Structure of F1-ATPase from the obligate anaerobe Fusobacterium nucleatum

The crystal structure of the F1-catalytic domain of the adenosine triphosphate (ATP) synthase has been determined from the pathogenic anaerobic bacterium Fusobacterium nucleatum. The enzyme can hydrolyse ATP but is partially inhibited. The structure is similar to those of the F1-ATPases from Caldalkalibacillus thermarum, which is more strongly inhibited in ATP hydrolysis, and in Mycobacterium smegmatis, which has a very low ATP hydrolytic activity. The βE-subunits in all three enzymes are in the conventional ‘open’ state, and in the case of C. thermarum and M. smegmatis, they are occupied by an ADP and phosphate (or sulfate), but in F. nucleatum, the occupancy by ADP appears to be partial. It is likely that the hydrolytic activity of the F. nucleatum enzyme is regulated by the concentration of ADP, as in mitochondria.


Introduction
The adenosine triphosphate (ATP) synthases, also known as F-ATPases or F 1 F o -ATPases, are multi-subunit enzymes found in energy-transducing membranes in mitochondria, chloroplasts and eubacteria [1,2]. They catalyse the synthesis of ATP from ADP and inorganic phosphate by using energy from a transmembrane electrochemical gradient of protons, known as the proton motive force (or pmf ). Alternatively, some eubacteria generate a sodium ion motive force (or smf ) to power the generation of ATP [3].
The subunits of ATP synthases are organized into membrane intrinsic and membrane extrinsic sectors [1,2]. The membrane extrinsic sector, known as F 1 -ATPase, is the catalytic part where ATP is formed from ADP and inorganic phosphate. It can be detached experimentally from the membrane domain in an intact state and retains the ability to hydrolyse, but not to synthesize, ATP. The F 1 -catalytic domains of bacterial ATP synthases are assemblies of five polypeptides. Three a-subunits and three b-subunits are arranged in alternation around a central stalk made from single copies of the g-and 1-subunits, and in the intact ATP synthase, the central stalk is associated with a ring of c-subunits in the membrane domain of the complex. Together, the g1-subcomplex and the c-ring constitute the enzyme's rotor. The turning of the rotor modulates the binding properties of the three catalytic sites which lie at three of the interfaces between a-and b-subunits, taking each of them through a cycle of substrate binding, and ATP formation and release. The single d-subunit sits on top of the a 3 b 3 -hexamer and, together with two identical b-subunits (or related but nonidentical b-and b 0 -subunits in some bacterial species and chloroplasts), forms part of the stator linking the external surface of the a 3 b 3domain to a single a-subunit in the membrane domain. Protons or sodium ions re-enter the bacterial cytoplasm via half channels at the interface between the c-ring and the a-subunit and deliver energy to impel the turning of the rotor.
Many ATP synthases can not only synthesize ATP, but under conditions of low pmf or smf and high intracellular ATP, they can operate in reverse, hydrolysing ATP to generate the pmf or smf, which is required for other cellular functions, such as transmembrane transport of small molecules, or, in motile bacteria, to drive the motor of the flagellum. Therefore, this hydrolytic mechanism has to be regulated in order to prevent wasteful hydrolysis of ATP. Since eubacteria live in a wide range of environments, their pmf or smf is influenced by external factors, such as pH, nutrients and oxygen tension, and can vary over a wide range [4]. Therefore, it is likely that diverse mechanisms of the regulation of bacterial ATP synthases will operate under these multifarious conditions of growth. In recent years, the need to understand these mechanisms of regulation has increased dramatically because of the increase of resistance to antibiotics of pathogenic microorganisms and the authentication of the ATP synthase of Mycobacterium tuberculosis, especially multidrug-resistant, extensively drugresistant and totally resistant strains, as the target for treating tuberculosis with the drug bedaquiline [5,6]. By implication, other bacterial ATP synthases could also be developed as drug targets for treating infectious diseases, but a rational approach to the design of new drugs requires the detailed structures of the ATP synthases from the pathogens, and an understanding of how they are regulated, and how their structures, mechanisms and modes of regulation differ from those of the human enzyme. As the structure [7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26] and regulation of the closely related bovine enzyme by the inhibitor protein IF 1 have been well studied [9,16,[27][28][29][30][31], the bovine enzyme provides an excellent surrogate for the human complex.
As part of this endeavour to develop bacterial ATP synthases as drug targets for treating infectious diseases, we have studied the structure and regulation of the F 1 -catalytic domain of the ATP synthase from the opportunistic periodontal pathogen Fusobacterium nucleatum, which is associated with a wide range of diseases including oral infections, adverse pregnancy outcomes, gastrointestinal disorders and atherosclerosis [32]. Also, it has been linked recently to the development and progression of colorectal cancer via inhibition of anti-tumour immune signalling pathways and the subsequent promotion of chemoresistance [33,34]. Fusobacterium nucleatum is an obligately anaerobic bacterium that grows by fermentative metabolism with glutamate as a substrate. A membranebound glutaconyl-CoA decarboxylase catalyses the decarboxylation of glutaconyl-CoA to crotonyl-CoA [35] and couples the free energy of the reaction to the transport of Na þ ions across the membrane, generating an smf [36]. The ATP synthase uses this smf to drive the synthesis of ATP required for catabolic and anabolic reactions [37].

Results and discussion
2.1. Biochemical properties of the F 1 -ATPase from Fusobacterium nucleatum The genes atpAGDC encoding, respectively, the a-, g-, b-and 1-subunits of the F 1 -catalytic domain of the ATP synthase from F. nucleatum (but lacking the d-subunit) were cloned into an expression vector with a His 10 -tag at the N-terminus of the 1-subunit and an intervening site for proteolytic cleavage. The purified recombinant enzyme contained the a-, b-, g-and 1-subunits (figure 1). The specific ATP hydrolytic activity at 378C of various preparations of the enzyme ranged between 3.5 and 9.4 U mg 21 of protein (figure 1; electronic supplementary material, figures S1-S3), similar to specific activities of 4.6-5.8 U mg 21 of the intact purified ATP synthase [37]. The apparent K m value for ATP was 0.12 mM (electronic supplementary material, figure S1). The activity of the enzyme rose with increasing pH from 6.5 to 8.5 where it reached a maximum and declined at higher values (electronic supplementary material, figure S2A). At 48C, the enzyme was stable over two weeks (electronic supplementary material, figure S2B). At temperatures of 458C and above, the activity rose substantially, attaining a maximum value of 43.5 U mg 21 at 658C (electronic supplementary material, figure S2C). From 65 to 758C, the activity declined, consistent with the melting temperature of the enzyme at 728C (electronic supplementary material, figure S2D). It is known that the c-rings of sodium-dependent ATP synthases are thermostable [37][38][39][40], but the exact molecular basis for the thermostability of c-rings and the F 1 -domain of the F. nucleatum enzyme remains unknown, although the general basis of The effect of ADP on ATP hydrolysis at an Mg 2þ : ADP ratio of 2 : 1 was assayed by the release of inorganic phosphate. A specific activity of 3.5 U mg 21 was set to 100%. (c) The effect of LDAO on ATP hydrolysis was measured with an ATP-regenerating assay. A value of 1 corresponds to a specific activity of 3.8 U mg 21 . Error bars represent the standard deviation of the mean from a biological triplicate. Where no error bars are shown, they are smaller than the diameter of the data points.
royalsocietypublishing.org/journal/rsob Open Biol. 9: 190066 thermostability of proteins is well understood [41,42]. As in F 1 -ATPases from other species, the ATP hydrolase activity of the enzyme was inhibited by Mg 2þ -ADP [43,44]. Preincubation of the enzyme with 2.5 mM Mg 2þ -ADP inhibited 45% of the ATP hydrolytic activity (figure 1b), and similar treatment with 25 mM Mg 2þ -ADP inhibited the enzyme completely. As in other species, the basal hydrolytic activity was stimulated by the addition of lauryldimethylamine oxide [LDAO; 0.05% (w/ v)], by a factor of three in this instance (figure 1c). As expected, ATP hydrolysis depended on the presence of Mg 2þ or Ca 2þ ions, but Mg 2þ at concentrations in excess of 2.5 mM was partially inhibitory (electronic supplementary material, figure S3). Both Ca 2þ and Mg 2þ ions stimulate the hydrolytic activity of the F 1 -ATPases from Escherichia coli [45,46] and from chloroplasts of Spinacia oleracea (spinach) [47]. However, the hydrolytic activities of the F-ATPases from Clostridium paradoxum [48], from the chemolithotrophic g-proteobacterium, Acidithiobacillus ferrooxidans [49] and from the bacterial thermophile, Geobacillus stearothermophilus [43], and bovine F 1 -ATPases, are lower in the presence of Ca 2þ than in the presence of Mg 2þ [50]. By contrast, in the F 1 -ATPase from F. nucleatum, the apparent K m value of Ca 2þ was lower, and the apparent V max was higher than with Mg 2þ . However, the intact ATP synthase complex from F. nucleatum is not active in the absence of Mg 2þ or in the presence of Ca 2þ [37].

Structure of F 1 -ATPase from Fusobacterium nucleatum
Crystals of the F 1 -ATPase complex from F. nucleatum have the unit cell parameters a ¼ 111.9 Å , b ¼ 200.2 Å , c ¼ 201.7 Å , b ¼ 102.28 and belong to the space group P2 1 with two F 1 -complexes in the asymmetric unit (referred to as molecules 1 and 2, respectively; figure 2a; electronic supplementary material, figure S4). The structure was solved to 3.6 Å resolution by molecular replacement with the a 3 b 3 -subcomplex, with no ligands or water molecules, taken from the structure of the F 1 -ATPase from Caldalkalibacillus thermarum containing the mutations Asp99Ala and Arg92Ala in the 1-subunit [52]. The statistics for data processing and refinement are summarized in electronic supplementary material, table S1. The quality of the electron density map is indicated in electronic supplementary material, figure S5, where representative segments and their interpretation are shown. Each of the final models of the two complexes contains three a-subunits (a E , a TP and a DP ), three b-subunits (b E , b TP and b DP ), and single copies of the g-and 1-subunits, and a total of 6420 amino acid residues were resolved.  figure S4). It results from the different position adopted by the foot of the central stalk relative to the a 3 b 3 -subcomplex in the two molecules, which in turn comes from a lattice contact involving residues 108-113 of the g-subunit of molecule 2 that is not present in molecule 1. In molecule 2, the foot of the g-subunit is rotated by about 128 in a clockwise direction, as viewed from above the 'crown' towards the membrane domain of the intact ATP synthase.

Catalytic a 3 b 3 -domain
As in other structures of F 1 -ATPases, each a-subunit and each b-subunit have three domains: an N-terminal domain with six b-strands, a central nucleotide-binding domain and a C-terminal domain consisting of a bundle of a-helices (seven in the a-subunits and four in the b-subunits). The six N-terminal domains associate together to form the 'crown' of the complex. The a 3 b 3 -domain and the entire F 1 -ATPase from F. nucleatum were compared with the equivalent structures from Mycobacterium smegmatis (6foc) [51] and C. thermarum (5ik2) [52] (figure 2b,c; electronic supplementary material, table S3). Similar comparisons were made with the a 3 b 3 -domains and F 1 -ATPases from chloroplasts in spinach royalsocietypublishing.org/journal/rsob Open Biol. 9: 190066 (6fkf ) [53], Paracoccus denitrificans (5dn6) [54], four bovine structures (4yxw, 2jdi, 1e79, 4asu) [10,12,15,23] and the a 3 b 3 -domains from G. stearothermophilus (4xd7) [55] and E. coli (3oaa) [56] and with entire F 1 -domains from the same species (electronic supplementary material, table S3). The most similar a 3 b 3 -domains were those from M. smegmatis (6foc) [51] and C. thermarum (5ik2) [52], and the structures of their F 1 -domains were the most closely related also. The least similar a 3 b 3 -domains were those from G. stearothermophilus (4xd7) [55] and E. coli (3oaa) [56]. Likewise, the least related F 1 -domains were also from G. stearothermophilus (4xd7) [55] and E. coli (3oaa) [56]. The high r.m.s.d. values for the F 1 -domains from G. stearothermophilus (4xd7) [55] and E. coli (3oaa) [56] arise because their 1-subunits are in the 'up' conformation, where the two C-terminal a-helices lie alongside the a-helical coiled-coil in the g-subunit, resulting in the a DP -b DP interface being displaced outwards (see below).
The nucleotide-binding sites in the three a-subunits have additional electron density that is compatible with each of them being occupied by an ATP molecule and an accompanying magnesium ion (figure 3a-c). Similarly, additional density in the nucleotide-binding sites of the b TP -and b DP -subunits provides strong evidence for the presence in each site of an ADP molecule with a magnesium ion (figure 3d,e). There is also some density in the nucleotide-binding site of the b Esubunit (figure 3f ), which is increased slightly in molecule 2 relative to molecule 1 (electronic supplementary material, figure S6), although the amino acid side chains that form the nucleotide-binding sites in the two molecules are essentially identical positions. Possible interpretations of this density are either that it is an ADP molecule at very low occupancy, or a citrate molecule, or a mixture of both (electronic supplementary material, figure S6). Citrate was present in the crystallization buffer, and it fits the density in molecule 1 better than in molecule 2 (electronic supplementary material, figure S6E,F). It has been found to be bound to the P-loop of RecA from M. smegmatis [57]. Although the crystallization buffer contained 500 mM ADP, similar to the conditions used with the F 1 -ATPase from C. thermarum (5ik2) [52] and M. smegmatis (6foc) [51], no ADP was added to the buffer for harvesting the crystals of the F. nucleatum F 1 -ATPase, and its absence probably accounts for the low occupancy in this site compared to the C. thermarum and M. smegmatis enzymes. The interpretation of the current data is not certain, and therefore neither a nucleotide nor citrate has been included in this site in the model. There is no evidence for the binding of either a magnesium ion or phosphate in the b E -subunit of F 1 -ATPase in F. nucleatum.

g-subunit
In the structure of the F 1 -ATPase from F. nucleatum, the g-subunit was resolved completely. As in other F 1 -ATPases, it has five a-helices, aH1-aH5. Helices aH1 and aH5 make an antiparallel, a-helical coiled-coil that occupies the central axis of the a 3 b 3 -domain, and aH2-aH4 are part of a Rossmann Figure 3. Occupancy of nucleotide-binding sites in the aand b-subunits of the F 1 -ATPase from F. nucleatum. An F o -F c difference density map for the complex was calculated with the nucleotides, Mg 2þ and water molecules at zero occupancy. The green mesh represents the difference density in the six nucleotide-binding sites contoured to 3.0 s. In (a-c), the a DP -, a TP -and a E -subunits; in (d -f ), the b DP -, b TP -and b E -subunits from molecule 1. In (a -c), the sites are occupied by an ATP molecule and an accompanying Mg 2þ (black sphere) with three water ligands (black crosses); the fourth, fifth and sixth ligands are provided by O2B and O2G of the ATP and the hydroxyl of aThr-176. In (d,e), the sites are occupied by an ADP molecule and an accompanying Mg 2þ (black sphere) with four water ligands (black crosses); the fifth and sixth ligands are provided by O2B of the ADP and the hydroxyl of bThr-156. In ( f ), the difference density in the vicinity of the P-loop cannot be interpreted with confidence, but it probably can be accounted for by an ADP molecule (without Mg 2þ ) at low occupancy or citrate.
royalsocietypublishing.org/journal/rsob Open Biol. 9: 190066 fold with five b-strands with the three a-helices between strands 1 and 2, 2 and 3, and 3 and 4. The lower part of the coiled-coil interacts with the N-terminal domain of the 1-subunit. Superimposition of the F. nucleatum g-subunit on orthologues showed that it is most similar to bacterial g-subunits from C. thermarum (5ik2) [52], E. coli (3oaa) [56], P. denitrificans (5dn6) [54] and also to the fragmentary structure of the g-subunit from M. smegmatis (6foc) [51], and to a lesser extent to the g-subunit in spinach chloroplasts (6fkf) [53] where aH1 is straighter, and the subunit has the additional b-hairpin involved its redox-linked regulatory mechanism (see below). The overall fold of these bacterial g-subunits ( figure 4) is also similar to that of the g-subunits from the enzymes from bovine (1e79) [15] and yeast (2hld) [58] mitochondria, although in the bacterial subunits aH1 extends further in a C-terminal direction and is less curved. The rotation of the g-subunit drives the synthesis of ATP in the F 1 -domains of F-ATP synthases with energy provided by the pmf (or smf), and each 3608 rotation in three 1208 steps generates three ATP molecules, one from each of the three catalytic sites of the enzyme [1,2]. During the hydrolysis of ATP, the energy provided by the ATP molecule drives rotation in the opposite sense. The hydrolytic 3608 cycle also has three 1208 steps [59,60], and the intervening pauses are known as the 'catalytic dwells', where the enzyme is poised to carry out, or is carrying out, ATP hydrolysis. At lower concentrations of ATP, a second pause, known as the 'ATP-binding dwell' [61,62], where the enzyme awaits the binding of the substrate, occurs 408 after the catalytic dwell, and in the mitochondrial enzyme, but not in bacterial enzymes, a third pause, the 'phosphate release dwell', has been observed by stopping rotation with the phosphate analogue, thiophosphate, 258 before the catalytic dwell [63].
In the wide range of high-resolution structures of F 1domains that have been resolved, the majority being structures of the bovine enzyme inhibited in a variety of ways, the 'foot' of the g-subunit consisting of the Rossmann-fold domain and the associated antiparallel a-helical coiled-coil of aH1 and aH5, has rotated to a range of positions. By contrast, the 'upper' part of the g-subunit, consisting of the N-terminal region of aH1, occupies the same position because of the intervention of a 'catch loop' provided by the adjacent b E -subunit [64]. This 'catch loop' holds the g-subunit and allows torsional energy to be stored somewhere below the catch. Once a critical point is reached, the stored energy is released in a quantum to generate the rotational step or sub-step. The role of the 'catch loop' is illustrated in electronic supplementary material, figure S7, where a selection of mitochondrial and bacterial structures, including the current one from F. nucleatum, have been superimposed. The rotational positions of the 'foot' domains in the various structures are summarized in electronic Figure 4. Comparison of the structure of the g-subunit of the F-ATPase from F. nucleatum with those of orthologues. (a) F. nucleatum (6q45; molecule 1) with the five a-helices numbered 1 -5 from N-to C-terminus; (b) C. thermarum (5ik2) [52]; (c) E. coli (3oaa) [56]; (d ) M. smegmatis (6foc) [51]; only the a-helices were resolved; (e) P. denitrificans (5dn6) [54]; ( f ) spinach chloroplasts (6fkf ) [53]; and (g) bovine mitochondria (1e79) [15].  [63]. In these rotational experiments, the phosphate release dwell is defined by the position adopted by the central stalk when rotation is inhibited by the phosphate analogue, thiophosphate, and in similar experiments, the F 1 -ATPase inhibitor protein IF 1 stopped rotation at the catalytic dwell. Therefore, the structure of bovine F 1 -ATPase inhibited with thiophosphate (4yxw) [9] provides a structural representation of the phosphate release dwell, and structures of bovine F 1 -ATPase inhibited by the monomeric form of IF 1 consisting of residues 1-60 (4tt3, 4tsf, 2v7q) [9,16] describe the catalytic dwell. Moreover, in 'ground-state' structures, for example (1bmf, 2jdi) [7,12], the enzyme has been arrested at approximately the same rotary position as in the thiophosphate-inhibited state [10], and therefore, it can also be interpreted as representing the phosphate release dwell. Likewise, in the structure of bovine F 1 -ATPase crystallized in the presence of phosphonate (4asu), rotation has been arrested at the same rotary position at approximately 308 as in the IF 1 -inhibited enzyme, and so it can be ascribed as representing the catalytic dwell [23]. Neither the structural data nor the 'single-molecule' rotary experiments [63] support alternative proposals based on simulations [65] that the structure of bovine F 1 -ATPase crystallized in the presence of phosphonate (4asu) [23] represents the ATP-binding dwell and that the 'ground-state' structures (e.g. 1bmf, 2jdi) [7,12] represent the catalytic dwell [65]. It is possible, but not certain, that the ATP-binding dwell is represented by the structure of bovine F 1 -ATPase inhibited by ADP and aluminium fluoride (1h8e) [19], where the g-subunit has rotated through 1058 (electronic supplementary material, figure S8 and table S4).
Based on the rotations of their g-subunits, the structures of those bacterial F 1 -ATPases where their hydrolytic activities seem to be regulated by the failure to release one or more of the products of hydrolysis, namely M. smegmatis (10.58 rotation) [51] and C. thermarum (11.78 and 13.28 for molecules 1 and 2, respectively) [52], lie at the position of the phosphate release dwell in the mammalian enzyme. Fusobacterium nucleatum (19.78 and 19.18 for molecules 1 and 2, respectively) falls between the position of the phosphate release and catalytic dwells in the mammalian enzyme (electronic supplementary material, table S4). The hydrolytic activity of the P. denitrificans enzyme is inhibited by the z-subunit, an orthologue in its inhibitory region of the inhibitory region of bovine IF 1 , and therefore it is likely that the P. denitrificans structure with a g-subunit rotation of 278 represents the catalytic dwell of the enzyme. Single-molecule experiments conducted with F 1 -ATPases from G. stearothermophilus and E. coli show that phosphate is released at the end of the catalytic dwell [66] and that ADP is released 258 before the catalytic dwell [67]. These positions approximate to the rotary positions in the human enzyme [63], but in the reverse order.
A comparison of the structures of the nucleotide-binding sites of the b E -subunits of F 1 -ATPases from F. nucleatum with those in the F 1 -ATPases from P. denitrificans (5dn6) [54], M. smegmatis (6foc) [51] and C. thermarum (5ik2) [52] (figure 5) illustrates that in this region, the F. nucleatum subunit is most similar to the P. denitrificans subunit, and that the C. thermarum and M. smegmatis proteins provide a second similar pair, that has a somewhat different conformation to the F.  royalsocietypublishing.org/journal/rsob Open Biol. 9: 190066 the P-loop, and in both F. nucleatum and P. denitrificans, it is displaced away from the g-subunit relative to C. thermarum and M. smegmatis. Region 2 in F. nucleatum and P. denitrificans is displaced towards the g-subunit relative to the C. thermarum and M. smegmatis subunits, and contains Tyr-332 (F. nucleatum numbering). However, this tyrosine residue (replaced by Phe-343 in M. smegmatis) contributes to one side of the adenine-binding pocket and is in approximately the same position in the four structures. By contrast, in region 3 of the F. nucleatum b E -subunit, Phe-411 (and the equivalent Phe-420 in P. denitrificans), which contributes to the opposite side of the adenine-binding pocket, is displaced away from the pocket by about 4 Å outwards relative to the equivalent residue, Phe-413, in C. thermarum, and therefore, the nucleotide would be expected to bind less strongly in F. nucleatum than in C. thermarum, as the structure of the F. nucleatum F 1 -ATPase suggests. The P. denitrificans enzyme was crystallized in the presence of ATP only, and ATP has never been observed bound to a b E -subunit in any structure of F 1 -ATPase. In region 4, the b E -subunit of F. nucleatum is more similar to the C. thermarum protein than to the P. denitrificans and M. smegmatis proteins. This region forms a loop leading into the a-helix in the top-right of figure 5b. In this loop is found residue Arg-182 (F. nucleatum numbering). In C. thermarum and M. smegmatis, this residue helps to coordinate the bound phosphate, and in F. nucleatum, the side chain is in a similar position, and yet no phosphate is evidently bound in this site. Currently, there is no clear explanation for why phosphate is not bound also in the F. nucleatum F 1 -ATPase.

1-subunit
The N-terminal domain of the 1-subunit is folded into a 10-stranded b-sandwich, and the C-terminal domain consists of a hairpin of two a-helices, lying alongside the b-sandwich ( figure 6). This conformation of two a-helices is known as the 'down' position. Superimposition of the structures of the 1-subunit from F. nucleatum via their N-terminal domains on those of orthologues demonstrates that the a-helices are in a similar position to those in E. coli (1aqt, 1bsn) [68,71,72], G. stearothermophilus (2e5y) [69] and C. thermarum (5ik2 and 5hkk) [52] (see electronic supplementary material, table S3). Similar to the structures of 1-subunits determined in the context of the intact F 1 -ATPases from E. coli (3oaa) [56] and M. smegmatis (6foc) [51], and in the structures of the intact ATP synthase from E. coli (5t4o) [73], no ATP molecule was bound to the 1-subunit in the F 1 -ATPase from F. nucleatum, and none was bound to the isolated 1-subunit or to the 1-subunit in the g1-subcomplex from Thermosynechococcus elongatus (5zwl) [70]. However, an ATP molecule with an accompanying Mg 2þ ion has been found bound to the 1-subunit in the F 1 -ATPase from C. thermarum (5hkk) [52] and to the isolated 1-subunit in G. stearothermophilus (2e5y) [69]. In the 1-subunits in C. thermarum (5hkk) [52], E. coli (3oaa, 1aqt) [56,68] and G. stearothermophilus (4xd7, 2e5y) [55,69], four conserved amino acids are involved in binding ATP [74]. They are Ile-88, Asp-89, Arg-92 and Ala-93 (C. thermarum numbering; see electronic supplementary material, figure S9). In F. nucleatum, the first two residues are conserved, but the arginine and alanine residues are replaced by serine and glutamic acid, respectively, and hence, the F. nucleatum 1-subunit lacks essential features for binding an ATP molecule at this site (electronic supplementary material, figure S9). As described below, the 1-subunit has been studied extensively in the context of regulating the hydrolytic activity of bacterial ATP synthases.

Regulation of bacterial ATP synthases
Eubacteria have evolved a variety of mechanisms for regulating the hydrolytic activity of their ATP synthases. In a-proteobacteria, exemplified by P. denitrificans, ATP hydrolysis appears to be inhibited by a protein called the z-subunit [54,75], royalsocietypublishing.org/journal/rsob Open Biol. 9: 190066 where the N-terminal inhibitory region binds to a catalytic interface under hydrolytic conditions in a closely related manner to the inhibitory action of the orthologous mitochondrial regulatory protein IF 1 on the mitochondrial ATP synthase [9,16,76]. Cyanobacterial ATP synthases are regulated by a mechanism that appears to be similar, but not identical, to the way that ATP synthases in the chloroplasts of green plants and algae are regulated. In the absence of light, when the pmf is low, ADP-Mg 2þ remains bound to one of the three catalytic sites of the chloroplast enzyme forming an inactive ADP-inhibited state of the enzyme [77,78]. This inhibited state is reinforced by the formation of an intramolecular disulfide bond in the g-subunit of the enzyme, which is thought to stabilize a b-hairpin structure formed by a unique additional sequence (residues 198-233) in the g-subunit. This b-hairpin wedges between the b-subunit and the central stalk, and may suppress futile ATP hydrolysis by preventing the rotation of the g-subunit [53]. When light is restored, the pmf increases and reduction of the disulfide bond by thioredoxin unlocks the ATP synthetic activity of the enzyme. In cyanobacterial ATP synthases, the g-subunits also contain a related insertion [79] that appears to inhibit ATP hydrolysis [70], but it lacks the segment containing the two cysteine residues, and so it cannot be regulated by a similar redox mechanism [80].
The role of the 1-subunit in the regulation of bacterial and chloroplast ATP synthases is an area of active study. The known structures of all bacterial [52,55,68,69,[71][72][73]81] and chloroplast [53,70] 1-subunits, and the orthologous d-subunit [15,58] in mitochondria, consist of an N-terminal domain folded into a 10-stranded b-sandwich, and a C-terminal domain folded into an a-helical hairpin. In the intact enzyme, the b-sandwich domain is involved in binding the 1-subunit to both the g-subunit in the central stalk of the F 1 -domain and the c-ring in the membrane domain of the enzyme. In the various structures of ATP synthases and F 1 -ATPases, the a-helical C-terminal region has been observed in one of two different conformations. In most structures, the two a-helices are associated closely with the b-sandwich, in the 'down' conformation, with an ATP molecule bound between the two domains, in the case of C. thermarum (5hkk) [52] and G. stearothermophilus (2e5y) [69]. In the intact ATP synthases from E. coli (5t4o) [73] and G. stearothermophilus (6n2y) [82] and in the F 1 -domain from E. coli (3ooa) [56] and G. stearothermophilus (2e5y) [55], the two a-helices assume a different 'up' conformation, where they penetrate into the a 3 b 3 -catalytic domain along the axis of the coiled-coil of the N-and C-terminal a-helices of the g-subunit [55,56,73,82]. In this conformation, the N-terminal domain has no bound ATP molecule. Therefore, it has been proposed that in the enzyme from G. stearothermophilus [55] but not in the E. coli enzyme, that a 'down'-'up' switch might provide a physiological mechanism that operates when the pmf and the concentration of ATP are low [69,74,83,84]. Under these conditions, the ATP molecule would leave the 1-subunit, allowing the two a-helices to dissociate from the b-domain and form the inhibitory 'up' conformation. In the thermophilic cyanobacterium, T. elongatus, where no nucleotide was observed bound to the isolated subunit (2ro6) [85], and in another structure of the g1 subcomplex where no nucleotide was present during the crystallization of the subcomplex (5zwl) [70], the 1-subunit was down in both instances. In this organism, it has been proposed that the ATPase is regulated by the g-subunit, in a similar fashion to the regulation of ATP hydrolysis in the chloroplast enzyme, but without the regulation via the oxidation and reduction of a disulfide linkage that occurs in the chloroplast enzyme [70]. However, the ATP synthases from C. thermarum [52] and M. smegmatis [51] appear not to conform to this mechanism of regulation. They can both synthesize ATP under appropriate conditions, but they hydrolyse ATP very poorly. The structures of their F 1 -catalytic domains are very similar to each other and also to the F 1 -domain of the ATP synthase from F. nucleatum, but despite being inhibited in ATP hydrolysis, the 1-subunit of the C. thermarum enzyme is in the 'down' position with an ATP molecule and a magnesium ion bound to it. Moreover, the subunit remained in the 'down' position, and the enzyme remained inhibited when the capacity of the 1-subunit to bind an ATP molecule was removed by mutation [52]. In the mycobacterial enzyme, the a-helical hairpin of its 1-subunit is truncated and incapable of binding an ATP molecule, and it also is in the 'down' position [51]. However, in the structures of the F 1 -catalytic domains from C. thermarum and M. smegmatis, a phosphate (possibly a sulfate in M. smegmatis) is bound to the most open of the three catalytic sites, suggesting that the hydrolytic activity of this enzyme may be inhibited by the failure to release one of the products of hydrolysis. In the same catalytic site, ADP is bound also in C. thermarum and is possibly present at low occupancy in M. smegmatis. In F. nucleatum, the 1-subunit is 'down' with no bound ATP, but, in contrast to the F 1 -ATPases from C. thermarum and M. smegmatis, it is intrinsically active in ATP hydrolysis, and that activity can be stimulated by LDAO. This behaviour is reminiscent of the behaviour of F 1 -ATPases from mitochondria, which are similarly intrinsically active and their activity is also stimulated by LDAO. One explanation of the stimulatory effect of LDAO is that it releases inhibitory Mg 2þ -ADP from the catalytic sites [43] and it has a similar effect on the a 3 b 3 g-subcomplex from G. stearothermophilus [86,87]. However, in the F 1 -ATPase from C. thermarum, this appears not to be the complete explanation as the activity of this enzyme, in addition to being stimulated by LDAO, is also partially activated by the removal of the C-terminal domain of the 1-subunit and could then be activated to its fullest extent by LDAO [88]. In the E. coli F 1 -ATPase, where the 1-subunit is permanently 'up' [56,73,89], LDAO has an additional effect as it influences interactions between the catalytic b-subunit and the 1-subunit. Thus, currently, the most plausible interpretation of the structure of the enzyme from F. nucleatum is that it represents the state of the enzyme that is partially inhibited by Mg 2þ -ADP, similar to the bovine F 1 -ATPase, and this partial inhibition can be relieved by LDAO. The exact molecular role of LDAO in activating these various F 1 -ATPases remains obscure. One possibility is that it loosens the structure of the nucleotide-binding domain so that the nucleotide is bound less tightly, and a similar explanation can be advanced to explain the activation of the F. nucleatum F 1 -ATPase by increased temperatures up to 658C. Structures of LDAO-activated F 1 -ATPases might help to resolve this issue.

Construction of expression plasmids
The genes atpAGDC from F. nucleatum encoding the a-, g-, band 1-subunits, respectively, of ATP synthase were amplified by a polymerase chain reaction from genomic DNA with the primers FusoF1for (5 0 -TTTTCCATGGATGAATATTAGAC-CAGAAGAAG-3 0 ) and FusoF1rev (5 0 -TTTTGGATCCTTAA TTATTCTTAGCATCTATTTTTG-3 0 ). The product was cloned into the expression vector pTrc99a (Amersham Biosciences). The translational initiation codons of the a-and b-subunits were changed to ATG, generating the expression construct pJP2. To facilitate the purification of enzyme, the sequence encoding either a His 10 -tag with a following cleavage site for the protease from tobacco etch virus (TEV) ( pJP3) or a His 10 followed by a 6-residue (Ser-Gly-Gly-Gly-GlyGly) linker, an intervening TEV protease cleavage site and another 6-residue (Ser-Gly-Gly-Gly-GlyGly) linker ( pJP5) were introduced at the 5 0 -end of atpC encoding the 1-subunit. For pJP3, the primers FusoHis_F1for (5 0 -TTTTT GAATTCCATCTGCTGTTGGATATCAACC-3 0 ) and Fuso-His_F2rev (5 0 -AAGATTCTCATGGTGATGGTGATGGTGAT GGTGATGGTGCATATTCCCTCCTTATTTTGCTAAATC-3 0 ) were used for the first fragment and FusoHis_F3for (5 0 -CAT-CACCATCACCATCACCATGAGAATCTTTATTTTCAGGGC CCTAGTTTTGATGTAAGTGTTGTAACAC-3 0 ) and Fuso-F1rev for the second fragment. For pJP5, the primers FusoHis_F1for and FusoHisLinker_F2rev (5 0 -ACCTGAGCC CTGAAAATAAAGATTCTCACCGCCACCGCCACCTGAA TGGTGATGGTGATGGTGATGGTGATGGTGCATATTCCCT CCTTATTTTGCTAAATC-3 0 ) for generating the first fragment and FusoHisLinker_F3for (5 0 -CACCATCACCATCACCATT CAGGTGGCGGTGGCGGTGAGAATCTTTATTTTCAGGG CTCAGGTGGCGGTGGCGGTCCTAGTTTTGATGTAAGTG TTGTAACAC-3 0 ) and FusoF1rev for amplifying the second fragment. In both cases, the two fragments overlapped by 30 nucleotides and were joined by overlap extension with the external primers FusoHis_F1for and FusoF1rev. The resulting fragments were cloned into EcoRI and BamHI sites in pJP2 producing the expression vectors pJP3 and pJP5. The sequences of all four genes were verified by DNA sequence analysis. A protein expressed from pJP3 was used in all assays. However, it was found that the TEV protease was unable to cleave the His 10 -tag. The protein used in the crystallization trials was expressed from pJP5 where the His 10 -tag was able to be removed by the TEV protease.

Expression and purification of F 1 -ATPase from
Fusobacterium nucleatum Expression plasmids pJP3 and pJP5 were transformed into E. coli DK8 (Dunc), together with the helper plasmid pRARE (Addgene). The cells were grown at 378C to an optical density of 0.4-0.8 at 600 nm in 2Â YT medium plus ampicillin (100 mg ml -1 ), chloramphenicol (34 mg ml 21 ) and 0.2% [w/ v] glucose. Expression from the trc-promoter was induced with 1 mM isopropyl b-D-1-thiogalactopyranoside, and the culture was incubated for 3-4 h at 378C and then for 16 h at 308C. The cells were harvested and washed with buffer (50 mM Tris -HCl pH 8.0 and 2 mM MgCl 2 ) and either used immediately or stored at 2208C. The yield of wet cells was 2 g l -1 . Cells (approx. 7 -10 g) were resuspended in the same buffer plus cOmplete EDTA-free protease inhibitor tablets (Roche) and DNase I (Roche), and disrupted by two passages through a Constant Systems cell disrupter at 31 kpsi. Cell debris was removed by centrifugation (

Biochemical methods
The F 1 -ATPase from F. nucleatum was analysed by SDS-PAGE on 4-12% NuPAGE Bis-Tris Mini gels (Life Technologies). Proteins were detected with Coomassie G-250 dye. The bands from the stained gel were excised, and the identities of the proteins were verified by mass-mapping of tryptic peptides in a MALDI-TOF mass spectrometer. Protein concentrations were measured with the DC protein assay kit (Bio-Rad) with bovine serum albumin as a standard. ATP hydrolysis was measured by an ATP-regenerating assay, at 378C unless otherwise stated, where one unit of activity is equal to 1 mmol of ADP produced per minute [92] or by the colorimetric assay of inorganic phosphate where one unit of activity is equal to 1 mmol of phosphate released per minute [93]. The influence of pH on the activity of the enzyme was examined in a three-buffer mixture composed of 50 mM each of MES-MOPS-Tris-HCl [94]. The inhibitory effect of increasing concentrations of Mg 2þ -ADP at a constant ratio of Mg 2þ : ADP 2 : 1 (w : w) on ATP hydrolysis was examined. The enzyme was pre-incubated with the Mg 2þ -ADP mixture for 10 min and then ATP hydrolysis was initiated by the addition of ATP.

Crystallization of F 1 -ATPase from Fusobacterium nucleatum
The enzyme was concentrated by ultrafiltration to 2-2.5 mg ml 21

Data collection, structure determination and refinement
Diffraction data were collected from two cryo-protected crystals of F 1 -ATPase from F. nucleatum at the Australian Synchrotron MX2 beamline [97], equipped with an ADSC Quantum 315r detector, and processed with XDS [98]. Because of radiation damage, four datasets from two crystals with the same space group and unit cell were merged with AIMLESS [99] during data reduction in CCP4 [100]. Molecular replacement was carried out with PHASER [101] using the a 3 b 3 subcomplex from the structure of the F 1 -ATPase from C. thermarum containing the mutations Asp89Ala and Arg92Ala in the 1-subunit (5ik2) [52] with nucleotides and other ligands removed. Rigid body refinement and restrained refinement using non-crystallographic symmetry restraints were performed with REFMAC5 [102]. In between each refinement round with REFMAC5, parts of the structure were rebuilt manually with COOT [103]. The stereochemistry of the structure was assessed with MolProbity [104]. Electron density maps were calculated with FFT [100], and images of structures and electron density maps were prepared in PYMOL [105].

Rotation of the g-subunit
Residues 23 -33 of the F. nucleatum g-subunit (and equivalent regions in other F 1 -ATPases) interact with the C-terminal domains of the a-and b-subunits, and this segment acts as a rigid body uninfluenced by contacts in the crystal lattice between adjacent F 1 -ATPase complexes. By contrast, residues 34-226 of the g-subunit, and the associated d-and 1-subunits lie outside the a 3 b 3 -domain, where their positions may be subject to such influences. Therefore, the rotations of residues 23-33 of the g-subunit in the various aligned structures were measured relative to the position of the same segment (residues 22-32) in the ground-state structure of azide-free bovine F 1 -ATPase (2jdi) [12]. These measurements were made by aligning the structures via the crown domains at the N-termini of a-and b-subunits and then by calculating the centre of mass of residues 22-33 of the g-subunit and determining the rotation angle around the pseudo-threefold axis of the a-subunits required to match its position with that of the equivalent segment in the bovine azide-free ground-state structure, which was taken as the reference point, set as 08 [10]. To calculate the rotation of the g-subunit in molecule 2 with respect to that in molecule 1, the distances between the Ca of residues 95 and 105 (helix 2 in the g-subunit) were measured in both molecules and then to their equivalent residues in the other model. These distances were extrapolated to a common origin giving the lengths of three sides of a triangle, allowing the angle to be calculated.
Data accessibility. The data available from the PDB accession code: 6q45. Authors' contributions. Conceptualization: J.E.W and G.M.C. Method-