Research Article

Structure and conformational plasticity of the intact Thermus thermophilus V/A-type ATPase

See allHide authors and affiliations

Science  23 Aug 2019:
Vol. 365, Issue 6455, eaaw9144
DOI: 10.1126/science.aaw9144

Innovations in an ATPase/ATP synthase

Enzymes that couple the chemical energy of adenosine triphosphate (ATP) to movement of ions across a membrane are present in all domains of life. Like their F-type cousins in mitochondria, chloroplasts, and most bacteria, vacuolar/archaeal (V/A-type) ATPases couple synthesis or hydrolysis of ATP to movement of protons across the membrane. To uncover mechanistic differences in energy coupling between F- and V/A-type enzymes, Zhou and Sazanov determined structures of a V/A-type ATP synthase from the bacterium Thermus thermophilus. With structures of multiple substates visible, the domain interfaces are made clear and a role for the elastic peripheral stalks is apparent in coupling rotational energy from Vo into the ATP-synthesizing V1 domain.

Science, this issue p. eaaw9144

Structured Abstract


The proton-translocating adenosine triphosphatase (ATPase)/ATP synthase family includes the F-type found in bacteria, mitochondria, and chloroplasts; the V-type (vacuolar) found in eukaryotic intracellular membrane compartments, and the A-type (archaeal) found in the plasma membrane of archaea and some eubacteria. These essential molecular machines share overall gross building plans and a rotary-catalysis mechanism. F-type and A-type enzymes function as proton motive force–driven ATP synthases, whereas V-type enzymes work as ATP-dependent proton pumps acidifying luminal environments. ATPases consist of a soluble F1/V1 domain synthesizing/hydrolyzing ATP and a membrane-embedded Fo/Vo domain translocating protons across membrane. V- and A-ATPases are similar structurally and differ from the F-ATPases by having additional peripheral stalks and connecting subunits between V1 and Vo.

Some eubacteria laterally acquired a V/A-type enzyme (including Thermus thermophilus, ThV1Vo). Rotation of the ThV1Vo membrane-embedded c12 ring relative to the stator subunit a is driven by sequential protonation and deprotonation of a conserved c subunit glutamate. Torque is transmitted via the connecting subunit d to the central stalk (DF shaft) rotating inside the V1 domain, resulting in ATP synthesis due to sequential conformational changes in subunits A3B3. The A3B3 domain itself is prevented from rotation by its connection to the two peripheral stalks.


High-resolution structures of the intact ATP synthases are known for F-type enzymes, whereas there are only models of the isolated V1 and Vo domains for V-type ATPases. The details of V1-Vo coupling are thus not understood. We activated ThV1Vo by thermal cycling and reconstituted it into lipid nanodiscs for structural characterization by cryo–electron microscopy (cryo-EM).


We determined structures of ThV1Vo at an overall resolution of up to 3.5 Å in three main rotational states and two substates of the lowest energy state. Global conformational plasticity of ThV1Vo, as signified by asymmetric V1 precession among rotational states, is a result of mechanical competition between rotation of the bent central rotor and stiffness of the stator. One adenosine diphosphate (ADP) is observed at the ABclosed interface, with the ABsemi interface void of nucleotide, indicating that the activation process removed one inhibitory ADP molecule. An additional putative inhibitory ADP′ molecule is found between the ABclosed and D subunits. ThV1-Vo coupling is achieved via close structural match between the DF shaft and subunit d, as well as electrostatic interactions between subunit d and the c12 ring. Visualization of the proton path in Ao reveals that A-ATPases lack one of the two key glutamates found in proton access half-channels of subunit a in eukaryotic V-ATPases.


Our structure of the intact V/A-type enzyme reveals how Vo and V1 domains are coupled and serves as a reference for the entire V-type ATPase family. The additional peripheral stalk in V/A-type ATP synthase helps to control the stator subunit a position when coupling c-ring rotation to one-off 120° rotation in V1. This is in contrast to mammalian F-ATP synthase with a single peripheral stalk, where F1 rotation is divided into several smaller substeps. Threefold symmetry in the c12 ring is a feature of ThV1Vo, which necessitates nonsynchronicity between the ThV1 and Vo dwell positions to avoid local low-energy minima. Our resolved substates demonstrate the central stalk flexibilities required to accommodate such rotational mismatch. This structure fills in the gap in the ATPase evolutionary tree.

Cryo-EM structure of the T. thermophilus V/A-type ATP synthase.

Left: Proton motive force drives the central shaft rotation, which is coupled to conformational changes and ATP synthesis in the V1 domain. V1 also undergoes a translational “precession” movement inducing bending of the central shaft and two peripheral stalks. Right: (Top) Hallmark features of the V1 ATPase domain; (bottom) mechanism of proton translocation through the membrane-embedded Vo domain.


V (vacuolar)/A (archaeal)-type adenosine triphosphatases (ATPases), found in archaea and eubacteria, couple ATP hydrolysis or synthesis to proton translocation across the plasma membrane using the rotary-catalysis mechanism. They belong to the V-type ATPase family, which differs from the mitochondrial/chloroplast F-type ATP synthases in overall architecture. We solved cryo–electron microscopy structures of the intact Thermus thermophilus V/A-ATPase, reconstituted into lipid nanodiscs, in three rotational states and two substates. These structures indicate substantial flexibility between V1 and Vo in a working enzyme, which results from mechanical competition between central shaft rotation and resistance from the peripheral stalks. We also describe details of adenosine diphosphate inhibition release, V1-Vo torque transmission, and proton translocation, which are relevant for the entire V-type ATPase family.

The adenosine triphosphate (ATP) synthase–adenosine triphosphatase (ATPase) family comprises membrane-bound protein complexes responsible either for ATP synthesis using a cross-membrane proton motive force [F-type and A (archaeal)-type], or for establishing proton motive force using the energy released from ATP hydrolysis [V (vacuolar)-type] (1). These molecular machines operate through a rotary-catalysis mechanism and share similar gross building plans but not the detailed architecture (Fig. 1C). Within this family, F-type enzymes produce ATP in bacteria, chloroplasts, and mitochondria, whereas V-ATPases are ubiquitously found in membranes of eukaryotic intracellular compartments and work as ATP-dependent proton pumps acidifying luminal environments (2). V-ATPases probably evolved from the A-ATPases originally found in archaeal plasma membranes (3, 4). A-ATPases work mostly as H+/Na+-dependent ATP synthases (5, 6) except for some strictly fermentative hyperthermophiles (3, 7). Some eubacteria acquired this enzyme by horizontal gene transfer and use it, depending on respective physiological needs, mostly as an ATP synthase or, in the case of anaerobic Enterococcus hirae, as an ATPase/ion pump (8, 9).

Fig. 1 Overall map and model of intact ThV1Vo in low-energy rotational state 1.

(A) Side and (B) top views color-coded as subunit A (green, cyan, and magenta), subunit B (yellow, salmon, and light gray), subunit D (slate), subunit F (orange), subunit E (teal and sand), subunit G (lime and hot pink), subunit d (pale yellow), subunit a (dark gray), and c subunits (brick). Peripheral stalks EG1 and EG2, as well as the asol and amem domains of subunit a, are indicated. In (B), catalytic ADP and additional ADP′ are shown as spheres. (C) Illustration of subunit composition of different types of ATPases and ATP synthases, color-coded as in (A).

Structurally, F-type ATP synthases consist of a soluble F1 domain that synthesizes or hydrolyzes ATP and a membrane-embedded Fo domain that translocates protons across membrane. F1 and Fo are connected by a central stalk rotating inside F1 and a stationary peripheral stalk. During ATP synthesis, proton motive force–driven rotation of the c ring in Fo is transmitted by the central stalk to power conformational changes in the F1, resulting in the synthesis of one ATP molecule per 120° rotation (as F1 is threefold symmetric, Fig. 1C). Subunit a next to the c ring is kept still during the process by attachment to the F1 domain by the peripheral stalk, forming the stator part of the enzyme. V- and A-ATPases have a similar V1/Vo arrangement but differ structurally from the better-studied F-type, most notably by having one (A-type) or two (V-type) additional peripheral stalks, as well as additional connecting subunits in the central stalk and hydrophilic “collar” domain of subunit a (Fig. 1C) (1012). Because of the similar architecture of A- and V-type enzymes, A-ATPases are also termed V/A-ATPases.

The eubacterium Thermus thermophilus possesses V/A-ATPase (ThV1Vo), which works as an ATP synthase (13, 14). This is an ~680-kDa complex consisting of nine subunit types in a stoichiometry of A3B3DFE2G2dac12, with uppercase letters denoting components of the soluble V1/A1 domain and lowercase letters denoting components of the membrane-embedded Vo/Ao domain (Fig. 1A) (10).

To date, complete atomic models for the intact ATPases are known only for the F-type (1517). For the V-type, structures of separate prokaryotic V1 (1820) and yeast Vo (21, 22) revealed details of the catalytic interface and proton pathway. However, full mechanistic interpretation of global rotational motion in V1Vo is limited by the low resolution (at best 5.0 Å) of the intact enzymes studied (1012, 23). The detailed mechanisms differ substantially between these two enzyme families; F-ATPases perform one catalytic 120° rotation in F1 in three smaller substeps (2428), whereas V-ATPases perform such rotation in one turn without substeps (19, 20, 29). This implies that V1-Vo coupling would require much higher flexibility of the central and peripheral stalks than in F1-Fo to link, over one full revolution, three large steps in V1 to ~12 smaller, per-c subunit steps in Vo. How this is achieved is not known. Furthermore, ThV1 purifies in an adenosine diphosphate (ADP)–inhibited state with complete loss of ATPase activity (14, 30). Existing ThV1 crystal structures were obtained in such a state (18, 31), whereas structural details of hydrolytically active ThV1 are not yet available. For V/A-ATPases, knowledge about Ao is especially limited, with no structures available. To resolve these issues, we performed structural studies on the entire active ThV1Vo. We determined the structure of its ground state at ~3.5-Å resolution, together with structures of the other two main rotational states and two substates. These structures provide insights into the mechanism of V/A- and V-ATPases in general, including overall flexibility, torque transmission, rotor–stator competition, ADP inhibition–reactivation, and proton translocation.

Overall structure of ThV1Vo

We purified native ThV1Vo from T. thermophilus HB8 and activated the enzyme by EDTA/thermal cycles to partially release autoinhibiting ADP (14). The average ATP hydrolytic activity after such a treatment was consistent with previous measurements (table S4) (13, 14). We reconstituted the enzyme into nanodiscs of the membrane scaffold protein MSP1E3D1 and POPC lipids to approximate native bilayer conditions, which resulted in an active enzyme (materials and methods and table S4). Cryo–electron microscopy (cryo-EM) data of the nanodisc preparation resulted in higher-resolution maps compared with the digitonin-dispersed preparation (~7 Å for state 1; not shown). A nanodisc density is clearly visible as a less ordered ring tightly wrapping around the hydrophobic Vo surface, and its circumference (~400 Å) is consistent with MSP1E3D1 length (fig. S2B).

Three-dimensional (3D) classification (fig. S1) resulted in three classic rotational states (state 1, 31% of particles, 3.5-Å overall resolution; state 2, 21%, 3.9 Å; and state 3, 19%, 4.1 Å) with respective rotor (subunit D with F attached, i.e., DF shaft) positions ~120° apart from each other (Fig. 2A and movie S1). Focus-refined maps using V1 (3.3-Å resolution), Vo (3.9 Å), and other customized (to better resolve peripheral stalks) masks were individually carved and combined to give a near-atomic resolution map of state 1 (fig. S2A). This map allowed us to build (see supplementary text) a complete atomic model of ThV1Vo (Fig. 1A, fig. S5, and tables S1 and S2), for which only polyalanine cryo-EM models were available previously (10, 23). Additionally, we identified two substates of state 1 (1L and 1R) with fixed Vo position but with V1 twisting as a whole either to the left or right by 8° to 10° (movie S2). Focused maps resulted in atomic V1 models combined with backbone Vo models for all four additional states (states 2, 3, 1L, and 1R; fig. S1 and table S1). In V1, V/A-type A and B subunits share a ~25% sequence identity with the catalytic β and noncatalytic α subunits of F1, respectively. Our ground-state (state 1) model exhibits the hallmark features of three conformationally different AB heterodimers named ABclosed, ABopen, and ABsemi-closed (hereafter, ABsemi) (Fig. 1B). Importantly, only one catalytic ADP is found at the ABclosed interface, whereas the ABsemi interface is free of nucleotide, which is in contrast to previous low-resolution V1 structures of autoinhibited enzyme and is likely due to the activation process during our sample preparation.

Fig. 2 Different rotational states and important interfaces of ThV1Vo.

(A) Symmetry axes of the c12 ring (solid line) and the A3B3 domain (dashed line) are plotted for the three main rotational states. (B) Top view of subunits ac12 in the three main states (1 to 3: green, cyan, and magenta, respectively) aligned by amem. Lines connecting the symmetry center of the 12 c_Glu63 residues to c(O)_Glu63 (spheres) in three states are plotted as yellow dashes, defining rotational angles. (C) In the transition from state 1 to 2 (left), V1 moves horizontally along a screw axis (cyan rod), combined with a twist. From state 1 to 1L (right), V1 rotates around a screw axis overlapping with the central stalk. Pivot points for peripheral stalks coiled-coil bending and globular-domain twisting are circled. Twisting axes of the globular domains relative to the attached subunit B β barrel are indicated as small red rods. (D) EG1 (left) coiled coil (green and teal) is tightly sandwiched against the β sheet in the proximal α/β domain (gray), whereas an additional helix (starred) in the asol distal α/β domain excludes tight interaction of its β sheet with EG2 (pink and sand) (right). (E) Alignments of central stalks in three rotational states show flexibility in top coiled coil for ThV1Vo but not for Bacillus or spinach chloroplast F-ATP synthases.

Global conformational plasticity of ThV1Vo

A fundamental issue in the current rotary-catalysis model of ATP synthesis or hydrolysis is that the ~120° rotation of central DF shaft cannot be readily coupled to the much smaller per-proton rotation of the c ring without twisting the rotor (10). Most ATPases possess 8, 10, or 14 c subunits without threefold symmetry in the c ring (1517, 21, 22, 3235), which may be a feature to prevent low-energy minima during rotation relative to threefold symmetric F1/V1. This is clearly not an absolute requirement, as the T. thermophilus c12 ring (36), Mycobacterium phlei c9 ring (37), and Spirulina platensis c15 ring (38) all have threefold symmetry. However, in single-molecule studies of intact ThV1Vo, 120° dwell positions (temporary stops) in V1 were often found to locate between two adjacent 30° dwell positions in Vo, rather than coincide with them (fig. S7C) (29). Therefore, low-energy minima of coinciding V1 and Vo dwell positions seem to be avoided, requiring the flexibility of the central stalk. In addition, unlike bovine F1, which dwells three times at 0°, 65°, and 90° during one 120° rotation (2428), V/A-ATPases of E. hirae and T. thermophilus demonstrate a one-off 120° rotation with only one 0° dwell position where all catalytic events happen (19, 20, 29). This further increases demands on the flexibility of the stalks, as the elastic energy must be accumulated in the twisted stalks during four consecutive 30° steps of Vo before being used to drive a single 120° rotation of V1, with the synthesis of one ATP molecule.

In our structures, almost precise 120° rotation of the c12 ring was measured for the three main rotational states 1 to 3 (Fig. 2B). This was likely due to the threefold symmetry in the ThV1Vo c12 ring, in contrast to the unequal rotations of spinach chloroplast F-ATPase with the c14 ring (15) and yeast V-ATPase with the c8c′c′′ ring (12). As the subunit D rotor, highly curved at the conserved Pro170 (fig. S8A), sequentially drives the three AB heterodimers to alter conformations of their catalytic sites (movie S1), the hydrophilic head of V1 is forced to tilt from a vertical position in different directions, a process known as “precession” (11, 15, 32). Precession angles, estimated as the angle between the V1 and Vo symmetry axes, indicate that V1 tilt is substantial but roughly similar for three rotational states (Fig. 2A). Therefore, rotation of the curved central stalk inside A3B3 causes V1 to wobble in space, which is restricted by clamping forces of peripheral stalks directly linked to the stator in membrane. Apparent ThV1 precession is thus driven by competition between the central and peripheral stalks, resulting in conformational deformations of all stalks in the process.

The motion trajectory of V1 between any two states is a combination of shift and rotation, and is dominated by a translational movement in the state 1 to 2 transition and by an off-axis rotation in the state 2 to 3, 3 to 1, and 1 to 1R transitions (Fig. 2C and fig. S7A). In the state 1 to 1L transition, the axis of such rotation overlaps with the central stalk, indicating that this V1 motion is a catalytic-like rotation in the ATP synthesis direction for ~8.0°, whereas the Vo position is fixed (Fig. 2C and movie S2). Such rotation due to a pure twist of the central stalk likely reflects its innate flexibility. When the enzyme is at work, the elastic energy stored in the twisted central stalk is used in torque transmission from the c12 ring to V1, along with the energy stored in the twists of the peripheral stalks. The state 1 to 1R transition is an ~10° off-axis twist in the ATP-hydrolysis direction (fig. S7, A and B, and movie S2), and both substates probably reflect different compensatory interactions between the central and peripheral stalks (supplementary text). The existence of substates to major states is probably a common feature of ATPases and has been reported in low-resolution studies of bovine mitochondrial F-ATP synthase (32). These substates may also be present in other rotational states of ThV1Vo, but could not be resolved in the current study because of the lower particle numbers in these states.

Elasticity of the stalks

Elasticity also exists in the two peripheral stalks as their coiled coils are extensively bent during V1 precession, pivoting around points where they are attached to subunit a (Fig. 2C). This is accompanied by the twisting of peripheral stalks in the area where they are attached to the V1 subunit B (Fig. 2C and fig. S7D), allowing V1 precession while restricting its rotation to avoid enzyme “slipping.” This in turn causes considerable flexing of the top part of the DF shaft during rotation, in contrast to the completely rigid rotor of chloroplast or bacterial F-ATP synthase (Fig. 2E) (15, 17). Mitochondrial enzyme also shows flexibility of the central stalk (32), which is likely due to the elaborated rigid structure of its peripheral stalk. For A- and V-types, the additional one or two peripheral stalks further restrict the central rotor, leading to its deformation in common with the mitochondrial enzyme (10, 11).

Our structures reveal a pronounced continuous antiparallel β sheet formed between the C-terminal (CT) globular domain of subunit E and the N-terminal (NT) β barrel of subunit B. This leads to a strong attachment of each peripheral stalk to the A3B3 domain by many hydrogen bonds between the two complementary β strands (Fig. 1A and fig. S7D). This β sheet serves as a pivoting hinge and is especially suited to accommodate complex relative motions in this area because it can be promiscuously twisted in different directions during V1 precession (fig. S7D). In the coiled-coil regions of peripheral stalks, bending is facilitated at the subunit F level as the two helices become easier to bend radially in an azimuthal arrangement (fig. S7E) (39).

ThV1Vo (10) and yeast V-ATPase (11) share the common feature that NTs of peripheral stalks are attached to the proximal and distal α/β domains flanking the horizontal coiled coil in the soluble part of subunit a (asol) rather than being embedded into the membrane as in the F-type enzymes (15, 16). The two peripheral stalks of ThV1Vo interact with asol differently. EG1 interacts with the open face of the proximal asol β sheet by extensive matching hydrophobic surfaces, whereas EG2 appears to be excluded from the distal asol β sheet by the subunit a helix 3 (a_H3) CT and interacts more loosely with loops at the edge (Fig. 2D). Because the distal α/β domain is not restricted by linkage to the fixed membrane part of subunit a (amem) as is the proximal α/β domain (Fig. 1A and fig. S6F), its connected peripheral stalk EG2 is generally more mobile than EG1, as seen in our structures. Such flexible coupling between peripheral stalks and subunit a helps to absorb V1 precession–induced twists and shifts, keeping them from being transmitted to amem and affecting its “stator” position in the membrane. Large and complex V1 precession movements are found in transitions between the five states of ThV1Vo resolved in this study (supplementary text). They allow accumulation and release of elastic energy stored in the twists of central and peripheral stalks to couple 120° steps in V1 to 30° steps in Vo.

A/B heterodimers and catalytic interfaces in V1

Each A/B subunit comprises an NT β-barrel domain, a central nucleotide-binding α/β domain, and a CT helical domain (fig. S6, A and B). An additional β-stranded bulge domain on the A subunit (fig. S6A) may prevent accidental peripheral stalk binding. Major conformational changes among the ABclosed, ABopen, and ABsemi heterodimers are found in their CT helical domains (Fig. 3B), as the subunit D coiled coil sequentially pushes against helix H13 of A/B subunits during its rotation (Fig. 3A). The force of the interaction is evident from the highly bent Aopen H13 CT tip compared with the other A subunits (Fig. 3A). Movements of the A/B CT helical domains relay to the central α/β domains to make the catalytic interface much narrower in ABclosed than in ABopen.

Fig. 3 Different AB heterodimer conformations and catalytic interfaces in V1.

(A) Top view of subunit D interactions with CT helical domains of different A/B subunits (transparent surfaces), with H13 (cartoon) and catalytic ADP (spheres) accentuated. (B) Side views of A/B subunits in different conformations aligned by their β-barrel domains (circled) to show substantial relocation of Aopen/Bclosed CT helical domains (black arrow). (C) ABsemi interfaces of the ADP-inhibited (blue, PDB 5y5y) and activated (green, our state 1) enzymes, with the ADP and key residues stopping subunit D rotation (dashed arrow) due to steric clashes (red star) shown as sticks. (D) Catalytic interface between Aclosed (green) and Bclosed (light gray) subunits with key residues (sticks) and ADP (salmon), Mg2+ (magenta sphere), and all polar interactions (gray dashes) indicated. (E) Left: Binding position of additional ADP′ molecule (spheres) in V1 shown along with the catalytic ADP at the ABclosed interface. Right: State 2 (blue) ADP′ (sphere), coordinated by a group of residues (sticks), would clash (red star) with bulky residues (spheres) if subunit D rotates (dashed arrow) to its position in state 3 (orange).

In the current structure of the active enzyme, clear ADP density is only found at the ABclosed catalytic interface (fig. S5), which presumably represents a state after the ATP hydrolysis and Pi release, but before the ADP release and new ATP binding (19, 20). The remaining ADP represents the leaving group after nucleophilic attack on the ATP γ-phosphate by a water molecule, as suggested for the bovine F1 structures (2427). It is coordinated by Aclosed Phe419, the “arginine finger” Bclosed Arg360, and an Mg2+ near the “P-loop” Aclosed Thr235 (Fig. 3D). Aclosed Lys234 and Arg258 likely position the now-gone γ-phosphate of ATP to receive the nucleophilic attack, whereas Aclosed Glu257, Arg329, and Ser385 likely position a water molecule (not resolved) as the nucleophile (Fig. 3D).

No nucleotide is found at the ABsemi interface, in contrast to the previous autoinhibited structures with ADP bound at both ABclosed and ABsemi interfaces (PDB 5y5y) (10). Naturally, side chains of above key residues come closer to the proposed ADP-binding site in our nucleotide-free ABsemi interface, but without major backbone rearrangement (Fig. 3C). Notably, the short horizontal Bsemi H10 is slightly bent in its CT because of the presence of ADP diphosphate in the autoinhibited structure (10). This leads to flipping of Bsemi His323 at its NT from an upright position in the active enzyme to a horizontal position that would clash with nearby D_Gln181 and stop subunit D rotation (Fig. 3C). Therefore, ADP entrapped at the ABclosed interface cannot be expelled by the central stalk (40) and the whole autoinhibited enzyme is arrested without further ATP hydrolysis possible.

Density for a possible additional ADP′ molecule (ADP prime, in contrast to catalytic ADP, see materials and methods) is also found in all three states (Fig. 3E and fig. S5) inside the doughnut-shaped cavity formed by the six A/B β-barrel domains and the subunit D CT helical tip. This ADP′ molecule interacts mostly with Aclosed Lys8, Bclosed Glu49, and Bclosed Arg274 (Fig. 3E). It is not on the entry or exit route of the catalytic nucleotide, but at a position similar to that of the bovine F1 inhibitor resveratrol (41). It is plausible that this embedded ADP′ might stabilize the autoinhibited state of ThV1Vo by sterically hindering subunit D rotation (Fig. 3E) (42) when environmental ADP concentration is elevated so that both catalytic ADP molecules are bound. As our preparation can hydrolyze ATP, evidently any such blockage can be overcome when the ABsemi interface is empty (as in our structure) or ATP bound (as under turnover). Therefore, the ADP′ site is a good candidate for an additional, possibly T. thermophilus–specific, site of reversible ADP inhibition.

Vo-V1 coupling and torque transmission

Unlike in F-ATP synthases, the central rotor of the ThV1Vo is composed of three segments: the DF shaft, the c ring, and the additional subunit d that connects the two (Fig. 4A) (10). Previous studies suggested a “screwdriver” mechanism of ThV1-Vo coupling, in which the DF-d association is of low affinity and its close interaction is enforced by external clamping forces from the two peripheral stalks (43). In our improved structure of this region, a compact rigid domain is formed by the bottom part of the subunit D coiled coil, subunit D short helix, and H1 helix of subunit F (Fig. 4B and fig. S6, C and D). It serves as the “screwdriver head” tightly plugged into the concave side of the d subunit by clamping peripheral stalks (Fig. 4A). This rigid domain sterically bends and breaks a long d helix into two shorter d_H7 and d_H8 while displacing d_H10-12 (Fig. 4B and fig. S6E) compared with the isolated d structure (PDB 1r5z) (44). Torque transmission is thus ensured by a close match between the interacting surfaces of the DF shaft and subunit d, aided by several specific hydrogen bonds, including that from the critical D_Glu74 (Fig. 4C) (43).

Fig. 4 V1-Vo coupling in the central rotor.

(A) Side view of the D-d-c12 assembly with the front six c subunits removed for clarity. Subunit F helix H5 might clash with the Bsemi loop (dashed arrows) as the central rotor rotates. The c12 ring upper and lower tunnels are separated by the circle of c_Arg36. (B) Subunit F fills the space between Asemi and subunit d (pale yellow), thus helping to conduct the pressure imposed by peripheral stalks at the V1-Vo interface. Several helices are displaced, bent, or broken by this pressure compared with the isolated subunit d structure (PDB 1r5z, pink). (C) Hydrogen-bonding (gray dashes) residues (sticks) responsible for the D-d association. (D) Complementary charges of subunit d and c12 ring illustrated by their surface electrostatic potentials.

Previous studies indicated that subunit F CT H5 plays a role in maintaining the correct rotational direction of the rotor (45), whereas its NT globular domain (fig. S6D) may be important in reinforcing torque transmission between V1 and Vo (46). Our structures (movie S1 and Fig. 4A) show that during DF shaft rotation, F_H5 may approach closely the loop region after H13 in the Bsemi and possibly also the Aopen subunit. Any such “clash” is probably a useful interaction, helping to couple DF rotation and A3B3 conformational changes. Because F_H5 is tilted from a vertical position, in vivo, it may be one of the factors favoring ATP synthesis rather than hydrolysis in ThV1Vo. Moreover, the compact NT domain of subunit F reinforces the rigidity of the screwdriver head part of subunit D by wrapping around it. This domain is tightly stuck in the space between Asemi and D short helix, so it would help in pushing the D short helix into subunit d when peripheral stalks clamp V1 onto Vo (Fig. 4B). It appears that subunit F works both as a ratchet and a pressure conductor to ensure efficient torque transmission between V1 and Vo.

The d-c12 interface is a distinctive feature of V-ATPases and it has been shown to be tight (43). A ring formed by c_Arg36 is observed inside the c12 ring, marking the narrowest point of the c12 tunnel with a constricted pore diameter of ~9 Å (Fig. 4D). It serves as a positive charge–enriched layer that separates the hydrophobic tunnel into a funnel-shaped upper part and a cylindrical bottom part (Fig. 4A). c_Glu47 forms a ring of negative charges on top of the tunnel, whereas bending of the inner c helices above c_Arg36 is permitted by the absence of side chains at c_Gly38 and c_Gly41 (fig. S9A). The tip of the cone-shaped subunit d sits deeply in this shallow upper tunnel of the c12 ring with striking electrostatic complementarity. A ring of positively charged residues on the subunit d surface forms a network of salt bridges with the c_Glu47 ring, whereas the tip of subunit d is enriched in negatively charged residues interacting with the c_Arg36 ring (Fig. 4D). The short d_H1 plugs into the cleft between the neighboring c subunits, where bulky residues such as d_Val14 and d_Tyr7 sterically prevent d-c12 slipping during rotation (Fig. 4A). Overall, buried surface areas between DF-d and d-c12 are roughly similar, but DF-d interactions are dominated by hydrogen bonds between specific residues and clamping forces by peripheral stalks, whereas d attachment to the c12 ring is dominated by electrostatic interactions. Interactions between d and the asol “collar” are nearly absent, as required for free rotation, although in state 1, a salt bridge between d_Arg38 and a_Glu57 is observed, which may be responsible for the lower energy of state 1 compared with states 2 and 3.

Vo and the proton channels

A “Brownian ratchet” model has long been proposed for proton translocation across biological membranes coupled to ATP synthesis or hydrolysis, which is accomplished by sequential protonation and deprotonation of a conserved c subunit glutamate (c_Glu63 in ThV1Vo) at two half-channels facing the opposite sides of the membrane (1, 21, 22, 33). More recently, the Brownian ratchet model has been modified to include a substantial role of electrostatic interactions at the subunit a to c-ring interface as a driving force for rotation (15, 47). Our structure of the ThV1Vo in state 1 provides an atomic model for the Ao domain. The proton-translocating half-channels, similar to those previously observed in other ATPases, are not fully enclosed here but rather are exposed to the membrane, which we modeled as a POPC bilayer equilibrated around Ao in CHARMM (movie S3) (48). A prominent cytoplasmic half-channel is delineated on one side by a_H12-16 and on the other side by the c12 ring and a few predicted lipids (Fig. 5A). The deep end of this half-channel is delineated by two long, highly tilted helices, a_H15 and a_H16, forming a striking hairpin (Fig. 5A), a distinctive feature of a subunits in all ATPases. The periplasmic half-channel is delineated by the U-turn half of the a_H15-16 hairpin and the c12 ring, as well as the extended loop between a_H11 and a_H12 (Figs. 5A and 6A). Lipid molecules are predicted to fill the void between the a_H10-H11 β-hairpin and a_H16 (Fig. 5B), as well as to plug the cleft between a_H15 and the c12 ring, completing the walls of the periplasmic half-channel (Fig. 5A). Both channels are lined by polar residues from subunit a, many of them conserved (fig. S10), forming proton translocation pathways (Fig. 6A).

Fig. 5 Vo (Ao) structure and two proton half-channels.

(A) Side view of the two proton half-channels (salmon surfaces) with amem helices (cyan cylinders) numbered. The β-hairpin (magenta) in amem interacting with c subunits (yellow) is circled, including also the predicted POPC (green sticks) plug. Area of interaction between a_H12/13 and the c ring is also circled on the left. (B) Rotation of circled part in (A) with interacting residues at a-c12 interface shown in sticks and distances labeled. Note the double-arginine checkpoint. (C) Top view of the ac12 subcomplex with the hydrogen bonds between c_Glu63 and neighboring c_Thr64 shown in black dashes. The c subunits lose this hydrogen bond as they enter hydrophilic half-channels near subunit a. (D) Surface electrostatic potentials of the amem subunit and c12 ring at the a-c12 interface. Note the negatively charged belt on the c12 ring formed by c_Glu63 and the positively charged spot on the amem long tilted helical hairpin formed by the double-arginine checkpoint.

Fig. 6 Mechanism of proton translocation in the ATP synthesis direction.

(A) Side and (C) top view of the proton pathway at the a-c12 interface as indicated by hydrogen bonds (black dashes) linking key residues and predicted water molecules (shown as sticks). (B) and (D) Illustration of proton movement (salmon arrow) during ATP synthesis with key proton donors/receivers, double-arginine checkpoint, and key water molecule (coordinated by a_Thr612) indicated by dots of different colors, with or without the c12 ring shown. Subunit a helices are shown as cyan cylinders.

Interactions at the a-c interface are in a dynamic equilibrium of breaking and rebonding, as the c12 ring rotates against the concave part of the a_H15-H16 hairpin with a matching hydrophobic surface (Fig. 5, B to D). a_Leu393 in the a_H10-H11 β-hairpin contacts c_Arg49 in chain Y of the model [c(Y)_Arg49], in the loop region between c helices (Fig. 5B), whereas the periplasmic termini of a_H12 and a_H13 contact the CT ends of c subunits (Fig. 5A). These interactions must be transient to allow c12-ring rotation, but they still appear to stabilize and control the subunit a to c-ring interface to a higher degree than in F-type ATP synthases (1517). Depending on the direction of the reaction, the key c_Glu63 receives a proton in one half-channel from one side of the membrane, allowing neutral c_Glu63 to enter the lipid bilayer and make a full turn facing the lipids. It then enters another half-channel, where it must be deprotonated so that the proton can be translocated to the other side of the membrane (Fig. 6, B and D). The coupling checkpoint ensuring that c_Glu63 loses its proton before it continues its rotation are key conserved a_Arg563 and Arg622 sitting on a_H15-H16 hairpin right between the two half-channels (Fig. 5B; fig. 6, A and C; and movie S3). There is only one such arginine in F-type enzymes, ensuring that only negatively charged deprotonated glutamate can pass further on by forming a salt bridge with arginine (1517, 33). However, V-type ATPases have a stricter checkpoint with two conserved arginines (21, 22, 35). In ThV1Vo, they form strong salt bridges to c(Z)_Glu63 such that the glutamate cryo-EM density is preserved (Figs. 5B and 6C and fig. S5). Other protonated c_Glu63 residues not facing subunit a adopt a more “inward” conformation stabilized by a hydrogen bond to Thr64 on the neighboring c subunit (Fig. 5C and fig. S9B). The c subunits themselves show not just a kink, but a substantial distortion of the helical structure around c_Glu63 (fig. S9A), which may contribute to the main chain flexibility as needed to receive and donate protons, in addition to c_Glu63 side chain flip. c_Gln34 and c_Thr30 neutralize the exposed backbone oxygens in this area (fig. S9A), but are not conserved in the less-bent c subunits of chloroplast F- or yeast V-type enzymes (15, 16, 33). It is possible that a higher distortion of the helical structure is needed in a thermophile to achieve the required flexibility in this area.

Mechanism of proton translocation

A mechanism of cross-membrane proton translocation can now be proposed for the ThV1Vo. During ATP synthesis, the c_Glu63 rotates into the cytoplasmic half-channel [c(O)_Glu63 in our structure] to be deprotonated (Fig. 6C and movie S3). a_Glu550, a homolog of the immediate proton donor a_Glu721 in yeast V-ATPase (fig. S10), likely receives the proton from c_Glu63 through a bridging water molecule [predicted by the Dowser program (49)]. The proton is then passed to a series of nearby protonatable residues, including a_Lys631, Arg482, His494, and His491, toward the cytoplasm (Fig. 6, A and C). The deprotonated c_Glu63 travels through the water-filled cavity until it encounters the checkpoint of a_Arg563 and a_Arg622. A highly conserved a_Glu627 immediately before the double arginine (fig. S10) could serve as a scavenger to deprotonate any escapist from a_Glu550 and then pass the proton to a_His557 to join the main pathway to the cytosol (Fig. 6, A and C).

This glutamate (a_Glu627) is strictly conserved among V- and A-type ATPases (a_Glu804 in yeast, fig. S10), along with the double arginine, and is possibly also required to compensate the positive charge of an additional arginine. Such a checkpoint is indeed stricter than the single arginine of the F-type, as suggested by the lower pKa (where Ka is the acid dissociation constant) of the bounding c_Glu in our model (~2.9) compared with that in the chloroplast (~6.3, PDB 6fkf) (15), as calculated by PROPKA (50). After passing it, the elevated pKa of 7.5 for c(Y)_Glu63 allows its reprotonation by a nearby water molecule(s) coordinated by c(X)_Thr64, a_Thr612, and the exposed backbone oxygen of a_Leu611 (Fig. 6A). a_His616 and conserved a_Asp365 with a_Glu426 connect these waters to the periplasm (Fig. 6, A and C). Notably, a_Leu611 sits at a substantial kink of a_H16, exposing its backbone in the cavity created by the kink (Fig. 6C). This feature is not conserved in the yeast V-ATPase because its corresponding helix is not as sharply bent. A key a_Glu789 in yeast periplasmic cavity, conserved among eukaryotic V-ATPases, transfers the proton from the c ring to subunit a through a bridging c_Tyr66 (22, 35), but this glutamate is replaced by threonine in eubacterial V/A-ATPases (a_Thr612 in T. thermophilus) or by asparagine in archaeal V/A-ATPases (fig. S10). Therefore, unlike eukaryotic V-ATPases, in which the key protonatable glutamates (Glu721 and Glu789 in yeast) are found on both sides of the double arginine, in A-ATPases, such glutamate is only found in the cytoplasmic cavity (a_Glu550 in ThV1Vo) (fig. S10). Lack of a direct deprotonator in the periplasmic cavity suggests that the proton transfer network in A-ATPases may be more controlled in the ATP synthesis direction, consistent with their physiological function.


Although different types of ATPases and ATP synthases all conform to similar gross building plans, structural and mechanistic details evolve to accommodate changing cellular environments and physiological roles. It has long been postulated that eukaryotic V-ATPase evolved from internalized A-ATP synthase of an archaeal ancestral cell, with subsequent duplication and fusion of c subunits (4, 51, 52). Bacterial V/A-ATPase, laterally acquired from the archaeal A-ATPase (8, 9), could represent an evolutionary bypass that mimics the ATP synthase to ATPase transition, as reflected in the functional reversal and subunit c duplication of V/A-ATPase from the aerobic T. thermophilus to the anaerobic E. hirae. The hallmark double-arginine checkpoint of V-type ATPases is retained in ThV1Vo, even though it works as an ATP synthase, probably because mutation to F-type single arginine would also require simultaneous mutations of nearby glutamate and other residues.

In a broader picture, the different number of peripheral stalks in the eukaryotic F- and V-type ATPases might also correspond to their respective physiological functions. In F-ATP synthases, rotation through the catalytic substeps is initiated in the c ring and the main task of the single peripheral stalk is to prevent the F1 from rotating with the rotor. In V/A-ATPases, more control is needed to keep subunit a still when coupling one-off 120° rotation in the V1 to c-ring rotation, perhaps reflected in the increased subunit a to c-ring interface discussed above. This enhanced V1/Vo mismatch also requires additional flexibility to store elastic energy, reflected in the addition of the second peripheral stalk of V/A-ATPases and large precession movements described here. In eukaryotic V-ATPases, the coupling task is perhaps even more challenging, as 120° steps (if this A-type feature is conserved in eukaryotes) are initiated in V1 and peripheral stalks need to accommodate these jumps while keeping subunit a still and the a to c-ring interface intact. This could be the reason for the evolution of the third peripheral stalk in eukaryotes. Our V/A-ATPase model fills in the gap in the ATPase evolutionary tree. Future structures of other V-type ATPases, such as those of yeast and E. hirae, might help to further elucidate how this enzyme adapts itself structurally to ATP synthesis or hydrolysis.

Materials and methods

Protein purification, nanodisc reconstitution, and ATPase activity assay

T. thermophilus HB8 cells were grown in-house as described previously (53, 54). All of the following steps were performed at room temperature. Pelleted cells (150 g) were resuspended with a homogenizer in 450 ml of lysis buffer (50 mM Tris, 5 mM MgCl2, pH 8.0) containing 0.001% phenylmethylsulfonyl fluoride (PMSF, Carl Roth) and one complete EDTA-free protease inhibitor tablet (Roche), before being lysed by passing through a cell disruptor once at 15 kilopounds per square inch (kpsi) and twice at 30 kpsi. Cell debris was removed by sequential centrifugation at 9874g for 20 min and at 22,217g for 35 min, supernatant from which was further ultracentrifuged at 225,100g for 3.5 hours to pellet the native membranes. Membrane pellet was solubilized in 150 ml of solubilization buffer [50 mM Tris, 5 mM MgCl2, 150 mM NaCl, 1% dodecyl-maltoside (DDM), 10% glycerol, pH 8.0] containing 0.001% PMSF (Carl Roth) and one complete EDTA-free protease inhibitor tablet (Roche) by constant stirring overnight. The insoluble material was removed by ultracentrifugation at 225,100g for 1 hour and the supernatant was passed through a 0.45-μm filter before being loaded onto a Hiload 26/10 Q-Sepharose column (GE Healthcare) pre-equilibrated in buffer A (50 mM Tris, 5 mM MgCl2, 150 mM NaCl, 0.02% DDM, 10% glycerol, pH 8.0). Eluted fractions from a linear gradient of 0 to 100% buffer B (50 mM Tris, 5 mM MgCl2, 250 mM NaCl, 0.02% DDM, 10% glycerol, pH 8.0) over four column volumes were analyzed by reducing SDS–polyacrylamide gel electrophoresis (SDS-PAGE). ATPase-containing fractions were pooled and concentrated on a 100-kDa molecular weight cutoff (MWCO) concentrator. Approximate DDM percentage in the concentrated fractions was estimated assuming that all detergent was retained.

The concentrated ATPase fraction from the Q-Sepharose column was buffer exchanged into activation buffer (100 mM K+/PO43–, 10 mM EDTA, 0.02% DDM, pH 8.0) using a PD MiniTrap- Sephadex G-25 desalting column (GE Healthcare). The sample then went through six thermal cycles of 65°C for 15 min and 4°C for 30 min to release entrapped ADP and to activate the enzyme. Additionally, this treatment disrupted the contaminating T. thermophilus GroEL-GroES chaperonine complex, allowing removal of individual GroEL/ES subunits at the subsequent gel-filtration step. For activity assay of the DDM-solubilized sample, the activated ATPase fractions were concentrated and further purified with a Superdex200 10/300 gl column (GE Healthcare) pre-equilibrated in buffer C (50 mM Tris, 100 mM NaCl, 5 mM MgCl2, 0.02% DDM, pH 8.0). For nanodisc reconstitution, POPC (Sigma-Aldrich) was dissolved in CHAPS–lipid-solubilization buffer (50 mM Tris, 100 mM NaCl, 5 mM MgCl2, 1% CHAPS, pH 7.4) using a single freeze–thaw cycle (55). T. thermophilus total lipids and polar lipids were extracted according to established protocols (56, 57) and then solubilized in DDM–lipid-solubilization buffer (50 mM Tris, 100 mM NaCl, 5 mM MgCl2, 2% DDM, pH 7.4) by vortexing at room temperature. Activated and concentrated ATPase fractions and solubilized MSP1E3D1 (Sigma-Aldrich) were mixed with either solubilized POPC in a 0.25:1:50 molar ratio or with T. thermophilus polar lipids in a 0.25:1:100 molar ratio. The mixtures were incubated for 1 hour at room temperature, before the addition of 600 g/liter washed Bio-Beads SM-2 (Bio-Rad) and further overnight incubation to initiate nanodisc formation. After Bio-Bead removal by centrifugal filtration, the samples were loaded onto the Superdex200 10/300 gl column pre-equilibrated in buffer D (50 mM Tris, 100 mM NaCl, 5 mM MgCl2, pH 8.0). Fractions eluted with the same buffer were analyzed by SDS-PAGE before concentrating with 100-kDa MWCO concentrators to ~6 mg/ml. Both the POPC and the T. thermophilus polar lipid nanodisc samples were immediately used for ATPase activity assays, and the former was also used for cryo grid preparation. Activity in POPC was lower than in T. thermophilus lipids but comparable to that in DDM, indicating that the POPC nanodisc supports ATPase activity (table S4). Samples not immediately used were stored in liquid nitrogen after addition of glycerol to 30%.

The ATP hydrolytic activities of DDM-solubilized or nanodisc-reconstituted samples were measured at 25°, 35°, and 40°C by the addition of 20 μg of treated ATPase to 1 ml of an enzyme-coupled ATP-regenerating system (50 mM Tris, 100 mM KCl, 5 mM MgCl2, 1 mM phosphoenolpyruvate, 50 μg/ml pyruvate kinase, 50 μg/ml lactate dehydrogenase, 0.2 mM NADH, 0.2 mM Mg-ATP, pH 8.0) with or without 0.05% DDM to start the reaction (14). The kinetics of ATP hydrolysis was monitored in real time as the NADH consumption rate, determined by the absorbance decrease at 340 nm in a UV-2600 spectrophotometer (Shimadzu). One unit of activity is defined as 1 μmole of ATP hydrolyzed per minute.

EM grid preparation, data collection, and image processing

For cryo grid preparation, 1% CHAPS (Carl Roth) dissolved in buffer D was added to the ~6 mg/ml nanodisc-reconstituted ATPase to give a final CHAPS concentration of 0.1% and a final total volume of 3 μl. This was necessary to prevent disruption of the enzyme on the air–water interface during blotting. A Quantifoil R 0.6/1 copper grid (Electron Microscopy Sciences) was glow discharged by an ELMO system (Cordouan Technologies) under an ~28 mA current for 2 min before sample application. The grid was then blotted with standard Vitrobot filter paper (Agar Scientific) for 2 s with blot force 25, wait time 15 s, drain time 1 s at 4°C, and 100% humidity in the chamber of a Vitrobot Mark IV (FEI) immediately before being snap-frozen in liquid ethane. The grid-containing box was thereafter stored in liquid nitrogen before loading into the microscope.

Raw micrographs were collected with a 300-kV Titan Krios electron microscope with direct electron detector FEI Falcon-IIIEC in counting mode using the data collection software EPU at a calibrated magnification of 1.085 Å⋅pixel−1 (×75,000) and a total dose of 50 e⋅Å−2 with movies divided into 49 frames collected over 63 s, with a defocus range of –1.0 to –2.0 μm. A total of 2700 raw micrographs were collected, from which 2564 good images were used for beam-induced motion correction using Unblur (58) followed by MotionCor2 (59) (fig. S1A). This two-step procedure improved the final result because MotionCor2 tended to fail when between-frames motions were very large because of the long exposure times. The following steps were performed in Relion (60, 61) unless otherwise stated. Contrast transfer function (CTF) parameters of the motion-corrected micrographs were estimated using CTFFIND4 (62). Particles were automatically picked using 2D class averages from manually selected particles that were extracted using a 480-pixel box and classified by reference-free 2D classification (fig. S1B), giving 181,245 good particles. These particles were 3D classified into six classes using a 30-Å low-pass–filtered initial model and a full mask, both generated from an previous low-resolution ThV1Vo structure (23) under a regularization parameter of 10. The three rotational states of ThV1Vo rotor separated into three classes with unequal particle numbers (39.4, 26.6, and 23.8 K), whereas the two substates of the most populated state 1 were additionally differentiated (23.0 and 15.6 K) (fig. S1C). The three main states were assigned because the rotational positions of the rotor relative to the stator were 120° apart from each other, whereas the two substates differed from state 1 as the whole hydrophilic domain is twisted to different sides. A 3D autorefinement of all six classes using the same initial map and full mask gave a 3.6-Å overall resolution of state 1 after postprocessing based on the gold standard Fourier shell correlation = 0.143 criterion (63) (figs. S3 and S4A); however, transmembrane helices were visible but not well resolved in the resulting map (fig. S1C). This was due to some remaining flexibility between V1 and Vo in each class, which leads to refinement being dominated by the larger V1. Therefore, many different masks were explored for focused refinement to achieve high-quality maps for all regions of the complex. This required a compromise between focusing on a particular area of interest and still having enough total protein volume for the refinement to work. Focused refinement of the hydrophilic domain including subunits A3B3DF and CT of EG2 (better ordered together with V1) gave near-atomic resolution maps for most states (3.4 Å for state 1, 3.6 Å for state 2, 3.9 Å for state 3, 3.8 Å for state 1L, and 4.3 Å for state 1R) (figs. S1C, S3, and S4B). Coiled-coil helices for the NT of both peripheral stalks in state 1 became visible after focused refinement using separate masks, including the membrane-embedded domain plus one of the two peripheral stalks (fig. S2A). Focused refinement of the membrane-embedded domain including subunits dac12 gave a near-atomic resolution map only for the most populated state 1 (~4.1 Å) (figs. S1C, S3, and S4C), whereas lower resolutions allowing secondary structure identification were obtained for the remaining states (5.8 Å for state 2, 6.8 Å for state 3, 5.9 Å for state 1L, and 7.4 Å for state 1R) (fig. S1C). Considering that the two substates differed from the main state 1 only in the hydrophilic region, particles in the main state 1 and two substates were grouped for focused Vo refinement to give a resolution of 3.94 Å. Postprocessing with the mask encompassing only the c12 ring and the membrane-embedded part of subunit a gave a resolution of 3.88 Å (figs. S1C, S3, and S4C). The CtfRefine procedure in Relion3.0-beta, followed by another round of autorefinement, led to slightly improved resolutions for state 1: 3.47 Å for V1Vo, 3.25 Å for V1, and 3.85 Å for Vo.

To generate a complete model for the ThV1Vo in state 1, several maps from the above described focused refinements were carved to generate high-resolution regional maps, which were then combined into a composite map in UCSF Chimera (64) by aligning each map to the map from full V1Vo refinement (fig. S2A). These aligned maps where then added together in Chimera after checking that all maps were on the same scale, generating a final map for refinement. The same process was also carried out for the other rotational states and substates, but with contributing maps coming only from respective V1 and Vo domain refinements.

Model building

The previous 4.7-Å hydrophilic domain structure (PDB 5y5y) and the remaining parts of the 5-Å complete structure (10) were separately fitted into the combined map of state 1 to generate an initial model. Residue numbering for subunit c started after signal peptide cleavage, and was therefore shifted compared with the Uniprot sequence but consistent with previous structures. This rough initial model was refined against the combined map with the Phenix suite phenix.real_space_refine program (65) using our protocols optimized for cryo-EM maps (66), with noncrystallographic symmetry (NCS) imposed for the c12 ring. Our refinement protocol allows optimization of B-factors such that electron radiation–damaged carboxylate side chains acquire high B-factors and do not lead to main-chain distortions. Secondary structure and exposure state of residues were predicted by the PredictProtein server (67) to assist with model building. The initial model was extensively manually corrected residue by residue in COOT (68) in terms of side-chain rotamers, helical register, and main-chain tracing, with references to previous crystal structures of ThV1Vo subunits A and B (PDB 3 gqb) (31), subunit G (PDB 3w3a) (18), subunit F (PDB 2d00) (69), peripheral stalks (PDB 3k5b and 3v6i) (39, 70), subunit d (PDB 1r5z) (44), and close homologs of the subunit a soluble domain (PDB 3rrk) (71) and membrane domain (PDB 6c6l) (22). Side chains were removed to generate a polyalanine model for the distal α/β domain in subunit a due to low resolution in this region (table S2). The corrected model was again refined by the phenix.real_space_refine (65) program with secondary structure and NCS restraints, the outcome of which was then manually checked in COOT (68) to correct any remaining errors. This iterative process was then performed for multiple rounds without NCS restraints until the model was satisfactory in terms of model–density correlation and agreement with reference structures after manual check. Good model geometry, as reflected by a MolProbity score of 1.71 and an EMRinger score of 2.89, was achieved for the final refined model of complete ThV1Vo in state 1 (fig. S4D and tables S1 and S2).

Initial models for substates to state 1 were generated by fitting individual subunits in state 1 model into substate maps. For states 2 and 3, A3B3 domain and rotor subunits (DFdc12) in state 1 were rotated 120° before fitting to their respective map. This was done because conformations of the three AB heterodimers changed as the rotor rotated within. Chain IDs of A3B3 subunits were renamed in COOT (68) to maintain the same scheme as in the state 1 model. These initial models again went through several rounds of real space refinement and manual corrections until convergence (fig. S4D). In the final models of states other than state 1, the side chains in the Vo domain were stripped to the Cβ atom because of lower resolution in this area. Additionally, in state 1R, the side chains in two peripheral stalks were also stripped to the Cβ atom because of low resolution.

Catalytic ADP density is unambiguous, whereas density of proposed additional inhibitory ADPs near the subunit D CT is less well defined (fig. S5). Some ligand is definitely bound in this position, as similar densities are visible in the same location in all states, being slightly clearer in state 2. The shape of density is roughly consistent with ADP (fig. S5), which is preliminary assigned as ADP′ in all states. Possible additional densities, resembling nucleotides to some extent, are observed nearby (near D_Glu207, Asemi Asp43, and Bclosed Glu53) in state 1 and also between subunits Asemi and Bsemi (near Asemi Ser56, Bsemi Gln46, and Bsemi Arg236) in states 1, 1L, and 1R. Their identities remain to be confirmed, so these potential additional ligands are not included in models.


Homologous sequences of individual ThV1Vo subunits were retrieved by the BLAST program, followed by manual inspection of annotations in Uniprot. Alignments of selected sequences were performed with the T-Coffee server (72), which were then used to build a conservation model in the Consurf server (73). Surface electrostatic potential was calculated with the APBS electrostatics plugin in Pymol with scale shown in units of kbT/e, where kb is the Boltzmann’s constant, T is temperature in Kelvin, and e is the charge of an electron.

Supplementary Materials

Supplementary Text

Tables S1 to S5

Figs. S1 to S10

References (7476)

Movies S1 to S3

References and Notes

Acknowledgments: We thank the ETH Zurich Scientific Center for Optical and Electron Microscopy (ScopeM) for the use of Titan Krios EM. Data processing was performed using the IST high-performance computer cluster. We thank A. Charnagalov (IST) for help in developing the ThV1Vo activation protocol and in isolating the T. thermophilus polar lipid extract. Author contributions: Conceptualization: L.A.S.; Investigation: L.Z. and L.A.S.; Writing – original draft: L.Z.; Writing – review & editing: L.A.S. and L.Z.; Supervision: L.A.S.; Funding acquisition: L.A.S. Competing interests: The authors declare no competing interests. Data and materials availability: The combined cryo-EM maps are deposited in the Electron Microscopy Data Bank with accession numbers EMD-4640, 4699, 4700, 4702, and 4703 for rotational states 1, 2, 3, and substates 1L and 1R, respectively. Atomic models have been deposited in the Protein Data Bank with accession numbers 6QUM, 6R0W, 6R0Y, 6R0Z, and 6R10 for rotational states 1, 2, 3, and substates 1L and 1R, respectively. All other data are available in the main text or the supplementary materials.
View Abstract

Navigate This Article