Gating mechanisms during actin filament elongation by formins

Formins play an important role in the polymerization of unbranched actin filaments, and particular formins slow elongation by 5–95%. We studied the interactions between actin and the FH2 domains of formins Cdc12, Bni1 and mDia1 to understand the factors underlying their different rates of polymerization. All-atom molecular dynamics simulations revealed two factors that influence actin filament elongation and correlate with the rates of elongation. First, FH2 domains can sterically block the addition of new actin subunits. Second, FH2 domains flatten the helical twist of the terminal actin subunits, making the end less favorable for subunit addition. Coarse-grained simulations over longer time scales support these conclusions. The simulations show that filaments spend time in states that either allow or block elongation. The rate of elongation is a time-average of the degree to which the formin compromises subunit addition rather than the formin-actin complex literally being in ‘open’ or ‘closed’ states.


INTRODUCTION
Actin is one of the most abundant proteins in eukaryotic cells and is important for numerous functions regulated by interactions with many other proteins (1)(2)(3). The dynamic actin network in the cell cortex enables cells to maintain specific shapes in addition to supporting the plasma membrane.
Actin filaments in lamellipodia drive cellular movements and contractile rings of actin filaments are responsible for cytokinesis (4). Transitions of actin between monomeric and filamentous states is a key feature of these dynamic systems (1).
Formins regulate actin assembly by nucleating and directing the elongation of unbranched actin filaments (5). Formin malfunctions are associated with cancer (6-9) and immune disorders (10), so characterizing formins is crucial for understanding these important diseases.
Formins consist of multiple domains, including a conserved formin homology 2 (FH2) domain that self-associates in a head-to-tail manner to form a homodimer that stabilizes actin filament nuclei (Figure 1). FH2 dimers stay processively associated with the filament's barbed end by "stepping" onto actin subunits as they incorporate into the growing filament. FH1 domains are located directly N-terminal to the FH2 domain and consist of flexible regions connecting multiple polyproline tracks that each bind a profilin-actin complex. Diffusive motions of the FH1 domains allow actin to transfer rapidly onto the barbed end.
Actin filament barbed ends associated with a formin FH2 domain elongate slower than free barbed ends (11,12). This effect is considered to arise from "gating," a rapid equilibrium between an "open state" (when an actin subunit can bind the barbed end) and a "closed state" (when the barbed end is blocked) (11)(12)(13). The fraction of time that a barbed end is in the "open" state is defined as the gating factor, which ranges from ~0.95 for mammalian formin mDia1 to ~0.5-0.7 for budding yeast formin Bni1 and to ~0.05 for fission yeast formin Cdc12 (12,14,15). Thus, barbed ends with mDia1 are mostly in the "open" state that does not slow elongation, while barbed ends with Cdc12 are mostly closed, strongly inhibiting elongation. Rapid transfer of actin from FH1 domains onto open barbed ends allows filaments to elongate rapidly, in spite of gating (16)(17)(18)(19).
Both steric interference and distortion of the barbed end may explain gating (20)(21)(22). A second hypothesis is that formins influence the twist angle of the terminal subunits at the barbed end of the filament (18,20). In the open state, the barbed end subunits are proposed to adopt the 167° twist typical of the middle of the filament, which creates a favorable binding site to add a subunit. In the "strained" closed state, the conformation of barbed end subunits is proposed to shift towards a 180° twist as observed in actin-FH2 co-crystals (21), a twist angle unfavorable for subunit addition.
Three models are proposed for the transition from the closed to the open state. In the "stairstepping" model one FH2 dissociates entirely or in part from the barbed end, relieving either strain or steric blocking before a new subunit binds. The "stepping second" model proposes a rapid equilibrium between closed states and open states; when an end is open, a new subunit can bind, followed by stepping of the trailing FH2 domain onto the new subunit. A third model is that the pair of FH2 domains moves in a screw-like fashion along the short-pitch helix of the elongating actin filament (23).
Understanding the mechanism of gating is important, because gating influences subunit addition from both solution and from FH1 domains (12,13,18,(24)(25)(26). Given the large body of experimental work without mechanistic tests, we sought to identify the gating mechanism of formins by examining three formins with different gating factors, mDia1, Bni1 and Cdc12. We used both allatom (AA), coarse-grained (CG) molecular dynamics (MD) and enhanced free energy sampling simulation techniques to examine the interactions between the FH2 domains of these formins with an actin filament. We found that both steric blocking and distortion of the barbed end can contribute to gating with implications for both the "stair-stepping" and "stepping second" models of processive elongation.

Homology models of FH2 domains associated with actin filament barbed ends
The first step in our comparison of the three formins was to build homology models of Cdc12 and mDia1 FH2 domains bound to actin, for which no structure yet exists. We based the new models on the only detailed model of FH2 domains bound to an actin filament barbed end (27). That model of the Bni1 FH2 domain on the barbed end a filament of seven actin subunits was based on a cocrystal structure of Bni1 FH2 and actin with a twist angle of 180° (21) and an X-ray fiber diffraction model of the actin filament with a twist angle of 167° (28). Baker et al. used the intermolecular contacts in the crystal structure to place the FH2 domains on the 167° barbed end and refined the model with 160 ns of AA MD simulations, during which conformational changes closed the contacts between the FH2 domains and actin and flattened the helical twist at the barbed end of the filament (27). A 3.4 Å crystal structure of a dimer of FMNL3 FH2 domains bound to actin has contacts between the FMNL3 FH2 domain and actin similar to those of the Bni1-FH2 domains (29).
However, this structure has a two-fold axis of symmetry between physically separated actin subunits, so it is less informative regarding the structure of FH2 domains bound to a helical actin filament than the Bni1-FH2-actin structure, upon which Baker et al. (27) and we based our models for MD simulations.
We created homology models of Cdc12 and mDia1 FH2 domains based on the crystal structure of the Bni1 FH2 and used Baker's model of the Bni1 FH2-actin filament (27) to align these FH2 models on the end of a filament consisting of seven subunits (see the Methods section for details). The FH2-actin complexes were solvated, ionized and then refined by extensive AA MD simulations (27) described in detail in the Methods section. These models with an FH2 dimer on subunits A2 and A3 of a seven-subunit filament correspond to the conformation immediately after the addition of a new actin subunit A1. We used 200 ns AA simulations to assess the quality and stability of the homology models, to document the contacts between FH2 domains and actin, and to determine differences between three FH2 domains interacting with actin. After 200 ns of simulation, we created structures corresponding to the step before the addition of subunit A1 by removing subunit A1 from the barbed ends of the models of the three seven-mer filaments. For computational efficiency we also removed actin subunit A7 at the pointed end, leaving five actin subunits in the filament. We extended the AA simulations of seven-mer and five-mer filaments to assess steric interference between the FH2 dimers and actin subunits binding to the barbed end of the filament and to determine how the FH2 domains influence the geometry of the terminal actin subunits.
Coarse-grained (CG) simulations of the FH2-actin models allowed us to study deviations from their initial configurations at time scales beyond the current range of the AA simulations.

Evaluation of the models of FH2 dimers on actin filament seven-mers after all atom MD simulations
At the end of the 200 ns simulations the total energies of the three models of formin dimers interacting with actin filaments (calculated by ProSA (30)) fell within the range of protein structures in the PDB determined by X-ray crystallography and NMR (Figure 2A). The z-scores indicate that the Bni1 model (z-score = -8.9) is the most accurate (as it is based on a crystal structure), and Cdc12 (z-score = -7.69) is more accurate than mDia1 (z-score = -6.45), because the sequence of the Bni1 template model is more similar to Cdc12 than mDia1. The z-score is calculated by taking both structural features and the sequences of FH2 domains into account. Forcing the FH2 primary sequence to fold into a structure other than an FH2 domain will give a very low z-score.
The models of the three formin/actin systems obtained from the AA simulations are very similar ( Figure 2E). All three models have the FHL (leading FH2 domain) and FHT (trailing FH2 domain) positioned head-to-tail to form a donut-like shape around the barbed end of an actin filament with seven subunits (21). Figure 2B shows the FH2 domain of Cdc12 interacting with the barbed end of an actin filament with the FHL domain engaged with actin A2 and the FHT domain engaged with actin A3, prior to FHT stepping onto the newly added actin subunit A1 at the barbedend. Each FH2 subunit consists of helical knob and post regions connected by a three-helix bundle (coiled-coil). The N-terminal lasso region of each subunit encircles the post of the other subunit and is connected to the knob of its own subunit by a flexible linker ( Figure 2C). The coiled-coil and post regions of the FH2 domains of the three formins align quite well ( Figure 2E), while other regions differ locally ( Figure 2F). The lasso regions of the FH2 domains have similar circular conformations, but the mDia1 FH2 lasso lacks a helix found in Cdc12 and Bni1. The helical parts of the linker regions align quite well with each other, but the unstructured parts differ in length and conformation.
The knob regions align the least well. The mDia1 knob has fewer helices than the Cdc12 and Bni1 knobs. Also, the locations of the helices and loops in the mDia1 knobs differ much more than those in the knobs of other two formins.

Do FH2 domains interfere sterically with the addition of an actin subunit at the barbed end?
We used AA simulations to test the steric blocking hypothesis for gating. The Baker et al.
simulations (27) of Bni1 brought the FHL and FHT domains into much closer contact with the actin subunits at the barbed end including small overlaps of the FHL domain with incoming actin subunit +A1 placed on the barbed end of the filament with the Oda filament (28) geometry. We measured the volume fraction of this overlap by determining which C-alpha atoms of FHL and actin subunit +A1 were separated by < 4 Å (diameter of C-alpha atom). The models of the other two formins were based on the Bni1 model after 160 ns of simulation (27), so their FH2 domains overlapped the +A1 site to the same extent as Bni1 at the beginning of the simulations ( Figures 3A, B, C). During the remainder of the 500 ns AA simulations, steric interference with subunit +A1 increased for Cdc12 and decreased for mDia1 with Bni1 in between ( Figure 3A).
During two rounds of AA simulations of five-mer filaments lacking subunit A1, the steric interference of FHT with subunit A1 increased to high levels for Cdc12 and fluctuated between none and low levels for Bni1 and mDia1 ( Figures 3D, E). The fractions of the time when FHT occupied > 1% of the volume fraction of A1 in the two replicates (350 ns for the first replicate and 200 ns for the second replicate) were, respectively, 64 and 21% for Cdc12, 19 and 18% for Bni1 and 4 and 2% for mDia1. Thus, both FHL and FHT of Cdc12 created more steric interference for subunit addition than Bni1 or mDia1.

How do the three formin FH2 domains influence the helical twist of the barbed end?
We tested the hypothesis that gating is caused by FH2 domains flattening the helical twist of the

How do interactions of FH2 domains with barbed ends explain the range of gating factors?
We used the results of the simulations to examine in detail how each formin interacts with actin subunits at the barbed end of a filament. We considered buried surface area, contacts and salt bridges.

Contacts:
We measured the contacts between all residues of the three FH2 domains and During AA simulations of five-mer filaments, flattening of the barbed end by Cdc12 FH2 was strongly correlated with the number of contacts between actin subunits A2 and A3 and the post regions of both FHT and FHL ( Figure S4 and Table S3). The sharp increase in the number of contacts for twist angles > 173° is also seen as an abrupt change in the time course data in Figure 5D (blue line is contacts between FHL post and A2). This correlation does not establish causality, but the additional contacts may provide the free energy change for the unfavorable change in the conformation of the filament. On the other hand, the contacts between the Cdc12 FH2 lasso or knob regions of Cdc12 and actin subunits A2 and A3 did not change as the filament flattened ( Figure S3).
During these simulations, neither Bni1 nor mDia1 caused a large change in the twist angles or the numbers of contacts of the knob, lasso or post regions with actin ( Figures S3, S4 and Table S3).
Other than this behavior of Cdc12, extensive analysis of the contacts and interaction energies between the FH2 domains and seven-mer filaments ( Figures 5, 7, S3 and S4 and Tables S3 and S4) did not show a simple, consistent correlation of the numbers of contacts or interaction energies with gating factors. Several measurements were not correlated with gating factors: the total interaction energy of Cdc12 was the largest (-713 kcal/mole), but interactions were stronger with mDia1 (-396 kcal/mole) than Bni1 (-204 kcal/mole); the lasso and post regions of mDia1 made fewer contacts with actin than Cdc12 and Bni1 ( Figure S2), while the linkers of mDia1 made more contacts than Cdc12 and Bni1; and the knob region of Bni1 FHT had fewer contacts with actin subunit A3 than Cdc12 and mDia1 FHT but more contacts with actin subunit A1.

Salt bridges:
The AA simulations of both five-mer and seven-mer filaments revealed that the number of salt-bridges between the FH2 knobs and the barbed end groove of actin correlate with gating (Table 1). At two time points in the simulations of seven-mer filaments and at the end of the simulation of five-mer filaments Cdc12 had more salt bridges occupied large fractions of the time than Bni1 or mDia1. Note that these salt bridges were dynamic. For example, the Cdc12 FH2 knob helices formed two additional salt bridges and the residues making some salt bridges rearranged between the middles and ends of the 500 ns AA MD simulations. The salt bridges varied more during simulations of the five-mer filaments (Table 1), but Cdc12 again formed more (8) salt bridges with the actin barbed end groove than Bni1 (2) or mDia1 (1).
During AA MD simulations of the seven-mer actin filaments the lasso/linker regions of the mDia1 FH2 domains formed more salt bridges with actins A1, A2 and A3, but both Cdc12 and Bni1 formed more salt bridges occupied more than 70% of the time ( Table 2). Only aspartic acid (D363) in the A2 subunit of the actin filament formed a salt bridge with the lasso/linker regions of all three formins ( Table 2). It was occupied 100% of the time in Cdc12, 85% in Bni1 and only 23% in mDia1.  Figure 6A. The FHT domains must rearrange to escape from these steric clashes. The energy landscapes in Figure 6A showed that the FHT domain of Cdc12 was more confined by high energy barriers (yellow and red regions) than Bni1 or mDia1. These high energy barriers for conformational rearrangements restricted the FHT domain of Cdc12 to a smaller area of the conformational space (including those with steric clashes with A1) than the FHT domains of Bni1 and mDia1. The lower conformational mobility of the Cdc12 FH2 domain may be related to large numbers of salt bridges (Table 1) and correlates with a larger distortion of the barbed end ( Figures 6A, C). Since the strength of the interactions between FH2 domains and actin play a role in actin filament nucleation (27), Cdc12 may be the best nucleator of these three formins.

Coarse-grained simulations over longer time scales
We used simulations of coarse-grained (CG) models of FH2-actin structures to study deviations from their initial configurations at large spatiotemporal scales beyond the range of AA MD simulations. CG MD simulations can reveal essential intra-and inter-molecular interactions by approximating other non-essential interactions. CG simulations also allowed us to test the importance of electrostatic interactions at the FH2-bound barbed ends by varying the dielectric constant at the protein surfaces.

How is gating related to the continuum of twist and steric interference states?
Our simulations demonstrate that stochastic thermal motions allow formin-actin filament complexes to sample a range of conformations on a nanosecond time scale. Both the helical twist of the terminal subunits and the degree of steric interference vary continuously in time and space rather than switching between discrete states like ion channels, where the gate is either fully open or closed.
Consequently, the probability that a diffusing actin subunit will bind to the barbed end likely declines as function of the steric blocking and deviation the twist of the terminal subunits from 167°. Therefore, the gating factor is a time-average of the degree to which the formin compromises subunit addition rather than literally being in 'open' or 'closed' states. Normally the probability that a collision between a diffusing monomer and the barbed end of a filament results in binding is high (~2%) (35). Gating should not change the rate constant for collisions with the end of a filament associated with an FH2 domain, but the probability of binding will be less than 2% depending on the particular formin.
Our simulations confirmed the original idea that FH2 domains can sterically block a barbed end (22) but reveal that thermal motions rather than dissociation from actin (21) can move an FH2 domain out of a blocking site to a conformation that allows addition of a new actin subunit.
The conformational mobilities of FH2 domains in our metadynamics simulations might be relevant to the hypothesis that an FH2 domain partially dissociated from a barbed end binds an incoming actin subunit and facilitates its association with the end of the filament (21). If so, the Cdc12 FH2 might encounter a higher energy barrier for the rearrangement than Bni1 or mDia1 and therefore have a lower probability of incorporating the actin subunit to the barbed end.

How do the equilibrium conformations of formins on barbed ends fluctuate in time?
The probability that a system visits any state depends on the difference in free energy between the equilibrium state and other higher energy states, which are visited less often. Our results show that on a nanosecond time scale each formin favors a different equilibrium state for the barbed end; Cdc12 favors larger helical twist angles with more steric blocking than Bni1 and mDia1. The MB-MetaD simulations ( Figure 6A) illustrate that FH2 domains must overcome energy barriers to deviate from the conformations sampled during AA simulations.
On the millisecond time scale relevant to actin filament elongation, barbed ends will visit a wide range of conformations from open 167° twists without steric blocking to closed 180° twists with steric blocking. However, the distributions will differ such that barbed ends with Cdc12 will be in closed conformations most of the time, while ends with mDia1 are in open conformations most of the time. Nevertheless, we expect that filaments with any of the formin FH2 domains will reach extreme angles a fraction of the time at longer time scales.

Why does Cdc12 slow barbed end elongation more than the other formins?
We compared three formins with different gating factors to identify physical factors that might contribute to their range of gating factors: ~0.95 for mDia1, 0.5-0.7 for Bni1 and ~0.05 for Cdc12.
Both the degree of steric interference and barbed end flattening by the three formins are proportional to their ability to inhibit elongation, with Cdc12 having the strongest and mDia1 the weakest effects, but why is this true?
MB-MetaD simulations show that energy barriers confine Cdc12 FH2 domains to a narrower range of steric blocking and twist states that interfere with elongation than Bni1 or mDia1 FH2 domains. Differences in buried surface area, numbers of contacts with actin or total nonbonded interaction energies might contribute to these barriers, but these parameters were not correlated with the gating factors of the three formins except for the association of barbed end flattening with numbers of contacts between the Cdc12 post with actin. On the other hand, CG simulations revealed substantial electrostatic effects that may be related to the numbers, locations and stabilities of salt bridges between FH2 domains and actin identified by the AA simulations. However, the relationship between salt bridges and gating is complex and incompletely understood.
The total numbers of stable salt bridges in three formin-actin complexes (Tables 1 and 2) are correlated with the equilibrium distributions of steric interference and flattening of barbed ends, but the locations of these salt bridges appear to matter. For example, Cdc12 with a large gating factor has more stable salt-bridges between the knob helices and FHL linkers and actin A2 than mDia1 with a small gating factor. However, the opposite is true for salt-bridges between the FHT linkers and actin A1 where mDia1 has more than Cdc12. Tight binding of FHL to A2 might limit the rearragements associated with FHL shifting from closed states to the open conformations without steric interference and with helical twists favorable for subunit addition.  (Table 2). However, our studies show that gating involves complicated, global conformational changes that may not be amenable to perturbation by simple mutations. For example, swapping linkers between these three formins changed the lifetime of the chimeric formins on the ends of growing filaments but did not alter gating (11). Therefore, structural studies and biophysical methods such as fluorescence resonance energy transfer may be more likely than mutations to provide more information about gating.

System Setup for All-atom Simulations
Crystal structures of mDia1 and Cdc12 FH2 domains associated with an actin filament are not available, so we generated their homology models by using our structure of Bni1/actin 7-mer obtained by all atom (AA) MD simulations (27) starting with the crystal structure of Otomo et al. (21). For homology modeling, amino acid sequences of mDia1 and Cdc12 from Uniprot (36), and pdb file of Bni1 (as a template) were supplied to SWISS-MODEL web server (37,38). MODELLER (39,40) was used to obtain the structures of missing residues based on ab initio methods. Both resulting Cdc12 FH2 subunits (FHT and FHL) have 402 residues consisting of residues 984 to 1385 and both resulting mDia1 FH2 subunits have 397 residues ranging from residue 754 to 1150. After running minimization, heating and equilibrium dynamics on the homology models of the Cdc12 and mDia1 FH2 domains, we formed the actin-formin complex by incorporating an actin filament composed of 7 subunits into the formin dimers. This was achieved by alignment of the formins onto the Bni1-actin complex by using MultiSeq and Stride tools of VMD (41). The definitions of features of FH2 domains were given in Supplemental Table S1. The accuracy of the homology models was assessed via ProSA (Protein Structure Analysis Tool) (Figure 2A) to calculate z-scores by using knowledge-based potentials of mean force (30). The z-score is obtained by calculating the deviation in the total energy of the target protein structure from an energy distribution of random protein structures on the Protein Data Bank (PDB). To confirm the quality of the models, we generated various structures of FH2 domains interacting with the barbed-end of an actin filament using template-based homology modeling (Bni1 based on Cdc12, Cdc12 based on mDia1, and mDia1 based on Cdc12). All atom simulations of these structures were run for about 95 ns to refine these models, which we compared with the crystal structure of Bni1 by using the template modeling score (TM-score) (42). All of these structures were found to have a TM-score between 0.5 and 1.0 (Table   S2), which indicates that they have about the same fold as the crystal structure of Bni1 (42,43). As the homology models are compared with the crystal structure of Bni1, the homology model of Bni1 FH2 has the highest TM-score (0.79) consistent with the expected quality of the homology models.
Thus, template-based homology modeling provides protein models with a high degree of similarity in their topological organization of the secondary structures to the original crystal structure.
The initial structures of mDia1 and Cdc12 FH2 domains bound to the barbed-end of the actin filament ( Figure 2B) were generated by aligning their FH2 domains with the final equilibrated structure of Bni1/actin 7-mer after 160 ns of AA MD simulations (27). Then, the FH2-actin complexes of Cdc12 and mDia1 were solvated and ionized using exactly same procedure to obtain the final configuration of the Bni1 FH2-actin complex. The systems were then solvated by leaving enough distance in each direction (at least 15 Å) to prevent the protein from interacting with its periodic image during the simulations. The N-terminus of the actin monomers was acetylated, and all other residues in the system were modeled in their standard states of protonation at a pH value of 7. The systems were neutralized with 0.18 M KCl. The Cdc12/actin and mDia1/actin systems were energy minimized by gradually releasing the restraints on the systems via the use of NAMD (44) with the Charmm27 force field with CMAP corrections (45). In the heating phase, the temperatures of the systems were gradually raised to 310 K during the simulation spanning 200 ps with restraints on the backbone atoms of the protein, ADP, the Mg 2+ ion and coordinating waters. The heating phase was followed by an equilibration phase in which the restraints on the protein backbone, ADP, the Mg 2+ ion and coordinating waters were gradually reduced from 10 kcal/mol/Å 2 to 0.1 kcal/mol/Å 2 over 400 ps. The AA MD simulations of the Cdc12/actin and mDia1/actin systems with unrestrained dynamics were run for 500 ns. The AA simulation of the Bni1 FH2/actin system was extended from 160 ns (27) for 340 ns to reach 500 ns of dynamics.

System Setup for Coarse-grained Simulations
To capture the longer time scale dynamics of FH2 dimers on the end of an actin filament, we created CG models using essential dynamics coarse graining (ED-CG) (33) and heterogeneous elastic network modeling (heteroENM) (34). EDCG is a systematic method that defines CG sites based on the collective protein motions, which are computed by using Principal Component Analysis (PCA) of all-atom trajectories. The CG sites of all regions in formin and actin were automatically determined by EDCG method. Each subunit in the complex consists of 45 CG sites, which correspond to approximately 8 and 9 residues per CG site, for FH2 and actin monomers. This CG level of resolution was chosen to obtain a model, which is both computationally efficient and capable of incorporating spatial structural elements of the complex. The assignment of CG sites was done separately for each subunit (Formin Homology Leading (FHL), Formin Homology Trailing (FHT), and actin subunits A1 to A7, numbered from the barbed end) to capture characteristic fluctuation profiles of individual subunits.
After the CG site assignment stage, intramolecular interactions of formin and actin were determined by using heteroENM. In heteroENM, CG sites within a certain cutoff distance are connected via effective harmonic springs, each with a specific spring constant, all collectively determined iteratively from the AA MD data to match the actual protein fluctuations observed in the AA simulations. The cutoff distances were chosen to obtain the best match between root mean square fluctuations (RMSF) of all-atom and CG simulations (50 Å (FH2) and 30 Å (actin) for Cdc12-actin, 60 Å and 20 Å for Bni1-actin, and 60 Å and 30 Å for mDia1-actin complexes, respectively).
Intermolecular interactions between FH2 and actin were described by 12-6 Lennard Jones (LJ) potentials and screened Coulomb potentials to effectively introduce dispersion and electrostatic forces into the system. According to the Derjaguin-Landau-Verwey-Overbreek theory, these interactions can be assumed to be additive and independent of each other, therefore intermolecular interactions can be written as follow: The dispersion and repulsive forces due to excluded volume were modeled using 12-6 LJ potential (2) and electrostatic forces were introduced by using screened Coulomb potential (3): The repulsive part of the 12-6 LJ potential was used to describe the excluded volume interactions. Epsilon parameters of the LJ potential were chosen to be 10 kcal/mol, and sigma parameters were obtained by calculating the radius of gyration of the CG sites. The cut-offs for the LJ potential were set to the sigma parameters, whereas that of screened Coulomb potential was set to 30 Å. The charge fitting method was used for assigning a charge to individual CG sites in order to match the charge distribution of the all-atom model (27). The Debye length (κ) in the screened Coulomb potential was chosen as 0.8 nm to mimic the physiological salt conditions. There is no consensus on the dielectric constant at the protein-protein interfaces, but the value is low and depends on the system. Implicit solvent MD simulations (27,46) and molecular mechanics Poisson-Boltzmann (MMPB) methods (47)

System Setup for Metabasin Metadynamics Simulations
The FH2 domains interacting with five-mer filaments were used as initial structures for metabasin metadynamics (MBMetaD) simulations (32).

SUPPORTING MATERIAL
Four figures and four tables are available at "this link".