Rotor/Stator Interactions of the ϵ Subunit in Escherichia coli ATP Synthase and Implications for Enzyme Regulation*

The H+-translocating F0F1-ATP synthase of Escherichia coli functions as a rotary motor, coupling the transmembrane movement of protons through F0 to the synthesis of ATP by F1. Although the ϵ subunit appears to be tightly associated with the γ subunit in the central stalk region of the rotor assembly, several studies suggest that the C-terminal domain of ϵ can undergo significant conformational change as part of a regulatory process. Here we use disulfide cross-linking of substituted cysteines on functionally coupled ATP synthase to characterize interactions of ϵ with an F0 component of the rotor (subunit c) and with an F1 component of the stator (subunit β). Oxidation of the engineered F0F1 causes formation of two disulfide bonds, βD380C-S108C ϵ and ϵE31C-cQ42C, to give a β-ϵ-c cross-linked product in high yield. The results demonstrate the ability of ϵ to span the central stalk region from the surface of the membrane (ϵ-c) to the bottom of F1 (β-ϵ) and suggest that the conformation detected here is distinct from both the “closed” state seen with isolated ϵ (Uhlin, U., Cox, G. B., and Guss, J. M. (1997) Structure 5, 1219–1230) and the “open” state seen in a complex with a truncated form of the γ subunit (Rodgers, A. J., and Wilce, M. C. (2000) Nat. Struct. Biol. 7, 1051–1054). The kinetics of β-ϵ and ϵ-c cross-linking were studied separately using F0F1 containing one or the other matched cysteine pair. The rate of cross-linking at the ϵ/c (rotor/rotor) interface is not influenced by the type of nucleotide added. In contrast, the rate of β-ϵ cross-linking is fastest under ATP hydrolysis conditions, intermediate with MgADP, and slowest with MgAMP-PNP. This is consistent with a regulatory role for a reversible β/ϵ (stator/rotor) interaction that blocks rotation and inhibits catalysis. Furthermore, the rate of β-ϵ cross-linking is much faster than that indicated by previous studies, allowing for the possibility of a rapid response to regulatory signals.

F 0 F 1 -ATP synthases are found embedded in the membranes of mitochondria, chloroplasts, and bacteria (for reviews see Refs. [1][2][3][4]. During oxidative-and photophosphorylation, synthases couple the transport of protons down an electrochemical gradient to the synthesis of ATP. The F 0 complex is composed of membrane-spanning subunits (ab 2 c 10 in Escherichia coli (5)) that conduct protons across the membrane, whereas F 1 (␣ 3 ␤ 3 ␥␦⑀) is an extrinsic complex with a catalytic site on each of the three ␤ subunits at an ␣/␤ interface (6). It is now widely accepted that a central rotor (␥⑀c 10 ), driven by a current of protons through F 0 , rotates relative to a stator (␣ 3 ␤ 3 ␦ab 2 ) to induce cyclical conformational changes between tight substrate binding and product release at alternating, cooperative catalytic sites on F 1 (Fig. 1A). Catalysis-dependent rotation of ␥ and ⑀ has been shown by various approaches with soluble F 1 (7)(8)(9)(10) and membrane-bound F 0 F 1 (11)(12)(13)(14). Rotation of the ring of c subunits has also been documented (15)(16)(17)(18)(19)(20), as likely required for completion of the proton pathway through F 0 . Recent studies have confirmed that the rotor turns in opposite directions during ATP synthesis and hydrolysis (21,22).
The structure of purified E. coli ⑀ subunit (23,24) is essentially identical to that of its homolog in bovine mitochondrial F 1 (MF 1 ) 1 (25). In each case, an N-terminal ␤-sandwich domain is connected by a loop to a smaller C-terminal domain that consists of a pair of antiparallel ␣-helices, with the second helix packed against one edge of the ␤-sandwich to give a compact "closed" structure ( Fig. 1B, green). Genetic truncations have shown that only the N-terminal domain of ⑀ is essential for coupled function (26 -29), and direct interaction of this domain with the c-ring of F 0 is supported by genetic and disulfide cross-linking studies with EcF 0 F 1 (30 -32), and by the low resolution structure of yeast MF 1 -c 10 (33).
In the bovine MF 1 structure (25), the closed conformation of the homolog of ⑀ appears to be fixed by interactions with a small accessory subunit that is unique to the mitochondrial synthase. The closed conformation of ⑀ can also be fixed in EcF 0 F 1 by disulfide cross-linking of substituted cysteines without loss of ATP synthesis activity (34,35). However, for some bacterial and chloroplast synthases, the C-terminal domain of ⑀ appears to modulate the ATPase activity in a manner that suggests a possible regulatory role (34, 36 -38). Consistent with such a role, the C-terminal domain of ⑀ has been observed to undergo significant conformational change in response to different nucleotides and/or the presence of an electrochemical gradient (for reviews, see Refs. 4, 37, and 39; see also Ref. 40). Cross-linking studies with EcF 0 F 1 have shown that ⑀Ser-108, located in the hairpin loop between the C-terminal helices, can achieve close contact with ␤Asp-380 and ␤Glu-381 in the DEL-SEED motif of a ␤ subunit (13,(41)(42)(43). Molecular dynamics simulations (44) suggest that the ␤-DELSEED motif functions as a tracking guide for rotation of ␥ between ␤ subunits. Thus an interaction between this motif and the C-terminal domain of ⑀ could interfere with subunit rotation, thereby inhibiting catalysis. The closed conformation of ⑀ shown in Fig. 1B does not predict such an interaction, because the ␤-carbons of ⑀Ser-108 and the nearest ␤DELSEED sequence are about 50 Å apart. However, a partially unfolded "open" conformation of the Cterminal domain of ⑀ was observed in a crystal structure of a complex of ⑀ with the central-stalk domain of ␥ (45). In this complex the C-terminal helices of ⑀ are separated from each other as well as from the ␤-sandwich and wrap around a portion of ␥ that, if docked into the MF 1 structure (Fig. 1C), would interact with the bottom of the ␣ 3 ␤ 3 hexamer. A disulfide cross-link (␥L99C-⑀S118C) was used to confirm that the Cterminal domain of ⑀ can adopt this open conformation in EcF 0 F 1 (35). However, this structure is also unable to account for ␤-⑀ cross-links involving ⑀S108C, because the ␤-carbons of ⑀Ser-108 and the nearest ␤Asp-380 are still far too distant (ϳ25 Å; Fig. 1C) to allow formation of a disulfide bond (ϳ5 Å (46)).
Because the available high resolution structures are unable to explain the ␤-⑀ cross-linking results, it has been suggested that ⑀ might release from the c-ring to move closer to the ␤ subunit (25). Results presented here show that this need not occur. Cross-linking experiments using well coupled EcF 0 F 1 provide evidence for a conformation of ⑀ in which ⑀Ser-108 contacts the DELSEED region of a ␤ subunit while, at the same time, ⑀Glu-31 of the N-terminal domain remains in contact with the polar loop of a c-subunit in F 0 . These results establish the ability of ⑀ to span the entire central stalk from the surface of the membrane to the bottom of F 1 . The presence of MgATP, MgADP, or MgAMP-PNP has different effects on the rate of ␤-⑀ cross-linking, consistent with a possible regulatory role for this stator/rotor interaction. Furthermore, the rapid rate of formation of the ␤-⑀ cross-link indicates a more dynamic flexibility of the C-terminal domain of ⑀ than suggested by previous studies. Pyruvate kinase and lactate dehydrogenase were purchased from Roche Molecular Biochemicals. MOPS was supplied by Research Organics. 5,5Ј-Dithiobis(2-nitrobenzoate) (DTNB) and KCN were obtained from Aldrich and dithiothreitol (DTT) from American Bioanalytical (Natick, MA). Tween 20 was purchased from Pierce, and oligonucleotides for site-directed mutagenesis were synthesized by Invitrogen. Pfu DNA polymerase I was from Stratagene, and restriction endonucleases were from New England Biolabs.
The strain/plasmid combination for expressing EcF 0 F 1 containing TABLE I ATP hydrolysis and proton transport by reconstituted membranes ␤D380C/⑀S108C/⑀E31C-F 1 was reconstituted with cQ42C-F 0 in F 1depleted membranes as described. Reconstituted membranes were incubated at 2 mg/ml for 1 h at 22°C in TME buffer with 2 mM DTT (line 1), 2 mM DTT, and 50 M DCCD (line 2), or 50 M DTNB (line 3). ATPase and proton-pumping activities were measured as described under "Experimental Procedures." For NADH-driven proton pumping, the values given for fluorescence quenching represent the percentage decrease in fluorescence intensity obtained after addition of 0.5 mM NADH. For ATP-driven proton translocation, quenching obtained upon addition of 1 mM ATP was measured relative to the fluorescence level reached after the transmembrane gradient of protons was dissipated by addition of 4 M FCCP. Values for NADH-and ATP-driven fluorescence quenching (assayed ϩDTT) were 77% and 83%, respectively, with native membranes expressing only the cQ42C mutation and 82% and 85%, respectively, for wild-type membranes. A, in this model, the a subunit contains two partial channels, each connected to a different side of the membrane. For a proton to traverse the membrane, it must move through one channel to the middle, bind to one of the c-subunits, and then be carried to the other partial channel by rotation of the c-ring. The c subunits are anchored to the ␥ subunit (part of the rotor), whereas the a subunit is anchored through b 2 ␦ to the ␣ 3 ␤ 3 hexamer (part of the stator). Hence, rotation of the c-ring relative to the a subunit in F 0 will drive the rotation of ␥ relative to the ␣ 3 ␤ 3 hexamer in F 1 . B and C, molecular models of F 0 F 1 were generated with VMD (62). ␣-Helices are represented as cylinders, and ␤-strands as ribbons. Subunits ␤ DP (red), ␤ E (blue), and ␥ (gold) are from a bovine MF 1 structure (25) and are docked with the c 10 -ring from the yeast MF 1 -c 10 structure (33) by superimposing the ␥ subunit ␣-carbons of each structure. In C, the open conformation of E. coli ⑀ (green) observed in a complex with a fragment of the ␥ subunit (45) was docked in a similar manner. In panel B the closed conformation of isolated E. coli ⑀ (green) (24) was aligned by superimposing its N-terminal domain with that of ⑀ in panel C (C␣ residues 3-80, root mean square deviation Ͻ 0.7 Å). Distances shown are between the ␤-carbons (yellow spheres, 150% of van der Waal radius) of ⑀Ser-108 and the nearest ␤Asp-380 (E. coli residue numbers). To estimate the distance between ␤Asp-380 and ⑀Ser-108 in the open conformation, the ⑀105-110 sequence, which was not resolved (45), was built with an ab initio loop modeling routine (63) (MODELLER version 6.0). Subunits in the c-ring are silver, except for the one (magenta) that shows the closest approach of cGln-42 to ⑀Glu-31, whose ␤-carbons are also indicated by yellow spheres.
the cQ42C mutation (YZ469/pYZ217 (31)) was kindly provided by Dr. Robert H. Fillingame. Rabbit polyclonal antibody against subunit c and rat monoclonal antibody against subunit ⑀ were provided by Dr. Robert H. Fillingame and Dr. Stanley D. Dunn, respectively. Mouse M2 monoclonal antibody against the FLAG-epitope was purchased from Sigma-Aldrich Co. The 125 I-labeled secondary antibodies against rabbit, rat, and mouse antibodies were obtained from Amersham Biosciences. Fluorescein-conjugated antibody against rat IgG (from goat) was supplied by Pierce. Other reagents and chemicals were the highest grade available.
Preparation of E. coli Membranes and Soluble F 1 -Membranes were isolated and washed, and soluble F 1 was prepared as described (7). Membranes prepared from cQ42C cells were depleted of F 1 during two washes with 10 mM Tris acetate, pH 7.5, 1 mM EDTA, 2 mM DTT, followed by two additional washes with TME buffer (50 mM Tris-HCl, pH 7.5, 5 mM MgCl 2 , 0.1 mM EDTA) containing 2 mM DTT. Final membrane pellets were resuspended in TME buffer containing 2 mM DTT, quickly frozen, and stored at Ϫ75°C. This procedure removed ϳ98% of the ATPase activity from the membranes.
Reconstitution of F 1 with F 1 -depleted Membranes-␤D380C/⑀S108C/ ⑀E31C-F 1 was incubated at 1 mg/ml with 2 mg/ml F 1 -depleted membranes containing cQ42C-F 0 in TME buffer for 30 min at 30°C in the presence of 2 mM DTT (11). Excess F 1 was removed by centrifugation at 40,000 ϫ g for 45 min. The membrane pellet was resuspended, washed twice with TME buffer containing 2 mM DTT, and finally resuspended in the same buffer at 4 -6 mg of protein/ml. Immediately prior to each experiment, membrane samples were diluted to ϳ2.5 mg/ml with TME buffer and passed through centrifuge columns of Sephadex-G-50 -80 (49) equilibrated with the same buffer to remove DTT.
Electrophoresis and Immunoblotting-SDS-PAGE was performed according to Laemmli (50) on 4 -15% gradient gels (Ready Gels, Bio-Rad), using 3-23 g of protein/lane. Because non-reducing conditions were used in gel assays for intersubunit disulfide cross-linking, it was necessary to block free cysteine residues. For this purpose, samples were treated for an additional 30 min at 22°C with 5 mM NEM and then denatured with SDS at 37°C, again in the presence of 5 mM NEM, before applying to gels. Protein bands were stained with SYPRO Orange (Molecular Probes, Inc.) and analyzed using an imager (STORM 860, Amersham Biosciences) in the blue-fluorescence mode as described previously (18). For the immunoblots shown in Fig. 3, proteins were transferred from an unstained gel to a polyvinylidene difluoride membrane (Invitrogen) in a Bio-Rad Mini Trans-Blot cell for 1 h at 250 mA in 10 mM CAPS buffer, pH 11, containing 10% methanol. The blotted membrane was blocked for 1 h with 5% nonfat, dried milk in TBST buffer (10 mM Tris-HCl, 150 mM NaCl, 0.05% Tween 20, pH 8.0), incubated for 2 h in TBST containing 1% bovine serum albumin with anti-⑀ (1:5000), anti-c (1:5000) or anti-FLAG M2 antibody (1:6000), and rinsed three times with TBST containing an additional 0.1 M NaCl. The immunoblot was then incubated for 1.5 h with 2 Ci of the appropriate 125 I-labeled secondary antibody, rinsed four times with TBST containing 0.1 M NaCl, air dried, and exposed to a standard phosphor screen (Molecular Dynamics) for 12-14 h. The exposed screen was then scanned with the imager in phosphorimaging mode. Analysis of 125 I-  1 and 2) and F 1 -stripped, cQ42C-F 0 membranes (lanes 3 and 4) were incubated at 1 mg/ml for 20 min at 22°C in the presence or absence of 50 M DTNB. Samples containing 23 g of membrane protein were then treated with 5 mM NEM, denatured, and subjected to SDS-PAGE under nonreducing conditions. Stained protein bands were quantitated as described under "Experimental Procedures." The molecular weight of the cross-linked product was estimated using "Mark-12" protein standards (Novex; not shown). Only the central part of the gel image is presented. reconstituted ␤D380C/⑀S108C/ ⑀E31C/cQ42C-F 0 F 1 membranes. In C, the ␤ subunit included an N terminus FLAG epitope (see text). Reconstituted membranes were treated in the presence or absence of 50 M DTNB as described for Fig. 2. Aliquots containing ϳ10 g of membrane protein were subjected to SDS-PAGE under non-reducing conditions and immunoblotting was performed as described under "Experimental Procedures." Besides the 74-kDa band identified here as ␤-⑀-c, other cross-linked bands are labeled as identified in previous studies (␤-⑀ (13), ⑀-c and c-c (31), and ␤-␤ (7, 13)). The sensitive exposures shown here were intended to visualize all cross-linked products. Hence, the relative intensities do not reflect the amount of protein in each band. Quantitation of stained bands (as in Fig. 2) and more quantitative anti-⑀ blots (not shown) confirm that 20 min of oxidation gives high yields of ␤-⑀-c, with only minor amounts of ␤-⑀, ⑀-c, and ⑀ bands remaining. labeled protein bands was done using ImageQuaNT 5.1 software (Amersham Biosciences). The immunoblots shown in Fig. 4 were treated with anti-⑀ primary antibody, washed as described above, and then incubated for 1.5 h in TBST containing 1% bovine serum albumin with fluorescein-conjugated anti-rat secondary antibody (1:1600), rinsed four times with TBST containing 0.1 M NaCl, and air-dried for 20 min. Quantitation of ⑀ in protein bands was performed with the imager in the blue-fluorescence mode.
Other Assays-The ATPase activity of membrane vesicles (2-3 g of protein/ml) was measured at 30°C with a coupled enzyme assay (51) in buffer containing 50 mM MOPS-Tris, pH 8.0, 50 mM KCl, 2 mM ATP, and 3 mM MgCl 2 . To prevent formation of a transmembrane proton gradient and to block NADH oxidation by the respiratory chain, 5 M FCCP and 5 mM KCN were also present in the assay. NADH-and ATP-driven uptake of protons by membrane vesicles was assayed by the quenching of fluorescence of 2.5 M acridine orange using 100 g of membrane protein/ml as described previously (52). Protein concentrations were determined by a modified Lowry assay (53).

Membrane-bound F 0 F 1 -ATP Synthase with Cysteine Substitutions in the ␤, ⑀, and c Subunits Retains Coupled Functions-
Previous studies with soluble and membrane-bound F 1 showed that ␤D380C (13) or ␤E381C (43), in the DELSEED motif of ␤, could form a disulfide cross-link with ⑀S108C in the C-terminal domain of ⑀. Other studies established that ⑀E31C, in the N-terminal domain of ⑀, could be cross-linked to cQ42C in the polar loop of a c subunit in F 0 (31). To test for the simultaneous interaction of ⑀ with ␤ and c subunits, we combined the ␤D380C, ⑀S108C, and ⑀E31C mutations in a single construct. To avoid complications due to cross-link formation between ␤D380C and the naturally occurring ␥Cys-87, the ␥C87S mutation was also included (54) but not specified in the enzyme nomenclature due to its silent role. Soluble ␤D380C/⑀S108C/ ⑀E31C-F 1 was purified and reconstituted with F 1 -depleted membrane vesicles containing cQ42C-F 0 . To determine the combined effects of these mutations, we measured the ATPase and proton-pumping activities of the reconstituted membranes. As shown in Table I, the membranes exhibited wild-type activities under reducing conditions (line 1). Furthermore, pretreating membranes with DCCD inhibited both ATP hydrolysis and ATP-driven proton transport (line 2), as observed with native wild-type membranes.
Oxidation of the Mutant ATP Synthase Yields a Disulfidelinked ␤-⑀-c Product-To test for disulfide cross-linking of subunits in ␤D380C/⑀S108C/⑀E31C/cQ42C-F 0 F 1 , membranes were treated with DTNB. As shown by nonreducing SDS-PAGE (Fig.  2), a cross-linked product of about 74 kDa is seen in the oxidized sample (lane 1) that is not present in the non-oxidized control (lane 2). The amount of protein remaining in the ␤ band of the oxidized sample (lane 1) is 34% less than that of the control (lane 2), indicating a high yield of cross-linking between ⑀ and one of the three ␤ subunits. As expected, the 74-kDa band was not observed with F 1 -stripped E. coli membranes (Fig. 2,  lanes 3 and 4). In experiments not shown, we also obtained the 74-kDa cross-linked product using 20 M CuCl 2 as oxidant in TME buffer lacking EDTA. However, CuCl 2 treatment required a longer incubation time, gave a lower yield of the 74-kDa product, and gave increased levels of undesired crosslinks (e.g. ␤-␤).
The size of the major cross-linked band is consistent with a predicted molecular mass of 73 kDa for a 1:1:1 ␤-⑀-c complex that would result upon formation of both the ␤D380C-⑀S108C and ⑀E31C-cQ42C cross-links. Immunoblot analysis confirmed the presence of subunits ⑀ (Fig. 3A) and c (panel B) in the 74-kDa band. To test for the presence of the ␤ subunit, ␤D380C containing the "FLAG" epitope at the N terminus (11) was exchanged into ␤D380C/⑀S108C/⑀E31C-F 1 using a subunit dissociation/re-association procedure (7). The resulting hybrid F 1 was bound to cQ42C-F 0 in F 1 -depleted membranes, and the reconstituted membranes were incubated with or without DTNB. Immunoblotting with anti-FLAG antibody confirmed the presence of ␤ in the 74-kDa band (Fig. 3C). We conclude that in functionally coupled, membrane-bound EcF 0 F 1 , ⑀ can be oriented so as to allow simultaneous cross-linking to a ␤ subunit in F 1 and a c subunit in F 0 .
Formation of the ␤-⑀-c Cross-linked Product Inhibits Catalytic Activity-Oxidation of membranes containing ␤D380C/ ⑀S108C/⑀E31C/cQ42C-F 0 F 1 with DTNB had no significant effect on NADH-driven proton pumping, but strongly inhibited ATP hydrolysis and ATP-driven proton pumping (Table I, line 3). It is likely that this inhibition is primarily due to the covalent tethering of the stator to the rotor, which occurs upon formation of the ␤D380C-⑀S108C disulfide bond (13). However, formation of c-c (Fig. 3B) and ␤-␤ (Fig. 3C) cross-links may also contribute (13,31).
The Rate of ␤/⑀ Cross-linking Is Influenced by the Type of Nucleotide Present-A number of previous studies have indicated that the filling of adenine nucleotide binding sites on F 1 can trigger significant conformational change in the C-terminal domain of ⑀, and several models for nucleotide-dependent regulation of F 0 F 1 -ATP synthase by ⑀ have been proposed (35,39,(55)(56)(57). It was therefore of interest to compare the effects of FIG. 4. Nucleotide-specific effects on the time course of ␤-⑀ cross-linking. Aliquots of the ␤D380C/⑀S108C/cQ42C-F 0 F 1 -reconstituted membranes were incubated at 0.45 mg/ml in TME buffer containing 5 M FCCP for 1 min in the presence of 2 mM ADP, ATP, or AMP-PNP and then oxidized with 50 M DTNB for the times indicated. NEM (5 mM final) was then added, and each sample was incubated for an additional 30 min. As a control, membranes were resuspended in TME buffer containing 5 M FCCP, and 5 mM NEM was added immediately before addition of DTNB. Aliquots containing 3 g of membrane protein were subjected to SDS-PAGE and immunoblotting under nonreducing conditions, and bands containing ⑀ were detected as described under "Experimental Procedures." The immunoblots shown in A illustrate the effects of ATP and AMP-PNP on the rate of formation of the ␤-⑀ cross-link. B shows the time course for ␤-⑀ disulfide formation in the presence of ATP (q), ADP (E), or AMP-PNP (). The relative yield of ␤-⑀ is normalized to the yield obtained in a sample oxidized for 60 s in the presence of MgATP. Data points are averages of four determinations, and error bars represent the standard error. Control immunoblotting experiments, using varied amounts of oxidized membranes, established that the assay provided a linear response to ⑀ in the ␤-⑀ band.
various nucleotides on the interaction of ⑀ with the ␤ and c subunits.
In preliminary experiments, we found the kinetics of ␤-⑀-c cross-linking to be complex due to the formation of two singledisulfide intermediates (␤-⑀ and ⑀-c). Hence, we chose to measure the rates of ␤-⑀ and ⑀-c cross-linking separately, using mutant EcF 0 F 1 that contained only one or the other cysteine pair. The kinetics of ␤-⑀ and ⑀-c formation were followed over a 10-to 60-s period. The rate of ␤-⑀ cross-linking was significantly different in the presence of MgATP, MgADP, or MgAMP-PNP (Fig. 4). In contrast, the rate of ⑀-c cross-linking was not influenced by the nature of the nucleotide added (data not shown). Regardless of the nucleotide present, the same maximal yields of the ␤-⑀ and ␤-⑀-c cross-links were obtained after a 20-min incubation with DTNB (data not shown).

DISCUSSION
Using functionally coupled E. coli ATP synthase, we demonstrate that the ⑀ subunit can span the central stalk region to interact simultaneously with a ␤ subunit of F 1 and a c subunit of F 0 . This possibility was suggested by the open conformation of ⑀ (Fig. 1C) observed in an ⑀/␥-fragment complex (45) and by a cross-linking experiment, which showed that ⑀ can adopt a similar conformation in EcF 0 F 1 (35). However, the latter study did not eliminate the possibility that the N-terminal domain of ⑀ might have to detach from the c-ring before the C-terminal domain of ⑀ can interact with the ␣ 3 ␤ 3 hexamer (25). Furthermore, based on a hybrid structure where the E. coli ⑀/␥-fragment complex and MF 1 were superimposed (45), we used ab initio modeling of the missing hairpin loop of ⑀ to determine that ⑀Ser-108 would still be about 25 Å from the nearest ␤Asp-380 (Fig. 1C). This large gap was also noted recently by Yoshida's group (58). Movement of ␤Asp-380 and ⑀Ser-108 into closer proximity would likely be prohibited with ⑀ in the open state, because the upward protrusion of the C-terminal helix ⑀ of already clashes sterically with a ␤ subunit (45). Thus, our ␤-⑀-c cross-linking results suggest that ⑀ can adopt an "extended" conformation in which the hairpin loop residues of the Cterminal domain project further from the surface of F 0 to place ⑀Ser-108 in contact with a ␤-DELSEED loop at the bottom of F 1 .
This could be accomplished by unfolding and linearizing the two C-terminal ␣-helices as proposed recently (Ref. 58 and Fig.  5A) or by swinging the C-terminal domain upward while retaining the antiparallel folding of the helices as we propose in Fig. 5B. We favor the latter structure based on an earlier cross-linking study with EcF 0 F 1 (34). Schulenberg and Capaldi (34) showed that an interdomain cross-link that locked ⑀ in the closed conformation stimulated the ATPase activity about 2-fold. In contrast, an intradomain cross-link between the two helices in the C-terminal domain had no effect, suggesting that the ability of ⑀ to modulate the ATPase activity of the synthase does not require the unfolding of this domain.
An important related issue is the question of how epsilon might regulate the activity of the ATP synthase. The extended unfolded conformation of the C-terminal domain of ⑀ shown in Fig. 5A was trapped by formation of a cross-link between ␥ and the C terminus of ⑀ (58). With an 80% yield in the cross-linking reaction, it was found that ATP hydrolysis was inhibited to a greater extent than ATP synthesis (75% versus 25%). A similar differential inhibition of hydrolysis and synthesis activities was reported earlier (35) with ⑀ trapped in the open conformation (Fig. 1C). In each case, the authors concluded that ⑀ can assume a conformation that allows ATP synthesis but not ATP hydrolysis. However, it has been documented that formation of the electrochemical gradient can be rate-limiting for ATP synthesis by bacterial membrane vesicles. For example, DCCD modification of a large fraction of ATP synthase complexes caused a 20-fold decrease in the rate of coupled ATP hydrolysis but only reduced the rate of ATP synthesis by 4-fold (59). Hence, our interpretation is that the conformations shown in Figs. 1C and 5A are devoid of any catalytic activity and that the synthesis activity observed was due to residual non-crosslinked synthase. Indeed, we propose that the structure shown in Fig. 5B is inactive due to interactions between the rotor and stator that prevent subunit rotation. However, this state is transient, and catalytic activity in both directions would be  Fig. 1B. A, the extended, unfolded conformation of ⑀ proposed by Suzuki et al. (58). To accommodate the ␥-⑀ disulfide (⑀135C-␥2C, not visible) observed in that study, the C-terminal helix of ⑀ is threaded between the ␤ E and ␣ E subunits, antiparallel to the N-terminal helix of ␥. This was achieved by adjusting phi and psi dihedral angles for residues between the N-and C-terminal domains of ⑀ (⑀81-86) and between the two ␣-helices (⑀106, 109, and 110), but significant steric collisions remain between ⑀ and the surrounding subunits. In this model, ␤Asp-380 and ⑀Ser-108 are ϳ12 Å apart (C␤-C␤) but could be closer with a small movement of the DELSEED loop of ␤ E or a partial rotary step. B, a proposed "extended hairpin" conformation that places ␤Asp-380 and ⑀Ser-108 within 6 Å (C␤-C␤). This was achieved by adjusting the phi and psi angles around three residues between the N-and C-terminal domains of ⑀ (⑀81, 83, and 86). Phi and psi angles were manipulated in stereo mode in the program Insight-II (Accelrys, Inc.). regulated by altering the time ⑀ spends in this orientation versus the closed active state (Fig. 1B). Molecular modeling gives no indication of significant steric hindrance in swinging the C-terminal domain of ⑀ between these two positions, and the transition only requires dihedral angle changes in the hinge region connecting the two domains.
Nucleotide-specific changes in the rate of ␤-⑀ cross-linking observed in Fig. 4 may reflect changes in the partitioning of ⑀ between the active conformation (Fig. 1B) and an inactive conformation (Fig. 5B). With high ATP concentrations in respiring cells, the net rate of ATP synthesis will slow, and it might be an advantage to the organism to limit synthesis/ hydrolysis cycling by inhibiting excess synthase. This would enhance metabolic efficiency if the conservation of energy by an idling synthase motor is less than 100%. Under such conditions, ⑀ would make more frequent contact with ␤, as detected by the increased rate of ␤-⑀ cross-linking in the presence of ATP (Fig. 4). In contrast, when a rapid ATP synthesis rate is necessitated by elevated ADP levels, it would be an advantage to have a larger fraction of active synthase. This predicts less frequent contact between ⑀ and ␤, as detected by the decreased rate of ␤-⑀ cross-linking in the presence of ADP (Fig. 4). Finally, if the synthase is already inhibited by the binding of MgAMP-PNP (14), it might be expected that the stimulus for a regulatory stator/rotor interaction would be diminished.
Results of this study also provide information about the conformational mobility of the C-terminal domain of ⑀. An earlier study with chloroplast thylakoids indicated that it undergoes a rapid conformational change (Ͻ30 s) in response to membrane energizing (60). In contrast, previous cross-linking studies with ⑀S108C in EcF 0 F 1 (43,61) suggested that different nucleotide conditions trap the C-terminal domain of ⑀ in relatively static alternate states, because selective cross-linking was achieved through slow, Cu 2ϩ -catalyzed oxidation (15 min to Ͼ1 h). In the present study, ␤D380C-⑀S108C cross-linking shows a sensitivity to different nucleotides but does so over a much briefer period of time (Fig. 4). Our results indicate that the C-terminal domain of ⑀ can sample alternate conformations on a much faster time scale (seconds) than previously recognized for the E. coli synthase. Such mobility would allow for a rapid response to regulatory signals.