Eukaryotic vacuolar H+-ATPases (V-ATPases) are rotary enzymes that use energy from hydrolysis of ATP to ADP to pump protons across membranes and control the pH of many intracellular compartments. ATP hydrolysis in the soluble catalytic region of the enzyme is coupled to proton translocation through the membrane-bound region by rotation of a central rotor subcomplex, with peripheral stalks preventing the entire membrane-bound region from turning with the rotor. The eukaryotic V-ATPase is the most complex rotary ATPase: it has three peripheral stalks, a hetero-oligomeric proton-conducting proteolipid ring, several subunits not found in other rotary ATPases, and is regulated by reversible dissociation of its catalytic and proton-conducting regions1,2. Studies of ATP synthases, V-ATPases, and bacterial/archaeal V/A-ATPases have suggested that flexibility is necessary for the catalytic mechanism of rotary ATPases3,4,5, but the structures of different rotational states have never been observed experimentally. Here we use electron cryomicroscopy to obtain structures for three rotational states of the V-ATPase from the yeast Saccharomyces cerevisiae. The resulting series of structures shows ten proteolipid subunits in the c-ring, setting the ATP:H+ ratio for proton pumping by the V-ATPase at 3:10, and reveals long and highly tilted transmembrane α-helices in the a-subunit that interact with the c-ring. The three different maps reveal the conformational changes that occur to couple rotation in the symmetry-mismatched soluble catalytic region to the membrane-bound proton-translocating region. Almost all of the subunits of the enzyme undergo conformational changes during the transitions between these three rotational states. The structures of these states provide direct evidence that deformation during rotation enables the smooth transmission of power through rotary ATPases.
The V-ATPase from S. cerevisiae consists of subunits A3B3CDE3FG3Hacxc′yc′′zde (Extended Data Fig. 1), where x, y, and z denote unknown stoichiometries. In the soluble V1 region, the three pairs of A- and B-subunits are arranged around the D-subunit of the central rotor. Each AB pair contains a single catalytic nucleotide-binding site and together the three AB pairs produce a pseudo-symmetric trimer of heterodimers, with each AB pair in a different conformation6. The three main conformations of the AB pairs are known as ‘tight’, ‘loose’, and ‘open’. These conformations are expected to bind ATP, ADP and phosphate, and no nucleotide, respectively, although their nucleotide content in crystal structures of catalytic regions from other rotary ATPases has depended on the crystallization conditions7,8. During catalysis, the inter-conversion between the catalytic conformations of the three AB pairs is coupled to rotation of the central rotor, which extends into the membrane via subunit d, to the cxc′yc′′z-ring. This rotary catalytic mechanism suggests that a population of intact V-ATPase complexes will be found in three or more structurally distinct rotational states. To investigate the rotational states of the eukaryotic V-ATPase, we isolated the complex from S. cerevisiae in detergent and imaged it by electron cryomicroscopy (cryo-EM) with a direct detector device camera (Extended Data Fig. 2). Classification of V-ATPase images enabled the identification of three distinct three-dimensional classes of varying size that gave maps at resolutions of 6.9 Å (47% of images), 7.6 Å (36% of images), and 8.3 Å (17% of images) (Extended Data Fig. 3a, b), with 106,445 particle images ultimately contributing to the three maps. Cross-sections through the maps show that α-helices are well resolved at these resolutions (Fig. 1 and Supplementary Video 1). The class with the largest population most closely corresponds to an earlier cryo-EM map of the S. cerevisiae V-ATPase where conformational separation was not performed9 but differs from the conformation identified for the Manduca sexta V-ATPase10.
Resolution does not appear to be homogenous within each three-dimensional map, suggesting further conformational heterogeneity within each class. However, additional classification strategies involving focusing classification on the regions thought to be variable11 did not produce meaningful classes. The resolution appears better in the soluble V1 region of the complex (Fig. 1a top and Extended Data Fig. 3c) than in the membrane-bound VO region (Fig. 1a bottom and Extended Data Fig. 3c). In the VO region resolution is best where the a-subunit interacts with the c-ring (pink arrow in Fig. 1a bottom), suggesting a rigid interaction within an otherwise dynamic complex. Crystal structures and homology models for V-ATPase subunits were docked into the three different maps and their conformations refined by molecular dynamics flexible fitting12 (Fig. 1b and Extended Data Fig. 4). The most striking structural difference between the three maps is the position of the central rotor, consisting of subunits D, F, d, and the c-ring (blue star in Fig. 1b). Each map therefore appears to correspond to a rotational state of the enzyme. Rotational state 3 is the most different from the other two and is the class with the fewest particle images. The unequal distribution of particle images in the three classes suggests that, in the absence of free nucleotide, the V-ATPase relaxes to three unequally populated states.
The proton-carrying c-ring of the yeast V-ATPase is a hetero-oligomer of subunits c, c′, and c′′, each possessing a single conserved glutamate residue (Glu137, Glu145, Glu108, respectively) that can bind and transport protons during catalysis13. Subunits c and c′ have four transmembrane α-helices each, with an additional amino (N)-terminal α-helix for c′′ that may not be membrane bound and is not necessary for function14. The cryo-EM maps show the c-ring to consist of an inner ring and an outer ring of transmembrane α-helices (Fig. 1a bottom and Fig. 2a, b magenta). The ring appears to have 20-fold symmetry (Fig. 1a bottom and Fig. 2b), suggesting that the extra N-terminal α-helix of the c′′-subunit may protrude into the ring, as seen in a recent crystal structure of a bacterial heteromeric c-ring15. The structure of the c-ring in each cryo-EM map accommodates a total of ten c-, c′-, or c′′-subunits (Fig. 2a, b) with each subunit contributing two α-helices to the outer ring and two α-helices to the inner ring. The modelled c-subunits fit the density with their N and carboxy (C) termini facing the luminal side of the membrane16. The presence of ten subunits in the c-ring is inconsistent with an earlier prediction of a 4:1:1 stoichiometry for subunits c, c′, and c′′, which would require a total of 6 or 12 subunits in the ring17,18. A result of the existence of ten subunits in the c-ring is that in a tightly coupled enzyme complete rotation of the c-ring driven by hydrolysis of three ATP molecules would deliver ten protons across the lipid bilayer. Consequently, the c-ring stoichiometry probably sets the ATP:H+ ratio at 3:10 for the S. cerevisiae V-ATPase, which is the same ratio found with the F-type ATP synthase in S. cerevisiae mitochondria19. The ATP:H+ ratio was not known previously for any eukaryotic V-ATPase. With a Gibbs free energy for ATP hydrolysis of 57 kJ mol−1 at 30 °C, this ATP:H+ ratio limits the maximum pH gradient or voltage established across the vacuolar membrane in S. cerevisiae to 3.0 units or 180 mV, respectively20.
The three maps provide the highest-resolution insight available into the structure of the membrane-bound portion of subunit a, which contains the channels that conduct protons to and from the c-ring. The transmembrane proteins in the VO region of the S. cerevisiae V-ATPase are subunits a, e, and the c-ring. However, the dodecylmaltoside-solubilized V-ATPase does not contain the e-subunit21. Consequently, any density in the VO region that cannot be attributed to the c-ring or detergent must be part of the a-subunit (Fig. 2b–d). Subunit a has been predicted to have eight transmembrane α-helices22. The density corresponding to the a-subunit has a complex fold and appears to have at least eight transmembrane α-helices and several well-defined structural elements above and below the expected position of the lipid bilayer. Strikingly, there are two highly tilted α-helices from the a-subunit that span the lipid bilayer where the a-subunit is in contact with the c-ring (Fig. 2c, red arrows). These α-helices contact a group of three α-helices from the c-ring (Fig. 2c, d). In F-type ATP synthases and V/A-ATPases each outer α-helix of the c-ring has a conserved proton-carrying Glu or Asp residue. In the eukaryotic V-ATPase every other α-helix on the surface of the c-ring has the conserved Glu residue. At the present resolution it is not possible to distinguish the outer helices with the conserved Glu residues from the outer helices lacking the Glu residues. Therefore, it is possible to fit the c-ring in the electron microscopy map in two different ways: one where a single Glu residue contacts the a-subunit and one where two different Glu residues contact the a-subunit. This latter arrangement would place two different c-subunits (Fig. 2c, circled in blue) in different chemical environments, enabling one c-subunit to exchange protons with the cytoplasm while the other exchanges protons with the organelle matrix. This arrangement could create two half-channels for proton translocation23 and was seen with the Thermus thermophilus enzyme24. The α-helices in the a-subunit are much easier to identify in the present maps at 6.9–8.3 Å resolution than in the map of the T. thermophilus V/A-ATPase at 9.7 Å resolution determined earlier24. The α-helices identified here in all three states (Extended Data Fig. 5a) are consistent with the density from the earlier map of the bacterial enzyme (Extended Data Fig. 5b left) but are not entirely consistent with the previously proposed locations of α-helices in that density (Extended Data Fig. 5b right).
With three catalytic nucleotide-binding sites, the V1 region is expected to operate as a three-step motor while the ten titratable Glu residues of the c-ring suggest that the VO region functions as a ten-step motor. This 3:10 ratio produces a symmetry mismatch between the V1 and VO regions. The new maps demonstrate how symmetry mismatch can be tolerated in rotary ATPases. The rotational position of subunit D of the central rotor was measured relative to the A3B3 hexamer (Fig. 3 upper, blue density and line), which itself twists within the rest of the enzyme between the three different states (Fig. 3, black lines). The rotation of subunit D corresponds to steps of 117° from state 3 to 1, 119° from state 1 to 2, and 124° from state 2 to 3, all of which are in good agreement with the 120° steps expected for the threefold pseudo-symmetric V1 region. In the VO region, rotation of one c-subunit against the a-subunit requires a 36° rotation of the c-ring (360°/10). The rotational position of the d-subunit and c-ring relative to the a-subunit could be measured precisely in rotational states 1 and 2 owing to the resolution in the membrane region of these maps. In rotational state 3, the rotation of the c-ring against the a-subunit could not be measured with the same confidence. For the transition from state 1 to 2, we measured a 139° rotation of the d-subunit and c-ring relative to the transmembrane portion of the a-subunit (Fig. 3 lower, cyan density and line). This rotation from state 1 to 2 matches the 144° rotation (36° × 4) expected from a four-tenths rotation of the c-ring and would be expected to deliver four protons across the lipid bilayer. The transitions from states 2 to 3 and 3 to 1 were measured at ∼101° and ∼120°, respectively. Together, these rotations match ∼216° of rotation (36° × 6), and are probably due to 108° (36° × 3) of rotation for each transition. Consequently, it appears that transitions from states 1 to 2, 2 to 3, and 3 to 1 transport four, three, and three protons, respectively. The observed rotational states reveal the conformations of the enzyme in vitro after the available ATP in solution has been hydrolysed. Therefore, it is possible that the conformations of the enzyme when rapidly hydrolysing ATP could be different from the conformations observed here. For example, when rapidly hydrolysing ATP, the stepping motions of the enzyme may be smoothed with 3.3 protons transported for each ATP hydrolysis event. It is also possible that some ‘slip’ occurs during proton translocation and on average less than one proton is transported for each c-subunit.
Overlaying the different conformations of various subunits in the complex suggests some of the structural changes that may occur during rotary catalysis (Fig. 4). These conformational changes are illustrated dramatically by interpolating between the three rotational state structures (Supplementary Video 2). The different conformations of the catalytic subunits of rotary ATPases have been reported from mitochondrial and bacterial F1-ATPases25,26 and bacterial V1/A1-ATPases8,27 but never before for a eukaryotic V-type enzyme and never before within an intact rotary ATPase. Different from crystal structures of isolated V1/A1-ATPase or F1-ATPase subcomplexes, the availability of structures of the intact enzyme in different rotational states enables comparison of ‘open’, ‘tight’, and ‘loose’ conformations of the AB pairs when the rotor is in different positions. The observed conformations of the A3B3 hexamer (Fig. 4a and Extended Data Fig. 6a–c) reveal a bend of the C-terminal domain of the A-subunit that closely resembles the conformational changes seen in the Enterococcus hirae V1/A1-ATPase8 and mammalian mitochondrial F1-ATPase25 rather than the nearly rigid movement of subunits in the T. thermophilus V1/A1-ATPase27. The equivalent ‘open’, ‘tight’, and ‘loose’ conformations from different AB pairs can be overlaid with near perfect fidelity (Extended Data Fig. 6d–f). The protein samples used to prepare cryo-EM grids were not supplemented with nucleotide and consequently the nucleotide occupancy of the different catalytic sites is unknown. Because the nucleotide occupancy of the different AB pairs is not known, it is possible that the observed states correspond to the intrinsic asymmetry seen in AB pairs and αβ pairs of the E. hirae and S. cerevisiae V1/A1- and F1-ATPase crystal structures lacking nucleotide, which closely resemble the conformations of the enzymes with bound nucleotide8,28.
Rotary ATPases have been proposed to have an elastic coupling between their catalytic and membrane-bound regions to smooth the transmission of power between ATP hydrolysis or synthesis and proton translocation3,4,5. This need for elastic coupling is exacerbated by the 3:10 symmetry mismatch of the V1 and VO regions: the enzyme must deform to allow the rotor to be simultaneously in the correct position in the catalytic V1 region and the membrane-bound VO region. Earlier studies have suggested that the central rotor of the E. coli ATP synthase is the compliant element in that enzyme, while the peripheral stalk remains rigid29. The current structures indicate that the extended helical part of the central rotor D-subunit, equivalent to the F-type ATP synthase γ-subunit, remains rigid during rotation while the part of subunit D in contact with subunit d bends (Fig. 4b, dark blue, and Supplementary Videos 2 and 3). Further, the orientation of the d-subunit changes with respect to the D- and F-subunits, wobbling to accommodate distortion of the enzyme during rotation (Fig. 4b, cyan, and Supplementary Videos 2 and 3). The catalytic A- and B-subunits push against the E- and G-subunits of the peripheral stalks during rotation. The E- and G-subunits of the peripheral stalks undergo a bending motion along their elongated coiled-coil region, reminiscent of the action of a cantilever spring (Fig. 4c, Extended Data Fig. 6g–i and Supplementary Videos 2 and 3). The N-terminal domain of the a-subunit swings parallel to the membrane, moving to and away from the rotation axis of the rotor like the arm of a record player. The bend in the a-subunit occurs at the narrow interface between its N- and C-terminal domains (Fig. 4d and Supplementary Videos 2 and 3). In comparison, the ‘head’ domain of the C-subunit twists like a torsion spring at the ‘neck’ domain and thereby maintains its connections between peripheral stalks 2 and 3 (Fig. 4e and Supplementary Videos 2 and 3). The existence of these conformational changes is not obvious from inspection of crystal structures of the individual subunits. However, when visualized as a movie, each subunit appears to have evolved to perform these motions. Overall, the structures presented here show that in the V-ATPase both the rotor and stator parts of the engine undergo coordinated conformational changes. This combination of flexibility and rigidity is probably a key attribute for the function of these highly efficient macromolecular machines.
No statistical methods were used to predetermine sample size.
S. cerevisiae V-ATPase samples were purified as described previously from 10 l fermenter growths of S. cerevisiae strain JTY002 (ref. 9). The enzyme was purified by affinity to 3×Flag tags at the C termini of the A-subunits, providing samples (30 μl at ∼10 mg ml−1) in buffer containing 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 0.02% (w/v) DDM, and 150 μg ml−1 3×Flag peptide, without any added nucleotide. Holey carbon-film-coated electron microscopy grids with regular arrays of 500–800 nm holes were prepared by nanofabrication30 and cryo-EM specimens were prepared as described previously9. A total of 3,685 30-frame movies were acquired with defocuses between 1 and 7 μm with an FEI Tecnai F20 microscope operating at 200 kV and equipped with a Gatan K2 Summit direct detector device camera. The camera was used in counting movie mode with five electrons per pixel per second for 15 s and 0.5 s per frame. This exposure rate resulted in one electron per square ångström per frame on the specimen. Frames were aligned using the new program alignframes_lmbfgs31 and averaged. Averaged frames were used for determination of contrast transfer function parameters with CTFFIND3 (ref. 32) and selection of coordinates for 200–250 particle-like features per image, some of which corresponded to true particle images, by templaand 3te matching with TMaCS33. These coordinates were then used to extract candidate particle images from the unaligned movie frames with individual particle motion correction using the alignparts_lmbfgs algorithm31. A measured ∼2% anisotropy in the magnification of the microscope was corrected in particle images by interpolation in Fourier space34, to produce a calibrated pixel size of 1.45 Å. Contrast transfer function parameters were corrected to account for the effects of this anisotropy34. A total of 790,642 candidate particle images were subjected to two- and three-dimensional classification with Relion35,36. Particle images in two-dimensional classes with averages that resembled projections of the V-ATPase and contained high-resolution features were selected for further analysis. Three-dimensional classes were obtained from 217,533 particle images and three classes containing 106,445 particle images were used to build the three final three-dimensional maps. Local resolution was estimated with ResMap37. Calculations with Relion were performed using the SciNet cluster38 and the SickKids High Performance Facility. Three-dimensional maps were segmented with UCSF Chimera39,40. Homology models were calculated either with Phyre2 (ref. 41) or with HHpred42 and MODELLER43 and atomic models from the PDB and homology models were docked into electron microscopy maps by MDFF12. Crystal structures were available for subunits C44 and H45, and the DF46 and EG47 subcomplexes. Homology models were built for subunits A, B8,27, d48, the N-terminal domain of subunit a49, and the c-subunit50. The magnitudes of subunit rotations between states were measured in UCSF Chimera and movies were generated with UCSF Chimera.
All new computer programs described above are available from https://sites.google.com/site/rubinsteingroup/home.
Electron Microscopy Data Bank
Protein Data Bank
Cryo-EM maps have been deposited in the Electron Microscopy Data Bank under accession numbers EMD-6284, EMD-6285, and EMD-6286. Atomic models have been deposited in the Protein Data Bank under accession numbers 3J9T, 3J9U, and 3J9V.
We thank P. Rosenthal, R. Henderson, V. Kanelis, and L. Kay for comments on the manuscript. J.Z. was supported by a Doctoral Postgraduate Scholarship from the Natural Sciences and Engineering Research Council of Canada and a Mary Gertrude l’Anson Scholarship. J.L.R. is the Canada Research Chair in Electron Cryomicroscopy. This work was supported by operating grant MOP 81294 from the Canadian Institutes of Health Research.
Extended data figures
Cross sections through the three maps, each showing a different rotational state of the V-ATPase.
Interpolation between the three observed rotational states of the V-ATPase.
Exploded view of subunits when interpolating between the three observed rotational states of the V-ATPase
Exploded view of subunits when interpolating between the three observed rotational states of the V-ATPase.