Individual Interactions of the b Subunits within the Stator of the Escherichia coli ATP Synthase*

Background: The peripheral stator stalk of Escherichia coli ATP synthase contains two b subunits. Results: Using disulfide bond formation, one b subunit was cross-linked to a, α, and δ and the other to β. Conclusion: The b subunits adopt distinct positions within the stator to generate stability. Significance: The different positions imply different roles in counteracting the torque generated by the rotor. FOF1 ATP synthases are rotary nanomotors that couple proton translocation across biological membranes to the synthesis/hydrolysis of ATP. During catalysis, the peripheral stalk, composed of two b subunits and subunit δ in Escherichia coli, counteracts the torque generated by the rotation of the central stalk. Here we characterize individual interactions of the b subunits within the stator by use of monoclonal antibodies and nearest neighbor analyses via intersubunit disulfide bond formation. Antibody binding studies revealed that the C-terminal region of one of the two b subunits is principally involved in the binding of subunit δ, whereas the other one is accessible to antibody binding without impact on the function of FOF1. Individually substituted cysteine pairs suitable for disulfide cross-linking between the b subunits and the other stator subunits (b-α, b-β, b-δ, and b-a) were screened and combined with each other to discriminate between the two b subunits (i.e. bI and bII). The results show the b dimer to be located at a non-catalytic α/β cleft, with bI close to subunit α, whereas bII is proximal to subunit β. Furthermore, bI can be linked to subunit δ as well as to subunit a. Among the subcomplexes formed were a-bI-α, bII-β, α-bI-bII-β, and a-bI-δ. Taken together, the data obtained define the different positions of the two b subunits at a non-catalytic interface and imply that each b subunit has a different role in generating stability within the stator. We suggest that bI is functionally related to the single b subunit present in mitochondrial ATP synthase.

F O F 1 ATP synthases utilize the energy of an electrochemical ion gradient (⌬ Hϩ or ⌬ Naϩ ) across biological membranes to catalyze the synthesis of ATP from ADP and inorganic phosphate. In many bacteria, the enzyme can also work in the reverse direction, generating a proton or Na ϩ motive force by hydrolysis of ATP. ATP synthases are rotary nanomachines that couple the translocation of ions in F O to ATP synthesis/ hydrolysis within the catalytic F 1 part. The flow of H ϩ or Na ϩ through two half channels within subunit a drives the rotation of the subunit c ring in F O as well as of the elongated ␥⑀ central stalk in F 1 . Subunit ␥ rotates inside a molecular bearing composed of the alternately arranged ␣ 3 ␤ 3 hexamer and generates cyclic conformational changes within the three catalytic nucleotide binding sites because of its eccentric rotation, thereby promoting ATP synthesis and its release. To counteract the tendency of the ␣ 3 ␤ 3 hexamer to follow the rotation of the rotor, a peripheral stalk, composed of a b dimer and subunit ␦ in most bacterial enzymes, is essential to hold the ␣ 3 ␤ 3 hexamer in position (1)(2)(3).
Peripheral stalks are present in all three evolutionarily related types of rotary ATPases. Interestingly, F O F 1 ATP synthases (F-type ATPases) contain only one peripheral stalk. A-type ATPases, which function primarily as ATP synthases but are evolutionarily more closely related to V-type ATPases, have two peripheral stalks, whereas eukaryotic vacuolar V-type ATPases working as ion pumps contain three peripheral stalks per enzyme complex (4,5). Furthermore, although each peripheral stalk of A-type as well as V-type ATPases studied so far is composed of a 1:1 heterodimer of nonhomologous subunits E and G (6 -11), the single peripheral stalk of F O F 1 ATP synthases shows variations in subunit composition dependent on the organism studied (12)(13)(14)(15)(16)(17)(18).
In mitochondrial F O F 1 , the peripheral stalk contains a single b subunit with two transmembrane helices together with the additional subunits d, F 6 , and OSCP 5 , the latter being homologous to bacterial subunit ␦. Structures of bovine mitochondrial stalk subcomplexes revealed that the hydrophilic region of sub-unit b forms a continuous, slightly curved ␣-helix that is stiffened by surrounding shorter helical stretches of subunits d and F 6 . The binding between OSCP and the C-terminal region of subunit b is strengthened by extensive ␣-helical interactions (17,18). In contrast, in chloroplasts and some bacteria, two different b-like proteins, named subunits I/II and subunits b/bЈ, respectively, are encoded, and both have been shown to be present in ATP synthase complexes purified, e.g., from spinach chloroplasts, Rhodobacter capsulatus, and the hyperthermophilic bacterium Aquifex aeolicus (12,14,16). Furthermore, heterodimer formation has been demonstrated for the soluble domains of subunits b and bЈ of the cyanobacterium Synechocystis PCC6803 (13) and for the chimeric b and bЈ constructs of Escherichia coli and Thermosynechococcus elongatus functionally assembled into E. coli ATP synthase (19,20).
In E. coli as well as most bacteria, the peripheral stalk contains a dimer of identical b subunits, each with a single transmembrane helix and a soluble domain extending from the membrane to the top of F 1 (15,21,22). Four distinct functional domains have been defined for the b dimer (23), starting at the N terminus with a mainly ␣-helical transmembrane domain (b  anchoring the protein in the membrane by strong direct interactions with subunit a (24,25). The tether domain (b  spanning the surface of the membrane and the region beneath the F 1 part appears to be a highly flexible region. Deletions of up to 11 amino acids or insertions of up to 14 amino acids can be positioned in this domain without loss of function (26,27). Furthermore, functional F O F 1 complexes can contain dimers of b subunits differing in the length of this region (28). Cross-linking data suggest that the dimerization domain (b  ) extends along the F 1 part at one of the non-catalytic (␣/␤) clefts (29,22). The C-terminal ␦-binding domain (b 123-156 ) is essential for the binding of F 1 . A disulfide cross-link can be formed between b158C and ␦M158C (30,31). Information about the structure of E. coli subunit b has been obtained from the analyses of peptides representing segments of subunit b, in particular of the membrane domain (fragment b , the tether domain (fragment b 30 -82 ), the dimerization domain (fragment b , and a part of the ␦-binding domain (b 140 -156 ). In each case, only monomeric subunit b segments yielded to structural analysis, revealing an extended, highly ␣-helical configuration (24,(32)(33)(34). A low-resolution structure of the dimeric, soluble b  domain exhibits a 160-Å-long, slightly curved shape (33) that resembles the density of the peripheral stalk observed in electron microscopy projections of F O F 1 (21). Whether the two b subunits are present in a staggered or in an in-register conformation and whether in a right-handed coiled coil, like in A-type ATPases (9,10), or in a left-handed coiled coil, which is observed in a great number of proteins (35), is still under discussion (20, 36 -38).
A number of studies have shown that in the E. coli enzyme, the two b subunits interact with, or are proximal to, subunits a, c, ␣, ␤, and ␦ (25, 29 -31, 39, 40). Because ATP synthase complexes do not have 2-fold symmetry, the two b subunits must occupy dissimilar positions within the enzyme and have different subunit-subunit interactions, but identifying features that distinguish the two positions has been difficult because the subunits themselves are identical, encoded by the same gene. How-ever, genetic complementation between two differently defective b subunits supports the view that each b subunit contributes uniquely to the function of the peripheral stalk (41). In addition, the use of two differently shortened C-terminal hydrophilic forms of subunit b (b syn ) suggested that the two b subunits have different roles in binding of F 1 (37).
In this study, we determined individual interactions of both b subunits within the stator of assembled F O F 1 by use of antibody binding studies as well as disulfide cross-linking, applying two or three different cysteine pairs at a time. Within region b 92-118 , a part of the b dimer known to be associated with the side of F 1 (29,42), one b subunit, which we designated b I, is proximal to subunit ␣, whereas the other one (b II ) can be linked to subunit ␤. They can also be distinguished by tagging the b I subunit with subunit a via the cross-linking pair bL16C/aV239C. Furthermore, the C-terminal region of only b I formed disulfides with subunit ␦, whereas the corresponding region of b II is accessible for antibody binding. The results clearly demonstrate that, although identical in sequence, the two b subunits have different positions and competences within the stalk. We suggest that subunit b I is functionally related to the single subunit b present in mitochondrial ATP synthase, whereas subunit b II may have a more stabilizing function like mitochondrial subunits d and F 6 .
Construction of Mutants-All plasmids were derived from plasmid pBWU13 (atpBEFHAGDC) (47). The endogenous cysteine residues within the subunits under investigation were changed to alanine using plasmid pFV2 as template or source for cloning (48). Cysteine substitutions were introduced by a two-step PCR overlap extension method using two mutagenic oligonucleotides containing the codon change necessary and two wild-type primers with the corresponding restriction sites (49). PCR products were directly transferred into the suitable pBWU13 derivative using unique sites for cloning (49). All mutations were verified by DNA sequencing of the cloned fragment through the ligation sites.
Preparative Procedures-F O subunits a, b, and c (43), F 1 (50), F O (43), and F O F 1 (51) were purified as described. mAbs against subunit b were prepared as described using purified subunit b as antigen (43). Antibodies from culture supernatants were puri-fied by caprylic acid and ammonium sulfate precipitation according to the method described by McKinney and Parkinson (52) and stored as lyophilized powder. Inverted membrane vesicles with and without F 1 were prepared from E. coli strains ML308-225 and CM1470 (53). Binding of F 1 to F 1 -stripped inverted membrane vesicles of strain ML308-225 and the influence of the anti-subunit b mAbs on the function of the ATP synthase was studied as described by Deckers-Hebestreit et al. (53).
Binding of mAbs to purified F O F 1 was performed as follows. First, F O F 1 and mAbs were independently purified on a Superdex SPX-200 gel filtration column to remove aggregates and, in the case of F O F 1 , incomplete protein complexes using 20 mM Tris-HCl (pH 7.8), 1 mM MgCl 2 , 0.2 mM phenylmethyl-sulfonyl fluoride, and 0.02% (w/v) n-dodecyl-␤-D-maltoside as buffer system. Homogeneous fractions were pooled and concentrated by ultrafiltration (exclusion size, 100 kDa; Millipore) to a final concentration of 30 M. Equimolar amounts of F O F 1 and mAbs were mixed and incubated for 1 h at 37°C. Subsequently, the mixture was separated by gel filtration using the same buffer system and analyzed by SDS-PAGE as well as immunoblotting for antibody binding.
Cross-linking with Copper-1,10-phenanthroline-Inverted membrane vesicles were prepared according to Krebstakies et al. (54) using TMG buffer (50 mM Tris-HCl (pH 7.5), 10 mM MgCl 2 , 10% (v/v) glycerol). In each case, the mutant strains used for membrane preparation were named according to the cysteine substitutions present within the F O F 1 subunits. For cross-linking experiments, membranes were adjusted to a final protein concentration of 5 mg/ml with TMG (pH 8.2) (or TMG (pH 7.5), as stated in the corresponding figure legends). The CuP stock solution was prepared in TMG (pH 8.2 or pH 7.5) and added to the samples to provide 1.5 mM CuSO 4 and 4.5 mM o-phenanthroline as the final concentrations. The cross-linking reaction was incubated for 1 h at 23°C and stopped by addition of 50 mM EDTA-KOH (pH 8.0) (500 mM stock solution) and 50 mM N-ethylmaleimide (500 mM stock solution in dimethyl sulfoxide), while incubating the samples for another 15 min at 23°C. An equal volume of double-concentrated sample loading buffer without 2-mercaptoethanol was added to solubilize the membranes and incubated for 15 min at 23°C (24,49).
Assays-Protein concentrations were determined by the method of Dulley and Grieve (55) or with the bicinchoninic acid protein assay as recommended by the supplier (Pierce). Proteins were separated by SDS-PAGE using 10% separating gels (56) with a Page-Ruler TM prestained protein ladder (Fermentas) as the standard.
Immunoblotting was performed according to Birkenhäger et al. (43) using 10 mM NaHCO 3 , 3 mM Na 2 CO 3 (pH 9.9), 20% (v/v) methanol as transfer buffer and the ECL Western blotting detection system (Amersham Biosciences) for visualization of immunolabeling. For cross-linking experiments, blot membranes were incubated with the corresponding primary antibodies raised in mice or rabbits (see figure legends), with IRDye TM 800DX-labeled goat-anti-mouse IgG (HϩL), or IRDye TM 700DX-labeled goat-anti rabbit IgG (HϩL) (Rockland or LI-COR) as secondary antibodies and finally detected with the two-channel infrared detection system Odyssey (LI-COR). Both secondary antibodies were affinity-purified for low cross-reactivities with serum proteins of other species, and, therefore, allowed the simultaneous detection of fluorescence (shown in red for IRDye TM 700DX and in green for IRDye TM 800DX) after immunolabeling of two proteins on one blot membrane. As a consequence, an overlay of both labels because of cross-linking of proteins appears as a yellow band. However, a clear yellow band can only be obtained in the presence of a 1:1 balance of the two fluorescent dyes. Because of the varying affinities and titers of the antibodies used (primary as well as secondary), the color of the yellow bands may shift to orange or lime green (chartreuse).
DCCD-sensitive ATPase activity of membrane vesicles was determined using a continuous flow system as described by Jäger et al. (57). ATP-and NADH-driven proton translocation was performed through ACMA (Sigma, catalog no. A-5806) fluorescence quenching as described by Birkenhäger et al. (43).

Epitope Mapping of mAbs Raised against Subunit b of the F O
Complex-mAbs were prepared against subunit b of the E. coli ATP synthase using subunit b purified under non-denaturing conditions dissolved in a cholate-containing buffer as antigen for immunization (43). For epitope mapping, the binding of the individual mAbs to N-or C-terminally truncated b subunits ( Fig. 1) as well as to b or b syn subunits (59) carrying single amino acid substitutions ( Fig. 2 and supplemental Table S1) was determined by immunoblot analyses using cell lysates. The results of the binding analyses for the different mAbs, summarized in Fig.  2, reveal that the epitopes of the different mAbs are spread over the hydrophilic part of subunit b. GDH 10-1A4 recognizes an epitope located at the C-terminal end of subunit b comprising amino acids b150 -156. Even the removal of only the last (two) C-terminal amino acids resulted in a complete loss of mAb binding (Fig. 1). In addition, an elongation of subunit b by two residues (b syn 158) reduced mAb recognition significantly, clearly indicating that the free C terminus is critical in mAb binding (Fig. 2). GDH 1-5A2 and GDH 10-1B1, which also bind within the ␦-binding domain of subunit b, cannot be distinguished by the use of truncated b subunits (Fig. 1). However, the analyses of substituted b subunits clearly demonstrate a slight shift between the binding sites of the two mAbs, b131-138 versus b128 -138 (Fig. 2). GDH 10-6D1 binds within region b77-85 of the dimerization domain, as observed with N-and C-terminally truncated proteins as well as substituted b subunits, whereas region b32-41 in the tether domain is recognized by mAb GDH 1-4D3.
Specificity of Anti-b mAb GDH 10-1A4 Binding-GDH 10-1A4 binds to the C-terminal region of subunit b, which is also known to be directly involved in binding of subunit ␦. Thus, we further investigated its binding characteristics. A direct binding of GDH 10-1A4 to subunit b in assembled F O F 1 was shown by purification of an F O F 1 -mAb complex by size exclusion chromatography (Fig. 3A). After incubation of purified F O F 1 with mAbs, the mixture was separated by gel filtration, showing a slight shift of the F O F 1 peak to a higher molecular mass ( Fig. 3A, peak A), with a shoulder representing residual free F O F 1 (peak B). A further peak represents free mAb mole-cules ( Fig. 3A, peak C) and is identical to that obtained in mAb purification (not shown). A silver-stained SDS gel showed that peak A separated under non-reducing conditions contained, in addition to the subunits of the F O F 1 complex, a molecular mass of approximately 150 kDa that can be detected in a corresponding immunoblot analysis exclusively by mouse IgG-specific antibodies conjugated with horseradish peroxidase (Fig. 3A, inset). The presence of only one single sharp F O F 1 -mAb peak as well as its elution volume indicates that only one mAb molecule was bound per F O F 1 complex. Corresponding results were obtained by application of the other mAb molecules (data not shown).
Binding of mAb molecules to membrane vesicles of insideout orientation in the presence and absence of F 1 was tested via a competitive inhibition ELISA using subunit b as a fixed antigen. GDH 10-1A4 showed competition between subunit b and F 1 -stripped inverted membrane vesicles as well as F 1 -containing inverted membrane vesicles of the atp wild-type strain ML308-225 (Fig. 3B). This indicates the accessibility of the antigenic determinants of subunit b even in the presence of the membrane. However, the greater amount of F 1 -containing membranes, compared with F 1 -stripped membranes, which is required to achieve competition, suggests that the presence of F 1 may hinder recognition of the epitope in some way. A comparable picture was obtained for the other mAbs, except GDH 1-4D3, which showed no binding to inverted membrane vesicles, indicating that region b 32-41 is at least partly shielded by the phospholipid bilayer. Taken together, although mAb GDH 10-1A4 recognizes the C-terminal part of the ␦-binding domain of subunit b, it is still able to bind to its epitope, even in the presence of F 1 and, therefore, subunit ␦.
Influence of Subunit b-specific mAb GDH 10-1A4 on the Function of F O -Independent of the presence or absence of F 1 , the incubation of inverted membrane vesicles with anti-b mAbs had no influence on the DCCD-sensitive ATPase activity or on the proton-translocating activity of the F O complex (Fig. 3C). Although, in the presence of F 1 , the generation of a pH gradient by energization with ATP as well as NADH could be followed by ACMA quenching, in the absence of F 1 , energization of the membranes even by use of NADH was not possible because of the passive proton translocation through the open F O complex. After preincubation with DCCD, however, a pH gradient could be generated by addition of NADH independent of added mAbs.
In contrast, the binding of F 1 to F 1 -stripped inverted membrane vesicles was affected after preincubation with mAbs, probably because of steric hindrance (Fig. 3C). With increasing amounts of mAb molecules up to saturation (not shown), the energization of the membranes with either NADH or ATP was reduced, as was the DCCD-sensitive ATPase activity, revealing that the antibodies interfered with the binding of F 1 to F O .
In summary, binding of mAb GDH 10-1A4 had no influence on the function of assembled F O F 1 , although its antigenic binding site overlaps with the region essential for binding of subunit ␦. This result can only be explained by a simultaneous binding of both molecules, subunit ␦ and mAb GDH 10-1A4, to the b dimer and demonstrates that one b subunit is far more important than the other for binding of subunit ␦, as has also been deduced from a completely different set of experiments by Wood and Dunn (37).
Effect of Cysteine Substitutions on the Function of F O F 1 -As a prerequisite for cross-linking experiments, the native cysteine residues present in the corresponding subunits were exchanged to alanine with no loss of function (supplemental Table S2). Indeed, a substitution of all 10 native cysteine residues present in F O F 1 to alanine has no effect on the ATPase activity of the enzyme (62). Furthermore, each of the mutants used for crosslinking, containing up to five individually introduced cysteine residues, grew similar to the wild type on succinate as the sole carbon and energy source, indicating a functional oxidative phosphorylation system (supplemental Table S2). To quantify the influence of the cysteine substitutions on the activity of ATP synthase, DCCD-sensitive ATPase activities of membrane vesicles were measured (supplemental Table S2). For most mutants, the maximal reduction in ATPase activity was by a factor of 2 in the presence of two cysteine substitutions or by a factor of 3 in the presence of four or five cysteine substitutions. Combinations of mutations causing a greater reduction in activity were not used for cross-linking experiments. However, although reductions in ATPase activity were observed, DCCD FIGURE 1. Epitope mapping by binding of mAbs to N-or C-terminally truncated b subunits. Cells synthesizing plasmid-encoded N-or C-terminally truncated b subunits (1100⌬uncFH/pRPG derivatives (45), KF92rA/pBMF derivatives (46), and DK8/pSD or pDM derivatives (23,29,30,32)) were separated on SDS-PAGE (20 g of cell protein/lane) and analyzed by immunoblotting with different mAbs using the ECL detection system for immunodecoration. The amount of protein was determined on the basis of the calculation that 1 ml of cells with an optical density (578 nm) of 1.0 contains ϳ160 g of protein (58). The differences observed for b and b syn truncated for the C-terminal amino acid bLeu-156 obviously depend on the varying expression level of the corresponding proteins because of different plasmid backgrounds. ؉, mAb binding; Ϯ, weak mAb binding; -, no binding; n.d., not determined. sensitivity, which indicates a precise coupling between F O and F 1 , was maintained in each case.
Identification of Amino Acid Pairs for Cross-linking of Subunit b to Subunits ␣ or ␤-To characterize the interactions of the b dimer in F O F 1 complexes in more detail, nearest neighbor analyses via disulfide-based intermolecular cross-linking are an ideal tool. The proximity of bA92C to region ␣464 -483 shown via benzophenone-4-maleimide cross-linking (29) was used as starting point to determine neighboring amino acid pairs between subunits b and ␣. Region ␣464 -483 has an ␣-helical conformation (1, 63) and, therefore, residues ␣Ser-466, ␣Ala-469, ␣Ala-473, ␣Arg-477, and ␣Ala-480, which are all present at the surface of the ␣ 3 ␤ 3 hexamer, were exchanged to cysteine. Membrane vesicles from mutants with each of these substitutions, along with bA92C, were analyzed for disulfide-linked b-␣ heterodimer formation mediated by copper-1,10-phenanthroline (CuP) using non-reducing SDS-PAGE and immunoblotting with subunit-specific antibodies (Fig. 4A). A direct detection of cross-linking products was carried out in one immunoblot analysis through the use of primary antibodies from different sources and detection by corresponding secondary antibodies labeled with different fluorophores. Heterodimers of b-␣ were observed in a high yield for residues ␣A473C and ␣R477C with bA92C, whereas ␣A480C showed only a weak signal. No cross-linking could be detected for ␣S466C and ␣A469C. When possible, cross-linking yields were estimated from both the appearance/increase of the crosslinked product and the reduction of the band intensities of the corresponding monomeric subunits. However, because of the presence of two b and three ␣ subunits within F O F 1 , but crosslinking of only one of each, and because of the signal saturation obtained by the conditions used for the immunoblot analyses, the removal of single polypeptides could hardly be determined in many cases. Nevertheless, even an increase of pH to 8.2, which lies close to the pK value of sulfhydryl groups under standard conditions, during the oxidation reaction with CuP had no influence, suggesting a high cross-linking yield for these interaction partners. In addition, membrane vesicles containing no cysteines within the two subunits studied and those carrying only the substitution bA92C were used as controls. As side reactions, the formation of b dimers and of a band with an apparent molecular weight of ϳ120 kDa could be observed in small amounts, which could both be completely abolished by reduction of the disulfide bonds with 2-mercaptoethanol (data not shown). The b dimer formation was not unexpected because the two b subunits are known to closely interact with each other (compare Refs. 22,23,28). To analyze the proteins present within the 120-kDa band, immunoblotting using antibodies raised against the individual eight F O F 1 subunits was performed, but the band could only be detected with antibodies against subunit ␣ (data not shown). Furthermore, using membranes with otherwise cysteine-free F O F 1 complexes, in which only ␣R477C is present, the cross-linking band could still be observed (data not shown), indicating the probability that an ␣ dimer had been formed, likely through an interenzyme reaction. Substitution of a cysteine residue in ␣ (or ␤) will result in incorporation of three thiol groups into F O F 1 . Although one may be proximal to the b dimer, the other two located at the surface of the hexamer will be freely available for other reactions. Because membrane vesicles are rich in F O F 1 , this intermolecular cross-linking was not unexpected.
The same approach was used to identify amino acid residues of subunit ␤ that are in close proximity to bA92C (Fig. 4B). Amino acid residues ␤Leu-347, ␤Gln-351, ␤Glu-352, ␤Arg-425, ␤Lys-428, and ␤Glu432 were substituted by cysteine. These residues are all located within a comparable distance to the non-catalytic cleft as those residues of subunit ␣ that were able to form an S-S bridge with residue bA92C. Intense cross-link formation of  Table S1 summarizing the complete analysis. The immunoblot analyses were performed as described in legend to Fig. 1. To verify that residue bAla-68 is not a part of the epitope of mAb GDH 10-6D1, GDH 1-5A2 has been shown as a control, revealing that, in general, the expression of the b syn A68C variant is lower. bA92C to ␤L347C and ␤Q351C was observed, whereas the reaction with ␤E352C was much weaker (Fig. 4B). Because of the localization of the interacting amino acid residues within an ␣␤ pair, it can be concluded that the b dimer is located at a non-catalytic ␣/␤ interface, as suggested previously (22,29).
In summary, Fig. 4C shows the localization of the amino acid residues of subunits ␣ and ␤ interacting with residue bAla-92. At this point, however, it is not apparent if both b subunits are involved in interaction with the ␣␤ pair (one b interacting with ␣ and the other with ␤) or whether only one of the two b subunits can interact at its position 92 with either subunit ␣ or ␤ depending on the cysteine residue provided in the experiment.
Identification of a Cross-linking Pair for Tagging One Subunit b with Subunit a-Subunit a is known to stably interact with the b dimer (25), and the cysteine pairs bN2C-aG227C and bN2C-aL228C allow cross-link formation (39). However, hydrophilic loops as well as N-terminal regions are in many cases highly flexible elements, and, therefore, we set out to identify a crosslinking pair between the more tightly packed transmembrane helices to increase the probability that a defined cysteine in the single subunit a might only react with the cysteine of one of the two b subunits. bLeu-16, located outside the known b-b interaction area (24), was exchanged to cysteine and combined with individual cysteine substitutions within region a237-250, which is proposed to be part of the C-terminal transmembrane helix of subunit a (64,65). The region was chosen on the basis of the finding that mutation bG9D can be partially suppressed by mutation aP240A/L in subunit a (66). The experiments shown  lanes 1 and 2), and analyzed by immunoblotting (lanes 3 and 4) using exclusively goat anti-mouse IgG antibodies conjugated with horseradish peroxidase for ECL detection. Lanes 1 and 3, purified F O F 1 ; lanes 2 and 4, peak fraction A. mAU, milliabsorption unit. B, competition between bound subunit b and free membrane vesicles. Microtiter plates were coated with purified subunit b (0.1 g/ml) and subsequently incubated with a mixture of different amounts of free membrane vesicles and culture supernatant of mAb GDH 10-1A4. The competitive inhibition ELISA was performed as described by Jäger et al. (57). f (dark gray), F 1 -containing inverted membrane vesicles of the atp wild-type strain ML308-225; ࡗ (black), F 1 -stripped inverted membrane vesicles of strain ML308-225; OE (light gray), inverted membrane vesicles of the atp mutant strain CM1470. The data represent average values of three independent measurements. C, influence of subunit b-specific mAb GDH 10-1A4 on the function of F O F 1 . Membrane vesicles (MV; 2 mg/ml) were incubated in the absence or presence of mAb GDH 10-1A4 (50 g/ml) as described under "Experimental Procedures." NADH-driven proton translocation by respiration and ATP-driven proton translocation by F O F 1 were measured via ACMA fluorescence quenching. DCCD-sensitive ATPase activities were determined as described. The presence of mAb GDH 10-1A4 on the membrane vesicles after the washing steps was determined by competitive inhibition ELISA before measuring ATPase and proton translocation activities. a, NADH-dependent ACMA fluorescence quenching after incubation of F 1 -stripped membrane vesicles with DCCD (40 M, 20 min at room temperature) for inhibition of F O . b, rebinding of F 1 to F 1 -stripped inverted membrane vesicles was performed after incubation with mAb, as described by Deckers-Hebestreit et al. (53).
in Fig. 5 demonstrate that only cysteine pair b16-a239 formed a cross-link with a high yield in the presence of CuP (note the conversion of virtually all subunits a to a-b). Therefore, the cross-linking pair b16-a239 was used in further experiments where several cross-linking pairs were combined to determine the interacting partners for both b subunits.

Generation of a-b-␣ and b-␤ Complexes by Combining Cysteine Pairs for Cross-linking-
In the next step, the cross-linking pair b16-a239 was combined with a second cysteine on b that would allow disulfide linkage to identified sites on either subunit ␣ or ␤. Furthermore, F O F 1 variants containing single cysteine residues in both ␣ and ␤ as well as the corresponding cysteine residue in the b subunit were used to identify preferred linkages between subunit ␣ or ␤ with one of the two b subunits. Note that in Fig. 6, lanes are identified by the pairs of subunits containing cross-linkable cysteines identified previously, with a-b always being b16-a239, whereas the varying positions of cysteines for b-␣ and b-␤ cross-links are indicated within the corresponding boxes. Fig. 6A shows a typical experiment with pairs b16-a239, b118-␣118, and b118-␤117, appropriate controls, and immunolabeling of all F O F 1 subunits involved in cross-linking, i.e. subunits a, b, ␣, and ␤. When only individual cross-linking pairs were present, strong, specific disulfide linkage was observed, although, in each case, a b-b band could also be detected. The heterodimers b-␣ and b-␤ were also observed in the absence of CuP, probably produced by oxygen present , and analyzed by immunoblotting using two antibodies simultaneously, as described under "Experimental Procedures." A, cross-linking between bA92C and residues of subunit ␣ of region ␣464 -483 exposed at the surface of the ␣ 3 ␤ 3 hexamer. Cross-linking yields obtained after incubation with CuP at pH 7.5 and 8.2, respectively, were compared. For immunolabeling, mouse anti-b antibodies (GDH 10-1B1, green) and rabbit anti-␣ antibodies (red) were applied. In each mutant, endogenous cysteine residues in subunits b (bCys-21) and ␣ (␣Cys-47, ␣Cys-90, ␣Cys-193, and ␣Cys-243) were substituted by alanine. St, molecular mass standard. B, cross-linking of bA92C with residues of subunit ␤ located at a noncatalytic cleft of F 1 . Rabbit anti-b (red) and mouse anti-␤ antibodies (green) were applied for detection. In each mutant, endogenous cysteine residues in subunits b (bCys-21) and ␤ (␤Cys-137) were changed to alanine. St, molecular mass standard; unspec., unspecific immunolabeling. C, structure of the E. coli ␣ 3 ␤ 3 hexamer (63) drawn by RasMol 2.7.2.1.1 with marked amino acid residues of subunits ␣ and ␤, which were cross-linked to subunit b at position bA92C. Red, high cross-linking yield; orange, intermediate cross-linking yield; yellow, low cross-linking yield; white, no cross-linking.
during incubation (Fig. 6A, lanes b-␣ and b-␤). In each case, disulfide bonds could be reduced by addition of 2-mercaptoethanol (data not shown). When the b-a pair was also present, we observed a shift of b-␣ as well as of b-␤ to a molecular mass corresponding to a-b-␣ or a-b-␤, and, indeed, the immunolabeling revealed the presence of all three subunits (Fig. 6A, lanes  b-a/b-␣ and b-a/b-␤). In addition, small amounts of an a-b-b product could be observed, because of the presence of two cysteine residues within subunit b, namely bL16C for linkage to subunit a and bE118C for linkage to subunit ␣ or ␤, whereas the amount of b dimer was strongly reduced. Surprisingly, when CuP was provided in the simultaneous presence of single cysteine residues in both subunits ␣ and ␤, only an a-b-␣ crosslinked product was formed, indicating that the b subunit bridged with subunit a is preferentially cross-linked to subunit ␣, whereas no b-␤ or a-b-␤ was observed (Fig. 6A, lane b-a/b-␣,␤). Thus, it appears from the results obtained with the set of cysteine substitutions used in Fig. 6A, that one of the b subunits, which we designated as b I , can be tagged with the a subunit through the b16-a239 pair and can also be linked to either subunit ␣ or subunit ␤, but preferentially to ␣ when the choice is available. In contrast, it appears that the second b subunit, b II , which is not tagged with subunit a, can be linked to neither ␣ nor ␤ in this case.
Comparable experiments performed for six other combinations of cross-linking pairs yielded a range of results as shown in Fig. 6B and summarized in Fig. 7. The cysteine residues present in F O F 1 in addition to b16-a239 are noted at the left side of the figure in the corresponding boxes. In each case, only the upper part of the immunoblot analysis is presented, showing the formation of b-␣, b-␤, a-b-␣, and a-b-␤, respectively. Bands with molecular masses corresponding to ␣-␣ or ␤-␤ were also detected after immunodecoration with subunit-specific antibodies, depending on the cysteine substitution present.
In each case, b-␣ and b-␤ cross-links could be observed, although the cross-linking yield of b106-␣118 was rather low. Upon combination with b16-a239, a shift of the b-␣ bands corresponding to a-b-␣ could be detected in each case, again, with only a weak band for b16-a239/b106-␣118. However, for the b-␤ cross-linking pairs, a different picture emerged after combination with a-b. A complete shift could only be observed for b118-␤117. Otherwise, depending on the b-␤ pair present, a moderate change (b103-␤122), weak change (b103-␤286, b92-␤351, b92-␤122), or no change (b106-␤287) of the b-␤ band to an a-b-␤ band was obtained (Fig. 6B and summarized in Fig.  7A). Remarkably, in the simultaneous presence of cysteine residues in subunits ␣ and ␤, an a-b-␣ complex was formed in each case, whereas subunit ␤ preferentially formed a b-␤ cross-linking product. Although there are two exceptions (no cross-linking between b118-␤117 and a weak a-b-␤ product for b16-a239/b103-␤122), the results obtained clearly demonstrate that it is possible to distinguish between the two b subunits by their different interaction partners, as summarized in Fig. 7B. These results strongly support the idea that the sulfhydryl group of aV239C forms a specific disulfide bond with the sulfhydryl group (bL16C) of only one of the two b subunits present (b I ).
In summary, for most positions within the region tested, cysteines in subunit b I (marked by subunit a) cross-linked strongly to subunit ␣. If ␣ contained no suitably positioned cysteine residue, most b I positions cross-linked also to ␤, as seen by the a-b-␤ band. These results imply that b I is located at an ␣/␤ interface, positioned most closely to ␣. In contrast, cysteines in subunit b II were usually linked to subunit ␤, as seen by the formation of b-␤ in the presence of the b I -a cross-link, except for residue bE118C. We saw no evidence that cysteines in b II would cross-link to sites on ␣. This would have been indicated by formation of b-␣ in the presence of the b-a, b-␣, and b-␤ cross-linking pairs. These results imply that subunit b II is positioned largely along the surface of the ␤ subunit.
Formation of an ␣-b I -b II -␤ Complex-To further test the observation that the two b subunits show a different interaction pattern toward subunits ␣ and ␤, cross-linking pairs b118-␣118 and b106-␤287 were combined with an additional cysteine residue at position b68, which allows b dimer formation in a high yield (59). If the hypothesis is correct, a covalently linked quaternary subcomplex composed of subunits ␣ and ␤ connected through the b dimer should be generated, as shown in Fig. 8B. Experimentally, each of the individual cross-linking pairs generated the expected homo-or heterodimer in addition to b-b, ␣-␣, or ␤-␤ as side products (Fig. 8A). By combining all three pairs, the formation of ␣-b-b-␤ as well as of ␣-b-b can be detected by immunolabeling as two newly formed bands with corresponding molecular masses (Fig. 8A). Although the bands were densely focused because of the behavior of proteins with high molecular masses in 10% N-[2-hydroxy-1,1-bis(hydroxymethyl)ethyl]glycine-SDS gels (56), the product yields appeared to be unusually high given the number of cross-links required. In the presence of CuP, free subunit b as well as b-␣ and b-␤ were completely absent, and b-b was drastically reduced compared with the samples containing individual cross-linking pairs, indicating that a large majority of the b subunit is involved in crosslinking. These results confirm that subunits b I and b II of the peripheral stalk can be clearly distinguished by their individual cross-linking properties, as described above.
Identification of the b Subunit Preferentially Involved in Binding of Subunit ␦-Subunit ␦ is known to interact through its N-terminal domain with the N-terminal region of subunit ␣ (67, 68), whereas the C-terminal region of ␦ is in contact with the C-terminal region of b, and one of the b subunits is much more critical for binding of ␦ than the other (30,31,37). To identify this b subunit, the same design as described above was applied, using the a-b cross-linking pair to discriminate between the two b subunits. In a first step, bE155C (30) was combined with subunit ␦ derivatives carrying cysteine substitutions within region ␦149 -172 to screen for a suitable b-␦ cross-linking pair. Residues ␦A149C, ␦I160C, and ␦G162C  of subunits b, ␣, and ␤ combined with a b-a (bL16C/aV239C) cross-linking pair to distinguish both b subunits. A, immunoblot analysis of the a-b I -␣ and b II -␤ complexes, respectively, generated by combination of the following cysteine pairs: b-a, bL16C/aV239C; b-␣, bE118C/␣K118C; b-␤, bE118C/␤E117C; and b-␣,␤, bE118C/␣K118C/␤E117C. Inverted membrane vesicles of the corresponding mutant strains were incubated with CuP as described in the legend to Fig. 4. Immunodecoration: left panel, rabbit anti-b (red) and mouse anti-a (green) antibodies. Right panel, mouse anti-␣ (green) and rabbit anti-␤ antibodies (red). In each mutant, the corresponding endogenous cysteine residues in subunits b, ␣, and ␤ (compare legend to Fig. 4) were substituted by alanine. Because of the use of 10% N-[2-hydroxy-1,1-bis(hydroxymethyl)ethyl]glycine-SDSpolyacrylamide gels (56), a high resolution of protein bands is observed within the range up to 50 kDa, whereas bands with higher molecular weights are observed as dense, highly focused bands (compare also the molecular mass standard). The yellow intensities observed in the blot analysis at the right side arise from a partial overlay because of the close proximity of the bands representing ␣ and ␤ as well as to a slight degradation of subunit ␣. St, molecular mass standard. B, sections of immunoblot analyses comparable with A showing the cross-linking products obtained by several combinations of different cysteinesubstituted subunits b, ␣, and ␤ as announced in the corresponding boxes at the left side. a-b always represents the presence of cross-linking pair bL16C-aV239C. For clarity, during the simultaneous presence of cysteine substitutions in subunits ␣ and ␤ for interaction with subunit b (last lane of the blot membranes), the complexes formed are marked in green for complexes involving subunit ␣ and in red for those containing subunit ␤.
were identified for disulfide bond formation (data not shown). However, a detection of the b-␦ cross-linked product was only possible with anti-␦-specific antibodies, whereas, surprisingly, the polyclonal anti-b antiserum, which exhibits an extremely high avidity to subunit b, cannot bind to the cross-link product formed. The dominant epitopes recognized by the antibodies present are apparently shielded by interacting subunit ␦ (Fig.  9C). Applying the different mAbs for detection revealed that the mAbs binding within region b123-156, namely the ␦-binding domain (23) obtained for lanes b-a/b-␣ and b-a/b-␤ (A) and for lanes b-a/b-␣,␤ (B) in Fig.  6. The numbering indicates the amino acid residues substituted by cysteine within the corresponding subunit. Solid bars, high cross-linking yields; dashed bars, low cross-linking yields.  Fig. 4. For immunolabeling, rabbit anti-b antibodies (red) and mouse anti-␣ (green, left panel) or mouse anti-␤ antibodies (green, right panel) were applied. In each mutant, the corresponding endogenous cysteine residues in subunits b, ␣, and ␤ (compare legend to Fig. 4) were substituted by alanine. St, molecular mass standard. B, schematic to illustrate the disulfide bonds formed between the different stator subunits allowing a clear discrimination between both b subunits present in F O F 1 . Black bars, zero-length cross-links present to obtain the quaternary ␣-b I -b II -␤ complex; gray bar, zero-length cross-link used to mark subunit b I with subunit a.; unspec., unspecific reaction of the secondary antibody IRDye TM 700DX-labeled goat-anti rabbit IgG (HϩL). 9D). However, with mAbs GDH 1-4D3 and GDH 10-6D1 with their epitopes located within the tether and dimerization domains, respectively, subunit b could be identified within b-␦, although the binding was comparatively weak (Fig. 9D).
Next, the b155-␦160 cross-linking pair was combined with b16-a239, and a cross-linking experiment with CuP was performed (Fig. 9A). As expected, no b-␦ product could be observed in the presence of only ␦I160C or bE155C, although bE155C showed strong b-b formation. b155-␦160 showed disulfide bond formation even in the absence of oxidizing agent, reaching a high cross-link yield in the presence of CuP. In each case, cross-links were abolished by the addition of 2-mercaptoethanol (data not shown). In the presence of both cross-linking pairs b155-␦160 and b16-a239, a complete shift of the band representing b-␦ toward a band with a molecular weight corresponding to an a-b-␦ product was observed after treatment with CuP using anti-␦ antiserum for detection (Fig. 9A, top  panel). However, with anti-b mAb GDH 1-4D3, three new bands were detected, whereas the b-␦ product was eliminated (Fig. 9A, center panel). Comparing the same immunoblot analysis as in the top panel decorated with anti-␦ (green) and anti-b antisera (red) (Fig. 9C), it can be seen that only the band with the highest molecular mass is decorated by anti-␦ antibodies, whereas the other two products were only detected by anti-b antiserum, indicating that the double band observed represents two different cross-linked products. From examination of the control lanes containing samples with cysteines at b16-a239 and b16-a239/b155, but lacking the substitution at ␦160, it is evident that, in addition to an a-b-␦ product, an a-b as well as an a-b-b cross-link was formed during treatment with CuP (Fig.  9A, center and bottom panels). Comparable results were obtained when combining b16-a239 with b155-␦149 and b155-␦162 (data not shown). Taken together, these results demonstrate that subunit b I is the direct interaction partner of subunit ␦ within the peripheral stalk, as summarized in Fig. 9B (black bars).

DISCUSSION
Despite many studies of the stator stalk of bacterial F-ATPases, in particular the b homodimer of E. coli, its relationship to the rest of the enzyme has remained ill-defined in many ways. In particular, the asymmetry of the enzyme requires that the two b subunits interact with other subunits in different ways. The b dimer itself appears to be intrinsically asymmetric because the helices of the dimerization domain form an unusual coiled coil in which they are offset rather than in-register (36). However, features distinguishing the different positions of the b subunits within F O F 1 have not been identified previously. At this point, single particle electron microscopic analyses have provided the most detailed images of the holoenzyme (69). No high-resolution structure of the holoenzyme is yet available, and it cannot be anticipated when one will be obtained. In this work, we have used two tools, mAb binding analyses and disulfide-based cross-linking of subunits in membrane-bound F O F 1 , to provide a more discriminating view of the relationships of the two b subunits to the remainder of the enzyme.
Binding between Subunits b and ␦-A shortening of the C-terminal region of both b subunits by four amino acids (bV153end) prevents the assembly of functional F O F 1 (46,41), whereas a heterodimer, in which only one of the two b subunits was shortened by molecular genetics, allows the formation of functional F O F 1 (41). Studies on the binding of F 1 to purified, soluble heterodimeric b syn subunits indicated that although one b subunit was most critical, the second b subunit played a lesser role (37). Here, the interaction of this region of one b subunit, designated b I , to the highly conserved C-terminal region of subunit ␦ could be observed, because bE155C was cross-linked to residues ␦A149C, ␦I160C, or ␦G162C, consistent with the cross-link pair described previously, b158C-␦M158C (30). Immunoblot analyses showed that cross-linking of b-␦ prevented not only the immunodecoration by mAb GDH 10-1A4 but also by mAbs GDH 1-5A2 and GDH 10-1B1 (Fig. 9D), revealing that the ␦-binding domain is mostly shielded by subunit ␦ in the presence of the covalent S-S bond, even under denaturing conditions. This finding fits well with the extensive helix-helix interaction observed for the C-terminal regions of subunit b and OSCP (homologous to bacterial subunit ␦) of bovine mitochondrial ATP synthase (18).
We observed that antibody GDH 10-1A4 was stably bound to its epitope in F O F 1 but less strongly than in F O , and that this binding had no influence on the functional activities of the wild-type F O F 1 complex. Together, these results imply that the C-terminal region of the second b subunit, here designated b II , which is less critical to the interaction with ␦, is accessible to the antibody, but to a more limited extent than when F 1 has been removed. In this context, it is worthwhile mentioning that the 10-Å cryoelectron microscopic structure of the A-type ATP synthase of Thermus thermophilus suggests that subunits E and G of the two peripheral stator stalks are differently associated with the A 3 B 3 hexamer, with only subunit E being closely apposed to the non-catalytic subunit B (11). Assuming a similar arrangement of the E. coli b dimer with the ␣ 3 ␤ 3 hexamer, it becomes feasible that the epitope of at least one of the two b subunits is accessible for mAb binding.
Tagging of One of the b Subunits by Subunit a-Several systems have been developed to allow for discrimination between the two b subunits within the E. coli homodimer. Using a genetic approach, two different b-encoding sequences were labeled with different tags (NH 2 -His tag and COOH-V5 tag) (28) or with diverse mutations (bR36I/bV153end) (41). In addition, chimeric b proteins were constructed by combination of b and bЈ DNA sequences from T. elongatus with the E. coli atpF gene (19,20). An in vitro approach utilized specific disulfide bond formation between purified b syn derivatives of different sizes (b34 -156 and b53-156) (37). Although valuable tools, these approaches have been limited by unequal expression, less-than-specific heterodimer incorporation into ATP synthase, or limited applicability. Furthermore, the individual locations of the two subunits have not been distinguished.
Here, we have identified pairs of cysteine-substituted positions between subunit b and subunits a, ␣, ␤, and ␦ that can be cross-linked through disulfide bond formation, and we then used these pairs in combinations to form multiple cross-links within ATP synthase solely by engineering the plasmid-encoded atp operon. This has allowed us to discriminate between the positions occupied by the two identical b subunits in the enzyme complex. Subunit a, known to tightly interact with the b dimer (25), was chosen to mark one of the b subunits by S-S bond formation between positions b16 and a239. The contact areas of the single, asymmetric subunit a must differ for the two b subunits. Using bL16C, which has little or no tendency to form b-b cross-links (24), a disulfide could be formed with aV239C in a high yield. A number of observations indicate an exclusive linkage of aV239C to bL16C of just one b subunit. First, in the presence of the b16-a239 cross-link, a complete shift of b-␦ to a-b-␦ was observed (Fig. 9, A and C, lane b-␦/a-b), and only one b subunit is involved in binding of subunit ␦ (see above). If cross-linking between subunits a and b was a statistical process, one would expect two bands corresponding to b-␦ and a-b-␦ in comparable amounts. Second, the b subunit tagged by subunit a (b I ) showed a preferred cross-linking to subunit ␣, whereas the unmodified subunit b (b II ) was exclusively linked to subunit ␤. The formation of an ␣-b-b-␤ cross-linked product in a high yield further supports our conclusion.
Individual Cross-links of the Two b Subunits to ␣ 3 ␤ 3 -The data show that, in general, the b subunit positions can be distinguished by their specific sites of disulfide cross-linking to either ␣ or ␤ (Fig. 9B, black and gray bars). The proximity of bA92C and region ␣464 -483 was inferred previously through cross-linking with benzophenone-4-maleimide, which provides a 10-Å spacer between the sites (29), and disulfide formation between bA92C and ␣R477C confirmed this. The sites of ␤ that supported disulfide formation to bA92C were located slightly above the region proposed after benzophenone-4-maleimide cross-linking (22,29). Using bA92C as starting point, several additional cross-linkable pairs between the b dimer and the ␣ 3 ␤ 3 hexamer were identified. However, before tagging subunit b I with subunit a, it was not evident that subunit ␣ exclusively interacts with subunit b I , whereas subunit ␤ predominantly makes contact to subunit b II .
The b cross-link sites in subunits ␣ and ␤ are located along a non-catalytic ␣/␤ interface, as proposed previously (22,29) and as shown by a structural analysis for the peripheral stalk of bovine mitochondrial ATP synthase (18). Our confirmation of this placement is significant because, unlike the experimental approaches used previously, it is done on the basis of work with the holoenzyme in its native membrane-bound state. Random movements of b 2 are restricted by specific interactions at both ends. We found that the S-S bond between b92-␣477 had no influence on the ATP-hydrolyzing activity (data not shown). Previous work (70) has shown that only small differences exist in the dwell times of the three catalytic sites, despite the asymmetry provided by the stator stalk. Evidently, the non-catalytic ␣/␤ interface, falling between two catalytic ␣␤ pairs, provides an unobtrusive placement for this asymmetric feature. However, in this context it is important to note that, although the ␣ subunit cross-links specifically to b I , the ␤ subunit may crosslink to either b I or b II , depending on the presence of an additional cysteine residue in subunit ␣ (Figs. 6 and 7). These observations argue in favor of b I being located at the ␣/␤ subunit interface, whereas b II may be along the adjacent surface of ␤. It should be noted that the recent 10-Å cryoelectron microscopic structure of the T. thermophilus A-type ATPase (11) showed each of the two heterodimeric EG stator stalks to be positioned along the outer surfaces of non-catalytic B subunits, with interactions only between subunits B and E. It seems unlikely to us that a similar arrangement of the ␣ and b subunits of E. coli F O F 1 would yield the results we obtained, in particular the b-␤ disulfide cross-links, suggesting that some variation in stator placement may exist within the family of rotary ATPases. Subunit b II may be less closely associated with ␣ 3 ␤ 3 than b I , given the lower levels of cross-linking to ␤ that was observed for most positions, but one cannot draw firm conclusions from weak or negative cross-linking results.
Subunit b I has also been shown to be proximal to subunit ␦. Comparing the peripheral stalks in bacterial F O F 1 complexes to that of mitochondria, it can be inferred that subunit b I is functionally related to the single subunit b present in mitochondrial ATP synthase, which makes the major contact to OSCP (17,18). Subunit b II might have a more stabilizing effect, like mitochondrial subunits d and F 6 , which appear to stiffen the continuous ␣ helix of subunit b through their shorter structural elements (18).
Staggered versus Non-staggered, In-register Conformation of the b Dimer-A staggered assembly within the dimerization domain of the b dimer with an offset of ϳ5.5 amino acid residues has been shown by different experimental approaches (20,36,37). The more N-terminally shifted b subunit (b N ) was shown to be more important for binding of subunit ␦ or the F 1 complex than the b subunit in the b C position (37). Therefore, it is reasonable to propose that the N-terminally shifted b N position corresponds to the b I subunit described in this study. If one compares the positions on ␣ and ␤ that formed disulfides with particular positions of b, there is in general a trend that the positions on ␤ are located farther up the surface of ␣ 3 ␤ 3 than those on ␣, consistent with b II being C-shifted and b I being N-shifted.
The bL16C-aV239C Cross-link-Identification of a highyield disulfide formation between aV239C and bL16C was quite unexpected. It is generally accepted that bLeu-16 is located within the lipid core of the membrane (24), with bTyr-24 and bTrp-26 at the cytoplasmic membrane/water interface (71). In addition, the binding characteristics of mAb GDH 1-4D3 revealed that at least a part of region b32-41 is shielded by the membrane, most probably by the polar head groups of the phospholipids. Although a structure for subunit a is still missing, Fillingame and co-workers (Ref. 64 and references therein) have analyzed the four C-terminal transmembrane helices of subunit a in great detail. In their topology model, residue aVal-239 is located close to the periplasmic surface of the membrane so that, even assuming the two helices to be adjacent, bL16C and aV239C would still be several angstroms apart. At this point we can only speculate on the explanation for such efficient cross-linking. Possibly, the membranespanning b helices may also be present in a staggered conformation, as proposed for the dimerization region. This would place bL16C of one subunit closer to the periplasmic surface. Alternatively, thermal fluctuations in the membrane domains may be greater than ordinarily expected. Finally, some modification of the topological models may be in order. In each case, additional studies examining these possibilities are required. In this context, it is worthwhile mentioning that a recent study of DeLeon-Rangel et al. (60) suggests that the two b subunits show individual interactions with different transmembrane helices of the four helix bundles of subunit a.