A Functionally Important Hydrogen-bonding Network at the βDP/αDP Interface of ATP Synthase*

ATP synthase uses a unique rotary mechanism to couple ATP synthesis and hydrolysis to transmembrane proton translocation. The F1 subcomplex has three catalytic nucleotide binding sites, one on each β subunit, at the interface to the adjacent α subunit. In the x-ray structure of F1 (Abrahams, J. P., Leslie, A. G. W., Lutter, R., and Walker, J. E. (1994) Nature 370, 621–628), the three catalytic β/α interfaces differ in the extent of inter-subunit interactions between the C termini of the β and α subunits. At the closed βDP/αDP interface, a hydrogen-bonding network is formed between both subunits, which is absent at the more open βTP/αTP interface and at the wide open βE/αE interface. The hydrogen-bonding network reaches from βL328 (Escherichia coli numbering) and βQ441 via αQ399, βR398, and αE402 to βR394, and ends in a cation/π interaction between βR394 and αF406. Using mutational analysis in E. coli ATP synthase, the functional importance of the βDP/αDP hydrogen-bonding network is demonstrated. Its elimination results in a severely impaired enzyme but has no pronounced effect on the binding affinities of the catalytic sites. A possible role for the hydrogen-bonding network in coupling of ATP synthesis/hydrolysis and rotation will be discussed.

F 1 F 0 -ATP synthase catalyzes the final step of oxidative phosphorylation and photophosphorylation, the synthesis of ATP from ADP and inorganic phosphate. F 1 F 0 -ATP synthase consists of the membrane embedded F 0 subcomplex with, in Escherichia coli, a subunit composition of ab 2 c 10 , and the peripheral F 1 subcomplex, with a subunit composition of ␣ 3 ␤ 3 ␥␦⑀. The energy necessary for ATP synthesis is derived from an electrochemical transmembrane proton (or, in some organisms, sodium ion) gradient. Proton flow, down the gradient, through F 0 is coupled to ATP synthesis on F 1 by a unique rotary mechanism. The protons flow through (half) channels at the interface of a and c subunits, which drives rotation of the ring of c subunits. The c 10 ring, together with F 1 subunits ␥ and ⑀, forms the rotor. Rotation of ␥ leads to conformational changes in the catalytic nucleotide binding sites on the ␤ subunits, where ADP and P i are bound. The conformational changes result in formation and release of ATP. Thus, ATP synthase converts electro-chemical energy, the proton gradient, into mechanical energy in form of subunit rotation and back into chemical energy as ATP. In bacteria, under certain physiological conditions, the process runs in reverse. ATP is hydrolyzed to generate a transmembrane proton gradient, which the bacterium requires for such functions as nutrient import and locomotion (for reviews, see Refs. 1-6). F 1 (or "F 1 -ATPase") has three catalytic nucleotide binding sites, with pronounced differences in their nucleotide binding affinity. The catalytic nucleotide binding sites are located on the ␤ subunits, at the interface to the adjacent ␣ subunit. In the original x-ray structure of bovine mitochondrial F 1 (7), one of the three catalytic sites was filled with the ATP analog AMP-PNP, 2 a second one with ADP (plus azide; see ref. 8), and the third site was empty. Hence, the ␤ subunits are referred to ␤ TP , ␤ DP , and ␤ E , and the ␤/␣ subunit pairs as ␤ TP /␣ TP , ␤ DP /␣ DP , and ␤ E /␣ E , respectively. The three catalytic ␤/␣ interfaces differ in their degree of interaction between both subunits. The ␤ DP / ␣ DP interface has the most extensive contacts between ␤ and ␣, the ␤ TP /␣ TP interface has less, and the ␤ E /␣ E interface has the least. These differences are most pronounced at the C-terminal domains of ␤ and ␣. Unique to the ␤ DP /␣ DP interface is an inter-subunit hydrogen-bonding network, reaching from the main-chain oxygen of ␤L328 3 and the side chain of ␤Q441 via the side chains of ␣Q399, ␤R398, and ␣E402 to the side chain of ␤R394. The guanidino group of ␤R394 forms a cation/ interaction with the phenyl ring of ␣F406 (Fig. 1, left-hand panel). At the ␤ TP /␣ TP interface, the ␣-helix carrying residues ␣Q399, ␣E402, and ␣F406 is rotated and has moved away from ␤ by 3-4 Å, preventing all these interactions (Fig. 1, right-hand panel). At the ␤ E /␣ E interface, the C-terminal domains of two subunits are even further apart.
The amino acids that make up the hydrogen-bonding network are conserved between bovine mitochondrial ATP synthase and the enzyme from E. coli. In the report presented here, we investigate the functional importance of this hydrogenbonding network at the ␤ DP /␣ DP interface of E. coli ATP synthase by perturbing it to varying degrees by site-directed mutagenesis. Single conservative amino acid substitutions were used to change charges and/or geometry of individual hydrogen bonds, single alanine substitutions removed one, two, or three possible hydrogen bonds, and a triple alanine mutant, ␤R394A/ ␤R398A/␤Q441A, prevented formation of all hydrogen bonds except one, as well as formation of the cation/ interaction. Analysis of the mutants demonstrated the functional impor-tance of the hydrogen-bonding network. A possible role for the network in energy coupling will be discussed.

EXPERIMENTAL PROCEDURES
E. coli Strains and Plasmids-The source of wild-type F 1 was strain SWM1 (9), the source of ␤Y331W mutant F 1 was strain pSWM4/JP17 (10). The template for mutagenesis was plasmid pSN6 (11); thus, all new mutants would carry in addition the ␤Y331W mutation, to allow fluorescence-based nucleotide binding measurements. Mutagenesis was performed using the QuikChange II XL kit (Stratagene). The mutagenic oligonucleotides were designed in such a way that, in addition to the desired mutation, a restriction site would be eliminated or generated, to facilitate screening: Plasmids containing the desired mutation were transformed into strain DK8 (12).
Preparation of Membranes and Enzymes and Functional Analysis of Mutant Strains and Enzymes-Growth yields of E. coli strains in limiting glucose and growth on succinate plates were measured as described previously (13). E. coli membranes were prepared as described (14). F 1 was isolated as described previously (15). NADH-and ATP-driven proton pumping in membranes was measured using acridine orange fluorescence quenching (16). The amount of F 1 in membrane preparations was estimated via Western blots, using an anti-␤ antibody (Agrisera, Vännäs, Sweden) or an anti-␣/anti-␤ antibody (kindly provided by Dr. Bill Brusilow, Wayne State University). The staining intensity was quantified using a Photodyne imaging system and the program Image J (National Institutes of Health). ATPase activities were measured in 50 mM Tris/ H 2 SO 4 , 10 mM ATP, 4 mM MgSO 4 , pH 8.0, at 23°C. Released phosphate was determined as described (17,18).
Fluorescence Measurements-A general outline of nucleotide binding experiments using ␤Trp331 fluorescence has been described previously (15). The experiments were performed on a spectrofluorometer type Fluorolog 3 (HORIBA Jovin Yvon, Edison, NJ), at 23°C. The buffer was 50 mM Tris/H 2 SO 4 , 2.5 mM MgSO 4 , pH 8.0. F 1 concentration was 30 -60 nM. ATP and ADP were added as indicated. To determine MgADP binding in the presence of scandium fluoride, the enzyme was incubated with 0.3 mM ScCl 3 , 10 mM NaF, 2.5 mM MgSO 4 , plus the indicated concentration of ADP (19). To correct for dilution and inner filter effects, parallel titrations with wild-type F 1 were performed. K d values were determined from the titration curves as described previously (15).
Modeling-Homology modeling, including energy minimization refinement, was performed using the program PRIME (Schroedinger Inc.). Templates were Protein Data Bank entries 1BMF and 1H8E, the structures of bovine mitochondrial F 1 with two (7) and three (20) catalytic sites occupied, respectively. Due to program restrictions regarding the number of amino acids, the N-terminal domains of the ␣ and ␤ subunits were removed.

RESULTS
Assaying the Mutants: An Overview-With the exception of residue ␤L328, which participates in the ␤ DP /␣ DP hydrogenbonding network via its main-chain carbonyl oxygen, the function of the other residues involved in the network was studied by mutational analysis. In addition, residue ␣F406, which forms a cation/ interaction with one of the terminal residues of the network, ␤R394, was included in the analysis. The results of the mutagenesis experiments are summarized in Table 1. To characterize the functional effect of each mutation on oxidative phosphorylation in vivo, we determined growth yields in 3 mM glucose and growth on plates containing succinate as the sole carbon source. After preparation of membranes from the mutant strains, we analyzed NADH-and ATP-driven proton gradient formation by measuring quenching of acridine orange fluorescence, and we measured ATPase activity. A low percentage of NADH-induced quenching is indicative of proton permeability of the membranes, which can be caused by oligomeric instability of the mutant ATP synthase complex; of the mutations tested here, only the ␣Q399C mutant showed a pronounced reduction in NADH-induced acridine orange quenching. The degree of quenching by ATP depends on the balance between ATP-driven proton pumping and proton permeability of the membranes. To determine if variations in enzymatic activities were direct functional consequences of the respective mutation or due to changes in the amount of enzyme on the membrane, we quantified the amount of F 1 present in the membrane preparations by Western blots. For most mutants, the amount of F 1 was within Ϯ25% of that observed for the parental strain pSN6/DK8. Exceptions were the ␤Q441A mutant, which had ϳ2/3 of the F 1 content of the control, the ␤Q441C mutant, which contained ϳ50% more F 1 than the control, and the ␣Q399A mutant, where no F 1 was found on the membrane.
The results of the experiments listed in Table 1 show that the hydrogen-bonding network plays an important role in the function of the enzyme. Many of the mutants showed impaired activity. In most cases, the effect of the mutations on ATP synthesis and hydrolysis was comparable. Although, in general, conservative substitutions had less influence on the enzymatic function less than an alanine replacement, in some cases these differences were smaller than expected. In the following, the results of the mutagenesis experiments will be analyzed in detail.
Mutation of ␤Q441-The side chain of residue ␤Q441 can form a hydrogen bond with the side chain of residue ␣Q399. The mutagenesis results suggest that this hydrogen bond can still be formed when the side chain of ␤Q441 is shortened by one methylene group to asparagine, because the ␤Q441N mutant has close-to-normal functional properties. Even the ␤Q441C mutant seems to have preserved some hydrogen-bonding capabilities. Abolishing the possibility for hydrogen bond formation in the ␤Q441A mutant impairs the enzyme significantly.
Mutation of ␣Q399-The side chain of residue ␣Q399 has a central position in the ␤ DP /␣ DP hydrogen-bonding network. It can form bonds to the side chains of ␤Q441 and ␤R398 as well as to the main-chain carbonyl oxygen of ␤L328. Like for ␤Q441, the results of the functional assays suggest that shortening the side chain of ␣Q399 in the ␣Q399N mutant leaves the functionally important hydrogen bonds largely unperturbed. Modeling (not shown) of the ␣Q399N mutation suggested that the hydrogen bonds to the side chains of ␤Q441 and ␤R398 might be preserved; in addition, a new hydrogen bond to the hydroxyl group of ␤Y444 seemed possible. Eliminating all potential hydrogen bonds in which the side chain of ␣Q399 is involved, in the ␣Q399A mutant, appears to prevent assembly of the enzyme. Visual inspection of the ␤ DP /␣ DP interaction site indicates that the ␣Q399C mutant should be capable to form at least one intersubunit hydrogen bond, most likely to ␤Q441. However, this residual hydrogen-bonding capability does not seem enough

Functional properties of strains and membranes containing mutations perturbing the ␤ DP /␣ DP hydrogen-bonding network
Growth yield in limiting (3 mM) glucose and growth on succinate plates were determined as in (13). Growth yield data were measured via the turbidity (A 590 ) and are expressed as percentage of the value for the positive control. Quenching of acridine orange fluorescence by NADH and ATP was measured as in (16). The determination of the amount of F 1 on the membranes is based on the quantitative evaluation of western blots by densitometric analysis (see "Experimental Procedures"). ATPase activities were determined in 50 mM Tris/H 2 SO 4 , 10 mM ATP, 4 mM MgSO 4 , pH 8.0, at 23°C. The relative ATPase activity in the last column is expressed as percentage of the activity of the positive control, corrected for the different amounts of F 1 on the membrane. Strain pSN6/DK8 served as positive control; it expresses ATP synthase containing a ␤Y331W mutation. ␤Y331W mutant ATP synthase is a normal, active enzyme (10,15,46). All mutant strains described in the table were derived from strain pSN6/DK8. Strain pUC118/DK8 does not express ATP synthase and served as negative control.  4 The plasmid containing the ␣F406W mutation was derived from plasmid pBOW1 (47), encoding ATP synthase with Trp-free F 1 . The plasmid containing the ␣F406C mutation was derived from pBWU13.4 (48), encoding wild-type ATP synthase. Growth yields are given as percentage of the growth yield of the wild-type control (pBWU13.4/DK8). From both mutant strains an active F 1 could be isolated. c Mutations were described previously (34,35). In all cases, wild-type enzyme was used as background. Growth yields and membrane ATPase activities are expressed as percentage of the respective values for the wild-type control. ␤R398W mutant F 1 was isolated and showed normal function (49). d Maximum standard deviation for the values in the respective column. All assays were run at least in triplicate, except for the determination of the amount of F 1 on the membrane which was done at least in duplicate.

Strain/mutation
to overcome the loss of the other hydrogen bonds; the ␣Q399C mutant is very strongly impaired.
Mutation of ␤R398-The guanidino group of ␤R398 can form hydrogen bonds with the side chains of ␣Q399 and ␣E402, plus an intra-subunit hydrogen bond to the main-chain carbonyl oxygen of ␤Q441. These hydrogen bonds seem of less importance for the functionality of the enzyme. Even a complete removal in the ␤R398A mutant gives an enzyme with considerable activity.
Mutation of ␣E402-The carboxylate group of ␣E402 can form hydrogen bonds with the guanidino functions of ␤R398 and ␤R394. Loss of the negative charge in the ␣E402Q mutant has only a minor effect on enzymatic function. Shortening the side chain by removal of one methylene group in the ␣E402D mutant reduces the activity by about a half, possibly by preventing formation of one of the hydrogen bonds. Loss of both hydrogen bonds in the ␣E402A mutant leaves an enzyme with overall Ͻ20% residual activity.
Mutation of ␤R394-The guanidino group of ␤R394 can form a hydrogen bond with the carboxylate group of ␣E402, plus a cation/ interaction with the phenyl ring of ␣F406. In addition, formation of intra-subunit hydrogen bonds with the hydroxyl function of ␤Y367 and the carboxylate group of ␤E440 seems likely. Complete removal of the hydrogen-bonding capability in the ␤R394A mutant reduces the activity severely. Interestingly, in the more conservative mutants, ␤R394K and ␤R394Q, a similar decrease in activity was observed. A possible explanation is that the lysine and glutamine side chains preferentially form intra-subunit hydrogen bonds, which are not required for the activity of the enzyme, but fail to provide the functionally important interactions with the ␣ subunit.
Mutation of ␣F406-The phenyl ring of ␣F406 can make a cation/ interaction with the guanidino group of ␤R394. Elimination of this interaction, in the ␣F406A mutant, results in a significant impairment of the enzyme. Earlier studies 4 had shown that the phenyl ring can be replaced by an indole ring system, in the ␣F406W mutant, without loss of activity. The ␣F406C mutant shows considerably higher activity the ␣F406A mutant, which might suggest that the cation/ interaction between the side chains of ␣F406 and ␤R394 has been substi-tuted by a hydrogen bond between the sulfhydryl group of the Cys and the guanidino group of the Arg, with partial preservation of functionality.
The ␤R394A/␤R398A/␤Q441A Triple Mutant-In an attempt to remove the ␤ DP /␣ DP hydrogenbonding network as completely as possible, without preventing assembly of the enzyme, we constructed a ␤R394A/␤R398A/␤Q441A triple mutant. (It is actually a quadruple mutant as the plasmid used to generate the mutations described here contained the ␤Y331W mutation, to allow fluorescence-based nucleotide binding measurements. As described earlier (10), the ␤Y331W mutation gives a normal, fully functional enzyme.) The ␤R394A/␤R398A/␤Q441A mutations would eliminate all possible hydrogen bonds plus the cation/ interaction, with the exception of one hydrogen bond between ␣Q399 and the backbone carbonyl oxygen of ␤L328. As expected, the data in Table  1 did not give any indication of assembly problems. This is also supported by the fact that we were successful in preparing F 1 from this strain (see below). As can be seen from Table 1, the ␤R394A/␤R398A/␤Q441A triple mutant has ϳ10% residual activity.
F 1 was prepared from the strain containing the ␤R394A/ ␤R398A/␤Q441A triple mutation by the standard procedure (15). The elution profile of the gel chromatography column used as the last purification step indicated that the isolated F 1 subcomplex had normal size and shape. SDS-PAGE of the final product revealed a normal subunit composition (data not shown). The isolated F 1 exhibited ATPase activity of ϳ10% of that of the parental ␤Y331W enzyme (0.56 unit/mg for ␤R394A/␤R398A/␤Q441Aϩ␤Y331W F 1 , as compared with 5.9 units/mg for ␤Y331W mutant F 1 ; see Refs. 10 and 21).
Nucleotide Binding to the ␤R394A/␤R398A/␤Q441A Mutant-The fluorescence of the tryptophan in position ␤331 was used to determine the affinities for binding of MgATP and MgADP to the three catalytic sites (10,15). The results of the titrations are shown in Fig. 2 (A and B). As can be seen, the affinities of ␤R394A/␤R398A/␤Q441Aϩ␤Y331W mutant F 1 do not differ significantly from those of the ␤Y331W control enzyme. For MgATP, the following K d values were determined: for ␤R394A/␤R398A/␤Q441Aϩ␤Y331W mutant F 1 , K d1 ϭ 21 nM, K d2 ϭ 0.9 M, K d3 ϭ 39 M; for ␤Y331W mutant F 1 , K d1 ϭ 20 nM, K d2 ϭ 1.5 M, K d3 ϭ 35 M. For MgADP, the values were as follows: for ␤R394A/␤R398A/␤Q441Aϩ␤Y331W mutant F 1 , K d1 ϭ 24 nM, K d2 ϭ 2.4 M, K d3 ϭ 53 M; for ␤Y331W mutant F 1 , K d1 ϭ 28 nM, K d2 ϭ 3.6 M, K d3 ϭ 41 M. These data show that the hydrogen-bonding network, which is present in the ␤Y331W control F 1 , but reduced to one possible hydrogen bond in the ␤R394A/␤R398A/␤Q441Aϩ␤Y331W enzyme, does not contribute to the affinity of any of the three catalytic sites. Specifically, it does not increase the affinity of the medium-affinity site 2, which was recently shown to be located on ␤ DP (22).  MgADP⅐AlF x and MgADP⅐ScF x are analogs of the transition state that is formed when MgATP is hydrolyzed to MgADP and P i . Interestingly, although in F 1 only the high affinity catalytic site 1 is catalytically active (23), a transition-state-like complex forms with MgADP⅐AlF x and MgADP⅐ScF x also at the medium affinity site; formation of this complex manifests itself as a significant increase in nucleotide binding affinity (19,24). Fig. 2C shows a titration with MgADP in presence of ScCl 3 and NaF. As can be seen from the increased affinities, the MgADP⅐ScF x complex is generated at sites 1 and 2 also in the absence of the ␤ DP /␣ DP hydrogen-bonding network. The measured K d values were as follows: for ␤R394A/␤R398A/␤Q441Aϩ␤Y331W mutant F 1 , K d1 Ͻ 2 nM, K d2 ϭ 35 nM, K d3 ϭ 90 M; for ␤Y331W mutant F 1 , K d1 Ͻ 2 nM, K d2 ϭ 58 nM, K d3 ϭ 66 M.
Modeling the ␤R394A/␤R398A/␤Q441A Mutant-The ␤R394A/␤R398A/␤Q441A triple mutant was modeled using either the original structure of bovine mitochondrial F 1 with two occupied catalytic sites (7) or the structure with three occupied catalytic sites (20) as template; in both templates, the conformation of the residues forming the ␤ DP /␣ DP hydrogenbonding network is very similar. Except for the "mutated" amino acid side chains, the resulting models (not shown) were very similar to the respective template, and therefore to each other. In both models the hydrogen bond between the mainchain oxygen of ␤L328 and the side chain of ␣Q399 were preserved. The geometry between the different secondary structural elements carrying the residues that form the network was not affected by the triple alanine substitution.
Conservation of Residues Forming the ␤ DP /␣ DP Hydrogenbonding Network-A BLAST (25) search resulted in several thousand sequences each for ␣ and ␤ subunits of ATP synthase. Of the residues forming the ␤ DP /␣ DP hydrogen-bonding network, ␤R394 was completely conserved (see Table 2 for selected species). ␣E402 and ␣F406 were conserved in Ͼ99% of all species. ␣F406 was replaced by Leu in some plant mitochondrial ATP synthases. In place of ␣E402 Asp (ϳ10 cases) and Ser (ϳ5) were found. ␣Q399 and ␤R398 were conserved in ϳ90% of all species. In place of ␤R398 mostly Lys and Phe were found, in place of ␣Q399 Ala and Ser. In light of the experimental data obtained here, which indicated that the preservation of two possible hydrogen bonds between ␣Q399 and ␤ might be required for normal function, the latter result was somewhat surprising. Of the residues under investigation here, ␤Q441 was the most variable. Replacements that can still form a hydrogen bond were found (Asp, Glu, Ser, Asn, and His) but also others where this is no longer possible (Gly, Ala, and Val).

DISCUSSION
In the study presented here we analyzed the functional importance of an inter-subunit hydrogen-bonding network that is unique to one of the three ␤/␣ interfaces in ATP synthase. This hydrogen-bonding network is located between the C-terminal domains of ␤ DP and ␣ DP . Directly involved in formation of inter-subunit contacts via hydrogen bonds are the main-chain carbonyl oxygen of ␤L328, the sides chain ␤Q441, ␣Q399, ␤R398, ␣E402, and ␤R394, plus, via a cation/ interaction, the phenyl ring of ␣F406 (Fig. 1). This network is present in all published structures of bovine mitochondrial F 1 , including the structure with all three catalytic sites occupied by nucleotide (20) and the recent structure obtained in absence of azide (26). Exceptions are the structures that contain the natural inhibitor protein, IF 1 (27,28). IF 1 is wedged between the C-terminal domains of ␤ DP and ␣ DP , keeping the interface between both subunits more open (more like at the ␤ TP /␣ TP interface; see Fig. 1), thus preventing formation of the hydrogen-bonding network (27). Instead, residues ␣Q399, ␣E402, ␣F406, and ␤R394 appear to contribute to binding of the inhibitor protein (27,28). In the structure of the yeast mitochondrial F 1 , the hydrogen-bonding network is present in only one of the three different enzyme conformations; in the two other conformations, the ␤ DP /␣ DP interface resembled the more open ␤ TP /␣ TP interface (29). However, as discussed below, there is experimental evidence for the functional importance of some of these residues also in the yeast enzyme. The absence of a comparable hydrogen-bonding network in the recent structure of F 1 from a thermoalkaliphilic Bacillus species (30) is not surprising. This structure contains no nucleotide, and all ␤ subunits are in the open conformation, resulting in clefts between the C-terminal domains of ␤ and ␣ at the catalytic interface.
The mutational analysis presented here led to the conclusion that the hydrogen-bonding network is indeed required for normal function of the enzyme. Reduction of the network to a single possible hydrogen bond in the ␤R394A/␤R398A/ ␤Q441A mutant reduced the enzymatic activity to ϳ10% of normal. Besides the ␣Q399A mutant, where no ATP synthase was assembled, the single mutants with the overall lowest activities were ␣Q399C, ␤Q441A, ␣F406A, and ␣E402A, followed by the three ␤R394 mutants. These findings indicate that the bonds at either end of the network are more important than the central ones (see Fig. 1). On one end, two hydrogen bonds between ␣Q399 and ␤ seem to be required for normal function; one of them has to be the bond to ␤Q441. On the other end, the hydrogen bond between ␣E402 and ␤R394 plus the cation/ interaction between ␤R394 and ␣F406 appear necessary. Ear-

TABLE 2 Conservation of residues forming the ␤ DP /␣ DP hydrogen-bonding network
Amino acids found in selected species in the position of residues that form the ␤ DP /␣ DP hydrogen-bonding network in the bovine mitochondrial enzyme (7). Residue numbers refer to the E. coli enzyme. lier results obtained with an ␣F406C mutant 4 suggest that the cation/ interaction can be replaced by a hydrogen bond.

Species
In general, analysis of the conservation of the residues forming the hydrogen-bonding network confirmed the functional importance of the interactions between ␣E402 and ␤R394, and between ␤R394 and ␣F406. A possible exception is a group of plant mitochondrial ATPases where ␣F406 has been replaced by Leu, preventing the cation/ interaction. Unfortunately, the sequences of the ␤ subunits are not known in these cases, making an analysis of the ␤/␣ interaction in these enzymes impossible. At the other end of the network, involving positions ␤441 and ␣399, a number of cases exist that do not fit into the pattern established here for E. coli ATP synthase (see Table 2). Apparently, some enzymes can compensate for the loss of hydrogen bonds from either of these positions. In the case of the thermophilic Bacillus sp. PS3 ATP synthase, a part of the hydrogenbonding network appears to be replaced by hydrophobic interactions. Residue ␤398 is not an Arg, but a Phe, which can interact with the Leu in position ␣403. Obviously, an Ala in position ␣399 fits better into this hydrophobic environment than a Gln. We are planning to characterize the ␤ DP /␣ DP interactions in the PS3 enzyme by mutational analysis as described here.
Although the present study is the first systematic analysis of the ␤ DP /␣ DP hydrogen-bonding network, the functional importance of some of the participating residues and their capability to form intra-subunit hydrogen bonds had been noticed before. Mutations that suppress 0 -lethality in the yeast Kluyveromyces lactis were identified in the genes that code for the ␣, ␤, and ␥ subunits of ATP synthase (31). Among these mitochondrial genome integrity, or mgi, mutations (32) found were ␣F406S,L and ␤R394G,I,T,V,K (31). None of the mutations abolished the ability of the respective strain to grow on glycerol, and all tested mutations (␣F406S and ␤R394G,K) showed some ATPase activity of isolated mitochondria (31). These results are comparable to those described here for mutations of residues ␣F406 and ␤R394. Introduction of a similar set of mutations in Saccharomyces cerevisiae (␣F406S and ␤R394G,I) gave similar results (32). In contrast, another study using S. cerevisiae (33) found that mutations of residue ␤R394 (in ␤R394I,T mutants) prevented growth on a nonfermentable lactate medium and eliminated mitochondrial ATPase activity. Mutational analysis of the function of residue ␤R398 in S. cerevisiae identified the residue as non-essential (33), similar to the results obtained here.
Residue ␤R398 was identified as necessary to confer E. coli with sensitivity to the antibiotic aurovertin; mutation to His, Cys, and Trp rendered ATPase and ATP synthesis activities resistant to the antibiotic (34,35). Furthermore, bacterial species where the equivalent residue is phenylalanine seem to be aurovertin-resistant (36). The x-ray structure of bovine mitochondrial F 1 complexed with aurovertin (37) showed that two molecules of the antibiotic were bound per F 1 , one to ␤ TP , the other to ␤ E . In both ␤ subunits residue ␤R398 contributed to aurovertin binding via hydrogen bond(s). The ␤ DP /␣ DP hydrogen-bonding network was unperturbed.
After having established the importance of the ␤ DP /␣ DP hydrogen-bonding network, the question about its role in the mechanism of ATP synthase remains. In the description of the original x-ray structure of F 1 -ATPase, the fact that the ␤ DP /␣ DP interface is more closed than the ones at ␤ TP /␣ TP or ␤ E /␣ E had led to the proposal that the catalytic site on ␤ DP might be the high affinity site (7). However, using fluorescence resonance energy transfer we could recently demonstrate that the catalytic site on ␤ TP is the high affinity site, whereas the ␤ DP site is the medium affinity site (22). Furthermore, in the present study it is shown that a nearly complete removal of the ␤ DP /␣ DP hydrogen-bonding network has no influence on nucleotide-binding affinities at any of the three catalytic sites. It is interesting to note in this context that the molecular modeling data suggested that elimination of the hydrogen-bonding network in the ␤R394A/␤R398A/␤Q441A mutant does not significantly affect the conformation of the protein backbone of the C-terminal domains of ␤ DP and ␣ DP . Specifically, the ␤ DP /␣ DP interface in the ␤R394A/␤R398A/␤Q441A mutant does not seem to be more "open" than in the wild-type enzyme.
Based on the localization of mgi mutations in K. lactis and their functional consequences, Wang et al. (32) concluded that the respective wild-type residues, including the two contributing to the hydrogen-bonding network, ␣F406 and ␤R394, might be "necessary for efficient coupling of ATP synthase, possibly by acting as support to fix the axis of rotation of the central stalk." Based on the data obtained here, we would like to propose a specific role for the ␤ DP /␣ DP hydrogen-bonding network in coupling ATP synthesis/hydrolysis and rotation. It is important to note that in the ␣ 3 ␤ 3 ring ␣ DP is opposite of ␤ E . ␤ E carries the low affinity catalytic site, which binds the incoming MgATP during multisite ATP hydrolysis and releases the newly formed MgATP during ATP synthesis (3,22,23,38,39). It is generally assumed that MgATP binding and release are energy-linked, i.e. the large conformational changes in ␤ accompanying these processes are coupled to rotation of ␥ (22, 39 -42). In ATP hydrolysis direction, MgATP binding and the associated closing of the ␤ E subunit appear to drive the 80°rotation substep; a recent molecular dynamics study supports this notion (43). When ␤ E closes it exerts a force on ␥, which has to be converted into a rotary motion of ␥, instead of a lateral motion, away from the axis of rotation, which would not allow the catalytic cycle to continue (Fig. 3). As the C terminus of ␣ DP is in the direction of such a lateral motion, hydrogen-bonding it to the C terminus of ␤ DP should make it more rigid, offering greater resistance to a lateral motion of ␥ (for a more detailed discussion, see the legend to Fig. 3). In ATP synthesis, the situation is similar. The rotating ␥ subunit, powered by the flow of protons through the F 0 subcomplex, has to be kept in a position where its rotation can be translated into opening of the low affinity catalytic site and release of the newly formed MgATP.
With the exception of the ␣Q399A mutant, which did not assemble, none of the mutants with a reduced set of hydrogen bonds was completely inactive. In the context of the model presented here this means that even the most impaired mutants had occasionally a rotation step, probably interspersed with non-productive events where ␥ moved laterally away from the rotation axis.
The model proposed here is supported by the finding that the bonds at either end of the network are more important than the central ones. The bonds at the end are sufficient to define the geometry of the interactions between ␣ DP and ␤ DP . The central bonds confer additional binding energy, which is, at least under the conditions applied her, not absolutely required; this appears to be another example of "overengineered" subunit-subunit interactions in ATP synthase (44).
The one or two hydrogen bonds that can be formed between the C-terminal domains of ␣ DP and ␤ TP (see Fig. 3) might play a role similar to that of the ␤ DP /␣ DP hydrogen-bonding network. In the bovine mitochondrial enzyme, the involved residues are ␣E355 and ␤S383 (bovine numbers). In E. coli these residues are not conserved but have been replaced by ␣N358 and ␤E369, respectively, which could also form a hydrogen bond. Alternatively, a hydrogen bond seems possible between ␣N358 and ␤Q365. The function of these residues will be tested in an approach similar to the one described here.
Finally, it should be noted that the functional importance of the ␤ DP /␣ DP hydrogen-bonding network gives an explanation as to why the medium affinity catalytic site has to be occupied during multisite catalysis (45). So far, only the roles of the other two sites were clearly defined. The high affinity site on ␤ TP performs catalysis, and nucleotide binding and release occur at the low affinity site on ␤ E . Without nucleotide bound to the medium affinity site, the ␤ DP subunit will assume an open conformation, which will prevent formation of the hydrogenbonding network, therefore impairing catalysis. FIGURE 3. Model of the possible role of the ␤ DP /␣ DP hydrogen-bonding network in ATP hydrolysis. A transection of F 1 at the level of the C-terminal domains of ␣ and ␤ is shown, as seen from the direction of the membrane; green, ␣ subunits; yellow, ␤ subunits; blue, ␥ subunit. A nucleotide bound to the catalytic site in the central domain of the respective ␤ subunit is indicated by an orange oval. The black lines between ␤ DP and ␣ DP indicate the hydrogen-bonding network. The single black line between ␣ DP and ␤ TP indicates the possible hydrogenbond(s) between the C-terminal domains of these subunits (see "Discussion"); the functional importance of the latter interaction has not yet been proven. As some of the subunit movements postulated here might be rather subtle, compared with the overall size of the F 1 subcomplex, the direction of movement of a specific subunit is indicated by a red arrow. The starting position (top panel) corresponds to the two-nucleotide x-ray structure (7) where the catalytic sites on ␤ TP and ␤ DP are filled with nucleotide, whereas the low affinity nucleotide binding site on ␤ E is empty and the ␤ E subunit is in an open conformation. MgATP binds to the open ␤ E site, and the site starts to close by moving the C-terminal domain toward the pseudosymmetry axis. The ␤ subunit transiently assumes a "halfclosed" conformation, "␤ HC " (middle panel), similar to the conformation observed for the low affinity site in the three-nucleotide structure (20), until it finally reaches the fully closed "␤ C " conformation (This ␤ C conformation could be very similar to the ␤ TP conformation that the subunit will assume after completion of the 120°r otation step.). Because of steric clashes, ␥ has to be pushed "out of the way" to allow the ␤ subunit to close completely. In the model, this can occur either by a lateral motion in the same general direction as the "pushing" C-terminal domain of ␤ HC (bottom left-hand panel), or it can be translated into a rotary motion (bottom right-hand panel). The latter movement is the desired one, as only it will allow the overall reaction to proceed. The lateral motion will exert some pressure on the C-terminal domains of ␣ DP , ␤ DP , and/or ␤ TP . Only if these domains can move away from the pseudosymmetry axis (bottom left), the lateral motion can occur (it does not necessarily mean that the enzyme would disintegrate under these circumstances, because it is still held together by ␣/␤ interactions in the N-terminal domain and in the nucleotide-binding domain.). According to the model, the ␤ DP /␣ DP hydrogen-bonding network and, possibly, the hydrogen bonds between ␣ DP and ␤ TP prevent the movement of the C-terminal domains away from the axis, thus forcing ␥ into a rotary instead of a lateral motion (bottom right).