Structure of the ATP synthase catalytic complex (F(1)) from Escherichia coli in an autoinhibited conformation.

ATP synthase is a membrane-bound rotary motor enzyme that is critical for cellular energy metabolism in all kingdoms of life. Despite conservation of its basic structure and function, autoinhibition by one of its rotary stalk subunits occurs in bacteria and chloroplasts but not in mitochondria. The crystal structure of the ATP synthase catalytic complex (F(1)) from Escherichia coli described here reveals the structural basis for this inhibition. The C-terminal domain of subunit ɛ adopts a heretofore unknown, highly extended conformation that inserts deeply into the central cavity of the enzyme and engages both rotor and stator subunits in extensive contacts that are incompatible with functional rotation. As a result, the three catalytic subunits are stabilized in a set of conformations and rotational positions distinct from previous F(1) structures.

INTRODUCTION

Adenosine triphosphate (ATP) is a key energy carrier in cellular metabolism. Most ATP is synthesized during oxidative- or photo-phosphorylation by the proton-translocating ATP synthase (FOF1-ATPase). This energy-transducing enzyme functions as a rotary motor and is conserved from bacteria to mitochondria and chloroplasts. A proton-motive force (PMF) is generated across a membrane during respiration or photosynthesis, and this PMF drives the transport of protons (Na+ in some bacteria) through the membrane-embedded FO complex of the ATP synthase. FO is connected by a peripheral stator and a central rotor to the extrinsic F1 complex, which contains the catalytic sites for ATP synthesis. Proton transport at the rotor stator interface in FO drives turbine-like rotation of the rotor’s -ring, which directly couples to the rotor subunits of F1. Subunit γ forms the main, asymmetric shaft of F1’s rotor, and transport-driven rotation of γ relative to the surrounding α3β3 complex of F1 drives alternating conformational changes in the three catalytic β subunits to drive net synthesis of ATP[1-3]. Rotational coupling is reversible and, if PMF drops below the energetic threshold needed to drive ATP synthesis, net ATP hydrolysis by the alternating catalytic sites on F1 can drive reverse rotation of γ and the –ring of FO, thus pumping protons across the membrane in the opposite direction. In vitro, F1 can be dissociated from membranes as a soluble ATPase, and crystal structures of mitochondrial F1 (MF1) have provided invaluable insights on the enzyme’s architecture and rotary mechanism[4-7]. However, few structural details are available for bacterial F1-ATPases[8-10], which have been exploited extensively for mechanistic studies[2,3].

The other subunit of F1’s rotor shaft is ε (δ in MF1). In all types of ATP synthase, ε’s N-terminal domain (NTD) binds to γ and directly couples to the –ring of FO. In bacteria and in chloroplasts, ε’s C-terminal domain (CTD) is thought to function as a mobile regulatory element that can change conformation in response to nucleotide conditions and/or PMF[2,11,12]. Growing evidence indicates that inhibition by εCTD involves direct contacts with catalytic β subunit(s). For instance, residues εS108 and βE381 of the Escherichia coli enzyme can be readily cross-linked in vitro and this interaction is modulated by nucleotide conditions. Meanwhile, there is no evidence for a regulatory role by the homolog of ε in mitochondrial ATP synthases, and residues analogous to E. coliβE381 and εS108 are more than 50 A apart in the structure of MF1[13]. Furthermore, in MF1 a unique mitochondrial subunit (known as εM) stabilizes ε’s homolog in a compact conformation that makes no direct contacts with α3β3. Finally, a distinct inhibitor protein has evolved for regulation of the mitochondrial enzyme[14]. Thus, there appear to be significant differences both in the composition of the rotor shaft and in regulation of catalytic activity of bacterial ATP synthases as compared to their mitochondrial homolog.

To provide an atomic description of a prototypical bacterial F1, we have determined the first high-resolution crystal structure of the ATP synthase catalytic complex (F1) from Escherichia coli in an auto-inhibited conformation. The structure provides a clear view of ε’s inhibitory conformation within the F1 complex and thereby sheds light on a regulatory feature that is unique to ATP synthases of bacteria and chloroplasts. Furthermore, bacterial ATP synthase, and not its mitochondrial counterpart, is the proven target for a recently discovered type of anti-tuberculosis drug[15]. Thus the structure for ε-inhibited F1 of E. coli ATP synthase will be particularly valuable for developing new antimicrobials that target bacterial but not mitochondrial ATP synthases.

RESULTS

Structure Determination and Overall Architecture of EF1

Crystallization studies of E. coli F1 (EF1) began nearly two decades ago, but progress was hindered by the limited homogeneity of purified samples of this multi-subunit enzyme that contains nine polypeptide chains (composition: α3β3γδε). Thus far, only a low-resolution, main chain model of EF1 depleted of the peripheral stator subunit δ has been reported[8]. Aiming for a high-resolution structure, we used high-throughput crystallization screening of EF1 depleted of subunit δ (henceforth called EF1) and identified a distinct crystal form that contains four EF1 complexes in the asymmetric unit (M.W. ~1.5 MDa). The diffraction quality of EF1 crystals was gradually improved by controlled dehydration in the presence of nucleotide. Complete diffraction data to 3.26 Å resolution were measured at the National Synchrotron Light Source (NSLS), beamline X25. A complete atomic model was built in a 4-fold averaged electron density map and refined to Rwork/Rfree ~24.3/26.4% at 3.26 Å resolution (Table 1, Supplementary Fig. 1). Sequence registers for all eight chains (α3β3γε) were confirmed using the anomalous signal of 89 selenium peaks. The general architecture of EF1 is analogous to that of MF1 and is illustrated in Figure 1a,b. A hexamer of alternating α- and β-subunits surround the upper region of the central rotor stalk, which consists of an antiparallel coiled-coil of the N- and C-terminal α-helices of γ (γNTH–γCTH, Fig. 1c). Nucleotide binding sites on β subunits are responsible for ATP synthesis and hydrolysis, while sites on α subunits are noncatalytic. In the α3β3 hexamer, the catalytic site of each β is at an interface with a specific α. Based on conserved β-γ interactions, the numbered β subunits correspond to MF1 nomenclature[4] with β1 = βDP,β2 = βE, β3 = βTP. Each of the three α subunits has a noncatalytic site with clear density for bound Mg·ATP or Mg·AMPPNP, although α3 has lower occupancy (Supplementary Fig. 2). Only one specific β subunit, β1, has nucleotide bound at its catalytic site (Fig. 1, Supplementary Fig. 2).

Table 1

Data collection and refinement statistics

	EF1-δ	SeMet-EF1-δ
Data collection
Space group	C2	C2
Cell dimensions
a, b, c (Å)	435.97, 183.00, 225.39	432.39, 181.71, 224.34
α, β, γ (°)	90.00, 108.99, 90.00	90.00, 109.44, 90.00
Resolution (Å)	30–3.26 (3.34–3.26)*	50–5.00 (5.18–5.00)*
Rsym	9.2(62.7)	17.4(58.5)
I/σI	15.4(1.5)	8.9(2.0)
Completeness (%)	98.5(91.7)	97.8(90.2)
Redundancy	2.5(2.1)	3.4(2.5)
Refinement
Resolution (Å)	15–3.26
No. reflections	252,275
Rwork/Rfree	24.31/26.48
No. atoms
Protein	99,621
Ligand/ion	16/32
Water	66
B-factors (Å2)**
Protein	99
Ligand/ion	91/93
Water	53
R.m.s. deviations
Bond lengths (Å)	0.004
Bond angles (°)	0.821

Values in parentheses are for highest-resolution shell.

Average B-factor values refer to EF1-1.

Figure 1

Overview of EF1 structure

(a) Side view of EF1 as a ribbon diagram, with α subunits omitted to reveal the portions of γ (yellow) and ε (magenta) within the central cavity. β subunits are colored in different shades of blue. Space-filling atoms are shown for ADP and SO4 bound on β1 and for the Cα of εSer108 and β1-Glu381 (7.3 Å apart). (b) View from above EF1 (53 Å cross-section, see bracket), including α subunits (green). For clarity, the only regions of γ and ε shown are γNTH, γCTH and ε109–138. Space-filling atoms are shown for all bound nucleotides (on α1, α2, α3, β1) and for residues of α and β subunits that contact ε109–138. (c) Rotated, magnified side view of γ and ε of EF1.

Below the α3β3 ‘head’, γ’s coiled-coil protrudes ~45 Å and is flanked on one side by γ’s globular Rossmann-fold domain, and on the other by the εNTD. In all ATP synthases, it is the base of γ and the εNTD that connect to the rotary -ring of FO (Fig. 1a), and the εNTD is essential for functional coupling of F1 to FO2. Both γ and ε are exceptionally well resolved in the EF1 structure, and the εCTD is the most unique feature (Figs. 1, 2). It adopts a highly extended state (denoted εX) that contacts five of the seven other subunits, including both domains of γ and the CTDs of α1, α2, β1 and β3, and the last half of the εCTD inserts deeply into the central rotor cavity (Fig. 1a,b). This contrasts with ε’s homolog in MF1 structures, in which the εCTD is far from α3β3 and folded compactly against the εNTD[13]. A similar compact state (denoted εC) is observed for isolated bacterial ε[16-18], and bacterial FOF1 retains coupled functions with ε trapped in the εC state[19]. Superimposing εNTD of εC and εX states reveals striking differences in the fold of the εCTD (Fig. 2). The final β-strand of the εNTD in εC (Fig. 2c, β-strand10) is unfolded in εX, forming a loop that begins the εCTD in EF1 (Fig. 2b, loop1). Following εloop1, εhelix1 starts and ends earlier in εX, so that εloop2 is longer (ε103–111; Fig. 2c vs. 2b) than in the εC state. In contrast, εhelix2 is shorter in εX (ε112–125) and the terminal segment, which we name the εhook (ε126–138), bends sharply away (73° crossing angle, εhelix2 vs. helix of εhook). The regions of εCTD that contact β1 and β3 in EF1 agree with previous chemical labeling and cross-linking studies[17]. Most telling, direct ε–β cross-linking[20] showed close contact for εSer108 and βGlu381 of β1 (Fig. 1b), and these sidechains are within hydrogen bonding distance in the EF1 structure (Fig. 3a). Proximity of εSer108 to βGlu381 (on any β) cannot be explained by the εC state (Fig. 2a) or by a distinct extended state of E. coliε seen in a complex of ε with only a truncated γ[21] (γ′–ε, Supplementary Fig. 3). Thus, only the εX conformation reported here in EF1 is consistent with biochemical data for ε–β inhibitory interactions.

Figure 2

Comparing ε’s compact and extended conformations

(a) Side view of EF1, omitting α subunits. Superimposed to ε’s extended state in EF1 (εX, magenta) is the compact state of ε(εC) observed for isolated E. coli ε[16] (only εCTD is shown in gray). Between the two conformations, εNTD aligns well (RMSD ~0.67 Å, ε2–81), while there is a difference of ~73 Å in the position of ε’s C-terminus (dashed line). In (b) and (c), green indicates segments of ε that differ in secondary structure between εX in EF1 (b) and isolated ε (c). In (b) shaded areas with circled numbers identify three regions of the εCTD that contact other EF1 subunits.

Figure 3

Interactions of εCTD within the central cavity of EF1

Selected contact residues are shown, colored by subunit. Dashed lines represent hydrogen bonds (black) or close electrostatic contacts (red); solid black lines show van der Waals contacts. (a) Region-2 contacts of εCTD with γ, β1, α1, and α2. Space-filling atoms are shown for residues in the coiled-coil interface of εhelix2–γNTH. (b) Region-3 contacts of εCTD withβ3 and γ; βTyr331 is shown as part of the adenine-binding pocket, but no nucleotide is bound to β3 in EF1.

Interactions of εCTD with Other EF1 Subunits

The εCTD has three regions of contact with other EF1 subunits (Fig. 2b). In region 1, ε’s loop1 and helix1 contact only γ (Figs. 1a, 2a); εloop1 forms a salt bridge with the γCTH (εArg85–γGlu224) and εhelix1 packs mainly against γ’s Rossmann-fold. In region 2 (Fig. 3a), εloop2 and εhelix2 contact five other chains (α1, α2,β1, β3, γ), with εhelix2 inserted into the central rotor cavity. εhelix2 and γNTH form an antiparallel coiled-coil, with substantial burial of hydrophobic residues, that is also stabilized by hydrogen bonds at both ends and by electrostatic contacts (γArg84, γLys30 with εAsp111). The position of εhelix2 between β1’s CTD and γ blocks specific β1–γ interactions that are seen in MF1 structures and thought to be important for rotational coupling[5,7,13]. The CTDs of α1 and α2 also contact εhelix2 from either side (Fig. 1b), apparently helping clamp εhelix2 in position (Fig. 3a). Contact region 3 spans the εhook (ε126–138), which contacts γNTH, γCTH and wraps partly around helix1 of β3’s CTD (Fig. 3b). The εhook also contacts β3 near βTyr331, a part of the adenine-binding pocket. In comparison, region 2 involves more specific bonds of εloop1 and helix2 with other subunits, but much more contact surface is buried in region 3 between β3 and εhook than between β1 and ε in region 2. Overall, contacts of εCTD with other subunits bury ~2900 Å2 of surface area and ~70% of this involves the segment of εCTD inserted within the central rotor cavity (ε109–138). Of the surface buried by ε109–138, ~56% is with α and β subunits. These extensive rotor–stator interactions are expected to prohibit rotation of γε relative to α3β3 when ε adopts the εX state. This agrees with a recent study of forced rotation of thermophilic bacterial F1 in which activation from an ε-inhibited state required much greater rotary torque than activation from an ADP-inhibited state[22]. Thus, the εX state observed crystallographically in EF1 correlates with the ε-inhibited state that blocks both hydrolysis and synthesis of ATP by EFOF1[23].

Distinct Features of Catalytic β subunits

EF1 is the first F1 structure determined in which only 1 of 3 catalytic sites has bound nucleotide (Fig. 1a,b, Supplementary Fig. 2), and the catalytic β subunits show a combination of conformational states not seen before (Fig. 4). Most MF1 structures are similar to the 1.9 Å ground-state of MF1[5]: β1 and β3 are each in a ‘closed’ state with bound nucleotide, but β2 has an open state without nucleotide, since contacts of its CTD with a convex surface of γ distort the nucleotide binding site[4]. In EF1, β2 adopts the usual open state and makes no contacts with the εCTD. β3 adopts the basic closed state but has no bound nucleotide, although its interface with the εhook causes minimal distortions relative to β3 of MF1. However, β2 and β3 each have SO42− bound at the P-loop (Supplementary Fig. 2), in a position nearly identical to that occupied by PO42− on β3 of nucleotide-free yeast MF1[24]. Finally, β1 cannot assume the usual closed state, due to insertion of εhelix2 between β1’s CTD and γ. β1 is also not in the open state, but adopts a half-closed state with bound ADP and SO42−. A unique MF1 structure, with nucleotide bound on all three β (denoted here as MF1-3filled), has one β in the same half-closed state (Fig. 4c) with bound ADP and SO42−, but at the β2 position that is typically in the open state[7]. The positions of bound Mg·ADP, SO42− and key ligand-binding residues also align closely between β2 of MF1-3filled and β1 of EF1 (Fig. 4d). Thus, the εCTD does not distort β1 into a unique state, but traps it in an intermediate conformation that was seen before, but in a different rotary position of β relative to γ.

Figure 4

Distinct conformations of three β subunits in EF1

Panels (a) and (b) show that EF1 subunit β1 (dark blue) has a conformation that is distinct from both the ‘open’ conformation of β2 (a, light blue) and the ‘closed’ conformation of β3 (b, cyan). Surveying known F1 structures, we found that β1 of EF1 superimposes best with the ‘half-closed’ conformation of β2 in the bovine MF1 structure that has nucleotide bound at all 3 catalytic sites (MF1-3filled)[7] (panel c, red). For the superimposition of panel (c), panel (d) focuses on details at the catalytic nucleotide-binding site. For β1 of EF1, (b) shows atom-colored sticks for bound ligands ADP and SO42− (Mg2+ = green sphere) and for five key residues (numbered); the corresponding residues, SO42− and Mg2+ are shown in red for β2 of MF1-3filled (bound ADP is oriented similar to ADP on β1 of EF1 and is omitted for clarity).

Correlations with Functional Rotary Mechanics

The surprising finding of a half-closed conformation of β1 in EF1 prompted us to compare the rotary arrangement of the three catalytic sites around γ in EF1 to that of MF1 structures. Single-molecule studies have shown that each 120° rotation (associated with net hydrolysis of one ATP) involves two sequential kinetic substeps: ~80° rotation follows ATP binding at one catalytic site, and ~40° rotation follows the catalytic pause limited by hydrolysis and release of product(s) at an alternate site[25,26]. Thus far, most MF1 structures were thought to exhibit one orientation of α3β3 around γ, but significantly different rotary positions were noted for MF1(3-filled)[7] and for one conformation of yeast MF1 (yF1I)[27]. It was suggested that MF1(3-filled) represents the catalytic dwell position (before the 40° step) and that yF1I represents the ATP binding position (before the 80° step)[28]. As a new approach to align F1 structures and compare their relative rotary positions, we identified a structural core of γ that has minimal deviations between different F1 structures (Supplementary Figs. 4, 5; Supplementary Methods). A stiff γ-core structure is considered necessary to drive alternating conformational changes in the β subunits during rotation; the γ-core identified here overlaps with stiff regions indicated by single-molecule studies with EF1[29] and includes most γ residues noted for torque generation in molecular dynamics studies with MF1[28]. The γ-core includes significant portions of γ’s coiled-coil and Rossmann-fold domains (Supplementary Fig. 5), and so provides a robust reference for superimposing F1 structures and comparing rotary positions of the three catalytic sites around γ (Supplementary Fig. 6). Figure 5a illustrates that the β subunits of MF1(3-filled) are rotated farthest in the direction of net ATP synthesis, as noted before, whereas those of EF1 are rotated farthest in the direction of ATP hydrolysis. The distinct rotary position of ε-inhibited EF1 is supported by electron microscopy studies of EF1[30] and by single-molecule fluorescence studies of EFOF1-liposomes[31]. The 43° rotary shift from MF1(3-filled) to EF1 correlates with the 40° step following the catalytic dwell, and rotating EF1 farther by 78° would superimpose its half-closed β1 with the half-closed β2 of MF1(3-filled). Thus, the εX state appears to trap EF1 in a rotary position close to the kinetic dwell before the next ATP binding event and 80° rotary step. Bound product(s) on the half-closed β before (EF1, β1) and after the 80° step (MF1(3-filled), β2) support the linkage of product dissociation to the 40° step and the original contention that MF1(3-filled) represents the rotary state post-hydrolysis but prior to product release from its half-closed β2[7].

Figure 5

Insights for rotary mechanics of ATP synthase

(a) View from below F1 with a ribbon diagram of the γNTH–γCTH coiled-coil (EF1) and mass-weighted ellipsoids for β subunits of three F1 structures superimposed by γ’s structural core (Supplementary Fig. 5). Specific βs are labelled for EF1 (shades of blue), and corresponding βs are shown for yF1II[27] (gray) and bovine MF1(3-filled) (red). Arrows indicate the rotation needed (in hydrolysis direction) to superimpose β1 of EF1 with β2 of MF1(3-filled) (78°) or to superimpose β2 of MF1(3-filled) with β2 of EF1 (43°). (b) and (c) side views of E. coli γ with the ellipsoids of either β2 (b) or β3 (c) for the three γ-aligned F1 structures. For each β ellipsoid in (b) and (c), two residues are shown: βAsp305 (in ‘catch 1’, β2–γ) and βIle376 (in ‘catch 2’, β3–γ). The red line (a,c) is the axis for the 78° rotation noted in (a).

Aligning F1 structures by their shared γ-core provides additional visual clues to the complex movement of β subunits relative to γ during functional rotation, which was suggested by normal-mode analysis[32]. This is illustrated in side views showing just the distinct positions of β2 (Fig. 5b) and β3 (Fig. 5c) for γ-aligned structures that span the 40° rotary step. The 80° rotary step has a central axis parallel to γ’s vertical shaft (red line, Fig. 5b,c), but the different positions of each β along the 40° step suggest a more complex pivoting around γ’s asymmetric features. Also, specific β-γ contacts or “catches”[4] may restrict the distinct pivoting of β2 vsβ3 across the 40° step, as hydrogen bonds of catch-1 (β2-γ, Fig. 5b) and catch-2 (β3-γ, Fig. 5c) are maintained through the range of rotary states shown. Thus, the three states of β2 remain close in the upper region near catch-1 but are farther apart at the base, whereas the β3 states are close at the base near catch-2 but farther apart in the upper region; positions of β1 around γ in different structures indicate β1 does not pivot much during the 40° step (not shown). The different pivoting of β2 vsβ3 should also correlate with opening or closing of the different α–β catalytic interfaces during rotation, which is thought to be important in modulating the functional states of the alternating sites[4]. Finally, γ-core alignment of F1 structures suggests that, during the 40° rotary step, the final segment of the γCTH (~20 residues) is bent in different directions (Supplementary Fig. 5b) by the “hydrophobic sleeve” region of α3β3 that surrounds it[4]. Flexibility of this final segment of γCTH is consistent with results of single-molecule studies and molecular dynamics simulations[33]. The direction of the γCTH bend correlates with F1’s rotary position being <20° (Supplementary Fig. 5b, circled) or >20° in the direction of ATP hydrolysis. This correlation holds true for all MF1 structures aligned by γ-core (not shown) with one exception (see Supplementary Fig. 6). Accordingly, the correlation between the γCTH bend and F1’s rotary position could suggest a rotary transition point at which torque between γ and α3β3 is sufficient to induce a distinct bend in the final segment of γCTH.

DISCUSSIONS

Physiological regulation of ATP synthases

The structure described in this paper reveals the first molecular view of the ε-inhibited state that can occur in ATP synthases in most bacteria and in chloroplasts: the εCTD adopts a highly extended conformation (εX) that partly inserts into the central rotor cavity, bridging between γ’s rotary stalk and surrounding catalytic subunits to prevent functional subunit rotation (Fig. 6a,c). This structural snapshot of the εCTD within E. coli F1 agrees with a wealth of cross-linking and functional data on ε’s inhibitory interactions with bacterial ATP synthases, so far not explained by structures of eukaryotic F1. In contrast, the mitochondrial homolog of ε is believed to be non-inhibitory, with its CTD clamped in the compact εC state by a unique mitochondrial subunit (Fig. 6d,f)[13]. Instead, eukaryotes evolved a separate protein, IF1, to inhibit mitochondrial ATP synthase[14]. Nevertheless, these distinct inhibitor proteins serve the same primary role, to block ‘wasteful’ ATP hydrolysis by FOF1 under conditions when the PMF across the membrane is low or absent. In mitochondria, respiration and PMF decline dramatically during cellular hypoxia, which occurs for instance during cardiac failure. Without PMF to drive ATP synthesis, FOF1 begins to work in reverse, but acidification of the mitochondrial matrix transforms IF1 into an active form that binds to and inhibits MF1, minimizing wasteful ATP hydrolysis and the odds of cell death. In plants, chloroplasts regularly lose PMF during long dark cycles, and inhibition by ε coordinates with a chloroplast-specific adaptation of γ to inactivate the ATP synthase in the dark[12]. Bacteria are more varied in their environmental and metabolic demands, and the physiological role of ε inhibition may be tuned to these differences in bacterial ecology; this is consistent with the variations in sequence and length of εhelix2 between different types of bacteria[11] and with the fact that some aerobic bacteria exhibit much stronger ε-inhibition on membranes than observed with E. coli[34]. Some bacteria can neither respire nor photosynthesize, but require their FOF1 to function as an ATPase-driven proton pump in order to maintain PMF and/or internal pH homeostasis[35]. Moreover, while facultative anaerobes such as E. coli can respire, they also need FOF1 to function as an ATPase-driven proton pump in anaerobic conditions, which can occur along the digestive tract of their hosts. Thus, similar to previous arguments[36], bacterial ε is not generally geared to inhibit FOF1 whenever thermodynamics favor ATP hydrolysis, but rather inhibits ATPase-driven proton pumping when it is wasted by failing to generate substantial PMF across the cell membrane. In E. coli, for instance, this can occur when high concentrations of membrane-permeant acids arise from fermentation or from their host’s digestive processes, and it is known that FOF1 is important for one acid-resistance mechanism of E. coli[37]. Nevertheless, further studies will be needed to determine specific environmental conditions where autoinhibition by ε subunit confers a selective advantage for growth or survival of different bacteria.

Figure 6

Comparing interactions with F1 for E. coli εCTD and the mitochondrial inhibitor IF1

Side and top views are shown for EF1 (a,b) and MF1+IF1(1–60)[6] (d,e). The same viewpoints of EF1 and MF1+ IF1(1–60) are aligned by the γ-core. Side views show only β1, β3, γ and ε (or its magenta-colored homolog, δM), plus unique mitochondrial chains εM (gray) and IF1 (red) (d,e). Solvent-excluded surfaces are shown for β1 and β3 (ADP shown if present), and ε106–138 (a) or IF1 (d). In (c) and (f) (view rotated 120° from (a/d), α and β subunits are omitted and a transparent, solvent-excluded surface is added for γ in EF1 (c) and in MF1+IF1(1–60) (f).

Comparisons of ATP synthase inhibitor proteins

Despite the distinct origins of the bacterial inhibitor εCTD and the mitochondrial inhibitor IF1, there are broad similarities in the way these two endogenous inhibitors interact with the F1 catalytic core (Fig. 6). In each case, the inhibitory segment is mainly α-helical, is inserted at the α1–β1 catalytic interface, and contacts the same five subunits (α1, α2, β1, β3, γ). Each inhibitory region buries extensive surface area (ε106–138, ~2100 Å2; IF1, ~2700 Å2) and has specific interactions buried deeply within F1. It was proposed that IF1 inserts at a prior rotary position, with a more open α–β interface, and then is buried by subsequent rotation and conformational changes in MF1[6]. With EFOF1, ε’s conformational change is blocked by preventing rotation of the –ring in FO[38], suggesting at least a rotary substep is linked to ε’s transition to or from the inhibited state. Rotational entrapment may also correlate with the bent shape of each inhibitor’s most buried end, although εCTD and IF1 are bent in different directions (Fig. 6a ); in the isolated γ′–ε complex[21], εhelix2 is long and unbent as in εC (Supplementary Fig. 3), suggesting the bent εhook in EF1 is induced by interactions with α3β3. Reversing inhibition of FOF1 also has a common factor. PMF stimulates dissociation of IF1 to activate MFOF1[39], and it was proposed that PMF-driven rotation in the direction of ATP synthesis causes IF1 to be expelled[6]. PMF activates the latent chloroplast enzyme and causes the εCTD to become exposed[12]. Activation by PMF also occurs with ATP synthases of bacteria, including E. coli[40], and preliminary evidence links this to relief of inhibition by εCTD[41]. Thus, whereas most bacteria retain in cis auto-inhibition by the εCTD, eukaryotes have evolved IF1 for a similar mode of inhibition, although it acts in trans.

Model for the regulatory transition between the εC and εX conformations

We propose a basic series of molecular events for the transition between the εC and εX states in bacterial ATP synthase, as summarized in Figure 7. An early step from εC to εX should be to disrupt the interface of εhelix2 with the εNTD, and several mutations in εNTD that alter inhibition are near εhelix2 in the εC state[42]. For some bacteria, the εC state can be stabilized by binding of ATP to a low-affinity site that bridges the εhelix2–εNTD interface, thus favoring active complexes when cellular ATP is abundant[18]. We speculate that the stability of the εhelix2–εNTD interface could also be influenced by the transmission of rotary torque between γ–εNTD and the –ring of FO; the apparent torsional compliance of the bottom regions of γ–εNTD[29], which interface with the —ring, could distort the εhelix2–εNTD interface when PMF induces torque through the –ring. Another possibility is that εhelix2–εNTD and/or εhelix1–helix2 interactions in εC may be directly influenced by the membrane potential component of PMF; εhelix1 and helix2 lie close and nearly parallel to the plane of the membrane in the εC state and contain many charged residues. Separation of εhelix1 from εhelix2 should also occur early, since the face of εhelix1 that contacts εhelix2 in εC instead interacts with γ in the εX state. We propose that a kinetically appreciable intermediate state of ε then forms by docking of εhelix1 to γ, as in εX, but with εhelix2 exposed and mobile below α3β3 (Fig. 7, center). This is supported by proteolysis studies in which cleavage of ε initiates in εhelix2[17,38]. Trypsin cleavage is slow for isolated ε (εC state) and for EFOF1 in the presence of MgADP and Pi, which favors the εX state; this is consistent with limited exposure of εhelix2 in the εC state (packed between εhelix1 and the εNTD) and in the εX state (buried within EF1’s central cavity). Trypsinolysis of ε in EFOF1 is much faster in the presence of MgAMPPNP, which may favor εC but also favors ε–α cross-linking instead of ε–β cross-linking[43]. Thus an intermediate state as shown, with εhelix1 bound to γ, would keep εS108 of εloop2 near α3β3 but would leave εhelix2 exposed for cleavage by trypsin. This intermediate should enhance the kinetics for insertion of εhelix2 into EF1’s rotor cavity in the next step, when transition of β1 towards the half-closed state and partial rotation of γ create an opening sufficient for εhelix2 to insert and form a coiled-coil with γNTH. Further subunit rotation would then occur, burying εhelix2 within the central cavity and disrupting the end of εhelix2 to form the εhook. Once stabilized with ε in the εX state, expelling εhelix2 to return FOF1 to an active state would probably require rotary torque from FO in the direction of ATP synthesis; consistent with this, it was noted earlier that PMF activates bacterial FOF1. Further experiments will be needed to elucidate details of this intriguing regulatory mechanism for ATP synthases of bacteria and chloroplasts.

Figure 7

Model for transition between εC and εX states in EF1

Models of EF1 are docked with the 10–ring of FO from a yeast MF1–10 structure[47] (EF1 and MF1 superimposed by γ-core). The membrane is depicted by the gray region. Not shown is the 2δ stator assembly, which has not yet been resolved. In each EF1 model, regions of ε are colored magenta (εNTD), gray (ε82–105) or red (ε106–138), and subunit α1 is omitted to view the central cavity. The determined structure of EF1 (right) has the rotor assembly (γε10) rotated ~40° (in hydrolysis direction) relative to the other two models. A specific subunit (orange) provides a visual reference for the rotation.

Concluding remarks

Functional ATP synthase is essential for higher organisms, but is also critical for the viability of pathogenic bacteria such as Streptococcus pneumoniae[44] and Mycobacterium tuberculosis[45]. Even enterohemorrhagic E. coli cannot compete for an intestinal niche if it lacks the ATP synthase[46]. Differences in structural complexity and regulation between bacterial and mitochondrial ATP synthases can be exploited to selectively inhibit the former. The structure of ε-inhibited EF1 presented in this paper provides a rational framework for developing antimicrobial agents that selectively mimic or stabilize the ε-inhibited state but do not inhibit mitochondrial ATP synthase.

METHODS

Protein expression and purification

EFOF1 was expressed from plasmid pJW1[48] in E. coli strain JP17[49]. Cell were grown and membranes isolated as described[48]. To incorporate selenomethionine (SeMet) into EFOF1, pJW1 was expressed in metB− strain LE392Δ(atpI–C)[50]; cells from rich-medium starter cultures were collected by centrifugation, washed with defined medium without methionine[51], then grown in 10 L of the same medium with 0.1 g of SeMet per liter. EF1 was purified as before[52], but the ion exchange step used 50 ml of Macro-Prep High Q resin (Bio-Rad) and a linear gradient from 0 to 350 mM NaCl (250 ml at 2 ml min−1). EF1 was depleted of subunit δ by gel filtration[8] (Sephacryl S-300, HiPrep 16/60, GE Life Sciences) at 22°C in the presence of 0.2% (w/v) lauryldimethylamine oxide; δ-depleted EF1 was dialyzed extensively (10 kDa MWCO) against column buffer without detergent, concentrated to >15 mg ml−1, frozen in liquid N2 and stored at −80°C.

Crystallization and X-ray data collection

Before crystallization, δ-depleted EF1 (or SeMet-substituted EF1, SeMet-EF1) was dialyzed at ~5 mg ml−1 vs. TE75 buffer (50 mM Tris-HCl, 0.1 mM Na2EDTA, pH 7.5) for 12–18 hr at room temperature (RT), including one buffer change, then concentrated by ultrafiltration to ≥20 mg ml−1. At this point, EF1 retained endogenous adenine nucleotides (mol per mol EF1: total, 2.74 ±0.16; noncatalytic, 1.03 ±0.23 ADP, 0.68 ±0.1 ATP; catalytic, 0.97 ±0.09 ADP, <0.12 ATP). EF1 and SeMet-EF1 were crystallized at RT by hanging-drop vapor diffusion. EF1 at 20 mg/ml (typically 3 μl) was mixed with an equal volume of 0.1 M MOPS-NaOH, pH 7, 75–150 mM MgSO4, 7–9% (w/v) PEG8000, 5 mM β-mercaptoethanol and equilibrated against 600 μl of the same solution. Crystals were screened for diffraction at NSLS beamlines X6A and X25 as well as macCHESS stations A1 and F1. Diffraction quality was improved by controlled dehydration in solutions containing 25% (v/v) glycerol, in the presence of 1 mM AMPPNP. A complete dataset to a resolution limit of ~3.26 Å was obtained at NSLS beamline X25 (Table 1). A 5.0 A dataset for a SeMet-EF1 crystal was collected at NSLS beamline X6A at the selenium edge (~0.972 Å) (Table 1). All data were processed and scaled in HKL-2000[53]. EF1 crystals belong to space group C2 with four EF1 complexes in the asymmetric unit.

Structure determination and refinement

The structure of EF1 was solved by molecular replacement with PHASER[54], using α3β3 of an MF1 structure (PDB entry 2CK3[55]) as search model. Initial phases were dramatically improved by iterative cycles of solvent flattening, histogram matching, and 4-fold non-crystallographic symmetry averaging with DM[56]. A 4-fold averaged map revealed striking electron density features for the γ and ε subunits, which were not present in the initial phasing model. The averaged 3.26 Å map allowed straightforward interpretation of most side chains. Anomalous scattering peaks from the SeMet-EF1 dataset helped confirm the register of each chain in EF1. An atomic model containing all 8 chains in EF1 (α3β3γε) was manually built in Coot[57] and refined with PHENIX[58]. Finally, complete atomic models were built for all four EF1 complexes in the asymmetric unit (referred to as EF1-1, EF1-2, EF1-3, EF1-4). Complexes EF1-1 (chains A H) and EF1-2 (chains I–P) have better-defined electron densities than EF1-3 (chains Q–X) and EF1-4 (chains Y, Z, a–f). For all structural analyses and illustrations described in this study, complex EF1-1 was used as reference. The pairs of α and β subunits that form the three distinct catalytic interfaces in EF1 are numbered 1–3 in the main text. For example, with complex EF1-1, these correspond to chains as follows: α1,β1 = C, D; α2, β2 = A, E; α3, β3 = B, F. In the final deposited model, all four EF1 complexes in the asymmetric unit include residues 25–511 of α1 (chains: C; K; S; a); residues 24–511 of α2 (chains: A; I; Q; Y); residues 26–511 of α3 (chains: B; J; R; Z); residues 2–459 of all β subunits (chains: D, E, F; L, M, N; T, U, V; b, c, d); residues 1–284 of each γ (chains: G; O; W; e) and residues 1–138 of each ε (chains: H; P; X; f). In all α chains, weak density is observed for the solvent-exposed loop 310–318 (residues EAFTKGEVK), which was modeled as poly-alanine in α1 (chains: C; K; S; a) and for residues Leu-448 and Ile-464, which were modeled as alanines in all α chains. Further, in α3 (chains: B; J; R; Z), residues 402–414 have poor density and were partially modeled as poly-alanine between residues 404–408 and 410–416. All β chains in the structure contain the spontaneous point mutation K81E, which is solvent-exposed and does not affect EF1 activity[59]. Finally in all γ chains, residues 60–61 have poor density and were modeled as alanines. Ligands bound to EF1 (Supplementary Fig. 2) include: Mg·ANP on every α chain; Mg·ADP and SO42− on the β1 subunit of each complex (chains D, L, T, b); a SO42− ion on each β2 chain (E, M, U, c) and on each β3 chain (F, N, V, d). Although a quantitative analysis of ligand occupancy is impossible at this resolution, the Mg·ANP bound to α3 chains (B, J, R, Z) is significantly less occupied than those bound to α1 and α2 chains. Likewise, the SO42− ion bound to β2 chains (E, M, U, c) has reduced occupancy as compared to that bound toβ3 chains (F, N, V, d). Additional strong peaks of density (4–6σ above background) were noted in the final Fo–Fc difference map (i) coordinating with ε-Ser65 and (ii) in a pocket inside the γ subunit. Finally, 66 water molecules were modeled in 3.5σ peaks of Fo–Fc density, mainly in proximity to EF1-1 subunit γ (chain G). The final model was refined to Rwork/Rfree ~24.3/26.4% at 3.26 Å resolution (Table 1). Structural figures were prepared with Chimera[60].