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

1In 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. 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 F1 and Fo, connected by central and peripheral stalks (Figure 1). The Fo 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 F1 catalytic sites.

2The Fo 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, 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 and the analogous V- and A-type ATPases as well as a low-resolution crystal structure of Paracoccus denitrificans F-ATPase are consistent with a two half-channel mechanism for the generation of rotation within the membrane. 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) through an aqueous half channel at the subunit a/c interface. 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. 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. The sequential binding of protons in combination with thermal fluctuations generates rotation within the complex in a manner akin to a turbine. The torque generated in the Fo motor is then transferred to the F1 motor by the central shaft consisting of subunits γ and ε. The N- and C-termini of subunit γ form a curved coiled-coil that extends into the central cavity of F1.

3The F1 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 α and β. Within the F1 motor, each αβ 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). These different catalytic states relate to the position of the curved coiled-coil of subunit γ in the central stalk, which drives the conformational changes associated with catalysis. To enable the central stalk to rotate relative to the F1 αβ heterodimers, the Fo and F1 motors need to be coupled. This coupling is mediated by the peripheral stalk that is constructed from subunits b and δ (Figure 1). Subunit b forms an amphipathic homodimeric coiled-coil that spans the periphery of the complex linking subunit a with subunit α, whereas subunit δ provides additional coupling of the C-termini of the b and α subunits.

4The bacterial F-ATPase can also function in reverse, employing ATP hydrolysis to generate a proton gradient across the membrane when needed. In vivo, subunit ε is believed to change conformation in an ATP-dependent manner to prevent rotation of the complex thereby conserving ATP when its concentration is low. This regulatory function is mediated by the C-terminal domain of subunit ε (εCTD) that, when not bound to ATP, opens to an extended conformation and inserts into the αβ heterodimers. The crystal structures of both the E. coli and Bacillus PS3 F1 motors in this autoinhibited state show the εCTD intercalating into the αβ heterodimers. However, in each structure, the F1 motor had been captured in a different conformation (E. coli – half-closed, closed, open and Bacillus PS3 – open, closed, open) which could either relate to inter species differences or crystal contacts and crystallization conditions.

5The crystal structure of the F-ATPase from P. denitrificans, that is closely related to E. coli (38% sequence identity over all subunits), shows a similar overall architecture to bovine F1Fo ATP synthase as well as to the main features of A/V ATPases. However, it is inhibited by the ζ-protein rather than by subunit ε, which is generally employed by bacterial F-ATPases for this purpose. Moreover, this crystal structure shows only one conformation of the rotary catalytic cycle.

6Here, 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 εCTD is in an extended conformation, stabilizing an overall F1 motor conformation similar to that seen in the thermophilic Bacillus PS3 F1 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 F1-delta interface, showing the three α 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.

7Cysteine-free E. coli F-ATPase, as described in where all 10 cysteines were replaced with alanines and a His-tag introduced on the β subunit, was solubilized in digitonin detergent and purified as described in the Materials and methods. This procedure provided pure protein (Figure 2—figure supplement 1) capable of ATP hydrolysis-driven proton pumping upon reconstitution into proteoliposomes (Figure 2—figure supplement 1). N,N`-dicyclohexylcarbodiimide (DCCD), a compound which selectively modifies Asp61 of subunit c at 50 µM completely abolished proton pumping (Figure 2—figure supplement 1B) and inhibited 90% of ATPase activity of isolated protein (Figure 2—figure supplement 1C). Such inhibition indicates coupling between the F1 and Fo motors.

8Protein 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. 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 (Figure 2A–C, and Figure 2—figure supplements 2 and 3). In these three conformations, the central stalk was progressively rotated 120° relative to the peripheral stalk.

9Even 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 β subunit, as well as helical and β sheet patterns observed in parts of the map in the F1 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 F1 motor to be of highest quality, the Fo 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 2D–F), with varying quality of the docked structures as indicated in Figure 2—figure supplement 6. The positions of the Cys-Ala mutants are depicted in Figure 2—figure supplement 7.

10The 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 and F1Fo ATP synthases from Bos Taurus, Yarrowia lipolytica and Polytomella, with the catalytic F1 motor attached to a proton powered membrane Fo motor and single central and peripheral stalks, differences in the individual motors and peripheral stalk were apparent. Comparison with the membrane-embedded 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 eight seen in mitochondrial F-type ATP synthase (Figure 2—figure supplement 8), consistent with labeling approaches. This difference suggested that the extra helices present in other rotary ATPase subtypes could have additional functions such as the dimerization seen in mitochondria.

11A movie generated by interpolation between the three states (Video 1) indicated that the F1 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/Fo interface (~bArg36) and one near the peripheral stalk/F1 interface (~bGln106), enabled the stalk to accommodate this eccentric movement of F1 (Figure 2—figure supplement 9).

12The maps showed a long right-handed coiled-coil dimer generated by the two b subunits of the peripheral stalk together with the globular δ subunit that anchors them to the catalytic head (Figure 3A). The quality of the map was sufficient to enable almost the entire of the δ subunit to be built as a polyalanine model (Figure 2D–F), whereas previous structural information was limited to the N-terminal domain. Interestingly, the peripheral stalk contacted all three α subunits via their N-terminal helices, but did so asymmetrically employing three different interfaces with each α 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 δ subunit binding in a single orientation across the top of the symmetrical αβ heterodimers.

13The 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. Furthermore, NMR analysis of the transmembrane domain of E. coli F-ATPase b subunit 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.

14All three reconstructions showed the complex in its autoinhibited state, with clear density for the εCTD extending deep into the central cavity of the F1 enzyme (Figure 3B). Fitting of the E. coli α3β3γε crystal structure into our cryo-EM maps showed that the β1 subunit had adopted a different more open conformation (Figure 3—figure supplement 1). In the above crystal structure, the εCTD contacts more subunits in F1 (α1, α2, β1, β2 and γ) compared to our cryo-EM reconstructions, where it contacted fewer subunits (α1, α2, β2 and γ). The conformation of our cryo-EM structure was more similar to that seen in the Bacillus PS3 structure, where it is proposed to be in an ‘open, closed, open’ conformation (Figure 5).

15To ascertain the enzymatic state of the F1 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 α contained nucleotide, as well as the ‘closed’ αβ 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 or the PS3 crystal structures.

16Single particle analysis revealed highly uniform homogeneity of the protein preparation, where 100% of F1Fo molecules observed demonstrated the extended conformation of εCTD. 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 ε was tested with N,N-dimethyldodecylamine N-oxide (LDAO), a well-known activator of the subunit ε 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, including earlier studies on the same cysteine-free F1Fo construct using the same batch of LDAO. Prior to addition of LDAO, the hydrolysis rate was 0.75 µmol 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 F1Fo was not described in the literature before and is consistent with our single particle data.

17Density in Fo 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 (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 c-ring to be decameric in E. coli F-ATPase. Furthermore, density inside the c-ring corroborates data suggesting it to be filled with phospholipids.

18By 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, 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).

19We 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 F1 and Fo motors, the inhibition mediated by the ε 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.

20Our reconstructions extended previous work by showing the structure of the complete peripheral stalk and how it is attached to both the F1 and Fo components. The peripheral stalk functions to counteract rotation of the F1 stator relative to the Fo stator as the central stalk rotates, but must also accommodate conformational changes in the F1 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, showed near parallel α-helices, based on an 11-residue hendecad sequence repeat, spanning the space between the Fo and F1 motors, that changed into a 15-residue quindecad sequence repeat along the F1 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, illustrating a strong evolutionary pressure for this fold and its function to prevent rotation between the αβ heterodimers and the a subunit while accommodating wobbling of the F1 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 F1 and Fo motors, enabling it to accommodate conformational changes in the catalytic head.

21In 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. Bacterial F-ATPases employ a unique method of regulation whereby the enzyme can be autoinhibited with the integral subunit ε. In all three rotational states of the E. coli F-ATPase, the εCTD 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 εCTD, 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 εCTD relative to the αβ 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 F1 state (α/βDP site proximal to the peripheral stalk) in the F1Fo ATP synthase dimer structure. The cryo-EM maps resembled the ‘open, closed, open’ conformation as seen in the Bacillus PS3 F1 crystal structure, despite different nucleotide binding positions. However, the major contacts formed by the F1 motor with the εCTD in our maps were similar to that of the E. coli crystal structure, except that one β 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 F1 could instead represent a partially bound state, especially since the crystals were soaked in 1 mM AMPPNP prior to freezing.

22The 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. Although the density corresponding to the c-ring 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, fusion and single molecule analysis.

23The model of subunit a generated from our cryo-EM maps confirmed that the Fo motor likely operates using two half channels separated by a conserved arginine that directs its rotation. 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 (Figure 6b and Video 2). In addition Gln252, which can be substituted with Arg210 and retain ATP synthase function, 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, 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 (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.

24In 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.

25The three models and maps were deposited in the pdb and emDB with codes 5T4O, EMD-8357 (State 1), 5 T4P, EMD-8358 (State 2), 5T4Q and EMD-8359 (State 3).

26A cysteine-free version of E. coli F-ATPase cloned in plasmid pFV2 and expressed in E. coli DK8 strain was used. Cells were grown at 37°C in LB medium supplemented with 100 μg/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 MgCl2, 0.1 mM EDTA, 2.5% glycerol and 1 μg/ml DNase I and processed by one pass in French press at 20 kPSI. Cellular debris was removed by centrifuging at 7700 × g for 15 min, and the membranes were collected by ultracentrifugation at 100,000 × 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 MgCl2, 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 MgCl2. The purified protein was then concentrated to 2 mg/ml for cryo-EM.

27Seventy microgram of F1Fo was reconstituted into extrusion-preformed 100 nm soybean phosphatidylcholine liposomes exactly as descried.

28Proton pumping by proteoliposomes was studied using quenching of a pH sensitive fluorescent probe 9-Amino-6-Chloro-2-Methoxyacridine (ACMA) exactly as described. The assay was performed with 100 µl of proteoliposomes in 2-ml cuvettes. The reaction was started with 0.25 mM ATP and stopped by 2 µM of the uncoupler FCCP.

29ATP hydrolase activity and its stimulation by LDAO was measured with ATP regenerating system using 5 µg of the protein with 1 mM ATP exactly as described.

30DCCD inhibition of proteoliposomes was done as described. DCCD inhibition of pure F1Fo was done as described, 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 MgCl2) with 50 µM 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.

31All the functional experiments presented here were repeated two to three times using the protein isolated by MS and shipped to RI at liquid N2 temperature. Results of typical experiments are shown.

32Aliquots of 4 μl of purified E. coli F-ATPase at a concentration of 3.58 μM 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 Å2 per second, and an exposure time of 2 s were used on the Falcon-II detector. 8640 movies were collected.

33MotionCorr 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. 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 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.

34Crystal and NMR structures of subunits from E. coli (αβγε - 3oaa, δ - 2a7u, b - 1b9u, 1l2p and 2khk) and related organisms (c - 3u2f and a - 5fik) were rigid body docked into the highest resolution cryo-EM map and the side chains ‘pruned’ to Cα. The sequence was mapped to subunit a using crosslinks as restraints. Subsequent manual model building and refinement was performed with Coot, Phenix and Refmac (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.