New Research In
Physical Sciences
Social Sciences
Featured Portals
Articles by Topic
Biological Sciences
Featured Portals
Articles by Topic
 Agricultural Sciences
 Anthropology
 Applied Biological Sciences
 Biochemistry
 Biophysics and Computational Biology
 Cell Biology
 Developmental Biology
 Ecology
 Environmental Sciences
 Evolution
 Genetics
 Immunology and Inflammation
 Medical Sciences
 Microbiology
 Neuroscience
 Pharmacology
 Physiology
 Plant Biology
 Population Biology
 Psychological and Cognitive Sciences
 Sustainability Science
 Systems Biology
Molecular noise of capping protein binding induces macroscopic instability in filopodial dynamics

Edited by Peter G. Wolynes, University of California at San Diego, La Jolla, CA, and approved April 30, 2009 (received for review December 15, 2008)
Abstract
Capping proteins are among the most important regulatory proteins involved in controlling complicated stochastic dynamics of filopodia, which are dynamic fingerlike protrusions used by eukaryotic motile cells to probe their environment and help guide cell motility. They attach to the barbed end of a filament and prevent polymerization, leading to effective filament retraction due to retrograde flow. When we simulated filopodial growth in the presence of capping proteins, qualitatively different dynamics emerged, compared with actinonly system. We discovered that molecular noise due to capping protein binding and unbinding leads to macroscopic filopodial length fluctuations, compared with minuscule fluctuations in the actinonly system. Thus, our work shows that molecular noise of signaling proteins may induce micrometerscale growth–retraction cycles in filopodia. When capped, some filaments eventually retract all the way down to the filopodial base and disappear. This process endows filopodium with a finite lifetime. Additionally, the filopodia transiently grow several times longer than in actinonly system, since less actin transport is required because of bundle thinning. We have also developed an accurate meanfield model that provides qualitative explanations of our numerical simulation results. Our results are broadly consistent with experiments, in terms of predicting filopodial growth retraction cycles and the average filopodial lifetimes.
 amplification
 stochastic chemical kinetics
 growth–retraction cycles
 stochastic switch
 mechanochemical sensing
Eukaryotic motile cells project fingerlike protrusions, called filopodia, to probe their environment and help guide cell motility (1). They play important roles in neuronal growth (2), wound healing (3), and cancer metastasis (4). The filopodial structure consists of parallel actin filaments, crosslinked into bundles by actinbinding proteins, all of which is enclosed by the cell's plasma membrane (1, 5). Despite their importance in eukaryotic biology and human health, the physical mechanisms behind filopodial regulation and dynamics are poorly understood, including what drives ubiquitous growth–retraction cycles and eventual filopodial disappearance (1–7).Using stochastic simulations of filopodial dynamics, we have discovered that molecular noise due to binding/unbinding of a capping protein results in macroscopic growth–retraction fluctuations, compared with minuscule fluctuations in the actinonly system. Because of rare fluctuations some filaments eventually retract all the way down to the filopodial base and disappear. In contrast to prior computational models that predicted stable filopodia at steady state, our simulations show that filopodial lifetimes are finite. We have also developed an accurate meanfield model that provides insights into filament disappearance kinetics.
A comprehensive computational model of a filopodium should contain the following features: mechanical interactions including membrane dynamics, protrusion force, and retrograde flow; chemical interactions, including actin polymerization and depolymerization; and biological signaling interactions that control the filopodia dynamics turnover. The first meanfield model for filopodial growth addressed the interplay between filament growth and diffusional actin transport (8). A subsequent work highlighted the importance of the interactions between the membrane and filament barbed ends (9). Our own previous study treated both polymerization and diffusion in a fully stochastic fashion (10). In that model, the filopodia grow to some steadystate length, and subsequently exhibit only slight fluctuations (10). Thus, no essential dynamics occurs after the steady state is reached, a finding similar to those of other prior filopodial simulations (8, 9), implying an essentially infinite filopodial lifetime and no turnover.
Although it is not well known whether the turnover is driven externally or internally, it is plausible that internal biochemical reaction network dynamics is a significant contributor. For instance, capping proteins bind to the barbed ends of actin filaments preventing polymerization (11). Their efforts are countered by formins (12), anticapping processive motors that attach to the barbed ends and may effectively increase polymerization rate up to fivefold (12). In this work, we investigated the influence of two regulating proteins on the filopodial turnover process. The mechanochemical model that we use here to describe the filopodial dynamics is fully stochastic, using the Gillespie algorithm to calculate simulation steps (Fig. 1). It consists of the following processes: (i) the diffusion of proteins from the cytosol at the filopodial base to the tip; (ii) the force applied by the membrane on individual filaments; (iii) the actin filament polymerization and depolymerization at the barbed end; (iv) the depolymerization at the pointed end and the induced retrograde flow v_{retr} of the filopodium as a whole; (v) capping of the filaments that stops polymerization (11); (vi) binding of formin to the barbed end that increases effective polymerization rate fivefold (12). The filopodium is split into compartments with diffusion realized as stochastic hops between them. Membrane force is taken into account by effectively decreasing the polymerization rates (see Materials and Methods and ref. 10 for more details). The parameters, such as reaction rates and concentrations, are given in Table 1.
The addition of capping proteins and formins results in very different, more complex growth dynamics (Fig. 2), in contrast to the actinonly model in ref. 10, where the observed length of the filopodium was stationary. In particular, filopodial length fluctuations become macroscopic, increasing from < 100 nm in a model without these proteins to a few micrometers, on the order of the length of filopodia. In addition, a clearly identifiable retraction phase appears, with a lifetime of ∼100 s. The mechanism is this: when an individual filament is capped, the retrograde flow makes it retract—the filament will eventually disappear if uncapping does not occur quickly enough. If the filament number becomes less than a minimum needed to overcome the membrane force, the filopodium collapses.
These largeamplitude oscillations of filopodial length observed in our simulations are the consequence of the amplification of molecular noise of capping protein. This amplification is possible because of the timescale separation between fast polymerization and retrograde flow processes and slow offrates of formin and capping protein. Because the regulatory proteins are present in very low concentrations (<100 nM), their noise is highly discrete (13–21), randomly driving backandforth transitions from fast growth to fast retraction for each filament. Such a highly fluctuating behavior should be advantageous from the point of view of efficiency of a filopodium as an environmental sensor, in analogy to nearcritical systems, where large fluctuations are indicative of large corresponding susceptibilities. Indeed, we found that the filopodial length is significantly more sensitive to the change in membrane force in a system with capping proteins and formins than in a system with just actin.
To compare our predictions with prior and future experiments, we computed filopodial lifetime distribution from 2,048 Gillespie trajectories (Fig. 2 Lower). Experimentally reported lifetimes are on the order of several minutes, consistent with our results (6, 22–26). We found that computed filopodial lifetimes strongly depend on the amplitude of the individual filament length fluctuations (Fig. 3).
To gain further insight into the observed noise amplification phenomenon, we investigated how the capping protein and formin concentrations influence the magnitude of the filopodial macroscopic fluctuations. To measure them, we created a stationary state by allowing the tips of disappearing filaments to stay at the filopodial base and eventually uncap. As one might have anticipated, the fluctuation amplitude grows with increasing capping protein concentration. However, formin quenches these oscillations such that the fluctuation amplitude depends mainly on the ratio of the capping protein to formin concentrations (Fig. 3 Upper). Overall, filopodial lifetime strongly depends on the amplitude of the individual filament length fluctuations (Fig. 3 Lower).
To shed light into the kinetics of filopodial retraction, we address the question of how such long timescale processes (hundreds of seconds) emerge from the much faster constituent kinetic rates (filament uncapping rate, k_{C}^{−} = (25 s)^{−1} is the slowest rate in our model). We elaborate below on a simple meanfield model to estimate the rate for filament disappearance. We consider a filament in the bundle of N anticapped filaments of stationary length, L, computed in our prior work (10) (see Table 1 for the definition of variables):
We keep track of a single filament as various binding/unbinding events occur. After formin unbinds (with offrate, k_{F}^{−}), the filament may either become capped and start retracting or rebind formin, with the latter being more likely. The ratio of probabilities for capping and anticapping is proportional to the ratio of the onrates for capping protein and formin, C_{C}k_{C}^{+} and C_{F}k_{F}^{+}, where C_{C} and C_{F} are capping protein and formin concentrations, respectively. The regulatory proteins are not consumed during polymerization (unlike actin), thus, we can assume bulk concentrations at the tip. Overall, the average capping time for a filament may be estimated as
The computational model for filopodial dynamics is essentially a multidimensional lattice on which a stochastic propagator enacts a random walk. If the filament is capped, it may either fully retract and disappear or become uncapped and grow back. Thus, many possible trajectories on the lattice must be considered. Each distinct trajectory is characterized by a certain probability of occurrence and the overall time for filament disappearance. The trajectories may be grouped according to how may times the filament has been capped and uncapped before it disappears. In the following analytical estimation, we calculate average time of disappearance in each group of paths described above by using a meanfield approximation, instead of carrying out a full path integral calculation (Fig. 4). Such a group of paths may be thought of as a particular event scenario. We then average the rates from each group with statistical weights of the scenarios; the weights are computed exactly.
In a bundle of N filaments the average time to wait for a capping event is
Thus, the average time for filament disappearance is
Another interesting result that we observed was that the longest filopodia grow to ≈4.5 times the size of those simulated without formins and capping proteins [in our prior work (10)]. The explanation for this follows: a key factor limiting the filopodial length is insufficient actin transport (8, 10). A meanfield estimation for the stationary length (Eq. 1) follows from equating the actin flux from the established gradient and actin consumption by N polymerizing barbed ends (10). Because the latter is proportional to N, a smaller number of filaments require less actin for growth; therefore, a thinner filopodium can grow longer before the diffusive Gactin transport again becomes a limiting factor. As filaments cap and disappear, N decreases, hence, the increase in the length. When too few filaments remain to oppose the force of the membrane load, the bundle withdraws, resulting in the eventual filopodial disappearance. One can speculate another interesting consequence of bundle thinning: it may induce mechanical buckling instability, due to diminution of the mechanical rigidity of the Factin bundle. In light of the recent suggestion that filopodia with >10 filaments are mechanically stable (27), our current finding of filament bundle thinning may turn out to be an important mechanism for mechanical collapse of filopodia due to buckling.
In summary, we have shown in this work that capping proteins exert a dramatic effect on filopodial dynamics. Introducing them allowed us to create a computational model that predicts a finite lifetime of filopodia. The resulting filopodial length dynamics has a remarkable feature: discrete noise of regulatory proteins that are in very low concentration becomes greatly amplified. Because of timescale separation, this slow discrete noise triggers the fast retrograde flow and polymerization processes resulting in oscillations of filopodial length macroscopic in space and time. This result suggests that experimentally observed filopodial growth–retraction turnover dynamics (5–7) may be partially driven by the internal noise of the filopodial mechanochemical network. Largeamplitude fluctuations in filopodia may be important with respect to their sensory role. In particular, a molecular system having oscillations of this magnitude is expected to be easily perturbed by small external forces, either mechanical or chemical in origin, in analogy with large susceptibilities seen in nearcritical systems with large fluctuations. The noise amplification described in this work, arising from discrete noise in a low copy number of some of the reaction network species, is related to stochastic switching in biochemical signal transduction, for example, when a cell needs to make a binary decision (28). A similar dynamic instability is observed in microtubular growth, although in filopodia the oscillations are switched by the binding noise, and in microtubular catastrophes and rescues it is enzymatic in origin (29). In addition, by promoting filament disappearance, capping proteins thin the filopodium, reducing its mechanical rigidity. We speculate that this may be an important mechanism for creating buckling instabilities. We also investigated in detail the filament retraction kinetics, and developed a meanfield analytical estimation for the rate of disappearance of individual filaments that matches the simulation data surprisingly well. We explained the extremely slow timescale of filament disappearance, compared with bare kinetic rates, by common occurrence of multiple capping, shrinking, and regrowing events, before the filament fully retracts as a result of a rare fluctuation.
Materials and Methods
In our stochastic model of filopodial growth (10), represented in Fig. 1, we assume that the reaction processes are confined to a spatial region having a linear dimension of ζ [socalled Kuramoto length (30)], such that particles diffuse across the region quickly compared with the typical reaction times. This allows us to discretize space into compartments and model protein diffusion along the filopodium as a random walk on a onedimensional lattice, with molecules hopping between these compartments at rates calculated from diffusion coefficients. At relevant concentrations, ζ is on the order of 100 nm at the tip of the filopodium. We chose a somewhat more conservative compartment size of l_{D} = 50 nm (10). At the filopodial base the protein concentrations are kept at their constant bulk values. Biochemical reactions within each compartment are simulated by using the Gillespie algorithm (10, 31). Polymerization rates at the tip are decreased by the membrane force via the Brownian ratchet model (8, 10). The experimentally reported uncapping frequency is about once every 30 min, however, it may be greatly increased through the actions of uncapping proteins, such as PIP2 (11). The variables and parameters for the model are given in Table 1. Further simulation details are elaborated below.
The Stochastic Model of Filopodial Growth.
A mature filopodium is a bundle of a few dozen actin filaments enclosed by the cell membrane with a protein complex at the tip (32). The filopodial base is usually located within a threedimensional actin mesh below, which forms the basis for the lamellipodium (32). In addition, bundling proteins crosslink the filaments in the bundle, but this was not included in our current model.
Typical growth and retraction speeds are within 0.1–0.2 μm/s, with filopodial lengths reaching 1–2 μm (33). In special cases, the filopodial length may reach nearly 100 μm (34). Since short filopodia are mostly straight, for simplicity we assume growth of straight filaments. Longer filopodia might tilt and bend. The possibilities of bending of the filaments and buckling of a filopodium should be addressed in future work.
Depending on the number of actin filaments, the filopodial diameter typically varies between d = 100 nm and 300 nm. In our simulations we used d = 150 nm, a number that was derived from minimizing the membrane free energy (9) and is reasonable if the number of filaments is not too large.
Our physicochemical model of filopodial growth consists of the following processes: (i) diffusion of proteins along the filopodial tube, providing passive transport from cell body to the filopodial tip; (ii) mechanical interactions between the membrane and individual filaments; (iii) actin filament polymerization and depolymerization processes at the barbed end; (iv) depolymerization at the pointed end and the induced retrograde flow v_{retr} of the bundle as a whole; (v) capping of the filaments that stops polymerization (11); and (vi) binding of formin to the barbed end which increases the effective polymerization rate fivefold (12).
Polymerization, Depolymerization.
Actin filaments (Factin) are asymmetric, with one end called “the barbed end” and the other end called “the pointed end.” Because of ATP binding and subsequent hydrolysis, chemical affinities for monomers (Gactin) are different at the two ends, such that the polymerization rate at the barbed end is much higher. This leads to the motion of the filament as a whole in the direction of the barbed end (whereas individual monomer units migrate from the barbed end to the pointed end). This process is called “treadmilling” and is the biochemical basis for the cytoskeletal dynamics (35, 36), including filopodial protrusion. An actin filament consists of two protofilaments, wound up in a righthanded helix, with a pitch distance of 37 nm. Diameter of one globular actin monomer is 5.4 nm. In our simulations we increase the filament length by δ = 2.7 nm on one polymerization event, since there are two protofilaments, and two monomers are needed to increase length by 5.4 nm.
The resulting doublehelical filament is mechanically robust, with a persistent length, L_{p} ∼ 10 μm. The buckling length, or the critical length at which one filament would buckle if subject to a force F, is
where k_{B} is the Boltzmann constant and k_{B}T = 4.1 pN nm at room temperature (8). The equation gives L_{b} ≈100 nm for a force F = 10 pN for a single filament. For weakly crosslinked bundles of N actin filaments, the buckling length of the bundle is
In our model, cytosolic molecules such as Gactin, formin, or capping protein, which are in the same compartment as the filopodial tip, can attach to one of the filament ends with a probability, given by the rates k_{A}^{+}, k_{FA}^{+}, or k_{C}^{+}. These monomers at filament ends can also stochastically dissociate with the corresponding rates k_{A}^{−}, k_{FA}^{−}, or k_{C}^{−}. In prior experiments, it was shown that the actin polymerization rate rather strongly depends on the membrane force (37). Theoretical explanations were also provided (8, 38, 39). For instance, the Brownian ratchet model considers membrane fluctuations at the tip of the filopodium (38, 40). That the membrane fluctuations are sufficient to allow a Gactin monomer to fit sterically atop the filament allows a polymerization with the rate k_{A}^{+}. Thus, the effective polymerization rate k_{A,n} on the n th filament equals the “bare” rate k_{A,0} times the probability of the gap opening at the tip of the n th filament. A convenient relation between the loading force f_{n} and the polymerization rate was derived earlier (40),
According to Eq. 6 the onrates for Gactin monomer addition (k_{A}^{+} for free barbed end; k_{FA}^{+} for the barbed end anticapped by formin) and capping protein (k_{C}^{+}) are modified after each timestep. The onrate for formin, k_{F}^{+}, was assumed to be independent of the membrane fluctuations due to diminished steric constraints (12). All of the offrates do not depend on the membrane position.
For these effective onrates, we needed to estimate the membrane force on each individual filament, which is discussed next.
Membrane Force.
According to prior studies fluctuations of a membrane sheet below micrometer scale relax on the microsecond to millisecond timescale (41–44). Chemical reactions occur on a much slower timescale (10^{−2}–10^{1} s), thus, due to this timescale separation, the membrane fluctuations may be assumed to be equilibrated at the time of each given chemical event. Each filament experiences an individual membrane force, f_{n}, that depends on the proximity of the filament tip to the average location of the membrane. Thus, the total membrane force f is distributed among the individual filaments, f = ∑_{n}f_{n}.
To calculate f_{n}, we assume that on average the force on a filament is proportional to the membranefilament contact dwelling probability, which is the probability that the membrane touches that filament. That, in turn, depends on the amplitude of the membrane fluctuation near the filament and the filament length. Longer filaments are more likely to be in contact with the membrane, and feel a stronger membrane force. If the membrane fluctuations are assumed to be described by a Gaussian distribution around the membrane average position, the dwelling probability p_{n} for the n th filament to be in contact with the membrane is proportional to the probability that the membrane height is found below the filament end where σ_{d} is the average membrane fluctuation amplitude (discussed next). Once p_{n} is obtained, the force f_{n} on each filament may be computed, where p = ∑_{n}p_{n} is a normalization factor. Then we use Eq. 6 to modify the polymerization rates.
Retrograde Flow.
The filopodial dynamics are strongly controlled by the polymerization at its tip (45, 46). As was mentioned earlier, the “treadmilling” contributes to a steady backward motion of the whole actin filament bundle, called the retrograde flow. In some cells, it is believed that specific myosin motors participate in creating the retrograde flow (47). All of this is subject to the regulation by the signaling proteins, but in our current computations, we neglect these subtleties of the retrograde flow process, and assume a constant average retrograde flow speed, v_{retr}. The retrograde flow speed v_{retr} is ≈20–200 nm/s (46, 48–51) and we take v_{retr} = 70 nm/s as the default value in our simulations.
Numerical Scheme.
Our computations are based on the Gillespie algorithm (31). In particular, at each simulation step two independent random numbers are chosen, where the first one determines the time of the next event t + δt (t is the time of last event) based on the reaction rates present in the system, and the second one determines which event occurs, based on the magnitude of individual event rates. The event is chosen among the following possibilities: (i) Gactin monomer, capping protein or formin hopping between filopodial compartments, (ii) individual filament polymerization or capping events, (iii) depolymerization or uncapping events, and (iv) binding or unbinding of formin. Then, we update current time, adding δt, the numbers of proteins in compartments, filament length, and which protein is on its end according to the event that happened. We then incorporate retrograde flow and update the length of filaments again. Finally, the membrane load is partitioned among the filaments, as described above, which results in recomputing individual filament polymerization rates, according to the Brownian ratchet model.
Acknowledgments
This work was supported by National Science Foundation Grant CHE715225. All calculations were carried out using the UNC Topsail supercomputer.
Footnotes
 ^{1}To whom correspondence should be addressed. Email: gpapoian{at}unc.edu

Author contributions: P.I.Z. and G.A.P. designed research, performed research, analyzed data, and wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.
References
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 Schafer DA,
 Jennings PB,
 Cooper JA
 ↵
 ↵
 ↵
 Sasai M,
 Wolynes PG
 ↵
 ↵
 Walczak AM,
 Onuchic JN,
 Wolynes PG
 ↵
 ↵
 Thattai M,
 van Oudenaarden A
 ↵
 ↵
 ↵
 ↵
 Gomez TM,
 Robles E,
 Poo M,
 Spitzer NC
 ↵
 ↵
 ↵
 PorteraCailliau C,
 Pan DT,
 Yuste R
 ↵
 VarnumFinney B,
 Reichardt LF
 ↵
 ↵
 Artyomov MN,
 Das J,
 Kardar M,
 Chakraborty AK
 ↵
 ↵
 Kuramoto Y
 ↵
 ↵
 Svitkina TM,
 et al.
 ↵
 ↵
 ↵
 Pantaloni D,
 Clainche CL,
 Carlier MF
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 ↵
 Mallavarapu A,
 Mitchison T
 ↵
 Heid PJ,
 Geiger J,
 Wessels D,
 Voss E,
 Soll DR
 ↵
 Lidke DS,
 Lidke KA,
 Rieger B,
 Jovin TM,
 ArndtJovin DJ
 ↵
 Forscher P,
 Smith SJ
 ↵
 ↵
 Huckaba TM,
 Lipkin T,
 Pon LA
Citation Manager Formats
Article Classifications
 Physical Sciences
 Physics
 Biological Sciences
 Biophysics and Computational Biology
Sign up for Article Alerts
Jump to section
You May Also be Interested in
More Articles of This Classification
Physical Sciences
Physics
Biological Sciences
Related Content
 No related articles found.
Cited by...
 Automated analysis of filopodial length and spatially resolved protein concentration via adaptive shape tracking
 Flightless I interacts with NMMIIA to promote cell extension formation, which enables collagen remodeling
 Capping protein is essential for cell migration in vivo and for filopodial morphology and dynamics
 Polarization of actin cytoskeleton is reduced in dendritic protrusions during early spine development in hippocampal neuron
 MyosinX: a MyTHFERM myosin at the tips of filopodia
 Derivation of Ca2+ signals from puff properties reveals that pathway function is robust against cell variability but sensitive for control