A tug of war between filament treadmilling and myosin induced contractility generates actin rings

  1. Qin Ni
  2. Kaustubh Wagh
  3. Aashli Pathni
  4. Haoran Ni
  5. Vishavdeep Vashisht
  6. Arpita Upadhyaya  Is a corresponding author
  7. Garegin A Papoian  Is a corresponding author
  1. Department of Chemical and Biomolecular Engineering, University of Maryland, College Park, United States
  2. Department of Physics, University of Maryland, College Park, United States
  3. Biological Sciences Graduate Program, University of Maryland, College Park, United States
  4. Biophysics Graduate Program, University of Maryland, College Park, United States
  5. Institute for Physical Science and Technology, University of Maryland, United States
  6. Department of Chemistry and Biochemistry, University of Maryland, United States

Abstract

In most eukaryotic cells, actin filaments assemble into a shell-like actin cortex under the plasma membrane, controlling cellular morphology, mechanics, and signaling. The actin cortex is highly polymorphic, adopting diverse forms such as the ring-like structures found in podosomes, axonal rings, and immune synapses. The biophysical principles that underlie the formation of actin rings and cortices remain unknown. Using a molecular simulation platform called MEDYAN, we discovered that varying the filament treadmilling rate and myosin concentration induces a finite size phase transition in actomyosin network structures. We found that actomyosin networks condense into clusters at low treadmilling rates or high myosin concentrations but form ring-like or cortex-like structures at high treadmilling rates and low myosin concentrations. This mechanism is supported by our corroborating experiments on live T cells, which exhibit ring-like actin networks upon activation by stimulatory antibody. Upon disruption of filament treadmilling or enhancement of myosin activity, the pre-existing actin rings are disrupted into actin clusters or collapse towards the network center respectively. Our analyses suggest that the ring-like actin structure is a preferred state of low mechanical energy, which is, importantly, only reachable at sufficiently high treadmilling rates.

Editor's evaluation

This important paper uses molecular simulations to explain how actomyosin networks transition from small clusters to the cortex or ring-shaped actin networks. The authors provide compelling evidence that variation in filament turnover rate and myosin concentration triggers a phase transition of these networks. The predictions of this model are consistent with observations made in T cells, where actin ring formation can be induced following their activation by antibodies.

https://doi.org/10.7554/eLife.82658.sa0

Introduction

In eukaryotic cells, actin filaments and myosin motors self-organize into a diversity of shapes (Blanchoin et al., 2014). A shell-like cortex is ubiquitously found under the cell membrane, which is characterized by a mesh-like geometry and plays an indispensable role in defining cellular shape and mechanochemical responses (Salbreux et al., 2012; Stewart et al., 2011; Bovellan et al., 2014). In immune cells such as T cells, the actin cortex reorganizes into a peripheral quasi-2D actin ring that sequesters different signaling complexes in separate concentric domains upon stimulation by antigen-presenting cells (Hui and Upadhyaya, 2017; Yi et al., 2012; Babich et al., 2012; Hammer et al., 2019; Murugesan et al., 2016). Ring-like actin geometries have also been widely found in other sub-cellular structures such as podosomes and axons (Collin et al., 2008; Xu et al., 2013). How actin filaments and associated motors and proteins assemble into such ubiquitous networks that control the shape of living cells and tissues remains poorly understood due to the complex and non-equilibrium nature of actomyosin networks.

In vitro networks reconstituted from purified proteins have been extensively used to derive the minimal set of determining conditions that govern the assembly, growth, and structural properties of actin networks. However, in vitro networks exhibit strikingly different higher order structures as compared to those in cellular networks. In notable contrast to the ring-like or shell-like networks ubiquitously seen in living cells, in vitro experiments primarily result in actomyosin networks comprised of clusters that originate from global geometric collapse due to myosin motor-driven contractility (Murrell and Gardel, 2012; Chuang et al., 2018; Linsmeier et al., 2016; Popov et al., 2016; Niu et al., 2017; Mak et al., 2016; Walcott and Sun, 2010; Komianos and Papoian, 2018). The origins of this disparity likely lies in the quantitatively different parameter spaces occupied by in vitro actin networks as compared to their cellular counterparts.

Actin filaments are highly dynamic, undergoing rapid polymerization and depolymerization, and are subject to contractile forces generated by myosin motors (Levayer and Lecuit, 2012; Fritzsche et al., 2016). Actin polymerization is polarized: monomeric actin (G-actin) binds to the barbed ends of filaments and polymeric actin (F-actin) dissociates from the pointed ends in a process called treadmilling (Bugyi and Carlier, 2010; Pollard, 2007; Ni and Papoian, 2019). We hypothesized that the differences between the predominant actomyosin architectures formed in vitro versus those observed in vivo may arise from the large difference in the corresponding treadmilling rates: in vitro networks reconstituted from purified proteins exhibit treadmilling rates that are often several-fold slower than those observed in vivo due to the lack of regulators that promote actin filament polymerization and disassembly (Kovar et al., 2006; Malik-Garbi et al., 2019; McCall et al., 2019; Jansen et al., 2015; Reymann et al., 2011). We further postulated that these differences in treadmilling rates render in vitro networks less resistant to myosin-induced collapse. A systematic way to explore how treadmilling rates and myosin contractility combine to shape actomyosin network architecture is essential to probe our hypothesis. This is a difficult experimental task, requiring careful manipulation of molecular machinery and actin polymerization kinetics. Such limitations can be overcome by computer simulations, which provide a powerful way to capture the complex chemistry and mechanics of the active cytoskeleton, and bring significant mechanistic insights.

In order to find a minimal set of conditions that lead to the formation of rings and cortices, we combined computer simulations via the open-access platform MEDYAN (Mechanochemical Dynamics of Active Networks; Popov et al., 2016) and experiments on live T cells. We find that the competition between actin filament treadmilling and myosin contractility determines the overall network morphology. Our simulations showed that the speed of actin filament treadmilling drives the network away from global centripetal actomyosin clustering, resulting instead in centrifugal condensation that creates ring-like and cortex-like structures, without tethering filaments to the boundary. On the other hand, increasing myosin motor activity or decreasing filament treadmilling rates lead to a centripetal collapse of actin networks, creating clusters in the network center. Our corroborating experiments on live T cells and simulations mimicking experimental conditions showed that, indeed, hyper-activating myosin II via Calyculin A (CalyA) or inhibiting filament treadmilling via Latrunculin A (LatA) disassembled pre-existing actin rings, causing the network to condense centripetally, resulting in clusters.

Furthermore, our computational analysis indicates that actin filaments located at the network periphery have lower mechanical energy as compared to those that form actomyosin clusters and hence represent the energetically preferred configuration. However, this energetic state is only achievable at sufficiently high treadmilling rates, while at lower treadmilling rates, the system gets trapped in long-lived states where actin filaments instead condense into clusters. In summary, our work shows that a tug of war between filament treadmilling and myosin-induced contraction determines the fate of actomyosin architectures: the energetically favorable ring/cortex states are kinetically accessible only at higher treadmilling rates. Our findings reveal that the assembly and stability of various cellular actin structures are crucially regulated by the fine-tuning of filament treadmilling, which can be achieved by the activation of accessory proteins, such as formin, profilin, and cofilin, via local biochemical signaling.

Results

Dissecting and modeling the T cell actin ring

In order to construct a molecular model of actin rings, we first examined the F-actin distribution in live Jurkat T cells expressing tdTomato-F-tractin (an indirect reporter of F-actin) and MLC-EGFP (myosin light chain). These cells were allowed to spread on an activating glass surfaces coated with anti-CD3 antibody and imaged with time-lapse total internal reflection fluorescence (TIRF) microscopy to visualize the dynamics of actin reorganization (Figure 1—video 1). Upon activation by stimulatory antibodies, the actin cytoskeleton in T cells reorganizes into a ring-like structure characterizing the immune synapse (Figure 1a-c and Hui and Upadhyaya, 2017; Yi et al., 2012; Babich et al., 2012; Hammer et al., 2019; Murugesan et al., 2016). The actin ring consists of an outer lamellipodial region and an inner lamellar ring. In the outer ring, Arp2/3 is activated by WASP near the membrane (Takenawa and Suetsugu, 2007), generating a branched actin network that largely excludes non-muscle myosin II (NMII) as shown in Figure 1c. The inner ring is enriched in actin filaments decorated with NMII which form actomyosin ‘arcs’ (Figure 1b–c). The central region is largely depleted of actin and NMII.

Figure 1 with 3 supplements see all
Actin and NMII distribution within actin rings in T cells and simulation.

(a–b) Representative snapshots of actin (a) and NMII (b) in actin rings of live Jurkat T cells activated on anti-CD3 antibody-coated coverslips. Actin is labeled by tdTomato F-tractin (magenta), and NMII is labeled by MLC-EGFP (green). (c) Merged fluorescence image (left panel) showing distribution of actin and NMII within the T cell actin ring. Inner and outer regions of the ring are indicated. Normalized fluorescence intensity profiles of F-actin and NMII (right) along the dashed line shown in the left panel. (a–c) Scale bar = 10 µm. (d) Setup of simulations using MEDYAN with the major cytoskeletal components labeled. (e) A representative snapshot of the simulation (left)and the corresponding distribution of actin and NMII along the diameter of the ring (right). Cactin=120μM,CNMII=0.1μM,Calphaactinin=4μM,CArp2/3=1μM,Cformin=0.3μM. Scale bar = 1µm.

To understand the biophysical determinants of ring formation and stability, we modeled the formation of actin ring systems using MEDYAN, a simulation platform that combines sophisticated, single molecule level treatment of cytoskeletal reactions, polymer mechanics, and mechanochemical feedback. Actin networks were simulated in a thin oblate cylinder with diameters between 3.8 µm and 10 µm, to mimic the lateral dimensions of small mammalian cells (Figure 1d). Model details can be found in Simulation Methods. We first modeled an actin network with Arp2/3 mediated branching near the periphery. Simulations show that this preferential activation of branching alone is sufficient to generate a lamellipodia-like actin ring, similar to the outer T cell ring, without any other cytoskeletal components, or filament tethering to the boundary (Figure 1—figure supplement 1). We then added the motor protein NMII, crosslinker alpha-actinin, and a filament nucleator formin, which are essential components for actin network remodeling and are ubiquitously found in actin rings and cortices (Salbreux et al., 2012; Blanchoin et al., 2014). Arp2/3 creates a dense dendritic actin mesh at the cell periphery (Svitkina and Borisy, 1999; Takenawa and Suetsugu, 2007), and we hypothesize that NMII is sterically expelled from this region as observed in T cells. To mimic in vivo conditions, we excluded NMII from the peripheral region which contains Arp2/3 mediated branched actin networks. Upon tuning the concentrations of cytoskeletal components and filament treadmilling rates, we found that the network self-organizes into and maintains an outer lammellipodia-like ring and an inner lamellar-like ring with similar actomyosin spatial distribution as the actin ring in T cells (Figure 1e). Also similar to T cells (Hui and Upadhyaya, 2017; Babich et al., 2012; Yi et al., 2012), simulated F-actin undergoes retrograde flow due to filament polymerization against the boundary and NMII generated contraction (Figure 1—figure supplement 2). Even without spatial restrictions on actin or myosin at the periphery, our simulated networks resemble the inner actomyosin ring found in T cells, suggesting that the formation of a ring-like actin structure is a consequence of actomyosin self-organization. We next focused on the origins of this inner actomyosin ring.

Building a minimal model for actin ring formation

To explore the minimal determinants of actin ring formation, we first modeled networks with only actin filaments at different average treadmilling rates (rTM= 0.57 s–1, 1.41 s–1, and 2.21 s–1) based on actin filament assembly kinetics reported from prior experiments (Fujiwara et al., 2007; Kovar et al., 2006; Fujiwara et al., 2018). These systems also include formin at a concentration of 100 nM (Pring et al., 2003; Ni and Papoian, 2019). We found that disordered actin networks were created at all treadmilling conditions tested (Figure 2—figure supplement 1, b). We quantified the spatiotemporal evolution of the network geometry by plotting the median of the radial filament density distribution (Rmedian) as a function of time (Figure 2—figure supplement 1, a). In NMII-free networks, we observed a relatively uniform filament density across the network regardless of treadmilling rates (Figure 2—figure supplement 1, c). In this case, the network geometry is dominated by stochastic filament treadmilling that is not spatially biased. The boundary plays an important role, as the boundary repulsion force inhibits barbed end polymerization such that filaments reaching the boundary rapidly depolymerize and eventually disassemble. The loss of filaments through depolymerization is compensated by the nucleation of new filaments, resulting in dynamic and disordered structures (Figure 2—video 1).

We next explored how these disordered networks behaved upon the introduction of crosslinking and motor contractility. We allowed the network to evolve for 300 s at different rTM as described above to reach a steady disordered state, and then added NMII and the actin crosslinker alpha-actinin to generate contractile forces. The addition of NMII and crosslinkers changed the steady state network geometry, as measured by Rmedian (Figure 2a). For slow treadmilling rates (rTM=0.56s-1), the addition of NMII and alpha-actinin resulted in the clustering of actin filaments (Figure 2b–iii, and Figure 2—video 1). This geometric pattern is consistent with prior in vitro and in silico studies on contractile actomyosin networks (Murrell and Gardel, 2012; Chuang et al., 2018; Linsmeier et al., 2016; Niu et al., 2017), where contractility can be defined as a symmetry breaking event accompanied by a geometric collapse of the network. The average local concentration of actin within the clusters was 234 µM (Figure 2—figure supplement 2), which is almost six-fold higher than the initial G-actin concentration (40 µM), suggesting a high degree of condensation. Although the size and location of actin clusters varied significantly across multiple trajectories (Figure 2—figure supplement 3), a decreasing Rmedian suggests that the overall collapse is centripetal (Figure 2b–iii).

Figure 2 with 10 supplements see all
NMII contractility induces geometric collapse of treadmilling actin filaments.

(a) Normalized medians of radial filament density distribution (Rmedian) at different treadmilling rates (rTM) are shown. The treadmilling rate is defined as the average number of actin monomers added per filament per second at the barbed ends - equivalent to the rate of F-actin depletion from the pointed ends - after reaching the kinetic steady state (See Simulation Methods and Figure 2—figure supplement 7 for details). 0.06µM of NMII and 4µM of alpha-actinin were added at 301s. The inset figure is a snapshot at t=300s of networks with rTM=2.05s-1. The shaded error bars represent the standard deviation across 5 runs. (b–c) Representative snapshots at each treadmilling condition (b) and their radial filament density distribution, ρ(r) (c) are shown. Dashed lines in (c) indicate the position of Rmedian. (d) Representative snapshot of ring-like networks with 80µM actin (left), and ρ(r) of actin rings with 40µM actin and 80µM actin (right) are shown (rTM= 1.35 s–1). (e) A snapshot of a spherical cortex-like network (left) and a slice showing the internal structure (right). (a,b,d,e) Actin filaments are magenta cylinders, NMIIs are green cylinders and linkers are blue cylinders in all snapshots. All scale bars are 1µm.

Our simulations suggest that actin networks are subject to two competing processes: treadmilling, which tends to homogeneously distribute filaments in the network, and NMII-mediated contractility, which tends to trap filaments into clusters. We thus explored changes in the actin network geometry by increasing the treadmilling rate while maintaining the same concentration of NMII. Although filament nucleation occurs stochastically throughout the entire network and there is no filament tethering near the boundary, we discovered that after addition of NMII to rapidly treadmilling networks (rTM=2.05s-1), filaments steadily accumulate at the network boundary (Figure 2a–i, and Figure 2—video 1). During this process, we observed that NMII deformed many filaments and gradually changed their orientation from being perpendicular to the boundary to parallel (Figure 2—video 2). Upon allowing the system to further evolve for several hundred seconds, we found that actin networks transformed into ring-like structures (Figure 2b–i). Networks with intermediate rTM=0.94s-1 form a mixture of clusters and rings (Figure 2b–ii, and Figure 2—video 1).

The resulting actin rings are highly condensed, with a thickness of a few hundred nanometers and exhibiting local actin concentrations similar to those found in actin clusters (263 µM). Increasing the initial G-actin concentration increases the thickness of actin rings (Figure 2d). Most filaments in actin rings are oriented parallel to the boundary (Figure 2—figure supplement 4), forming small actin clusters that undergo azimuthal flow (Figure 2—videos 1; 2). Analogous ring-like patterns were observed on a larger system with a diameter of 10 µm (Figure 2—figure supplement 5). In a spherical system, networks evolved into hollow spherical cortex-like geometries under similar conditions (Figure 2e, Figure 2—figure supplement 6, and Figure 2—video 3).

Competition between filament treadmilling and NMII contractility determines network morphology

To further examine how treadmilling rate regulates the formation of distinct actomyosin architectures, we performed extensive simulations at different treadmilling rates. Indeed, rTM emerges as a key control parameter that governs the steady state network geometry. Below a critical rTM, which is 0.94 s-1 in our simulations, networks geometrically collapse into clusters, while above this critical rTM, they preferentially evolve into ring-like geometries (Figure 3a and b). The radial distribution of the ring state is characterized by higher Rmedian and smaller standard deviation compared with the cluster phase. Interestingly, Rmedian as a function of rTM displays a sharp increase as the network transitions from the cluster state to the ring state (Figure 3b). Because the Rmedian trajectories after adding NMIIs are almost linear before reaching a steady state, we quantified the network remodeling speed by measuring the slopes of the linear part of the Rmedian trajectories. We found that the network remodeling speed is positively correlated with rTM (Figure 3c), indicating that rTM is an important factor driving network structural evolution.

Figure 3 with 1 supplement see all
Treadmilling rate and NMII concentration regulate network structure transitions.

(a) Normalized medians of radial filament density distribution (Rmedian) as a function of time at different rTM(0.56 s-1 to 2.05 s-1) are shown. The shaded colors represent the standard deviation of means for 5 runs. (b) The box plot shows the average Rmedian in the last 500 s of simulation at each treadmilling rate. Solid line connects the mean Rmedian at each rTM. (c) The box plot shows the speed of network remodeling, measured as the slope of the linear part of Rmedian after 300 s. The solid line connects the mean remodeling rates at each rTM. (a–c) Cactin=40μM,CNMII=0.06μM,Calpha-actinin=4μM. (d) Steady state Rmedian at different rTM(0.56 s-1 to 2.05 s-1) and CNMII (0.02–0.2 µM). (e) Representative snapshots of steady state actin network structures at different rTM and CNMII. Representative snapshots of trajectories in (a) are shown in the dashed box. (d-e) Cactin=40μM,Calpha-actinin=4μM. Actin is depcited as magenta cylinders and NMII as green cylinders. Scale bar = 2 µm.

We next varied the NMII concentration (CNMII) at different treadmilling rates, obtaining a phase diagram delineating actin network morphologies (Figure 3d–e). The phase diagram indicates that the critical rTM for actin ring formation increases as CNMII increases. Networks collapse into clusters for rTM below the critical value. The higher CNMII is, the more likely a cluster tends to localize to the geometric center of the network, indicating that NMII induced contractility drives the centripetal condensation. When rTM and CNMII are both low, the network becomes disordered (for example, see Figure 3e, rTM=0.56s-1 and CNMII=0.02μM). Similarly, increasing network contractility by tuning alpha-actinin concentration also results in a transition from ring-like networks at low linker concentrations to bundles and clusters at higher concentrations (Figure 3—figure supplement 1).

Inhibition of actin dynamics disrupts actin rings in live cells and in silico

In order to further understand how treadmilling regulates ring-like actin networks, we experimentally disrupted F-actin dynamics in live Jurkat T cells expressing EGFP-F-tractin. Since it is not feasible to directly control the treadmilling rate in experiments, we used the actin inhibitor, Latrunculin-A (LatA), which decreases the polymerization rate and increases the depolymerization rate by sequestering G-actin and accelerating phosphate release from ADP-Pi-actin (Lodish, 2000; Yarmola et al., 2000; Fujiwara et al., 2018). Upon the formation of the actin ring at the contact zone, LatA (at different concentrations) was added to spreading cells and the resulting effect on the rings was monitored with time-lapse imaging. In order to compare with simulations, we used the fluorescence intensity as a reporter of F-actin levels and calculated a normalized Rmedian to quantify the evolution of the actin network under varying degrees of LatA inhibition compared to vehicle control (Figure 4a). With weak inhibition (CLatA= 250 nM), the ring like structure is perturbed but largely preserved for several minutes (Figure 4b, and Figure 4—video 1). At higher doses of LatA (CLatA= 500 nM and 1 µM, Figure 4c, and Figure 4—video 1), Rmedian rapidly decreases (Figure 4d), indicating a collapse of the network towards the geometric center of the cell. The rate of centripetal collapse of the actin network increases with increasing CLatA (Figure 4e). The dismantling of the actin ring is also accompanied by the formation of F-actin clusters or bundles (Figure 4b–c).

Figure 4 with 2 supplements see all
Inhibition of actin dynamics induces collapse of actin rings in live T cells and in silico.

(a–c) Timelapse montages of Jurkat T cells expressing F-tractin-EGFP spreading on anti-CD3 coated glass substrates. Cells were treated with (a) vehicle control (0.1% DMSO), (b) 250 nM LatA, or (c) 500 nM LatA between 300 and 360 s after contact with activating surface. The first post-treatment image is labeled as 0 s. Timelapse images illustrate the centripetal collapse of the actin ring upon treatment with LatA. Timescales of this collapse depend on the concentration of LatA as can be seen from the timestamps on the images. Scale bar is 10 µm. (d) Quantification of the spatial organization of the actin network using the normalized median of radial filament density distribution. Shaded error bars represent the standard deviations across trajectories (7–11 cells per condition). (e) Box plots showing the rate of centripetal collapse, measured as the slope of the Rmedian distribution after inhibition. (f–h) Timelapse montages of simulations of (f) control, (g) weak inhibition, and (h) strong inhibition. Treadmilling rates in these conditions are 2.05s-1, 1.80s-1, and 0.50s-1, respectively. Indicated rTM is the averaged treadmilling from 1500 s to the end of simulations. Scale bar indicates 1 µm. (i) Medians of radial filament density distribution at different conditions. (j) Rate of centripetal collapse, measured as the slope of the Rmedian distribution after inhibition. (i, j) The shaded color and error bars represent the standard deviation across trajectories, n=5 runs per condition.

To compare with these experimental observations, we perturbed actin network assembly in silico after ring-like networks were established. Based on recent work on reconstituted actin networks under LatA treatment (Fujiwara et al., 2018), we reduced the polymerization rate constants and increased the depolymerization rate constants to mimic the effect of LatA on sequestering G-actin and accelerating depolymerization to closely model the T cell experiments. Actin rings (no inhibition, Figure 4f) were created in the same way as shown in Figure 2a–i. Upon the formation of stable actin rings at 800 seconds, we perturbed actin filament polymerization to different extents to mimic weak and strong LatA inhibition (Figure 4—video 2). We found that actin rings persist under weak inhibition (Figure 4g), while they collapse into clusters under strong inhibition (Figure 4h). The disruption of actin filament assembly also reduces rTM from 2.05s-1 to 1.80s-1 (weak inhibition) and 0.50s-1 (strong inhibition), respectively. Measurements of Rmedian and the rate of collapse (Figure 4i and j) at different inhibition conditions reveal the centripetal collapse of the ring network, reproducing the above-described experimental observations.

Enhancement of NMII activity leads to centripetal contraction of actomyosin rings in T cells and in silico

In order to validate the role of NMII activity in regulating ring-like actin networks, we next altered NMII dynamics in live Jurkat T cells. Under vehicle control (DMSO), actin rings are relatively stable over the timescale of 10 min, and the F-actin distribution displays a steep transition from a depletion zone at the cell center to a high-intensity plateau (Figure 5—figure supplements 1 and 2a). Calyculin A (CalyA) application to enhance NMII activity (Ishihara et al., 1989) leads to an increase in contractility and a centripetal collapse of the actin network (Figure 5a, and Figure 5—video 1), as quantified by the decrease of Rmedian (Figure 5c). On the other hand, upon treatment with Y-27632, an inhibitor of NMII’s upstream regulator, Rho kinase (Uehata et al., 1997), which decreases myosin based contractility, the network becomes more disordered and displays a shallower transition from the central depletion zone to the peripheral plateau (Figure 5b, and Figure 5—video 1). We quantified these changes by calculating the slope of the normalized F-actin intensity from the center to plateau region. As shown in Figure 5d, the slope remained constant over time under vehicle addition, while it decreased upon Y-27632 addition, indicating that the network becomes more diffuse and disordered, and the ring integrity is compromised with loss of myosin contractility. These results confirm that NMII is a central regulator of actin network structure, and high NMII activity is antagonistic to actin ring formation.

Figure 5 with 4 supplements see all
Enhancement or inhibition of NMII regulates actin structure in live T cells and in silico.

(a–b) Time lapse montages of Jurkat T cells expressing F-tractin-EFGP spreading on anti-CD3-coated glass substrates (left) and the normalized radial F-actin intensity (right). After achieving maximal spreading, cells were treated with (a) 50 nM CalyA, or (b) 100 µM Y-27632. Scale bar is 10 µm. (c) The normalized median of radial filament density distribution Rmedian. n=12 cells for vehicle (0.5% DMSO), and n=14 cells for CalyA. Two sample t-test was performed for the first point before drug addition (ns, p=0.83) and 600 s after drug addition (****, p<0.0001). (d) The slope of the intensity profiles over the transition region from the center to the peripheral plateau as a function of time. n=25 cells for vehicle (0.1% DMSO), and 24 for Y-27832. Two sample t-test was performed for the first point (before drug addition) and the last point (660 s after drug addition). (e–f) Timelapse montages of simulations (left) and the normalized radiaul filament density distribution ρ(r) at different times (right) mimicking actin rings in (e) CalyA treatment by increasing NMII levels, and (f) Y-27632 treatment by reducing NMII levels. An actin ring containing 80 µM actin, 0.18 µM NMII, and 4 µM alpha-actinin was pre-initialized as described in the Simulation Methods, and the NMII perturbation was performed at 0 s. The control condition is shown in Figure 5—figure supplement 2. Scale bar is 1 µm. (g) The evolution of Rmedian for different levels of NMII addition. Blue curve is control, and other curves are simulations with indicated levels of NMII added to mimic the CalyA experiment. (h) The slope of the intensity profiles over the transition region from the center to the peripheral plateau as a function of time for simulations of Y-27632 addition. Blue curve is control while orange curve represents simulations after reduction of NMII concentration by 0.04 µM. Two sample t-test was performed for the first three points (before inhibition) and the last three points (510 s to 600 s after drug addition). (g–h) n=5 runs per condition. (a–h) In all figures, 0 s represented the first time point recorded after drug addition (for experiments) or NMII addition/depletion (for simulations). (c,d,g,h) Shaded colors and error bars represent the standard deviation across cells or simulation trajectories.

We then validated the role of NMII in shaping actin structure using MEDYAN simulations. To reduce the computational time, we first initialized actin ring networks and then increased or decreased NMII concentrations (see Materials and methods for simulation setups). Under control conditions, we tuned the actin and NMII concentrations to mimic the conditions tested in T cells (Figure 5—figure supplement 2b). In agreement with experiments, enhancing NMII levels induces centripetal collapse of the network (Figure 5e, and Figure 5—video 2), and the speed of the collapse is proportional to the amount of NMII added to the system (Figure 5f). These results also indicate that a confined boundary is not required for the maintenance of actin rings. On the other hand, upon reduction of NMII levels, the actin ring becomes more disordered (Figure 5—video 2) and the slope of the center to plateau F-actin distribution decreases (Figure 5g and h), in agreement with Y-27632 inhibition experiments.

Energetic origins of structural polymorphism in active networks

We next explored the chemical and mechanical properties of actin networks at various treadmilling rates. We found that the numbers of F-actin filaments, bound linkers, and bound motors remain nearly constant across different rTM, while distributions of diffusive molecules, such as G-actin and nucleators, also did not show spatial localization, being uniformly distributed throughout the simulation volume (Figure 6—figure supplement 1). These observations suggest that ring-like architectures do not form because of the enrichment of soluble constituent molecules near the periphery.

The lack of enrichment of soluble molecules in the periphery suggested a possible energetic origin of the structures. We thus examined the mechanical energy (UMech) of the system, which primarily arises from filament bending in our simulations. For fixed concentrations of NMII (0.06µM) and crosslinker (4 µM), we found that UMech decreases with increasing rTM (Figure 6a). In addition, UMech undergoes a sharp reduction when rTM reaches the critical threshold, with UMech of actin rings being two- to threefold lower than that of clusters. Moreover, we found that UMech is negatively correlated with Rmedian, regardless of the structural state (Figure 6b). Since higher Rmedian indicates localization of actin filaments at the network periphery, this negative correlation indicates that configurations with the lowest mechanical energy are those with a ring-like geometry. These results suggest that the peripheral arrangement of actin filaments is more energetically favorable than more distorted configurations found in centripetal clusters.

Figure 6 with 1 supplement see all
Energetic origins of actin rings.

(a) The box plot shows the steady state UMech at each treadmilling rate. UMech is the sum of the bending energy of actin filaments and the stretching energy of filaments, motors, and linkers. The solid line connects the mean UMech at each rTM. (b) Mechanical energy (UMech) and the corresponding Rmedian at different rTM as indicated in the legend. Each data point represents the average UMech and Rmedian per 100 s of the last 500 s of simulation. (a–b) Cactin=40μM,CNMII=0.06μM,Calpha-actinin=4μM, with varying rTM as shown in Figure 3a–c. n=5 runs per condition. (c) A graphical description showing the proposed energy landscape for generating actin cortices. (d) Schematic showing the formation of actin ring/cortex versus clusters. At low treadmilling rates, networks are dominated by myosin-driven contraction, leading to centripetal collapse into clusters (lower). Faster filament treadmilling allows networks to overcome the myosin-driven centripetal motion, where filaments tend to move to the network periphery due to lower energy (upper).

Discussion

Detailed mechanochemical modeling using MEDYAN shows that active actin networks exhibit a striking morphological transition upon changes in the filament treadmilling rate. We found that two distinct types of dynamic structures emerge due to the interplay between treadmilling rates and NMII contractility in an initially disordered network: (1) actin clusters formed in slow-treadmilling or high CNMII networks and (2) ring-like and cortex-like structures spontaneously assembled in fast-treadmilling and low CNMII networks. This geometric transition does not require filament tethering to the boundary or spatially biased filament assembly. We also observed a sharp transition in the system’s mechanical energy during the transformation from a multi-cluster network to a ring architecture. Such a sharp change in morphology and mechanical energy, induced by tuning filament treadmilling speed, is indicative of a finite size phase transition.

While phase transitions in many biomolecular systems are often driven by passive biomolecular interactions (Li et al., 2012; Brangwynne et al., 2009), in this work we identified a phase transition in cytoskeletal networks that is induced by non-equilibrium actomyosin dynamics. Our analysis shows that the formation of actin rings and cortices arises from the competition between filament treadmilling and myosin induced contraction. The addition of myosin motors and crosslinkers to an initially disordered actin network induces contractile forces, creating actomyosin clusters having higher mechanical energy (Figure 6c). In the language of dissipative structures (Glansdorff et al., 1973), actomyosin clusters are thereby trapped in a non-equilibrium, metastable state that cannot easily transition to a final steady state structure with lower mechanical energy. Rapid filament treadmilling provides a mechanism for escaping these traps (Kim et al., 2014; Popov et al., 2016; McCall et al., 2019), giving rise to smaller clusters that rapidly dissolve and reappear. In this state, the network has more freedom to remodel its structure in order to lower the mechanical energy. Indeed, our analysis suggests that as the actin filament distribution shifts to the network periphery, the smaller curvature at the boundary results in a decrease in filament bending, thereby lowering the mechanical energy of the network (Figure 6a–b). As a consequence, actin filaments at high treadmilling speeds rapidly accumulate at the network periphery, contributing to the build up of an actin ring in flattened volumes or actin cortices in fully 3D spherical geometries (Figure 6d, upper). In contrast, networks undergoing slow filament treadmilling are trapped in cluster-like configurations that have higher mechanical energy. The latter networks are dominated by myosin-driven contractility, leading to a highly non-ergodic state in which actin filaments undergo centripetal collapse (Figure 6d, lower).

Although some other modeling studies have studied how network morphology and contractility are regulated by treadmilling rates or stochastic motion of actin filaments (Vavylonis et al., 2008; Kim et al., 2014; Mak et al., 2016; Oelz et al., 2015), the formation of ring-like or cortical shell-like networks under active force and their underlying mechanisms have not been examined before. In this work, we examined the impact of actin filament treadmilling and myosin contractility on actin structure using a computational model and validated our findings in experiments. Although results from the simulations are quantitatively in agreement with experiments, we note some of the limitations of our model and suggest future directions to improve our simulations. First, we note that we did not explicitly include some significant properties of actin networks in vivo due to the prohibitively high computational overhead associated with modeling Arp2/3-mediated branching and steric interactions. Second, while we used concentrations of cytoskeletal proteins and their spatial distribution that are largely in agreement with the literature, precise measurements of these will significantly improve the simulations.

Since both the ring state and the cluster state conceptually have lower structural entropy compared to the uniform disordered state, we believe that the driving force for actin ring formation is energetic in origin. However, additional work is needed to quantitatively estimate the entropic contribution to actomyosin network self-organization to further validate this argument. The contribution of filament orientation and length to the formation of ring structure remains to to be determined. Some studies have observed that the assembly of ring-like cytoskeletal structures can be achieved by generating long filaments that are mechanically compressed by confinement, or by tethering filaments to the network boundary or membrane (Miyazaki et al., 2015; Dmitrieff et al., 2017; Nguyen et al., 2018; Adeli Koudehi et al., 2019; Litschel et al., 2021). We have shown that forming long filaments is not necessary for generating actin rings, however, filament length can still be an critical parameter in modulating actin network morphology and should be explored in the future. Furthermore, it is likely that filament binding to the cell membrane (Litschel et al., 2021) or the spatially biased localization of actin assembly regulators, such as Arp2/3 (Murugesan et al., 2016), can further enhance the formation of ring-like structures.

In summary, we have shown that rapid treadmilling and the presence of myosin are sufficient to create ring-like or cortex-like actomyosin networks in a system with confined boundaries. These observations suggest that T cells may modulate the actin treadmilling speed or myosin activity upon stimulation by antigen-presenting cells, which generates the actin ring, which is a hallmark of the immunological synapse. On the other hand, cell types that do not assemble ring-like or shell-like actin structures may have intrinsically slower filament treadmilling or higher myosin contractility. Studying these and other regulatory processes will bring new mechanistic insights into the organization and dynamics of cortices/rings and their defects, which occur in primary immunodeficiencies, autoimmune disorders, and cancers.

Materials and methods

Cell culture and transfection

Request a detailed protocol

E6.1 Jurkat T cells (a gift from Brian C. Schaefer, Uniformed Services University, MD, USA) were grown in RPMI medium supplemented with 10% Fetal Bovine Serum (FBS) and 1% penicillin-streptomycin at 37°C in a CO2 incubator. Transfections were performed with 2×105 cells using 1 µg of plasmid by electroporation using a Neon electroporation kit (Thermo Fisher Scientific). Prior to imaging, cells were transferred to CO2 independent L-15 medium (Fisher Scientific). Cells tested negative for mycoplasma contamination using MycoAlert Mycoplasma Detection Kit (Lonza).

Plasmids and reagents

Request a detailed protocol

pEGFP-C1 F-tractin-EGFP was a gift from Dyche Mullins (Addgene plasmid # 58473) (Belin et al., 2014). The tdTomato-F-tractin plasmid was a gift from Dr. John A. Hammer and the MLC-EGFP plasmid was a gift from Dr. Robert Fischer, National Heart, Lung, and Blood Institute. Latrunculin A was purchased from Sigma Aldrich Calyculin A was purchased from Cell Signaling Technology, Y-27632 was purchased from Selleck Chemicals, and dimethyl sulfoxide (DMSO) was purchased from Thermo Fisher Scientific.

Preparation of glass coverslips

Request a detailed protocol

Sterile eight-well chambers (Cellvis) were incubated with 0.01% poly-L-lysine solution in distilled water for 10 min and then dried at 37°C for 1hr. Poly-L-lysine-coated chambers were then incubated with anti-human CD3 antibody (HIT3a clone, Thermo Fisher Scientific) in PBS at a concentration of 10 µg/mL for 2 hr at 37°C or overnight at 4°C. Following incubation, the chambers were washed five times with L-15 and warmed prior to imaging.

Microscopy

Request a detailed protocol

Transfected T cells were seeded on anti-CD3 coated glass coverslips and allowed to activate for 5 min. Chambers were maintained at 37°C using a stage-top incubator (Okolab). Latrunculin A or vehicle (DMSO) were added at specified concentrations 5 min after seeding the cells. Fluorescence and interference reflection microscopy (IRM) images were acquired using an inverted microscope (Ti-E, Nikon, Melville, NY) with a scientific CMOS camera (Prime BSI, Photometrics, Tucson, AZ) with a frame interval of 2s. F-tractin-EGFP was imaged using total internal reflection fluorescence (TIRF), using a 60X, 1.49 NA oil immersion objective. One background image was captured during every session in order to perform background subtraction.

For inhibitor experiments with Calcyulin-A and Y-27632, 50 nM Calyculin-A, 100 µM Y-27632 or vehicle (DMSO) were added after the cells had formed an actin ring. TIRF images were acquired as above with a frame interval of 2 s using a 100X, 1.49 NA oil immersion objective.

Image analysis

Request a detailed protocol

Initial preprocessing of images was done using Fiji (Schindelin et al., 2012). A custom MATLAB script was written to perform background subtraction. The IRM or actin images were used to find the outline and centroid of the cells. 50 uniformly spaced lines were drawn from the centroid and these 50 line profiles were pooled together to generate a histogram of intensities as a function of a normalized distance to the centroid. The median of the distribution of intensities (and hence F-actin) was estimated for each time point. Custom MATLAB script can be found in the repository in the Data Availability Statement.

To calculate the slope of center to plateau F-actin distribution, a cell mask was drawn for each cell (Figure 5—figure supplement 1, left - yellow outline) using a minimum threshold intensity. The centroid (red dot) of the masked cell was identified, and 50 equally spaced lines joining the centroid to the mask edge were drawn and the intensity profile was averaged over all these lines. This plot gives a single intensity line profile from cell centroid to cell edge for a cell at a given time point. Similarly, the line profiles for all the other time points spaced 30 s apart are obtained and normalized using the mean intensity of the cell to account for the effects of photobleaching. The resultant normalized line profile curves are now representative of how the actin distribution changes over time inside the cell (Figure 5—figure supplement 1, right). The intensity profiles typically display a linear regime before they plateau near the cell edge. The linear region of the line profile curves are fit to straight lines (Figure 5—figure supplement 1, right - shaded red and blue lines) to find their slope at each time point and the changes in the slopes over time are then compared across different chemical perturbations.

Simulation methods

Simulation setup overview

Request a detailed protocol

In this work, we employed an open-access mechanochemical platform for simulating active matter (MEDYAN Popov et al., 2016) to investigate the spatiotemporal evolution of actin networks under different treadmilling and myosin motor conditions. MEDYAN accounts for two overlapping phases and their interactions. (1) Diffusing G-actin and unbound formins, NMII and linkers are spatially dissolved in a solution phase. In this phase, the network is discretized into compartments based on the Kuramoto length of G-actin, which is the mean-free path that G-actin molecules are expected to diffuse before undergoing their next reaction (Hu and Papoian, 2010). Diffusing chemical species are assumed to be well-mixed within each compartment, and inter-compartment transports are modeled as stochastic diffusion reactions. (2) Polymeric filaments and bound species comprise the continuous polymeric phase which is overlaid on the solution phase. The polymeric phase is mechanically active, where filament bending, stretching, and steric interactions are taken into account. Bound motors and linkers are modeled as harmonic springs based on the mechanical properties of NMII and alpha-actinin. A boundary repulsion potential restricts filaments within the volume boundary. Filament polymerization is affected by interactions with the boundary, following the Brownian Ratchet model (Peskin et al., 1993). The following chemical reactions stochastically occur among the two phases: filaments can polymerize, depolymerize, and interact with myosin and crosslinker; formins are able to bind to G-actin and nucleate filaments; filaments that are only two monomers long can be rapidly destroyed. The chemical reaction modeling engine is based on an efficient and statistically accurate Next Reaction Method (NRM) (Gibson and Bruck, 2000), which is a variant of the Gillespie Algorithm (Gillespie, 1977).

We initialized de novo cytoskeletal networks in MEDYAN with small seed filaments, 40 µM diffusing G-actin, and 100 nM filament nucleators based on their reported cytoplasmic concentrations in cells (Wu and Pollard, 2005; Kiuchi et al., 2011; Dominguez and Holmes, 2011). Most of the simulations were carried out in a thin oblate geometry, having a diameter ranging from 3.8 µm to 10 µm and an effective height of 200 nm. The spherical simulation volume has a diameter of 4 µm. We tuned the barbed end polymerization rate and pointed end depolymerization rate to model the effects of treadmilling promoters such as formin, profilin, and cofilin. To monitor the actual speed of treadmilling, we define rTM as the average barbed end elongation rate, which is also equal to the shortening rate of the pointed end at steady state. Networks were allowed to assemble with only filament polymerization, depolymerization, nucleation, and disassembly for 300 s. At 300 s, 0.06 µM NMII and 4 µM alpha-actinin crosslinkers are added. The local density of clusters and rings were measured using a customized density based clustering algorithm.

Mechanical models

Request a detailed protocol

Unlike the traditional bead-spring model, the semi-flexible filaments are represented as connected cylinders. The equilibrium length (under zero force) of each cylinder elements varies from 2.7 nm (1 actin monomer) to a maximum of 108 nm (40 actin monomers). Addition of each actin monomer would increase the length of the first or last cylinders by 2.7 nm, and vice versa. Polymerization will create a new cylinder if the cylinder has reached its maximum length. Filaments have a very large aspect ratio, that is, the persistence length of a filament (20μm) is much larger than its diameter (10nm). Thus, it is reasonable to ignore the radial stretching/compression and only allow the axial stretching/compression of a cylinder, which is written as

Ufilamentstr=12Kfilamentstr(lf-lf,0)2.

lf is the actual length of cylinder under force, and lf,0 is the equilibrium length based on the number of actin monomers on this cylinder (each monomer is 2.7 nm). Radial filament deformation is modeled as bending between two connected cylinders:

Ufilamentbending=Kfilamentbending(1-cos(θ-θ0)),

where θ is the angle between the two consecutive cylinders under force, while θ0 is the equilibrium angle that is set to be 0.

A novel volume exclusion exclusion potential is implemented to prevent cylinders overlapping, which is written as

UVol=li,ljδUrirjdlidlj,

where δUri-rj=1/ri-rj4 is the pair potential between two points located on the two interacting cylinders. ri amd rj are the distances between any two points along the cylinder i and j, respectively. This potential can provide a steep enough volume exclusion effect while remain analytically solvable.

Bound NMIIs and linkers are modeled as harmonic springs, and the stretching energy is written as

UNMII/linkerstr=12KNMII/linkerstr(lNMII/linker-lNMII/linker,0)2.

lNMII/linker,0 is the equilibrium length of a linker, which are initialized when a linker /NMII binding reaction occurs as the distance between the paired binding site. lNMII,0 is reset every time a motor walking reaction occurs.

In order to confine all the filaments within the simulation boundary, an exponential boundary repulsion potential is implemented. In the thin oblate system, the actual height of the network is set to be 400nm, and the diameter to 4000 nm. However, filaments would occasionally move out of the mechanical boundary due to rapid treadmilling, leading to simulation failures. To prevent this, we shift the boundary barrier slightly inside the network by a0, and the exponential boundary repulsion is written as

Uboundary=ϵboundarye-(d-a0)/λ,

where ϵboundary=100pNnm is the repulsive energy constant, d is the distance between boundary and filament element, and λ=2.7nm is the screening length. The boundary shifting factor a0 is chosen to be 100nm based on experience. The existence of a0 restricts the effective network boundary to height =200 nm and diameter =3800 nm.

The mechanical model parameters can be found in Table 1.

Table 1
Mechanical parameters.
NamesParametersReferences
Cylinder stretchingKfilamentstr=100pN/nmPopov et al., 2016
Cylinder bendingKfilamentbending=672pNnmOtt et al., 1993
Filament volume exclusionKvol=105pN/nm4Popov et al., 2016
Linker stretchingKlinkerstr=8pN/nmDiDonna and Levine, 2007
NMII stretchingKNMIIstr=2.5pN/nm per headVilfan and Duke, 2003
Boundary repulsionϵboundary=100pNnmThis work

Chemical models

Request a detailed protocol

The chemical engine of MEDYAN is powered by Next Reaction Method (NRM)(Gibson and Bruck, 2000), which is a variant of the Gillespie algorithm (Gillespie, 1977). Overall, the NRM stochastically solves the chemical Master Equation by generating a trajectory of chemical events. In this work, we simulated the following chemical reactions: diffusion, filament polymerization, filament depolymerization, filament nucleation, destruction of filaments, binding of myosin motors and linkers, and motor walking.

The diffusion of molecules is modeled as a single molecule transfer process between neighboring compartments, which follows our stochastic chemical reaction protocol as

DMi0,j0,k0DMi1,j1,k1,

where a diffusing molecule (DM) originally located in compartment i0,j0,k0 is transferred to a neighboring compartment i1,j1,k1. The copy number of this diffusing molecule species is decreased by 1 in compartment i0,j0,k0 and is increased by 1 in compartment i1,j1,k1.

Actin filament (F-actin) polymerization and depolymerization occur at both barbed end (BE) and pointed end (PE) of a filament. These reactions are written as

G-actinF-actin,
F-actinG-actin

It should be noted that G-actin is dissolved in the solution phase, while F-actin is in the polymeric phase.

The nucleation reaction is presented as a two-step reaction based on the mechanism of formin nucleation (Pring et al., 2003; Ni and Papoian, 2019):

Step 1:Formin+G-actinintermediate,
Step 2:G-actin+intermediateFBE-actin+F-actin+PE-actin.

The intermediate is an arbitrary molecule that consists of a formin and a G-actin molecule. We assume step 1 is the rate-limiting step and step 2 is a fast step, thus this intermediate would rapidly react with a G-actin molecule and become a short filament consisting of one F-actin molecule at the pointed end (PE-actin), a regular F-actin molecule, and another F-actin molecule at the formin bound barbed end (FBE-actin). For simplicity, polymerization and depolymerization at FBE are the same as regular barbed end reactions. Formin can dissociate from a filament, which releases a formin molecule into the solution phase and creates a regular F-actin barbed end (BE-actin) on that filament:

F-actin+FBE-actinBE-actin+Formin.

Since new filaments are constantly created by nucleation,, the filament destruction process is required to establish a steady state which maintains a constant total number of filaments. The destruction reaction occurs exclusively when a filament has only two F-actin molecules (a BE-actin and a PE-actin), which destroys this filament and releases two diffusing G-actin molecules as

BE-actin+PE-actin2G-actin.

The binding reactions of myosin motors and linkers are carried out with a slightly different protocol. Firstly, the system will search for all possible binding site pairs on actin filaments and stochastically choose one for binding reaction. The two binding sites of a pair must be located at different filaments. The distance between the two binding sites ranges from 175 to 225 nm for NMII mini filament (Pollard, 1982), and 30–40 nm for alpha-actinin crosslinker (Meyer and Aebi, 1990). After the binding site pair is determined, the binding reaction convert a diffusing motor or linker to a bound motor or linker with two ends attaching to the two binding sites, creating a mechanical linkage. This linkage vanishes when an unbinding reaction occurs, releasing the motor or linker to the diffusing pool. It should be noted that NMII mini filament is an ensemble of 15–30 myosin heads (Verkhovsky et al., 1995), and we model the entire ensemble as a while. To take the variation of the number of myosin heads into account, the number of myosin heads of each NMII mini filament is chosen stochastically for each reaction, and the reaction rate for each NMII binding event is then scaled by the number of myosin heads.

In an active cytoskeleton, myosin motors consume energy from ATP hydrolysis and actively walk along filaments, which is one of the most important sources of contractile force generation. In MEDYAN, a motor stepping reaction is implemented to mimic this effect. For a bound NMII, the stepping reaction is written as

NMIIiNMIIi+1,

where i and i+1 are the NMII locations on the filament before and after walking. NMII is a barbed end walking motor, thus i+1 represents the next binding site towards the barbed end.

Parameters for diffusion and chemical reactions can be found in Table 2.

Table 2
Parameters for diffusion and reactions.
NamesParametersReferences
DiffusionDactin,arp2/3,CP=20μM2/sHu and Papoian, 2010
ActinkonBE=11.6-34.8(μMs)-132 and this work
konPE=1.3(μMs)-1
koffBE=1.4s-1
koffPE=0.8-2.4s-1
Destructionkdestruction=1.0-1.9s-1This work
Nucleationknu=0.005s-1Ni and Papoian, 2019
Formin dissociationkoffformin=0.01s-1Fritzsche et al., 2016
Alpha-actininkonα=0.7(μMs)-1Wachsstock et al., 1993
konα=0.3s-1
NMII head bindingkonM=0.2s-1Kovács et al., 2003
konM=1.7s-1Popov et al., 2016

Mechanochemical models

Request a detailed protocol

Many cytoskeletal reactions, including actin polymerization, myosin motor binding and stepping, and linker binding, are mechanosensitive. To capture this feature, MEDYAN implements mechanochemical models that explicitly allow force-dependent chemical reaction rates.

The effect of boundary force on filament polymerization is described by the Brownian Ratchet model (Peskin et al., 1993), which models the force sensitive polymerization rate kpoly as:

kpoly=kpoly0exp(-Fext/Fpoly,0),

where kpoly0 is the bare polymerization rate under zero external force, Fext is the boundary repulsive force exerted on the filament ends, and Fpoly,0 is the characteristic polymerization force based on the thermal energy and the size of actin monomers.

We used a simple exponential equation to model the slip bond property of alpha-actinin crosslinker:

klinker,unbind=klinker,unbind0exp(Flinker,stretching/Flinker,unbind),

where klinker,unbind0 is the unbinding rate constant under zero external force, and Flinker,unbind is the characteristic unbinding force of alpha-actinin. Flinker,stretching is the stretching force on the linker, while a compressive force on the linker does not trigger the slip bond.

In this work, we model NMII binding as a catch bond, as adapted from the Parallel Cluster Model (Erdmann et al., 2013), such that the force loaded on NMII can reduce its unbinding rate constant:

kNMII,unbind=βkNMII,unbind0Nheadsexp(-FextNheadsFNMII,unbind),

where β is a tunable parameter, kNMII,unbind0 is the unbinding rate constant under zero force, Fext is the total stretching force applied on the NMII, and Nheads is the number of NMII heads.

The NMII walking rate is also mechanochemically sensitive and can be modeled with a Hill type force-velocity relation:

kwalk=kwalk0Fstall-Fext/NheadsFstall+FNMII,pulling/(Nheadsξ),

where Fstall is the stall force of a single NMII head, FNMII,pulling is the pulling force on NMII in the opposite direction of walking movement, and ξ is a tunable parameter.

The mechanochemical model parameters can be found in Table 3.

Table 3
Mechanochemical dynamic rate parameters.
NamesParametersReferences
Characteristic polymerization forceFpoly,0=1.5pNFooter et al., 2007
Characteristic linker unbinding forceFlinker,unbind=17.2pNFerrer et al., 2008
NMII duty ratioρ=0.1Kovács et al., 2003
NMII stall forceFstall=12.62pN per headErdmann et al., 2013
Tunable parametersβ=0.2Popov et al., 2016
γ=0.05pN-1
ξ=0.1

Simulation protocol

Request a detailed protocol

The relaxation time for local deformations of actin networks (Falzone et al., 2015) is much shorter than the timescale of typical chemical events such as motor stepping (Kovács et al., 2003) or filament polymerization (Fujiwara et al., 2007), thereby creating a significant separation of timescales. Hence, the mechanical equilibrium process can be viewed as a pseudo-adiabatic process that can be separated from chemical reactions. Based on this hypothesis, the simulation can be carried out in the following steps:

  1. Chemical reactions occur that evolve the time of the system stochastically.

  2. Pausing chemical reactions when the time step reaches a preset value, which is 10ms in this work. The system then mechanically minimizes the total energy.

  3. Reaction rates are updated based on the tension acting on NMIIs/linkers and load force acting on actin filament barbed ends after mechanical minimization.

  4. Step 1 is repeated based on the updated reaction rates.

This protocol is iterated until we reach 2000 s of simulation time, or until we reach the wall time limit on the Deepthought2 High-Performance Computing cluster at University of Maryland, College Park, whichever comes first.

Defining treadmilling rate and treadmilling inhibition simulation setups

Request a detailed protocol

Although treadmilling in cells is a complex system that involves hundreds of reactions (Bugyi and Carlier, 2010; Floyd et al., 2017), it is simplified to four reactions in this work by considering polymerization and depolymerization at both barbed ends and pointed ends. When a steady state is established, the net barbed end growth rate will equal the net pointed ends reduction rate (averaged over the system), maintaining a constant average filament length. Therefore, we can define a kinetic steady state for treadmilling by monitoring the average filament length of the network as shown in Figure 2—figure supplement 7. We found that such a kinetic steady state could be established after 1000 s in all conditions, and at this state, the average barbed end elongation rate is almost the same as the average pointed end shrinkage rate. Hence, we quantify the average treadmilling rate rTM as the average barbed end elongation rate after 1000 s.

While the treadmilling rate is an elegant and robust way of quantifying the speed of actin network assembly, it is extremely hard to measure in vivo. An alternative way to quantify the speed of actin network remodeling is to measure the turnover timescale, which has been widely studied via an experimental technique called Fluorescent Recovery After Photobleaching (FRAP). To compare with experiments, in our simulation we used a method mimicking the FRAP to calculate the turnover halftime (t1/2, the time required for a network to reach 50% turnover) as developed in our previous work (Ni and Papoian, 2019), and we obtain t1/2168s for the slowest treadmilling condition, and t1/248s for the most rapid treadmilling case. It should be noted that our longest t1/2 is similar to the turnover timescale of some reconstituted networks (McCall et al., 2019), and our shortest t1/2 is comparable to that of in vivo actin cortices (Salbreux et al., 2012). The details of turnover halftime measurement in MEDYAN and how it is related to treadmilling has been discussed in depth in a prior computational study (Ni and Papoian, 2019).

We utilized kinetic parameters measured in vitro (Fujiwara et al., 2007) as the baseline to assemble the slow treadmilling networks. To explore suitable parameters for rapidly treadmilling networks, we looked into the effects of formin and ADF/cofilin. An earlier work (Kovar et al., 2006) has shown that the presence of formin can boost the polymerization rate at the barbed end several-fold over the baseline. For simplicity, we imitated this effect by increasing the barbed end polymerization rate constant (konBE). ADF/cofilin can also promote treadmilling by severing filaments. Importantly, the fragment that contains the pre-existing pointed end is very unstable and would undergoes rapid disassembly (McCall et al., 2019). This observation allows us to mimic the effect of ADF/cofilin by simply increasing the depolymerization at the pointed end (koffPE). For example, we increase the konBE and koffPE to three-fold in the actin ring network as shown in Figure 1a–c (rTM=2.05s-1).

Calculation of local actin concentration for clusters and rings

Request a detailed protocol

In this work, we used a density-based clustering method to define regions that contain actin clusters and rings, and calculated the local F-actin concentration within these regions. We first a generated pixelated map by dividing the network into 100nm×100nm bins and calculated the F-actin concentration within each bins (Figure 2—figure supplement 2a). We then grouped connecting bins with concentration higher than a threshold (160 µM) into clusters (Figure 2—figure supplement 2b). Clusters with size less than 4 bins were ignored. The local actin concentration within clusters was calculated as the average F-actin concentration of these clusters. The local actin concentration within actin rings is calculated using the same method (Figure 2—figure supplement 2c-d).

Simulation setups of Latrunculin A, Calyculin A, and Y-27632 modeling

Request a detailed protocol

Earlier works have shown that LatA affects filament treadmilling in two ways: (1) it sequesters G-actin and (2) it accelerates the phosphate release from ADP-Pi-actin thereby reducing filament polymerization while increasing depolymerization at both ends (Lodish, 2000; Yarmola et al., 2000; Fujiwara et al., 2018). To simulate such effects in the actin ring perturbation simulations, we explore a parameter space that mimicked the effect of LatA treatment: we disrupted rTM by reducing the filament polymerization rate and increasing the depolymerization rates. In the weak inhibition case, we decreased konBE to 11.6(μMs)-1, increased koffBE to 2.1s-1, and maintained koffPE at 2.4s-1. In the strong inhibition case, konBE was decreased to 3.48(μMs)-1, koffBE was increased to 11.2s-1, and koffPE was increased to 4.8s-1. In all simulations, pointed end polymerization rate was set to be constant at 1.3(μMs)-1. Treadmilling rate is consequentially reduced as a result of such disruption.

Calyculin A is an enhancer of NMII activity by inhibiting myosin light chain ATPase, while Y-27632 inhibits Rho kinase, a upstream regulator of NMII. Thus, we model their effects by increasing or decreasing the NMII levels after actin ring formation to match the T cell experiment. In the CalyA experiment, actomyosin ring collapses while maintaining the ring-like geometry. We realize that such ‘whole ring contraction’ is difficult to achieve at the low actin concentration (Cactin=40μM) that we used other conditions. At low actin concentration, enhancing NMII activity often simultaneously cause centripetal collapse as well as the local collapse that disassemble the ring-like structure, due to lack of filament-filament connectivity. To overcome this issue, we double the actin concentration to Cactin=80μM and adjust CNMII to 0.18 µM in the model. Such a high concentration of actin and motor protein significantly reduces the computational efficiency, therefore we initialize the ring-like actin structure instead of starting from a disordered network. In the control condition as shown in Figure S9, network will slightly contract but can maintain the ring-like structure.

Data availability

Source Data files for experiments and the modeling code are available in Digital Repository at the University of Maryland(DRUM): https://doi.org/10.13016/9t26-ovid.

The following data sets were generated
    1. Ni Q
    2. Wagh K
    3. Pathni A
    4. Ni H
    5. Vashisht V
    6. Upadhyaya A
    7. Papoian GA
    (2022) Digital Repository at the University of Maryland
    Data for "A tug of war between filament treadmilling and myosin induced contractility generates actin ring".
    https://doi.org/10.13016/9t26-ovid

References

  1. Book
    1. Lodish HF
    (2000)
    Molecular Cell Biology
    W.H Freeman.
    1. Ott A
    2. Magnasco M
    3. Simon A
    4. Libchaber A
    (1993) Measurement of the persistence length of polymerized actin using fluorescence microscopy
    Physical Review. E, Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics 48:R1642–R1645.
    https://doi.org/10.1103/physreve.48.r1642

Decision letter

  1. Alphee Michelot
    Reviewing Editor; Institut de Biologie du Développement, France
  2. Anna Akhmanova
    Senior Editor; Utrecht University, Netherlands
  3. Anatoly BK
    Reviewer

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "A tug of war between filament treadmilling and myosin induced contractility generates actin ring" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Anna Akhmanova as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Anatoly B. Kolomeisky (Reviewer #1).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

You will find that the reviewers have a very positive opinion of your work. Their comments should be easy to address. Please respond to them point by point.

Reviewer #1 (Recommendations for the authors):

I have several specific comments and suggestions that might improve the paper:

1) It would be nice to discuss why rings are observed only in a few types of cells, but not in all of them. Is it known?

2) NMII on page 5 should be properly defined first - it is not done here.

3) On page 6, the authors said that they "excluded NMII from the peripheral region..." But what would happen if simulations started with motor proteins uniformly distributed over the system?

4) I would add a brief discussion that entropic terms most probably are not important for this system, and thus the mechanical energy provides a valid thermodynamic quantity to decide about the proper phase. This is because one could naively argue that in the ring structures the entry is reduced due to higher density.

Reviewer #2 (Recommendations for the authors):

A] Expanding on public review:

– Additional simulation controls are needed:

While the mean length of filaments according to the depolymerization rate is given, the length distribution in the different conditions is not provided – while this distribution has a strong effect on the ring-like organization of actin in confinement. Moreover, in the simulation of lat A treatment, a control that in-silico actin concentration matches the in-vivo one is missing.

– Confusing description of rings/experiment – simulation discrepancies

For instance, the ring in the control of latA treatment (Figure 4a) seems much more focused than that of WT (Figure 1A). In figure 5, supp. 2 it is hard to understand why the simulated ring is contracting in the control condition. Is time 0 not the stationary solution? Then, how is time 0 chosen?

Also, in figure 1, the ring seems much more realistic than in the following.

I am assuming the authors focused on a simpler system with fewer ingredients to establish the phase diagram, but this needs to be made clearer. Also, why not do a phase diagram with a simpler system, and a more realistic system to compare to experiments?

For now, the comparison between experimental results and simulations is not as strong as claimed. Either the conclusions could be toned down a bit, or the discrepancies should be clearly discussed.

– MEDYAN license

This is not directly relevant to this article but falls under the public review guideline "the utility of the methods and data to the community".

While MEDYAN being an open source software is a great boon for the community, it is crippled by its own license. The item 3 "Users can modify the MEDYAN source code for their own academic and research purposes, but cannot redistribute modified MEDYAN source code that differs in any way from Papoian lab's current MEDYAN distribution" goes against the core idea of open source scientific software: if a team decide to build upon MEDYAN, they will not be able to publish the modified code, and thus will not be able to share their results in a significant manner.

Moreover, the guideline "cannot redistribute any other codes that use or extend any MEDYAN source code" prevents anyone from sharing wrappers and utilities for MEDYAN. Most scientific software uses the GPL (most restrictive) or MIT (most permissive) license with great success.

Our main value as scientists is the production and sharing of knowledge – such a restrictive license does not seem to achieve this goal.

– Patches as a metastable state: this is a bit confusing or over-stated. First, this is a highly out-of-equilibrium active system, and I do not think "metastable" quite applies. Both are fundamentally unstable states. A more likely claim is that treadmilling allows for a faster relaxation of filament elastic energy.

B] Additional comments:

– This is a very well-written article that is highly interesting – yet easy to read.

– The claim that ring organization is not understood, not found in vitro, is far-fetched. First, actin rings in vitro have been observed, e.g. Miyazaki et al. 2015 to name one. Second, simulation of actin rings exists, cf Vavylonys 2008, Hang 2015, Nguyen 2018, Koudehi 2016; see also Dmitrieff 2017 for a ring of microtubules. In all these systems, confinement and filament length plays a major role in a ring formation, but the interplay between the rate of treadmilling and motor activity has not yet been really discussed to my knowledge. This is why not only the mean, but also the distribution of filament length has to be documented, and the role of confinement has to be discussed. The existing literature has to be more discussed.

– The difference in filament orientation in Figure 2 supp 4 is striking. I suspect that this could make for a much more striking phase diagram (by computing some order parameter for instance) than the current phase diagram.

https://doi.org/10.7554/eLife.82658.sa1

Author response

Reviewer #1 (Recommendations for the authors):

I have several specific comments and suggestions that might improve the paper:

1) It would be nice to discuss why rings are observed only in a few types of cells, but not in all of them. Is it known?

Why different cell types display distinct higher order structures such as actin rings is not well understood. In some cell types, ring-like actin networks are formed to subserve distinct biological functions. For example, actin rings form in T cells upon stimulation by antigen-presenting cells, where the actin ring dynamically aggregates receptors to initialize the formation of immunological synapses. The assembly of actin ring may also be a consequence of preferential localization of actin regulatory proteins such as Arp2/3 and formins. In neuronal axons, actin organizes into circumferential rings that are evenly spaced along the long axis, providing a structural framework for membrane channel organization. The main purpose of our manuscript is to determine the minimal required conditions for the formation of actin rings. Our results suggest that lack of ring-like structures in some cell types is potentially due to low filament treadmilling rates or high myosin activity, which can be tested experimentally in these systems. We have now highlighted some of these points in the Discussion section (pg, 19-20).

2) NMII on page 5 should be properly defined first - it is not done here.

NMII stands for non-muscle myosin II. We have added this definition to the manuscript.

3) On page 6, the authors said that they "excluded NMII from the peripheral region..." But what would happen if simulations started with motor proteins uniformly distributed over the system?

The Arp2/3 based actin network at the ring periphery forms a dense meshwork1,2, which can sterically exclude NMII mini filaments. Because simulating steric interactions of myosins with actin filaments in dense dendritic networks is computationally expensive (which is why this is not yet currently implemented in MEDYAN), we have taken a simpler approach by excluding NMII from the peripheral region to mimic the steric interactions. If NMIIs are allowed to access all regions in our simulations, we expect that the inner ring and the outer ring will merge. We have elaborated on the reasoning for excluding NMII from peripheral regions in the manuscript (pg. 6).

4) I would add a brief discussion that entropic terms most probably are not important for this system, and thus the mechanical energy provides a valid thermodynamic quantity to decide about the proper phase. This is because one could naively argue that in the ring structures the entry is reduced due to higher density.

We thank the reviewer for bringing up this point. Because the system is far from equilibrium, it is difficult to quantitatively estimate the entropic contribution. Conceptually, the uniform disordered state should have higher structural entropy, while both the ring state and the contracted clusters should have lower entropy. This argument suggests that the driving force for the ring state is energetic in origin. We now mention this in the Discussion section (pg. 20).

Reviewer #2 (Recommendations for the authors):

A] Expanding on public review:

– Additional simulation controls are needed:

While the mean length of filaments according to the depolymerization rate is given, the length distribution in the different conditions is not provided – while this distribution has a strong effect on the ring-like organization of actin in confinement. Moreover, in the simulation of lat A treatment, a control that in-silico actin concentration matches the in-vivo one is missing.

We thank the reviewer for bringing up this point. We have now added a figure showing the filament length distribution at different treadmilling rates (Figure 2-supp. 7b). We find that the filament length (mean length ~ 0.4 um) is much smaller than the confinement size (the diameter of the network is 4um) even in some extreme cases. Thus, we believe filament length is not a critical parameter for ring formation in our simulations. This is different from other in vitro work where long filaments lead to ring formation. We have now added this in the Discussion section of the manuscript (pg. 20).

We note that actin concentration in vivo has not been well quantified, including in T cells. We use 40 µM as the actin concentration, on the same order of 100 µM as reported in other cell types3–5. We now explicitly mention the choice of actin concentration in the Simulation Methods section, and have added additional references to address this point (pg. 24).

– Confusing description of rings/experiment – simulation discrepancies

For instance, the ring in the control of latA treatment (Figure 4a) seems much more focused than that of WT (Figure 1A).

We have replaced Figure 4a with a more representative cell. We note that actin rings in T cells display some degree of heterogeneity in terms of cell size, actin density, ring thickness, etc. Figure 1a shows a clear separation of the inner ring and the outer ring, but in some cases the separation is not as clear. Some of the observed heterogeneity may be also due to differences in the levels of the exogenously expressed fluorescently labeled F-tractin.

In figure 5, supp. 2 it is hard to understand why the simulated ring is contracting in the control condition. Is time 0 not the stationary solution? Then, how is time 0 chosen?

In Figure 5e,f and Supp. Figure 2, time 0 is the timepoint where we start to add or remove NMII, while time 0 in the experiment is when the drug was added. We edited the caption to make this clearer. The ring is observed to slowly contract even in the control case due to a different simulation setup compared to the stationary ring in Figure 2 and 3. There are two main changes: (1) we increased the total actin concentration to 80 µM, and (2) we initialized a ring-like actin network rather than starting from a disordered network. Stationary rings are formed at a lower actin concentration (40 µM) and low myosin concentrations, but introducing additional myosin to this ring network leads to the formation of contractile clusters instead of a contracting ring. A higher actin concentration (80 µm) is required to obtain a contracting ring upon the introduction of additional myosin. To reduce computational costs, we simulated a pre-formed ring. However, the pre-formed ring at 80 µM actin exhibits a slow contraction. Although we cannot create a stationary ring under these conditions, we note that the control ring contracts significantly slower than upon the addition of excess myosin. We therefore believe that our results are in qualitative agreement with the experiments, given the complexity of the system. The details and rationale are discussed in Method section “Simulation setups of Latrunculin A, Calyculin A, and Y-27632 modeling”(pg. 34-35). It should be noted the actin rings in T cells also tend to slowly contract even under control conditions (some quantifications can be found in Figure 5c and Figure 5-supp.2a).

Also, in figure 1, the ring seems much more realistic than in the following.

I am assuming the authors focused on a simpler system with fewer ingredients to establish the phase diagram, but this needs to be made clearer. Also, why not do a phase diagram with a simpler system, and a more realistic system to compare to experiments?

For now, the comparison between experimental results and simulations is not as strong as claimed. Either the conclusions could be toned down a bit, or the discrepancies should be clearly discussed.

We agree that in vivo cellular systems are much more complex than we can simulate. There are two main reasons to focus on a simpler system: (1) our main goal is to explore the minimal requirements for actin ring formation, and (2) simulating more complex systems would have significantly higher computational costs. This high computational cost is also why we cannot work on a more realistic system (i.e. add Arp2/3 and increase actin concentration) to more directly compare with experiment. We have revised the section on the comparison between the simulation and experiment to highlight the limitations of the work, and proposed potential future improvements to the simulation in the Discussion section (pg. 19-20).

– MEDYAN license

This is not directly relevant to this article but falls under the public review guideline "the utility of the methods and data to the community".

While MEDYAN being an open source software is a great boon for the community, it is crippled by its own license. The item 3 "Users can modify the MEDYAN source code for their own academic and research purposes, but cannot redistribute modified MEDYAN source code that differs in any way from Papoian lab's current MEDYAN distribution" goes against the core idea of open source scientific software: if a team decide to build upon MEDYAN, they will not be able to publish the modified code, and thus will not be able to share their results in a significant manner.

Moreover, the guideline "cannot redistribute any other codes that use or extend any MEDYAN source code" prevents anyone from sharing wrappers and utilities for MEDYAN. Most scientific software uses the GPL (most restrictive) or MIT (most permissive) license with great success.

Our main value as scientists is the production and sharing of knowledge – such a restrictive license does not seem to achieve this goal.

We thank the reviewer for bringing up this concern. First, we would like to note that highly influential scientific software come with licenses at all levels of permissiveness. For example, among broadly used molecular dynamics codes, Gromacs is issued under LGPL, OpenMM under MIT License and LGPL, while NAMD, AMBER and CHARMM have restrictive bespoke licenses (in case of AMBER and CHARMM, they are not even free for academic users). Nevertheless, all these software thrive and are productively used, where the nature of the license may influence some users’ decision on which molecular dynamics code to use.

In the context of mesoscale biomolecular simulations, Cytosim and Affines are GPL, while MEDYAN has a more restrictive license, giving choices to the users who are concerned about the license issue. Interestingly, for over 10 years (2007-2016) Cytosim was closed sourced, with only the executable being available, until it was open sourced in 2016.

To address the substance of the reviewer’s concern: we certainly would love for the users to write and distribute “wrappers and utilities for MEDYAN”. We will explore how MEDYAN’s license can be changed to encourage these possibilities and ask the tech transfer office to help. We would like to point out that the intellectual property for MEDYAN does not belong to our laboratory. It belongs to the University of Maryland, which has a tech transfer office that makes final decisions on software licenses based on an internal review involving attorneys, taking into account, in particular, the commercial potential for the software, among other factors.

We are hoping that the above discussion shows that choosing a license for a complex scientific software, such as MEDYAN, is a difficult endeavor, involving various legal and technical concerns. With that said, we will do our best to increase permissiveness of MEDYAN’s license to address some of the reviewer’s concerns.

– Patches as a metastable state: this is a bit confusing or over-stated. First, this is a highly out-of-equilibrium active system, and I do not think "metastable" quite applies. Both are fundamentally unstable states. A more likely claim is that treadmilling allows for a faster relaxation of filament elastic energy.

To avoid confusion, we have removed the word metastable from the introduction. However, we would like to note that in the context of Prigogine’s framework of dissipative structures, some structures may represent long-lived states that eventually transition to a final steady state structure. In this sense, we think of these long-live cluster states as metastable. In the revised text, we mentioned this point in the Discussion section (pg. 19).

B] Additional comments:

– This is a very well-written article that is highly interesting – yet easy to read.

– The claim that ring organization is not understood, not found in vitro, is far-fetched. First, actin rings in vitro have been observed, e.g. Miyazaki et al. 2015 to name one. Second, simulation of actin rings exists, cf Vavylonys 2008, Hang 2015, Nguyen 2018, Koudehi 2016; see also Dmitrieff 2017 for a ring of microtubules. In all these systems, confinement and filament length plays a major role in a ring formation, but the interplay between the rate of treadmilling and motor activity has not yet been really discussed to my knowledge. This is why not only the mean, but also the distribution of filament length has to be documented, and the role of confinement has to be discussed. The existing literature has to be more discussed.

We agree with the reviewer: the assembly of actin rings has been observed in vitro and in silico, but these rings typically require filaments (or filament bundles) longer than the confinement dimension or require membrane filament tethering. We have added these references and revised the manuscript to discuss these points (pg. 20).

– The difference in filament orientation in Figure 2 supp 4 is striking. I suspect that this could make for a much more striking phase diagram (by computing some order parameter for instance) than the current phase diagram.

This is an interesting point, but we believe that filament orientation is less important in the cluster case. Filament orientation is vital when located at the network periphery, where filaments perpendicular to the boundary will quickly depolymerize due to boundary exclusion and the Brownian Ratchet effect. However, many clusters are located far away from the boundary such that filaments can maintain treadmilling regardless of their orientation. That is why we believe a phase diagram of filament orientation may not be particularly informative.

Reference:

1. Svitkina, T. M. and Borisy, G. G. Arp2/3 complex and actin depolymerizing factor/cofilin in dendritic organization and treadmilling of actin filament array in lamellipodia. J. Cell Biol. 145, 1009–1026 (1999).

2. Takenawa, T. and Suetsugu, S. The WASP-WAVE protein network: connecting the membrane to the cytoskeleton. Nat. Rev. Mol. Cell Biol. 8, 37–48 (2007).

3. Kiuchi, T., Nagai, T., Ohashi, K. and Mizuno, K. Measurements of spatiotemporal changes in G-actin concentration reveal its effect on stimulus-induced actin assembly and lamellipodium extension. J. Cell Biol. 193, 365–80 (2011).

4. Wu, J.-Q. and Pollard, T. D. Counting Cytokinesis Proteins Globally and Locally in Fission Yeast. Science (80-. ). 310, 310–314 (2005).

5. Dominguez, R. and Holmes, K. C. Actin structure and function. Annu. Rev. Biophys. 40, 169–86 (2011).

https://doi.org/10.7554/eLife.82658.sa2

Article and author information

Author details

  1. Qin Ni

    Department of Chemical and Biomolecular Engineering, University of Maryland, College Park, College Park, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Investigation, Visualization, Methodology, Writing – original draft, Project administration, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0738-1817
  2. Kaustubh Wagh

    Department of Physics, University of Maryland, College Park, College Park, United States
    Contribution
    Conceptualization, Data curation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    Contributed equally with
    Aashli Pathni, Haoran Ni and Vishavdeep Vashisht
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8514-027X
  3. Aashli Pathni

    Biological Sciences Graduate Program, University of Maryland, College Park, College Park, United States
    Contribution
    Formal analysis, Investigation, Methodology, Writing – review and editing
    Contributed equally with
    Kaustubh Wagh, Haoran Ni and Vishavdeep Vashisht
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4196-890X
  4. Haoran Ni

    Biophysics Graduate Program, University of Maryland, College Park, College Park, United States
    Contribution
    Software, Investigation, Methodology, Writing – review and editing
    Contributed equally with
    Kaustubh Wagh, Aashli Pathni and Vishavdeep Vashisht
    Competing interests
    No competing interests declared
  5. Vishavdeep Vashisht

    Biophysics Graduate Program, University of Maryland, College Park, College Park, United States
    Contribution
    Formal analysis, Investigation, Visualization, Writing – review and editing
    Contributed equally with
    Kaustubh Wagh, Aashli Pathni and Haoran Ni
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9367-0278
  6. Arpita Upadhyaya

    1. Department of Physics, University of Maryland, College Park, College Park, United States
    2. Biophysics Graduate Program, University of Maryland, College Park, College Park, United States
    3. Institute for Physical Science and Technology, University of Maryland, College Park, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Methodology, Writing – original draft, Project administration, Writing – review and editing
    For correspondence
    arpitau@umd.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1496-919X
  7. Garegin A Papoian

    1. Biophysics Graduate Program, University of Maryland, College Park, College Park, United States
    2. Institute for Physical Science and Technology, University of Maryland, College Park, United States
    3. Department of Chemistry and Biochemistry, University of Maryland, College Park, United States
    Contribution
    Conceptualization, Software, Supervision, Funding acquisition, Investigation, Methodology, Writing – original draft, Project administration, Writing – review and editing
    For correspondence
    gpapoian@umd.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8580-3790

Funding

National Science Foundation (CHE-2102684)

  • Garegin A Papoian

National Science Foundation (PHY-1806903)

  • Garegin A Papoian

National Science Foundation (PHY-1607645)

  • Arpita Upadhyaya

National Institutes of Health (R35 GM145313)

  • Arpita Upadhyaya

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank A Chandrasekaran, C Floyd, and J Komianos for helpful discussions and feedback on the manuscript. This work was supported by National Science Foundation grants CHE-2102684 and PHY-1806903. AU acknowledges support from NSF grant PHY 1607645 and NIH grant R35 GM145313. Computational resources were provided by Deepthought2 HPC at University of Maryland.

Senior Editor

  1. Anna Akhmanova, Utrecht University, Netherlands

Reviewing Editor

  1. Alphee Michelot, Institut de Biologie du Développement, France

Reviewer

  1. Anatoly BK

Publication history

  1. Preprint posted: June 6, 2021 (view preprint)
  2. Received: August 12, 2022
  3. Accepted: October 6, 2022
  4. Accepted Manuscript published: October 21, 2022 (version 1)
  5. Version of Record published: November 21, 2022 (version 2)

Copyright

© 2022, Ni et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 503
    Page views
  • 141
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Qin Ni
  2. Kaustubh Wagh
  3. Aashli Pathni
  4. Haoran Ni
  5. Vishavdeep Vashisht
  6. Arpita Upadhyaya
  7. Garegin A Papoian
(2022)
A tug of war between filament treadmilling and myosin induced contractility generates actin rings
eLife 11:e82658.
https://doi.org/10.7554/eLife.82658

Further reading

    1. Computational and Systems Biology
    2. Neuroscience
    Sergio Oscar Verduzco-Flores, Erik De Schutter
    Research Article Updated

    How dynamic interactions between nervous system regions in mammals performs online motor control remains an unsolved problem. In this paper, we show that feedback control is a simple, yet powerful way to understand the neural dynamics of sensorimotor control. We make our case using a minimal model comprising spinal cord, sensory and motor cortex, coupled by long connections that are plastic. It succeeds in learning how to perform reaching movements of a planar arm with 6 muscles in several directions from scratch. The model satisfies biological plausibility constraints, like neural implementation, transmission delays, local synaptic learning and continuous online learning. Using differential Hebbian plasticity the model can go from motor babbling to reaching arbitrary targets in less than 10 min of in silico time. Moreover, independently of the learning mechanism, properly configured feedback control has many emergent properties: neural populations in motor cortex show directional tuning and oscillatory dynamics, the spinal cord creates convergent force fields that add linearly, and movements are ataxic (as in a motor system without a cerebellum).

    1. Computational and Systems Biology
    2. Immunology and Inflammation
    Mingyao Pan, Bo Li
    Short Report Updated

    T cells are potent at eliminating pathogens and playing a crucial role in the adaptive immune response. T cell receptor (TCR) convergence describes T cells that share identical TCRs with the same amino acid sequences but have different DNA sequences due to codon degeneracy. We conducted a systematic investigation of TCR convergence using single-cell immune profiling and bulk TCRβ-sequence (TCR-seq) data obtained from both mouse and human samples and uncovered a strong link between antigen-specificity and convergence. This association was stronger than T cell expansion, a putative indicator of antigen-specific T cells. By using flow-sorted tetramer+ single T cell data, we discovered that convergent T cells were enriched for a neoantigen-specific CD8+ effector phenotype in the tumor microenvironment. Moreover, TCR convergence demonstrated better prediction accuracy for immunotherapy response than the existing TCR repertoire indexes. In conclusion, convergent T cells are likely to be antigen-specific and might be a novel prognostic biomarker for anti-cancer immunotherapy.