Subunit δ Is the Key Player for Assembly of the H+-translocating Unit of Escherichia coli FOF1 ATP Synthase*

Background: FOF1 ATP synthases are rotary nanomachines transporting protons across the membrane during catalysis. Results: Escherichia coli FOF1 is assembled from subcomplexes with subunit δ functioning as clamp to generate stable FO. Conclusion: δ guarantees that the open proton channel is concomitantly assembled within coupled FOF1. Significance: We investigate how a proton-translocating unit is assembled while simultaneously maintaining the low membrane proton permeability essential for viability. The ATP synthase (FOF1) of Escherichia coli couples the translocation of protons across the cytoplasmic membrane to the synthesis or hydrolysis of ATP. This nanomotor is composed of the rotor c10γϵ and the stator ab2α3β3δ. To study the assembly of this multimeric enzyme complex consisting of membrane-integral as well as peripheral hydrophilic subunits, we combined nearest neighbor analyses by intermolecular disulfide bond formation or purification of partially assembled FOF1 complexes by affinity chromatography with the use of mutants synthesizing different sets of FOF1 subunits. Together with a time-delayed in vivo assembly system, the results demonstrate that FOF1 is assembled in a modular way via subcomplexes, thereby preventing the formation of a functional H+-translocating unit as intermediate product. Surprisingly, during the biogenesis of FOF1, F1 subunit δ is the key player in generating stable FO. Subunit δ serves as clamp between ab2 and c10α3β3γϵ and guarantees that the open H+ channel is concomitantly assembled within coupled FOF1 to maintain the low membrane proton permeability essential for viability, a general prerequisite for the assembly of multimeric H+-translocating enzymes.

F O F 1 ATP synthases comprise different structural and functional entities that couple the translocation of ions across the membrane to the synthesis or hydrolysis of ATP via a rotary mechanism. In Escherichia coli, the membrane-integrated, H ϩ -translocating F O complex consists of subunits ab 2 c 10 , whereas the peripherally associated F 1 part with its catalytic centers comprises subunits ␣ 3 ␤ 3 ␥␦⑀. As the rotor c 10 ␥⑀ is driven by the transport of protons through two half-channels formed by subunit a as well as the c 10 ring in F O , the rotation of ␥⑀ causes conformational changes in the catalytic nucleotidebinding sites within the ␣ 3 ␤ 3 hexamer of F 1 , thereby provoking ATP synthesis and its release. To counter the tendency of the ␣ 3 ␤ 3 to follow the rotation of the rotor, a peripheral stalk consisting of b 2 ␦ stabilized by subunit a acts as a stator and holds ␣ 3 ␤ 3 in position (1,2).
In E. coli, insertion of subunit c into the membrane depends on YidC insertase (3), and formation of a c 10 subcomplex is independent of the presence of other F O F 1 subunits (4). Recent studies on AtpI revealed a participation probably in a chaperone-like manner in the assembly of a stable c ring (5)(6)(7). Insertion of subunits a and b into the membrane involves the Sec translocon, the signal recognition particle pathway, and for subunit a also YidC (8 -10). Furthermore, a stable insertion of subunit a into the membrane is strictly dependent upon the co-insertion of the other F O subunits, whereas b and c are inserted independently (11). Overexpression of subunits ␣, ␤, and ␥ allowed complex formation with high ATPase activity in the cytoplasm (12), and for interaction between subunits ␦ and ␣, the assembly of ␣ with other F 1 subunits is a prerequisite (13).
The different subunits of E. coli ATP synthase are translated from a polycistronic atp mRNA, and a balanced stoichiometry is obtained by translational coupling between the cistrons as well as regulation by mRNA secondary structure (14,15). In most bacteria, the cistrons are arranged in clusters separating those for F O from those for F 1 , an arrangement fitting well with the proposal that both have been evolved from functionally unrelated ancestor protein complexes (16 -18). Furthermore, this suggests that ATP synthases are probably assembled from subcomplexes, and studies on the assembly of the yeast mitochondrial ATP synthase support this assumption (19).
The goal of this study was to gain insight into the assembly pathway of the F O complex of the E. coli ATP synthase with special emphasis on the H ϩ -translocating unit. Accordingly, the analyses of single-subunit knock-out mutants ⌬a, ⌬b, and ⌬␦, in which the synthesis of subunits a, b, and ␦, respectively, is prevented due to insertion of an early stop codon into the corresponding gene, were combined with intermolecular disulfide bond formation of cysteine-substituted subunits or affinity purification of partially assembled F O F 1 complexes. ⌬a allowed the formation of F O F 1 Ϫa in amounts comparable with wild type. ⌬␦ revealed the presence of an ab 2 as well as a c 10 ␣ 3 ␤ 3 ␥⑀ subcomplex; both could be assembled into a functional ATP synthase by a time-delayed synthesis of subunit ␦. ⌬b only contained the F O F 1 core complex c 10 ␣ 3 ␤ 3 ␥⑀. In each case, subunit ␦ was essential for the integration of the b dimer into the core complex independent of the interaction of b 2 with subunit a. This demonstrates that ␦ functions as a clamp between ab 2 and c 10 ␣ 3 ␤ 3 ␥⑀ to generate the H ϩ -translocating machinery within F O .

EXPERIMENTAL PROCEDURES
Mutagenesis-All plasmids used are listed in Table 1. In most cases, a two-step PCR method was used to generate the early stop codons and the cysteine substitutions in the individual subunits. For cysteine substitutions, the plasmids used as template DNA contain alanine codons instead of the endogenous cysteine codons in the atp genes mutated. Mutant PCR fragments were transferred into pBWU13 or its derivatives using single or double cutters for restriction. To study subunit interactions in the absence of other F O F 1 subunits, atpF or atpEF were cloned into plasmid pET-22b via NdeI/EcoRI with the EcoRI site directly located downstream of the corresponding stop codon. For expression of the atp genes under control of the inducible/repressible promoter P araBAD , a KpnI site was introduced in pFV2 downstream of the weak, constitutive atp promoter P3 (29). Subsequently, the atp operon was cloned into pBAD33 via KpnI (49 bp upstream of the stop codon of atpI) and XbaI (320 bp downstream of the 3Ј-end of atpC). The atpH gene was cloned into plasmid pET-22b via NdeI/EcoRI, and the start codon was subsequently exchanged to TTG. The presence of each mutation was confirmed by DNA sequencing.
Bacterial Strains and Growth Conditions-E. coli strains used are listed in Table 1. The atp deletion strain HB1(DE3) was obtained by P1 co-transduction of ⌬atpBEFHAGDC and ilv::Tn10 (Tet R ) using DK8 (20) as donor and BL21(DE3) (Novagen) as recipient strain (4). E. coli strain DK8 transformed with plasmid pBWU13 or its derivatives or strain HB1(DE3) transformed with pET-22b derivatives were grown in minimal medium with 0.5% (v/v) glycerol or in LB medium with 100 g/ml ampicillin as described (4). All DK8 cells transformed with pBWU13 derivatives grew on succinate as a nonfermentable carbon source indicating a functional oxidative phosphorylation system, whereas no growth was observed in knock-out mutants (data not shown).
Time-delayed in Vivo Assembly of Subunit ␦ into Preformed F O F 1 Ϫ␦-The time-delayed in vivo assembly system used to study the time-delayed integration of subunit ␦ into preformed F O F 1 subcomplexes missing subunit ␦ has been characterized in detail by Brockmann et al. (30). DK8 transformed with three different plasmids was grown in TYGPN medium (31) at 37°C with 100 g/ml ampicillin, 30 g/ml chloramphenicol, and 50 g/ml kanamycin. The medium was preincubated overnight with 10 units/ml ␤-galactosidase from Kluyveromyces lactis (Sigma) for removal of lactose known to be present in varying amounts in yeast extract used for preparation of TYGPN medium (TaKaRa Single Protein Production System, TaKaRa Bio Inc.). After inoculation, the medium was supplemented with 0.03% (w/v) arabinose for induction of P araBAD -controlled atp genes. At OD 578 nm ϭ 0.3, the araBAD promoter was repressed by adding 0.5% (w/v) glucose and 0.045% (w/v) D-fucose (33,24). After degradation of atp mRNA within 20 min, the expression of atpH and T7 gene1 (28) was induced by the addition of 0.1 mM IPTG 6 for 1 h.
Copper Cross-linking-Inverted membrane vesicles were prepared in the presence of protease inhibitor mix without EDTA (Sigma) according to Krebstakies et al. (23) using TMG buffer (50 mM Tris-HCl, pH 7.5, 10 mM MgSO 4 , 10% (v/v) glycerol). Cross-linking of membrane vesicles was carried out as described by Ballhausen et al. (4), incubating 5 mg/ml membrane protein in the presence of 1.5 mM copper-1,10-phenanthroline (CuP) in TMG, pH 7.5 or pH 8.2, for 1 h at 23°C. The reaction was stopped by the addition of 50 mM EDTA and 25 mM N-ethylmaleimide, and after 20 min, a quarter volume of 4ϫ SDS sample buffer without 2-mercaptoethanol (36) was added.
Purification of Polyhistidine-tagged Proteins-For isolation of F O F 1 via a His 6 tag fused N-terminally to subunit ␤, membranes were prepared from a 3-liter cell culture according to Wise et al. (37), and membrane proteins were extracted as described by Pänke et al. (38). After centrifugation, the solubilized proteins were incubated for 1.5 h on ice with 1 ml of Ni 2ϩ -nitrilotriacetic acid-Sepharose FF (GE Healthcare). Subsequently, the matrix was washed with 50 mM Tris-HCl, pH 7.5, 5 mM MgCl 2 , 1% (w/v) n-octyl-␤-D-glucopyranoside, 20 mM imidazole, 10% (v/v) glycerol and eluted with the same buffer containing 150 mM imidazole.
The solubilization of F O F 1 with a His 12 tag fused N-terminally to subunit a was performed according to Krebstakies et al. (22). 5 mg of membranes prepared in TMG buffer were pelleted, resuspended in 50 mM Tris-HCl, pH 7.5, 10% (v/v) glycerol, 0.1 mM PMSF, 1.4% (w/v) n-dodecyl-␤-D-maltoside, and stirred for 30 min at 16°C. After centrifugation, the supernatant was adjusted to 150 mM NaCl, 0.1 mM PMSF, and 20 mM imidazole (pH 7.5) and incubated end-over-end for 15 min with the matrix of a His SpinTrap TM column (GE Healthcare). Handling was performed as recommended by the supplier using 50 mM Tris-HCl, pH 7.5, 10% (v/v) glycerol, 0.1 mM PMSF, 150 mM NaCl, 20 mM imidazole, 1% (w/v) sodium cholate as binding buffer and the same buffer containing 500 mM imidazole as elution buffer. All buffers used included protease inhibitor mix without EDTA.
Assays-Protein concentrations were determined with the BCA assay (Pierce). Proteins were separated by SDS-PAGE using 10% separating gels (39) with PageRuler TM prestained protein ladder (Fermentas) as standard. Immunoblotting was performed according to Birkenhäger et al. (40). For immunolabeling, polyclonal antibodies raised in mice (anti-␣, anti-␤, anti-␦, anti-⑀) or rabbits (anti-␣, anti-␤, anti-␥, anti-b, anti-c) and monoclonal antibodies raised in rat (anti-␥) or mice (anti-a, anti-b, anti-c) against the individual subunits of F O F 1 were used as primary antibodies as indicated in the figures. IRDye TM 800DX-labeled goat-antimouse IgG, IRDye TM 800DX-labeled goat-anti-rat IgG, or IRDye TM 700DX-labeled goat-anti-rabbit IgG (Rockland) was applied as secondary antibody and detected with the two-channel system Odyssey (LI-COR). The secondary antibodies are affinity-purified for low cross-reactivities, and they allowed the simultaneous detection of fluorescence (shown in red or green) after immunolabeling of two proteins on one blot membrane. In the case of cross-linking of two different proteins, an overlay of both labels appears as a yellow band. Due to the varying avidities of the antibodies used, the color of the yellow bands may change to orange or lime green.
ATPase activities of membrane vesicles were determined as described (41). ATP-and NADH-driven proton pumping via ACMA fluorescence quenching was performed according to Birkenhäger et al. (40).

Formation of b Dimer Independent of Other F O F 1 Subunits-To
investigate the formation of the b dimer in the absence of other F O F 1 subunits, we performed nearest neighbor analyses by use of disulfide bond formation after substituting a single amino acid residue within the polypeptide chain to cysteine. We have chosen three cysteine substitutions that are spread over the polypeptide chain, namely bQ10C, bL65C, and bS139C, and compared the b-b formation in the presence and absence of other F O F 1 subunits. CuP was applied for generation of disulfide bonds using membranes with a bC21A substitution as a Cys-less control. Crosslinking was analyzed by immunoblotting after non-reducing SDS-PAGE (Fig. 1A). The cross-linking yields obtained for b-b assembled in F O F 1 (Fig. 1A, left) were comparable with those described previously (42)(43)(44). For bS139C, a nearly complete removal of monomeric b was observed, indicating a high crosslinking yield; however, the signal of b-b was very low, and the apparent molecular mass was lower than that observed for the other b-b cross-linking products, as has also been observed by McLachlin and Dunn (44). Both results imply a different protein folding within b-b for bS139C compared with dimerization by residues bQ10C or bL65C.
In the absence of other F O F 1 subunits, the immunoblot analysis also showed b-b formation with high cross-linking yields (Fig. 1A, right). In the presence of CuP, monomeric b was completely missing, and the cross-linking products were detected in comparable rates using the same antibody as described above. However, for bS139C, the cross-linking product was split into two bands, indicating the presence of two different conformations for b-b, both comparable with the different apparent molecular masses found for cross-linking of subunit b in F O F 1 . Furthermore, for bL65C and bS139C, dimer formation could already be observed even without CuP as oxidant, suggesting that the interacting SH-groups are even susceptible to oxidation by oxygen. In conclusion, membrane-bound b subunits showed dimer formation independent of the presence of other F O F 1 subunits.
No Interaction between b Dimer and the c 10 Ring in the Absence of Other F O F 1 Subunits-In the absence of other F O F 1 subunits, individually synthesized, membrane-bound subunits b and c have been shown to be organized in homooligomeric subcomplexes; namely, the c subunits are arranged in a ring of 10 monomers (4), whereas the b subunits are present in dimeric form (Fig. 1A). To answer the question of whether both subcomplexes interact with each other, as observed in F O F 1 using the cV74C/bN2C variant (25), the same experimental design as described above was used. For expression of individual subunits bN2C and cV74C, the modified genes atpEF were cloned in the same gene order, retaining the intergenic region as in the atp operon, and protein interaction was examined in membranes by oxidation with CuP. Despite massive overproduction of subunits b and c in the membranes due to the expression of these subunits with the pET system, no b-c cross-linking product could be detected in the corresponding immunoblot analysis (Fig. 1B), whereas b-c cross-linking was present in F O F 1 -containing membranes, as expected. In addition, c-c as well as b-b were monitored as a cross-linking side reaction. Although the proportion of monomeric and dimeric c subunits was comparable in both samples, although the expression level of subunit c is completely different, the dimerization of subunit b even in the absence of CuP as oxidant was unusually high (Fig. 1B), indicating that the flexible N-terminal region forced b-b formation when subunit c was not in close vicinity. In summary, the data demonstrate that the individually assembled membranebound subunit b dimer has no contact with the oligomeric c 10 ring when both subunits are expressed simultaneously in the absence of other F O F 1 subunits.
Presence of F O F 1 Subunits in Membranes of Single-subunit Knock-out Mutants ⌬a, ⌬b, and ⌬␦-Subunit a and subunit ␦ are both known to be tightly bound to the subunit b dimer to stabilize the stator part of F O F 1 versus the strong forces present during rotation of c 10 ␥⑀ in catalysis. Quantitative binding experiments revealed K d values in the range of 2-3 nM for the b␦ interaction in F O F 1 (22), and an ab 2 subcomplex can be purified from wild type membranes via an N-terminal His 12 tag fused to subunit a (41). To elucidate the need of subunit a and/or ␦ for the assembly of the b dimer into the F O F 1 complex, we have constructed single-subunit knock-out mutants, in which one of the F O F 1 subunits, namely a, b, or ␦, is no longer synthesized. For that purpose, the ⌬atpB-C strain DK8 was transformed with pBWU13 derivatives (Table 1) in which the synthesis of one of the eight F O F 1 subunits was prevented due to insertion of an early stop codon into the corresponding gene of the otherwise unchanged atp operon. In mutant ⌬␦, the stop codon was inserted at codon 11 of the atpH gene (␦Y11end) and in ⌬b at codon 7 of atpF (bI7end), whereas in ⌬a the stop codon was introduced at codon position aW231, a mutation that has been well characterized (11,45). Characterization of membranes of mutants ⌬a and ⌬␦ by immunoblotting revealed the presence of all F O F 1 subunits except the one knocked out ( Fig. 2A), indicating that the insertion of the additional stop codon induced no polar effect on the expression of the cistrons located downstream to the mutation in the polycistronic atp mRNA. However, it is important to note that no truncated subunit aW231end could be observed in all experiments performed, as has been discussed in detail by Hermolin and Fillingame (11).
In contrast, in membranes of ⌬b, also subunits a and ␦, which are tightly bound to the b dimer in F O F 1 , are completely missing in addition to subunit b. Such an interdependence between F O F 1 subunits during assembly has been previously described for subunit a to be interdependent on subunits b and c (11,46).
The immunoblot analysis ( Fig. 2A) showed that in membranes of mutant ⌬a, the amount of F O F 1 subunits present was comparable with wild type, whereas it was largely reduced in membranes of ⌬b and ⌬␦. To obtain roughly comparable signals in immunoblotting, we therefore increased the amount of protein for ⌬␦ by a factor of 7. This fits very well for all F O F 1 subunits except subunit c, which now revealed largely increased amounts of protein compared with wild type. For subunit c, comparable intensities for wild type and ⌬␦ were obtained by separation of identical amounts of membrane protein, as can be observed for the ⌬b mutant as well. Furthermore, intermolecular cross-linking via oxidation of bicysteine-substituted subunit c (cA21C/cM65C) (4, 47) produced an equal pattern of cross-link formation with all intermediate oligomers possible, stopping with the formation of decamers in either case (Fig.  2B). These observations demonstrate that the ringlike structure of the subunit c oligomer is not only independently assembled in the membrane (compare Refs. 4 and 47) but maintains high stability against degradation by proteases and is, therefore, present in ⌬␦ and ⌬b in amounts comparable with wild type in contrast to the other subunits of F O F 1 , indicating the presence of free c 10 subcomplexes.
The ATPase activities of membrane vesicles of the knock-out mutants compared with wild type exhibited a similar behavior ( Table 2). Although subunit a is missing and the proton pathway cannot be established (30), the ATPase activity of ⌬a is in a range comparable with wild type activities, providing evidence for a stable assembly of the remaining F O F 1 subunits. In membranes of ⌬␦ and ⌬b, the ATPase activities were reduced by a factor of 5-7, corresponding well to the amount of F O F 1 proteins present in the membrane. In addition, the interaction areas of the ␣ 3 ␤ 3 hexamer with subunit ␥ were verified by disulfide bond formation. Membranes containing subunit pairs with single cysteine substitutions were applied, confirming contact sites between subunits ␣-␥ (␣A334C/␥L262C) and ␤-␥ (␤D380C/␥C87) (Fig. 2B) comparable with wild type (27). Therefore, it can be concluded that at least an F O F 1 core complex composed of subunits c 10 ␣ 3 ␤ 3 ␥ exhibiting ATPase activity is assembled in membranes of ⌬␦ and ⌬b.
Cross-linking of Subunit b with Its Interacting Subunits ␣, ␤, and c in Mutants ⌬a and ⌬␦-In wild type F O F 1 , subunit b has contact with subunits a and c of F O and subunits ␣, ␤, and ␦ of F 1 . To characterize in more detail the status of b present in membranes of ⌬a and ⌬␦, nearest neighbor analyses were performed with wild type and mutant membranes containing the amino acid substitution pairs bL65C, bA92C/␣R477C, bA92C/ ␤Q351C, or bN2C/cV74C (Fig. 3); the localization of these residues within F O F 1 is summarized in Fig. 3A. To facilitate reading, the terms wild type (WT), ⌬a, and ⌬␦ exclusively refer to the mutations leading to partially assembled F O F 1 complexes independent of the presence or absence of cysteine substitutions. All cysteine pairs used are well characterized to generate disulfide bond formation in wild type with high yield (Fig. 3, B-E; compare Refs. 42, 44, and 48) and were applied as indicators for stable interactions between subunits in partially assembled F O F 1 complexes. The formation of b-b was comparable in ⌬a and ⌬␦ with that observed for solely expressed subunit b (Fig. 3B). In ⌬a, intense b-␣ and b-␤ heterodimer formation was observed after oxidation with CuP, indicating a proximal localization between b 2 and ␣ 3 ␤ 3 in this partially assembled F O F 1 complex (Fig. 3, D and E, left). However, no cross-linking could be obtained between subunits b and c in ⌬a, suggesting that the

Assembly of E. coli F O F 1 ATP Synthase
right positioning of the amino acid residues of both subunits involved needs the presence of subunit a (Fig. 3C, left). Both cysteine residues are located within terminal regions of the polypeptide chains, which are in general more flexible and, therefore, in most cases a benefit for cross-link formation. However, in this case, a stabilization by subunit a seems to be necessary for disulfide bond formation between b and c, and instead b-b formation is favored. In contrast, in ⌬␦, no cross-linking of subunit b to any of its nearest neighbors could be observed, indicating that in the absence of ␦, b 2 had no contact with a partially assembled F O F 1 core complex, at least within the area tested. In general, the ATPase activities observed were within the same range as those measured in the absence of the cysteine substitutions necessary for cross-linking (Table 2).
In the case of b-␣ cross-linking, as a side reaction, an ␣-␣ cross-linking was observed, which was analyzed by Brandt et al. (48) to be an interenzyme cross-link reaction. Substitution of a cysteine residue in ␣ resulted in incorporation of three thiol groups into F O F 1 . While one thiol group is proximal to the b dimer (b-␣ cross-linking in wild type or ⌬a F O F 1 ), the other two, located at the surface of the hexamer, will be freely available for other reactions. Because wild type membranes are rich in F O F 1 , such an intermolecular cross-linking is not unexpected. However, in membranes of mutant ⌬␦, the amount of F O F 1 in the membrane was reduced approximately by a factor of 7. Therefore, the probability of interenzyme cross-linking was also reduced, and the amount of ␣-␣ varied from experiment to experiment (data not shown).
Taken together, the results revealed that the b dimer can be assembled with the core F O F 1 complex in the absence of sub-unit a, forming a stable complex with interactions in the F 1 part comparable with functional F O F 1 , whereas an integration of b 2 into F O F 1 is not possible in the absence of subunit ␦.
Purification of Partially Assembled F O F 1 Complexes via N-terminally His 6 -tagged Subunit ␤ from Mutants ⌬a and ⌬␦-To further characterize the partially assembled core F O F 1 complex formed in ⌬a and ⌬␦, a His 6 tag was fused to the N terminus of subunit ␤, usually applied to purify ATP synthase complexes from detergent-solubilized wild type membranes (26,38). For ⌬a and ⌬␦, this approach was used to determine which F O F 1 subunits are stably associated with subunit ␤. After purification via Ni 2ϩ -nitrilotriacetic acid-agarose chromatography, the purified complexes were analyzed by immunoblotting in comparison with their corresponding membranes using antibodies against subunits b and ␥. Both subunits were in contact with subunit ␤ in wild type F O F 1 but did not interact with each other (Fig. 4A). In membranes, both subunits were present in comparable amounts after adjustment of the amount of membrane protein of ⌬␦ by a factor of 7, as described above (compare Fig.  2A and Table 2). After purification of F O F 1 , subunit ␥ was again present in comparable amounts in both mutants, as expected, without any adjustment. In contrast, for subunit b, a different picture emerged; whereas in ⌬a the amount of b was comparable with wild type F O F 1 , subunit b was not co-purified and, therefore, was completely missing in the subcomplex purified via His 6 -␤ by affinity chromatography from ⌬␦. To further characterize the subcomplexes isolated using His 6 -␤ as the bait, immunoblot analyses were carried out for all F O F 1 subunits (Fig. 4B). In ⌬a, all subunits except a, which was knocked out, were present in amounts comparable with wild type, verifying that an F O F 1 Ϫa complex can be stably assembled. In contrast, in ⌬␦, subunits ␦ and b could not be detected, and in addition, subunit a was only present in extremely low amounts and, therefore, barely visible. The results demonstrate that although subunits a and b are present in the cytoplasmic membrane of ⌬␦, they cannot be assembled into F O F 1 in the absence of subunit ␦, as has also been observed for subunit b by cross-linking (compare Fig. 3). Nevertheless, a protein complex, which comprises subunits ␣, ␤, ␥, ⑀, and c, could be isolated from ⌬␦ membranes by use of His 6 -␤ as ligand in affinity chromatography, implying the presence of an c 10 ␣ 3 ␤ 3 ␥⑀ subcomplex.
Purification of Partially Assembled F O F 1 Subcomplexes via N-terminally His 12 -tagged Subunit a from Mutant ⌬␦-Accordingly, in the next step, we studied the status of subunit a present in the membrane of ⌬␦ by affinity chromatography using a His 12 tag N-terminally fused to subunit a (23,41). A modified purification protocol was applied to membranes of wild type and ⌬␦. The immunoblot analysis using anti-a and anti-b antibodies simultaneously revealed the presence of both subunits in the eluate for wild type and ⌬␦ (Fig. 4C). In addition, the eluates obtained were screened for the presence of other F O F 1 subunits by immunoblotting (Fig. 4D). Although from wild type membranes, the entire ATP synthase complex was purified, only subunit b could be co-purified from membranes of ⌬␦ using His 12 -a as bait, indicating that a separate subcomplex is formed. Taken together, in the absence of subunit ␦, two F O F 1 subcomplexes are independently present in the cytoplasmic membrane, one comprising subunits ␣, ␤, ␥, ⑀, and c and the other containing the residual subunits a and b.
Formation of an ab 2 Subcomplex in Membranes of Mutant ⌬␦-To verify the presence of an independent ab 2 subcomplex in ⌬␦, we used an experimental set-up in which two cysteine pairs were combined for cross-linking of two different subunit contact areas. As shown in Fig. 5A for wild type membranes, in the presence of CuP as oxidant, bE155C located in the C-terminal region of subunit b was able to generate a b-b product, whereas the cysteine pair bA13C/aN238C located in the transmembrane region of both subunits formed a disulfide bond between subunit a and one of the b subunits (b-a). After combining both cross-linking pairs, an a-b-b cross-linked complex was obtained in high yield in the presence of CuP (b-b/b-a), exhibiting a molecular mass of ϳ65 kDa, as expected. Due to the presence of the cysteine substitutions, the ATPase activity was reduced by a factor of approximately 2 (Table 2); however, the DCCD sensitivity remained unchanged (78.4% inhibition), indicating a tight coupling between F 1 and F O in those membranes.
After introducing the corresponding cysteine residues into subunits a and b of mutant ⌬␦, the addition of CuP allowed the formation of an a-b cross-linking product in the presence of cysteine residues bA13C/aN238C (Fig. 5B) and, furthermore, the formation of a tertiary a-b-b product when, in addition, cysteine residue bA155C was provided (Fig. 5C), suggesting that an ab 2 subcomplex was purified using His 12 -a for affinity chromatography. In each case, b-b was observed as byproduct. In summary, membranes of ⌬␦ contained two F O F 1 subcomplexes, an ab 2 subcomplex and a c 10 ␣ 3 ␤ 3 ␥⑀ subcomplex, both verified by a combination of affinity purification and zero length cross-linking and the latter exhibiting ATPase activity, whereas in membranes of ⌬a, an F O F 1 complex lacking only subunit a (F O F 1 Ϫa) was assembled, showing ATPase activities comparable with wild type F O F 1 .

Time-delayed in Vivo Assembly of Subunit ␦ into Preformed F O F 1 Subcomplexes Generating a Functional ATP Synthase-The
presence of the two subcomplexes, ab 2 and ␣ 3 ␤ 3 ␥⑀c 10 , in membranes of ⌬␦, just missing knocked out subunit ␦, implies that ␦ is probably the last subunit being assembled into F O F 1 , as has been suggested by Rak et al. (19) for yeast mitochondrial F O F 1 . To answer the question, we applied our time-delayed in vivo assembly system (30) for a time-delayed integration of subunit ␦ into preformed F O F 1 subcomplexes missing subunit ␦.
The experimental design of the time-delayed in vivo assembly system is depicted in Fig. 6. After inoculation, P araBAD was induced by arabinose to allow synthesis of F O F 1 subunits except ␦ (F O F 1 Ϫ␦). During this growth phase, the lac operator-controlled promoters were repressed (compare Fig. 8B) due to pre- incubation of the medium with ␤-galactosidase for removal of residual lactose and the use of TTG as a start codon for atpH. At OD ϭ 0.3, P araBAD was repressed by the simultaneous addition of the catabolite repressor glucose and the anti-inducer D-fucose, a non-metabolizable analog of L-arabinose, and, after further growth for 20 min, nearly all atp mRNA present in the cell was degraded (see below). Due to this time delay, the de novo biosynthesis of F O F 1 subunits was completely prevented. Sub-sequently, IPTG induction of lac operator-controlled promoters enabled the individual synthesis of subunit ␦, which might be assembled together with the preformed subcomplexes ab 2 and c 10 ␣ 3 ␤ 3 ␥⑀ (F O F 1 Ϫ␦ϩ␦) to generate functional F O F 1 .
The stability of both subcomplexes formed was controlled by immunoblotting applying antibodies against subunits a, b, ␣, and ␥ to analyze both subcomplexes individually (Fig. 7A). Cells of F O F 1 Ϫ␦ were grown to OD ϭ 0.3 in the presence of arabinose (Ara) before the addition of glucose and D-fucose (Glu/Fuc), and in defined time intervals, samples were taken for analysis of F O F 1 subunits in cell lysates. Whereas the amount of ␣ and ␥ was stable over a period of at least 60 min, the immunodetection of a and b revealed a continuous degradation of ab 2 , starting at the latest after 30 min, and implies that the ab 2 subcomplex is more susceptible to degradation by proteases than the core complex c 10 ␣ 3 ␤ 3 ␥⑀. In parallel, the decomposition of the atp mRNA was controlled by rt-RT-PCR, revealing a complete degradation 15 min after the addition of Glu/Fuc (Fig. 7B). Therefore, the delay time for the degradation of atp mRNA, being simultaneously the starting point for the synthesis of ␦, was set to 20 min.
Different cell batches of F O F 1 Ϫ␦ were grown, each with the additives added at the time points addressed in Fig. 6. As a control, samples expressing wild type F O F 1 were handled correspondingly. Under Ara-induced conditions, atp mRNA was present within the cells, as shown by rt-RT-PCR (Fig. 8A), although the amount was reduced in F O F 1 Ϫ␦ by a factor of approximately 3 compared with wild type. In samples containing Glu/Fuc independent of the presence of Ara, atp mRNA was almost completely absent (0.3-2%), revealing that the synthesis of F O F 1 subunits is prohibited.
Immunoblotting using antibodies against subunits b and ␥ as representatives of the two subcomplexes formed also revealed a reduction for F O F 1 Ϫ␦ compared with F O F 1 under Ara-induced conditions (Fig. 8B). In the presence of Glu/Fuc but in the absence of Ara, no F O F 1 subunits were detectable in F O F 1 Ϫ␦, whereas extremely low amounts could be observed in wild type F O F 1 , which were quantitated to be below 2% by determining ATPase activity (Fig. 8C) but remained below the detection limit for ATP-driven H ϩ translocation (Fig. 8D). Samples induced with Ara and subsequently repressed by Glu/Fuc showed reduced but significant signals by immunolabeling, and again the ATPase activities observed correlated very well (Fig. 8,  B and C). In addition, the extent of subunit degradation after repression with Glu/Fuc (in total for 80 min) was comparable with that observed in Fig. 7A. Subunit ␦ was not detectable in membranes of F O F 1 Ϫ␦; however, after expression of atpH, it was overproduced beyond stoichiometric amounts (F O F 1 Ϫ␦ϩ␦; Fig. 8B). Importantly, a stabilizing effect, especially on subunit b, could be observed due to induction of atpH expression, whereas the  amount of ␥ as well as membrane-bound ATPase activity remained unchanged (Fig. 8, B and C). After repression of atp transcription by Glu/Fuc, degradation of subunits was observed for F O F 1 as well as F O F 1 Ϫ␦ϩ␦. Nevertheless, in the absence of subunit ␦ (F O F 1 Ϫ␦), the degradation of b is increased, supporting the notion that ␦ protects the ab 2 subcomplex against proteolytic digestion probably by its integration into the holoenzyme, whereas the stability of the c 10 ␣ 3 ␤ 3 ␥⑀ remained unaffected.
The acidification of the lumen of membrane vesicles by ATPdriven H ϩ translocation was measured using ACMA as a pH-sensitive dye and applied to determine the assembly of a functional F O F 1 complex from subcomplexes (Fig. 8D). Under Ara-induced conditions, the fluorescence quenching rates obtained for F O F 1 were within the range expected (40,49). After the addition of Glu/Fuc in the presence of Ara, the quenching rate was only slightly decreased, although the ATPase activity revealed that the amount of F O F 1 in the membrane had been reduced to one-third, a discrepancy only at first glance that is probably due to a less dense protein packing within the membrane and, therefore, a significantly decreased membrane proton permeability (23,50), leading to a stronger acidification of the vesicle lumen. Under repressed conditions, no H ϩ transport could be observed for F O F 1 as well as F O F 1 Ϫ␦, although H ϩ pumping via the respiratory chain with NADH as substrate revealed intactness of the vesicles (data not shown), indicating a tight repression of P araBAD by Glu/Fuc. In contrast, F O F 1 Ϫ␦ showed extremely low H ϩ pumping rates (below 3%) in the presence of Ara. However, these rates were independent of the presence/absence of Glu/Fuc, implying that now and then spontaneously an H ϩ -conducting unit could be formed from the two individual subcomplexes containing subunits a and c, an assumption corroborated by the complete absence of ATPdriven H ϩ transport in otherwise identical experiments performed with F O F 1 Ϫa (30); an observation that requires further investigations to clarify whether subunit ␦ increases the affinity of ab 2 for the F O F 1 core complex or whether it changes the conformation of subunit a to an active one. However, after synthesis of subunit ␦ (F O F 1 Ϫ␦ϩ␦) the quenching rate increased to 21.4%, demonstrating that functional F O F 1 was assembled (Fig.  8D). The rate obtained fits well with the corresponding ATPase activity, and furthermore, both values are in good correlation with the activities of wild type F O F 1 (in the presence of Ara and Glu/Fuc), indicating that both enzyme complexes independent of their assembly pathway exhibit comparable ATPase activities. Furthermore, ATP-driven H ϩ transport was completely abolished after preincubation of F O F 1 as well as F O F 1 Ϫ␦ϩ␦ membranes with DCCD (data not shown). DCCD, which modifies specifically the protonated carboxyl groups of Asp-61 of subunit c at pH 8.0, inhibits the rotation of the c 10 ring and, thereby, H ϩ translocation. Taken together, a functional F O F 1 complex was assembled from subcomplexes ab 2 and c 10 ␣ 3 ␤ 3 ␥⑀ by time-delayed synthesis of subunit ␦. FIGURE 6. Design of the time-delayed in vivo assembly system and its different states during cell growth. ⌬atpB-C strain DK8 was transformed with three plasmids bearing different resistance genes and different origins to gain compatibility: (i) a pBAD33 derivative (p15A ori; Cm R ) with structural genes of the atp operon carrying an early stop codon (␦Y11end) in atpH (atpBEFH*AGDC) under control of P araBAD ; (ii) a pET-22b derivative (pMB1 ori, Ap R ) carrying the atpH gene under control of the IPTG-inducible T7 lac promoter using the weak start codon TTG for atpH; (iii) a pSC101 derivative (Kan R ) containing T7 gene1 under control of the IPTG-inducible T5N25 lac promoter, encoding the RNA polymerase specific for promoters of phage T7. Left, cells were inoculated with OD ϭ 0.05; P araBAD was induced by arabinose to allow expression of the atpBEFH*AGDC genes, and the F O F 1 subunits except for subunit ␦ (F O F 1 Ϫ␦) were synthesized. During this growth phase, the lac operator-controlled promoters are completely repressed. Middle, at OD ϭ 0.3, P araBAD is repressed by the simultaneous addition of glucose and D-fucose, and after further growth for 20 min, the atp mRNA present within the cell is completely degraded. Due to this time delay, the de novo biosynthesis of F O F 1 subunits is prevented. Right, time-delayed IPTG induction of lac operator-controlled T7 and T5N25 promoters for 1 h enables the synthesis of subunit ␦ assembling together with the preformed F O F 1 subcomplexes ab 2 and c 10 ␣ 3 ␤ 3 ␥⑀ (F O F 1 Ϫ␦ϩ␦), a functional F O F 1 complex. Red cross, repressed or uninduced state of promoters.

DISCUSSION
Generation of the H ϩ -translocating Unit-The F O complex of E. coli ATP synthase is composed of subunits a, b, and c present as ab 2 c 10 . Wild type membranes depleted of F 1 revealed an F O complex functional in H ϩ conduction as well as rebinding of F 1 (40,49). Furthermore, purification of F O subunits and reconstitution into liposomes revealed the same characteristics (51,52). In contrast, after synthesis of F O subunits in the absence of F 1 , only low levels of proton permeability were observed in membranes as well as after reconstitution into liposomes, implying a dependence on F 1 for the proton translocation by F O (53,54), although the additional presence of subunit ␦ enhances the proton conduction through F O (55,56).
Our results indicate that subunit ␦ is the key player in generating the functional H ϩ -translocating unit composed of ac 10 . The position of b 2 in F O F 1 is restricted by its specific interactions with subunits a and ␦ at both terminal regions. The characterization of partially assembled F O F 1 in ⌬a and ⌬␦, however, revealed that both interactions do not influence each other; the binding of a and ␦ to b 2 are two unrelated processes apparently without major conformational changes. Whereas in ⌬a, b 2 interacts with ␦, thereby enabling its integration into F O F 1 , in ⌬␦, a specific ab 2 subcomplex is formed by specific interaction with subunit a. Moreover, timedelayed binding of the second interaction partner (subunit ␦ (Fig.  8) or subunit a (30)) is possible without any preassigned assembly sequence, leading to the formation of a functional F O F 1 complex in each case. As a consequence, it can be concluded that subunit ␦ is functioning as a clamp to induce a first contact between ab 2 and c 10 (via ␣ 3 ␤ 3 ␥⑀) to generate the H ϩ -conducting unit within the membrane because the exclusive presence of both subcomplexes in ⌬␦ is not sufficient to constitute an open H ϩ channel. By this arrangement, two important characteristics of F O F 1 are respected. First, an intermediate assembly product exhibiting uncontrolled H ϩ conduction is avoided, which would lead to a breakdown of the proton motive force and, therefore, be lethal for the cell. Instead, it is guaranteed that an open H ϩ channel is concomitantly assembled within a coupled F O F 1 complex, thereby preventing membrane proton permeability. Second, due to the rotation of the c 10 ring versus the stator part ab 2 , it is essential for function that the affinity between stator and rotor subunits in F O is low. In addition, the recent 9.7 Å cryoelectron microscopy structure of the A-type ATP synthase of Thermus thermophilus (57) revealed a surprisingly small contact area between the rotor ring and the I protein (equivalent to subunit a of F-type ATP synthases). This low affinity is elegantly bypassed by subunit ␦ holding both complexes of F O , ab 2 as well as c 10 (via ␣ 3 ␤ 3 ␥⑀), in position, with b 2 providing the necessary stiffness.
Nonetheless, as described above, once the F O complex has been generated, a depletion of F 1 is possible without major changes in its H ϩ -conducting capabilities, indicating that the first contact between ab 2 and c 10 enabled by ␦ triggers a kind of induced fit within subunits a and c that later on facilitates binding between F O subunits in the absence of F 1 . However, whether subunit ␦ simply increases the affinity of ab 2 for the rest of the complex or whether it changes the conformation of F O to an active one, as suggested previously (55,56), remains to be solved in further investigations.
Assembly of E. coli ATP Synthase from Subcomplexes-Our results support the notion that E. coli F O F 1 is assembled from different subcomplexes (summarized in Fig. 9). The formation of b 2 is independent of the presence of other F O F 1 subunits, as shown by zero length cross-linking. With the same experimental design, it has recently been demonstrated that the c 10 ring is also stably organized in the absence of other F O F 1 subunits (4). Furthermore, a simultaneous but exclusive expression of subunits b and c in stoichiometric amounts allowed no b-c cross-link formation using a contact site characterized for wild type F O F 1 (25). Although both subunits are overexpressed, no interaction between subunits b and c could be observed in the absence of other F O F 1 subunits, which is in contrast to the model for assembly of E. coli F O F 1 proposing a b 2 c 10 subcomplex as an intermediate product (59,60) but in accordance with models proposed for yeast mitochondrial ATP synthase (19,58,61).
Characterization of ⌬␦ and ⌬b revealed the presence of the membrane-bound core complex c 10 ␣ 3 ␤ 3 ␥⑀ possessing ATPase activity with rates comparable with wild type in relation to the amount of protein present. Furthermore, due to the absence of subunits a, b, and ␦ in ⌬b, it can be concluded that core complex formation occurs independently of other subcomplexes. Until now, only little experimental data regarding the order of the assembly of the E. coli F 1 part have been available. However, in vitro reconstitution experiments (13,62,63) support an assembly in the cytoplasm prior to its binding to the c ring comparable with that described for yeast mitochondrial F 1 (19,64,65). Therefore, an assembly of the F O F 1 core complex from two individual subcomplexes c 10 and ␣ 3 ␤ 3 ␥⑀ is proposed, with ␣ 3 ␤ 3 ␥ forming the minimal unit exhibiting ATPase activities (12,62,66) and subunit ⑀ being necessary for a stable binding to the membrane via c 10 (63,67). Preliminary experiments on chromosomally expressed ATP synthase containing His 6 -␤ enabled the purification of subunits ␣, ␤, ␥, and ⑀ exhibiting ATPase activity. 7 In mutant ⌬␦, a separate ab 2 subcomplex is present in the membrane in addition to the ATP-hydrolyzing F O F 1 core complex. Furthermore, time-delayed synthesis of subunit ␦ enabled a subsequent assembly of functional ATP synthase, supporting the notion that both subcomplexes are present as native assembly intermediates. In addition, a reduction in synthesis of subunit c due to a point mutation in the ribosome binding site of atpE revealed the presence of free ab 2 subcomplexes that can be reconstituted in liposomes with subunit c after purification to form F O complexes functional in passive H ϩ translocation as well as F 1 binding (23). Due to the plasmid-encoded overexpression of F O F 1 subunits, free excess subunits are inevitably present in the cell, and therefore, the assembly of active F O F 1 complexes from these isolated subunits during time-delayed synthesis of subunit ␦ is at least conceivable. However, the observation that the amount of subunit ␥ present in the membrane as well as the membrane-bound ATPase activity remained unchanged after time-delayed expression of subunit ␦ revealed that ab 2 as well as c 10 ␣ 3 ␤ 3 ␥⑀ are true assembly intermediates instead of dead end products of incomplete assembly.
Last Subunit Being Assembled into F O F 1 -Whereas in ⌬a, a single subcomplex just missing subunit a is assembled, in ⌬␦, two subcomplexes, ab 2 and c 10 ␣ 3 ␤ 3 ␥⑀, are present in the mem-7 G. Deckers-Hebestreit, unpublished observation.  . Model for the assembly of E. coli ATP synthase. The assembled c 10 ring and b 2 are independently present in cytoplasmic membranes. Formation of ␣ 3 ␤ 3 ␥⑀ is not known in detail for E. coli. For F O F 1 from yeast mitochondria, it has been proposed that two subcomplexes, ␣ 3 ␤ 3 hexamer and ␥⑀, are assembled, which combine to form ␣ 3 ␤ 3 ␥⑀, before it subsequently binds to the membrane-bound c 10 ring (58,19). The membrane insertion of subunit a is interdependent on the presence of b and c. Integration of b 2 into the complex is dependent on subunit ␦, whereas binding of a as well as ␦ to b 2 possibly has no preferred sequence. SEPTEMBER  brane. Furthermore, in both cases, a time-delayed synthesis of the missing subunit enabled the formation of functional ATP synthase, suggesting that each subunit can be assembled as the last subunit into preformed subcomplexes. Comparable results have been described for Bacillus PS3 as well as mitochondrial F O F 1 . Thermophilic Bacillus PS3 F O F 1 lacking subunit a (F O F 1 Ϫa) can be isolated in stable form from E. coli membranes, and after co-reconstitution with independently purified or cell-free synthesized subunit a, both components form a functional enzyme complex in liposomes (68,69). In addition, from human 0 cells, a stable F O F 1 complex lacking subunits a (Atp6p) and A6L (an additional subunit not present in bacterial F O F 1 ) was purified (70). On the other hand, the data of Rak et al. (19) revealed that F O F 1 is assembled in a modular way from at least three different modules, F 1 , the c ring, and a stator subcomplex. No information is given, however, for the assembly of OSCP (homologous to bacterial ␦) into F O F 1 , although it has been deduced that OSCP might may be last subunit being assembled.

Assembly of E. coli F O F 1 ATP Synthase
This seems to be contradictory at first sight; however, it is probably due to the manipulative conditions applied. In each case, the use of mutants provokes an arrest in the assembly pathway at a certain point, whereas in wild type cells, all subunits are nearly isochronously at hand by their simultaneous translation from polycistronic mRNA (15,71). Apparently, subunit a as well as subunit ␦ could bind to preformed b 2 , probably simultaneously due to their autonomous interaction, therefore implying that in vivo a subcomplex of ab 2 ␦ is formed prior to its binding to c 10 ␣ 3 ␤ 3 ␥⑀. This is further supported by the observation that in ⌬b in addition to b, also subunits a and ␦ were absent. In vitro studies with purified subunit ␦ and the hydrophilic part of subunit b (comprising amino acid residues 34 -156) revealed the formation of a b 2 ␦ complex (72). On the other hand, subunit ␦ is highly susceptible to proteolytic digestion and, therefore, rapidly degraded in most single-subunit knockout mutants producing partially assembled F O F 1 complexes. 7 A proteolytic fragment of subunit ␦ comprising amino acid residues 1-134 has been characterized (73), although the corresponding protease has not yet been identified. In addition, in the absence of subunit ␦, ab 2 as well as ␣ 3 ␤ 3 ␥⑀ were also degraded by proteases in appreciable amounts. Certainly, this rapid turnover indicates a tight regulation of ␦ and underlines its importance in generating functional F O F 1 . Whether ␦ makes the first contact with its C-terminal region with b 2 (74), forming an ab 2 ␦ subcomplex, or whether it is first bound with its N-terminal region to the F O F 1 core complex via subunit ␣ (13, 32), forming a c 10 ␣ 3 ␤ 3 ␥␦⑀ subcomplex, is the next question to be investigated.