size and dynamics of already formed (lo) domains (reduced/stalled growth), compared with a free membrane or one with associated static F-actin. Although at this point we do not fully understand the reason for the reduced membrane domain segregation, it is likely that the active actomyosin churns up the phase-segregating bilayer to create either domains smaller than the optical limit or simply mixes up the membrane. In addition, our data indicate that the membrane domains influence the actin organization, imposing a minimal length scale for actin filaments on the order of the membrane domain diameter, below which domains will not be affected by actomyosin dynamics.

The effect of slowing of domain growth may be amplified in our system by frictional coupling to the substrate that is likely to influence domain growth and dynamics (61). The influence of F-actin on domain dynamics is expected to be much stronger in a freestanding membrane system such as observed in GUVs (40, 43) or support-free planar bilayers (62). Recent theoretical work from the laboratory of one of us suggests that the phase-segregating active composite membrane will exhibit specific characteristics in the distribution of domain sizes and its scaling behavior (63), and the experiments shown here exemplify the number of effects that such an active composite system can have on their dynamics.

The active composite nature of the cell surface also has implications for cellular function in the context of signaling, as several signaling proteins resident in the cell membrane have the capability to interact with cortical actin either directly (64), via linkers (65), or indeed via transbilayer interactions of inner leaflet lipids (60). Signaling that originates at the cell surface may be influenced, enhanced, or facilitated by interacting with actomyosin-related proteins (such as nucleators) that locally regulate the dynamics of actin such that cell surface proteins are organized optimally for their function, resulting in enhanced reaction rates on the cell surface (66) or an efficient reading of a spatially inhomogeneous external ligand (67); indeed, any function in which the spatiotemporal dynamics of cell surface proteins is important is a potential target.

We emphasize that the system we have described here is a minimal realization of an active composite layer, where we have included only key ingredients: small actin filaments of prescribed length confined to a thin film on a supported membrane bilayer by using an actin-binding binding membrane components (HYE) and, finally, ATP and myosin minifilaments. Our results on the actin configurations obtained after consuming the ATP in the system are consistent with the prediction of our earlier theory [and indeed others have predicted this for related systems (52, 68)]. The length scale of clustering and actomyosin structures demonstrated here are larger than those proposed for the cell in the context of GPI-anchored proteins, but this can be attributed to the differences in length scales (L_actin = 2–7 µm in vitro, L_actin = 0.1–0.3 µm in vivo; L_myoII = 0.5–1.2 µm in vitro, L_myoII = ∼0.2 µm in vivo), composition, and local environment between our in vitro system and the cell. The time required for aster formation at low ATP levels (30–60 s) is longer than expected in cells, likely due to the jamming of actin filaments during this process. However, the actin dynamics observed in the remodeling state (<1 s) more closely resembles the turnover of nanoclusters in the cell (0.1–1 s⁻¹) (11). The ease with which this in vitro system reproduces the essential signatures of an active composite argues for its relevance to the understanding of membrane organization of the living cell.

Materials and Methods

Active actin and myosin were purified from chicken breast with minor modifications of protocols described earlier (69, 70). His-tagged C-terminal fragment of Ezrin and its isoforms (HYE, HKE, and HYE-RA) as well as CP were purified from bacterial lysates using affinity chromatography. Supported lipid bilayers were formed by small unilamellar vesicles of DOPC and NTA-lipids in custom-built experimental chambers on cleaned glass slides, to localize the His-tagged proteins. F-actin length was regulated by titrated amounts of CP during F-actin polymerization or by shear forces through pipetting F-actin. After addition of polymerized actin, myosin II was added to the chamber in ATP-containing buffers and appropriate salt conditions to generate minifilaments (51). Phase-segregating lipid mixtures of multiple lipid species and cholesterol were prepared in chloroform, deposited on experimental chambers formed on freshly cleaved mica supports (∼100-μm thickness), and rehydrated for bilayer formation (42). Fluorescence imaging was carried out on a custom-designed total internal reflection fluorescence microscope, and electron microscopy on myosin II filaments was performed following established protocols (71). Image analysis was carried out by pipelines developed specifically for the purpose of each experiment by using available routines in Image J and Matlab, which are available upon request. More detailed descriptions of the material and methods used in this study are provided in SI Materials and Methods.

SI Materials and Methods

Reagents.

If not otherwise stated, reagents were purchased from Sigma-Aldrich. The lipids 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC), cholesterol, 1,2-dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl)iminodiacetic acid)succinyl] (nickel salt) [DGS-NTA(Ni²⁺)], and 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine-N-(lissamine rhodamine B sulfonyl) (RhoPE) were obtained from Avanti Polar Lipids.

Actin and Myosin II Purification and Labeling.

Actin was purified from chicken breast following the protocol from Spudich and Watt (69) and kept on ice in monomeric form in G-buffer (2 mM Tris base, 0.2 mM ATP, 0.5 mM TCEP-HCl, 0.04% NaN₃, 0.1 mM CaCl₂, pH 7.0); myosin II was obtained from chicken breast following a modified protocol from Pollard (70) and kept in monomeric form in myoII-buffer (500 mM KCl, 1 mM EDTA, 1 mM DTT, 10 mM Hepes, pH 7.0). Chicken actin and myosin II were fluorescently labeled with maleimide-Atto565, and maleimide-Atto633, respectively (Atto-Tec GmbH). For maleimide labeling, monomeric protein was reduced for 30 min on ice with 5 mM TCEP followed by buffer exchange using a PD-10 desalting column (GE Healthcare Life Sciences) with buffer lacking reducing agent (for actin: 0.2 mM ATP, 0.1 mM CaCl₂, 5 mM Hepes, pH 7.0; for myosin: 500 mM KCl, 1 mM EDTA, 10 mM Hepes, pH 7.0). Maleimide dyes were added in 5 molar excess, incubated for 15 min on ice, and the reaction was quenched with 10 mM DTT for 5–10 min. Monomeric actin and myosin were then polymerized at room temperature (RT) by adjusting the buffer conditions (for actin: 50 mM KCl, 1 mM EGTA, 10 mM Hepes, pH 7.2; for myosin II: 50 mM KCl, 1 mM EDTA, 10 mM Hepes, pH 7.2) and separated from unbound dye by centrifugation [for actin: TLA100.3 (Beckman Coulter) at 200,000 × g and RT for 20 min; myosin II: 10,000 × g at 4 °C for 20 min], removal of the supernatant, resuspension of the pellet, and dialysis in protein monomer buffer. For myosin II, functional proteins are separated from dead heads by a round of binding and unbinding to F-actin at 5:1 actin-to-myosin ratio (switch from no ATP to 3 mM ATP) followed by a spin at 200,000 × g for 10 min at 4 °C in a TLA100.3 rotor. The supernatant containing functional myosin II is dialyzed against myoII-buffer.

Expression and Purification of Protein Constructs.

The EzrABD constructs [HYE, HKE, HYE(R579A)] are based on the human C-terminal Ezrin cDNAs described in ref. 48, kindly provided by Helen Morrison, Leibniz Institute of Age Research, Jena, Germany. We added on the N-terminal side of the EzrABD [or EzrABD(R579A)] 10×His domains followed by a yellow fluorescent protein and a single arginine as linker, expressed the protein in Escherichia coli, and purified it as described in detail elsewhere (49). The binding of affinity of the EzrABD for F-actin is reported to be in the nanomolar range (48), and the 10×His domain provided a stable binding to bilayers containing 2%_mol DGS-NTA(Ni²⁺) lipids at buffer conditions used in this study (Fig. S1A).

To generate fluorescently tagged CP, the α1- and β2-isoforms of murine heterodimeric capping protein were cloned into pETM20 and pETM33, respectively, and an N-terminal SNAP-tag was fused to the β-subunit. Both subunits were coexpressed in E. coli (Rosetta) for 16 h at 18 °C and purified by immobilized metal ion affinity chromatography (IMAC) over a 5-mL HiTrap Chelating column (GE Healthcare Life Sciences) followed by combined TEV/Prescission cleavage of the N-terminal tags overnight on ice. After desalting over a HiLoad Desalting column (GE Healthcare Life Sciences), uncleaved protein and free tags, both of which still contained a His-tag, were removed by recirculation over the IMAC column (GE Healthcare Life Sciences). The flowthrough containing the fully cleaved heterodimer was subjected to ion-exchange chromatography over a Mono Q column and labeled with a 1.5-fold molar excess of the fluorescent SNAP substrate (SNAP surface Alexa Fluor 647) (New England BioLabs) overnight on ice. The labeled protein was separated from free dye by gel filtration over a Superose 6 column (GE Healthcare Life Science). Protein was snap-frozen in liquid nitrogen in storage buffer [10 mM Tris⋅Cl (pH 7.5), 50 mM NaCl, 0.5 mM TCEP, 20% (vol/vol) glycerol]. Addition of the N-terminal SNAP-tag did not affect capping activity as measured by polymerization of pyrene-actin in bulk or capping in single-filament total internal reflection fluorescence microscope assays.

Supported Lipid Bilayer and Experimental Chamber Preparation.

Glassware for lipid handling and SLB formation was cleaned with Hellmanex III (Hellma Analytics) following the manufacturer’s instructions, immersed in 3 M NaOH and sonicated for 20 min (Bath Sonicator; Optics Technology). Between the steps, glass was thoroughly rinsed with MilliQ water, and finally blown dry with N₂ (glass slides) or placed for 3 h in a drying oven at 80 °C (glass vials). For the experimental chamber, the middle part of 0.2-mL PCR tubes (Tarsons Products) without their lid were stuck to cleaned cover slides (borosilicate no. 1; 25-mm diameter; Warner Instruments) using UV glue (NOA88; Norland Products) and an UV ozone cleaner (PSD-UV8T; Novascan) for curing. For imaging, the cover slide was mounted on a microscope stage in a fitting metal chamber (RC40-HP; Warner Instruments).

SLBs were formed by fusion of small unilamellar vesicles (SUVs) (5). Lipid mixtures containing 98% mol DOPC and 2% mol DGS-NTA(Ni²⁺) were mixed together with 1 mL of chloroform (Electron Microscopy Sciences) in a cleaned amber glass vial (B7990-2A; Thermo Fisher Scientific), dried under a N₂ stream, placed in a vacuum desiccator overnight, and resuspended in SUV storage buffer [150 mM NaCl, 20 mM Hepes, 5% (wt/vol) sucrose, pH 7.2] to a final lipid concentration of 4 mM. SUVs of ∼80-nm diameter were formed by the extrusion method using a lipid extruder (Avanti Polar Lipids) with an 80-nm pore size polycarbonate filter membrane (GE Whatman).

For SLB formation, 10–20 µL of SUV mix were added to 100 µL of SLB formation buffer (150 mM NaCl, 2 mM CaCl₂, 20 mM Hepes, pH 5.5), incubated for 20 min at RT, and washed 10 times with SLB working buffer (150 mM NaCl, 20 mM Hepes, 0.1 mg/mL β-casein, pH 7.2). For experiments with F-actin, SLB working buffer was replaced by 100 µL of KMEH (50 mM KCl, 2 mM MgCl₂, 1 mM EGTA, 20 mM Hepes, pH 7.2). Use of Hepes as pH buffer instead of imidazole allowed a more stable binding of His-tagged proteins to DGS-NTA(Ni²⁺) lipids.

Formation of Phase-Separating SLBs.

Based on a method described in ref. 42, a lipid mix of 32.6%_mol DOPC, 32.6%_mol DPPC and 32.7%_mol cholesterol with additional 2%_mol DGS-NTA(Ni²⁺) and 0.1%_mol RhoPE was prepared in chloroform at a total lipid concentration of 2 mM. A circular mica sheet (grade V-1, 12-mm diameter; SPI Supplies) was freshly cleaved to ∼100-µm thickness and fixed on a cover slide (borosilicate no. 1; 24 × 40 mm; Thomas Scientific) with UV glue. A drop of lipid mixture was placed on the mica glass; the mica glass was immediately placed on a swing bucket rotor and centrifuged for a minute at 5,000 × g and RT to spread the lipid film, followed by 20 min in a vacuum desiccator to ensure total evaporation of the chloroform. The mica glass was then mounted on a cover slide using double-sided sticky tape, and SLB formation was initiated by injection of 200 µL of SLB working buffer and incubation for 20 min at 43 °C. After 10 times wash with KMEH, HYE followed by F-actin, myosin and ATP regenerator were added sequentially (see below). For each experimental condition, samples were prepared individually, and the chambers were sealed with silicon grease to avoid any evaporation and placed on the microscope stage at 30 °C.

Formation of Actomyosin Network.

In a typical experiment, 10 nM HYE was added to SLBs and incubated for 40 min followed by three washes with KMEH. During this incubation time, F-actin was polymerized. Unlabeled G-actin was mixed with 10%_mol Rhodamine, Atto-565– or Atto-633–labeled G-actin, and 10%_vol of 10× ME buffer (100 mM MgCl₂, 20 mM EGTA, pH 7.2), and incubated for 2 min to replace G-actin–bound Ca²⁺ ions with Mg²⁺. Next, capping protein and polymerization buffer were added to induce F-actin polymerization in a test tube at a final G-actin concentration of 5 µM in KMEH supplemented with 2 mM ATP and 1 mg/mL BSA. After 20- to 30-min incubation, the desired amount of F-actin was transferred to SLBs using blunt-cut 200-µL pipette tips. An incubation of 30 min allowed the F-actin layer to bind to the SLB at steady state; labeled or unlabeled myosin was added afterward and evolution of the actomyosin system was observed for 30–90 min. To induce remodeling of the actomyosin system, Mg-ATP (100 mM) was added to a final concentration of 1 mM or 20× ATP regenerating mix (10 mM Mg-ATP, 100 mM creatine phosphate, and 100 U/mL creatine kinase) to reach a 1× dilution (72). The regenerating mix worked at 37 °C at least for 20 min (Fig. S4C). The open chamber design allowed the addition of each component (e.g., 1 µL of 100 mM ATP into 100 µL of buffer) from top without induction of flows that would perturb the actin network.

Experimental conditions for the presented experiments are depicted in Table S1.

Electron Microscopy.

Electron microscopy on myosin II filaments was performed following the protocol in ref. 71.

Image Acquisition.

Images were generally acquired using total internal reflection fluorescence (TIRF) microscopy on a Nikon Eclipse Ti equipped with a TIRF unit fed by a fiber coupled to an Agilent monolithic laser combiner MLC400 with the laser lines 405, 488, 561, and 653 nm (Agilent Technologies). Images were collected with a 100×, 1.49 N.A. objective either with an EMCCD camera (Evolve 512; Photometrics) yielding a pixel size of 154 nm or with a sCMOS camera (Zyla 5.5 sCMOS; Andor) yielding a pixel size of 65 nm. Multicolor, time lapse, and stream acquisition was controlled and recorded using the software µManager (https://micro-manager.org/).

For measures of membrane particle diffusion, we performed fluorescence recovery after photobleaching (FRAP) on a Zeiss LSM 5 live (Zeiss) equipped with a FRAP unit.

Microscope control and data collection were done via the LSM 5 software package. Alternatively, we performed fluorescence correlation spectroscopy (FCS) using the Confocor 2 module on a Zeiss 510 Meta confocal microscope. The samples were imaged at 488-nm laser line illumination and focused with a 40×, 1.2 N.A. water-immersion lens using the descanned detectors in the confocal scan head; the laser beam for FCS was parked at a specified z distance from the coverslips, based on high counts per molecules, indicating the position of the supported lipid bilayer. The FCS spot size was typically of 200-nm radius. Each measurement comprises fluorescence intensity time traces collected over six iterations of 10 s each. The intensity data are autocorrelated using the on-board hardware correlator. The autocorrelation amplitude versus time data were fitted using maximum entropy method (MEMFCS)-based algorithms (9) to obtain distributions of diffusion timescales.

For phase-separating SLBs on mica, we used a Nikon Eclipse Ti microscope in combination with a confocal spinning disk unit (Yokogawa CSU-22 scan head), and images were collected with an EMCCD camera (Andor Ixon + 897) using a 100×, 1.4 N.A. oil or a 20×, 0.75 N.A. air objective (both Nikon).

For SIM, samples were fixed with 2.5% (wt/vol) glutaraldehyde for 60 min (Electron Microscopy Sciences) and imaged on a home-built instant structured illumination microscope with a 100×, 1.45 N.A. Olympus TIRF oil objective and a pixel size of 55 nm that was set up by Hari Shroff and coworkers (73) at the Marine Biology Laboratory (Woods Hole, MA). For stimulated emission depletion (STED) microscopy, we used a standard Leica TCS STED system in combination with a 100×, 1.4 N.A. Leica “STED white” oil objective, and recorded images of unfixed samples with a pixel size of 30 nm.

Image Analysis.

Basic image analysis, such as measurements of myosin II length and extraction of intensity profiles, was performed in ImageJ (imagej.nih.gov/ij/). F-actin length distribution was measured with the ImageJ plugin NeuronJ (www.imagescience.org/meijering/software/neuronj/). All further analysis was performed with Matlab (MathWorks) as described below.

Actin area occupation.

For the quantification of actin binding to SLBs, images of rhodamine-labeled actin were thresholded to select regions of bound actin filaments only. The area occupation fraction was then computed as the number of selected pixels versus the total number of pixels.

FRAP analysis.

Intensity profiles of the bleached region and a nonbleached control region were analyzed to obtain intensity–time traces, which are corrected for photobleach and normalized with respect to the starting intensity. A single-exponential function was then fitted to these traces, and the exponential fit parameter B is used to compute the diffusion coefficient corresponding to the following equation (74):

$$D=\frac{0.88w^2}{4{\tau }_{1/2}}=\frac{\mathrm{-}0.88w^{2}B}{4\mathrm{In}\left(\mathrm{0,5}\right)},$$

with w equal to the radius of the bleached area and τ_1/2 the time at which the fluorescence signal has recovered 50% of its prebleach value.

Identification of ATP-depleted actomyosin phases.

Images of membrane-bound actin taken after myosin addition and consumption of limited ATP (60- to 90-min incubation) were manually classified into the three described phases on the basis of filament organization. The polar aster phase was characterized by disconnected agglomerations of filaments that were, when resolvable, arranged in a radial fashion around the center of the clump. Filaments in the apolar connected network phase formed a mesh-like arrangement, with concentrated foci due to myosin action. The bundled phase was defined as having a dense packing of actin filaments into locally aligned regions. To confirm that the phases were qualitatively distinguishable from each other, we computed correlation functions on images of each phase following manually segregation: the density (that is, intensity) autocorrelation function to distinguish between the two aster phases and the orientation autocorrelation to distinguish the connected network phase from the bundled filament phase.

The intensity autocorrelation, $\chi \left(\overrightarrow{x}\right)=\langle {\delta }_l\left(\overrightarrow{y}\right){\delta }_l\left(\overrightarrow{y}+\overrightarrow{x}\right)\rangle$, was computed with a fast Fourier transform and summed over the angular coordinate to obtain χ(r). Representative examples of χ(r) may be found in Fig. S2D, depicting the rapid decay and subsequent oscillations characteristic of the polar aster phase, and the typically slower decay found in the connected aster phase.

The orientation of actin filaments was addressed as described in ref. 75. Pixels corresponding to actin filaments were identified by computing the Laplacian of the intensity and thresholding for negative values. The resultant binary image was convolved with a Gaussian and the matrix $M_{i,j}=\langle {\partial }_{i}l,{\partial }_{j}l\rangle w$ was computed for each pixel (where $\langle f,g\rangle w$ denotes the integration of the product fg over space weighted with a Gaussian of size w and centered at the pixel of interest). The local orientation $\overrightarrow{n}=\left(\cos \theta ,\sin \theta \right)$ was obtained from the direction of the largest eigenvector of this matrix, and its spatial autocorrelation function $\langle \delta \overrightarrow{n}\left(\overrightarrow{x}\right)\cdot \delta \overrightarrow{n}\left(\overrightarrow{x}+\delta \overrightarrow{x}\right)\rangle$ was computed. The results of this analysis may be found in Fig. S2 B and C, with the slower decay indicating the presence of aligned bundles.

Characterization of the polar aster phase.

The size of asters in the polar aster phase was extracted by suitably thresholding the actin intensity image and measuring the area of the resultant regions. The length scale corresponding to the area a was computed as $\sqrt{\alpha /\pi }$.

To ascertain the relative position of asters in an automated fashion, we had to resort to a more sophisticated analysis. Regions of local accumulation of actin density were identified by a two-step segmentation protocol. First an exponential was fit to the spatial intensity autocorrelation function χ(r) to obtain a correlation length scale χ_l. The original image was then convolved with a Gaussian of SD ∼ χ_l and local maxima were identified, corresponding to the centers of aster-like structures. All segmented images were inspected manually for accuracy. From the segmentation data, we computed the distribution of nearest neighbor distances, which was well peaked about its mean value. This indicated a well-defined lattice spacing between the asters and hence identifying the phase as a disordered lattice.

Identification of steady state in movies of actin dynamics.

Upon addition of myosin II or excess ATP to membrane-bound actin, the system underwent a period of structural rearrangement. To identify when the system had attained a structural steady state, the spatial autocorrelation χ(r) was computed for each frame of an actin movie. The resultant series of autocorrelation functions were then manually inspected for the attainment of steady state, as representatively shown in Fig. S4 B and C.

Particle image velocimetry.

Actin movies were preprocessed for PIV by time averaging to give an effective frame interval of 20 s. Only movies in which a visually identifiable flow occurred were chosen for analysis to facilitate manual verification. The PIV procedure was performed using the MATLAB package PIVLab (76), the results of which were verified manually by comparing the flow vectors obtained with the flow observed in the movie. Following verification, a histogram of flow magnitudes (that is, the magnitude of the flow vector at each point in the movie) was constructed and a typical flow rate identified from the peak of the distribution.

Cross-correlation of actin and HYE dynamics.

Two color movies of the actin meshwork and the associated membrane marker were bleach corrected and analyzed for dynamic features as follows. First, the spatial intensity autocorrelation function was computed frame-by-frame in the actin channel (as described above) and used to select the portion of the movie in which the actin dynamics has achieved a steady state. Individual frames were then coarse-grained into 5 × 5 pixel bins (∼770 × 770 nm²), and the average fluorescence from each region was extracted as a function of time for both channels, I_actin(t) and I_SLB(t). From this, we computed the cross-correlation ${\chi }_{cc}\left(\tau \right)=\langle \delta I_{actin}\left(\tau \right)\delta I_{SLB}\left(t+\tau \right)\rangle$ for each region, and the Pearson’s correlation coefficient was extracted. The extent of local actin dynamics was gauged by computing the variance of I_actin(t) and these two quantities were compared.

Preprocessing of HYE movies.

To quantitatively characterize features of the HYE intensity, we first had to remove artifacts arising from (i) photo bleaching, (ii) the illumination gradient, and (iii) small, static agglomerates of HYE. The region of near-uniform illumination was determined from the images and the average intensity of that region extracted as a function of time: I_frame(t). This was fit to an exponential function to determine the bleaching rate, which was used to correct pixel intensities in each frame of the movie. To mitigate the effect of the illumination gradient as well as of any static HYE agglomerates, the time average of the movie was then subtracted from each frame. Furthermore, static agglomerates were identified in the first frame of the movie by segmenting the Laplacian of the intensity and neglecting these regions from further analysis.

Intensity distribution.

Pixel intensities from the region of uniform illumination (∼150 pixels square) were then collected from the last 40 frames of each movie and their probability distribution constructed: P(I). To compare across movies, we plotted the distribution of normalized intensity P(i), where

$$i=\frac{i-\langle I\rangle }{\langle I\rangle }.$$

The distributions were then characterized by computing the associated moments and L-moments, the latter chosen for their tolerance of outliers (potentially arising from artifacts that slipped past the preprocessing). In particular, the SD and the L-skewness were computed and compared across conditions (Fig. S4 E and F).

Number fluctuation analysis.

The number fluctuation analysis was performed as described in ref. 15. Briefly, regions of successively larger size, A_box, were chosen in the field of view of a fluorescent HYE movie, and the integrated intensity from each region was computed as a function of time, I_a(t). This function was then processed by subtracting a rolling average of five frames, <I_a(t)>_S and then dividing by the same, giving i(t) = (I_a − <I_a(t)>_S)/<I_a(t)>_S. The SD of i(t) was then computed to give an estimate of <δN²>^0.5/<N>, the logarithm of which was then plotted against the logarithm of the interrogation area, ln(A_box). We fit this to a linear function over a range of areas such that effects of the illumination gradient were mitigated. The obtained slope corresponds to α − 1, where α is the number fluctuation exponent, <δN²>^0.5 ∼ <N>^α.

Domain segmentation and shape analysis.

Domain size distributions from 20× images of RhoPE were extracting by a two-scale segmentation process. Smaller domains were identified by segmentation of the Laplacian of the intensity, whereas larger domains were identified by top-hat filtering with a large structuring element and then thresholding by Otsu’s method.