F₁-ATPase of Escherichia coli

Background: Bacterial ATP synthases are autoinhibited by the subunit ϵ C-terminal domain.

Results: Nucleotide hydrolysis is required to form the ϵ-inhibited state, which also responds dynamically to different ligand conditions.

Conclusion: ϵ inhibition initiates at the catalytic dwell angle, but reversible rotation over ∼40° is probably involved in nucleotide effects on the inhibitory state of ϵ.

Significance: ϵ inhibition may provide a new target for antimicrobial discovery.

Abstract

F₁-ATPase is the catalytic complex of rotary nanomotor ATP synthases. Bacterial ATP synthases can be autoinhibited by the C-terminal domain of subunit ϵ, which partially inserts into the enzyme's central rotor cavity to block functional subunit rotation. Using a kinetic, optical assay of F₁·ϵ binding and dissociation, we show that formation of the extended, inhibitory conformation of ϵ (ϵ_X) initiates after ATP hydrolysis at the catalytic dwell step. Prehydrolysis conditions prevent formation of the ϵ_X state, and post-hydrolysis conditions stabilize it. We also show that ϵ inhibition and ADP inhibition are distinct, competing processes that can follow the catalytic dwell. We show that the N-terminal domain of ϵ is responsible for initial binding to F₁ and provides most of the binding energy. Without the C-terminal domain, partial inhibition by the ϵ N-terminal domain is due to enhanced ADP inhibition. The rapid effects of catalytic site ligands on conformational changes of F₁-bound ϵ suggest dynamic conformational and rotational mobility in F₁ that is paused near the catalytic dwell position.

Introduction

ATP synthases play a key role in energy metabolism in most living organisms and achieve energy coupling as dual engine rotary nanomotors (1–3). The F-type ATP synthase of Escherichia coli (Fig. 1), a bacterial prototype, is composed of core subunits that all have homologs in the ATP synthases of mitochondria and chloroplasts (4). The membrane-embedded F_O complex (ab₂c₁₀) acts like a turbine to transport protons across the membrane, and the external F₁ complex (α₃β₃γδϵ) contains three cooperative catalytic sites for ATP synthesis or hydrolysis. The ring of c-subunits, with the critical proton transport sites, is the rotor complex of F_O and connects to the central rotor stalk of F₁, composed of γ and the N-terminal domain (NTD)² of ϵ. The three catalytic β subunits alternate with three α subunits to surround the upper half of the asymmetric rotor stalk of γ, and the δ-b₂ connection forms a peripheral stator stalk anchoring α₃β₃ to the other stator subunit of F_O, a. In vitro, F₁ from eukaryotes and bacteria can be dissociated from F_O as a soluble, rotary motor ATPase, and these F₁-ATPases have been useful for both mechanistic studies and the determination of high resolution structures.

FIGURE 1.

Architecture of bacterial ATP synthase and alternate conformations of subunit ϵ. The smaller image (bottom left) depicts the E. coli ATP synthase, with F_O subunits spanning the membrane bilayer (shaded box); the arrow across the bilayer indicates the direction of proton (H⁺) transport during net ATP synthesis. F_O subunits (ribbons, a (dark red), b₂ (gray), and c₁₀ (green)) and F₁ subunit δ (orange ribbon) are from a homology-modeled assembly (67). All other F₁ subunits are from determined structures and are surface-rendered in the F_OF₁ model but displayed as ribbons in the magnified view of E. coli F₁ (3α (green), 3β (shades of blue), γ (yellow), ϵNTD(1–87) (light pink), and ϵCTD(88–138) (dark pink)). The F_OF₁ model shows ϵ in the ϵ_C or compact conformation (Protein Data Bank entry 1BSN), docked to γ of EF₁-δ (Protein Data Bank entry 3OAA). The magnified ribbon diagram shows the ϵ-inhibited F₁-δ structure (Protein Data Bank entry 3OAA) and omits the foremost α subunit to reveal the extended conformation of ϵ (ϵ_X); for comparison, a ribbon model of the ϵ_C state is shown offset to the right. The ribbon diagram of each ϵ conformation shows space-filling side-chain atoms (colored by element) predicted in silico for mutations ϵA101C/L121C. Space-filling atoms are also shown for ADP and SO₄²⁻ on the one occupied catalytic β subunit (chain D). The molecular graphics were prepared with Chimera (81).

Despite general conservation between bacterial and mitochondrial ATP synthases, it has been demonstrated that bacterial ATP synthase can be an effective target for antibacterial treatment. It is the target of a novel class of compounds that are bactericidal for actively replicating and dormant mycobacteria (5, 6) and that show promising effects against multidrug-resistant tuberculosis in phase II clinical trials (7). However, the lead compound is only effective against a narrow spectrum of mycobacteria, and, because it targets the H⁺-transporting sites of F_O, adapting this scaffold to target other pathogenic bacteria introduces a significant risk of cross-reaction with mitochondrial ATP synthase. Recently, our group determined the first crystal structure of a bacterial F₁-ATPase that is in an autoinhibited state mediated by the C-terminal domain (CTD) of its ϵ subunit (8). Inhibition by ϵ may serve regulatory roles in ATP synthases of bacteria (2, 9) and chloroplasts (10) but does not occur in mitochondrial ATP synthase, which has a distinct inhibitor protein (11). Recent studies confirmed that the bacterial ϵCTD inhibits ATP synthesis as well as hydrolysis (12, 13), indicating that ϵ inhibition may provide a new target for future development of antimicrobial drugs selective for bacteria. With that in mind, the current study focuses on improving our biochemical understanding of how the catalytic F₁ complex of E. coli ATP synthase is inhibited by ϵ.

As shown in Fig. 1, the ϵ subunit has two domains. The ϵNTD, essential for the F₁ rotor connection to the c-ring in F_O (2, 9), is a β-sandwich fold and exhibits a similar conformation and association with γ in several structures of bacterial F₁ (8, 14) and mitochondrial F₁ (MF₁) (15, 16); essentially the same ϵNTD structure is also seen for isolated bacterial ϵ (17–19). However, the α-helical ϵCTD has been observed in dramatically different conformations (Fig. 1). A compact conformation (the ϵ_C state) has a coiled-coil between its two α-helices, and the second helix packs against the ϵNTD. The ϵ_C state has been observed for isolated bacterial ϵ (17–19) and in one bacterial F₁ structure (14). In structures of MF₁ (15) and MF₁·c-ring (20), the homolog of ϵ appears to be locked in the ϵ_C state by a mitochondria-specific subunit. E. coli ATP synthase can synthesize and hydrolyze ATP when ϵ is restricted to the ϵ_C state (21), in which the ϵCTD does not contact any F₁ subunits (Fig. 1, left). In contrast, in the recently determined structure of E. coli F₁ (Fig. 1) (8), an extended conformation of the ϵCTD (ϵ_X state) contacts five other subunits, and its terminal half is inserted into the central cavity of F₁. The position and subunit contacts of the ϵCTD within the E. coli F₁ structure correlate well with many biochemical studies of ϵ inhibition and interaction with other F₁ subunits (reviewed in Refs. 2 and 9). The extensive buried surface of the ϵCTD within the F₁ structure and its interactions with two catalytic β subunits suggest that this form of the enzyme represents an inactive state. This correlates with results of “single-molecule” (SM) studies of F₁ from E. coli (22) and other bacteria (23–26), showing that ϵ can induce or extend long “pauses” (seconds) during which γ does not rotate in the presence of substrate MgATP. Some SM studies concluded that ϵ inhibits by stabilizing or extending an ADP-induced inhibitory pause that occurs at the catalytic dwell (22, 24, 27), whereas another recently concluded that ADP- and ϵ-induced inhibitions are separate processes for cyanobacterial F₁ (25). Some studies also concluded that ϵ inhibition includes or is dominated by changes to one or more intrinsic kinetic steps along the catalytic pathway (12, 22, 28, 29). In the current study, we adapt an optical assay to directly measure the kinetics of binding and dissociation for E. coli F₁·ϵ and correlate these with inhibitory effects for wild type (WT) and mutant forms of ϵ. Our biochemical evidence confirms that inhibition by the CTD of ϵ initiates at the catalytic dwell but also shows that ϵ inhibition competes with formation of the ADP-inhibited state. Further, whereas ϵ inhibition initiates at the catalytic dwell, we also show that the balance between active and ϵ-inhibited states responds dynamically to changing nucleotide conditions.

EXPERIMENTAL PROCEDURES

Plasmid Constructs for Affinity-tagged E. coli ϵ Subunit

A plasmid described previously (30) (noted here as pH₆ϵ) encodes a His₆-tagged ϵ (H₆-ϵ), with a tobacco etch virus protease cleavage site following the N-terminal His₆ tag. Site-directed mutagenesis was used to create the following mutations in H₆-ϵ. A pair of Cys mutations, ϵA101C/L121C, was created on pH₆ϵ using the QuikChange Multi site-directed mutagenesis kit (Stratagene) with primers 5′-CATGGAAGCGA-AACGTAAGTGTGAAGAGCACATTAGGAG-3′ (ϵA101C) and 5′-GCTCAGGCGTCTGCGG-AATGCGCCAAAGCGATC-3′(ϵL121C). H₆-ϵ that expresses only the ϵNTD (H₆-ϵ88stop; see Ref. 31) was created with the QuikChange-II XL kit (Stratagene) (forward primer, 5′-CAATTCGCGGCCAGTAAGTCGACGAAGCG-3′). Plasmid pBKH2 was created to add a biotin acceptor peptide (Bap) before the N-terminal His₆ tag on H₆-ϵ. This BapH₆-ϵ has 49 residues before the native initial Met of ϵ, and tobacco etch virus cleavage would yield ϵ with three extra N-terminal residues (GAM). It was created with pH₆ϵ as a template, using PCR to generate a 514-bp amplicon with restriction sites added before (XhoI) and after (BamHI) the gene for H₆-ϵ (primers, 5′-CGACTCGAGCATGTCGTACTA-CCATCACC-3′ and 5′-CTCGGATCCTTACATCGCTTTTTTGGTCAAC-3′). Following cleavage with XhoI and BamHI, this amplicon was cloned into the same sites of pDW363 (32), replacing the malE gene, to create pBKH2. BirA (E. coli biotin holoenzyme synthetase) is co-expressed from pBKH2, allowing in vivo biotinylation of the biotin acceptor peptide on BapH₆-ϵ (32). BapH₆-ϵ expressed with ϵ88stop or ϵA101C/L121C had poor protein yields, so ϵ was also expressed as a fusion protein following an N-terminal maltose-binding protein (MBP), a cleavage site for PreScission protease (GE Healthcare), and the Bap tag. This MBP-Bap-ϵ has a 31-residue segment between MBP and the initial Met of ϵ, and after cleavage by PreScission protease, Bap-ϵ would retain a 25-residue N-terminal Bap tag. The vector for this construct, pMal-PPase, was derived from pMAL-c2e (New England Biolabs), with a PreScission protease cleavage sequence after malE (33), and an NdeI site was removed by cleavage and polymerase fill-in. The sequence encoding the Bap tag was PCR-amplified from pDW363, with flanking restriction sites before (StyI) and after (NdeI and BamHI) the Bap sequence (primers, 5′-CATCCCAAGGCTGGAGGCCTGAACGATATTTTC-3′ and 5′-CTCGGATCCCATATGGCCACCAGTGTCCTCGTG-3′). After cleavage with StyI and BamHI, this amplicon was inserted into StyI-BamHI sites of pMal-PPase to generate pMAL-PP-Bap. The atpC gene for WT ϵ was extracted from p3U (34) as a 625-bp NdeI-XbaI fragment and ligated into the same sites of pMAL-PP-Bap to create pBKH8, encoding a fusion protein of MBP-Bap-ϵ. Plasmids encoding MBP-Bap-ϵ with mutation ϵ88stop (pBKH9) or ϵA101C/L121C (pBKH10) were produced in the same way, but the NdeI-XbaI inserts were 799 bp (extra sequence downstream of atpC) because those mutant atpC genes had been passed through an intermediate vector.

Purification of Proteins

SDS-PAGE (35) (Bio-Rad Ready Gels, 12% or 4–15%) was used to analyze the purity of all protein preparations, with staining by SYPRO Orange (Invitrogen) and scanning on a Typhoon 9410 imager (GE Healthcare; 488-nm laser excitation, 526-nm SP emission filter). With gels to test for internal disulfide bonding in Bap-ϵA101C/L121C, concentrated gel sample buffer contained 0.5 mm N-ethylmaleimide instead of 2-mercaptoethanol (βME). Concentration of protein samples was determined by a modified Lowry assay (36). E. coli F_OF₁ was expressed, and F₁ was purified and depleted of subunit δ to form F₁(−δ) as described (8). F₁(−δ) was depleted of subunit ϵ by immunoaffinity chromatography (37), using three passages of 5–7 mg of F₁−δ through an anti-ϵ column (3 ml). At this stage, upon dilution and passage through two sequential centrifuge columns, luciferase assays (8) showed that F₁(−δϵ) had the following endogenous nucleotide content (mol/mol F₁(−δϵ) ± S.E. from four different samples): noncatalytic sites, 0.89 ± 0.13 ATP and 0.83 ± 0.04 ADP; catalytic sites, 0.1 ± 0.08 ATP and 1.49 ± 0.17 ADP.

Wild-type and mutant forms of H₆-ϵ were expressed in E. coli BL21 strain “T7 Express lysY” (New England Biolabs). H₆-ϵ was purified by affinity chromatography (column with 10 ml of TALON resin; Clontech) as described (30) but with ϵ-buffer at pH 7.5 (20 mm Tris-HCl, 100 mm NaCl, pH 7.5). After loading, the column was washed with buffer + 10 mm imidazole (10 column volumes) and buffer + 15 mm imidazole (4 column volumes) before elution with buffer + 100 mm imidazole (4 column volumes). Residual impurities were removed from H₆-ϵ by gel filtration (HiPrep-16/60, Sephacryl S100 HR, GE Healthcare Life Sciences), and pure H₆-ϵ was exchanged into ϵ-buffer + 10% (v/v) glycerol before storage at −80 °C. For some experiments, the His₆ tag was removed by treatment with 25 units/ml AcTEV^TM protease (Life Technologies) protease for 6 h. The sample was then loaded on the TALON column, and untagged ϵ was collected in the flow-through, concentrated, and frozen in ϵ-buffer + 10% glycerol.

BapH₆-ϵ was expressed from pBKH2 in E. coli strain BL21. Cells were grown at 22 °C in a medium containing Luria broth, biotin (12.2 mg/liter), 4.5 m sorbitol, and 1 m betaine. Protein expression was induced by adding isopropyl 1-thio-β-d-galactopyranoside when the A₅₉₅ reached 0.5. Cells were grown for 4 h after induction until the A₅₉₅ reached ∼1.0. BapH₆-ϵ was partially purified by TALON chromatography (as for H₆-ϵ). Fractionation with ammonium sulfate was then used; BapH₆-ϵ that precipitated between 25 and 55% saturation was dissolved in ϵ-buffer + 10% glycerol. Finally, BapH₆-ϵ was purified by gel filtration (Sephadex G-50 column, 44 cm × 1-cm diameter), concentrated by ultrafiltration (Vivaspin 6 concentrator, 5000 molecular weight cut-off), frozen in liquid N₂, and stored at −80 °C.

To express biotinylated MBP-Bap-ϵ mutants, E. coli strain DH5α was co-transformed with pBirAcm (which expresses BirA; Avidity (Aurora, CO)) and either pBKH9 (MBP-Bap-ϵ88stop) or pBKH10 (MBP-Bap-ϵA101C/L121C). Each strain was grown overnight at 37 °C in LB with ampicillin (100 μg/ml) and chloramphenicol (25 μg/ml). Overnight cultures were used to inoculate 2 liters of the same medium plus 0.4% glucose and 0.1 mm biotin. Cells were grown at 37 °C to A₅₉₅ ∼0.5, 0.4 mm isopropyl 1-thio-β-d-galactopyranoside was added to induce expression of MBP-Bap-ϵ and BirA, and growth continued for ∼3.5 h at 37 °C. Cells were harvested by centrifugation and washed once with column buffer (20 mm Tris-HCl, pH 7.5, 200 mm NaCl, 1 mm EDTA). Cells were lysed by sonication and centrifuged at 11,290 × g for 30 min, and the supernatant was passed through a 0.45-μm filter. This sample was mixed with 5 ml of amylose-agarose resin (New England Biolabs; pre-equilibrated with column buffer) and incubated at 4 °C, with rocking, for 2 h. The amylose resin was then sedimented (1000 × g, 3 min), the supernatant was discarded, and the resin was washed five times by centrifugation with 40 ml of column buffer + 5 mm βME. For the ϵA101C/L121C mutant, 5 mm βME was present throughout purification. The resin was incubated with 0.1 mg of PreScission Protease (4 °C, 3 h, in 15 ml of column buffer + 5 mm βME) to release Bap-ϵ, which was collected in the supernatant and in a subsequent wash of the resin with 10 ml of column buffer + 5 mm βME. Ultrafiltration (Vivaspin-20, 5000 molecular weight cut-off) was used to concentrate the Bap-ϵ from 25 to 1 ml and exchange it into ϵ buffer + 10% glycerol, including 1 mm βME for Bap-ϵA101C/L121C. Concentrated Bap-ϵ was frozen in liquid N₂ and stored at −80 °C. For some experiments, His₆- or Bap-tagged ϵA101C/L121C was treated with 5,5′-dithiobis(2-nitrobenzoate) (DTNB) to induce disulfide bonding between closely approaching cysteines (38). The sample was first passed through a Biogel P6 centrifuge column (39) (pre-equilibrated with 20 mm MOPS-Tris, 50 mm KCl, pH 8 (MTK8)) to remove βME. Tagged ϵA101C/L121C (∼30 μm) was then incubated with 50 μm DTNB for 15 min or less at room temperature and passed through a second centrifuge column to remove excess DTNB.

ATPase Assays

A coupled enzyme assay (40) was used for continuous monitoring of ATP hydrolysis, and assays were done at 30 °C. Decrease in NADH concentration was monitored at 340 nm in a Hewlett-Packard 8453 spectrophotometer. The standard assay buffer was MTK8 supplemented with 1 mm phosphoenolpyruvate and 0.3 mm NADH. MgATP substrate was added from stock solutions of Mg acetate and Na₂ATP; concentrations and ratios of Mg²⁺/ATP are noted for specific experiments. Pyruvate kinase (rabbit muscle, Roche Applied Science catalog number 109045) and lactate dehydrogenase (Porcine heart, Calbiochem catalog number 427211) were each present at 0.1 mg/ml in assays with excess Mg²⁺ versus ATP; for assays with excess ATP versus Mg²⁺, pyruvate kinase was 0.2 mg/ml. In some assays, hydrolysis of GTP was measured rather than ATP. For assays measuring inhibition by ϵ, BSA (fatty-acid free) was added at 0.5 mg/ml, EDTA was added to 0.1 mm, and F₁(−δϵ), ϵ, and ATP were added to final concentrations and preincubated at 30 °C for 10 min; the assay was initiated by adding magnesium acetate from a concentrated stock and mixing. Hydrolysis rates were measured at steady state, typically 12–15 min after adding Mg²⁺. For each data set varying ϵ concentration, a fixed concentration of F₁(−δϵ) was used (F_T = 0.6 or 1.2 nm), and hydrolysis rates were fit by nonlinear regression (Prism, GraphPad, Inc.) to the following equation,

where A₀ is the rate measured for F₁(−δϵ) alone, A_i is the rate measured at each ϵ concentration, ϵ_i, A_e is the rate fitted for ϵ-saturated F₁(−δϵ), and K_I is the apparent dissociation constant fitted for F₁·ϵ binding. For assays of inhibition by azide, magnesium acetate, ATP, sodium azide, and ϵ (if any) were added first, and the assay was initiated by adding F₁(−δϵ) from a concentrated stock; steady-state rates were measured as above.

Kinetic Assays of Binding and Dissociation between F₁(−δϵ) and Biotinylated ϵ, Using Biolayer Interferometry (BLI)

An Octet-RED system and streptavidin-coated sensors (FortéBio, SA biosensors, catalog number 18-5019) were used to monitor BLI kinetics of protein-protein binding and dissociation, analogous to surface plasmon resonance techniques (41). MTK8 buffer included 0.5 mg/ml BSA to minimize nonspecific binding of proteins to the sensors and as carrier protein for nanomolar dilutions of F₁(−δϵ) or ϵ. All steps were done at 30 °C, with each sensor stirred in 0.2 ml of sample at 1000 rpm and a standard measurement rate of 5 s⁻¹. Wild-type or mutant forms of biotinylated ϵ (BapH₆-ϵ, or Bap-ϵ after cleavage from MBP-Bap-ϵ) were immobilized on SA biosensors. Levels of in vivo biotinylation varied between ϵ samples, so preliminary BLI titrations were done for each biotinylated ϵ to determine the Bap-ϵ concentration and loading time needed for optimal BLI kinetic responses in subsequent binding of F₁(−δϵ) to the ϵ-loaded sensors. Minimal loading of ϵ was found to be favorable for kinetic responses to F₁(−δϵ) binding, so most subsequent experiments loaded biotinylated ϵ to yield 0.2–0.4 nm of BLI signal per sensor. Use of reference sensors with immobilized biotinylated ϵ but without added F₁(−δϵ) confirmed that added ligands (nucleotides, Mg²⁺, EDTA, and azide) did not alter the BLI signal for immobilized ϵ. To correct for BLI base-line drift and minimal nonspecific binding of F₁(−δϵ) to sensors, all BLI experiments included one or more reference sensors in parallel for which biotinylated ϵ was omitted from the ϵ-loading step, but F₁(−δϵ) was included in the association step, usually at the highest concentration of F₁(−δϵ) used for each experiment. FortéBio's analysis software (version 6.4) was used for reference subtraction, Savitsky-Golay filtering, and global fitting of kinetic rates for F₁·ϵ binding and dissociation.

RESULTS

Inhibition of E. coli F₁ by ϵ with and without the ϵCTD

Upon in vitro dissociation of E. coli F₁ from the membrane, ϵ becomes more inhibitory but can dissociate upon dilution of F₁, relieving inhibition of F₁-ATPase activity (2). For most experiments in this study, we used F₁ that was depleted of δ and ϵ subunits, or F₁(−δϵ). The stator subunit δ does not significantly affect F₁-ATPase activity (42) but was removed because its dissociation from F₁ could interfere with assays below for F₁·ϵ binding and dissociation. Fig. 2 compares inhibition of F₁(-δϵ) by WT and mutant forms of H₆-ϵ, and Table 1 summarizes the inhibition parameters from regression curves of Fig. 2 and an additional data set. As noted before (30), the N-terminal His₆ tag on WT ϵ did not significantly alter inhibition compared with WT ϵ that had the tag removed (Table 1). Also, inhibition was not altered by the N-terminal Bap tag added to ϵ (WT and mutants) for kinetic assays of F₁·ϵ binding and dissociation (not shown). For WT ϵ, values for the inhibitory constant K_I and residual activity of ϵ-saturated F₁ agree with earlier estimates (29, 43). We obtained nearly the same parameters for ϵ inhibition in assays with ATP < K_m (not shown), consistent with noncompetitive inhibition by ϵ versus ATP (29, 44). We also show that the >90% inhibition by saturating WT ϵ was unaffected by excess Mg²⁺ (Fig. 2A), although F₁(−δϵ) alone was inhibited >50% by 1 mm excess Mg²⁺ (Table 1).

FIGURE 2.

Effects of truncating or cross-linking the ϵCTD on inhibition of F₁(−δϵ). A, F₁(−δϵ) was preincubated for 10 min with 0.1 mm EDTA and ATP (black and orange symbols) or GTP (green symbols) and the indicated concentrations of wild-type H₆-ϵ (♦) or H₆-ϵ88stop (●). Hydrolysis was initiated by adding magnesium acetate. Final concentrations of added ligands were as follows: 2:1 mm ATP/Mg²⁺ (black); 1:2 mm ATP/Mg²⁺ (orange); 1:2 mm GTP/Mg²⁺ (green). B, ATPase assays as for A, but H₆-ϵA101C/L121C was used with 1 mm DTT present (▴) or with an ϵA101C–L121C disulfide bond (■). See “Experimental Procedures” for assay details and regression analysis. With H₆-ϵ88stop, data points are averages from three (ATP) or two (GTP) experiments, and standard error bars are included but are smaller than the symbols for most points. Results of regression analyses for these and for a data set with untagged WT ϵ are summarized in Table 1.

TABLE 1.

Inhibition of F₁(−δϵ) by variants of subunit ϵ

Results of nonlinear regression for data shown in Fig. 2 and a data set with untagged WT ϵ. Equation 1 was used (see “Experimental Procedures”).

ϵ	NTP/Mg2+ ratio	Activity of F1(−δϵ) (A0)a	Activity of ϵ-saturated F1 (Ae/A0)b ± S.E.c	KI ± S.E.c
	mm	μmol min−1 mg−1	%	nm
Wild type	ATP 2:1	78	7.9 ± 0.6	0.49 ± 0.02
Wild-type H6-ϵ	ATP 2:1	78	6.8 ± 0.8	0.67 ± 0.04
	ATP 1:2	37	7.9 ± 0.5	0.46 ± 0.02
	GTP 1:2	94	7.3 ± 0.7	0.87 ± 0.04
H6-ϵ88stop	ATP 2:1	73	76.3 ± 1.2	12.4 ± 1.9
	ATP 1:2	36	54.6 ± 1.4	7.2 ± 0.8
	GTP 1:2	95	84 ± 2.0	13 ± 7
H6-ϵ101C/121C, + 1 mm DTT	ATP 2:1	80	31 ± 1	1.2 ± 0.1
	ATP 1:2	41	27 ± 1	0.98 ± 0.06
H6-ϵ101C/121C, disulfide-bonded	ATP 2:1	80	69 ± 2	23 ± 3.7
	ATP 1:2	38	58 ± 1	31 ± 2

^a Hydrolysis activity units are μmol·min⁻¹·mg⁻¹ F₁(−δϵ).

^b Activity of ϵ-saturated F₁(−δϵ), A_e, is listed as a percentage of A₀, the measured activity of F₁(−δϵ) alone.

^c S.E., standard error from nonlinear regression.

To test for inhibition by ϵ lacking its CTD, we used ϵ88stop, one of the largest C-terminal deletions that still allows assembly of F_OF₁ that is functionally coupled, both in vivo and in vitro (31). In Fig. 1, both conformations of ϵ are colored darker for the C-terminal region that is absent in ϵ88stop. As shown in Fig. 2A, H₆-ϵ88stop caused much less inhibition than WT H₆-ϵ and had a >15-fold larger K_I, confirming that the ϵCTD is responsible for the majority of inhibition. However, unlike WT H₆-ϵ, the maximal extent of inhibition by H₆-ϵ88stop almost doubled to ∼45% in the presence of excess free Mg²⁺. Inhibitory effects of free Mg²⁺ are linked to inhibitory MgADP bound at a catalytic site on F₁ from E. coli (45, 46), from other bacteria (47), from mitochondria (48, 49) and chloroplasts (50, 51), and hydrolysis of GTP is less sensitive to this type of inhibition (45, 48, 52). For example, with 1 mm excess Mg²⁺, GTPase turnover by F₁(−δϵ) is ∼2.5-fold faster than ATPase (Table 1). We show that WT H₆-ϵ exhibits similar high affinity inhibition for GTPase and ATPase (Fig. 2A and Table 1). However, H₆-ϵ88stop inhibited GTPase much less, ∼16% maximal, both with excess free Mg²⁺ present (Fig. 2A) and without it (not shown). Thus, observed partial inhibition of ATPase by the ϵNTD is largely due to increased MgADP inhibition in the absence of the ϵCTD. This can also explain why ϵ truncated after ϵ94 (with only about half of the first helix remaining) inhibited E. coli F₁ ∼50% because the assays contained 2 mm excess free Mg²⁺ (53). The effects of the ϵNTD are distinct from the >90% inhibition caused by the ϵCTD of intact WT ϵ, which is not sensitive to the effects of excess Mg²⁺.

As an alternative to removing the ϵCTD, we also used the ϵA101C/L121C mutant (21).³ This cysteine pair can form a disulfide bond in nearly 100% yield (Fig. 3) that cross-links the two α-helices of the ϵCTD in the ϵ_C conformation, preventing ϵ from switching to the ϵ_X conformation (see Fig. 1). As shown in Fig. 2B, this disulfide linkage prevented high affinity inhibition of F₁(−δϵ) as effectively as removing the ϵCTD. However, the partial inhibition observed was less sensitive to excess free Mg²⁺ than with H₆-ϵ88stop; this suggests that ϵCTD/ϵNTD interactions in the ϵ_C state can influence interactions of ϵNTD with γ that alter catalytic behavior. With DTT present to prevent the disulfide bond, H₆-ϵA101C/L121C could access the ϵ_X state and showed high affinity inhibition (K_I ∼1 nm), similar to that with WT H₆-ϵ (K_I ∼0.5 nm). However, the activity of F₁ saturated with reduced H₆-ϵA101C/L121C was 4-fold greater than with WT H₆-ϵ. In the structure of ϵ-inhibited F₁ (8), ϵLeu-121 is in a coiled-coil interface with the γ N-terminal helix, and the ϵL121C mutation probably perturbs this interface, favoring more F₁ complexes in the active state on average. This supports the concept that, with ϵ-saturated F₁, the residual ATPase activity (7–8% with WT ϵ) is due to the time-averaged fraction of F₁ complexes in which ϵ is not in the inhibitory ϵ_X conformation.

FIGURE 3.

SDS-PAGE analysis of disulfide bond formation with ϵA101C/L121C. Samples were applied to 12% SDS-PAGE under non-reducing conditions. Lane 1 of each gel, molecular mass markers (approximate kDa noted for gel A). Gel A, samples of H₆ϵA101C/L121C (1 μg/lane) were taken immediately after DTT was removed from the sample by centrifuge column (lane 2) or after 2 min of reaction with 50 μm DTNB (lane 3). Gel B, samples of (MBP-cleaved) Bap-ϵ (5 μg/lane) were taken after removing DTT as above (lane 2) or after reaction with 50 μm DTNB for 15 min (lane 3). With 2-mercaptoethanol in the gel sample buffer (not shown), each ϵA101C/L121C migrated only at the upper band position (above 14 kDa standard), confirming that the lower band seen here was a faster migrating band due to the internal ϵA101C–L121C disulfide bond.

Kinetics of F₁·ϵ Binding and Dissociation, Assayed by BLI

In preliminary assays, H₆-ϵ was loaded on BLI sensors coated with Ni²⁺-nitrilotriacetic acid, but slow dissociation of WT H₆-ϵ from the sensors prevented accurate measures of the slow dissociation rate of F₁ from WT H₆-ϵ. To achieve more stable and specific attachment of ϵ to the sensor surface, ϵ was engineered with an N-terminal Bap tag, so that a specific lysine could be biotinylated in vivo (32). Biotinylated Bap-ϵ could be stably bound to streptavidin-coated sensors (supplemental Fig. S1), and BLI was then used to measure binding and dissociation kinetics of F₁(−δϵ). For each Bap-ϵ variant, 4–7 sensors were used in parallel, with F₁(−δϵ) concentrations varied ≥10-fold in the association samples, and association/dissociation kinetics were fit globally to determine the rate constants (Table 2). K_D values derived from the rate constants correlate well with inhibitory K_I values (Table 1) for WT ϵ, ϵ88stop, and disulfide-bonded ϵA101C/L121C. Representative kinetics for binding/dissociation of F₁(−δϵ) with sensors containing WT ϵ or ϵ88stop are shown in Fig. 4. The ϵCTD did not significantly alter the association rate, indicating that only ϵNTD/γ interactions are involved in initial F₁·ϵ binding. In contrast, removing the ϵCTD (Fig. 4) or preventing it from adopting the ϵ_X state (disulfide-bonded ϵA101C/L121C) increased the dissociation rate by ≥80-fold (Table 2). For sensors loaded with biotinylated WT BapH₆-ϵ, only a small fraction of bound F₁(−δϵ) could be observed to dissociate in buffer only (Fig. 4), but results presented below show that essentially all F₁(−δϵ) on the sensor is reversibly bound. Additional assays (not shown) included excess, non-biotinylated WT H₆-ϵ in the dissociation phase and confirmed that the observed, slow dissociation rate was not due to rebinding of F₁(−δϵ) to the sensor. Thus, the much slower dissociation of F₁(−δϵ)/WT-ϵ is probably due to strong bias of bound WT ϵ to reside in the ϵ_X state, with the ϵCTD buried within F₁. However, from the K_D values (Table 2), note that the ϵCTD contributes only ∼20% to the net free energy for F₁·ϵ binding (ΔΔG, −10 or −12 kJ/mol for WT ϵ versus ϵ88stop or disulfide-bonded ϵA101C–L121C, respectively). This does not mean the ϵ_X state of ϵCTD has only weak interactions with other F₁ subunits; rather, the small contribution to net binding energy is probably due to the loss of favorable interactions between γ and α₃β₃ that are blocked by insertion of the ϵCTD.

TABLE 2.

Binding/dissociation rates and K_D values for F₁(−δϵ) with variants of subunit ϵ

Biotinylated ϵ	ka ± S.E.a	kd ± S.E.a	KD
	m−1 s−1	s−1	nm
Wild type	2.0 × 105 ± 0.1%	4.8 × 10−5 ± 0.1%	0.24
Wild type, +1 mm EDTA/ATP	2.3 × 105 ± 0.1%	4.1 × 10−3 ± 0.1%	17
ϵ88stop	3.1 × 105 ± 0.6%	3.8 × 10−3 ± 0.4%	12.2
ϵΑ101C/L121C, disulfide-bonded	2.1 × 105 ± 0.3%	6.6 × 10−3 ± 0.2%	32

^a S.E., standard error from global fitting analysis (presented as a percentage of the parameter's value).

FIGURE 4.

Binding and dissociation kinetics for F₁(−δϵ) and sensor-bound biotinylated ϵ, with or without ϵCTD. Biotinylated Bap-ϵ (wild type or ϵ88stop, as noted) was loaded onto streptavidin biosensors. At time 0, each sensor was transferred from buffer alone to a sample containing 15 nm F₁(−δϵ). After 900 s (vertical dashed line), each sensor was moved into buffer without F₁(−δϵ). Black lines, experimental data; red lines, the kinetic fits for each ϵ, from global regression of kinetic data at varied concentrations of F₁(−δϵ). Kinetic parameters and K_D values derived from the fittings are summarized in Table 2.

Effects of F₁ Ligands (Mg²⁺, Nucleotides, P_i) on Conformational Bias of Bound, Full-length ϵ

From the crystal structure of ϵ-inhibited F₁ (8), the extensive surface area of ϵCTD that is buried within the central cavity of F₁ suggests that the ϵ_X state of ϵ does not directly dissociate from F₁; the slow dissociation observed in Fig. 4 probably occurs due to dynamic transition of ϵ between ϵ_X and conformations like ϵ_C in which the ϵCTD is outside of the central rotor cavity. Thus, factors that influence the fraction of F₁ complexes with ϵ in the ϵ_X state should alter the kinetics of F₁·ϵ dissociation. To show that the conformation of ϵ on E. coli F₁ and F_OF₁ can be influenced by nucleotides and other ligands that interact with catalytic sites, early studies used static assays, such as the capacity to form a β-ϵ cross-link; cross-linking of β-ϵ was minimized by non-hydrolysis conditions, such as ATP/EDTA or MgAMPPNP but maximized by post-hydrolysis conditions (MgADP/P_i) (54, 55). The β-ϵ cross-linking residues (56) are within hydrogen-bonding distance in the structure of ϵ-inhibited F₁ but should be at least 35 Å apart with ϵ in the ϵ_C state (8). Here, we use the BLI assay for F₁·ϵ binding/dissociation for more dynamic analyses of how different ligands may shift the conformation of F₁-bound ϵ between the ϵ_X state and other conformations of the ϵCTD that allow faster F₁·ϵ dissociation. In control assays (not shown), the various ligands tested did not alter the rate at which F₁(-δϵ) dissociated from biotinylated ϵ88stop, confirming that the ligand effects are specific to the ϵCTD of WT ϵ. F₁(−δϵ) was bound in parallel to multiple sensors with biotinylated WT ϵ, and Fig. 5 shows dissociation of F₁(−δϵ) when sensors were exposed to different ligands. For Fig. 5A, F₁(−δϵ) was bound to all sensors in MTK8 buffer + BSA, and dissociation in this buffer was slow (Fig. 5A, curve 4, ∼5.4 × 10⁻⁵ s⁻¹ ± 0.2%). This is consistent with access to the ϵ_X state when F₁ is in a post-hydrolysis conformation because isolated F₁(−δϵ) retained ∼1.5 ADP (mol/mol) but negligible ATP at catalytic sites (see “Experimental Procedures”). Added MgADP/P_i (Fig. 5A, curve 5) appeared to stabilize the ϵ_X state, consistent with prior β-ϵ cross-linking results (55). However, a similar effect was achieved by adding only Mg²⁺ and P_i (Fig. 5A, curve 6), suggesting that the endogenous ADP in isolated F₁(−δϵ) was sufficient to stabilize the ϵ_X state upon the addition of Mg²⁺ and P_i. The importance of P_i in stabilizing this state (55) was also observed here because MgADP alone (Fig. 5A, curve 2) allowed a significant fraction of F₁ to dissociate faster.

FIGURE 5.

Effects of other ligands on dissociation of F₁ from sensor-bound wild-type ϵ. F₁(−δϵ) (50 nm) was incubated for ∼10 min in MTK8 + BSA buffer only (A) or plus 1 mm ATP/EDTA (B) and then incubated with BLI sensors containing WT BapH₆-ϵ to equilibrate F₁·ϵ binding. Signals for bound F₁(−δϵ) varied slightly between sensors (6–9%), so data were normalized for display. Results show the kinetics of F₁(−δϵ) dissociation upon moving sensors into buffer with different ligands. C and D, selected results from A and B, respectively, for the initial 60 s of dissociation. When present, Mg²⁺ was at 2 mm; all other ligands were 1 mm. Curve 1, ATP/EDTA; curve 2, MgADP; curve 3, MgATP; curve 4, buffer only; curve 5, MgADP/P_i; curve 6, Mg²⁺/P_i.

In contrast to slow F₁·ϵ dissociation under post-hydrolysis conditions, the addition of 1 mm ATP/EDTA caused ∼94% of F₁·ϵ to dissociate ∼80-fold faster (Fig. 5A, curve 1, 4.2 × 10⁻³ s⁻¹ ± 0.1%) and with <3-s transition to faster dissociation (Fig. 5C). This effect is due to ATP because EDTA alone had a minimal effect on F₁ dissociation from immobilized WT ϵ (not shown). Further, by including 1 mm ATP/EDTA during association and dissociation phases, global analysis of F₁·ϵ binding/dissociation shows that ATP/EDTA did not alter the F₁·ϵ binding rate but gave a dissociation rate and K_D similar to values for ϵ88stop (Table 2). The presence of nonhydrolyzable MgAMPPNP (2:1 mm) during the dissociation phase (not shown) also accelerated F₁·ϵ dissociation, indicating that nucleotide binding alone is sufficient to shift F₁ to a conformation that does not allow the ϵCTD to insert into F₁ and form the ϵ_X state. Also, the ability of MgADP/P_i to stabilize the inhibitory state of ϵ was readily reversible; even when F₁(-δϵ) was bound to WT ϵ/sensors for 45 min with MgADP/P_i present, switching the sensors to buffer with MgAMPPNP immediately caused >90% of F₁(−δϵ) to dissociate at the faster rate (not shown; 3.7 × 10⁻³ s⁻¹ ± 0.1%).

For the experiment in Fig. 5B, F₁(−δϵ) was bound to immobilized WT ϵ in the presence of 1 mm ATP/EDTA, so that most F₁·ϵ complexes would not have ϵ in the slowly dissociating ϵ_X state at the time the sensors were moved to dissociation wells. As expected, F₁ dissociation was fast with ATP/EDTA present (Fig. 5B, curve 1). With buffer only (Fig. 5B, curve 4), most F₁ still dissociated fast. This could indicate that ATP bound during the F₁·ϵ association phase dissociated slowly or that endogenous ADP had dissociated from F₁ during the association phase due to the ATP/EDTA present. MgADP alone (Fig. 5B, curve 2) slowed dissociation of most F₁, but MgADP/P_i (Fig. 5B, curve 5) or Mg²⁺/P_i (Fig. 5B, curve 6) effectively reversed the ATP/EDTA effect so that almost all F₁ dissociated very slowly. Mg²⁺ was essential for this effect; without it, F₁ dissociation in the presence of 1 mm P_i (not shown) was nearly identical to that in buffer alone (Fig. 5B, curve 4). The effect of Mg²⁺ was enhanced by submillimolar P_i (not shown, K½ ∼0.2 mm P_i), similar to how P_i enhanced MgADP protection of F₁-bound ϵ from trypsin (55).

With Mg²⁺/P_i, there was no apparent lag in reverting F₁·ϵ to slow dissociation (Fig. 5D, curve 6). This suggested that, after F₁·ϵ association with ATP/EDTA present, rapid reversion by Mg²⁺/P_i to slow F₁·ϵ dissociation required hydrolysis of ATP that remained bound at a catalytic site. In the dissociation step, added Mg²⁺ could complex with the bound ATP, and hydrolysis would return F₁ to the catalytic dwell step, at which insertion of the ϵCTD into the central cavity appears to occur. The bound MgADP/P_i present would then stabilize F₁ with ϵ in the ϵ_X state. The experiment shown in Fig. 6 tested this possibility. With non-hydrolyzable MgAMPPNP present during F₁·ϵ association, dissociation of most F₁·ϵ was fast in the presence of MgAMPPNP (Fig. 6, curve 1) or in buffer alone (Fig. 6, curve 2), similar to the effects of ATP/EDTA in Fig. 5B. However, with MgAMPPNP present during association, inclusion of Mg²⁺/P_i during dissociation failed to prevent fast dissociation of most F₁ (Fig. 6, curve 3), in contrast to the parallel control with ATP/EDTA in association (Fig. 6, curve 4). These results indicate that, with catalytic nucleotide bound in a prehydrolysis state, the rotary conformation of F₁ does not allow insertion of the ϵCTD into the central rotor cavity, but hydrolysis at the catalytic dwell allows the ϵCTD access to insert and form the ϵ_X state.

FIGURE 6.

Contrasting effects of Mg²⁺/P_i on F₁·ϵ dissociation after preincubating F₁ with ATP/EDTA or MgAMPPNP. F₁(−δϵ) (50 nm) was incubated ∼10 min in MTK8 buffer containing 2:1 mm MgAMPPNP (curves 1–3) or 1 mm ATP/EDTA (curve 4) and incubated with BLI sensors containing WT BapH₆-ϵ. BLI signals for bound F₁(−δϵ) were normalized as in Fig. 5. Kinetics of F₁ dissociation were monitored in the presence of 2:1 mm MgAMPPNP (curve 1), buffer only (curve 2), or 2:1 mm Mg²⁺/P_i (curves 3 and 4).

In the experiments of Fig. 4, hydrolysis conditions had complex effects on F₁·ϵ dissociation. With or without ATP/EDTA during F₁·ϵ binding, MgATP in the dissociation phase (curve 3) induced a small or negligible rate of F₁·ϵ dissociation during the initial 60 s (Fig. 5, C and D). Comparable with conditions for Fig. 5B, assays for ϵ inhibition of F₁-ATPase (Fig. 2) included an ATP/EDTA preincubation, and ϵ-saturated F₁ had initial ATPase activity (1–2 min, not shown) that was ∼85% of the steady-state, inhibited rate. Thus, compared with fast F₁ dissociation in the continued presence of ATP/EDTA, hydrolysis of MgATP initially reverted F₁·ϵ complexes to slow dissociation, rapidly re-establishing the bias toward the inhibitory ϵ_X state (Fig. 5D, curves 1 and 3). This is consistent with noncompetitive inhibition of ϵ versus MgATP and, combined with other results above, indicates that ϵ accesses the inhibitory ϵ_X state following hydrolysis at the catalytic dwell step. On the longer time scale of Fig. 5 (A and B), hydrolysis conditions increased the dissociation rate for a fraction of F₁·ϵ complexes. As indicated by other experiments below, this slow effect on F₁·ϵ dissociation is probably due to gradual competitive transition of some active complexes to the ADP-inhibited state, which favors faster dissociation of ϵ. The fast dissociating fraction was substantially smaller for F₁·ϵ complexes formed in the presence of ATP/EDTA (Fig. 5B), probably because the ATP/EDTA preincubation minimizes initial inhibition (not shown) due to ADP-inhibited complexes.

Inhibition of F₁(−δϵ) by Azide and the ϵCTD Are Competing Processes

The above results support prior SM mechanics studies (22–24, 26) that concluded that ϵ inhibition pauses rotation at the catalytic dwell position. In the absence of ϵ, long pauses at the catalytic dwell have been documented and attributed to inhibitory MgADP (57, 58). However, there have been conflicting conclusions about the relationship between inhibitory MgADP and ϵ inhibition for F₁ of different bacterial species (22, 24, 25, 27). We investigated this by testing interactions between ϵ inhibition and inhibition by sodium azide, which acts by stabilizing the MgADP-inhibited state (45, 59–61). We first tested inhibition of F₁(−δϵ) by azide, with or without excess WT or mutant forms of H₆-ϵ present. Inhibition of F₁(−δϵ) alone showed a K_I of ∼5 μm azide, whether assays were done with 2:1 mm Mg/ATP (Fig. 7) or with 1:2 mm Mg/ATP (not shown). Thus, for E. coli F₁, azide inhibition is separated from the step that confers sensitivity to inhibition by excess free Mg²⁺. The K_I for azide was not altered by bound H₆-ϵ mutants that could not access the ϵ_X state due to truncation (ϵ88stop) or disulfide bonding (ϵA101C–L121C). The presence of 100 nm WT ϵ reduced the activity of F₁(−δϵ) ∼10-fold, but the residual activity was still inhibited by excess azide. However, bound WT H₆-ϵ increased the K_I for azide ∼5-fold (Fig. 7). Also, saturating F₁(-δϵ) with reduced H₆-ϵA101C/L121C, which was less inhibitory than WT H₆-ϵ, yielded an intermediate K_I value for azide inhibition. These results indicate that forming the inhibitory ϵ_X state competes with azide's capacity to inhibit F₁-ATPase. To test whether azide also competed with formation of the ϵ-inhibited state, F₁(−δϵ) was bound to immobilized WT BapH₆-ϵ in buffer alone, and F₁·ϵ dissociation was measured under hydrolysis conditions with varied concentrations of azide (Fig. 8A). Increasing azide concentrations caused greater fractions of F₁·ϵ to dissociate at a faster rate, and the hyperbolic dependence on azide (Fig. 8B) yielded a K½ of ∼14 μm, comparable with the mid-range of K_I values for azide inhibition with or without WT H₆-ϵ (Fig. 6). Taken together, the results of Figs. 7 and 8 show competition between (i) the ability of the ϵCTD to insert, forming the inhibitory ϵ_X state, and (ii) the ability of azide to bind to and stabilize the MgADP-inhibited state of F₁.

FIGURE 7.

Effects of excess ϵ variants on inhibition of F₁-ATPase by azide. Steady-state ATPase rates were measured for F₁(−δϵ) in the presence of varied concentrations of sodium azide. Assays were done in the absence of ϵ or in the presence of excess variants of H₆-ϵ. Each data set was fit to the equation, v = V/(1 + [azide]/K_I), and then ATPase rates were normalized to V (μmol·min⁻¹·mg⁻¹ F₁(−δϵ), without azide) = 100%. Without ϵ, similar K_I values (μm) were obtained for assays done with 2:1 mm ATP/Mg²⁺ (V = 78 ± 3, K_I = 4.7 ± 0.7, not shown) or 1:2 mm ATP/Mg²⁺ (○; dashed line, V = 29.3 ± 0.5, K_I = 5.3 ± 0.3). Assays were done with 1:2 mm ATP/Mg²⁺ in the presence of 100 nm wild-type ϵ (●; V = 3.8 ± 0.04, K_I = 24.5 ± 0.9) or 100 nm ϵ88stop (▵; V = 17.9 ± 0.3, K_I = 4.6 ± 0.2) or with 2:1 mm ATP/Mg²⁺ in the presence of 100 nm ϵA101C/L121C + 5 mm DTT (▾; V = 30.9 ± 0.3, K_I = 13.5 ± 0.4) or in the presence of 300 nm disulfide-bonded ϵA101C/L121C (▿; V = 75.6 ± 0.7, K_I = 6.2 ± 0.2). In control assays (not shown), 5 mm DTT had no direct effect on ATPase activity of F₁(−δϵ).

FIGURE 8.

Effects of azide on dissociation of F₁ from sensor-bound, wild-type ϵ. A, kinetics for F₁·ϵ dissociation measured by BLI. Prior to the F₁ dissociation phase shown, BLI sensors had been loaded with biotinylated wild-type ϵ and then incubated for 900 s in buffer with 20 nm F₁(−δϵ); the normalized BLI signal at time 0 represents bound F₁(−δϵ). Dissociation of F₁(−δϵ) from immobilized ϵ was done in hydrolysis conditions (1 mm ATP, 2 mm Mg²⁺) with increasing concentrations of azide, as indicated. Dissociation kinetics were biphasic, with “fast” (1–2.5 × 10⁻³ s⁻¹) and “slow” (0.8–1.6 × 10⁻⁴ s⁻¹) fractions (nonlinear regression fits not shown but essentially superimpose with data at scale shown; R² > 0.9995 for all fits). B, the azide dependence of F₁·ϵ dissociation. The fraction of complexes that dissociated fast versus slow shows hyperbolic dependence on azide concentration, with K½ = 14 ± 2.3 μm; R² = 0.9935.

DISCUSSION

Correlating Results with Functional and Rotational States of the Enzyme

Hydrolysis by the three alternating catalytic sites of F₁ drives a full 360° rotation of the γ central rotary shaft, so hydrolysis of each ATP molecule involves a 120° rotation of γ relative to α₃β₃. As depicted schematically in Fig. 9, SM microscopy studies of bacterial F₁-ATPases (reviewed in Refs. 62 and 63) have shown that each 120° rotation is comprised of two observable rotary substeps (solid arrows); MgATP binding at one alternating catalytic site drives an ∼80° rotation of γ to the catalytic dwell angle, whereas the subsequent ∼40° rotation is limited by the intrinsic rates of hydrolysis and P_i dissociation at the other alternating catalytic sites. At 120°, Mg²⁺ and ADP dissociate from one site before or in concert with binding of MgATP at another site to drive the next 80° substep. Fig. 9 also includes bars along the rotational arc that depict the observed angular position of γ (relative to α₃β₃) in crystal structures of F₁ or F₁·c-ring complexes. This is based on alignment of structures by a structurally conserved, apparently stiff core of the γ subunit that we identified previously (8). A recent analysis of MF₁ structures concluded that the position of γ could not be correlated with rotational angle during the catalytic cycle, arguing that the part of γ protruding below α₃β₃ is variably displaced by lattice contacts in different crystals (64). However, biophysical characterization of the stiffness of γ indicates that only the lowest portion of γ, near its interface with the c-ring of F_O, is extremely pliable (65). Also, the 99 residues of γ that we identified as a structurally conserved “γ-core” include (i) most of the γ coiled-coil that is inside α₃β₃, (ii) the first ∼15 residues of the γ C-terminal helix that protrude below α₃β₃, and (iii) 41 residues of the γ Rossmann fold domain that pack beside the protruding part of the γ C-terminal helix (see supplemental Figs. 4 and 5 of Ref. 8). We have updated our analysis of the rotational angle of γ to include 34 F₁ or F₁·c-ring structures and find that all γ subunits superimpose well with the γ-core (supplemental Table S-I). Thus, the distribution of F₁ structures along the rotary arc of γ in Fig. 9 should be useful for comparing structural states with functional and rotational data.

FIGURE 9.

Proposed paths for inhibition by ϵCTD or ADP/azide, relative to rotary catalytic substeps and angular positions of γ in F₁ structures. A 120° segment is shown for the γ rotational cycle, corresponding to net hydrolysis of 1 ATP. Two rotary substeps, observed by SM microscopy, are depicted by solid arrows (direction is for rotation of γ during hydrolysis). Bars along the inner edge of the 80–120° arc indicate the rotary angle of γ in 34 F₁ crystal structures (see supplemental Table S-I). Bars are spaced every 5°, and the height represents the number of F₁ structures aligned near each angle ± 2.5°; shortest bar, one structure; longest bar, 12 structures. Bars are shaded for structures of bovine F₁ (black), yeast F₁ (gray), and E. coli F₁ (checkered). The dotted arrows below the bars indicate the proposed paths to and from the ADP-inhibited state, as stabilized by azide at ∼95° (61). The dashed arrow shows the ∼40° rotary shift proposed to occur during transition into (counterclockwise) or out of (clockwise) the ϵ-inhibited state. The two F₁ aligned past 120° are also shown near 0° because this position should be the starting point for the next 120° turn.

Although there is no current consensus for correlating the rotary position of γ in known structures with the dwell states observed in SM assays after 80 and 40° substeps (63, 66, 67), we assign γ to be at 80° for one MF₁ structure, with all three catalytic sites filled with nucleotide (Protein Data Bank entry 1H8E), because structural considerations (68) and molecular dynamics simulations (69) suggest that it is closest to the catalytic dwell state. Most MF₁ structures have no nucleotide in the β_E site due to an open conformation that distorts the nucleotide-binding pocket, but 1H8E has a “half-closed” conformation of β_E with bound MgADP and SO₄²⁻ (thought to mimic P_i binding). Another recent MF₁ structure (Protein Data Bank entry 4ASU) also has a nucleotide in all three catalytic sites, but its β_E is much closer to the open conformation and has ADP but no bound SO₄²⁻ (or P_i) or Mg²⁺ (64). The 4ASU structure was proposed to represent the catalytic intermediate from which final products Mg²⁺ and ADP dissociate, and its rotary position at ∼123° (Fig. 9) correlates with SM studies indicating that ADP dissociates after ∼40° rotation to one of the 120° dwell positions (70–72). The structural indication that Mg²⁺ dissociates before ADP (64) suggests that the inhibitory effect of free Mg²⁺ occurs at the 120° position. This can explain results showing that excess free Mg²⁺ does not affect inhibitory transitions that occur at the catalytic dwell step at 80°; free Mg²⁺ does not affect the rate at which actively rotating E. coli F₁(−ϵ) switches to the ADP-inhibited (paused) state (22), and, as shown here, free Mg²⁺ does not alter inhibition of E. coli F₁ by ϵCTD or azide.

Although our results indicate that ADP and ϵ inhibition begin as competing processes after hydrolysis at the catalytic dwell, F₁ structures of these inhibited states have γ positioned at angles that are distinct from the catalytic dwell; in Fig. 9, azide-inhibited MF₁ (61) has γ at 93°, and ϵ-inhibited E. coli F₁ (8) has γ rotated much further to 123° (checkered bar). In contrast, SM studies concluded that both ADP and ϵ inhibition cause long paused/inactive states at the catalytic dwell at 80° (22–24, 57). With a bead attached to γ or ϵ, it is possible that SM microscopy could have overlooked dynamic oscillations between 80 and 120° because those assays can exhibit broad angular distributions of events. For example, during long paused periods (up to 1–2 s) without net rotation, a γ-attached bead on E. coli F₁(−ϵ) showed rapid, ongoing angular fluctuations spanning at least ±30° (Fig. 3A in Ref. 22). This could represent dynamic rotational oscillations in E. coli F₁ or could be a technical limitation because the bead was attached to γ by a single cysteine, allowing for significant flexibility in the linkage. On the other hand, there is prior evidence that functional rotation is needed for transition to and from the ϵ-inhibited state; with E. coli F_OF₁ in liposomes, a chemical treatment that blocks rotation of F_O also prevented nucleotide-dependent changes in the conformation of ϵ (54). With F₁ exposed to MgADP/P_i, conditions that stabilize the ϵ-inhibited state (Fig. 5), cryoelectron microscopy of E. coli F₁ showed a unique, ϵ-dependent position of γ and a unique position of ϵ relative to α₃β₃ (73). Also, with F_OF₁-liposomes, SM fluorescence assays showed a shift in the position of the ϵ NTD relative to the F_O stator for active versus inactive complexes (74). Thus, we propose that the ϵCTD begins inserting into bacterial F₁ near the catalytic dwell (80°) but then induces partial rotation to ∼120° to achieve the final ϵ-inhibited state, with the last half of the ϵCTD buried in the central cavity of F₁ (8). This is similar to a proposal that insertion of the mitochondrial inhibitor protein (IF₁) into MF₁ involves rotational steps (75), and IF₁-inhibited MF₁ has γ rotated ∼27° past the catalytic dwell in Fig. 9. The transitions of azide-inhibited MF₁ and ϵ-inhibited E. coli F₁ to distinct rotary angles may help explain the competition between these inhibitory paths. Azide-inhibited MF₁ has azide bound with MgADP in the closed, high affinity β_D site, but azide inhibition is unlikely to occur in the ϵ-inhibited state of E. coli F₁ because insertion of the ϵCTD and rotation of γ shift β_D to a distinct “half-closed” conformation. Conversely, if E. coli F₁ first shifted to the ADP-inhibited state, azide would probably stabilize the closed β_D state and so prevent opening of the α_Eβ_D interface with γ that is necessary to allow insertion of the ϵCTD. In Fig. 9, the broken lines below the arc indicate the proposed rotational paths leading to and from ADP- or ϵ-inhibited states. In SM tests of forced rotation, TF₁ was preferentially activated from the ADP-inhibited state (paused at ∼93°; Fig. 9) by >40° rotation forward, and, consistent with proposed release of ADP near 120° (Fig. 9), added ADP suppressed or reversed rotational activation (76). In contrast, forced rotation of up to 120° in either direction failed to reactivate ϵ-paused TF₁ (26). Based on the asymmetric insertion of ϵCTD within F₁ (8), we suspect that reactivation from the ϵ_X-inhibited state (Fig. 9, near 120°) occurs with the lowest activation barrier by reverse rotation of γ/ϵNTD toward the catalytic dwell angle at 80° (Fig. 9, dashed arrow). Such reactivation by rotation in the direction of ATP synthesis may be indicated by a study with E. coli F_OF₁-liposomes; with MgADP and P_i present, prior exposure to proton motive force activated the initial rate of subsequent ATPase activity up to 9-fold (77).

Dynamics of ϵ Conformational Changes and the Influence of Nucleotides/Ligands

Under conditions for ATP hydrolysis, SM assays with a 60-nm gold bead attached to γ showed that E. coli F₁ complexes switch back and forth between actively rotating and paused states every few seconds, with or without ϵ bound to F₁ (22). Our results on F₁·ϵ dissociation kinetics are consistent with such rapid exchange between active and inactive states. As shown in Fig. 5 (C and D), there were no more than a few seconds lag in ligand-dependent switching between the slowest and fastest modes of F₁·ϵ dissociation. Exposure to saturating nucleotide without hydrolysis (ATP/EDTA or MgAMPPNP) induced the fastest and essentially monophasic F₁·ϵ dissociation (Figs. 5 and 6), indicating that few F₁·ϵ complexes remained in or could regain access to the ϵ_X-inhibited state. However, most F₁·ϵ complexes still dissociated slowly upon exposure to hydrolysis conditions (Fig. 5), consistent with noncompetitive inhibition by ϵ versus ATP for E. coli F₁ (29, 44). Also, the K_I for ϵ inhibition of steady-state hydrolysis (Table 1, ∼0.5 nm) is similar to the K_D for F₁·ϵ binding (Table 2, 0.24 nm), which was measured without added nucleotides. Thus, it is unlikely that catalytic binding of ATP or MgAMPPNP directly increases the rate at which the ϵ_X-paused state returns to an active form. Rather, we propose that the intrinsic rates to and from the ϵ_X-inhibited state are fast enough (seconds or less) to allow ligands to influence subsequent conformational changes once F₁ exits from the ϵ-inhibited state. Once the ϵCTD escapes from the central cavity of F₁ near 80°, F₁ would be most likely to rotate forward, completing an active 40° step (Fig. 9, solid arrow). At that point, binding of nucleotide would drive an ∼80° step toward the next catalytic dwell but, without hydrolysis, would trap F₁ in a conformation and rotary position that would not allow the ϵCTD to reinsert; thus, F₁·ϵ complexes would dissociate at the faster rate observed with only γ/ϵNTD interactions. With hydrolysis conditions, each return to a catalytic dwell step would provide the same low probability for insertion of the ϵCTD, and, as indicated by SM studies (22), the long durations of the paused/inhibited states (1 to 3.5 s) relative to the limiting catalytic steps (1–2 ms) would result in a large fraction of inhibited complexes during steady-state hydrolysis.

In an SM study with E. coli F₁ (22), the presence of ϵ did not alter the duration time of active complexes (0.5–1 s), but ϵ-paused states had longer duration times (up to 3.5 s) than ADP-paused states (∼1 s). This is consistent with other indications that ϵ inhibition predominates over ADP inhibition. Oxyanions, which are thought to activate F₁ by promoting release of inhibitory ADP (78), activate E. coli F₁ more if ϵ is absent (28), and recently it was shown that the oxyanion selenite optimally activates ϵ-depleted E. coli F₁ ∼10-fold⁴ with excess free Mg²⁺ present but only activates ϵ-saturated F₁ 2–2.5-fold (79). Also, because our results show that inhibition by the ϵCTD is distinct from the ADP-inhibited transition, the unaffected duration of the active state (22) suggests that a prior common step is rate-limiting for transitions to either ADP- or ϵ-inhibited states. This common step is probably the end of hydrolysis that precedes P_i release because ϵ reduces the rate of P_i release ∼15-fold following “unisite” hydrolysis (28). P_i stabilizes F₁ with ϵ in the ϵ_X state, which could mean that P_i rebinds to β_D (with bound MgADP) and delays an active 40° rotation, allowing more time for a possible transition to the ϵ_X-inhibited state. However, we cannot rule out that the P_i effect could be due in part to binding to other β(s), which show SO₄³⁻ bound at the “P-loop” in ϵ-inhibited F₁ (8).

Conclusions

This study sheds further light on how the CTD of subunit ϵ inhibits the catalytic F₁ complex of a bacterial ATP synthase. Most significantly, results reveal that ATP hydrolysis is required for insertion of the inhibitory ϵCTD into F₁ at the catalytic dwell step and that ϵ inhibition competes with conversion to an ADP-inhibited state of the enzyme. With insertion of the ϵCTD starting at the catalytic dwell (∼80°), the dynamic response of the conformation of ϵ to catalytic site ligands and the structurally observed γ angle of ∼123° in ϵ-inhibited F₁ suggest dynamic, reversible rotation over the 40° substep. Our results also show that the ϵCTD has a small energetic contribution to net binding of ϵ to F₁. Thus, there is potential for antibiotic development by discovering or designing compounds that enhance or mimic ϵ inhibition of bacterial ATP synthases. The BLI assays established here for kinetics of F₁·ϵ binding and dissociation should be valuable in further analyzing which ϵCTD residues and interactions are critical for ϵ inhibition in F₁-ATPase from E. coli and other bacteria. Of course, the BLI assay cannot be used to study ϵ inhibition in membrane-bound ATP synthase because ϵ does not dissociate from intact F_OF₁. The extent of ϵ inhibition can vary widely for membranes isolated from different bacteria. For example, E. coli membranes exhibit substantial ATPase activity, whereas mycobacterial membranes are devoid of ATPase activity but can be activated by treatment that probably damages the ϵ subunit (80). Thus, additional approaches will be needed to determine what other factors influence ϵ inhibition in ATP synthases of different bacterial species.