Cryo-EM reveals distinct conformations of E. coli ATP synthase on exposure to ATP

ATP synthase produces the majority of cellular energy in most cells. We have previously reported cryo-EM maps of autoinhibited E. coli ATP synthase imaged without addition of nucleotide (Sobti et al. 2016), indicating that the subunit ε engages the α, β and γ subunits to lock the enzyme and prevent functional rotation. Here we present multiple cryo-EM reconstructions of the enzyme frozen after the addition of MgATP to identify the changes that occur when this ε inhibition is removed. The maps generated show that, after exposure to MgATP, E. coli ATP synthase adopts a different conformation with a catalytic subunit changing conformation substantially and the ε C-terminal domain transitioning via an intermediate ‘half-up’ state to a condensed ‘down’ state. This work provides direct evidence for unique conformational states that occur in E. coli ATP synthase when ATP binding prevents the ε C-terminal domain from entering the inhibitory ‘up’ state.


Introduction
The majority of metabolic energy in cells is generated by F 1 F o ATP synthase, a biological rotary motor that converts the proton motive force (pmf) to adenosine tri-phosphate (ATP) in both oxidative phosphorylation and photophosphorylation Walker, 2013). The enzyme consists of two reversible rotary motors, termed F 1 and F o , coupled together by one central and one peripheral stalk, with the simplest subunit composition found in bacteria such as Escherichia coli (Figure 1-figure supplement 1). The F o motor spans the membrane and converts the potential energy of the pmf into rotation of the central stalk that in turn drives conformational changes in the three catalytic sites of the a 3 b 3 F 1 motor subunits to generate ATP. Moreover, this process is reversible so that, if the pmf drops below the threshold needed to power ATP synthesis, the motor has the ability to reverse and, in some bacterial species, operates primarily as a proton pump driven by ATP hydrolysis. Regulation of these ATPase/synthase activities is particularly important in times of cellular stress, primarily to prevent wasteful ATP consumption. However, different regulatory mechanisms are used by different F 1 F o subtypes (bacterial, chloroplastic, mitochondrial) and species (Sielaff et al., 2018;Morales-Rios et al., 2015;Gledhill et al., 2007;Hahn et al., 2018;Stewart, 2014;Pullman and Monroy, 1963).
It has been postulated that bacterial ATP synthases are regulated by several mechanisms, with conformational changes of the e subunit and catalytic nucleotide occupancies likely playing major roles (Sielaff et al., 2018;Hyndman et al., 1994;Feniouk et al., 2006). A long-standing question surrounds the role of the C-terminal domain of the e subunit (eCTD). The eCTD is known to change conformation and block rotation of the central stalk in the F 1 motor from several bacteria, but this role is absent in the enzyme of mitochondria (Sielaff et al., 2018;Laget and Smith, 1979;Cingolani and Duncan, 2011;Sobti et al., 2016;Yagi et al., 2007;Shirakihara et al., 2015). In the active bacterial enzyme, the eCTD is proposed to adopt a condensed or 'down' conformation (Krah et al., 2017), and this state has been observed for isolated e subunit from E. coli (Uhlin et al., 1997;Wilkens and Capaldi, 1998a) and Geobacillus stearothermophilus (or Bacillus PS3, hereafter termed PS3) (Yagi et al., 2007) as well as in isolated F 1 from Caldalkalibacillus thermarum (Ferguson et al., 2016). In autoinhibited states observed in crystal structures of F 1 from E. coli and PS3, the eCTDs are extended in similar 'up' conformations, so that a helix inserts into the F 1 central cavity and appears to block rotation of the complex by binding to both the central rotor subunit g and several surrounding a and b subunits (Sielaff et al., 2018;Cingolani and Duncan, 2011;Mendoza-Hoffmann et al., 2018) (Figure 1-figure supplement 2). However, there is a key structural difference between these 'up' states: in E. coli, the eCTD consists of two helices (here termed eCTH1 and eCTH2; see Figure 1-figure supplement 2) separated by an extended loop, with each helix interacting with a different region of the g subunit (Cingolani and Duncan, 2011;Rodgers and Wilce, 2000); whereas in PS3, eCTH1 and eCTH2 instead join to form one continuous helix (Shirakihara et al., 2015). In PS3 and a related Bacillus species, ATP is known to bind to the e subunit and stabilise the 'down' conformation (Yagi et al., 2007;Krah et al., 2017;Kato-Yamada and Yoshida, 2003;Kato-Yamada, 2005;Imamura et al., 2009), and a clear mechanism of regulation by the eCTD and ATP levels has been proposed (Shirakihara et al., 2015). However, the physiological importance of the eCTD in ATP synthases of other bacteria, such as E. coli, has remained uncertain. Crosslinking studies have suggested that the eCTD of E. coli ATP synthase could act as a ratchet, preventing hydrolysis -but not synthesis -of ATP (Tsunoda et al., 2001). However, deletion studies performed in vivo, in which the E. coli eCTD was removed, showed minimal impact on cell growth or viability (Shah and Duncan, 2015), although deletion of just the five terminal residues results in decreased respiratory growth due to increased inhibition of ATP synthesis (Taniguchi et al., 2011). For the highly latent enzyme from C. thermarum, an early study indicated the eCTD is involved in inhibition (Keis et al., 2006), but thus far crystallographic studies of its F 1 complex have shown the eCTD in the down conformation irrespective of ATP binding to e (Ferguson et al., 2016). Despite confusion over the possible importance of the eCTD for the bacteria noted above, one recent study demonstrated that the eCTD can impact the virulence of Streptococcus pneumoniae. F 1 F o is essential for viability of S. pneumoniae and functions in pH homeostasis and in its acid tolerance response (Ferrándiz and de la Campa, 2002;Cortes et al., 2015). In a mouse model of pneumococcal bacteraemia, a frameshift mutation that scrambled the sequence of the eCTD was found to increase pneumococcal survival in the spleen, most likely by reducing killing by splenic macrophages (Gerlini et al., 2014). Therefore, it is important to characterize more fully the bacterial eCTD and its different conformational states and interactions within bacterial ATP synthases.
In contrast to the crystallographic reports mentioned above with isolated e subunits or soluble F 1 , our previous cryo-EM study (Sobti et al., 2016) provided maps of the intact E. coli ATP synthase. These data were obtained using purified protein without the addition of exogenous nucleotide, thus likely representing a state autoinhibited by e, here termed ATP synthase AI . Here we have used similar methodologies to investigate the effect of ATP on the conformation of E. coli F 1 F o ATP synthase, by cryo-freezing and subsequently imaging the enzyme after a brief incubation with 10 mM MgATP. The maps obtained (termed ATP synthase +ATP hereafter) indicate that, in the presence of MgATP, the eCTD became detached from the body of the enzyme, coupled with a large change in one of the three catalytic b subunits. Further analysis identified an intermediate in which the eCTD was in a 'half-up' conformation, suggesting that the enzyme can still rotate, even when the e subunit is not in its condensed down position, as well as an eCTD 'down' conformation likely showing an active intermediate. Further inspection of our maps demonstrated peaks present in all six nucleotide-binding sites in the ATP synthase +ATP enzyme (three catalytic b sites and three noncatalytic a sites), in contrast to the autoinhibited protein in which only four binding pockets (a single catalytic b site and three noncatalytic a sites) were occupied with nucleotide. Together with our previous work, these results reveal three distinct positions taken up by the eCTD in response to different catalytic nucleotide-bound states, providing further insight into the mechanism of ATP synthase regulation in E. coli and related species.

Results
Structure of E. coli F 1 F o ATP synthase in the presence of ATP To determine the structure of the E. coli F 1 F o ATP synthase in the presence of ATP, the detergentsolubilized E. coli enzyme (Sobti et al., 2016) was incubated with 10 mM ATP (containing 50 mM ADP [n = 3; stdev = 2 mM]) and 10 mM MgCl 2 for 30 s at 20˚C prior to preparing grids for cryo-EM. Sample preparation took a further 15 s before cryogenic temperature was reached, with approximately 0.25 mM ATP hydrolysed to ADP in the total time for preparation (45 s) (Figure 1-figure supplement 3). The time used for these experiments was optimized to be within the linear range of ADP production, so that uninhibited ATP synthase could be observed, with an ADP concentration of~0.3 mM at the time of freezing. 10 mM ATP and a 45 s preparation time were chosen to emulate the concentrations seen in E. coli growing under aerobic conditions, where 9.6 mM ATP and 0.6 mM ADP has been observed (Yaginuma et al., 2014;Bennett et al., 2009).
The grids produced after incubation with ATP were similar to those used in our previous study of the autoinhibited enzyme and micrographs showed clear ATP synthase particles. A dataset of 8509 movie micrographs was collected at 200 kV accelerating voltage, of which 7858 were selected based on image quality. 579,942 particles were picked and 319,315 of these appeared to be of the intact complex ( Figure 1-figure supplement 4). Three different rotational conformations could be filtered from these data (each containing 97,095, 72,757 and 51,534 particles) that were related by a 120˚rotation of the central stalk, displaying the hallmark dynamic movements predicted for ATPases/synthases . The quality of the maps obtained for each conformation was comparable, with resolutions of 5.0 Å , 5.3 Å and 5.5 Å (Figure 1a and Figure 1-figure supplements 5 and 6). However, the proportion of particles corresponding to each reconstruction differed from our previous study (33% State 1, 44% State 2 and 23% State 3 in this study, compared with 46% State 1, 30% State 2 and 24% State 3 in Sobti et al., 2016), suggesting that a larger number of molecules were now observed in State 2 rather than State 1. Single molecule studies have also observed the enzyme populating the rotational dwell states in different proportions (Sielaff et al., 2019), suggesting that each dwell state may be at a different energy minimum. Inspection of the nucleotide-binding pockets in the a and b subunits revealed two further peaks, in addition to the four nucleotide peaks we reported previously for the autoinhibited enzyme (Sobti et al., 2016)  Re-refinement of the ATP synthase AI maps presented in our previous study (Sobti et al., 2016) improved their nominal resolution to 5.7 Å , 6.6 Å and 7.2 Å (previously 6.9 Å , 7.8 Å and 8.5 Å , respectively), and enabled features to be seen more clearly (Figure 1b and Figure 1-figure supplement 8). The re-refined maps were used to compare ATP synthase before and after incubation with ATP. Compared with ATP synthase AI , the ATP synthase +ATP reconstructions showed a large conformational change in one of the three b subunits (b DP ) and that the eCTD was absent from the maps (Figure 1). The conformation of the catalytic subunits in the ATP synthase +ATP map closely resembles that of the isolated F 1 enzyme from C. thermarum (Ferguson et al., 2016) (Figure 1-figure supplement 9), except that no density was detectable for the eCTD. Even inspecting the unmasked unfiltered half maps at low threshold, it was difficult to formulate a clear idea of the position of the eCTD, suggesting that under these conditions it may be sampling multiple conformations. To identify the position of the eCTD, the data were processed more extensively and refined into sub-classifications ( Figure 1-figure supplement 6). Although the resolution obtained in these subclasses was lower than in the general classification and they displayed the same rotational states, they showed additional weak density corresponding to the eCTD in either a condensed 'down' position ( Figure 2a and In the 'down' position, weak density was present for both eCTH1 and eCTH2 with the helices arranged adjacent to one another, attached to the N-terminal region of the e subunit. However, in the intermediate 'half-up' position weak density was only observed for the eCTH1 bound to the g subunit, with no identifiable density for the eCTH2, suggesting that this helix was mobile.

Peripheral stalk structure
The peripheral stalk of ATP synthase is constructed from a right-handed coiled coil and holds the catalytic a and b subunits stationary relative to a rotating central stalk (Wilkens and Capaldi, 1998b;Lee et al., 2010;Walker and Dickson, 2006). Our previous maps (Sobti et al., 2016) described the overall peripheral stalk architecture in E. coli F 1 F o ATP synthase, to which density was attributed as belonging to two b subunits and one d subunit. In light of recent higher resolution cryo-EM maps of the eukaryotic F 1 F o ATP synthase from Spinacia oleracea chloroplasts (Hahn et al., 2018), we attempted to assign the density of the peripheral stalk more precisely in the new maps. Of particular interest was a small density peak (Figure 3a-c, pink density) that cannot be identified in other structures of related ATP synthases (Morales-Rios et al., 2015;Hahn et al., 2018). However, this density can be identified in all cryo-EM maps produced of E. coli F 1 F o ATP synthase thus far (EMD-8357, EMD-8358, EMD-8359 and the maps presented in this study). The length of the density relating to Figure 2. Two conformations of subunit e are observed in sub-classification of ATP synthase +ATP maps. Processing and refinement of the ATP synthase +ATP dataset using Relion (Scheres, 2012) identified two conformations for the eCTD. These show a 'down' conformation (a) where the eCTH1 and eCTH2 are bound to the N-terminal region of the e subunit and a 'half-up' conformation (b) where the eCTH1 was still bound to the g subunit but the eCTH2 was not visible even at low thresholds. DOI: https://doi.org/10.7554/eLife.43864.015 The following figure supplement is available for figure 2: the b subunits was measured to investigate whether this additional density could be attributed to the b subunits, and found to be 233 Å and 227 Å (termed b 1 and b 2 , respectively; Figure 3d). Together with the prediction that residues 66-122 of the b subunits form an unstaggered coiled coil in the isolated dimer (Del Rizzo et al., 2002), this suggests that the additional density peak represents the C-terminus of just the b 2 subunit. The differing conformation between the b 1 and b 2 subunits seen here appears to be unique to some bacterial ATP synthases and is likely due to their being a homodimer in these systems. Other related F-ATP synthases, such as Paracoccus denitrificans, contain two different b subunits (termed b and b') that differ in length and sequence (Morales-Rios et al., 2015) and therefore can adopt their own individual folds. In E. coli ATP synthase, each b subunit is required to adopt two different conformations using the same amino acid sequence. This may be due to the requirement of the homodimer binding to asymmetric a and d surfaces, whereby the b 1 subunit forms the majority of the contacts between the F 1 motor while the b 2 subunit folds back onto itself providing rigidity to the peripheral stalk. Most recently, a high-resolution structure of PS3 ATP synthase has been reported using cryo-EM (Guo and Rubinstein, 2018). Although the maps and models are not currently available, similar density can be seen suggesting that the break in symmetry of the b subunits observed at the top of the F 1 motor is conserved across species which contain a homodimeric b subunit. Moreover, analysis of our improved maps of the E. coli ATP synthase AI (particularly State 2, which shows the clearest density in the F o region) corroborates the results from the far higher resolution structure of the PS3 enzyme, showing a similar overall fold in the membrane F o region to spinach chloroplast ATP synthase (Hahn et al., 2018), although following slightly different trajectories (Figure 3-figure supplement 1). However, because the maps of the E. coli enzyme show only weak density in this region it is difficult to draw a firm conclusion on this point.

Discussion
The structural information obtained here using E. coli F 1 F o ATP synthase incubated with MgATP likely reflects conformations uninhibited by the eCTD. Although the maps generated are of limited resolution, the position of the catalytic subunits and eCTD were clearly identifiable. Two different eCTD conformations were observed: a 'down' conformation, in which eCTH1 and eCTH2 are arranged next to one another and interacts with the N-terminal region of the e subunit, and a 'halfup' conformation in which the eCTH1 lies against the g subunit (Figure 2). Because the ATP and ADP concentrations present were comparable to those observed in intact cells (Bennett et al., 2009) it is likely that both these conformations are present in E. coli. Furthermore, analysis of the maps in light of the recent high-resolution structure of related S. oleracea enzyme has highlighted a unique C-terminal fold of the peripheral stalk b subunits in E. coli ATP synthase that derives from the homodimeric nature of this subunit in this system.
A considerable body of work has described the inhibition of F 1 -ATPases by ADP, Mg 2+ and eCTD (Sielaff et al., 2018;Hyndman et al., 1994;Laget and Smith, 1979;Cingolani and Duncan, 2011;Sobti et al., 2016;Richter, 2004;Zhou et al., 1988;Avron, 1962;Drobinskaya et al., 1985;Guerrero et al., 1990). Because the sample in the present study was imaged under conditions in which ADP accumulated at a linear rate (Figure 1-figure supplement 3), we propose that the maps represent structures of E. coli F 1 F o ATP synthase in which the autoinhibitory state of the eCTD is prevented. These results indicate that ATP induces the eCTD to exit from the central cavity of F 1 and so facilitate rotation of the enzyme. The contribution of the eCTD to regulate E. coli F 1 F o ATP synthase activity has previously been unclear. Our structural studies on E. coli F 1 F o ATP synthase support a mechanism whereby the eCTD is primed to act as a 'safety lock" (Feniouk and Junge, 2005) or 'ratchet" (Tsunoda et al., 2001;Cipriano and Dunn, 2006) to prevent ATP synthase from entering the ADP/Mg 2+ inhibited state (Shah et al., 2013) even when ADP and Mg 2+ concentrations are high (Figure 4). The 'half-up' state that we identified in our cryo-EM maps has been previously hypothesised (Cingolani and Duncan, 2011) and studies on the enzyme in the absence of the eCTH2 support the hypothesis that the eCTD 'half-up' conformation can act as a weak inhibitor, partially inhibiting without insertion into the F 1 motor (Nakanishi-Matsui et al., 2014). It is unlikely that the 'half-up' conformation we observe here is an artefact generated by the cysteine residue free construct used in this study, because in our previous work (Sobti et al., 2016) the same cysteine-free enzyme did not show the 'half-up' state. In addition, previous studies using gC87S mutant protein did not show any change in e inhibition (Duncan et al., 1995), and in the structure of the native enzyme ( Figure 1-figure supplement 10) loops in the g subunit separate the cysteines in subunit g and residues in the eCTH1, so that they are not in direct contact.
Structural data from this and previous studies suggest that loading multiple catalytic sites with nucleotides disrupts access of the eCTD to the inhibitory 'up' state in E. coli (Figure 4). In a nucleotide-free solution, the F 1 -ATPase adopts an autoinhibited state (seen in the crystal structure of isolated E. coli F 1 -ATPase (Cingolani and Duncan, 2011) and the ATP synthase AI cryo-EM map (Sobti et al., 2016); Figure 4a and b). The isolated F 1 -ATPase shows one b subunit (b DP ) in a 'halfclosed' state, with ADP bound, which interacts directly with the 'up' conformation of the eCTH2 (Figure 4a). In the ATP synthase AI cryo-EM map, the same b subunit is unoccupied and has adopted an open conformation that no longer interacts with the eCTH2, which could facilitate a reversible mechanism whereby the helix is able to escape from its inserted position. When compared with isolated F 1 , the reduced interactions with the 'up' state of subunit e seen in the complete F 1 F o complex may explain why e inhibition is significantly reduced upon rebinding of E. coli F 1 to F o . In the ATP synthase +ATP cryo-EM maps, all catalytic sites are at least partially occupied with nucleotide (Figure 1-figure supplement 7) and the eCTD is seen in two conformations that no longer contact the a and b subunits (Figure 4c and d). Because we only observed the eCTD in the 'up' conformation in the absence of ATP, these results indicate that nucleotide binding to the catalytic subunits, in either one or a combination of catalytic sites, is important for the release of the eCTD. Binding of ATP is unlikely to accelerate release of the eCTD from the 'up' state, as that would conflict with the known noncompetitive nature of e inhibition with E. coli F 1 and F o F 1 . Rather, as the eCTH2 escapes from the central cavity of the ATP synthase AI conformation (Figure 4b), rapid binding of ATP would induce a transition from the open (Figure 4b, 'O') to the closed conformation (Figure 4c/d, 'C'), preventing re-entry of eCTH2 until the next catalytic dwell position. Finally, the 'half-up' conformation seen for the eCTD supports the prediction that the eCTH2 can remain mobile, anchored to g by the eCTH1, so that it can re-enter the central cavity at a later step in the cycle (Cingolani and Duncan, 2011). It is also possible that this eCTH1 interaction alone may be able to modulate enzyme activity, as seen in a previous deletion study (Nakanishi-Matsui et al., 2014).
The 'down' conformation of the e subunit has been observed for the isolated e subunit of PS3 with ATP bound (Yagi et al., 2007) and on F 1 from C. thermarum (Ferguson et al., 2016). For ATP synthase +ATP cryo-EM maps containing the eCTD in the 'down' conformation, close inspection showed very weak density, consistent with a fraction of these complexes showing nucleotide bound to the e subunit. Docking of the C. thermarum (Ferguson et al., 2016) e subunit with ATP bound, showed good correlation between the position occupied by the nucleotide and additional density seen in the ATP synthase +ATP map (Figure 1-figure supplement 11). Partial occupancy would be . Proposed mechanism for the release of eCTD from the inhibitory 'up' conformation in E. coli F 1 -ATPase. Structural models presented as maps (above; E. coli crystal structure in (a) presented as low smooth map for comparison) and simple schematics (below) to describe the possible structural movements of E. coli F 1 -ATPase during activation. In a nucleotide-free solution, the F 1 -ATPase adopts an inactive state (seen in the crystal structure of isolated E. coli F 1 -ATPase (Cingolani and Duncan, 2011) and the ATP synthase AI cryo-EM map (Sobti et al., 2016;(a) and (b) respectively). The isolated F 1 -ATPase shows one of the b subunits in a 'half-closed' state (labelled as HC), interacting with the 'up' conformation of the eCTH2 (a). In the cryo-EM maps of the intact enzyme (b), the same b subunit adopts an open conformation (labelled as O) where it no longer interacts with the eCTH2, allowing the reversible release of the helix from its inserted position. In the presence of ATP, the eCTD is seen in two conformations (c and d) that no longer contact the a and b subunits, and the aforementioned b subunit then adopts a closed state (labelled as C) that prevents the eCTD from returning to the 'up' conformation. The eCTD can either be in a 'half-up' (c) conformation, with the eCTH2 mobile, or a 'down' conformation (d) with both eCTH1 and eCTH2 in a condensed form. DOI: https://doi.org/10.7554/eLife.43864.020 consistent with the concentration of ATP employed in our study being slightly below the K d of~22 mM observed for ATP binding to isolated E. coli e subunit (Yagi et al., 2007), and so because weak density attributable to nucleotide bound to the eCTD was only seen with the down conformation, it is possible that the 'half-up' conformation, with the associated probable disorder of eCTH2, might result when nucleotide is not bound to the e subunit. Further work, for example, using a range of ATP concentrations, will be needed to resolve this point.
Although the movements of the eCTD are well defined by this study, the resolution of the reconstructions is limited compared to others (Guo and Rubinstein, 2018;Murphy et al., 2019). We believe this is likely due to inherent flexibility in the complex, possible varying subunit or lipid stoichiometry (Chorev et al., 2018), the detergent used, ice thickness, particle density and the lower accelerator voltage used during image acquisition (200 vs. 300 kV). Moreover, we only observed three rotational states whereas seven were seen in bovine mitochondrial ATP synthase (Zhou et al., 2015) and thirteen in Polytomella sp. mitochondrial ATP synthase (Murphy et al., 2019). We did employ methods similar to those used in these studies in an attempt to observe rotary sub-steps in the E. coli enzyme, but were unsuccessful. Although the complex should be undergoing rotation, due to the ATP hydrolysis that is occurring during freezing, single molecule studies have shown that in order for the complex to spend significant time in rotary sub-steps it had to be imaged under high drag, such as in 30% polyethylene glycol with a large object attached (Ishmukhametov et al., 2010). Indeed, the poor density seen at the membrane interface of the subunit b dimer, points towards these reconstructions containing multiple conformations, which may contribute to limiting the resolution obtained. It is possible that further work to obtain a much larger dataset, containing many more particles, may enable identification of more of the rotary sub-steps that are likely present in this sample.
Recent work on Polytomella sp. mitochondrial ATP synthase (Murphy et al., 2019) has described this ATP synthase in great detail, showing a flexible coupling between the F 1 and F o motors mediated by a hinge region in OSCP (analogous to the d subunit in E. coli ATP synthase). Comparison of the peripheral stalks in this study, by superposition of either the F o stator or the N-terminal region of subunit d (Figure 3-figure supplement 2), shows that the peripheral stalk likely bends and twists as the enzyme undergoes rotary catalysis, as suggested in previous studies (Hahn et al., 2018;. We do not observe movement in the delta/OSCP hinge, most likely due to our inability to resolve rotary substates in our maps. In previous studies on E. coli ATP synthase, the enzyme has been shown to become inactive four hours after purification, possibly due to aggregation (Ishmukhametov et al., 2010). In this study, the enzyme was purified rapidly (see Materials and methods) and EM grids were made shortly after purification to ensure active enzyme was being imaged. The images obtained showing free monodisperse protein with little evidence of aggregation. Furthermore, the enzyme was purified in digitonin rather than the mixture of phosphatidylcholine, octyl glucopyranoside, sodium deoxycholate and sodium cholate used previously (Ishmukhametov et al., 2010;Ishmukhametov et al., 2005). Digitonin was selected because enzyme purified in the original mixture of lipids and detergents was not stable on the size exclusion chromatography (SEC) employed as a final purification step prior to grid freezing. Because the enzyme had been purified in a different detergent and further purified using SEC, we assessed the activity of the digitonin solubilized enzyme using ATP regeneration assays after 0, 1, 2, 3, 4 and 8 hr to compare with data taken on enzyme purified in previous studies (Ishmukhametov et al., 2010) (Figure 1-figure supplement 12). Importantly the enzyme showed little change in activity over 8 hr, suggesting that E. coli ATP synthase purified with digitonin and SEC may be less prone to inactivation or aggregation.
The present study has exploited the ability to manipulate sample conditions immediately before freezing to obtain images of E. coli ATP synthase at a time when ADP formation from ATP proceeds linearly, and hence likely unhindered, to identify the possible conformational changes that occur to remove the eCTD from the autoinhibitory position. This outcome was achieved by optimizing experimental design, considering particle concentration, as well as mixing and blotting time. Interestingly, employing similar methods on the related bacterial A/V type ATPase from Thermus thermophilus, showed that no conformational changes occur in the catalytic or related F subunits upon addition of the ATP transition state mimic fluoroaluminate (Davies et al., 2017). This could be due to different nucleotide affinity, operating temperature or indeed function of the complexes (Nakano et al., 2008).
In summary, this study presents structural data of E. coli ATP synthase imaged in the presence of~9.75 mM ATP and~0.3 mM ADP, comparable to the concentrations present in the intact bacterium. Compared to our previous work on the autoinhibited enzyme in the absence of ATP, the cryo-EM reconstructions presented here show that a single b subunit and the C-terminal domain of subunit e change conformation and position substantially upon addition of ATP. Together, our studies provide strong support for the hypothesis that the e subunit acts as an emergency brake to minimize wasteful hydrolysis of cellular ATP in situations where the ATP-to-ADP ratio is low. As the eCTD has recently been shown to impact pathogenesis of at least one bacterial species (Gerlini et al., 2014), the different conformations of the eCTD observed in these studies may have potential to guide future development of drugs that could target this bacteria-specific inhibitory mechanism.

Materials and methods
ATP was purchased in the form of adenosine 5-triphosphate disodium salt hydrate (Sigma A7699).

Protein purification
The same preparation was used as in Sobti et al. (2016). However, the protein sample was concentrated to 11 mM (6 mg/ml) to improve particle density. An in-depth description of the methods used to isolate this protein is described below with approximate timings for each step.
Cysteine-free E. coli ATP synthase (with all cysteines residues substituted with alanine residues and a His-tag introduced on the b subunit) was expressed in E. coli DK8 strain (Ishmukhametov et al., 2005). Cells were grown at 37˚C in LB medium supplemented with 100 mg/ ml ampicillin for 5 hr. The cells were harvested by centrifugation at 5,000 g, obtaining~1.25 g cells per litre of culture [~20 min for centrifugation step]. 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 with three freeze thaw cycles followed by one pass through a continuous flow cell disruptor at 20 kPSI [~1 hr for lysis]. Cellular debris was removed by centrifuging at 7,700 Â g for 15 mins, and the membranes were collected by ultracentrifugation at 100,000 Â g for 1 hr [~1 hr 45 min for centrifugation steps]. 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) [~1.5 hr for resuspension]. Insoluble material was removed by ultracentrifugation at 100,000 g for 30 min. The complex was then purified by binding on Talon resin (Clontech) [~1 hr] and eluted in 150 mM imidazole [~1 hr for affinity step]. The protein was further purified with 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, 1 mM digitonin and 2 mM MgCl 2 [~1 hr for gel filtration step]. The purified protein was then concentrated to 11 mM (6 mg/ml) [~15 min for concentration step], and snap frozen and stored for grid preparation. The total time from membranes to freezing was~6 hr.

ATPase enzymatic assay
The reaction mixture was started by vigorously mixing 27 ml of 11 mM (6 mg/ml) purified cysteinefree E. coli F 1 F o ATP synthase and 3 ml of 100 mM ATP/100 mM MgCl 2 (pH 8.0) and incubated at 22˚C. 5 ml of the reaction mixture was removed at 15, 30, 45, 60, 120 and 240 s time points and stopped immediately by adding 25 ml of ice-cold 0.4 M perchloric acid and mixed vigorously with a pipette. Samples were subsequently spun at 1700 x g at 4˚C for 5 min. 30 ml of 1 M KH 2 PO 4 and 7.5 ml of 3 M KOH was added to neutralize the mixture before a further spin at 1700 x g for 5 min at 4˚C. The samples were then stored at À80˚C prior to analysis via HPLC-UV.

Analysis of ADP and ATP content by HPLC-UV
ADP and ATP in enzymatic reactions were detected as described previously in Micheli et al. (1993) with slight modifications. Briefly, 50 ml of sample was subjected to HPLC coupled with UV detection (Agilent 1200 series). ADP and ATP were separated on a Supelcosil C18 column (5 mM, 250 Â 4.6 mm) by gradient elution using mobile phase A (0.1 M KH 2 PO 4 pH 5.5 containing 6 mM tetrabutylammonium phosphate) and mobile phase B (100% methanol) at 1 mL/min. The gradient consisted of 7% mobile phase B (0-6 min) and 30% mobile phase B (6-11 min). The column was re-equilibrated with 7% mobile phase B for 9 min. To obtain the ADP concentration in commercially bought ATP, a 10 mM ATP sample was loaded, and ADP and ATP were detected as described previously in Huang et al. (2010) and subtracted from subsequent readings. Briefly, ADP and ATP were separated on a Supelcosil C18 column (5 mM, 250 Â 4.6 mm) by isocratic elution using 39 mM KH 2 PO 4 pH 6.5 at 1 mL/min. For both chromatographic methods, ADP and ATP were detected by UV 254nm and quantified by area comparison with authentic commercial standards (Sigma Aldrich, USA) and data analysed using ChemStation software (Agilent Technologies, USA).
Cryo-EM grid preparation 1 ml of 100 mM ATP/100 mM MgCl 2 (pH 8.0) was added to an aliquot of 9 ml of purified cysteinefree E. coli F 1 F o ATP synthase (Sobti et al., 2016;Ishmukhametov et al., 2005) at 11 mM (6 mg/ml) and the sample was incubated at 22˚C for 30 s, before 3.5 ml was placed on glow-discharged holey gold grid (Ultrafoils R1.2/1.3, 200 Mesh). Grids were blotted for 3 s at 22˚C, 100% humidity and flash-frozen in liquid ethane using a FEI Vitrobot Mark IV (total time for sample application, blotting and freezing was 15 s).

Data collection
Grids were transferred to a Thermo Fisher Talos Arctica transmission electron microscope operating at 200 kV. Images were recorded automatically using EPU, yielding a pixel size of 0.98 Å . A total dose of 50 electrons per Å (Walker, 2013) was used, with the first 30 electrons spread over the initial 31 frames and the final 20 electrons captured as the final frame. The total exposure time was 62 s and the data were collected using a Falcon III in counting mode. 8509 movie micrographs were collected in two data collections (Figure 1-figure supplement 6).

Data processing 200 kV ATP synthase +ATP dataset
MotionCorr2 (Zheng et al., 2017) was used to correct local beam-induced motion and to align resulting frames, with 5 Â 5 patches. Defocus and astigmatism values were estimated using Gctf (Zhang, 2016) and 7858 micrographs were selected after exclusion based on ice contamination, drift and astigmatism.~1000 particles from each data set were manually picked and subjected to 2D classification to generate templates for autopicking in RELION-3.0-beta (Scheres, 2012). The automatically picked micrographs were manually inspected to remove false positives, yielding 579,942 particles. These particles were re-extracted from motion corrected images that were dose-weighted and did not contain the final frame. These particles were then subjected to four rounds of 2D classification generating a final dataset of 319,315 particles. To reduce model bias, an independent initial map was made with RELION-3.0-beta using a subset of particles (13,461 particles) from the first data collection. Each dataset was classified into 3D classes using this initial model in RELION-3.0-beta, and similar classes were pooled yielding maps for ATP synthase +ATP , related by a rotation of the central stalk (total: 221,386 particles; three states: 97,095, 72,757 and 51,534 particles). Subclassifictions were merged based on whether the eCTD could be observed in the 'half-up' or 'down' state (45,446 and 40,262 particles respectively). Final refinements were performed in cryoSPARC (Punjani et al., 2017). See Figure 1-figure supplement 6 for a flowchart of the data processing strategy.

Map deposition
Maps for the ATP synthase +ATP , eCTD 'half-down' and eCTD 'down' were deposited to the EMDB with accession codes EMD-9345, EMD-9346 and EMD-9348 respectively. Maps for the three ATP synthase AI were updated in the EMDB.
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.