Cryo-EM structures of the autoinhibited E. coli ATP synthase in three rotational states

A molecular model that provides a framework for interpreting the wealth of functional information obtained on the E. coli F-ATP synthase has been generated using cryo-electron microscopy. Three different states that relate to rotation of the enzyme were observed, with the central stalk’s ε subunit in an extended autoinhibitory conformation in all three states. The Fo motor comprises of seven transmembrane helices and a decameric c-ring and invaginations on either side of the membrane indicate the entry and exit channels for protons. The proton translocating subunit contains near parallel helices inclined by ~30° to the membrane, a feature now synonymous with rotary ATPases. For the first time in this rotary ATPase subtype, the peripheral stalk is resolved over its entire length of the complex, revealing the F1 attachment points and a coiled-coil that bifurcates toward the membrane with its helices separating to embrace subunit a from two sides. DOI: http://dx.doi.org/10.7554/eLife.21598.001


Introduction
In most cells, the bulk of ATP, the principal source of cellular energy, is synthesized by ATP synthase. This molecular generator couples ion flow across membranes with the addition of inorganic phosphate (Pi) to ADP thereby generating ATP (Iino and Noji, 2013;Stewart et al., 2014). Most bacteria, including Escherichia coli have only one type of rotary ATPase, referred to as F-type ATPase. Like the analogous complexes in other kingdoms, it is based on two reversible motors, termed F 1 and F o (Negrin et al., 1980), connected by central and peripheral stalks (Wilkens and Capaldi, 1998a) (Figure 1). The F o motor spans the membrane converting the potential energy of the proton motive force (pmf) into rotation of the central stalk that in turn drives conformational changes in the F 1 catalytic sites.
The F o motor is constructed from subunits a, b and c (Figure 1). Subunit c assembles into a ring, thought, in E. coli, to have decameric stoichiometry (Jiang et al., 2001;Ballhausen et al., 2009;Düser et al., 2009;Ishmukhametov et al., 2010), whereas subunits a and b associate to form a helical bundle adjacent to this ring. Recent sub nanometer electron cryo-microscopy (cryo-EM) reconstructions of F-type (Allegretti et al., 2015;Zhou et al., 2015;Kühlbrandt and Davies, 2016;Hahn et al., 2016) and the analogous V-and A-type ATPases (Zhao et al., 2015;Schep et al., 2016) as well as a low-resolution crystal structure of Paracoccus denitrificans F-ATPase (Morales-Rios et al., 2015) are consistent with a two half-channel mechanism for the generation of rotation within the membrane (Vik and Antonio, 1994;Junge et al., 1997). All structures confirm that a four-helix bundle of subunit a, inclined by 20-30˚to the membrane plane, forms a crucial structural component. In this mechanism, protons from the bacterial periplasm access a conserved negatively charged carboxylate in subunit c (Asp61 in E. coli [Hoppe et al., 1982]) through an aqueous half channel at the subunit a/c interface (Steed and Fillingame, 2008). Neutralizing this carboxylate enables the c-ring to rotate within the hydrophobic membrane and to access a second aqueous half channel that opens to the cytoplasm into which the protons are released (Pogoryelov et al., 2010). A conserved arginine residue in helix-4 of subunit a (Arg210 in E. coli) prevents the c-ring rotating in the opposite direction and short-circuiting of the system (Lightowlers et al., 1987;Cain and Simoni, 1989;Mitome et al., 2010). The sequential binding of protons in combination with thermal fluctuations generates rotation within the complex in a manner akin to a turbine (Oster and Wang, 1999;Pogoryelov et al., 2010;Aksimentiev et al., 2004). The torque generated in the F o motor is then transferred to the F 1 motor by the central shaft consisting of subunits g and e (Wilkens et al., 1995). The N-and C-termini of subunit g form a curved coiled-coil that extends into the central cavity of F 1 .
The F 1 motor is the chemical generator in which ATP is synthesized. The motor comprises a ring of three heterodimers, each containing an active site at the interface of subunits a and b. Within the F 1 motor, each ab dimer has a different conformation at any point in time and can be either empty, bound to ADP and Pi, or bound to ATP (open, half-closed, closed) (Abrahams et al., 1994;Yoshida et al., 2001). These different catalytic states relate to the position of the curved coiled-coil of subunit g in the central stalk, which drives the conformational changes associated with catalysis. To enable the central stalk to rotate relative to the F 1 ab heterodimers, the F o and F 1 motors need to be coupled. This coupling is mediated by the peripheral stalk that is constructed from subunits b and d (Figure 1). Subunit b forms an amphipathic homodimeric coiled-coil that spans the periphery eLife digest ATP synthase is a biological motor that produces a molecule called adenosine triphosphate (ATP for short), which acts as the major store of chemical energy in cells. A single molecule of ATP contains three phosphate groups: the cell can remove one of these phosphates to make a molecule called adenosine di-phosphate (ADP) and release energy to drive a variety of biological processes.
ATP synthase sits in the membranes that separate cell compartments or form barriers around cells. When cells break down food they transport hydrogen ions across these membranes so that each side of the membrane has a different level (or "concentration") of hydrogen ions. Movement of hydrogen ions from an area with a high concentration to a low concentration causes ATP synthase to rotate like a turbine. This rotation of the enzyme results in ATP synthase adding a phosphate group to ADP to make a new molecule of ATP. In certain conditions cells need to switch off the ATP synthase and this is done by changing the shape of the central shaft in a process called autoinhibition, which blocks the rotation.
The ATP synthase from a bacterium known as E. coli -which is commonly found in the human gut -has been used as a model to study how this biological motor works. However, since the precise details of the three-dimensional structure of ATP synthase have remained unclear it has been difficult to interpret the results of these studies.
Sobti et al. used a technique called Cryo-electron microscopy to investigate the structure of ATP synthase from E. coli. This made it possible to develop a three-dimensional model of the ATP synthase in its autoinhibited form. The structural data could also be split into three distinct shapes that relate to dwell points in the rotation of the motor where the rotation has been inhibited. These models further our understanding of ATP synthases and provide a template to understand the findings of previous studies.
Further work will be needed to understand this essential biological process at the atomic level in both its inhibited and uninhibited form. This will reveal the inner workings of a marvel of the natural world and may also lead to the discovery of new antibiotics against related bacteria that cause diseases in humans. of the complex linking subunit a with subunit a, whereas subunit d provides additional coupling of the C-termini of the b and a subunits (Carbajo et al., 2005;Wilkens et al., 2005).
The bacterial F-ATPase can also function in reverse, employing ATP hydrolysis to generate a proton gradient across the membrane when needed (von Ballmoos et al., 2008). In vivo, subunit e is believed to change conformation in an ATP-dependent manner to prevent rotation of the complex (Capaldi et al., 1992;Rodgers and Wilce, 2000;Yagi et al., 2007;Imamura et al., 2009) thereby conserving ATP when its concentration is low. This regulatory function is mediated by the C-terminal domain of subunit e (eCTD) that, when not bound to ATP, opens to an extended conformation and inserts into the ab heterodimers. The crystal structures of both the E. coli and Bacillus PS3 F 1 motors in this autoinhibited state show the eCTD intercalating into the ab heterodimers (Cingolani and Duncan, 2011;Shirakihara et al., 2015). However, in each structure, the F 1 motor had been captured in a different conformation (E. coli -half-closed, closed, open [Cingolani and Duncan, 2011] and Bacillus PS3 -open, closed, open [Shirakihara et al., 2015]) which could either relate to inter species differences or crystal contacts and crystallization conditions.
The crystal structure of the F-ATPase from P. denitrificans (Morales-Rios et al., 2015), that is closely related to E. coli (38% sequence identity over all subunits), shows a similar overall architecture to bovine F 1 F o ATP synthase as well as to the main features of A/V ATPases. However, it is inhibited by the z-protein rather than by subunit e, which is generally employed by bacterial F-ATPases for this purpose. Moreover, this crystal structure shows only one conformation of the rotary catalytic cycle.
Here, we present three cryo-EM maps along with molecular models of E. coli F-ATPase in its autoinhibited state, determined to resolutions of 6.9, 7.8, and 8.5 Å . In all three reconstructions, the eCTD is in an extended conformation, stabilizing an overall F 1 motor conformation similar to that seen in the thermophilic Bacillus PS3 F 1 ATPase structure. Density for the peripheral stalk extends the entire length of the complex and its coiled-coil bifurcates towards the N-terminus to enter the membrane as two separate helices that clamp the a subunit to the c ring. Moreover, our maps allowed us to interpret the complete F 1 -delta interface, showing the three a subunit N-termini in distinct orientations. Each map also confirmed the c-ring stoichiometry to be decameric, which to date has been only characterized by crosslinking and single molecule analyses. We used our maps in combination with published crosslinking and mutagenesis information to generate a molecular model of the complex in three states. These models provide crucial structural information on a key complex that extends our understanding of the mechanism of rotary ATPases in general, together with information on the bacterial ATP synthase, which is seen as an important antimicrobial target in organisms related to E. coli such as Mycobacterium tuberculosis (Ahmad et al., 2013).

Complete molecular models of three different F-ATPase conformations
Cysteine-free E. coli F-ATPase, as described in Ishmukhametov et al. (2005) where all 10 cysteines were replaced with alanines and a His-tag introduced on the b subunit, was solubilized in digitonin detergent and purified as described in the Materials and methods. This procedure provided pure protein ( (Cook et al., 2003;Peskova and Nakamoto, 2000;Tsunoda et al., 2000).
Protein was further examined by cryo-EM without addition of nucleotides. 395,140 particles were picked, of which 216,711 were used in refinement. Three different conformations of the complex were identified using 3D classification in RELION (Scheres, 2012). The particles in each subset were then refined to generate sub-nanometre reconstructions, to a resolution of 6.9, 7.8 and 8.5 Å ( In these three conformations, the central stalk was progressively rotated 120˚relative to the peripheral stalk. Even though the resolution of the reconstructions varied throughout the complex, it was sufficient to resolve individual helices. Additional density of the N-terminal His-tag of the b subunit, as well as helical and b sheet patterns observed in parts of the map in the F 1 motor region illustrate the high quality of the maps, with the c-ring density being poorest ( Figure 2-figure supplement 4). Local resolution estimates showed the region corresponding to the F 1 motor to be of highest quality, the F o motor with moderate detail and, the detergent micelle being clearly the worst region of the map (Figure 2-figure supplement 5). Docking of high-resolution crystal and NMR models of different components into the maps followed by manual building and refinement enabled virtually complete molecular models of the three different states to be built ( Figure  The cryo-EM maps provided novel insights into the architecture and function of the E. coli F-ATPase. Thus, although its overall architecture was similar to that of F-ATPase from P. denitrificans (Morales-Rios et al., 2015) and F 1 F o ATP synthases from Bos Taurus (Zhou et al., 2015), Yarrowia lipolytica (Hahn et al., 2016) and Polytomella (Allegretti et al., 2015), with the catalytic F 1 motor attached to a proton powered membrane F o motor and single central and peripheral stalks, differences in the individual motors and peripheral stalk were apparent. Comparison with the membraneembedded motors from other sub nanometre cryo-EM maps indicated the E. coli F-ATPase had a simpler stator architecture, containing only seven helices in the a and b subunits rather than the   (Wada et al., 1999). This difference suggested that the extra helices present in other rotary ATPase subtypes could have additional functions such as the dimerization seen in mitochondria (Hahn et al., 2016). A movie generated by interpolation between the three states (Video 1) indicated that the F 1 motor rocks or wobbles during the catalytic cycle  as previously predicted, although of course the structures described here do not represent the complex in its uninhibited active synthesizing form. Two pivot points, one near the peripheral stalk/F o interface (~bArg36) (Welch et al., 2008) and one near the peripheral stalk/F 1 interface (~bGln106), enabled the stalk to accommodate this eccentric movement of F 1 (Figure 2-figure supplement 9).

Subunit d interaction with subunit b dimer and all three a subunits
The maps showed a long right-handed coiled-coil dimer generated by the two b subunits of the peripheral stalk together with the globular d subunit that anchors them to the catalytic head ( Figure 3A). The quality of the map was sufficient to enable almost the entire of the d subunit to be built as a polyalanine model ( Figure 2D-F), whereas previous structural information was limited to the N-terminal domain (Wilkens et al., 2005). Interestingly, the peripheral stalk contacted all three a subunits via their N-terminal helices, but did so asymmetrically employing three different interfaces with each a subunit ( Figure 4). Although the resolution of the map was insufficient to assign the precise interface, the binding of the peripheral stalk to three anchor points in different geometries would provide a molecular key that would result in the d subunit binding in a single orientation across the top of the symmetrical ab heterodimers.  The peripheral stalk bifurcates into the membrane The b subunits formed a homodimeric coiled-coil that spaned almost the entire complex (212 of 232 Å ) with their N-termini bifurcating just above the membrane to generate two separate helices within the membrane ( Figure 3A). This was unexpected, albeit reminiscent of the yeast F-type ATP synthase dimer (Figure 2-figure supplement 8C), where subunit eight is an evolutionary derivative of the bacterial b subunit (Hahn et al., 2016). Furthermore, NMR analysis of the transmembrane domain of E. coli F-ATPase b subunit (Dmitriev et al., 1999) showed a helical structure that was interrupted by a rigid 20˚bend at residues 23-26 that result in a structure consistent with the b subunit bifurcation.
Inhibition of the E. coli F-ATPase by central stalk subunit e All three reconstructions showed the complex in its autoinhibited state, with clear density for the eCTD extending deep into the central cavity of the F 1 enzyme ( Figure 3B). Fitting of the E. coli a 3 b 3 ge crystal structure (Cingolani and Duncan, 2011) into our cryo-EM maps showed that the b1 subunit had adopted a different more open conformation (Figure 3-figure supplement 1). In the above crystal structure, the eCTD contacts more subunits in F 1 (a1, a2, b1, b2 and g) compared to our cryo-EM reconstructions, where it contacted fewer subunits (a1, a2, b2 and g). The conformation of our cryo-EM structure was more similar to that seen in the Bacillus PS3 structure To ascertain the enzymatic state of the F 1 motor, we generated a difference density map between the cryo-EM density and our built model. Remarkably, clear peaks were seen at the nucleotide binding pockets, indicating that the non-catalytic binding sites in subunit a contained nucleotide, as well as the 'closed' ab heterodimer ( Figure 5). These were modeled as ATP and ADP respectively based on known orientations of these nucleotides from high-resolution crystal structures. Interestingly, this did not correspond to the nucleotide binding states of either the E. coli (Cingolani and Duncan, 2011) or the PS3 crystal structures (Shirakihara et al., 2015).
Single particle analysis revealed highly uniform homogeneity of the protein preparation, where 100% of F 1 F o molecules observed demonstrated the extended conformation of eCTD. Our structural data were supported by an enzymatic assay (Figure 3-figure supplement 2), where the extent of ATP hydrolysis inhibition of the protein by subunit e was tested with N,N-dimethyldodecylamine N-oxide (LDAO), a well-known activator of the subunit e inhibited protein. LDAO used at 0.4% (weight/volume) concentration has been shown to stimulate ATP hydrolysis by the E coli protein 3-4 times (Dunn et al., 1990;Peskova and Nakamoto, 2000), including earlier studies (Ishmukhametov et al., 2008(Ishmukhametov et al., , 2016 on the same cysteine-free F 1 F o construct using the same batch of LDAO. Prior to addition of LDAO, the hydrolysis rate was 0.75 mmol ATP/min/mg protein but surprisingly in presence of 0.4% LDAO it was stimulated~13 times. To our best knowledge, such a high LDAO stimulation of ATP hydrolysis by E. coli F 1 F o was not described in the literature before and is consistent with our single particle data. The E. coli F o motor -architecture of a proton channel Density in F o defined the overall architecture of the membrane-embedded motor together with two invaginations of the detergent micelle that have previously been proposed to facilitate proton translocation (Allegretti et al., 2015;Kühlbrandt and Davies, 2016) (Figure 6-figure supplement 1). While the overall density of the c-ring was relatively weak, 10 peaks of density were clearly present when viewed from above ( Figure 6-figure supplement 2), confirming the stoichiometry of the cring to be decameric in E. coli F-ATPase (Jiang et al., 2001;Ballhausen et al., 2009;Düser et al., 2009;Ishmukhametov et al., 2010). Furthermore, density inside the c-ring corroborates data suggesting it to be filled with phospholipids (Oberfeld et al., 2006).
By combining the helical density from the cryo-EM maps ( Figure 6A), with models previously suggested for the related bovine subunit, together with crosslinking data and transmembrane topography prediction for the E. coli F-ATPase (Jiang and Fillingame, 1998;Valiyaveetil and Fillingame, 1998;Moore and Fillingame, 2008;Wada et al., 1999), it was possible to build a molecular model of the a subunit ( Figure 6B). The crosslinks mapped to two clusters ( Figure 6-figure supplement 3), allowing a likely sequence register for the model to be proposed. This was consistent with the two half channel hypothesis, placing Arg210 of subunit a adjacent to Asp61 of the c-ring ( Figure 6B). Interestingly, density for the c subunit is clearest adjacent to Arg210 of subunit a suggesting this area to be well ordered ( Figure 6-figure supplement 4).

Discussion
We have generated cryo-EM maps of a bacterial F-ATPase providing new insights into this rotary ATPase subtype. These maps enabled the generation of a molecular model that presents a framework onto which the vast array of information available on the widely studied E. coli enzyme can be mapped, including the attachment of the peripheral stalk to the F 1 and F o motors, the inhibition mediated by the e subunit, and the stoichiometry of the c-ring. In addition, the model confirmed the presence of key features such as the near 'horizontal' helices angled at 20-30˚relative to the membrane, indicating that this feature is conserved and is a signature of F, V and A-type ATPases.
Our reconstructions extended previous work (Wilkens and Capaldi, 1998b) by showing the structure of the complete peripheral stalk and how it is attached to both the F 1 and F o components. The peripheral stalk functions to counteract rotation of the F 1 stator relative to the F o stator as the central stalk rotates, but must also accommodate conformational changes in the F 1 motor during catalysis. The peripheral stalk is based on a long right-handed coiled-coil dimer, that is the hallmark of all rotary ATPase peripheral stalks (Lee et al., 2010), showed near parallel a-helices, based on an 11residue hendecad sequence repeat, spanning the space between the F o and F 1 motors, that changed into a 15-residue quindecad sequence repeat along the F 1 motor enabling it to accommodate conformational changes . Although sequence identity is low (22%), the overall fold of the soluble portion of the peripheral stalk was strikingly similar to that of Thermus thermophilus A-ATPase (Lee et al., 2010), illustrating a strong evolutionary pressure for this fold and its function to prevent rotation between the ab heterodimers and the a subunit while accommodating wobbling of the F 1 motor. The bifurcation of the peripheral stalk coiled-coil into two separate helices in the membrane was unexpected, but this arrangement enabled the peripheral stalk to bind to the a subunit in two regions. This in turn increased the distance about the fulcrum of interaction, which may help clamp the a subunit to the c-ring and counteract rotation and pivoting relative to the rotor. The cryo-EM maps also indicated that the peripheral stalk is able to flex about two hinges adjacent to the F 1 and F o motors, enabling it to accommodate conformational changes in the catalytic head.
In addition to its fundamental importance in cell metabolism, the regulation of ATP synthase is also an attractive antibiotic discovery target for pathogenic bacteria closely related to E. coli (Ahmad et al., 2013). Bacterial F-ATPases employ a unique method of regulation whereby the enzyme can be autoinhibited with the integral subunit e. In all three rotational states of the E. coli F-ATPase, the eCTD had an extended conformation, albeit with a different proportion of particles observed at each state (46%, 30% and 24%) (Figure 2-figure supplement 3), suggesting State one to be the lowest energy. No reconstruction at any stage of data processing contained density corresponding to a closed/down conformation of the eCTD, and this along with the strong stimulation of ATPase activity by LDAO, suggests that the majority of the protein to be in an autoinhibited form. Although the position of eCTD relative to the ab subunits was similar to that of the mitochondrial inhibitor protein (IF1), the observation that it bound to all three states was different to that seen for IF1 that is bound to a single rotational F 1 state (a/b DP site proximal to the peripheral stalk) in the F 1 F o ATP synthase dimer structure (Hahn et al., 2016). The cryo-EM maps resembled the 'open, closed, open' conformation as seen in the Bacillus PS3 F 1 crystal structure, despite different nucleotide binding positions. However, the major contacts formed by the F 1 motor with the eCTD in our maps were similar to that of the E. coli crystal structure, except that one b subunit changed conformation substantially (Figure 3-figure supplement 1). Because our cryo-EM study was performed in the absence of externally added nucleotide, it is likely that the structures correspond to the autoinhibited conformations in solution, and the crystal structure of the isolated E. coli F 1 could instead represent a partially bound state, especially since the crystals were soaked in 1 mM AMPPNP prior to freezing (Cingolani and Duncan, 2011).
The c-ring is responsible for the rotation of the complex and contains the conserved carboxylate that binds the proton. Different species have varying numbers of subunits in their ring, believed to 'gear' the motor tailoring them to their environment and ranging from 8 to 15 subunits (Stock et al., 1999;Pogoryelov et al., 2007;Watt et al., 2010;Stewart et al., 2013). Although the density corresponding to the cring was quite weak, 10 peaks can be discriminated in the density at either end of the ring ( Figure 6-figure supplement 2). This confirmed the c-ring stoichiometry of E. coli F-ATPase that had previously been suggested to be decameric by crosslinking (Ballhausen et al., 2009), fusion (Jiang et al., 2001) and single molecule analysis (Düser et al., 2009;Ishmukhametov et al., 2010).
The model of subunit a generated from our cryo-EM maps confirmed that the F o motor likely operates using two half channels separated by a conserved arginine that directs its rotation (Vik and Antonio, 1994;Junge et al., 1997). Importantly, in this context, our model placed Arg210, which is believed to mediate the rotation of the c-ring, adjacent to the conserved carboxylate residue, Asp61, of the c subunit that has been shown to bind protons (Vik and Video 2. View of F o motor during ATP synthesis. Same as main text Figure 6b, but with rotating c-ring in the foreground. DOI: 10.7554/eLife.21598.028 Antonio, 1994;Pogoryelov et al., 2010) (Figure 6b and Video 2). In addition Gln252, which can be substituted with Arg210 and retain ATP synthase function (Ishmukhametov et al., 2008;Hatch et al., 1995), is positioned on a proximal helix with a similar distance to Asp61 of subunit c ( Figure 6B and Figure 6-figure supplement 5). Moreover, mainly charged residues, that have been shown to be aqueous accessible , map to two channel-like areas exposed to solvent by the invagination of the detergent micelle ( Figure 6B and Figure 6-figure supplement 5). Furthermore, residue A217, which has been shown to be sensitive to arginine mutation, and therefore been suggested to be near or part of the aqueous pocket (Cain and Simoni, 1989), is positioned next to the periplasmic half channel ( Figure 6B and Figure 6-figure supplement 5). Additionally. residues E219 and H245, which when substituted for one another result in a functional enzyme, are proximal to one another in our model (Cain and Simoni, 1988) (Figure 6-figure supplement 5). Further analysis of inter subunit crosslinking fits our model well (Figure 6-figure supplement 3), with the distances between the c-ring and a subunit being minimal.
In summary, our models show a new level of detail for the bacterial F-ATPase, providing a template for further experiments as well as to guide future antibiotic discovery in related pathogenic bacteria.

Protein purification
A cysteine-free version of E. coli F-ATPase cloned in plasmid pFV2 and expressed in E. coli DK8 strain was used (Ishmukhametov et al., 2005). Cells were grown at 37˚C in LB medium supplemented with 100 mg/ml ampicillin, for 4-5 hr. The harvested cells were resuspended in lysis buffer containing 50 mM Tris/Cl pH 8.0, 100 mM NaCl, 5 mM MgCl 2 , 0.1 mM EDTA, 2.5% glycerol and 1 mg/ml DNase I and processed by one pass in French press at 20 kPSI. Cellular debris was removed by centrifuging at 7700 Â g for 15 min, and the membranes were collected by ultracentrifugation at 100,000 Â g for 1 hr. The ATP synthase complex was extracted from membranes at 4˚C for 1 hr by resuspending the pellet in extraction buffer consisting of 20 mM Tris/Cl, pH 8.0, 300 mM NaCl, 2 mM MgCl 2 , 100 mM sucrose, 20 mM imidazole, 10% glycerol, 4 mM digitonin and EDTA-free protease inhibitor tablets (Roche). The complex was then purified by binding on Talon resin (Clontech) and eluted in 150 mM imidazole. The protein was further purified and sugars removed by size exclusion chromatography on a 16/60 Superose six column equilibrated in a buffer containing 20 mM Tris/Cl pH 8.0, 100 mM NaCl, 4 mM digitonin and 2 mM MgCl 2. The purified protein was then concentrated to 2 mg/ml for cryo-EM.

Protein reconstitution into proteoliposomes
Seventy microgram of F 1 F o was reconstituted into extrusion-preformed 100 nm soybean phosphatidylcholine liposomes exactly as descried (Ishmukhametov et al., 2016).

Functional assays
Proton pumping by proteoliposomes was studied using quenching of a pH sensitive fluorescent probe 9-Amino-6-Chloro-2-Methoxyacridine (ACMA) exactly as described (Ishmukhametov et al., 2016). The assay was performed with 100 ml of proteoliposomes in 2-ml cuvettes. The reaction was started with 0.25 mM ATP and stopped by 2 mM of the uncoupler FCCP.
ATP hydrolase activity and its stimulation by LDAO was measured with ATP regenerating system using 5 mg of the protein with 1 mM ATP exactly as described (Ishmukhametov et al., 2016).
DCCD inhibition of proteoliposomes was done as described (Ishmukhametov et al., 2016). DCCD inhibition of pure F 1 F o was done as described (Ishmukhametov et al., 2005), with the following modification. Ten microgram of the protein was incubated in 1 ml buffer A (50 mM MES, pH 6.4, 100 mM KCl, 1 mM MgCl 2 ) with 50 mM DCCD for 30 min at room temperature. Control sample contained 1% ethanol instead of DCCD. Reaction was started by mixing the inhibited protein with 1 ml of buffer A containing all the components of ATP regenerating system.
All the functional experiments presented here were repeated two to three times using the protein isolated by MS and shipped to RI at liquid N 2 temperature. Results of typical experiments are shown.

Cryo-EM grid preparation and data collection
Aliquots of 4 ml of purified E. coli F-ATPase at a concentration of 3.58 mM were placed on glow-discharged holey carbon grids (Quantifoils Copper R2/2, 200 Mesh). Grids were blotted for 2 s and flash-frozen in liquid ethane using an FEI Vitrobot Mark IV. Grids were transferred to an FEI Titan Krios transmission electron microscope that was operating at 300 kV. Images were recorded automatically using the FEI EPU software, yielding a pixel size of 1.4 Å . A dose rate of 29 electrons (spread over 20 frames) per Å 2 per second, and an exposure time of 2 s were used on the Falcon-II detector. 8640 movies were collected.

Data processing
MotionCorr (Li et al., 2013) was used to correct local beam-induced motion and to align resulting frames. Defocus and astigmatism values were estimated using CTFFIND4 , and 252 micrographs were excluded due to drift or excessive ice contamination. 1208 particles were manually picked and subjected to 2D classification to generate templates for autopicking in RELION (Scheres, 2012). The automatically picked micrographs were manually inspected to remove false positives, finally yielding 395,140 particles. These particles then underwent two rounds of 2D classification to generate 22 classes with 311,887 particles. The final particles were classified into four 3D classes using a previously generated model from a low-resolution data set of the same sample (unpublished), low-pass filtered to 60 Å . The resolution was estimated using Fourier Shell Correlation (FSC = 0.143, gold-standard). Three of the four classes containing 104,510 (State 1), 67,829 (State 2) and 53,587 (State 3) particles were movie-refined and post-processed in RELION producing maps at 7.4, 7.8, 8.5 Å , respectively (Figure 2-figure supplement 10). State 1 was further processed using masked classification (Bai et al., 2015) with residual signal subtraction with a mask created by removing parts of the detergent micelle. Three out of the four classes from this classification containing 95,345 particles were combined and refined to generate the final 6.9 Å map. Figure 2-figure supplement 3 is a summary of these methods, shown as a flowchart. Local resolution of different parts of the complex was estimated using RELION and ResMap (Kucukelbir et al., 2014).

Model building
Crystal and NMR structures of subunits from E. coli (abge -3oaa [Cingolani and Duncan, 2011], d -2a7u [Wilkens et al., 2005], b -1b9u [Dmitriev et al., 1999], 1l2p [Del Rizzo et al., 2002] and 2khk [Priya et al., 2009]) and related organisms (c -3u2f [Symersky et al., 2012] and a -5fik [Zhou et al., 2015]) were rigid body docked into the highest resolution cryo-EM map and the side chains 'pruned' to Ca. The sequence was mapped to subunit a using crosslinks as restraints. Subsequent manual model building and refinement was performed with Coot (Emsley et al., 2010), Phenix (Adams et al., 2010) and Refmac (Murshudov et al., 2011) (excluding subunit c due to weak density), with crosslinks again used as external restraints (a summary of the types of models used to build the initial model can be found in Figure 2-figure supplement 6). Nucleotide occupancy was determined by first building the model without any nucleotide present, and then segmenting the map and selecting any density with 15% overlap with atoms and deleting this density. Nucleotide was subsequently docked into this difference density using the known positions from previous structures. Once a complete model was built of the highest resolution map, this was docked and refined to the other two maps to create three models. The three models and maps were deposited in the pdb and emDataBank with codes 5T4O, EMD-8357 (State 1), 5 T4P, EMD-8358 (State 2), 5T4Q and EMD-8359 (State 3). Data statistics shown in Figure 2-source data 1. Alastair G Stewart The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Major datasets
The following datasets were generated: