Defining the Domain of Binding of F1 Subunit ε with the Polar Loop of F0 Subunit c in theEscherichia coli ATP Synthase*

We have previously shown that the E31C-substituted ε subunit of F1 can be cross-linked by disulfide bond formation to the Q42C-substitutedc subunit of F0 in the Escherichia coli F1F0-ATP synthase complex (Zhang, Y., and Fillingame, R. H. (1995) J. Biol. Chem.270, 24609–24614). The interactions of subunits ε and care thought to be central to the coupling of H+ transport through F0 to ATP synthesis in F1. To further define the domains of interaction, we have introduced additional Cys into subunit ε and subunit c and tested for cross-link formation following sulfhydryl oxidation. The results show that Cys, in a continuous stretch of residues 26–33 in subunit ε, can be cross-linked to Cys at positions 40, 42, and 44 in the polar loop region of subunit c. The results are interpreted, and the subunit interaction is modeled using the NMR and x-ray diffraction structures of the monomeric subunits together with information on the packing arrangement of subunit c in a ring of 12 subunits. In the model, residues 26–33 form a turn of antiparallel β-sheet which packs between the polar loop regions of adjacent subunitc at the cytoplasmic surface of thec 12 oligomer.

The H ϩ -transporting, F 1 F 0 -ATP synthase of Escherichia coli utilizes an H ϩ electrochemical gradient to drive ATP synthesis during oxidative phosphorylation (1). Similar enzymes are found in mitochondria, chloroplasts, and other bacteria. The enzymes are composed of two sectors, termed F 1 and F 0 . The F 1 sector contains the catalytic sites for ATP synthesis, and when released from membrane, it shows ATPase activity. The F 0 sector traverses the membrane and functions as the H ϩ transporter. When F 1 is bound to F 0 , the complex acts as a reversible, H ϩ -transporting ATP synthase or ATPase. In E. coli, F 1 is composed of five types of subunits in an ␣ 3 ␤ 3 ␥␦⑀ stoichiometry, and F 0 is composed of three types of subunits in an a 1 b 2 c 12 stoichiometry (2)(3)(4). The structure of much of the ␣ 3 ␤ 3 ␥ portion of F 1 has been solved by x-ray diffraction analysis and shows subunit ␥ extending through the center of a hexamer of the larger, alternating ␣ and ␤ subunits (5,6). During catalysis, the ␥ and ⑀ subunits have been shown to rotate in 120°steps between the three alternating catalytic sites in the ␤ subunits (7)(8)(9)(10)(11)(12)(13). Subunits ␥ and ⑀ are thought to rotate as a unit because they can be cross-linked to each other with minimal inhibitory effects on ATPase activity (14,15).
The relation of structure and mechanism in F 0 is less thoroughly understood. The largely hydrophobic subunit a folds in the membrane with five transmembrane helices (16,17), at least two of which likely interact with subunit c during proton transport (18 -20). Subunit b is anchored in the membrane via a single transmembrane helix that is connected to a polar, elongated cytoplasmic domain that is thought to play a key role in fixing F 1 to F 0 (21). Subunit c is a protein of 79-amino acid residues that folds in the membrane in a hairpin-like structure. The two hydrophobic transmembrane ␣-helices are joined by a more polar loop region that is exposed to the F 1 binding side of the membrane. Aspartyl 61, lying at the center of transmembrane helix-2, is thought to be the site of H ϩ binding during transport (22). The polar loop was proposed to play a key role in coupling H ϩ transport to ATP synthesis or hydrolysis based upon the uncoupled phenotypes of mutants with substitutions in the conserved Arg 41 -Gln 42 -Pro 43 sequence of the polar loop (22,23). The "uncoupled" phenotype of the cQ42E mutant proved to be suppressed by second site substitutions in Glu 31 of F 1 subunit ⑀ (24), and this led to cross-linking studies demonstrating a physical proximity between the polar loop and subunits ␥ and ⑀ of F 1 (14,25,26). A recently determined NMR structure of monomeric subunit c conforms well with folding predictions made from biochemical and genetic analysis (27).
Recent experiments now indicate that the c 12 oligomer of F 0 is organized in a ring with transmembrane helix-1 on the inside and transmembrane helix-2 on the outside (2,4,28) and with the a and b subunits associating at the periphery of the ring (2,20). Such an arrangement is consistent with low resolution electron and atomic force microscopic images (29 -31). The structural data fit well with rotary models where H ϩ transport at the a-c interface is proposed to drive rotation of the oligomeric c ring as the Asp 61 carboxylate is protonated and deprotonated from alternate access channels on each side of the membrane (10,(32)(33)(34). The rotation of subunit c is proposed to drive the rotation of the ␥⑀ unit in F 1 via a fixed linkage between subunit c and the ␥ and ⑀ subunits (14) although other explanations have been proposed (22,25). The elongated cytoplasmic domain of subunit b is thought to extend from the membrane surface to the top of F 1 as a "second stalk", or stator, to hold F 1 fixed as subunit ␥ rotates at the center of the molecule (21,(35)(36)(37).
In this study, the interacting regions of the polar loop of subunits c and subunit ⑀ are more thoroughly defined by disulfide cross-linking of Cys introduced into the two subunits. The experimental design and interpretation was aided by a structural model for subunit ⑀ derived by NMR (38) and x-ray crystallography (39). The Cys residues of subunit ⑀ that form cross-links with subunit c localize to a span of residues 26 -33 which fold as two strands of antiparallel ␤-sheet connected by a loop. A structural model for the subunit-subunit interaction is developed from the structures of the monomeric subunits and information on the organization of the c-oligomer, using distance constraints from the cross-linking data reported here. The model indicates that the segment of antiparallel ␤-sheet encompassing residues 26 -33 of subunit ⑀ packs in a space between neighboring polar loops of the subunit c oligomer.

EXPERIMENTAL PROCEDURES
Mutant Construction and Expression-The plasmids constructed in this study are derivatives of plasmid pYZ201 (25), which carries the eight structural genes of unc operon coding F 1 F 0 (bases 870 -10172; Ref. 40). 1 The Cys substitutions in subunit c and subunit ⑀ were introduced by oligonucleotide-directed mutagenesis using the strategy described previously (25). The plasmids were expressed in strain OM204 (41), a strain in which the unc operon is deleted from the chromosome.
Membrane Preparations and Cross-link Analysis-Cells were grown and membranes prepared by the methods described (25). To catalyze disulfide bond formation, aliquots of 200-l membrane vesicles at 10 mg/ml in TMG buffer (50 mM Tris-HCl, pH 7.5, 5 mM MgCl 2 , 10% (v/v) glycerol) were treated with 20 l of a mixture of 15 mM CuSO 4 and 45 mM 1,10-phenanthroline in 50% ethanol. After a 1-h incubation at room temperature, the reaction was stopped by addition of 20 l of 0.5 M Na 2 EDTA and 20 l of 0.5 M N-ethylmaleimide (NEM) 2 in dimethyl sulfoxide, and the sample was incubated for a further 60 min. The sample was then diluted with 230 l of 2ϫ SDS sample buffer (125 mM Tris-HCl, pH 6.8, 4% SDS, 20% glycerol, and 0.02% bromphenol blue) and incubated at 30°C for 1 h. The solubilized membrane proteins were separated by SDS-polyacrylamide gel electrophoresis using a 7.5-15% acrylamide gradient and the Tris-Tricine buffer described by Schä gger and von Jagow (42). After electrophoretic transfer to nitrocellulose paper (43), immunostaining was carried out using the Enhanced Chemiluminescence System (Amersham Pharmacia Biotech). The rabbit antiserum to subunit c used was that described by Girvin et al. (44). Antibodies that nonspecifically cross-reacted with E. coli membrane proteins were removed by preabsorption with membranes prepared from a mutant strain with a deleted unc operon (44). The mouse monoclonal antibody to subunit ⑀ (13-A7, ⑀II; Ref. 45) was a gift from Dr. R. Capaldi (University of Oregon, Eugene, OR).
Structural Modeling of Subunit-Subunit Interaction-A model for subunit c interaction in the c 12 oligomer has been derived from the NMR model (27) using distance constraints derived from the cross-linking data of Jones et al. (28). 3 A subunit c dimer, taken from the oligomer model, was manually docked to the N-terminal domain of subunit ⑀ (residues 1-87) so that the distances between the ␣-carbons of crosslinked residues were Ͻ12 Å. The range for ␣-carbon distances in naturally occurring disulfide bonds in proteins is 4 -7.5 Å (46). A somewhat wider distance constraint range of 4 -11 Å was used in the molecular mechanics calculations done here to allow for possible thermal motions and distortions of structure on cross-link formation. The shortest of the two distances between the Cys ␣-carbon in each of the two subunits c and the Cys ␣-carbon in ⑀ in the manually docked structure was used to impose a distance constraint. The positions of all the atoms of the c subunits were fixed except for residues 39 -45, which were left unrestrained. The backbone angles in subunit ⑀ were restrained to their value in the x-ray structure (39) using quadratic restraints. Energy minimization was performed with CVFF (constant valence force field) using the steepest descents and then conjugated gradient methods as implemented in DISCOVER 3.0 (Molecular Simulations Inc.) until the maximum derivative was below 0.1 kcal mol Ϫ1 Å Ϫ1 . 4

Properties of Cys-substituted Double
Mutants-All of the mutants constructed grew on succinate minimal medium, which indicates formation of a functional ATP synthase ( Table I). Two of the ⑀-substituted, cQ42C mutants showed very little membrane ATPase activity and nondetectable amounts of ⑀ subunit on immunoblotting (Table I). In these two cases, ⑀V25C/cQ42C and ⑀G27C/cQ42C, we conclude that the F 1 -F 0 interaction is probably stable under in vivo conditions but that the F 1 -ATPase and ⑀ subunit disassociate from the membrane during membrane preparation.
Cross-linking of Cysteine in Various ⑀-Substituted/cQ42C Mutants-We had previously shown that Cys at position 31 in subunit ⑀ cross-links with Cys in positions 40, 42, or 43 in the polar loop of subunit c (25). In this study, Cys was substituted in a series of positions proximal to position 31 in subunit ⑀, and cross-link formation was tested with Cys at position 42 in subunit c. An experiment comparing cross-linking in cQ42C/ ⑀E29C and cQ42C/⑀E31C mutant membranes illustrates a number of typical features (Fig. 1). An ⑀-c cross-linked product was observed in membranes prepared from both mutants. The cross-linked product identified as ⑀-c was found at an identical position on blots stained with either anti-⑀ or anti-c antibody. 5 Comigration of the product on the two types of blots was confirmed using several types of gels of varying acrylamide concentration. The ⑀-c product seen in untreated membranes is presumed to form by autoxidation as the membranes are isolated from the cell. The extent of cross-link formation was enhanced by Cu(II)(phenanthroline) 2 (CuP) catalyzed oxidation. Cross-linking was largely reversed by subsequent treatment with 0.5 mM dithiothreitol.
The intensity of the ⑀-c-immunostained product that is detected with anti-⑀ antibody does prove to be misleading. Beginning with the anti-⑀ blot shown in Fig. 1, note the much greater intensity of staining of the ⑀-c product in cQ42C/⑀E29C versus cQ42C/⑀E31C membranes despite the loading of equal amounts of membrane protein in all lanes. Note also for the ⑀E29C membranes that the changes in intensity of ⑀-c are considerably greater than the changes in intensity of monomeric ⑀, i.e. the changes are not in the inverse proportions expected in a precursor-product relationship. We interpret this to mean that the ⑀-c heterodimer of the ⑀E29C mutant protein binds antibody better than the ⑀E29C monomer and also better than the ⑀-c heterodimer of the ⑀E31C mutant. This interpretation is 1 The unc DNA numbering system corresponds to that used by Walker et al. (40). 2 The abbreviations used are: NEM, N-ethylmaleimide; CuP, Cu(II) (phenanthroline) 2 ; Tricine, N-tris(hydroxymethyl)methylglycine. 3 O. Y. Dmitriev, M. E. Girvin, P. C. Jones, and R. H. Fillingame, submitted for publication. 4 A coordinate file of the final model is available by E-mail at dmitriev@iris.bmolchem.wisc.edu. 5 A number of minor bands were also detected with the anti-⑀ antibody, the intensity and number of which vary from experiment to experiment. They appear to be artifactual because similar bands are detected in mutant membranes lacking subunit ⑀. Most of the proteins reacting nonspecifically are removed by the stripping procedure used to remove F 1 , i.e. washing membranes with 1 mM Tris-HCl and 0.5 mM EDTA, but this treatment also removes monomeric ⑀ (see Fig. 2). qualitatively confirmed by the relative intensities of the bands on the anti-c blot. In the untreated membrane samples, the intensity of the anti-c immunostained ⑀-c product is much greater for ⑀E31C than for ⑀E29C, i.e. just the reverse of the pattern seen with anti-⑀ antibody. We have concluded that the best way to approximate the extent of ⑀-c formation may be to compare the amounts of monomeric ⑀ remaining after various treatments. Using this criteria, the ⑀E31C mutant would appear to form at least as much and possibly more ⑀-c product by either autoxidation or CuP-catalyzed oxidation.
Other mutants also show ⑀-c products whose staining intensities differ considerably with the two antibodies. In the experiment shown in Fig. 2, membranes were washed with 1 mM Tris-HCl, pH 8.0, 0.5 mM Na 2 EDTA, a procedure which removes F 1 and uncross-linked subunit ⑀ and other nonspecific immunoreactive proteins. The ⑀S28C, ⑀E29C and ⑀G30C ⑀-c products stain much more intensely with anti-⑀ antibody than with anti-c antibody. Conversely, the ⑀-c product in the ⑀G33C mutant stains much less intensely with anti-⑀ body than with anti-c antibody. Other ⑀-c products stain with nearly equivalent intensities in the two blots. In summary, the blots do not provide a good quantitative means of distinguishing the extent of cross-link formation in different mutant membranes.
A survey of ⑀-c cross-linking in a series of double-Cys-substituted pairs is shown in Fig. 3. In the presence of CuP, most or all of the Cys-substituted subunit ⑀ was cross-linked with Cys 42 -substituted subunit c when Cys was substituted as positions 26, 28, 29, 30, 32, and 33 in subunit ⑀. Detectable crosslinking also occurred in the absence of CuP in each of these mutants. No cross-linking was observed with Cys substituted at positions 24, 25, 27, 34 and 38 of subunit ⑀. In the case of the ⑀V25C mutant, subunit ⑀ was not incorporated into the membrane. This was also the case for the ⑀G27C mutant (not shown). In other experiments, the cQ42C/⑀S10C pair was also shown to not form a cross-link (Table I). Most of the ⑀-c-crosslinked product formed in the various membranes shown in Fig.  3 was reduced by treatment with dithiothreitol. The results from several similar experiments are summarized in Table I.  Fig. 4, Cys substituted at either position 40 or 42 was readily cross-linked to Cys at positions 28, 31, and 32 of subunit ⑀. We conclude that subunit ⑀ must be able to interact with either face of the loop region when it binds to the oligomer of subunit c. DISCUSSION The surface of subunit ⑀ lying proximal to subunit c has been mapped by cross-linking experiments to a region encompassing residues 26 -33, which in the NMR and x-ray diffraction structures of subunit ⑀ (38, 39) reside in a loop of antiparallel ␤-sheet (Fig. 5A). The NMR and x-ray diffraction structures agree closely and show subunit ⑀ to be a protein of two distinct domains. The N-terminal domain of 86 residues folds in a FIG. 2. CuP-catalyzed cross-linking of Cys 42 in subunit c with Cys at various positions in subunit ⑀ analyzed by immunoblots of "stripped" membranes. The cross-linking reaction was carried out with whole membranes, and the reaction was quenched with EDTA and NEM. Membranes were then stripped of F 1 by incubation in 1 mM Tris-HCl, pH 8, 0.5 mM EDTA, 10% (v/v) glycerol, and the stripped membranes were collected by centrifugation and solubilized in SDS sample buffer. Immunoblots prepared with anti-⑀ and anti-c antibodies are marked as described for Fig. 1. Numbers at the top indicate the position of the Cys in subunit ⑀.

FIG. 1. Immunoblots showing cross-link formation between subunits c and ⑀ in membrane vesicles of Cys-substituted mutants.
Mutations are indicated by referring to the positions of the Cys residues of subunit c and subunit ⑀. Membranes, prepared in the absence of dithiothreitol, were either treated with CuP (ϩCu) or not treated (0Cu). Following quenching of the reaction with EDTA and NEM, the treated or untreated membrane vesicles were centrifuged and resuspended in TMG buffer containing (ϩDTT) or lacking (0DTT) 0.5 mM dithiothreitol and incubated for 30 min at 22 C. Following solubilization with SDS and electrophoresis of 50-g samples, acrylamide gels were blotted to nitrocellulose paper and probed with anti-subunit ⑀ or anti-subunit c antibodies. The positions of subunit ⑀, the ⑀-c dimer, and subunit c monomer, dimer (c 2 ), trimer (c 3 ), and tetramer (c 4 ) are indicated. Some of the dimers, trimers, and tetramers of subunit c result from incomplete disaggregation of the c 12 oligomer in SDS sample buffer. Most of the staining seen at the position of c 4 is because of an immunoartifact in the membrane. The positions of molecular mass markers, with the molecular mass given in kDa, are shown at the side of the blots. Note that subunit c electrophoreses anomalously relative to these markers.
10-stranded ␤-sandwich and the C-terminal domain of 45 residues is formed from two ␣-helices arranged in an antiparallel coiled coil. Much of the C-terminal domain appears to be nonessential because it can be deleted without effect on ATP synthase function (47). Cross-linking and chemical modification experiments indicate that the surface of the ␤-sandwich, including residues His 38 and Ser 10 , neighbor the ␥ subunit and that His 38 lies close to the surface of F 0 (14,38,39,48). Residue 31 of ⑀ must also lie proximal to the surface of F 0 because the ⑀E31C-substituted protein can be cross-linked to Cys at positions 40, 42, and 43 of subunit c (25). The loop including residues 26 -33 protrudes from the "bottom" of subunit ⑀ as a well defined lobe (Fig. 5A), and most Cys replacements in this loop were cross-linked to Q42C subunit c in the experiments The interacting surfaces of subunit ⑀ and subunit c have been modeled beginning with a model for the c 12 oligomer described elsewhere. 3 In the modeling, equivalent distance constraints were imposed for each cross-link formed because of difficulties in quantitatively distinguishing the extent of cross-link formation. In the model, the loop of antiparallel ␤-sheet that is centered around ⑀Glu 29 packs between the polar loops of two c subunits (Fig. 6A). The model also depicts the ⑀Glu 31 residue lying close enough to the conserved and essential cArg 41 residue to interact electrostatically and also close to cGln 42 (Fig.  6B). The positioning of these side chains in the model should be interpreted with caution because the model is derived without use of side chain distance constraints. However, with these precautions, the general proximity of residues in the model does provide a reasonable explanation for the uncoupling effects of the cQ42E mutation (23) in that charge-charge repulsion would be expected between the cGlu 42 and ⑀Glu 31 carboxylates. The charge-charge repulsion explanation is supported by the differences in pH dependence of neutral versus positively charged ⑀-31 suppressor substitutions in restoring function to the cQ42E mutant (24). The smaller uncoupling effects of some substitutions in ⑀Glu 31 (49), versus the cQ42E mutation, are not as easily rationalized by the model. Immunoblots of whole membranes are shown after probing with anti-⑀ antibody. Whole membranes, prepared in the absence of dithiothreitol, were treated with CuP (ϩ) or not treated (0), and the reaction was quenched with EDTA and NEM. Cross-linked products in CuP-treated membranes were reduced with dithiothreitol (ϩDTT) as described in Fig. 3. In the model shown in Fig. 6A, it is notable that the space between loops of subunit c is essentially filled by the packing with subunit ⑀. Because residue 205 of subunit ␥ is also known to cross-link with residues 42, 43, and 44 of subunit c (14,26), it seems likely that the region around ␥-205 packs between a different set of c subunits than those interacting with ⑀. The shielding of two separate pairs of subunit c by the binding of subunits ␥ and ⑀, respectively, may explain the observations of Watts and Capaldi (50) on the functional effects of NEM modification of Cys 42 in cQ42C mutant membranes. Function was retained during the initial phase of modification of approximately 60% of the subunits and then lost during modification of the last 40%. The inhibitory phase of NEM modification may correspond to reaction with Cys 42 at the c-␥-c or c-⑀-c interfaces. In the currently envisioned rotary models, subunits ␥ and ⑀ remain fixed to a set of c subunits and turn as the c-oligomer rotates (14,33,34). The binding interaction must be of sufficient strength to withstand a considerable torque, estimated at exceeding 40 pN nm under load, or approximately 12 kcal mol Ϫ1 for each 120°step in which ATP is synthesized (8,34,(51)(52)(53). From the structural model deduced here, it seems likely that the binding energy is derived from the combined interaction of subunit ⑀ with the loop region of one set of subunit c and of subunit ␥ with an adjacent set of subunit c.