Introduction

The majority of ATP, the energy currency of life, is synthesized by oxidative phosphorylation or photophosphorylation, catalyzed by ATP synthase1,2,3. F-type ATP synthases, FoF1, exist in the inner membrane of mitochondria, the plasma membrane of bacteria, and the thylakoid membrane of chloroplasts, while Archaea and some bacteria express the homologous V-type ATP synthases, called V/A-ATPases4,5,6,7,8. FoF1 consists of a hydrophilic F1 domain containing three catalytic sites9, and a hydrophobic Fo domain housing a proton translocation channel10,11 (Fig. 1a). The movement of protons through the translocation channel of Fo drives rotation of the γ-stalk and this catalyzes the conversion of ADP to ATP at the α3β3 dimer catalytic sites referred to as αEβE, αTβT, and αDβD. Upon dissociation from the Fo domain, the F1 domain can catalyze the reverse reaction, i.e. hydrolysis of ATP driven by rotation of the central γ subunit inside the cylinder comprising α3β3 (Fig. 1b, c). The catalytic sites are located at interface between the α and β subunit in each catalytic dimer. Most of the residues involved in ATP binding and hydrolysis are found in the β subunit, although residues in the α subunit are also involved12,13.

Fig. 1: Structures and rotation mechanism of FoF1.
figure 1

a Cryo-EM structure of FoF1 ATP synthase in the step-waiting conformation. Each subunit is colored differently. The F1 domain contains three catalytic αβ dimers which surround the γ subunit. b, c Side views (upper) and bottom section views (lower) of the F1 domain in the step-waiting (0°) and 81° structure, respectively. The three catalytic dimers are represented different colors: αE (marine blue) and βE (light blue), αT (moss green) and βT (light green), and αD (purple) and βD (light purple). The βD subunit in step-waiting adopts a more open structure, termed as DHO. The γ subunits are represented as a gray tube in the center of both α3β3 sub-complexes. d A proposed scheme for chemo-mechanical coupling during a 120° rotation step of the F1 domain driven by ATP. In this model, ATP binding to the F1 domain immediately initiates the 80° rotation with an associated release of ADP. ATP is hydrolyzed at the 80° dwell position with no associated rotation of γ. It is the release of Pi from the enzyme which is suggested to drive the final 40° rotation.

Full size image

The binding change mechanism of ATP synthesis at the F1 domain by rotation of the γ subunit relative to α3β3 was firstly proposed by P. Boyer9,14. According to this mechanism, the three catalytic dimers adopt different conformational states, and they interconvert sequentially between three the different conformations as catalysis proceeds. Strong support for this asymmetrical F1 arrangement was confirmed by the X-ray crystal structure of mitochondrial F1 (MF1)13, which revealed the three catalytic β subunits in different conformational states and with different nucleotide occupancy at the catalytic sites; closed βDP with ADP, closed βTP with ATP analog, and open βE with no bound nucleotide.

Rotation of F1 driven by ATP hydrolysis was directly demonstrated in single molecule observations using bacterial F1 from Geobacillus stearothermophilus (Bacillus PS3 sp.)15. When using 40 nm gold beads with almost negligible viscous resistance, F1 pauses at the 80° dwell position during the 120° rotation step at an ATP concentration close to the Km16. Histogram analysis of the frequency of dwell times for each dwell position suggests a model in which binding of ATP to F1 at 0° causes an initial 80° rotation step of γ subunit, followed by a 40° rotation of the γ subunit due to hydrolysis of ATP and release of phosphate at the 80° dwell position (Fig. 1d)16,17. In addition, measurement of the rotational velocity of the rotating probe with viscous resistance or external force suggests that F1 converts the hydrolysis energy of ATP into torque with high efficiency18,19. However, these single-molecule experiments only allow observation of the motion of the γ subunit and hence do not provide information on the changes occurring at each catalytic site.

In order to elucidate the entire F1 rotation mechanism driven by ATP, we set out to determine the cryo-EM structures of the catalytic intermediates of the F1 domain during rotation. We trap these intermediate states by freezing cryo-EM grids at different time points or under different reaction conditions allowing us to build up a picture of the chemo-mechanical cycle of these motor proteins step by step20. Previously, we determined several intermediate structures of V/A-ATPase from Thermus thermophilus revealing to a high level of detail the ATP hydrolysis cycle in these enzymes. The results showed that catalytic events occur simultaneously at the three catalytic sites of A3B3 stator, driving the 120° rotation of the central rotor21. Sobti et al. also performed snapshot analysis using a mutant F1 exhibiting slow ATP hydrolysis and obtained structures corresponding to both the 0° and 80° rotation angles of γ subunit22. They proposed, however, a different rotation model of rotary ATPase, in which ATP binding and ATP hydrolysis drive the initial 80° rotation and the 40° rotation of F1, respectively. As a result, the coupling of the chemo-mechanical cycle of rotary ATPases remains controversial, 60 years after Boyer first predicted the rotary catalytic mechanism of ATP synthase.

Here, we describe multiple reaction intermediates of the wild type of FoF1 from G. stearothermophilus captured using cryoEM. These structural intermediates of the F1 domain fill in the missing pieces in the rotation scheme and have allowed us to redraw the complete chemo-mechanical coupling scheme of rotation in FoF1 powered by ATP hydrolysis.

Results

Sample preparation and analysis

The ATP hydrolytic activity of G. stearothermophilus FoF1 is very low due to the up-form of the C-terminus of ε subunit23,24. In this study, we used the ΔεCT-FoF1 mutant, less susceptible to ε inhibition by truncation of ε-C-terminal helix than the wild-type24. The ΔεCT-FoF1 expressed in E. coli membranes was solubilized in DDM, then purified as described in Methods section. The ΔεCT-FoF1 exhibited ATPase activity, which obeyed simple Michaelis–Menten kinetics with a Vmax of 164 s−1 and a Km of 30 μM (Supplementary Fig. 1a). The purified ΔεCT-FoF1 was subjected to nucleotide depletion treatment by dialysis against EDTA-phosphate buffer, in order to deplete endogenous nucleotide. The enzymatic properties of the nucleotide depleted enzyme (ND-ΔεCT-FoF1) were almost identical to the non-depleted enzyme (Supplementary Fig. 1b). In contrast to the results obtained by Paik et al. using F1-ATPase25, our purified FoF1 showed no initial lag phase (Supplementary Fig. 1c). Instead, the ATPase activity of the enzyme gradually increased, indicating that FoF1 transitions from an initial inactive state to an active state without adopting the ADP inhibited state.

Both depleted and non-depleted forms of the purified enzymes were concentrated then subjected to a cryo-grid preparation as described below. The ΔεCT-FoF1 is referred to simply as FoF1, hereafter.

To exclude the possibility that the nucleotides present in the resulting structure were endogenous, we used ND-FoF1 without endogenous nucleotides for structural analysis at high [ATP]. When conducting reactions at low [ATP], the reaction solution typically contains 9 µM FoF1 and 25 µM ATP. Therefore, using ND-FoF1 at low [ATP] is likely to lead to ATP depletion in the solution, as most of the ATP will bind to the three α subunits in F1 domain. To avoid this issue, we used FoF1 without nucleotide depletion treatment for structural analysis at low [ATP].

Structures of 81° and post-hyd (83°) at high [ATP]

Cryo-grids were prepared using a reaction mixture containing 26 mM ATP and 9 μM ND-FoF1 to final concentration. The reaction mixture was incubated for 20 s at 25 °C, and then loaded onto a holey grid, followed by flash freezing. To prevent depletion of ATP in the reaction mixture due to the ATPase activity of FoF1, the concentration of ATP in the reaction mixture was set to 26 mM. This ensured that even after a 20-s reaction time, a saturating concentration of ATP remained in the solution, allowing us to obtain the structure under ATP-saturated conditions.

The flow charts showing image acquisition and reconstitution of the 3D structure of FoF1 at high [ATP] are summarized in Supplementary Fig. 2b. The particles containing three rotational states (154 k particles) were subjected to 3D classification focused on the F1 domain in order to obtain sub-states of the F1 domain. As the result of this classification, three sub-classes of the F1 domain were obtained with different γ positions, termed the 80°, 100° and 120° structures (Supplementary Fig. 2b). Each structure was further classified by a masked classification.

The two F1 domain 80° structures were obtained from 56 k particles using an αDβD covering mask (Supplementary Fig. 2b). The two 80° structures shared the canonical asymmetric hexamer which adopts the closed αTβT, closed αDβD, and open αEβE conformation (Fig. 2a, Supplementary Fig. 4). The angle of the γ subunits in the two 80° structures relative to the 0° structure were 81° and 83°, respectively (Figs. 2a, 3b). The nucleotide bound to the αDβD differed between the two structures. In the αDβD of the 81° structure, ATP was identified at the catalytic site (Fig. 2c, left) and we termed the structure as 81°. In contrast, the 83° structure obtained at high [ATP] contained ADP and Pi and is designated post-hydrolyzed (post-hyd) (Fig. 2c, center). This indicates that ATP bound to αDβD is hydrolyzed at 81° or between 81° and 83°, and that the structural change due to ATP hydrolysis is small.

Fig. 2: Structure of 81° (left), post-hyd (center), and 101°(right) at high [ATP].
figure 2

a Cross section of the F1 domain showing the catalytic sites viewed from the Fo side. Each catalytic dimer is shown in ribbon representation and colored as detailed in Fig. 1. The bound nucleotides are represented as spheres. b Side view structure of αDβD dimer. The left panel is a magnified view of the catalytic interface at each αDβD. c Structure of the catalytic site in αDβD. Amino acid residues and bound nucleotide and Pi are represented as sticks. Lower panels; Stick representation model of ATP/ADP and Pi with EM density in αDβD. The distance between γ and β phosphate of the ATP in each conformational state is shown in blue (Å). Each distance was calculated using Chimera software. d Structure of the catalytic site in αEβE. The EM density of ATP / Pi is superimposed onto the model. The distance between Pi and γ phosphate of ATP is shown in yellow (Å).

Full size image
Fig. 3: Structure comparison of 6 intermediates captured during the 40° step at high [ATP].
figure 3

a Cross section of the F1 domain showing the catalytic site. 81°, post-hyd, 91°, 101°, and 120° structures are arranged from left to right. The Cα displacement relative to the next structure (right side) is indicated by the red-white color gradient. The dashed square indicates the area of this figure shown in the zoomed in view in b–f. b–f Comparison of each structure to the following one. b compares the γ subunit and its surroundings, and the different rotation angles between each γ subunit and that of the next structure are indicated by different colored lines. c shows an upper view of βD, d compares βD from a side view, and e, f compare αT (120° and step waiting) and αE (91° and 101°), respectively. Each cartoon chain is colored as 81°(gray), 83°(green), 91°(blue), 101°(pink), 120°(light green), and step-waiting (orange). g Structures of the catalytic site of 91° in αEβE. The key side chains involved in coordinating Pi and nucleotide are shown in stick representation. Semi-transparent electron density for the Pi and nucleotide is shown over the stick representations. h Superimposition of the side chains of αEβE in 91°(orange) with those of αEβE in 101° (light blue).

Full size image

The overall structures of αTβT and αEβE in the 81° and post-hyd states are mostly identical, but the αDβD structures differ slightly. In the post-hyd, the CT domain of βD was in a slightly more open conformation relative to that in 81° (Fig. 2b).

In both 81° and post-hyd, ATP and Pi were identified at the catalytic sites of αEβE. The Pi is occluded by β/164 K, β/190E, β/191 K, β/252D, β/256 R, and α/365R (Fig. 2d, Supplementary Fig. 6f). The Pi binding site structure of αEβE is highly similar to that of the yeast F1 βE Pi-binding site26. Since βE has an open structure, the site occupied by Pi is separated from the site where ATP is bound. Therefore, in this state the γ-phosphate of ATP does not directly interact with the bound Pi. For other structures obtained under high [ATP], as described in the next section, ATP was identified at the catalytic sites of αEβE (Supplementary Fig. 4d). This indicates that αEβE is capable of binding of ATP to the catalytic site, regardless of the rotary angle of the γ subunit.

Structures at 91°, 101° and 120° rotation angles at high [ATP]

Two additional structures found to be rotated a further 8° and 18° relative to the post-hyd (83°), respectively, were classified (Supplementary Fig. 2b). We termed these the 91° and 101° structures (Fig. 3b). In both structures, ATP was identifiable at the catalytic site of αEβE, but Pi is additionally bound to the 91° as well as the 81° and post-hyd forms (Fig. 2d and Supplementary Fig. 4d). In contrast, Pi is not present at the catalytic site of αEβE in the 101° structure, indicating that Pi is released during the 10° rotation of the γ subunit from 91° to 101°. The αEβE in 101° adopts a more open structure than that seen in the 91° structure due to movement of the CT domain of αE (Fig. 3f and Supplementary Fig. 4d). This more open arrangement results in a more exposed catalytic site in αEβE and thus a decrease in affinity for Pi in the 101°. The geometry of the amino acid residues in βE that coordinate Pi changes due to the release of Pi (Fig. 3h), but no significant change in the conformation of the main chain in βE was observed (Fig. 3f), suggesting that the release of Pi directly does not cause the opening of αEβE.

Two 120° structures were classified (Supplementary Fig. 2b) according to structural differences in αTβT. One 120° structure, termed step-waiting, has a more closed αTβT than the second, termed 120°. αTβT in 120° is almost identical to that of the 101° structure (Fig. 3a and Supplementary Fig. 7a). The difference in αTβT between 120° and step-waiting is the result of movement of the CT domain of the αT subunit (Supplementary Fig. 7c). Thus, sequential structural changes occur; a structural change from 101° to the 120° associated with the 19° rotation of the γ subunit, followed by closing of αTβT which occurs between 120° and step-waiting. Notably, the 19° step of the γ-subunit from the 101° to the 120° occurs independently of the catalytic cycle of ATP, such as Pi release and ATP hydrolysis in the F1 domain.

Comparing the four conformational states, 81°, post-hyd, 91°, 101°, the difference in βD among these structures is more marked than that of βE and βT. As the rotational angle of the γ subunit increases, the βD adopts a more open structure due to movement of the CT domain but remains bound to both ADP and Pi (Fig. 3c and Supplementary Fig. 5a, b). The motion of the CT domain in βD continues until the rotation angle of the γ subunit reaches 101° (Fig. 3a, c). The αDβD dimer in the 101° and 120° structures can be described as half open (αDβDHO). This conformational change of αDβD also occurs largely independently of the catalytic cycle of ATP, with the exception of the release of Pi occurring between 91° and 101°. This suggests that the opening of αDβD is largely driven by the release of strain on the molecule accumulated in the initial 80° rotation step.

Structures of F1 domain at low [ATP]

To obtain the structure of the F1 domain awaiting ATP binding, FoF1 containing endogenous nucleotides was mixed with the reaction solution containing 25 µM ATP and 0.2 mg/ml of pyruvate kinase and 4 mM phosphoenolpyruvate in order to regenerate consumed ATP. The solution was incubated at 25 °C for 60 s and then loaded onto a holey grid followed by flash freezing. The flow charts showing image acquisition and reconstitution of the 3D structure of FoF1 at low [ATP] are shown in Supplementary Fig. 3b. From the combined 75 k particles for the 80° structure, three intermediate structures, two 83° and one 81° were classified. The two post-hyd structures are almost identical, including the angle of the γ subunit, apart from a slightly more open structure of the βD being evident in one of the forms.

Both 81° and post-hyd structures obtained at low [ATP] were very similar to the corresponding structures obtained at high [ATP] (Supplementary Table 1), but there were significant differences. In the 81° and post-hyd structures obtained at low [ATP], Pi was identified at the catalytic site of αEβE, but no ATP density was observed (Fig. 4a, Supplementary Fig. 8a). In addition, ATP molecule was not found in other obtained structures at low [ATP] (Fig. 4b and Supplementary Fig. 8a) indicating that ATP does not bind to αEβE at low [ATP]. This contrasts with the binding of ATP to all structures of αEβE obtained at high [ATP] (Supplementary Fig. 4d).

Fig. 4: Structures captured at low [ATP].
figure 4

a Cross section of the F1 domain in 81° showing the catalytic sites viewed from the Fo side. The bound nucleotides are represented as spheres. The electron density of the nucleotides within the enclosed circles are shown in larger view on the right. The distance between γ (or Pi) and β phosphate of the ATP in each conformational state is shown in blue (Å). Each distance was calculated using Chimera software. b Cross section of the F1 domain in ATP-waiting (120°). c Comparison of the Pi binding site in αEβE in the 91° and 120° structures with electron density indicated for the Pi. d Comparison of the catalytic site of the αEβE in ATP waiting (cream) and step waiting (purple). ATP in step waiting is represented by the orange and blue sticks. The step waiting structure was captured at high [ATP].

Full size image

The density of nucleotides in the αDβD at 81° under low [ATP] is not well defined (Fig. 4a, right). This implies that ATP and (ADP + Pi) are in equilibrium at the catalytic site of αDβD in 81°. Additionally, this observation suggests that ATP hydrolysis at the catalytic site does not directly cause rotation of the γ subunit, consistent with the binding change mechanism where ATP synthesis at the catalytic site is independent of rotation of the γ subunit1.

From the combined 332 k particles for 120° structure at low [ATP], three structures of the F1 domain, 91° and 120° and step-waiting, were identified (Supplementary Fig. 3b). The 101° structure was not captured in this condition. At high [ATP], the F1 domain of 120° and step-waiting were composed of the αDβDHO with ADP and Pi bound, αTβT with ATP, and αEβE with ATP (Supplementary Fig. 4c–e). However, at low [ATP], ATP nor Pi density was identified at the catalytic site of αEβE in either the 120° and step-waiting states. We refer to the step-waiting containing an empty αEβE as ATP-waiting.

In summary, ATP was not identified at αEβE in all five structures, 81°, post-hyd, 91°, 120°, and ATP-waiting, obtained at low [ATP]. Comparing these results with the nucleotide occupancy of αEβE in all structures obtained at high [ATP] indicates that ATP is capable of binding to αEβE independent of the rotation angle of γ in the F1 domain (Fig. 5).

Fig. 5: ATP driven rotation scheme of FoF1.
figure 5

a Under low [ATP] conditions, the catalytic site in αEβE of ATP waiting remains empty. b The step-waiting is formed by binding of ATP to αEβE of ATP waiting. c The step-waiting initiates the 80° rotation step of the γ subunit coupled with structure transition of the three αβ dimers; αEβE to αTβT with a zippering motion caused by binding of ATP to αEβE, αTβT to αDβD, and αDβDHO to αEβE with associated release of ADP via an unzippering motion of αDβDHO. d ATP bound to αTβT is hydrolyzed in the αDβD dimer of 81° just after the 81° rotation. e, f Once ATP bound to αDβD is hydrolyzed, an unzippering motion of αDβD (purple arrows) proceeds via a 10° rotation step of γ, resulting in the structural change of hydrolysable to 91° through post-hyd. The outward motion of αE in 91°(light blue arrow) in concert with a further 10° rotation induces release of Pi, resulting in 101°which adopts a more open αEβE. g The final rotation from 101° to 120° occurs without structural change in any of three catalytic dimers. h Further motion of the CT domain of αE induces the structural between 120° and, i ATP waiting without any associated rotation of the γ subunit. j The step-waiting (120°) is formed by binding of ATP to αEβE of ATP waiting. The αEβE of six intermediates (81°, post-hyd, 91°, 101°, 120°, and ATP waiting) are occupied with ATP at high [ATP], indicating that ATP binds to empty αEβE at any rotation angle of γ (light blue dash arrows). k 360° rotation of FoF1. CryoEM maps of FoF1 obtained at high [ATP] are placed in a circular arrangement according to the angle of the γ subunit. The three state maps are represented by yellow (state1), blue (state2), and green (state3), respectively.

Full size image

Rotation scheme of FoF1

In this study, we have captured multiple structures of F1 domain and from these we can reveal the rotation scheme that occurs during the full 120° rotation (Fig. 5a–j, Supplementary movie 1). The first 80° rotation occurring via step-waiting of rotational state 1 is coupled with structural transition of three catalytic dimers, αEβE to αTβT, αTβT to αDβD, and αDβDHO to αEβE. The resultant 81° is in rotational state 2 (Fig. 5d). We further classified the three rotational states of FoF1 from the single-particle images of each F1 domain structure, enabling us to reproduce the 360° rotation scheme of FoF1 upon hydrolysis of the three ATPs (Fig. 5k). Comparison and interpretation of these intermediate structures provides us with several important insights into the coupling of the chemo-mechanical cycle of ATP hydrolysis driven rotation of FoF1.

Tri-site mechanism

ATP binds to empty αEβE of the F1 domain at any rotation angle, as shown in Fig. 5. Based on single molecule observation experiments for F1-ATPase, the coupling scheme for chemo-mechanical cycle of F1-ATPase proposed that rotation occurs simultaneously with ATP binding to the structure at a rotation angle of 0° (Fig. 1d)16,17,18. In contrast, structural snapshots of V-type ATP synthase (V/A-ATPase) revealed that the 120° rotations occur after all three catalytic sites are occupied by nucleotides (Supplementary Fig. 10a)21. The snapshot analysis of FoF1 structures in this study indicates that binding of ATP to empty αEβE in ATP-waiting results in step-waiting in which all three catalytic sites are occupied by nucleotides, then the first 80° rotation starts (Fig. 5c). This means that ATP binding to the F1 domain does not immediately cause the 80° rotation. Our results clearly demonstrate that FoF1 and V/A-ATPase share a common tri-site mechanism in which nucleotide occupancy transitions between 2 and 3 binding sites during continuous catalysis2.

The 80° dwell

A further major insight is that the 80° dwell during the 120° step observed in F1-ATPase is the result of waiting for ATP hydrolysis at the αDβD. In the αDβD of 81°, the density of ATP is confirmed, whereas ADP+Pi is bound in the post-hyd (83°) of αDβD (Fig. 2c), indicating that hydrolysis of ATP is completed at αDβD in post-hyd.

Notably, the αDβD in 81° has the most closed conformation, adopting a slightly more open conformation in the following post-hyd (Fig. 2b). In the αDβD of 81°, the γ-phosphate of ATP interacts with β/Glu190 and β/Arg192 residues, and the adenine ring interacts with the β/Tyr341 in the CT domain (Fig. 2c and Supplementary Fig. 5c). Therefore, ATP bound to the catalytic site of αDβD stabilizes the closed conformation of the β subunit. Once ATP at the catalytic site of αDβD is hydrolyzed, αDβD is then able to structurally transition to a more open conformation (Supplementary Figs. 5c, d). Indeed, the structure of αDβD gradually becomes more open as the rotation of the γ subunit proceeds, to the half open αDβDHO state visible in the 120° ATP/step waiting state (Fig. 3c). Based on the dwell time analysis at the 80° rotation angle of F1, it has been proposed that two or more catalytic events, including ATP hydrolysis and Pi release, contribute to the 80° dwell16,17. Our results support that the dwell at this rotation angle is due mainly to ATP hydrolysis, but is also dependent on the structural transition from the 81° to the post-hyd.

The subsequent 40° rotation step

The third finding is that the final 40° rotation of the 120° step is composed of three short sub-steps (Fig. 5), that are largely uncoupled from ATP hydrolysis occurring in the F1 domain. In previously suggested rotational model of F1, the dissociation of Pi at 80° is reported to drive the last 40° of rotation (Fig. 1d)17,27. However our structures reveal that the Pi remains both following rotation beyond 80°, and indeed the structure of αEβE without Pi in 101° indicates that release of Pi occurs during the structural transition from 91° to 101°. The rotation of the γ-subunit associated with Pi release is only 10° (Fig. 5f, g), suggesting that the contribution of conformational changes to Pi release (or Pi release by conformational change) is small. Other conformational changes, such as the 8° step of post-hyd to 91° and the 19° step between 101° and 120°, occur independently of the ATP hydrolysis cycle. In addition, the αTβT of ATP-waiting adopts a more closed conformation after the structural transition from 120° to step-waiting without rotation of the γ subunit (Fig. 3e). The only explanation for these spontaneous conformational rearrangements, uncoupled from chemical reactions, is that they are caused by the release of strain inside the molecule. The 80° and 120° (0°) structures identified in the F1 domain of FoF1 are also observed in F1-ATPase22. Therefore, the strain inside the molecule that drives the 40° step is most likely of F1 origin. For instance, a comparison of the γ subunit in the 101° and 120° structures shows a slight structural difference (Supplementary Fig. 9). Together, our results strongly suggest that these structural changes, including the 10° step of 91° associated with Pi release, is driven by the release of the internal molecular strain accumulated by the initial 80° rotation.

Principle of rotation mechanism of ATP synthases

Taken together, these structural snapshots of the F1 domain indicate that FoF1 functions via a similar mechanism to the ATP-driven rotation of V/A-ATPase that we have previously demonstrated21 (Supplementary Fig. 10a). The binding of ATP to empty ABopen, corresponding to αEβE, in V2nuc with two nucleotides bound forms V3nuc with nucleotides bound to all three AB dimers (ABopen with ATP, ABsemi with ATP, and ABclosed with ADP and Pi). Three distinguishable catalytic events occur at the three AB dimers simultaneously and these events contribute to the first 120° step of the rotor in a concerted manner (Supplementary Fig. 10b)21. Considering each 120° step, the rotation mechanism of FoF1 and V/A-ATPase is almost identical. ATP binds to the F1 domain of ATP-waiting, resulting in the step-waiting where the three catalytic sites are occupied with nucleotides, corresponded to V3nuc. Step-waiting comprises αEβE with ATP, αTβT with ATP, and αDβDHO with ADP and Pi. Each conformational change, from αEβE to αTβT, αTβT to αDβDHO, and αDβDHO to αEβE occurs simultaneously, accompanied by the full 120° rotation of the of γ-subunit, hydrolysis of ATP in the αTβT and release of products (ADP and Pi) from αDβDHO (Supplementary Fig. 10d). This structural transition is driven in part by the change from the ATP-bound αEβE state to the more closed αTβT state resulting from a zipper motion of αEβE that occurs upon binding ATP28. Our structural analysis indicates that the ATP bound to αTβT is hydrolyzed as a result of the conformational change from αTβT to αDβD which occurs just after the 80° rotation angle (81° to 83°) (Fig. 5). This subsequently leads to the conformational change of αTβT to αDβDHO that occurs during the further rotation to 120° (Supplementary Fig. 10d). The conformational change from αTβT to αDβDHO is likely to occur spontaneously since it involves ATP hydrolysis, an exergonic reaction. The full 120° rotation of the γ-subunit, coupled with spontaneous reactions occurring at both αEβE with ATP and αTβT with ATP, reduces the affinity for ADP and Pi at αDβDHO, resulting in the release of ADP during the first 80° rotation step and Pi during the following 40° rotation step.

Discussion

The proposed mechanism of rotation of FoF1 in this study differs considerably from the model proposed by previous cryo-EM structural snapshots of the F1-ATPase, where the 40° rotation is driven by hydrolysis of ATP accompanied by simultaneous release of ADP and Pi22. This difference may be due to the use of a mutant F1-ATPase (β/E190D) that significantly slows ATP hydrolysis at αDβD. As a result, only the 80° structure with αDβD bound to ATP was identified in the previous study, which would have led to the conclusion that the 40° step was driven by ATP hydrolysis.

This study provides insight into the structural changes occurring in the F1 domain during ATP hydrolysis. In the proton-driven ATP synthesis reaction, ADP is used as a substrate and is bound to the αEβE instead of ATP. The structural changes observed during ATP hydrolysis are the reverse of those that occur during ATP synthesis. Moreover, our findings shed light on the reason why ATP synthases prefer ADP as a substrate. The obtained structure reveals that Pi can bind to the αEβE with or without nucleotide binding, and that this Pi binding can prevent the closing of the ATP-bound αEβE. As a result, ATP binding to αEβE is inhibited, with ADP preferentially binding.

In addition, our study demonstrates that the FoF1 ATP synthase uses the chemical energy of ATP hydrolysis to drive the 80° rotation of the γ subunit and this causes the internal molecular strain which drives the last 40° rotation step, resulting in minimal heat dissipation of the chemical energy. As a result, the coupling of the chemo-mechanical cycle in F1 domain is achieved at almost 100 % efficiency, as previously demonstrated for F1-ATPase by single molecule observation experiments2,18,19.

Under high ATP conditions, the presence of ADP resulting from ATP hydrolysis suggests the potential existence of an ADP-inhibited structure in the reaction solution. Indeed, the addition of Lauryldimethylamine oxide (LDAO), which is known to enhance the activity of F1-ATPase in the ADP-inhibited state25, significantly increased the ATPase activity of FoF1 used in this experiment (Supplementary Fig. 1d). This suggests that ADP-inhibited structure of FoF1 is present in the reaction mixture. However, the structure identified as the ADP-inhibited state have not been obtained in this study. Previous studies have suggested that the dissociation of Pi, resulting from ATP hydrolysis, occurs before the formation of the ADP-inhibited structure29. However, in the structures obtained in this study, ADP was found to bind in the presence of Pi in all catalytic sites, indicating that these structures represent rotational intermediates rather than a dead-end structure, such as the ADP-inhibited structure.

One possible reason for the inability to isolate structures definitively identified as ADP-inhibited conformations is that the ADP-inhibited structures and the rotational state conformations closely resemble each other, making it challenging to classify them as distinct classes. The specific characteristics of the ADP-inhibited structure remain unclear. Further investigations using conditions that induce ADP inhibition or conducting snapshot analysis of mutant variants susceptible to ADP inhibition are necessary to identify the ADP-inhibited structure.

Methods

Sample preparation

For ΔεCT-FoF1 ATP synthase from G. stearothermophilus, we deleted the C-terminal domain (83–133 amino acids) of ε from the expression vector PTR19-ASDS27. The expression plasmid was transformed into E. coli strain DK8 in which the endogenous ATP synthase was deleted. Transformed E. coli cells were cultured in 2xYT medium for 24 h. Cultured cells were collected by centrifugation at 5000 x g and suspended in lysis buffer (50 mM Tris-Cl pH 8.0, 5 mM MgCl2, and 10% [w/v] glycerol). The cells were disrupted by sonication using Branson Sonifier. Membrane fraction was collected by centrifugation at 85,000xg for 30 min, then were resuspended in solubilization buffer (50 mM Tris-Cl, pH 8.0, 5 mM MgCl2, 10% [w/v] glycerol, and 2% n-dodecyl-D-maltoside [DDM]). The suspension was mixed for 1 hr at room temperature, then ultracentrifuged at 85,000 x g. The enzyme in the supernatant was affinity purified using a nickel-nitrilotriacetic acid (Ni2+-NTA) column. For protein samples used under high ATP conditions, endogenous nucleotides were removed from ΔεCT-FoF1 by dialysis with phosphate EDTA buffer containing 200 mM sodium phosphate, pH 8.0, 10 mM EDTA and 0.03% DDM at 25 °C for 24 h. Samples were concentrated to ~500 µl by an Amicon ultra (100 k cut), and then subjected to gel permeation chromatography using a SuperoseTM 6 Increase column equilibrated with 20 mM Tris-Cl, pH 8.0, 150 mM NaCl, 0.03 %DDM. Peak fractions containing FoF1 were collected and concentrated to ~300 μl using an Amicon ultra (100 k cut) for grid preparation.

Measurement of ATPase activity

The ATPase activity of purified FoF1 was assessed at 25 °C using a NADH-coupled assay30. The assay mixtures contained 50 mM Tris-HCl (pH 8.0), 100 mM KCl, 5 mM MgCl2, 2.5 mM phosphoenolpyruvate (PEP), 100 μg/ml pyruvate kinase (PK), 100 μg/ml lactate dehydrogenase, and 0.2 mM NADH, and a range of concentrations of ATP-Mg. The reaction was initiated by adding 2 pmol of FoF1 to 2 ml of the reaction mixture. ATPase activity of FoF1 was measured by monitoring NADH oxidation over time by absorbance at 340 nm.

CryoEM grid preparation

Holey Quantifoil R1.2/1.3 Mo grids were used for high [ATP] conditions and UltraAuFoil R1.2 and 1.3 grid for low [ATP] conditions were used. Before using holly grids, they were treated to 1-min glow discharge by the Ion Bombarder (Vacuum Device).

For high [ATP] conditions, 3 µl of 200 mM ATP-Mg was added to 20 µl of 10 µM of enzyme solution, and then the solutions mixed well by pipetting. The mixtures containing 26 mM ATP-Mg, 9 µM enzyme, 17 mM Tris-Cl, pH 8.0, 130 mM NaCl, ~0.03% DDM were incubated for 20 sec at 25 °C. Then, 3 μL enzyme solution was loaded onto the grid and blotted for 6 sec with a blot force of 10, drain time of 0.5 sec, and 100% humidity using a FEI Vitrobot (ThermoFisher), followed by flash freezing by liquid ethane. For low [ATP] conditions, 2 μl of reaction buffer containing 0.2 M Tris-Cl pH 8.0, 250 μM ATP, 40 mM PEP, 1 M KCl, 5 mg/ml PK was added to 18 μl of 10 μM enzyme solution, and then mixed well by pipetting. The reaction mixtures containing enzyme were incubated at 25 °C for 60 sec. Aliquots of 3 μL of the reaction mixtures were loaded onto grids and immediately flash frozen as described above. The prepared cryo-grids were stored in liquid nitrogen until use.

Cryo-EM data acquisition

Cryo-EM imaging was performed using a Titan Krios (FEI/Thermo Fisher Scientific) operating at 300 kV acceleration voltage and equipped with a K3 electron detector (Gatan) in electron counting mode (CDS). SerialEM software was used for data collection. CryoEM movies were collected at a nominal magnification of 88,000 with a pixel size of 0.88 Å/pix. The defocus range was 0.8-2.0 μm, and data were collected at 50 frames. The total electron dose was 60 electrons/Å2.

Image processing

The detailed procedures for single-particle analysis are summarized in Supplementary Figs. 2 and 3. RELION 4.031 and CryoSPARC v3.332 were used for image analysis. The conversion of file formats between RELION and CryoSPARC was executed using csparc2star.py in pyem. For both high ATP (5922 movies) and low ATP (8200 movies) conditions, beam-induced drift was corrected using MotionCor233, and CTF estimation was performed using CTFFIND4.134. Particle picking was performed by Topaz35. In each dataset, good particles were sorted by 2D classification after template picking from hundreds of images, and 2000-6000 particles were used to train the Topaz model. 1,118,093 high [ATP] particles were and 2,654,860 low [ATP] particles were picked by Topaz and then these were subjected to 2D classification by CryoSPARC, which further selects particles of good quality. Heterogeneous refinement was used to further select particles and classify into different rotation states. The total number of particles selected and classified was 241,668 at high [ATP] and 557,503 at low [ATP] conditions. All particles were combined and a focused 3D refinement was performed on the F1 domain. The F1 structures with different rotation angles were classified by 3D classification without alignment with masks for the F1 domain. Both CTF refinement and Bayesian polishing were carried out multiple times. Finally, we obtained structures with multiple rotational states at 2.6-4.2 Å resolution, estimated by the gold standard Fourier shell correlation (FSC) = 0.143 criterion (Supplementary Figs. 2c and 3c).

Model building and refinement

The atomic model was built using the previous cryo-EM structure of bacterial FoF1 PDB 6N2Y. The βDP has a very different conformation in our structures, so F1 excluding βDP was fitted as a rigid body and βTP was fitted to the βDP density. The initial model was manually modified using COOT and ISOLDE36. The manually modified model was refined by the phenix.real_space_refinement program37. Manual corrections and refinement were iterated until model parameters were improved. The final model was evaluated by MolProbity38 and EMRinger39. The statistical data and data corrections for all structures have been summarized in Supplementary Table 2A–D.

Reporting summary

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