Solution Structure of the ε Subunit of the F1-ATPase from Escherichia coli and Interactions of This Subunit with β Subunits in the Complex*

The solution structure of the ε subunit of theEscherichia coli F1-ATPase has been determined by NMR spectroscopy. This subunit has a two-domain structure with an N-terminal 10-stranded β sandwich and a C-terminal antiparallel two α-helix hairpin, as described previously (Wilkens, S., Dahlquist, F. W., McIntosh, L. P., Donaldson, L. W., and Capaldi, R. A. (1995) Nat. Struct. Biol. 2, 961–967). New data show that the two domains interact in solution in an interface formed by β strand 7 and the very C-terminal α-helix. This interface involves only hydrophobic interactions. The dynamics of the ε subunit in solution were examined. The two domains are relatively tightly associated with little or no flexibility relative to one another. The ε subunit can exist in two states in the ECF1F0 complex depending on whether ATP or ADP occupies catalytic sites. Proteolysis of the ε subunit in solution and when bound to the core F1 complex indicates that the conformation of the polypeptide in solution closely resembles the conformation of ε when bound to the F1 in the ADP state. Chemical and photo-cross-linking show that the ε subunit spans and interacts with two β subunits in the ADP state. These interactions are disrupted on binding of ATP + Mg2+, as is the interaction between the N- and C-terminal domains of the ε subunit.

An F 1 F 0 -type ATP synthase is found in the inner mitochondrial membrane, the inner membrane of bacteria, and the thylakoid membrane of chloroplasts where it functions to convert the free energy of the proton motive force into the chemical energy source ATP (for recent reviews see Refs. [1][2][3]. This large enzyme complex is composed of two major parts, a watersoluble F 1 made of three ␣, three ␤, and one copy of each of the ␥, ␦, and ⑀ subunit; and a membrane-embedded F 0 consisting of 1 a, 2 b, and 9 -12 c subunits. The overall molecular weight of the complex is 520,000. There are three catalytic nucleotide binding sites located on the ␤ subunits of the F 1 and one proton channel formed by the a and c subunits in the membraneembedded F 0. The two parts of the complex are linked by two 45-Å stalks, a central one formed by the ⑀ and part of the ␥ subunit and a peripheral one, constituted by the hydrophilic portions of the two b subunits of the F 0 and the ␦ subunit of the F 1 (4,5). The function of the central stalk is to transmit energy between the proton channel in the membrane and the catalytic nucleotide binding sites on the ␤ subunits via conformational changes in the stalk-forming proteins. The peripherally located second stalk functions as a "stator" or "scaffold" to form a rigid link between the catalytic domain and the a subunit of the F 0 with respect to a mobile domain formed by the ␥, ⑀, and c subunit ring (6,7).
The 2.8-Å crystal structure of the mitochondrial F 1 provides a molecular picture of the ␣ and ␤ subunits and part of the ␥ subunit (8). The ␣ and ␤ subunits are arranged alternatingly in a hexagon around a central cavity in which the N and C termini of the ␥ subunit are located. Unfortunately, part of the ␥ subunit and the entire ⑀ subunit (Escherichia coli nomenclature), although present, were not resolved in the crystal structure, probably because they were disordered.
Structural features of the N-and C-terminal domains of the ⑀ subunit have been obtained in this laboratory by NMR spectroscopy (9). These data show that the N-terminal domain of the subunit is folded in a 10-stranded ␤ sandwich, and the C-terminal third of the protein is arranged in an antiparallel two ␣-helix hairpin. A possible arrangement of the two domains with respect to one another was suggested based on spin labeling experiments and on a small number of long range NOEs 1 between the two domains. However, because all structure calculations had been performed for the two domains independently, the proposed interaction site between the two domains remained speculative. While the work described here was in progress, a crystal structure of the E. coli ⑀ subunit at 2.8 Å was obtained (10), which confirmed the domain structure seen in NMR and provided new information on the interaction of the two domains. To establish that the arrangement of the two domains relative to one another was not an artifact of the crystallization, we have completed the structure determination of the ⑀ subunit in solution by NMR. The dynamics of the polypeptide have been investigated. In the intact ECF 1 and ECF 1 F 0 , the ⑀ subunit exists in at least two arrangements (11,12). Proteolysis studies show that in solution the ⑀ is in the same conformation as when ADP is present on the enzyme.
NMR Spectroscopy and Data Analysis-NMR spectroscopy and data analysis were performed essentially as described previously (9). A total of 1321 experimental restraints have been collected for the full-length polypeptide, including 1212 NOE distance restraints, 30 hydrogen bond distance restraints, and 79 dihedral angle restraints. NOE intensities were sorted visually into the three classes: strong (1.8 -3 Å), medium * This work was supported by National Institutes of Health Grant 58671. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
(1.8 -4 Å), and weak (1.8 -5.5 Å). Hydrogen bonds were only added when the accepting oxygens were unambiguously determined from structure calculations without or with only a subset of hydrogen bonds. Structure calculations were performed with the program X-PLOR (15) according to standard protocols. A template with ideal local geometry was created based on standard bond length and angles. This template was then used to calculate a starting set of 30 embedded structures using distance geometry. The 30 structures were regularized and repeatedly refined by using simulated annealing and molecular dynamics protocols until stable. All 30 structures converged to give the same fold with 3 structures having NOE violations greater than 0.5 Å. All 30 structures were averaged, and the average was energy minimized to satisfy standard bond lengths and angles. 15 N relaxation parameters were measured based on the inverse detection scheme from a series of two-dimensional spectra (16 -18) as described by Zhou et al. (19). Overall correlation times were calculated from T 1 /T 2 ratios on a residue-byresidue basis as described by Kay et al. (16).
Proteolysis of Isolated and F 1 -bound ⑀ Subunit-ECF 1 and an isolated ⑀ subunit were cleaved with trypsin at a ratio of 1:2000 for ECF 1 and 1:80 for the ⑀ subunit protease to protein (w/w) in 50 mM MOPS/Cl, pH 7, 10% glycerol, 0.5 mM EDTA at a protein concentration of 2.5 and 0.1 mg/ml, respectively. Proteolysis was stopped by the addition of 5 mM phenylmethylsulfonyl fluoride (from a 200 mM stock solution in ethanol). Proteolysis was either carried out in the presence of 2 mM ATP ϩ 0.5 mM EDTA or after turnover of 2 mM ATP in the presence of 2.5 mM MgCl 2 ϩ 0.5 mM EDTA. The rate of proteolysis for the isolated ⑀ subunit FIG. 1. Backbone amide spectrum of the 138-residue E. coli F 1 -ATPase ⑀ subunit. The spectrum has been recorded with a 0.35 mM solution of uniformly 15 N-labeled ⑀ subunit in 5 mM potassium phosphate buffer, pH 7.4, 6% D 2 0, and 3 mM NaN 3 in a 8-mm triple-resonance gradient probe at 20°C. Labels of residues that form the C-terminal ␣ domain are drawn in red. The second primary amide proton of the side chain of Gln 127 is located at a proton chemical shift of 5.6 ppm. All other amides and primary side chain amides have been assigned in the spectrum; however, the intensity of some backbone amide resonances is weak because of solvent exchange (see text). The amide proton of His 109 is exchange-broadened beyond detectability at pH 7.4, and it can be detected at pH values smaller than 7 (approximately 9.1/122 ppm). Protons of primary amides are connected by horizontal lines.
was identical under both nucleotide conditions. Cleavage products were analyzed by SDS-polyacrylamide gel electrophoresis in 10 -18% gradient gels.
Photo-and Chemical Cross-linking Using ⑀M138C-ECF 1 containing the mutation ⑀M138C (20) was prepared using ammonium sulfate precipitation followed by two subsequent spin columns in 50 mM MOPS/Cl, pH 7, 10% glycerol, 0.5 mM EDTA. The enzyme at a protein concentration of 10 mg/ml was reacted with either TFPAM-3 (21, 22) (200 M for 1 h in the dark) or EDC (0.5 mM for 30 min) at room temperature. Excess reagent was removed by one spin column in 50 mM Tris/Cl. The enzyme was converted to the ADP state before cross-linking by the addition of 2 mM ATP and 2.5 mM MgCl 2 . Half of the photolyzed sample was treated with 0.5 mM EDC as described above. Finally, all three samples (EDC-treated, photolyzed, and photolyzed and EDC-treated) were reacted with 0.8 mM NbfCl followed by one spin column to remove excess NbfCl. Cross-linked and labeled products were analyzed by denaturing polyacrylamide gel electrophoresis as described above. The composition of cross-linked bands was analyzed by monoclonal antibody blots as described (23).

RESULTS
Solution Structure of the ⑀ Subunit-Previously, we determined the global fold of the ⑀ subunit from NMR spectroscopy (9). This showed that the N-terminal two-thirds of the protein is folded in a 10-stranded ␤ sandwich with the C-terminal third as an antiparallel two ␣-helix hairpin. The solution structure of the full 138-residue ⑀ subunit, which uses a total of 1321 experimental restraints of which 822 are within the N-terminal 10-stranded ␤ sandwich and 499 within the C-terminal two ␣-helix hairpin, has now been completed. Fig. 1 shows a 1 H-15 N correlation spectrum of a uniformly 15 N-labeled ⑀ subunit. Resonances of residues in the N-terminal ␤ sandwich are labeled in black, whereas amides in the C-terminal ␣-helical domain are in red. The ⑀ subunit has the tendency to aggregate in solution at concentrations higher than 1 mM and at pH values below 7.4. The relatively high pH needed for the solution structure determination leads to solvent exchange broadening of backbone amide resonances from residues Ser 17 , Ser 28 , Gly 37 , Ala 39 , Ile 50 , His 56 , Gly 74 , Glu 91 , Ser 108 , His 109 , and Gly 110 . All these residues are either in turn or loop positions or at the beginning of secondary structure elements throughout the structure. Fig. 2a shows the minimized average of all 30 structures, and a summary of the NMR analysis is given in Table I. The dimensions of the ⑀ subunit are approximately 30 ϫ 25 ϫ 17 Å for the N-terminal ␤ sandwich and 45 ϫ 18 ϫ 15 Å for the C-terminal antiparallel two ␣-helix hairpin. A tentative orientation of the two domains with respect to each other was inferred previously from line broadening of amide resonances of residues between Ala 124 and Lys 130 caused by a nitroxide spin label introduced into the ␤ domain at position 10, as well as from a limited number of long range NOEs between the Cand N-terminal domains (9).
The NMR data here define the interface between the N-and FIG. 2. Solution structure of the E. coli F 1 -ATPase ⑀ subunit. a, the minimized average structure for the entire family of 30 refined structures. b, the interface between the N-terminal ␤ domain and the very C-terminal ␣-helix. Based on the NOE data, the interface is formed by 5 residues within the N-terminal domain (shown in green) and 7 residues from the second ␣-helix (red). The side chain of Ile 131 is not shown, and the long range NOE involving this residue is between the amide proton and the ⑀ protons of Phe 61 . a was created with Molscript, and b was created in RasMol2.6. C-terminal domain by 30 long range NOEs involving the side chains of 12 residues, 5 from the N-terminal domain (Gly 48 , Phe 61 , Tyr 63 , Gly 86 , and Gln 87 ) and 7 from the C-terminal ␣-helix (Ala 126 , Gln 127 , Val 130 , Ile 131 , Leu 133 , Thr 134 , and Ala 137 ). This interface is shown in Fig. 2b. The two domains are held together entirely by hydrophobic contacts, as there is no indication of any hydrogen bonds, salt bridges, or other polar interactions.
Less well defined is the interaction between the two Cterminal ␣-helices themselves. Only a few (17) long range NOEs could be identified because of the overlap of chemical shift values for the interacting protons. These are between the side chains of residues 91, 92, 94, 95, 98, 102, and 105 of the first helix and residues 131, 128, 125, 121, and 114 of the second helix. As reported previously, the backbone amide protons in the two C-terminal ␣-helices are undetectable in D 2 O buffer because of rapid exchange with solvent D 2 O (9). However, amide protons of most of the residues in the ␣-helices can be detected in D 2 O at pD values lower than 7.4 (data not shown).
Backbone Dynamics of the ⑀ Subunit-NMR allows examination of the dynamics of a protein and, in the case of ⑀, the flexibility of the two domains with respect to one another. A first indication that the N-and C-terminal domains of ⑀ are in tight contact in the solution structure is the observation that there are at least 30 resolved long range NOEs between residues of the N-and C-terminal domains (see above). On average, these have the same intensity (not decreased) compared with other long range NOEs within the N-terminal domain (not shown). To examine the flexibility of the two domains with respect to one another in more detail, a backbone 15 N{ 1 H} NOE spectrum was recorded, and the backbone relaxation parameters were measured. A 15 N{ 1 H} NOE spectrum of the ⑀ subunit is shown in Fig. 3. Despite the fact that some of the peaks are weakened because of the solvent exchange (the same ones that are weak in amide correlation spectra recorded at pH 7.4, see above), all of the peaks, including the ones of residues in the linker region and the interface, have a positive intensity, consistent with the ⑀ subunit tumbling as one entity in solution. Fig. 4 shows a summary of the measured relaxation parameters for a total of 100 of the 138 residues. The average T 1 and T 2 times for the ⑀ subunit backbone amides are 571 Ϯ 27 ms (residues 3-136, excluding residues 14, 17, 39, 74, and 107 because of their large errors) and 65 Ϯ 3 ms, respectively.
The measured T 1 and T 2 relaxation parameters shown in Fig. 4   two domains in ⑀. Only Leu 89 shows a somewhat larger than average T 1 , which is not uncommon for a turn position. Moreover, residues involved in the interface between the two domains (Gly 48 , Phe 61 , Tyr 63 , Gly 86 , Gln 87 , Ala 126 , Gln 127 , Val 130 , Ile 131 , Leu 133 , Thr 134 , and Ala 137 ) show none of the unusual relaxation behavior expected for residues undergoing conformational exchange. Thus, any such conformational change must be faster than 10 Ϫ3 -10 Ϫ4 s or slower than the T 2 relaxation time above.
The Conformation of the Isolated ⑀ Subunit Corresponds to the Conformation of the Subunit When Bound to ECF 1 in the ADP State-Previous experiments from this laboratory showed that the ⑀ subunit can adopt at least two conformations, when bound to ECF 1 (11) or in ECF 1 F 0 (12), depending on whether the enzyme is in the ATP form (e.g. ATP/EDTA or AMP-PNPϩMg 2ϩ ) or in the ADP state (ADP/Mg/P i present). For example, the rate with which trypsin cleaves the ⑀ subunit is different whether ADP or ATP is in catalytic sites, with the ADP form of ⑀ being more resistant to protease attack (11). This difference in rates of cleavage of ⑀ under the two conditions is clearly evident in the experiment shown in Fig. 5. With ECF 1 in ATP (lanes 2-5), the first cleavage product observed runs above and subsequent products run below the unmodified ⑀ subunit. In the presence of ADP (lanes 7-10), the rate of cleavage of ⑀ is  6 -10). Pure ⑀ subunit was trypsinized at a ratio of 1:80 protease to protein. The addition of either ATP/EDTA or ADP/Mg/P i had no effect on the rate or the fragment pattern of ⑀. In each experiment, aliquots were removed after 2, 8, 32, and 128 min, and the protease was stopped by addition of 5 mM phenylmethylsulfonyl fluoride from a 200 mM stock solution in ethanol. Phenylmethylsulfonyl fluoride was added to the sample prior to trypsin in lanes 1, 6, and 11. Lanes 1-10 contain 50 g of ECF 1 each, and lanes 11-15 contain 2 g of ⑀ subunit each. much slower under otherwise identical proteolysis conditions (see also Ref. 11). Lanes 11-15 of Fig. 5 show the cleavage pattern of the isolated ⑀ subunit in the presence of much higher levels of trypsin than used for ECF 1 . Similar to the ⑀ subunit bound to ECF 1 with ADP in catalytic sites, there is only very slow cleavage of the polypeptide. The various cleavage sites obtained after prolonged trypsin treatment of the isolated ⑀ subunit were determined from mass spectrometric analysis. Peptides with molecular weights for residues 1-123, 1-100, 1-99, 1-98, and 1-93 were generated sequentially in time course experiments. These are the same cleavage products found by trypsin treatment of ⑀ in ECF 1 in the presence of either ATP or ADP. It is the rate of the cleavage, not the site of cleavage, which varies under different nucleotide conditions in ECF 1 . These sites are in both ␣-helices of the C-terminal domain. The different nucleotide conditions had no effect on cleavage rates in isolated ⑀ subunit (results not shown).
Arrangement of the ⑀ Subunit from Cross-linking Studies-The orientation of the N-terminal ␤ sandwich domain in the ATP synthase complex is well defined based on a large number of chemical and photo-cross-linking results (20,22,24,25). From these data, interfaces to the ␥ subunit of the F 1 and to the c subunit oligomer of the F 0 could be identified, involving one face and the bottom of the ␤ sandwich, respectively.
Several cross-linking studies have shown that there is binding of the ⑀ subunit to ␣ and ␤ subunits and that this is through the C-terminal ␣-helix-loop-␣-helix domain. For example, cross-linking can be induced between a ␤ and ⑀ subunit by the water-soluble carbodiimide EDC (26), a cross-link subsequently shown to involve Glu 381 of ␤ and one of three Ser residues (106 -108) in ⑀ (27). Cross-linking from ␣ or ␤ to ⑀ can also be obtained by replacing Glu 381 of ␤ (24) or Ser 411 of ␣ (28) along with Ser 108 of ⑀ with Cys residues. Finally, a Cys replacing the C-terminal residue Met 138 of ⑀ can be cross-linked to a ␤ subunit when modified with the photoactivatable cross-linker TFPAM-3 (20). As Ser 108 and Met 138 are at the opposite ends of the helix-loop-helix region of ⑀, it seems likely that this subunit spans two ␤ subunits.
To test this possibility, cross-linking from the two sites was performed sequentially using ECF 1 from the mutant ⑀M138C. As shown in Fig. 6, EDC reaction, or photoactivation after TFPAM-3 modification of Cys 138 alone, gave a major crosslinked product of M r(app) . 70,000. In contrast, when both reactions were performed sequentially, a significant amount of cross-linked product of M r(app) . 130,000 was produced. This large product must contain ⑀ cross-linked to two ␤ subunits, i.e. it is a ␤-⑀-␤ adduct. The presence of ⑀ and ␤ in the 130,000 cross-link product was confirmed by monoclonal antibodies (results not shown).
In one set of experiments, ECF 1 from the mutant ⑀M138C was modified with NbfCl after the cross-linking reaction (29). No fluorescence because of bound NbfCl was observed in the ␤-⑀ cross-link generated with either EDC or via photo-cross-linking from the Cys at position 138. However, bound reagent was readily apparent in a noncross-linked ␤ subunit (not shown).

DISCUSSION
The 2.8-Å crystal structure of a major part of the F 1 from beef heart mitochondria has provided important insight into cooperative catalysis by ATP synthases (8). However, to fully understand energy coupling within this enzyme, it will almost certainly be necessary to obtain a high resolution structure of the entire complex. The lack of any high resolution structural data for the stalk region and membrane-bound F 0 makes it particularly difficult to establish the molecular mechanism for the coupling of ATP hydrolysis to proton translocation.
We have been examining individual purified subunits of the stalk region to study their structure and interaction. Previously, we obtained structural models of the N-and C-terminal domains of the ⑀ subunit from NMR spectroscopy (9). Here, we describe the solution structure of the entire polypeptide, which adds important information on the interaction sites between the two domains, and the dynamics of the molecule as a whole. The interaction between the two domains involves side chains from ␤ strand 7 of the N-terminal domain and one face of the very C-terminal ␣-helix. No interactions were observed between the first C-terminal ␣-helix (residues 91-106) and the N-terminal domain. Recently, the crystal structure of the E. coli ⑀ subunit has been determined at 2.3-Å resolution (10). This crystal structure confirmed our previously determined structures of the two domains and also showed a tight interaction of the N-and C-terminal domains mainly via hydrophobic residues. The tight interaction between the two domains of ⑀ seen in the crystal structure determination could have been enforced by the crystal packing. The fact that the same interaction is now seen by NMR adds confidence to it being a preferred conformation of the polypeptide. In conjunction with the NMR structure determination, we have been examining the interactions of the ⑀ subunit with other subunits of ECF 1 F 0 by cross-linking experiments. Here, we extend these studies and show that the C-terminal domain of the ⑀ subunit binds in such a way as to span two ␤ subunits, with the region around Ser 108 interacting with one and Met 138 being close to the other. The distance between Ser 108 and Met 138 from the structure shown in Fig. 2 is approximately 40 Å and, therefore, sufficient to interact with two ␤ subunits at the same time.
It is clear from the x-ray structure of MF 1 (8), as well as from biochemical studies, that the three ␤ subunits are in different conformations that appear to be determined by nucleotide bind- ing and by the interaction with the small subunits. In the crystal structure, one ␤ subunit has AMP-PNP bound (called ␤ TP ), a second ␤ subunit has ADP bound (␤ DP ), and the third ␤ subunit is empty (␤ E ). From cross-linking studies, we have defined the three ␤ subunits as follows: ␤␥ is the ␤ subunit in which a Cys at position 381 (in the DELSEED region) can be cross-linked to an intrinsic Cys (Cys 87 ) of the ␥ subunit; ␤⑀ is the ␤ subunit in which Cys 381 cross-links to ⑀Cys 108 ; ␤ free is the ␤ subunit that is not cross-linked to either ␥ or ⑀ (24,29). From the crystal structure, ␤␥ is equivalent to ␤ TP . Labeling studies have shown that NbfCl preferentially binds to ␤ free (29), which is equivalent to ␤-E based on a crystal structure of the mitochondrial enzyme reacted with NbfCl. 2 In the experiments reported here, NbfCl labeled free ␤ subunit but not ␤ subunits involved in the ␤-⑀ cross-link products (␤-⑀ via 108 or 138 and ␤-⑀-␤). Therefore, the ␤ subunit ⑀Met 138 is close to must be ␤␥(␤ TP ), whereas the ␤ subunit cross-linking to ⑀ via position 108 has to be ␤ DP . In summary, the ⑀ subunit spans between ␤ TP and ␤ DP .
The ⑀ subunit can exist in at least two different conformations when bound to the core F 1 complex (11,12,30). With ADP/Mg/P i present, ⑀ is less sensitive to proteolytic degradation compared with the "active" state of the complex, when ATP is in catalytic sites. The structure described here and in Uhlin et al. (10) appears to be the ADP state, based on trypsin cleavage studies. On ATP binding to F 1 , protease digestion at several sites in the C-terminal domain is facilitated, and these sites are in both ␣-helices and the backbone within the region of interaction with the N-terminal domain. This can only happen if the helical hairpin comes off its binding site on the ␤ subunits and the N-terminal ␤ sandwich and unfolds to allow the trypsin cleavage. The release of ⑀ from interaction with ␣ and ␤ subunits on ATP binding is evident in the loss of the EDC-induced cross-link between ⑀Ser 108 and the DELSEED region in ␤ DP (11), as well as of the photochemical cross-link from ⑀M138C to ␤ TP (20) on reaction with ATP ϩ EDTA or AMP-PNP ϩ Mg 2ϩ .
Based on the NMR data, the interaction between the two domains of the ⑀ subunit appears strong, with an isolated ⑀ subunit behaving as a rigid molecule in solution. Protease digestion studies of the isolated ⑀ subunit show that separations of the two domains by disruption of the interface between the two is not favored but does occur (rate much slower than T 2 ; see "Results"). The implication is that ATP binding and the consequent conformational changes in ␤ subunits generate an energetically unstable arrangement of the ⑀ subunit.
It is now generally accepted that the coupling between the proton channel and catalytic sites is accompanied by a threestepped rotation of a domain formed by the ␥ (31-34) and ⑀ subunits (35), and it is further assumed that a ring of 12 c subunits in the F 0 is also part of the rotating unit (36,37). In previous studies, we showed that dicyclohexylcarbodiimide inhibition of the F 1 F 0 ATP synthase can lock ⑀ in either the ADP or ATP state, depending on whether the carbodiimide reaction took place with ADP or ATP in catalytic sites (12). These data are consistent with the conformational switch in ⑀ being coupled to a rotary motion of the ␥⑀c 9 -12 domain and/or proton translocation. At the same time, there is evidence that the ⑀ subunit in the ATP state resides preferentially at ␣ subunits rather than attached to ␤ subunits. This is seen in cross-linking studies (28) and in electron microscopy studies (38). Thus, a cysteine at position 108 in the ⑀ subunit cross-links preferentially to the ␣ subunit when going from the ADP to the ATP state, and a gold particle bound to the ⑀ subunit at position 38 can be seen superimposed on either an ␣ (ADP form) or a ␤ subunit (ATP form). Taken together, our data suggest that the rotation of the ␥⑀c domain in the intact F 1 F 0 ATP synthase requires conformational rearrangements of the ⑀ subunit in each step. The possibility that rotation is six-stepped, rather than three-stepped, warrants further study.