Main

Many processes driven by actin, such as cell division, depend on its ATPase activity1. In its monomeric form (G-actin), actin exhibits very weak ATPase activity (7 × 10−6 s−1) (ref. 4) but polymerization into filaments (F-actin) triggers a conformational rearrangement that allows actin to hydrolyse ATP within seconds (0.3 s−1) (ref. 5). The cleaved inorganic phosphate (Pi) is not released immediately after hydrolysis (release rate 0.006 s−1) (ref. 6), yielding the intermediate ADP-Pi state of F-actin7. After the exit of Pi, ADP-bound F-actin represents the ‘aged’ state of the filament, which can then be depolymerized back to G-actin. In vivo, this cyclic process is tightly regulated by various actin-binding proteins (ABPs), of which a subset is capable of sensing the actin nucleotide state8,9. As a prominent example, ABPs of the ADF/cofilin family efficiently bind and sever ADP-F-actin to promote actin turnover but only bind with weak affinity to ‘young’ actin filaments in the ATP or ADP-Pi state10,11,12.

In addition to ABPs, the divalent cation that associates with the actin-bound nucleotide, Mg2+ or Ca2+, also strongly affects polymerization rates. It is now accepted that Mg2+ is the predominant cation bound to actin in vivo13,14. However, because Ca2+-ATP-bound G-actin exhibits slower polymerization kinetics and a higher critical concentration of polymerization15,16,17, it has been used as standard in actin purifications18, many in vitro studies and most G-actin crystal structures19. What causes the slow polymerization rates of Ca2+-actin remains unknown.

Since 2015, numerous cryo-EM studies have shown the F-actin architecture in all nucleotide states20,21,22 and in complex with a variety of ABPs such as cofilin23,24 and myosin25,26,27. However, previously published F-actin structures were solved at moderate resolutions of about 3–4.5 Å and therefore did not show sufficient details to model solvent molecules and exact positions of amino-acid side-chains. Hence, key mechanistic events in F-actin aging, such as ATP hydrolysis, which strongly depends on water molecules, remain unknown. Here we present 6 cryo-electron microscopy (cryo-EM) structures of rabbit skeletal α-actin filaments at approximately 2.2 Å resolution in 3 functional states, polymerized in the presence of Mg2+ or Ca2+. The structures illuminate the F-actin architecture in unprecedented detail and underpin the critical role of solvent molecules in actin filament assembly and aging.

Structures of F-actin reveal solvents

First, by using an optimized cryo-EM workflow (Extended Data Fig. 1 and Methods), we determined structures of Mg2+-F-actin in three relevant nucleotide states (ATP, ADP-Pi and ADP) at resolutions of 2.17–2.24 Å (Fig. 1a–d, Extended Data Figs. 2, 3a–f and 4a–c and Supplementary Table 1; Methods).

Fig. 1: Cryo-EM reconstructions of F-actin at 2.2 Å resolution.
figure 1

a, Local-resolution filtered, sharpened cryo-EM density map of F-actin in the Mg2+-ADP-BeF3− state. The subunits are labelled on the basis of their location along the filament, ranging from the barbed (A−2) to the pointed (A2) end. The central actin subunit (A0) is blue and the other four subunits are grey. Actin subdomains (SD1–4, also known as Ia, Ib, IIa and IIb) are annotated in the central subunit. Densities corresponding to water molecules are red. b–g, Cryo-EM densities of the nucleotide-binding pocket in F-actin in the Mg2+-ADP-BeF3− (b), Mg2+-ADP-Pi (c), Mg2+-ADP (d), Ca2+-ADP-BeF3− (e), Ca2+-ADP-Pi (f) and Ca2+-ADP (g) states. Mg2+ and Ca2+ are shown as green spheres. Water molecules that directly coordinate the nucleotide-associated cation are magenta. For the Ca2+-ADP structure (g), one coordinating water is hidden behind the Ca2+ ion.

Full size image

The unprecedentedly high resolutions of the F-actin reconstructions allowed for the modelling of hundreds of solvent molecules and we accordingly observed clear densities for the nucleotide and the associated Mg2+ ion with its coordinating water molecules (Fig. 1b–d and Supplementary Video 1). The overall conformations of all Mg2+-F-actin structures are highly similar, with a Cα atom root-mean square deviation of <0.6 Å and no changes in helical rise and twist (Supplementary Table 1 and Extended Data Fig. 4g), indicating that the differences are in the details (see below). Although earlier studies predicted extra Mg2+ and Pi binding sites outside of the F-actin nucleotide-binding pocket19,28, we did not find evidence for these secondary ion-binding sites in any of our reconstructions.

We solved F-actin structures using ADP complexed with beryllium fluoride (BeF3−, also referred to as BeFx)29 to mimic the short-lived ATP state of the filament (Fig. 1a,b). In the Mg2+-ADP-BeF3− F-actin structure (2.17 Å), we observed unambiguous density for the modelling of ADP-BeF3− in the nucleotide-binding site (Fig. 1b). Notably, the nucleotide conformation of Mg2+-ADP-BeF3− in F-actin resembled Mg2+-ATP in G-actin (Fig. 2b,c and Extended Data Fig. 5a), confirming the suitability of ADP-BeF3− as ATP mimic.

Fig. 2: Water relocation during the G- to F-actin transition.
figure 2

a, Schematic cartoon representation of actin flattening during the G- to F-actin transition. b–e, Nucleotide conformation and inner-coordination sphere of the divalent cation in Mg2+-ATP-G-actin (Protein Data Bank (PDB) 2V52) (b), Mg2+-ADP-BeF3− F-actin (c), Ca2+-ATP-G-actin (PDB 1QZ5) (d) and Ca2+-ADP-BeF3− F-actin (e). Bond lengths are annotated in angstroms. f,g, Water relocation in Mg2+-actin (f) and Ca2+-actin (g). In f and g, the left panel shows water and amino-acid arrangement in ATP-G-actin. Amino acids are pink for Mg2+-actin and light-brown for Ca2+-actin, whereas the cartoon representation is shown in grey. Arrows depict the movement of amino-acid regions for the transition to F-actin. The middle panel shows overlay of the amino-acid positions in ADP-BeF3− F-actin (blue for Mg2+-actin and cyan for Ca2+-actin) with the solvent molecules in the G-actin structure. The water molecules in the SD3/1 cavity of ATP-G-actin are shown as semitransparent spheres. Arrows indicate the direction of water relocation. Finally, the right panel shows the water and amino-acid arrangement in ADP-BeF3− F-actin. Nuc, nucleophilic.

Full size image

To elucidate the mechanistic basis for the slower polymerization kinetics of Ca2+-actin, we solved Ca2+-F-actin structures in complex with ADP-BeF3−, ADP-Pi and ADP at resolutions of 2.15–2.21 Å (Fig. 1e–g, Extended Data Figs. 2, 3g–l and 4d–f, Supplementary Table 2 and Supplementary Video 2). The reconstructions showed that, even though Ca2+-actin displays slow polymerization and fast depolymerization kinetics17, it adopts stable conformations in the filamentous state. Globally, the Ca2+-F-actin structures are comparable to those of Mg2+-F-actin, with no changes in helical rise and twist and a Cα atom root-mean square deviation of <0.6 Å (Extended Data Fig. 4h–k), indicating that the change from Mg2+ to Ca2+ does not induce large conformational rearrangements in the filament.

Water relocation triggers ATP hydrolysis

Upon polymerization, subdomains 1 and 2 (SD1 and SD2) of the actin monomer rotate about 12.4°, leading to a more compact arrangement in the filament (Fig. 2a and Extended Data Fig. 6a,b), commonly referred to as flattening30. A comparison of crystal structures of G-actin in the ATP state31 with our F-actin structures allows for a description of the G- to F-actin transition in the context of solvent molecules. We first analysed the water molecules directly bound to the nucleotide cation. In Mg2+-ADP-BeF3− F-actin, Mg2+ is coordinated by Pβ of ADP, a fluoride moiety of BeF3− and four water molecules, defining a hexa-coordinated, octahedral coordination, similar to that of Mg2+ in ATP-G-actin (Figs. 1b and 2b,c). Our F-actin structure thus provides experimental evidence that Mg2+ retains its water coordination during the G- to F-actin transition, which was previously only predicted on the basis of molecular dynamics data19. We next inspected the potential relocation of water molecules near the nucleotide-binding pocket. In Mg2+-ATP-G-actin (PDB 2V52)31, there is a large cavity (about 7 Å in diameter) that accommodates several ordered water molecules in front of the ATP γ-phosphate between SD3 and SD1 (SD3/1 cavity, Fig. 2f). Actin flattening results in the upward displacement of the H-loop (residues 72–77) and the movement of the proline-rich loop (residues 108–112) and the side-chains of Q137 and H161 towards the nucleotide (Fig. 2f). As a result, the SD3/1 cavity becomes narrower (about 5 Å in diameter) and a cavity in the SD1 (deemed SD1 cavity) opens up (Extended Data Fig. 6a). Because of the narrowing of the SD3/1 cavity, several water molecules would clash with amino acids and therefore need to relocate to the SD1 cavity through a path which involves the movement of a water molecule (Wx) that is bound in between both cavities (Fig. 2f and Supplementary Video 3). Notably, the relocation of water molecules into the SD1 cavity does not impact those coordinating the nucleotide-bound Mg2+ ion (Fig. 2c,f).

After the conformational change from G- to F-actin, only three water molecules remain in the SD3/1 cavity that are not coordinated by Mg2+. One of them is hydrogen-bonded to the side-chain of Q137 (Fig. 3a and Extended Data Fig. 6c). Owing to the rearrangement of the nucleotide-binding site in F-actin, this water molecule is much closer (3.6 Å) to the Pγ-analogue BeF3− than in Mg2+-ATP-G-actin (>4 Å distance from the Pγ and 4.6 Å in PDB 2V52; ref. 31) (Fig. 3a and Extended Data Fig. 6c). As no other ordered water molecules align in front of the nucleotide, the water molecule that is hydrogen-bonded to Q137 is likely to represent the nucleophile (Wnuc) that hydrolyses ATP in F-actin. The O–Be–Wnuc angle in the structure is 144° (Fig. 3a and Extended Data Fig. 6c,i), whereas an angle of >150° is required for efficient nucleophilic attack19. Although it cannot be excluded that nucleotide orientation is slightly altered between ADP-BeF3−-bound and ATP-bound F-actin, inspection of the reconstruction showed that the density for Wnuc is extended (Extended Data Fig. 6e), indicating that the position of Wnuc is not fixed, allowing it to move into a position that brings the O–Be–Wnuc angle >150° while remaining hydrogen-bonded to Q137. In other words, Wnuc probably exchanges between hydrolysis-competent and hydrolysis-less-competent configurations.

Fig. 3: Structural insights into ATP hydrolysis and Pi release.
figure 3

a,b, Isolated amino-acid and water arrangement near the nucleotide in Mg2+-ADP-BeF3− F-actin (a) and Mg2+-ADP-Pi F-actin (b). Regions unimportant for interactions are depicted as smaller sticks. Amino acids and the proposed nucleophilic water (Wnuc) and assisting water (Wbridge) are annotated. c,d, Internal solvent cavities near the Pi binding site in ADP-Pi (c) and ADP (d) structures of Mg2+-F-actin. The upper panel shows the F-actin structure as surface with the bound Pi and water molecules. In the lower panel, F-actin is shown in cartoon representation and the amino acids forming the internal cavity are annotated and shown as sticks. Hydrogen bonds are depicted as dashed lines. The position of the proposed back door is highlighted in purple in the upper panel. All distances are shown in angstroms.

Full size image

Although Q137 positions Wnuc in close proximity to Pγ, the Q137 side-chain cannot accept a proton to act as a catalytic base for the hydrolysis. We found no other amino acids that are close enough to interact with Wnuc. Instead, Wnuc resides at about 4.2 Å from a neighbouring water molecule (Wbridge) (Fig. 3a), which is not close enough to form a hydrogen bond but the movement of Wnuc into a hydrolysis-competent position would also place Wnuc in hydrogen-bonding distance to Wbridge. By forming hydrogen bonds with D154 and H161, Wbridge may represent a Lewis base with a high potential to activate Wnuc and potentially act as an initial proton acceptor during hydrolysis, followed by transfer of the proton to D154, as previously predicted by simulations32,33 or, alternatively, to H161. In conclusion, we propose that Q137 coordinates Wnuc but that the hydrogen-bond network comprising Wbridge, D154 and H161 is responsible for the activation of Wnuc and proton transfer. Indeed, the ATP hydrolysis rates of the Q137 to alanine (Q137A) actin mutant are slower but not abolished34, whereas the triple mutant Q137A/D154A/H161A-actin exhibits no measurable ATPase activity35.

Slow polymerization of Ca2+-actin

We next inspected the G- to F-actin transition in Ca2+-actin structures. The Ca2+ ion in G-actin is coordinated by the Pβ and Pγ of ATP and five water molecules in a hepta-coordinated, pentagonal-bipyramidal arrangement (Fig. 2d)36,37. By contrast, in Ca2+-ADP-BeF3− F-actin, the Ca2+ ion loses one coordinating water and displays an octahedral coordination sphere (Fig. 2e). How does the G- to F-actin transition lead to changes in Ca2+-coordination? Globally, the flattening of Ca2+-actin triggers rearrangements that are analogous to those observed in Mg2+-actin (Extended Data Fig. 6a,b), with a similar relocation of ordered water molecules from the narrowing SD3/1 cavity to the widening SD1 cavity (Fig. 2g). However, in Ca2+-G-actin, one of the relocating water molecules (Wx) resides within the coordination sphere of Ca2+, indicating that the hydration shell of the Ca2+ ion needs to be altered for the G- to F-actin transition to occur. Thus, our analysis rationalizes why the inner-sphere coordination of Ca2+ changes from hepta-coordinated, pentagonal-bipyramidal in ATP-G-actin to hexa-coordinated, octahedral in ADP-BeF3− F-actin (Fig. 2d,e,g and Extended Data Fig. 5b). The required rearrangement of the Ca2+-coordination sphere could pose a kinetic barrier for the G- to F-actin transition, which provides a structural basis for the slower polymerization kinetics of Ca2+-actin compared to Mg2+-actin.

We also assessed the ATP hydrolysis mechanism of Ca2+-actin, which exhibits a five times slower ATP hydrolysis rate (0.06 s−1) than Mg2+-actin (0.3 s−1) (ref. 5). A structural comparison between Ca2+-ATP-G-actin (PDB 1QZ5) and Ca2+-ADP-BeF3− F-actin demonstrates that the water corresponding to Wnuc, which is hydrogen-bonded to Q137, locates closer to Be in F-actin (3.7 Å) than to Pγ in G-actin (4.5 Å) (Extended Data Fig. 6d,h,i). Thus, the induction of ATP hydrolysis is comparable in Ca2+-actin and Mg2+-actin. However, in Ca2+-ADP-BeF3− F-actin, the distance between Q137 and Wnuc is 3.4 Å (3.2 Å in Mg2+-actin) and the O–Be–Wnuc angle is 137° (144° in Mg2+-actin) (Extended Data Fig. 6c–i), making the position of Wnuc similar, but slightly less favourable for nucleophilic attack, providing a likely explanation for the slower hydrolysis rate of Ca2+-F-actin.

Pi release occurs in a transient state

We next analysed how ATP hydrolysis affects the F-actin nucleotide arrangement. In the Mg2+-ADP-Pi state, the cleaved Pi moiety is separated from ADP by at least 2.9 Å (Figs. 1c and 3b and Extended Data Fig. 5a), indicating that ADP and Pi do not form a covalent bond. After Pi release, the Pi-binding site is occupied by a water molecule, which also holds true for structures of monomeric Mg2+-ADP-G-actin38 (Extended Data Fig. 5a). Taken together, our structures show that the Mg2+-coordination shell is octahedral and that Mg2+ resides at a fixed position beneath the Pβ moiety of the nucleotide in all stable states of G-actin and F-actin.

After ATP hydrolysis in Ca2+-F-actin, the coordination of Ca2+ is also octahedral in the ADP-Pi state (Fig. 1f and Extended Data Figs. 5b and 6h). Interestingly, one coordinating water molecule is replaced by the side-chain of Q137 (Extended Data Fig. 5c–e). Finally, following Pi release, the Ca2+-ADP-F-actin structure shows that the Ca2+ ion changes position so that it is directly coordinated by both the Pα and Pβ of ADP (Fig. 1g and Extended Data Fig. 5b) and four water molecules in an octahedral arrangement. Thus, in contrast to Mg2+, the Ca2+ ion position is not fixed in F-actin and its coordination changes considerably during the ATPase cycle. The absence of discrete differences in amino-acid conformation between the ADP states of Ca2+- and Mg2+-F-actin suggests that the faster depolymerization rates of Ca2+-F-actin may be caused by differences in long-range filament mechanostability or conformations at filament ends.

Pi is thought to exit from the F-actin interior through the so called ‘back door’39, which is formed by the side-chains of R177 and N111 and the backbones of methylated histidine 73 (H73) and G74 (Fig. 3c,d and Extended Data Fig. 5f,g). In this model, S14 switches rotameric position to change its hydrogen-bonding interaction from the backbone amide of G74 to the one G158, thereby allowing Pi to approach the back door, where R177 would mediate its exit22,39. On the basis of a lower-resolution reconstruction, the back door was proposed to be open in ADP-F-actin22. However, the S14–G74 hydrogen bond is intact and the back door is closed in our 2.2 Å structures of both the ADP-Pi and ADP states of Mg2+-F-actin (Fig. 3c,d) and Ca2+-F-actin (Extended Data Fig. 5f,g). Thus, unexpectedly, our structures show that the back door closes again after Pi release, indicating that the F-actin conformation that allows for the exit of Pi is a transient state. In fact, the proposed rotameric switch of S14 towards G158 alone would not result in an opened back door, which suggests that larger rearrangements are required for Pi release, indicating that the release mechanism remains incompletely understood. We envision that Pi release could be further explored by molecular dynamics simulations or time-resolved cryo-EM in future research, guided by our structures as high-quality starting models.

Coupling of filament centre to periphery

We next examined the nucleotide state-dependent conformational mobility of the D-loop (residues 39–51) and the carboxy terminus at the intrastrand (or longitudinal) interface in the actin filament21,40. The intrastrand arrangements in the current high-resolution Mg2+-F-actin structures are largely consistent with those in previous reconstructions21, with a mixture of open/closed D-loop conformations in ‘young’ ATP-bound filaments and a predominantly closed D-loop arrangement in ‘aged’ ADP-F-actin (Extended Data Fig. 7). In the Mg2+-ADP-BeF3− F-actin structure, we could separate two intrastrand conformations through a focused classification approach (Extended Data Fig. 8a,b). The first conformation (about 37% of the particles, 2.32 Å resolution) represents the open D-loop, where the D-loop bends outwardly and interacts with the extended C terminus of the adjacent actin subunit. In the second conformation (about 63% of the particles, 2.32 Å resolution), the C terminus remains extended but turns away from the inwardly folded, closed D-loop (Extended Data Fig. 7). In Mg2+-ADP-Pi F-actin, the C terminus forms a compact, folded α-helix and the D-loop is predominantly closed, whereas the Mg2+-ADP-F-actin structure resembles the extended C terminus and closed D-loop conformation of the Mg2+-ADP-BeF3− state (Extended Data Fig. 7).

We next analysed how conformational changes in the nucleotide-binding pocket are transmitted to the filament surface. Surprisingly, we could not identify a direct communication path between the D-loop and the nucleotide-binding site (Extended Data Fig. 8c–e). Hence, our structures do not explain why the intrastrand interface can adopt two conformations in the ATP state of Mg2+-actin. However, we were able to identify the structural basis for the nucleotide-dependent conformation of the C terminus. After ATP hydrolysis, Q137 in the nucleotide-binding pocket moves upward by about 0.4 Å in the ADP-Pi state so that it resides within 3.1 Å of Pi (Figs. 3b and 4 and Supplementary Video 4). This upward movement of Q137 triggers a sequence of small movements in the SD1; the proline-rich loop (residues 108–112) moves slightly forward and triggers the relocation of the E107–R116 salt bridge, which allows the penultimate residue C374 to flip into a hydrophobic pocket, permitting R116 to interact with the carboxylate group of the C-terminal residue F375 (Fig. 4 and Supplementary Video 4). Altogether, these changes result in a compact, folded C-terminal helix, which then unfolds again when Q137 moves downward after Pi release (Extended Data Fig. 9). In conclusion, our data suggest that Q137 and its surrounding residues represent a major region that is capable of sensing the nucleotide state and transmitting it to the periphery.

Fig. 4: Structural coupling of the nucleotide-binding site to the filament surface.
figure 4

Top: differences in the SD1 of F-actin in the Mg2+-ADP-BeF3− and Mg2+-ADP-Pi structures. Residues thought to be important for the movement are annotated. Bottom: zoom of the nucleotide-binding site (1) and C-terminal region (2) of the SD1. Arrows depict the direction of the putative movement from the Mg2+-ADP-BeF3− to the Mg-ADP-Pi structure. All distances are shown in angstroms. Distances shown in the Mg2+-ADP-BeF3− structure are blue, whereas those in the Mg-ADP-Pi structure are orange.

Full size image

Nucleotide-state sensing by cofilin

It has been proposed that the intrastrand interface represents a major site for ABPs such as cofilin to sense the nucleotide state of F-actin. Cofilin binds and changes the helical twist of actin filaments by wedging itself between the C terminus and D-loop23,24,41,42, which may be inhibited by the open D-loop conformation21. In the structures of Ca2+-F-actin, we observed similar arrangements of the intrastrand interface compared to those of Mg2+-F-actin, except that the open D-loop conformation is adopted to a lesser extent in Ca2+-ADP-BeF3− F-actin (Extended Data Fig. 7). We therefore proposed that if the D-loop arrangement represents the dominant recognition signal, cofilin would efficiently bind and sever Ca2+-F-actin regardless of the nucleotide state. To assess this, we incubated cofilin-1 and Mg2+- or Ca2+-bound F-actin in three nucleotide states and measured cofilin-dependent filament severing. The assays showed that, comparable to Mg2+-F-actin, cofilin-1 only substantially severs the ADP state but not to ADP-BeF3− and ADP-Pi-bound Ca2+-F-actin (Extended Data Fig. 10a). Therefore, the D-loop conformation does not represent the only sensor for cofilin-1 binding. What other mechanism does cofilin use to sense the nucleotide state? Our structures show that BeF3− and Pi make hydrogen-bonding interactions with S14 of SD1; and the backbones of G158 and V159 of the SD3 (Extended Data Fig. 10c–h). Thus, the γ-phosphate moiety forms a bridge between the two subdomains, which is absent in the ADP state. Previous studies indicated that the tight binding of cofilin to F-actin necessitates a change in helical twist of the filament24, which involves the rotation of SD1 and 2 (ref. 42; Extended Data Fig. 10b). This rotation involves the movement of the loop of S14, which is not possible when S14 is hydrogen-bonded to the γ-phosphate of the nucleotide in ATP or ADP-Pi-F-actin (Extended Data Fig. 10c–i). Our results therefore support the previously proposed model that cofilin cannot form a strong complex with F-actin when the γ-phosphate moiety is present23 and that it potentially senses the mechanical properties of the filament. This is in agreement with numerous biochemical observations that ADP/cofilin proteins can only sever actin filaments when Pi or BeF3− is removed from the active site10,11,43.

Conclusions

The structures of F-actin at about 2.2 Å resolution show the filament architecture and the arrangement of the nucleotide-binding pocket in unmatched detail, allowing us to revise certain statements about the flexibility and stability of F-actin. Traditionally, the structure of F-actin has been described as polymorphic44, whereas ‘aged’ ADP-F-actin is regarded as a structurally destabilized form of the filament45. By contrast, our structures are remarkably similar in all solved nucleotide states, showing that ADP-F-actin should not be regarded as destabilized but rather as a ‘primed state’, which exhibits faster depolymerization rates at the filament ends and is sensitive to cofilin binding and severing due to the absence of the γ-phosphate moiety. This model is highly consistent with a recent study, which showed that the nucleotide state affects the bending and mechanical properties of the filament, rather than large amino-acid rearrangements46. We furthermore show how the mechanism of ATP hydrolysis in F-actin and the slow polymerization rates of Ca2+-actin depend on the positions of water molecules, emphasizing that high-resolution structures are crucial for explaining these important aspects of filament assembly and aging. Our optimized cryo-EM workflow now also paves the way for high-resolution structures of F-actin bound to ABPs, which will enhance our understanding of cytoskeletal remodelling. Finally, we envision that our solvent-molecule visualizing structures of F-actin may serve as high-quality templates for the development of actin-binding small molecules, which may be tailored for imaging and, perhaps, even therapeutic applications47,48.

Methods

Protein purification

Skeletal α-actin was purified from rabbit muscle acetone powder through an established protocol that was described previously18,21,49. A total of 0.5 g of frozen muscle acetone powder was thawed, resuspended in 10 ml of G-buffer (5 mM Tris pH 7.5, 0.2 mM CaCl2, 0.2 mM ATP, 0.5 mM tris(2-carboxyethyl)phosphine (TCEP), 0.1 mM NaN3) and stirred for 25 min at 4 °C. The suspension was then filtered and the pellet was again resuspended in 10 ml of G-buffer and subjected to the same stirring procedure. After filtering, the 20 ml of filtered solution was ultracentrifuged at 100,000g for 30 min to remove any remaining debris. The supernatant was collected and actin was polymerized by the addition of 2 mM MgCl2 and 100 mM KCl (final concentrations) for 1 h at room temperature. To remove ABPs bound to actin, solid KCl was added to the solution to bring the KCl concentration to 800 mM and the mixture was incubated for 1 h at room temperature. Then, the actin filaments were pelleted by ultracentrifugation at 100,000g for 2 h and resuspended in 5 ml of G-buffer. Actin was depolymerized by dialysis in 1 l of G-buffer for 2 d, with one buffer exchange per day. This ensured that Ca2+ was the divalent cation bound in the active site of G-actin. On the third day, the solution was ultracentrifuged at 100,000g for 30 min and G-actin was recovered from the supernatant. The 2 d procedure of actin polymerization, high-salt wash and depolymerization by dialysis in G-buffer was repeated once more to ensure removal of all impurities and ABPs. After depolymerization, purified G-actin was flash frozen in liquid nitrogen in 50 µl aliquots at a concentration of 28 µM and stored at −80 °C until further use.

Human cofilin-1 was purified as described previously50.

Reconstitution of F-actin in different functional states

Structural studies were performed on rabbit skeletal α-actin, which is identical to human skeletal α-actin in amino-acid sequence. G-actin aliquots were thawed and ultracentrifuged for 1 h at 100,000g to remove aggregates. For structures determined with Mg2+ as nucleotide-associated cation, G-actin (28 µM) was mixed with 0.5 mM EGTA and 0.2 mM MgCl2 to exchange Ca2+ for Mg2+ 5–10 min before polymerization. In all subsequent steps, buffers contained CaCl2 for the isolation of F-actin with Ca2+ as divalent cation or MgCl2 for the isolation of F-actin with Mg2+ as divalent cation. Actin polymerization was induced by the addition of 100 mM KCl and 2 mM CaCl2/MgCl2 (final concentrations). Actin was polymerized at room temperature for 2 h and subsequently overnight at 4 °C. The next morning, filaments were isolated through ultracentrifugation at 100,000g for 2 h.

For the aged ADP-F-actin states, the filament pellet was resuspended in F− buffer: 5 mM Tris pH 7.5, 100 mM KCl, 2 mM CaCl2/MgCl2, 2 mM NaN3, 1 mM dithiothreitol (DTT). F-actin was used for cryo-EM sample preparation about 1 h after pellet resuspension.

When choosing an ATP analogue for structural studies, we considered that previous work from our group has shown that the widely used non-hydrolysable ATP analogue AppNHp (also known as AMP-PNP) is a suboptimal ligand for F-actin because its degradation product, the ADP analogue AppNH2, exhibits higher affinity for F-actin51 and hence accumulates in the active site during filament preparation21. We therefore opted to use ADP-BeF3− as mimic of ATP. To obtain these ADP-BeF3− states of F-actin, aged ADP-bound filaments were resuspended in F− buffer supplemented with 0.75 mM BeF2 and 5 mM NaF. Because the on-rate of BeF3− for ADP-F-actin is relatively slow29, the filaments were incubated in this buffer for >6 h before cryo-EM sample preparation to ensure saturation with BeF3−.

To isolate F-actin in the ADP-Pi state, we resuspended the actin pellet in F− phosphate buffer: 5 mM Tris, 50 mM KCl, 2 mM CaCl2/MgCl2, 2 mM NaN3, 1 mM DTT, 50 mM potassium phosphate pH 7.5. To remove any potential precipitates of calcium phosphate and magnesium phosphate, we filtered the buffers directly before use. The filaments were incubated in F− phosphate buffer for >6 h before cryo-EM sample preparation.

Cryo-EM grid preparation

A total 2.8 µl of F-actin sample (3–16 µM) was pipetted onto a glow-discharged R2/1 Cu 300 mesh holey-carbon grid (Quantifoil). After incubating for 1–2 s, excess solution was blotted away and the grids were plunge frozen in liquid ethane or a liquid ethane/propane mixture using a Vitrobot Mark IV (Thermo Fisher Scientific). The Vitrobot was operated at 13 °C and the samples were blotted for 9 s with a blot force of −25.

Cryo-EM grid screening and data collection

Grids were prescreened on a 200 kV Talos Arctica Microscope (Thermo Fisher Scientific) equipped with a Falcon III detector (Thermo Fisher Scientific). Typically, low-magnification grid overviews (atlases) were collected using EPU (Thermo Fisher Scientific). Afterwards, around two holes per grid square were imaged at high magnification for a total of five grid squares to visualize F-actin. The grids that displayed optimal filament concentration and distribution were then retrieved from the microscope and stored in auto grid boxes (Thermo Fisher Scientific) in liquid nitrogen until further use for high-resolution data collection.

All datasets were collected on a 300 kV Titan Krios microscope (Thermo Fisher Scientific) equipped with a K3 detector (Gatan) and a postcolumn energy filter (slit width of 15 eV). Videos were obtained in super-resolution mode at a pixel size of 0.3475 Å, with no objective aperture inserted. All datasets were collected on the same microscope at the same magnification of ×130,000, to ensure that the resulting cryo-EM density maps could be compared directly without issues caused by pixel size discrepancies. Using EPU, we collected about 10,000 videos per dataset in 60–80 frames at a total electron exposure of about 72–90 e− Å−2. The defocus values set in EPU ranged from −0.7 to −2.0 µm. The data quality was monitored live during acquisition using TranSPHIRE52. If necessary, the microscope was realigned to ensure optimal imaging conditions. An overview of the collection settings used for each dataset can be found in Supplementary Tables 1 and 2.

Cryo-EM image processing

For each dataset, video preprocessing was performed on the fly in TranSPHIRE52, the super-resolution videos were binned twice (resulting pixel size of 0.695 Å), gain corrected and motion corrected using UCSF MotionCor2 (ref. 53), contrast transfer function (CTF) estimations were performed with CTFFIND4.13 (ref. 54) and F-actin segments were picked using the filament picking procedure in SPHIRE_crYOLO55,56 using a box distance of 40 pixels per 27.8 Å and a minimum number of six boxes per filament. The resulting particles were extracted in a 384 × 384 pixel box and further processed into the pipeline of helical SPHIRE v.1.4 (ref. 57). The number of extracted particles differed per dataset and ranged from 1,296,776 (Mg2+-ADP dataset) to 3,031,270 (Ca2+-ADP-Pi dataset) particles. For each dataset, the particles were two-dimensionally classified in batches of 20,000 particles using ISAC2 (ref. 58) (sp_isac2.py). All classes were then pulled together and manually inspected and those that represented non-filament picks and ice contaminations were discarded. A virtual substack was created of the remaining particles (sp_pipe.py isac_substack) and the particles were subjected to three-dimensional helical refinement52 using meridien alpha. This refinement approach within SPHIRE imposes helical restraints tailored to the helical sample to facilitate the refinement process but does not apply helical symmetry. Hence, symmetrization artifacts during refinement are avoided. We refined the F-actin structures with a restrained tilt angle during exhaustive search (--theta_min 90 –theta_max 90 –howmany 10) and used a filament width of 140 pixels (97.3 Å) and a helical rise of 27.5 Å to limit shifts larger than one subunit to prevent duplication of particles. For the first processed dataset, EMD-11787 (ref. 47) was lowpass filtered to 25 Å and supplied as initial model for the refinement. The first meridien alpha refinement of each dataset was performed without a mask; the resulting three-dimensional density map of this refinement was then used to create a soft mask using sp_mask.py that covered about 85% of the filament (326 pixels in the Z-direction). The global refinement was then repeated with the same settings but with the mask applied. These masked refinements yielded F-actin reconstructions at resolutions of 2.6–3.0 Å. The particles were then converted to be compatible with Relion59 using sp_sphire2relion.py. Within Relion 3.1.0, the particles were subjected to Bayesian polishing60 for improved estimation of particle movement trajectories caused by beam-induced motion; and to CTF refinements61 to estimate per-particle defocus values and to correct for beam tilt, threefold (trefoil) astigmatism, Cs and fourfold (tetrafoil) astigmatism and anisotropic magnification. We then performed three-dimensional classification without image alignment (8 classes, 25 iterations, tau2fudge 4) to remove particles that did not contribute high-resolution information to the reconstruction. Typically, one or two high-resolution classes containing most particles were selected and the other low-resolution classes were discarded. Finally, after removal of duplicates, this set of particles was subjected to a masked refinement with solvent flattening Fourier shell correlations (FSCs) and only local searches (initial sampling 0.9°) in Relion using the map (lowpass filtered to 4.0 Å), mask and particle orientations determined from SPHIRE. These refinements yielded cryo-EM density maps at resolutions of 2.15–2.24 Å according to the gold-standard FSC = 0.143 criterion. The final maps were sharpened with a negative B-factor and corrected with the modulation transfer function of the K3 detector. Local-resolution estimations were performed in Relion.

To separate the closed and open D-loop conformations in the Mg2+-ADP-BeF3− reconstruction, the good 2,228,553 particles of this dataset were subjected to a focused classification without image alignment in Relion. We created a soft mask around an inter-F-actin contact comprising the D-loop of the central actin subunit and the C terminus of the subunit directly above. Initial attempts to separate the D-loop conformations into two classes using a single density map as initial model were unsuccessful because all particles would end up in a single class with a mixed closed/open D-loop population. We therefore classified the particles into two classes using two references; the jasplakinolide-bound, Mg2+-ADP-Pi (in-house structure) and Mg2+-ADP structures as templates for, respectively, open and closed D-loop conformations. A particle separation into two classes with two initial references was chosen because the two conformations of closed and open D-loops were visible in the refined, non-sharpened reconstruction computed through all 2,228,553 good particles. Potential alternative conformations of the D-loop adopted by only a marginal number of F-actin particles are not distinguishable by our classification strategy. After optimization of the tau2fudge parameter, which required a high value due to the small size of the mask compared to the full box, this classification without image alignment (two classes, 25 iterations, tau2fudge = 500) yielded two classes with clearly distinguishable D-loop conformations. The particles belonging to each class (834,110 for the open D-loop and 1,394,443 for the closed D-loop) were then selected and separately refined in Relion with solvent flattening FSCs using the map with mixed D-loop conformation (filtered to 8.0 Å, at which the D-loop conformation is indistinguishable) as reference and the SPHIRE-mask covering 85% of the filament for both refinements. These refinements were performed with the default tau2fudge = 1 to prevent any overfitting. The high tau2fudge value used during the classification without image alignment was only used to sort particles and not for further map processing and analysis. The resulting maps of the open D-loop (2.32 Å) and closed D-loop (2.32 Å) particles showed, respectively, the expected open and closed D-loop conformations at the region that was used for focused classification. The D-loop conformations remained a mix between open and closed in actin subunits within the map that were not used for focused classification.

Model building, refinement and analysis

To build the F-actin models in the high-resolution density maps, the structure of F-actin in complex with an optojasp in the cis state47 (PDB 7AHN) was rigid-body fitted into the map of F-actin Ca2+-ADP state. We modelled five actin subunits in each map to capture the entire interaction interface within the filament because the central subunit interacts with four neighbouring protomers. The central actin subunit in the map was rebuilt manually in Coot62 and the other actin subunits were adjusted in Coot by applying non-crystallographic symmetry using the central subunit as master chain. The structure was then iteratively refined using Coot (manually) and phenix real-space refine63 with non-crystallographic symmetry restraints but without imposing any geometry restraints. The structures of all other states were built by rigid-body fitting of the Ca2+-ADP structure in the map belonging to each F-actin state, followed by manual adjustments in Coot. These structures were then refined through a similar protocol of iterative cycles in Coot and phenix real-space refine. All solvent molecules (ions and water molecules) were placed manually in Coot in the central actin subunit and were then placed in the other subunits using non-crystallographic symmetry. Because the local resolution of each F-actin reconstruction is highest in the centre and lower at the periphery of the map, we inspected all water molecules in each structure manually before the final phenix refinement; water molecules with poor corresponding cryo-EM density were removed. A summary of the refinement quality for each structure is provided in Supplementary Tables 1 and 2. For the structural analysis, the central actin subunit in the structure was used, unless stated otherwise. The solvent cavities in the structures were calculated using the CASTp 3.0 web server64. All figures that depict cryo-EM density maps and protein structures were prepared in UCSF ChimeraX65. The helical parameters reported in Supplementary Tables 1 and 2 were estimated from the atomic model of five consecutive subunits independently fitted to the map as described previously66.

Cofilin severing assays

F-actin in different nucleotide states was prepared as for cryo-EM experiments (see above). Severing assays were performed in 20 μl volumes by incubating 5 μM of F-actin with 5, 10 or 20 μM of cofilin for 30 min at room temperature, followed by centrifuging the samples at 120,000g in a TLA120.1 rotor for 15 min at 4 °C. After centrifugation, aliquots of the supernatant and pellet fractions were separated by SDS–polyacrylamide gel electrophoresis and analysed by densitometry using Image Lab software v.6.0.1 (Bio-Rad) and plotted using GraphPad. The data points are available as Source data.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.