Aerobic Growth of Escherichia coli Is Reduced, and ATP Synthesis Is Selectively Inhibited when Five C-terminal Residues Are Deleted from the ϵ Subunit of ATP Synthase*

Background: Bacterial ATP synthases are autoinhibited by subunit ϵ. Results: Altering the regulatory interactions of ϵ increases inhibition of ATP synthesis and reduces respiratory growth of E. coli. Conclusion: The ϵ subunit can have distinct regulatory interactions during ATP synthesis versus hydrolysis. Significance: Inhibition by ϵ provides a bacteria-specific means to target ATP synthase for antibiotic development. F-type ATP synthases are rotary nanomotor enzymes involved in cellular energy metabolism in eukaryotes and eubacteria. The ATP synthase from Gram-positive and -negative model bacteria can be autoinhibited by the C-terminal domain of its ϵ subunit (ϵCTD), but the importance of ϵ inhibition in vivo is unclear. Functional rotation is thought to be blocked by insertion of the latter half of the ϵCTD into the central cavity of the catalytic complex (F1). In the inhibited state of the Escherichia coli enzyme, the final segment of ϵCTD is deeply buried but has few specific interactions with other subunits. This region of the ϵCTD is variable or absent in other bacteria that exhibit strong ϵ-inhibition in vitro. Here, genetically deleting the last five residues of the ϵCTD (ϵΔ5) caused a greater defect in respiratory growth than did the complete absence of the ϵCTD. Isolated membranes with ϵΔ5 generated proton-motive force by respiration as effectively as with wild-type ϵ but showed a nearly 3-fold decrease in ATP synthesis rate. In contrast, the ϵΔ5 truncation did not change the intrinsic rate of ATP hydrolysis with membranes. Further, the ϵΔ5 subunit retained high affinity for isolated F1 but reduced the maximal inhibition of F1-ATPase by ϵ from >90% to ∼20%. The results suggest that the ϵCTD has distinct regulatory interactions with F1 when rotary catalysis operates in opposite directions for the hydrolysis or synthesis of ATP.

with isolated ⑀ from two species (16,17) and in one bacterial F 1 structure (18). In mitochondrial F 1 structures, the ⑀ homolog appears to be locked in the ⑀ C conformation by a unique mitochondrial subunit (19). Thus far, an extended state of the ⑀CTD has only been observed within the enzyme in a crystal structure of Escherichia coli F 1 (13), in which the latter half of the ⑀CTD inserts into the central cavity of F 1 and has extensive contacts with other subunits. The second ␣-helix of the ⑀ X state (⑀112-125) contacts five other subunits, with apparent H-bonds and/or salt bridges to ␣1, ␣2, ␤1, and ␥ subunits. The terminal segment of ⑀CTD (⑀126 -138) was called the ⑀-hook in the ⑀ X state, because it bends around ␥ and "hooks" the CTD of another catalytic subunit, ␤3 (Fig. 1B). The ⑀-hook buries extensive surface area within F 1 but has minimal specific interactions, with perhaps one H-bond (⑀Ile 131 amide to ␤3-Asp 372 side chain). The final C-terminal segment of ⑀ varies significantly in sequence and length between diverse bacterial species (9). High resolution structures have been determined for bacterial ⑀ from two other species, Bacillus PS3 (17) and Caldalkalibacillus thermarum TA2.A1 (18); both superimpose well with the ⑀ C conformation of E. coli ⑀ but are shorter at the C terminus by 4 and 3 residues, respectively. Despite this, the activity of each ATP synthase can be strongly inhibited by the shorter ⑀CTD (10,20,21). Thus, we postulated that the final segment of the ⑀-hook might be dispensable or even destabilizing for inhibition by the ⑀CTD. Inhibition by ⑀ is not essential for respiratory growth of E. coli, because significant growth on a nonfermentable carbon source can be achieved with the entire ⑀CTD genetically truncated (22,23). Interestingly, however, an early mutagenic study with E. coli found that combined deletion of 4 C-terminal and 15 N-terminal residues disrupted growth by oxidative phosphorylation, whereas the N-terminal truncation alone did not (24). In this study, we genetically deleted just the final five amino acids of ⑀CTD (⑀134 -138) to generate an E. coli ⑀⌬5 mutant. Growth on a nonfermentable carbon source was reduced ϳ60% by ⑀⌬5, whereas complete deletion of the ⑀CTD (⑀88stop) reduced growth by ϳ20%. With isolated membranes, ⑀⌬5 reduced the ATP synthesis rate by Ͼ2.7-fold, but did not alter ⑀ inhibition of ATPase activity. Thus, the ⑀⌬5 truncation has distinct effects on ATP synthesis versus hydrolysis. Since a new class of effective antibacterial agents has been found to target the ATP synthase (25), our results show that regulation by the ⑀CTD provides a bacteriaspecific target for future development of antibacterials against the ATP synthase.

Experimental Procedures
Plasmids and Mutagenesis-Plasmid p3DC (26), which encodes subunits ␤ and ⑀, was used as template to truncate the terminal five amino acids of ⑀CTD via site-directed mutagenesis (27). Primer 5Ј-CAGCTGCGCGTTATCGAGTTGTA-ATAAAAAGCGATGTAACACCGGC-3Ј (mutations underlined) was used to replace codons for ⑀Thr 134 /Lys 135 with two ochre stop codons (bold type). DNA sequencing (Upstate Medical University core facility) was used to confirm that only the desired mutations were created. DNA encoding ⑀ with the stop codons (⑀⌬5) was extracted in a NdeI-XbaI restriction fragment and used to replace the corresponding fragment of pAU1 (26) to obtain pAU1⑀⌬5. The NdeI-XbaI fragment was also used to move ⑀⌬5 into pBKH8 (15), creating pBKH8⑀⌬5. The ⑀88stop truncation was also moved in a NdeI-XbaI fragment from pBKH9 (15) into pAU1. The ␤M209L mutation was originally a gift from A. E. Senior (28) and was transferred to pAU1 in a SacI-EagI fragment.
Phenotypic Assay for Respiratory Growth-F 0 F 1 , either WT or with the mutants noted ( Fig. 2 and Table 1), was expressed from the atp operon on low copy plasmid pAU1. For most phenotypic growth tests, pAU1 constructs were transformed into strain LE392⌬(atpI-C) (29). Individual bacterial colonies were inoculated into 10 ml of Luria Bertani broth (LB; Lennox type, Sigma Aldrich) ϩ ampicillin (0.1 mg/ml) and grown overnight at 37°C, with shaking at 200 rpm in a 125-ml Erlenmeyer flask. Cells were then diluted into 10 ml of fresh LB ϩ ampicillin to obtain an A 600 of 0.1. When growth reached A 600 ϳ0.8, cells were diluted 100-fold into defined minimal salts medium (30) including 1 mM MgSO 4 , 0.1% (v/v) trace elements (30), 0.06% cas-amino acids (BD Difco), 6 M thiamine, 0.1 mg/ml ampicillin, 50 g/ml methionine, and 30 mM succinate as the nonfermentable carbon source. Growth at 37°C was measured with 0.4 ml of culture per well in a 48-well transparent microtiter plate with lid (catalog no. 677102; Greiner Bio-one), with triplicate samples for each distinct pAU1 construct. Growth was monitored every 15 min by A 600 , using a plate reader (Biotek Synergy HT or TECAN Infinite F200). Plates were shaken at 88.6 rpm (TECAN) or at "slow" setting (Biotek) for 20 -30 h. Some assays were repeated with the same plasmids in a distinct atp-deletion strain, DK8 (31), so the defined growth medium included 0.3 mM isoleucine and valine and omitted methionine. The DK8 strains showed less stringent differences in growth between WT and negative controls in initial tests. For more consistent performance, defined medium for DK8 had reduced cas-amino acids (0.03%) and succinate (6 mM). Also, to reduce carryover of LB, DK8 cells from starter cultures were sedimented and resuspended with defined medium before final dilution into defined medium for the assay.
Isolation of Inverted Membrane Vesicles-Strain LE392⌬ (atpI-C) containing pAU1 (WT or ⑀ mutants) was inoculated from individual colonies into 10 ml of LB ϩ ampicillin (0.1 mg/ml) and grown overnight in a 125-ml Erlenmeyer flask at 37°C with shaking (200 rpm). Cells were then diluted into 2 liters of defined minimal salts medium (30) to obtain an A 600 of ϳ0.05. Additions were as noted earlier except that 30 mM glucose and 1% glycerol were the carbon sources. The cells were grown at 37°C with constant aeration and were harvested during logarithmic growth phase. Inverted membrane vesicles (membranes) were prepared as described before (32), but with a final exchange into 50 mM MOPS-Tris, 10% (v/v) glycerol, 5 mM magnesium acetate, pH 7.5.
Detection of F 0 F 1 Content in Membrane Vesicles by Immunoblotting-Membrane samples (at least two amounts each) and known amounts of purified F 1 were subjected to SDS-PAGE (35) on precast 4 -20% gradient polyacrylamide gels (Bio-Rad) at 200 V for 33 min. Proteins were then transferred to a polyvinylidene difluoride membrane (Invitrogen) in a Bio-Rad Mini Trans-Blot cell at 200 mA for 1 h using 1ϫ electrophoresis buffer (35) ϩ 10% (v/v) methanol. The blot was blocked with TBST (10 mM Tris-Cl, 150 mM NaCl, pH 8, 0.05% Tween 20) ϩ 5% (w/v) nonfat dried milk and then washed three times for 5 min each with TBST (0.3 M NaCl total). The blot was then incubated for 1 h with the primary rabbit anti-␤ antibody (1:200) in TBST ϩBSA (10 mg/ml); this anti-␤ antibody (antibody AS05-85; Agrisera) was previously tested for this purpose with E. coli membranes (36). The blot was washed three times as above, followed by 1 h with a fluorescent goat anti-rabbit secondary antibody (antibody 35553; Thermo Scientific), 1:1000 dilution in TBST ϩBSA. After three final washes as above, the blot was air-dried, and fluorescence was detected on a Typhoon 9410 imager (GE Healthcare Life Sciences) with a 532-nm laser and 526-nm short pass filter. Signals for ␤ from known amounts of F 1 provided a linear response range that was used to quantify the amount of ␤ in different membrane samples.
ATP Hydrolysis-ATP hydrolysis rates were measured at 30°C with a coupled enzymes assay (37) as described (15). Assays with membranes contained 5 mM magnesium acetate and 2 mM ATP, 5 mM KCN to inhibit NADH oxidation by the electron transport chain and 5 M FCCP as uncoupler to prevent generation of PMF. Assays to measure ⑀ inhibition of isolated F 1 (-␦⑀) included preincubation of F 1 (-␦⑀) with ⑀, 2 mM ATP, and 0.1 mM EDTA, and the values for K I and maximal inhibition with Bap-⑀⌬5 were determined as before for WT-⑀ and ⑀88stop (15). Assays of NADH oxidation by the electron transport chain were done with the same conditions but without ATP or coupling enzymes, and ϮKCN; oxidation rates were Ͼ88% inhibited by KCN, confirming that most NADH oxidation was through the electron transport chain in all membranes.
ATP Synthesis-Assays of ATP synthesis by membranes were modified from (38). The membranes were diluted to 0.105 mg/ml final in 1910 l of synthesis reaction buffer (50 mM MOPS-Tris, pH 7.5 ϩ 10 mM magnesium acetate) in a 1 ϫ 1-cm cuvette. Aeration was achieved throughout the assay by stirring with a cylindrical magnetic stirrer with cross-cut channels. Reactions were done at ambient temperature (ϳ22°C). Membranes were allowed to equilibrate for 2 min after dilution into the cuvette. NADH (50 l of 0.1 M stock) was added to 2.5 mM final concentration, and after 1 min to establish PMF, ATP synthesis was started by adding 40 l of ADP/P i mixture to obtain 1 mM ADP and 3 mM P i final (total assay volume, 2.0 ml). Over 4 min, 100 l of reaction was withdrawn at 1-min intervals and added to 400 l of ice-cold stop solution (1% TCA, 2 mM EDTA) with vortexing, and the quenched samples were kept on ice. For each membrane sample tested, a control time course was done with 10 M FCCP present to prevent PMF formation; this corrected for (i) minimal ATP synthesis from ADP by contaminating pyruvate kinase and (ii) residual ATP in the assay (primarily from the ADP stock). For each quenched sample in duplicate, 10 l was added to 390 l of ice-cold luciferase assay buffer (0.1 M Tris acetate, 2 mM EDTA, pH 7.5). Samples of ATP standards were treated with stop solution and diluted as above to provide a linear response over 0.25-12 pmol in the final measurement. Samples could be frozen at this point, if needed. For each neutralized sample, 100 l was transferred to a well of a white, opaque 96-well microtiter plate (catalog no. 236108; Nunc), which was then equilibrated to ambient temperature. The plate was placed in a Synergy HT microplate reader (Biotek) equipped with autoinjectors. For each sequential sample well, 50 l of luciferase reagent (ATP bioluminescence assay kit CLS II; Roche Diagnostics) was injected, and luminescence was measured for 10 s (top path, no emission filter; integration, 1 s; gain, 135). The rates of synthesis for control samples (ϩFCCP) were minimal and were subtracted from rates with energized membranes to obtain ATP synthesis rates caused by the ATP synthase. All membranes assayed showed linear rates of ATP syn-thesis over the assay period (all linear fits used had R 2 values of Ͼ0.98).
Proton Pumping-Proton pumping activity of membranes was measured by monitoring fluorescence quenching of ACMA, which reflects ⌬pH (39). Membranes were diluted to 0.1 mg/ml in assay buffer (20 mM MOPS-Tris, pH 7.5, 50 mM KCl, 5 mM magnesium acetate) ϩ 1 M ACMA and equilibrated for ϳ9 min, and the assay was started by adding NADH or ATP to drive proton pumping. Total assay volume was 2 ml in a 1 ϫ 1-cm fluorescence cuvette, and aeration was maintained by stirring, as in the ATP synthesis assays. Assays were done at 30°C on a Fluoromax-4 or Fluorolog-3 (Horiba Scientific) with excitation/emission wavelengths (nm) of 430/560, excitation/ emission slits of 5/4 nm, gratings set at 1200, integration time of 0.5 s, and interval time of 7.5 s. For WT and each ⑀ mutant, at least two separate preparations of membranes from different cell growths were tested.

Results
Effects of ⑀CTD Truncations on Aerobic Growth-To observe whether ⑀CTD truncations affect in vivo function of the ATP synthase, bacteria expressing WT or mutant forms of F 0 F 1 were grown with a nonfermentable carbon source, succinate, so growth required oxidative phosphorylation (41). As shown in Fig. 2 and Table 1, growth on succinate was negligible for cells expressing F 0 F 1 with a control mutation, ␤M209L, which allows assembly of normal levels of ATP synthase on the membrane but renders it essentially inactive (28,42). Deletion of the entire ⑀CTD (⑀88stop) reduced respiratory growth yield only ϳ20%, with little effect on growth rate. This is consistent with another group's study in which ⑀88stop allowed respiratory growth on acetate and caused a minimal decrease in growth yield on limiting glucose (22). In contrast, deleting only five C-terminal amino acids from ⑀ (⑀⌬5) reduced growth yield and growth rate (Table 1) by ϳ60%. Growth assays were repeated with the same plasmids expressed in a distinct ⌬atp-operon host strain, DK8. Initially, the DK8 strains showed less robust differences in phenotypic growth, possibly because of greater C 4 -dicarboxylate transporter activity (43). For subsequent assays, succinate concentration was reduced 5-fold, and results were similar to the effects of mutations seen in Table 1 and Fig.  2 (DK8 growth yields relative to WT: ⑀88stop, 95%; ⑀⌬5, 46%; ␤M209L, 9%). The greater phenotypic defect of ⑀⌬5 was not simply due to poor expression or assembly of F 0 F 1 because ⑀⌬5 membranes showed higher F 0 F 1 content than for ⑀88stop (Table 1). Thus, the entire ⑀CTD can be removed with minimal effects, but the small ⑀⌬5 truncation perturbs the regulatory interactions of ⑀CTD so that the capacity for in vivo oxidative phosphorylation is significantly degraded.
Effects of ⑀CTD Truncations on in Vitro Functions of Membrane-bound ATP Synthase-To further examine why ⑀⌬5 is more deleterious than ⑀88stop in vivo, membranes were isolated and tested for effects of ⑀CTD truncations on activities of F 0 F 1 in vitro. Membrane ATP hydrolysis was measured with excess uncoupler present in all conditions, to ensure that activity was not inhibited by "back pressure" from PMF. Nearly all ATPase activity of WT and mutant membranes was likely due to F 0 F 1 because sodium azide, a catalytic site inhibitor, reduced ATPase Ն98%. In direct comparison, ⑀⌬5 and ⑀88stop membranes showed 63 and 60% ATPase activity versus WT (Table  2). However, when results were normalized for the F 0 F 1 content in membranes, intrinsic ATPase activity was 2.6-fold higher in the complete absence of the ⑀CTD (⑀88stop). This is consistent with prior demonstrations that WT membrane ATPase activity doubled when ⑀ inhibition was disrupted (12,44). In contrast, the ⑀⌬5 truncation did not significantly alter the intrinsic ATPase activity of F 0 F 1 in membranes. This is also supported by the effects of LDAO, a detergent that is known to activate ATPase of E. coli F 0 F 1 and F 1 mostly by disrupting ⑀ inhibition (45,46). LDAO activated ATPase activity to the same extent for WT and ⑀⌬5 membranes but less for ⑀88stop membranes, which were already activated by the absence of the ⑀CTD (Table  1). These results suggest that the loss of ⑀'s 5 C-terminal residues does not significantly alter the inherent energetic balance between active and ⑀-inhibited forms of F 0 F 1 in membranes.
As suggested previously for ⑀88stop (22), it is possible that partial functional uncoupling of F 1 from F 0 contributes to the in vivo phenotypic defect of ⑀⌬5. One result of this could be that some ATPase activity is not thermodynamically linked to PMF; under respiratory conditions that would drive net ATP synthesis through well coupled F 0 F 1 , unregulated ATP hydrolysis LE392(⌬atpI-C) cells expressing F 0 F 1 from pAU1were grown on defined medium with succinate as the sole carbon source. The ⑀ subunit expressed was WT (•), ⑀88stop (f), or ⑀⌬5 (E); a negative control expressed WT-⑀ but catalytically defective ␤M209L subunit (OE). Each data set is averaged from triplicate cultures grown in parallel; analyses for multiple experiments are in Table 1.

TABLE 1 Effect of ⑀CTD truncations on aerobic growth on succinate
The values with ranges are means ϮS.E. with the number of independent experiments for each assay noted in parentheses.

Strain
Growth yield Growth rate a Amounts of F 0 F 1 in membrane were quantified by anti-␤ antibody (Experimental Procedures). The values were normalized relative to WT. b NS, not significant. c ND, not determined, but previously measured as equivalent to WT (28).
would create a futile cycle that reduces the efficiency of cellular energy conversion. To test for this, ATP hydrolysis was measured after treating membranes with DCCD, a covalent modifier of the c-ring that blocks proton transport through F 0 . For well coupled F 0 F 1 complexes, blocking proton transport with DCCD also inhibits ATP hydrolysis (47). As shown in Table 2, the ⑀88stop truncation did not significantly alter the sensitivity of membrane ATPase to DCCD, although the original study by Cipriano and Dunn (22) showed slightly reduced DCCD inhibition for ⑀88stop membranes. The ⑀⌬5 truncation resulted in a small and insignificant decrease in inhibition by DCCD ( Table 2). Membranes were also tested for possible effects of ⑀CTD truncations on ATP synthesis. As shown in Table 2, ⑀⌬5 membranes showed a Ͼ2.7-fold lower rate for ATP synthesis. This was not due to reduced PMF, because NADH-driven respiration generated similar ⌬pH gradients for WT and ⑀⌬5 membranes ( Fig. 3 and Table 2). The lower ATP synthesis rate was also not due to the ϳ50% lower F 0 F 1 content in ⑀⌬5 membranes because ⑀88stop membranes had even lower F 0 F 1 content (Table 1) but had ATP synthesis rates similar to that of WT pAU1 membranes (Table 2). This is consistent with prior studies showing that F 0 F 1 content of haploid membranes exceeds that necessary for ATP synthesis rates in vivo (48) and in vitro (49). In fact, because ⑀⌬5 membranes had more F 0 F 1 than ⑀88stop or WT haploid membranes, their 2.7-fold lower synthesis rate probably reflects an even greater intrinsic inhibition of ATP synthesis by the ⑀⌬5 subunit. Control assays were also included to test whether reduced ATP synthesis by ⑀⌬5 membranes was due in part to any uncoupled ATPase activity. The ATP synthesis rates of ⑀⌬5 and WT membranes were not significantly altered by the presence of 10 M AMPPNP, which inhibits ATPase but not ATP synthesis (50). Together, these results indicate that the ⑀⌬5 truncation directly increases ⑀ inhibition of ATP synthesis by F 0 F 1 .
Another test for possible coupling defects between F 1 and F 0 is to monitor the kinetics of proton pumping by isolated, inverted membranes. Altered coupling between F 1 and F 0 might allow uncontrolled, passive flux of protons, which would decrease the capacity to generate PMF by respiration or by ATPase-driven proton pumping (51). As shown in Table 2 and Fig. 3A for NADH-driven proton pumping, ⑀⌬5 membranes generated similar or better PMF than did WT, but ⑀88stop membranes generated partially reduced PMF. To test for F 0 -specific proton leaks, membranes were treated with DCCD before addition of NADH. DCCD had a similar effect on NADH-driven proton pumping for WT and ⑀⌬5 membranes (Fig. 3), so ⑀⌬5 did not cause any increased leak through F 0 . Fig.  3 also shows that the lower PMF achieved with ⑀88stop membranes (Table 2) was not due to greater proton leaks, because (i) upon depletion of NADH, the gradient collapsed with a time course similar to WT and ⑀⌬5 membranes, and (ii) DCCD had a minimal effect on proton pumping by ⑀88stop membranes. The reduced PMF generated with ⑀88stop membranes was not FIGURE 3. Respiratory generation of proton motive force by membranes. Proton pumping was measured by quenching of fluorescence of the dye ACMA (see "Experimental Procedures"). A, respiration was initiated by addition of NADH to 0.5 mM. Dashed lines represent proton pumping by untreated ⑀88stop (blue), ⑀⌬5 (red), and WT (black) membranes. Solid lines represent proton pumping by the same membranes after treatment with DCCD to block possible proton leakage through F 0 . Once the NADH was depleted, the relaxation of ACMA fluorescence quenching reflects all intrinsic membrane transport processes that contributed to collapse of the ⌬pH. B, ATP was added to 1 mM to initiate proton pumping by F 0 F 1 , and after ϳ150 s, FCCP was added to 5 M to collapse the PMF. Table 2 summarizes statistical results from multiple experiments for both NADH-and ATP-driven pumping. due to gross changes in the capacity of the electron transport chain, because two preparations of ⑀88stop membranes showed NADH oxidation rates at least as fast as with WT membranes (0.8 and 0.7 mol/min/mg protein, respectively). 4 In any case, the current proton pumping results show that the ⑀CTD truncations do not cause any increased proton leak in the membrane preparations. Proton pumping was also tested when PMF was generated by ATP hydrolysis via F 0 F 1 , and results were similar for WT, ⑀⌬5, and ⑀88stop membranes ( Table 2 and Fig.  3B). Cipriano and Dunn (22) noted a more significant defect in ATPase-driven proton pumping for ⑀88stop; there is no apparent reason for this discrepancy, although different host strains of E. coli were used. Effects of ⑀⌬5 Truncation on Interactions of ⑀CTD with Isolated F 1 -In vitro, the catalytic complex of ATP synthases can be released from the membrane as a soluble F 1 -ATPase. Isolated bacterial F 1 is strongly inhibited by ⑀ but, upon dilution, ⑀ can dissociate, activating the enzyme (2,9). Previously, BLI kinetic assays of protein-protein interactions showed that the conformation of the ⑀CTD controls dissociation of ⑀ from E. coli F 1 (15), and ⑀ probably does not dissociate at all when it adopts the inhibitory extended conformation (⑀ X ), with part of the ⑀CTD buried in the central cavity of F 1 (13,15). Here, BLI was used to test whether the ⑀⌬5 truncation changes the interactions of ⑀CTD with F 1 . With F 1 bound to immobilized ⑀⌬5 in buffer alone, 87% of F 1 /⑀⌬5 complexes dissociated very slowly in buffer only (Fig. 4, trace 1, and Table 3). This is similar to the behavior of F 1 /WT-⑀ (15), but the difference in their slow dissociation rates is near the limit of sensitivity for BLI in these conditions. Although WT-⑀ on F 1 is strongly biased toward the tightly bound inhibitory state, transition in and out of that state is dynamic (14,15); addition of excess ATP in the BLI dissociation step (with EDTA present to prevent hydrolysis) rapidly shifts F 1 /WT-⑀ complexes to dissociate ϳ80-fold faster, as if the ⑀CTD were completely absent (Fig. 4, traces 3 and 6). ATP/ EDTA in the dissociation step produced faster, essentially monophasic dissociation of F 1 /⑀⌬5 complexes (Fig. 4, trace 2) with no noticeable lag, but with a rate ϳ6-fold slower than for F 1 /WT-⑀ ( Table 3). As seen before (15), when F 1 /WT-⑀ was bound in the presence of ATP/EDTA, subsequent exposure to Mg 2ϩ allowed hydrolysis and rapid switching to the ⑀-inhibited state at the catalytic dwell, and post-hydrolysis conditions (MgADP/Pi) in the dissociation step stabilized ⑀ in the tightly bound form (Fig. 4, trace 4). F 1 /⑀⌬5 complexes also showed rapid reversal to a tightly bound state on switching from ATP/ EDTA during F 1 /⑀⌬5 association to MgADP/P i in the dissociation phase (Fig. 4, trace 5). Overall, these results indicate that although the ⑀CTD can still undergo dynamic transitions between different conformations, the absence of five terminal residues from the ⑀CTD significantly stabilizes a tightly bound state of ⑀⌬5 on F 1 relative to the dissociable state.
The ATPase activity of isolated E. coli F 1 is inhibited Ͼ90% by bound WT-⑀ (9,15). Because ⑀⌬5 showed a bias toward tight binding, we investigated whether this correlates with greater inhibition. An N-terminal Bap tag on ⑀ does not affect its inhibition of isolated F 1 (15). As shown in Fig. 5, the K I of 0.7 nM for Bap-⑀⌬5 is similar to the K I for WT-⑀ (0.5 nM) but does not reflect the increased stability of the tightly bound state as indicated by the BLI assays of F 1 /⑀⌬5 binding. Furthermore, maximal inhibition by ⑀⌬5 was only ϳ20%. This is similar to the ϳ24% maximal inhibition by ⑀88stop under the same conditions, although the K I for ⑀88stop is nearly 20-fold weaker because of the complete absence of the ⑀CTD (15). This surprising finding indicates that the shorter ⑀CTD of ⑀⌬5 still con-  Table 3. Note that data for trace 6 are reproduced from Ref. 15. Fig. 4 Dissociation parameters are given for the nonlinear regression lines shown in Fig. 4. For the top four sample rows, all parameter values have standard errors Ͻ2%, and fits have R 2 values Ͼ 0.996. For the bottom two rows, insufficient dissociation occurred for reliable fitting; 10 Ϫ6 s Ϫ1 is the slowest rate that can be fit reliably under the experimental conditions.

Conditions for association
Conditions for dissociation ⑀ (Fig. 4  tributes to tight binding to F 1 , but that the five terminal residues of ⑀ are critical for strong inhibition of F 1 -ATPase activity. However, ⑀⌬5 does inhibit ATP synthesis and hydrolysis by F 0 F 1 on membranes ( Table 2), suggesting that F 0 -F 1 interactions are important for the ⑀⌬5 subunit to achieve inhibition of ATP hydrolysis.

Discussion
Earlier studies with F 0 F 1 of E. coli (52) and B. PS3 (53) suggested that the extended ⑀CTD inhibits ATPase but not ATP synthesis, based on disulfide cross-links to trap the ⑀CTD in extended states. However, it is not clear that those ␥-⑀ crosslinks occurred in native conformations of the enzyme. For example, the E. coli cross-linking sites (␥99, ⑀118) were based on a structure of an isolated complex of truncated ␥ with ⑀ (54), but are 28 Å apart (C␣-C␣) in the structure determined for ⑀-inhibited F 1 (13). Subsequent studies showed that deleting the ⑀CTD increased the ATP synthesis rate 3-fold for B. PS3 F 0 F 1 (10) and activated ATP synthesis more than it activated ATPase for E. coli F 0 F 1, (11). Thus, it is clear that the ⑀CTD can inhibit both ATP hydrolytic and synthetic directions of rotary catalysis in bacterial ATP synthases. Here, in vitro results for F 0 F 1 containing the ⑀⌬5 subunit further show that altering interactions of the ⑀CTD with F 1 can preferentially increase inhibition of ATP synthesis (Table 2).
Prior studies with membrane-bound E. coli F 0 F 1 (12,44) indicate that, on average, ϳ50% of F 0 F 1 complexes are in an ⑀-inhibited state. Current results with ⑀88stop membranes support this, because the intrinsic ATPase activity is 2.6-fold greater in the absence of the ⑀CTD (relative ATP hydrolysis in Table 2). ATP synthesis results with ⑀88stop membranes also likely reflect a greater fraction of active F 0 F 1 complexes without the ⑀CTD: compared with WT, ⑀88stop membranes showed about the same synthesis rates (Table 2), although they contained ϳ4-fold less F 0 F 1 (Table 1) and generated lower PMF by NADH oxidation (Fig. 3). In the presence of MgADP/Pi, PMF activates F 0 F 1 in E. coli membranes (55), probably because of release from the ⑀-inhibited state (13,15). Thus, without the inhibitory ⑀CTD, ⑀88stop membranes in this study likely contained a higher fraction of active F 0 F 1 complexes and so achieved high ATP synthesis rates even with a reduced PMF. In contrast, ⑀⌬5 membranes showed ATP synthesis rates nearly 3-fold less than those for WT or ⑀88stop (Table 2), even though F 0 F 1 content was ϳ2-fold greater in ⑀⌬5 than in ⑀88stop membranes ( Table 1). The low synthesis rate was not due to uncoupling, because ⑀⌬5 membranes generated a greater NADHdriven pH gradient than did WT and showed no greater F 0 -specific proton leak (Fig. 3). On the other hand, ⑀⌬5 membranes showed intrinsic ATPase rates, activation by LDAO, and ATPase-driven proton pumping that were very similar to the values obtained with WT membranes (Table 2). Thus, the ⑀⌬5 truncation specifically increased ⑀ inhibition of ATP synthesis without increasing inhibition of ATP hydrolysis or uncoupling ATPase from proton pumping. Further, without interactions with F 0 , ⑀⌬5 subunit bound isolated F 1 with high affinity (Fig. 4) but inhibited F 1 -ATPase minimally, as seen with the ⑀NTD alone (Fig. 5). This suggests that contacts of the ⑀-hook with the CTD of subunit ␤3 (Fig. 1B) are important for inhibition of F 1 -ATPase.
Differential effects on ATP synthesis versus hydrolysis modes have been noted for other inhibitors (reviewed in Ref. 50). For example, azide or AMPPNP inhibit ATP hydrolysis but not ATP synthesis, whereas some fluorescent analogs of ADP inhibit ATP synthesis more than hydrolysis. However, what mechanisms might explain how the ⑀⌬5 truncation selectively increases inhibition of ATP synthesis? Thus far, only one ⑀-inhibited state has been observed structurally (13). If one assumes that the observed ⑀ X state is responsible for inhibition of both synthesis and hydrolysis, then the ⑀⌬5 truncation could preferentially increase the energy barrier for activation from the ⑀ X state during rotation in the direction of ATP synthesis. Control of ⑀ conformation by rotational direction has been proposed before (56). Such directional asymmetry has been demonstrated for an ADP-inhibited state that pauses the enzyme at a specific rotary angle: in single-molecule studies with B. PS3 F 1 , magnetically driven torque reactivated the enzyme after 40°of forced rotation in the direction of hydrolysis but not after 120°i n the direction of ATP synthesis (57). For regulation by the ⑀CTD, an alternative is the bidirectional ratchet model, in which the ⑀CTD has distinct regulatory interactions with F 1 during opposite directions of rotary catalysis (6,58). With this model, the ⑀⌬5 truncation could preferentially enhance the stability of the inhibitory state that forms primarily during ATP synthesis mode. Our present results on interactions of the ⑀⌬5 subunit with isolated F 1 seem more consistent with this second model: kinetic assays for F 1 /⑀ interactions (Fig. 4) indicate that the tightly bound state of F 1 /⑀⌬5 reverses to a dissociable state more slowly than for F 1 /WT-⑀, but the tightly bound state of ⑀⌬5 causes minimal inhibition of F 1 -ATPase activity (Fig. 5).
The existence of a distinct F 1 /⑀CTD interaction state is also consistent with our recent collaborations to study conformational changes of the ⑀CTD by single-molecule FRET with probes on ␥ and on helix-1 of the ⑀CTD. Initial studies with isolated F 1 (59,60) showed bimodal distribution of FRET efficiencies that correlate with the ⑀ C and ⑀ X states, and nucleotides shifted the balance between the two FRET states in agreement . The specific activity of F 1 (-␦⑀) alone was 60.9 mol/min/mg. For each data set, the curve shown is from a nonlinear regression fit to a quadratic equation described in Ref. 15. For ⑀⌬5, regression indicated maximal inhibition of F 1 (-␦⑀) ϭ 20% (95% confidence interval, 19 -22%), and K I ϭ 0.68 nM (95% confidence interval, 0.33-1.02 nM); R 2 ϭ 0.972 (GraphPad Prism). In parallel with ⑀⌬5 assays, control assays confirmed that 100 nM WT-⑀ inhibited F 1 (-␦⑀) Ͼ85%.
with our bulk assays of F 1 /⑀ interactions (15). Subsequent studies with FRET-labeled F 0 F 1 -liposomes revealed a trimodal distribution of FRET efficiencies in the presence of MgATP that cannot be explained by the two known orientations of helix-1 of ⑀ (13, 16). Thus, it seems likely that the ⑀CTD can form distinct interactions with F 1 during opposite directions of rotary catalysis and that ⑀⌬5 preferentially stabilizes or promotes formation of the tightly bound state that inhibits the ATP synthesis direction.
For the direction of ATP hydrolysis, single-molecule rotation assays (14,61,62) and our recent enzymological study (15) show that inhibition by the ⑀CTD initiates at the catalytic dwell angle after the hydrolytic step. In contrast, the only available structure of ⑀-inhibited F 1 appears to be paused after further 40°r otation to an angle near the next dwell for ATP binding (13). Some rotational data could suggest dynamic oscillation between these two angles during a long inhibitory pause (Ref. 14 and Fig. 3A), so perhaps these represent two positions of the ⑀CTD that have distinct regulatory effects during opposite directions of rotary catalysis. In detail, ⑀⌬5 might also cause some type of mechanical slip between F 1 and F 0 only during rotation in the ATP synthesis direction, but further tests are needed to explore these possibilities.
Correlation of the ⑀⌬5 Phenotypic Growth Defect with Inhibited ATP Synthesis-Reduced ATP synthesis rate was the only significant functional defect identified in vitro with ⑀⌬5 membranes, and this is likely the primary reason that cells expressing ⑀⌬5 grew poorly by oxidative phosphorylation. With the entire ⑀CTD absent, cells showed better phenotypic growth, and rates of in vitro ATP synthesis were normal, even though ⑀88stop membranes contained less F 0 F 1 . An earlier study reported that deletion of 10 C-terminal residues from E. coli ⑀ also allowed normal growth yield on succinate (23), indicating that in vivo ATP synthesis is more effective than with ⑀⌬5. Together, these results suggest that residues between ⑀128 -133 are important for inhibition of ATP synthesis.
It should be noted that the pAU1 construct used here expresses the entire atp operon, and ⑀⌬5 membranes contained ϳ4-fold greater F 0 F 1 than in haploid membranes. Even haploid expression of E. coli F 0 F 1 is not rate-limiting for ATP synthesis in vivo (48), so the low rate measured for in vitro ATP synthesis by ⑀⌬5 membranes probably represents a greater intrinsic inhibition by ⑀⌬5. Thus, ⑀⌬5 should cause an even larger defect in phenotypic growth in a strain expressing lower, haploid levels of F 0 F 1 , and we are currently reengineering our expression system to test for this.
Summary-Overall, our results are consistent with the idea that the ⑀CTD may be fine-tuned in different bacterial species to regulate ATP synthesis and hydrolysis functions according to the distinct metabolic/environmental demands of each species (2,9). We showed that a minor truncation of the ⑀-hook selectively increased inhibition of ATP synthesis and reduced the capacity for cell growth on a nonfermentable carbon source. ATP synthases from two Gram-positive species appear to be missing the last 3-4 residues of the ⑀ hook (17, 18) but still show strong inhibition of ATPase by ⑀ (20,21). This could suggest that inhibitory behavior in different species involves co-evolution of one or more subunits that interact with the ⑀CTD. This correlates with results of recent computational studies of coevolution in protein complexes, which used interactions of ␥ and ⑀ as a test case (63,64). There are also indications that ⑀ inhibition also occurs in the enzyme of several species of Mycobacterium (65,66), and the CTD of most mycobacterial ⑀ subunits is ϳ17 residues shorter than that of E. coli, although different possible alignments make it uncertain how much of the hook and/or helix-2 are absent (67,68). Mycobacterial ATP synthase is the target of a new class of antibiotics, the diarylquinolines, and the lead drug, bedaquiline, has been approved for treatment of multidrug-resistant tuberculosis (25,69). Modified diarylquinolines have been developed to attack other Gram-positive pathogens including Staphylococcus aureus but so far, these show significant inhibition of mitochondrial ATP synthase (70). Bacterial ATP synthase function is also essential or important for the viability or virulence of Gram-negative pathogens (71)(72)(73)(74). Thus, it will be important to explore how ⑀ inhibits ATP synthases in a range of bacterial pathogens. Results of the current study support the concept that ⑀ inhibition can provide a bacteria-specific means to target the ATP synthase for development of future antibiotics.