Rotation of the ε Subunit during Catalysis by Escherichia coli FOF1-ATP Synthase*

We report evidence for catalysis-dependent rotation of the single ε subunit relative to the three catalytic β subunits of functionally coupled, membrane-bound FOF1-ATP synthase. Cysteines substituted at β380 and ε108 allowed rapid formation of a specific β-ε disulfide cross-link upon oxidation. Consistent with a need for ε to rotate during catalysis, tethering ε to one of the β subunits resulted in the inhibition of both ATP synthesis and hydrolysis. These activities were fully restored upon reduction of the β-ε cross-link. As a more critical test for rotation, a subunit dissociation/reassociation procedure was used to prepare a β-ε cross-linked hybrid F1 having epitope-tagged βD380C subunits (βflag) exclusively in the two noncross-linked positions. This allowed the β subunit originally aligned with ε to form the cross-link to be distinguished from the other two βs. The cross-linked hybrid was reconstituted with FO in F1-depleted membranes. After reduction of the β-ε cross-link and a brief period of catalytic turnover, reoxidation resulted in a significant amount of βflag in the β-ε cross-linked product. In contrast, exposure to ligands that bind to the catalytic site but do not allow catalysis resulted in the subsequent cross-linking of ε to the original untagged β. Furthermore, catalysis-dependent rotation of ε was prevented by prior treatment of membranes with N,N′-dicyclohexylcarbodiimide to block proton translocation through FO. From these results, we conclude that ε is part of the rotor that couples proton transport to ATP synthesis.

F O F 1 -ATP synthases are found embedded in the membranes of mitochondria, chloroplasts, and bacteria (1,2). During oxidative phosphorylation and photophosphorylation, synthases couple the movement of protons down an electrochemical gradient to the synthesis of ATP. The F O sector is composed of membrane-spanning subunits (ab 2 c 12 in Escherichia coli) (3) that conduct protons across the membrane, whereas the F 1 sector (␣ 3 ␤ 3 ␥␦⑀) is an extrinsic complex that contains catalytic sites for ATP synthesis. F 1 can be removed from the membrane in a soluble form that functions as an ATPase, and rebinding F 1 to F O in membranes restores the capacity to catalyze net ATP synthesis. A high-resolution structure for bovine F 1 shows a hexamer of alternating ␣ and ␤ subunits surrounding a single ␥ subunit. The three catalytic sites of F 1 are located on the three ␤ subunits at ␣/␤ subunit interfaces (4).
A model for energy coupling by F O F 1 -ATP synthases that has gained widespread support is called the binding change mechanism (1). According to this proposal, the major energy-requiring step is not the synthesis of ATP at catalytic sites but rather the simultaneous and highly cooperative binding of substrates to and release of products from these sites (5,6). Furthermore, it was proposed that these affinity changes are coupled to proton transport by the rotation of a complex of subunits that extends through F O F 1 . Rotation of the ␥ subunit in the center of F 1 is thought to deform the surrounding catalytic subunits to give the required binding changes (7), whereas rotation of the c subunits relative to the single a subunit in F O is believed to be required for completion of the proton pathway (8 -10).
The rotary aspect of the binding change mechanism remained a popular but speculative idea for a number of years until a critical test became possible following the publication of a high resolution structure for F 1 (4). Focusing on a ␤/␥ intersubunit point of contact identified in the structure, we introduced a Cys into the ␤ subunit at a position (␤380) that would place it in close proximity to a naturally occurring Cys on the ␥ subunit (␥C87). When the resultant ␤D380C-F 1 was exposed to an oxidant, a rapid and specific ␤D380C-␥C87 disulfide crosslink was formed (9,11). Using a subunit dissociation/reassociation approach with the ␤-␥ cross-linked enzyme, we incorporated radioisotope-or epitope-labeled ␤ subunits into the two noncross-linked ␤ subunit positions. Following reduction of the cross-link and a short burst of ATP hydrolysis (9,12) or synthesis (13), labeled and unlabeled ␤ subunits in the hybrid F 1 showed a similar capacity to form a disulfide bond with the ␥ subunit indicating that ␥ had rotated relative to the three ␤ subunits during catalysis. Subsequently, additional evidence for subunit rotation during ATP hydrolysis was obtained using immobilized chloroplast F 1 with a spectroscopic probe attached near the C terminus of the ␥ subunit. Recovery of polarized absorption after photobleaching was used to monitor the rotational motion of ␥ during ATP hydrolysis by the tethered F 1 on a time-resolved basis (14). Finally, in a dramatic visual demonstration, a fluorescent actin filament attached to one end of the ␥ subunit of immobilized bacterial F 1 was seen by fluorescence microscopy to undergo multiple unidirectional rotations during ATP hydrolysis (15). Now that catalysis-dependent rotation of ␥ relative to the catalytic ␤ subunits is well established (16 -19), it is important to identify other components of the rotor. A likely candidate is the ⑀ subunit. It forms a tight 1:1 complex with purified ␥ (20), and cross-linking ⑀ to ␥ causes less than a proportional amount of inhibition (21)(22)(23). In contrast, cross-linking ⑀ to either ␤ (24) or ␣ (25) in soluble F 1 strongly inhibits ATPase activity. The fact that ⑀ is randomly oriented in F O F 1 relative to the ␣ subunit that interacts with the single ␦ subunit is also consistent with ⑀ being part of the rotor (26). Finally, it was recently reported that ATP hydrolysis promotes the rotation of ⑀ in immobilized F 1 from chloroplasts (27) and thermophilic bacte-ria (28). In work presented here, we have extended our crosslinked hybrid approach to demonstrate rotation of ⑀ relative to the three ␤ subunits during catalysis by functionally coupled membrane-bound F O F 1 .

EXPERIMENTAL PROCEDURES
Materials-NADH, ATP, ADP, phosphoenolpyruvate, carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP), 1 N,NЈ-dicyclohexylcarbodiimide (DCCD), N-ethylmaleimide, and hexokinase were supplied by Sigma; pyruvate kinase and lactate dehydrogenase were supplied by Boehringer Mannheim; 5,5Ј-dithiobis(2-nitrobenzoate) (DTNB) was supplied by Aldrich; and dithiothreitol (DTT) was supplied by American Bioanalytical (Natick, MA). Oligonucleotides for site-directed mutagenesis were synthesized by Life Technologies, Inc. Pfu DNA polymerase I was from Stratagene, and all restriction endonucleases were from New England Biolabs. Anti-Flag M2 antibody was obtained from Eastman Kodak, 125 I-labeled anti-mouse antibody was from Amersham Pharmacia Biotech, and [ 32 P]P i was from ICN. Other reagents and chemicals were the highest grade available.
Mutagenesis and Plasmid Construction-Mutant constructs p3U␤D380C/␥C87S and p3U␤ flag D380C/␥C87S were described previously, and these combined mutations have minimal effects on the F O F 1 function in vivo (normal phenotypic growth on succinate) or on ATPase activity of purified F 1 (11,12). The ⑀S108C mutation, originally created by Aggeler et al. (21), also has only minimal effects on the function of F O F 1 in vivo or of purified F 1 . Here a polymerase chain reaction-based, site-directed method (29) (Pfu DNA polymerase I) was used to introduce ⑀S108C into p3U, which expresses all 8 F O F 1 subunits. The antisense mutagenic primer, 5Ј-ATTAGCAGCTGTCACGGCGAC-3Ј, has a single base change that generates the ⑀S108C mutation and creates a PvuII restriction site. The product was digested with KpnI and NdeI and cloned into the corresponding sites of the p3U␤D380C/␥C87S vector to produce p3U␤D380C/␥C87S/⑀S108C. The mutated region of uncC was sequenced to confirm the presence of ⑀S108C and absence of any additional mutations. To express mutant F O F 1 , each construct was transformed into strain AN887, which has a Mu insertion that blocks expression of the chromosomal unc operon (30). Expression of p3U␤D380C/ ␥C87S/⑀S108C yields membranes with normal levels of DCCD-sensitive ATPase and purified F 1 with ATPase activity (35 mol⅐min Ϫ1 ⅐mg Ϫ1 ) comparable with that of wild-type F 1 .
Preparation of Hybrid F 1 and Reconstitution with F 1 -depleted Membranes-Aliquots of F 1 stock solutions were passed through centrifuge columns containing Sephadex G-50 -80 (34) equilibrated with MTKE buffer (20 mM Mops-Tris, 50 mM KCl, 0.1 mM EDTA, pH 8.0). The enzymes were diluted to 1 mg/ml with the same buffer and treated with 50 M DTNB for 40 min at 22°C. Excess DTNB was removed by column centrifugation. ␤-⑀ cross-linked ␤D380C/␥C87S/⑀S108C-F 1 and epitopelabeled ␤ flag D380C/␥C87S-F 1 were dissociated into subunits by repeated freezing and thawing in 50 mM MES, 1 M LiCl, 5 mM ATP, 0.5 mM EDTA, pH 6.1. The dissociated enzymes were mixed in a 1:1 ratio and allowed to reassemble as hybrid complexes as described previously (9,35). F 1 hybrid containing the ␤D380C-⑀S108C disulfide cross-link can contain ␤ flag D380C only in the two noncross-linked ␤ positions. F 1 containing the wild-type ⑀ subunit is incapable of forming a ␤-⑀ crosslink and is therefore silent in subsequent experiments. Hybrid F 1 (0.5 mg/ml) was rebound to F O in F 1 -depleted membranes (2 mg of protein/ ml) by incubation in TM buffer for 30 min at 30°C. Excess F 1 was removed by centrifugation at 100,000 ϫ g in a Beckman Airfuge for 1 min. The membrane pellet was resuspended and washed twice with TMG buffer (TM buffer containing 50 mM glucose) and finally resuspended in TMG buffer at about 4 mg of protein/ml.
For the DCCD-inhibited control, membranes were incubated with 0.5 mM DCCD and 5 mM MgCl 2 for 30 min at 22°C, conditions that give specific modification of c subunits in F O (36,37). Enzyme for the "noncross-linked" control was prepared as described for hybrid F 1 except that a ␤-⑀ cross-link was not formed prior to subunit dissociation by omitting DTNB treatment. Hence this form of F 1 can have ␤ flag at all three ␤ subunit positions.
Electrophoresis and Immunoblotting-SDS-PAGE was performed according to Laemmli (38) on 4 -15% gradient gels (Ready Gels, Bio-Rad). For nonreducing conditions, samples were denatured in the presence of 0.5 mM N-ethylmaleimide instead of 2-mercaptoethanol to block sulfhydryls without cleaving disulfides. Proteins were transferred from the gel to a polyvinylidene difluoride membrane (Novex) in a Bio-Rad Mini Trans-Blot cell for 1 h at 250 mA using a buffer containing 25 mM Tris, 192 mM glycine, 10% methanol, and 0.005% SDS (39). The blotted membrane was blocked for 1 h with 5% nonfat, dried milk in TBST (10 mM Tris-Cl, 150 mM NaCl, 0.05% Tween 20, pH 8.0), incubated for 2 h with anti-Flag M2 antibody (Eastman Kodak) at 0.4 mg/ml in TBST with 1% bovine serum albumin, 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 125 I-labeled anti-mouse secondary antibody (19 Ci/g, Amersham Pharmacia Biotech) in 10 ml of TBST with 1% bovine serum albumin, rinsed five times with TBST containing an additional 0.1 M NaCl, air dried, and exposed to a Phosphor Screen for 12 h. 125 I-Labeled bands were quantitated using a PhosphorImager (model 425E, Molecular Dynamics) and ImageQuant software.
Other Assays-The ATPase activity of soluble F 1 was measured at 30°C by a coupled enzyme assay (40) using 5 mM ATP, 2 mM MgCl 2 , and 0.5 g of F 1 /ml in MTK buffer (20 mM Mops-Tris, 50 mM KCl, pH 8.0). For measurement of ATP hydrolysis by membrane vesicles at 2-3 g of protein/ml, 5 M FCCP and 5 mM KCN were also added to prevent formation of a transmembrane proton gradient and to block NADH oxidation by the respiratory chain. The synthesis of ATP by E. coli membranes was determined by a coupled enzyme assay as described (13). Protein concentrations were determined by a modified Lowry assay (41).

RESULTS
Disulfide Bond Formation between ␤D380C and ⑀S108C in the Triple Mutant, ␤D380C/␥C87S/⑀S108C-F 1 -Our approach in testing for rotation of the ⑀ subunit requires the reversible formation of a specific covalent linkage between a ␤ subunit and the single copy of ⑀. Guided by previous cross-linking studies (24,42), we combined the ⑀S108C and ␤D380C mutations in a single construct. To prevent cross-linking between ␤ and ␥ subunits, the ␥C87S mutation was also included (11). As shown by nonreducing SDS-PAGE (Fig. 1), a rapid and near complete disappearance of ⑀ is accompanied by the appearance of a new band at 67 kDa when F 1 (lane 2 versus lane 1) or membrane-bound F O F 1 (lane 7 versus lane 6) containing these mutations is oxidized by DTNB. The high yield of cross-linked product correlates to a 90 -95% loss of ATPase activity, and the same results were obtained when samples were oxidized in the presence of Mg 2ϩ , MgATP, or MgADP/azide (data not shown). The apparent size of the cross-linked product (67 kDa) is consistent with the predicted molecular mass of 65 kDa for a 1:1 complex between ␤ and ⑀. Furthermore, immunoblotting confirmed the presence of both ⑀ and ␤ in the 67-kDa band (data not shown). As expected for a disulfide linkage, cross-linking and inactivation are fully reversed by brief exposure to dithiothreitol (lane 3). Finally, oxidation of F 1 lacking the ⑀S108C mutation shows no 67-kDa band nor does the ⑀ band disappear (lane 4). We conclude that the 67-kDa product results from formation of a specific disulfide cross-link between ␤D380C and ⑀S108C.
Exchange of ⑀ Subunits between Soluble F 1 Complexes-A second requirement for our F 1 -hybrid approach in testing for rotation of ⑀ is that ⑀ must not dissociate from F 1 during the course of the experiment. If this occurred, ⑀ could rebind in a manner that would allow it to cross-link to a different ␤ than the one with which it was originally aligned, thus giving a false indication of subunit rotation. This requirement presented a potential problem in using soluble F 1 because ⑀ is known to undergo reversible dissociation from the E. coli enzyme (43). Hence, the experiment presented in Fig. 2 was conducted to 1 The abbreviations used are: FCCP, carbonyl cyanide p-trifluoromethoxyphenylhydrazone; DCCD, N,NЈ-dicyclohexylcarbodiimide; DTNB, 5,5Ј-dithiobis(2-nitrobenzoate); DTT, dithiothreitol; PAGE, polyacrylamide gel electrophoresis; Mops, 4-morpholinepropanesulfonic acid; MES, 4-morpholineethanesulfonic acid. determine the rate of ⑀ subunit exchange between members of an F 1 population. The strategy behind this assay was to mix two different forms of the enzyme: one containing ⑀S108C and ␤ lacking the Flag epitope (␤D380C/␥C87S/⑀S108C-F 1 ), and the other containing epitope-tagged ␤ and wild-type ⑀ (␤ flag D380C/ ␥C87S-F 1 ). Because wild-type ⑀ cannot cross-link to ␤ flag , the only way that a ␤ flag -⑀ cross-link can form is if an ⑀S108C subunit dissociates from the triple mutant enzyme and rebinds to a Flag-tagged double mutant F 1 that has released its own wild-type ⑀. DTNB was added at the times indicated, and the Flag epitope in the cross-linked product was measured in the immunoblot shown in Fig. 2A. As expected, the amount of Flag in the ␤-⑀ band increased with time. The t1 ⁄2 for the exchange of ⑀ was found to be about 1 min (Fig. 2B). Because this is on the same time scale as our assay for subunit rotation, we conclude that the soluble E. coli enzyme is unsuitable for this test. In contrast, ⑀ does not dissociate from F O F 1 (as confirmed below in Fig. 3, lane 8) and in fact is required for binding F 1 to F O (44). Hence, tests for ⑀ subunit rotation were conducted with reconstituted hybrid F O F 1 .
Rotation of ⑀ Relative to the ␤ Subunits during ATP Hydrolysis by F O F 1 -Preliminary studies (not shown) confirmed that DTNB-treated ␤D380C/␥C87S/⑀S108C-F 1 can be dissociated into subunits, reassembled, and reconstituted with F O in F 1depleted membranes without breaking the ␤D380C-⑀S108C disulfide bond. Membrane-bound, ␤-⑀ cross-linked F 1 is catalytically inactive. However, treatment with DTT to reduce the disulfide cross-link restores ATP hydrolysis (9.5 mol⅐min Ϫ1 ⅐mg Ϫ1 ) and synthesis (30 nmol⅐min Ϫ1 ⅐mg Ϫ1 ) activities. Furthermore, when reconstituted membranes were preincubated with DCCD, the hydrolysis and synthesis of ATP were inhibited by 80 and Ͼ98%, respectively. These results demonstrate that cross-linked ␤D380C/␥C87S/⑀S108C-F 1 can rebind to F O to form an F O F 1 complex that is functionally coupled following reduction of the disulfide.
Using a subunit dissociation/reassociation procedure developed previously (9), we formed a ␤-⑀ cross-linked hybrid F 1 containing ␤ flag D380C subunits exclusively in the two noncross-linked ␤ positions. The resulting hybrid provides a means of distinguishing the ␤ subunit that is initially oriented to allow cross-linking to ⑀ from the other two ␤ subunits. Hybrid F 1 was rebound to F 1 -depleted membranes, and excess soluble F 1 was removed. To test for possible rotary movement of ⑀, the reconstituted membranes were briefly reduced with DTT, exposed to various ligands, and reoxidized with DTNB. In the absence of rotation, ⑀S108C would be expected to cross-link to the original unlabeled ␤ subunit. However, if ⑀ has rotated, ⑀S108C would be positioned to cross-link to Flag epitope-labeled ␤ in a significant fraction of the F O F 1 complexes resulting in the appearance of epitope in the ␤-⑀ band. The results are shown in the immunoblot in Fig. 3. When reconstituted membranes were reduced and exposed to conditions for ATP hydrolysis, a significant amount of Flag epitope was detected in the ␤-⑀ crosslinked band following reoxidation (Fig. 3, lane 4). In contrast, when reconstituted membranes were reduced and exposed to ligands that bind to catalytic sites but do not allow catalytic turnover, little Flag epitope was found in the ␤-⑀ bands (Fig. 3,  lanes 5 and 7). The amount of epitope in the ␤-⑀ band was also low when membranes were reduced and reoxidized in the presence of MgADP but absence of azide (not shown).

FIG. 2. Exchange of ⑀ subunits between soluble F 1 complexes.
A, ␤D380C/␥C87S/⑀S108C-F 1 and ␤ flag D380C/␥C87S-F 1 were diluted to 1 mg/ml with MTKE buffer, mixed in a 1:1 ratio, and incubated at 22°C. At the times indicated, 50 M DTNB was added, and after 10 min an aliquot containing 6 g of F 1 was denatured under nonreducing conditions and subjected to SDS-PAGE. For the lane marked Control, incubation was for 60 min without subsequent addition of DTNB. Immunoblotting and PhosphorImager detection of the Flag epitope were done as described under "Experimental Procedures." B, the amount of Flag in each ␤-⑀ band is plotted versus time of incubation. possibility that the ⑀ subunit can exchange between F O F 1 complexes. An additional control of this type showed that even when the initial ␤-⑀ disulfide was reduced for 5 min prior to the addition of ATP, the exchange of ⑀ between F O F 1 complexes was not detectable (data not shown). Taken together, the results provide compelling evidence for rotation of ⑀ relative to the ␤ subunits during catalytic turnover.
To determine the maximal level of Flag epitope expected in the ␤-⑀ band if ⑀S108C has an equal chance of reacting with any of the three ␤ subunits after catalytic turnover, a noncrosslinked control was run (Fig. 3, lane 3). In this case, ␤D380C/ ␥C87S/⑀S108C-F 1 was dissociated in the presence of a source of ␤ flag without forming an initial cross-link between ⑀ and one of the ␤ subunits. Consequently, during reassociation ␤ flag could assemble in each of the three ␤ subunit positions. When this enzyme was used to reconstitute F O F 1 , DTNB oxidation resulted in the highest level of ␤ flag observed in the ␤-⑀ band (Fig.  3, lane 3). This value was used to calculate the maximal expected ␤ flag in the ␤-⑀ band of hybrid F 1 if the orientation of ⑀ relative to the ␤ subunits is randomized by catalysis-driven subunit rotation (Fig. 4, 100%). Conditions for ATP hydrolysis yielded 65% of this maximal value (Fig. 4, MgATP). A value less than the calculated maximum is not surprising in view of the expectation that not all F O F 1 complexes in reconstituted membranes will be catalytically active during a brief exposure to MgATP. Of notable significance is the fact that very little ␤ flag appeared in the ␤-⑀ band in the absence of catalytic turnover (Fig. 4, 5-7%).
It is well known that covalent modification of one or more c subunits by DCCD blocks proton translocation through F O and inhibits both ATP synthesis and hydrolysis by F O F 1 (37). In addition, we recently reported that DCCD modification of F O prevents the catalysis-dependent rotation of the ␥ subunit in membrane-bound F O F 1 (12,13). A similar experiment was carried out here to test for functional coupling between ⑀ rotation in F 1 and proton conduction through F O . For this purpose, membranes were treated with DCCD under conditions that selectively modify the c subunits of F O . As shown in Fig. 3 (lane 6) and Fig. 4 (ϩDCCD), exposure of DCCD-inhibited membranes to MgATP yielded only 3% of the calculated maximal amount of Flag epitope in the ␤-⑀ band. This emphasizes the tight functional coupling of subunit rotation in F 1 to proton translocation through F O . DISCUSSION The results presented demonstrate catalysis-dependent rotation of the ⑀ subunit in functionally coupled membranebound F O F 1 . In view of earlier evidence for the rotation of ␥ in F 1 (9,14,15) and in F O F 1 (12,13), we conclude that ␥ and ⑀ constitute part of the rotor that couples proton transport through F O to the required binding changes in F 1 . The fact that DCCD modification of subunit c in F O prevents catalysis-dependent rotation of ␥ (12, 13) and ⑀ (Fig. 3, lane 6) supports the possibility that subunit rotation in F 1 is coupled to subunit rotation in F O . However, it remains to be determined whether the c subunit complex of F O constitutes the remaining portion of the rotor (8,9,14).
The intact E. coli F O F 1 complex was used in these studies to avoid two potential problems that might have been encountered with soluble F 1 . The first relates to the well established ability of ⑀ to inhibit catalysis when F 1 is separated from F O (43). This could hinder attempts to detect catalysis-dependent subunit rotation, particularly if ⑀ inhibits by preventing the rotation of ␥. A second difficulty could arise from the fact that ⑀ readily dissociates from F 1 . As noted earlier, when monitoring the orientation of ⑀ relative to the three ␤ subunits as a means of detecting subunit rotation, it is essential to rule out dissociation and rebinding of ⑀ as an alternative cause for its reorientation. The use of F O F 1 avoided both of these problems because in the native complex, ⑀ does not inhibit activity (45) nor does it dissociate from the complex (Fig. 3, lane 8).  For lanes 1, 2, and 4 -7, ␤-⑀ cross-linked hybrid F 1 having Flag-tagged ␤ exclusively in the two noncross-linked ␤ positions was prepared and reconstituted with F 1 -depleted membranes as described under "Experimental Procedures." Membranes were suspended at 4 mg of protein/ml in buffer containing 50 mM Tris acetate, 5 mM MgSO 4 , 50 mM glucose, 5 M FCCP, pH 7.5 (TMGF buffer) and subjected to the following treatments. Lane 5, membranes were incubated with 10 mM DTT for 30 s, passed through a centrifuge column equilibrated with TMGF buffer, and collected in a tube containing DTNB (50 M final), and incubated for 10 min at 22°C. Lanes 4 and 7, same as for lane 5 except that the incubation mixture and centrifuge column buffer also contained 1 mM ATP (lane 4) or 0.5 mM ADP and 0.5 mM NaN 3 (lane 7). Lane 1, same as for lane 5 except that DTT was omitted. Lane 2, same as for lane 5 except that the column effluent was collected in the absence of DTNB. Lane 6, same as for lane 4 except that membranes were pretreated with DCCD to modify F O as described under "Experimental Procedures." Lane 3, same as for lane 4 except that F O F 1 was reconstituted using F 1 that could have ␤ flag at all three ␤ positions (see "Experimental Procedures"). Lane 8, as for lane 4 except that F 1 -depleted membranes were reconstituted separately with ␤D380C/␥C87S/ ⑀S108C-F 1 or ␤ flag D380C/␥C87S-F 1 and then mixed in a 1:1 ratio. Hexokinase was included at 3 units/ml in samples 1, 2, 5, and 7 to prevent any potential ATP hydrolysis. The uncoupler FCCP was present in all samples to prevent formation of a transmembrane proton gradient. An aliquot of each sample containing 6 g of membrane protein was subjected to SDS-PAGE under nonreducing conditions, and immunoblotting was performed as described under "Experimental Procedures." The results shown are typical of three replicate experiments.  Fig. 3 was determined as described under "Experimental Procedures." The value obtained for a noncross-linked control (Fig. 3, lane 3) was multiplied by 0.8 to correct for its greater ␤ flag content compared with the cross-linked hybrid F 1 (Fig. 3, lanes 4 -7). This value was set to 100% representing an equivalent probability that ⑀ will be oriented to allow cross-linking to any of the three ␤ subunits following catalytic turnover.
The t1 ⁄2 for exchange of ⑀ between ␤D380C/␥C87S/⑀S108C-F 1 and ␤ flag D380C/␥C87S-F 1 molecules was found to be about 1 min (Fig. 2), whereas at the same temperature, the t1 ⁄2 for dissociation of ⑀ from our wild-type F 1 is about 5 min. 2 This suggests that these mutations weaken the interaction of ⑀ with F 1 resulting in an increased dissociation rate. This is not surprising in view of the fact that ␤D380 is in close contact with ␥C87 and ⑀S108 as evidenced by the facility with which substituted cysteines can cross-link. The results imply that one or more of the native residues contributes to the stability of the interaction of ⑀ with the rest of F 1 .