The structure of bovine F1‐ATPase inhibited with ADP and beryllium fluoride at 2.0 Å resolution contains two ADP.BeF3− complexes mimicking ATP, bound in the catalytic sites of the βTP and βDP subunits. Except for a 1 Å shift in the guanidinium of αArg373, the conformations of catalytic side chains are very similar in both sites. However, the ordered water molecule that carries out nucleophilic attack on the γ‐phosphate of ATP during hydrolysis is 2.6 Å from the beryllium in the βDP subunit and 3.8 Å away in the βTP subunit, strongly indicating that the βDP subunit is the catalytically active conformation. In the structure of F1‐ATPase with five bound ADP molecules (three in α‐subunits, one each in the βTP and βDP subunits), which has also been determined, the conformation of αArg373 suggests that it senses the presence (or absence) of the γ‐phosphate of ATP. Two catalytic schemes are discussed concerning the various structures of bovine F1‐ATPase.
The ATP synthase (F1Fo‐ATPase) in the inner membranes of mitochondria is a multisubunit enzyme that uses energy from the proton motive force (PMF) generated by electron transfer to catalyse the synthesis of ATP from ADP and inorganic phosphate (Boyer, 1997; Walker, 1998; Senior et al, 2002). It has a membrane domain Fo, attached by central and peripheral stalks to the catalytic domain, F1, which lies outside the membrane in the matrix of the organelle. The central stalk consisting of subunits γ, δ and ε penetrates into the core of the F1 catalytic domain, and the three α‐subunits and the three β‐subunits that together with subunits γ, δ and ε make up the F1 domain are arranged in alternation around a six‐fold axis of pseudo‐symmetry passing through the central stalk (Abrahams et al, 1994). The foot of the central stalk (Gibbons et al, 2000) is attached firmly to a ring of hydrophobic c‐subunits in the membrane domain (Stock et al, 1999). The central stalk and the c‐ring rotate as an ensemble, and the rotation is driven by energy from the PMF (Yoshida et al, 2001). This rotation modulates the nucleotide‐binding properties and catalytic activities of each catalytic β‐subunit by bringing about a series of structural transitions that takes each of them through states with high, medium and low affinities for substrates and products.
The F1 domain can be released intact from the membrane and retains the ability to hydrolyse ATP. Over the past 10 years, bovine F1‐ATPase has been studied intensively by X‐ray crystallography. The first crystal structure, called ‘the reference structure’ below, was determined with crystals grown in the presence of ADP and the nonhydrolysable ATP analogue AMP‐PNP (Abrahams et al, 1994). In this structure, AMP‐PNP was found in one β‐subunit (βTP), ADP in the second (βDP) and the third (βE) had adopted an open conformation that was incapable of binding a nucleotide. On the basis of the reduced solvent accessibility to the catalytic site, and the arrangement of the catalytic residues, the βDP subunit was designated as the catalytic (tight) binding site. As MgADP, but not phosphate, was bound to this subunit, it was suggested that the structure represents the ‘ADP‐inhibited state’. Subsequently, the structures of the enzyme inhibited by the antibiotics efrapeptin (Abrahams et al, 1996) and aurovertin B (van Raaij et al, 1996), and of the enzyme covalently inhibited with 4‐chloro‐7‐nitrobenzofurazan (Orriss et al, 1998) and dicyclohexyl‐carbodiimide (DCCD) (Gibbons et al, 2000) were determined. All of them were very similar to the reference structure. In each of them, MgADP was bound to the βDP subunit, and the βE subunit remained empty. When crystals were grown in the presence of a high concentration (5 mM) of AMP‐PNP and the usual concentration of ADP, the βE subunit still remained empty (Menz et al, 2001a). However, when they were grown from enzyme inhibited with ADP and aluminium fluoride (and aluminium fluoride was not maintained in the mother liquor), the resulting structure (Braig et al, 2000) had an AlF3 group bound close to ADP in the βDP subunit, mimicking a catalytic intermediate. The presence of aluminium fluoride resulted in small movements of the catalytic residues (αArg373, βGlu188, βArg189) in the βDP catalytic site, but the overall conformation of the enzyme was still very similar to that of the reference structure. These experiments provided overwhelming evidence that the reference structure (and the related structures) represents a conformation on the catalytic pathway of ATP hydrolysis by F1‐ATPase. Other crystals of F1‐ATPase inhibited by ADP and aluminium fluoride, where excess aluminium fluoride and ADP were maintained in the mother liquor, led to a structure where the ADP.AlF4− complex was bound to both βTP and βDP subunits, and in addition ADP and a sulphate moiety (mimicking phosphate) were found in the βE subunit (Menz et al, 2001b). This structure, referred to below as the (ADP.AlF4−)2F1 structure, displayed a novel ‘half‐closed’ conformation of the βE subunit in which the formerly open or empty site had closed partially forming both an adenine‐binding pocket that allowed ADP to bind and a distinct phosphate‐binding site that was occupied by sulphate. In this structure, the central stalk had rotated significantly in a clockwise direction as viewed from the foot of the central stalk relative to the reference structure. The coordination distances of the AlF4 group bound to the βDP subunit suggested that this structure was mimicking a transition state in the catalytic pathway. Another structure of F1‐ATPase inhibited by the regulatory protein, IF1, appears to represent a prehydrolysis state where ATP is poised for hydrolysis (Cabezón et al, 2003).
This series of structures has helped to define the catalytic pathway of ATP hydrolysis by F1‐ATPase, but structural information about an enzyme–substrate complex with ATP bound to the catalytically active βDP subunit has been lacking, and all attempts to obtain crystals with ATP or an ATP analogue bound to the βDP subunit have been unsuccessful. Beryllium fluoride is well established as a phosphate analogue (Petsko, 2000), and it has been used to study a variety of NTPases including G‐proteins (Bigay et al, 1987), F‐actin, microtubules (Combeau and Carlier, 1988; 1989) and rabbit muscle kinases (Garin and Vignais, 1993). Its use has resulted in a number of protein structures in which the beryllium fluoride is mimicking phosphate. In others, it is associated with ADP. In the structures of the myosin motor domain from Dictyostelium discoideum (Fisher et al, 1995; Bauer et al, 1997; Gulick et al, 2000), of scallop myosin S1 (Himmel et al, 2002) and of nucleotide diphosphate kinase (Xu et al, 1997) the ADP.BeF complex was interpreted as mimicking bound ATP, whereas in a UMP/CMP kinase a BeF2 group formed a strong bridge between the terminal oxygens of the bound nucleotides (ADP and UDP) (Schlichting and Reinstein, 1997). ADP and beryllium fluoride together inhibit bovine F1‐ATPase (Dupuis et al, 1989).
This paper describes the crystal structure of bovine F1‐ATPase inhibited by ADP and beryllium fluoride at 2.2 Å resolution. In this structure, an ADP.BeF complex appears to mimic ATP bound to both βDP and βTP subunits. Two possible reaction schemes for ATP hydrolysis are discussed in the context of the known crystal structures of the bovine enzyme.
Structure determination of BeF3−‐inhibited bovine F1‐ATPase
The crystals of bovine F1‐ATPase inhibited with BeF3− and ADP (BeF3−‐F1) were grown by microdialysis in the presence of beryllium chloride and sodium fluoride. The crystallization buffer did not contain AMP‐PNP. The structure was solved by molecular replacement (see Materials and methods) using data to 2.2 Å resolution. Data processing and refinement statistics are summarized in Table I.
The refined model consists of residues αE 24–510, αTP 23–401, 410–510, αDP 13–510, βDP 9–475, βE 9–474, βTP 9–474, γ 1–47, 67–90, 105–116, 127–148, 159–173 and 201–272. The electron density for the δ‐ and ε‐subunits was too weak to allow reliable modelling. The mobility of these subunits appears to be correlated to the a dimension of the unit cell, as they are well defined in both the DCCD‐inhibited structure (Gibbons et al, 2000) (a=267 Å) and the (ADP.AlF4−)2F1 structure (Menz et al, 2001b) (a=268 Å), but not in other structures with a in the range 282–286 Å. This is probably because the smaller cell edge is associated with additional lattice contacts between the γ‐ and δ‐subunits of one complex and the N‐terminal region of the αE subunit of an adjacent molecule.
Structure determination of the ADP bound form of bovine F1‐ATPase
Since the BeF3−‐F1 crystals were grown with ADP as the only nucleotide present, the structure of F1‐ATPase with ADP as the only bound nucleotide (ADP‐F1) was determined to allow comparison with the BeF3−‐inhibited structure. The ADP‐F1 crystals were grown under the same conditions as the reference crystals (Lutter et al, 1993) except that AMP‐PNP was omitted from both the inside and outside buffers, and the ADP concentration was increased from 5 μM to 2 mM in the outside buffer. The structure was solved by molecular replacement (see Materials and methods), but the resolution was limited by crystal quality to 2.85 Å. Data processing and refinement statistics are summarized in Table I.
The refined structure consists of αE 24–510, αTP 24–401, 410–510, αDP 19–510, βDP 9–475, βE 9–474, βTP 9–474, γ 1–44, 77–90 and 209–272. One residue (Asp411) in the αTP subunit is in a disallowed region of the Ramachandran plot, but this region of the protein is poorly ordered, with B‐factors above 100 Å2.
Overall structure of BeF3−‐F1
When the beryllium fluoride‐inhibited structure is compared to the reference structure (Braig et al, 2000), the root mean square deviation (r.m.s.d.) in Cα positions for the entire complex is 0.41 Å, ranging from 0.30 Å (βTP) to 0.62 Å (γ) for individual subunits. The relatively large r.m.s.d. for the γ‐subunit probably arises because it is more mobile: the average B‐factors for individual subunits range from 38.7 Å2 (αDP) to 82.9 Å2 (γ).
Nucleotide‐binding sites of the BeF3−‐inhibited structure
The difference electron density map calculated after the first round of refinement (using a model with ADP as the bound nucleotide) showed positive density in both the βDP and the βTP nucleotide‐binding sites adjacent to the β‐phosphate of the bound ADP. Beryllium fluoride (BeF3−) gave an excellent fit to the electron density in both subunits (Figure 1). There was no evidence for bound nucleotide in the third β‐subunit, βE. The presence of two BeF3− molecules in the structure is consistent with the stoichiometry estimated in biochemical studies (Dupuis et al, 1989).
The P‐loop residues of the βDP and βTP subunits were superimposed, and the positions of the catalytic residues were compared (Figure 2A). The only significant difference is in the position of the guanidinium group of αArg373, which is ca.1 Å further away from the BeF3− group in the βTP subunit. Other side‐chain positions differ by less than 0.3 Å. However, comparison of the water structure in the catalytic sites revealed a significant shift in the position of the water molecule lying between the beryllium and the carboxylate group of βGlu188. This water molecule is 2.6 Å from the beryllium in the βDP subunit, but 3.8 Å distant in the βTP subunit. The βDP catalytic site of BeF3−‐F1 is also very similar to the βDP catalytic site of the (ADP.AlF4−)2F1 structure (which is believed to represent the pentacovalent intermediate). Following superposition of the P‐loop residues, the side‐chain positions differ by less than 0.5 Å (Figure 2B).
By superimposing the catalytic site of the βTP subunit of BeF3−‐F1 onto the βTP subunit of the reference structure, it became apparent that the beryllium fluoride group overlaps the γ‐phosphate of AMP‐PNP, demonstrating that in terms of the geometry of binding ADP.BeF3− is a good mimic of AMP‐PNP (and ATP). The noncatalytic nucleotide‐binding sites in the three α‐subunits all contain ADP, with no evidence for bound sulphate or beryllium fluoride.
Structure of bovine ADP‐F1
The structure of ADP‐F1 is also very similar to the reference structure, with an r.m.s.d. in Cα positions for the entire complex of 0.54 Å. The r.m.s.d. values for individual subunits range from 0.32 Å (αDP) to 0.70 Å (γ). When compared with the BeF3−‐F1 structure, the overall r.m.s.d. is 0.55 Å, with the values for individual subunits ranging from 0.28 Å (βDP) to 0.55 Å (γ).
The nucleotide‐binding sites of the βDP and βTP subunits and all three α‐subunits have ADP bound, whereas the βE subunit has no bound nucleotide. The conformations of the catalytic residues of the βDP catalytic site are very similar (within 0.6 Å when superimposed by the P‐loop residues) in the ADP‐F1 and BeF3−‐F1 structures. However, in the βTP catalytic site, αTPArg373 adopts a very different conformation in ADP‐F1 (Figure 3), which has previously only been observed in the DCCD‐inhibited structure (Gibbons et al, 2000). The re‐arrangement of the Arg373 side chain is accompanied by a shift in its α‐carbon position of 2 Å (adjacent residues shift less than 0.5 Å), and somewhat larger shifts (up to 3 Å) in residues 421–427 in the βTP subunit, including Phe424, which forms part of the nucleotide‐binding pocket (Figure 3). In its new conformation, the Nε of Arg373 can form a hydrogen bond to the 2′ hydroxyl of the ADP bound to the βTP subunit.
Comparison with other structures of bovine mitochondrial F1‐ATPase
The overall structure of BeF3−‐F1 is very similar to that of the reference structure, and also to the majority of the previously determined structures of bovine F1‐ATPase in a variety of inhibited states (Abrahams et al, 1994; 1996; van Raaij et al, 1996; Orriss et al, 1998; Braig et al, 2000; Gibbons et al, 2000; Menz et al, 2001a). The two exceptions are the (ADP.AlF4−)2F1 structure (Menz et al, 2001b) and the complex with the natural inhibitor protein (Cabezón et al, 2003). The structural similarities include the catalytic residues in the βDP and βTP subunits. The r.m.s.d. values for all atoms in the P‐loop and the side chains of βLys162, βGlu188, βArg189 and αArg373 between the BeF3−‐inhibited and the reference structures are 0.29 and 0.45 Å for the βDP and βTP subunits, respectively. In the BeF3−‐F1 structure, the BeF3− group mimics the γ‐phosphate of ATP, and this structure provides the first picture of how ATP binds to the βDP subunit.
In view of the similarities in the crystallization conditions of the (ADP.AlF4−)2F1 and the BeF3−‐F1 structures, which differ primarily in the replacement of aluminium fluoride by beryllium fluoride, it is not apparent why the BeF3−‐F1 structure did not bind ADP in the βE subunit. One difference between the two structures is the length of the crystallographic a axis (a=283.5 Å for BeF3−‐F1, a=268 Å for (ADP.AlF4−)2F1). This difference suggests that the shorter a axis favours the nucleotide‐bound half‐closed conformation of the βE subunit. However, the βE subunit is not involved in lattice contacts in either structure, and therefore the βE conformation must be influenced indirectly via intermolecular interactions involving the γ‐ and δ‐subunits. While lattice contacts may influence the conformation and nucleotide occupancy of the βE subunit, they cannot be the sole factor because the DCCD‐inhibited F1 structure, which also has a short a axis (a=267 Å; Gibbons et al, 2000), has the ‘open’ (nucleotide free) conformation of the βE subunit. In recent experiments, both the open and half‐closed conformations of the βE subunit have been observed within the same crystal (S Bartoschek, S Contessi, MG Montgomery, AGW Leslie and JE Walker, unpublished work), suggesting that the energetic barrier between the two conformations is rather small, and could be influenced by crystal packing forces. However, there is no simple explanation that accounts for all the observations.
Which β‐subunit is catalytically active?
The βDP subunit rather than the βTP subunit was identified as being closest to the catalytically active conformation in the original structure of the bovine enzyme because the nucleotide is more buried at the βDP/αDP interface than at the βTP/αTP interface. The presence of MgADP in the βDP subunit is also consistent with the enzyme being in the MgADP‐inhibited form, in which MgADP (but no phosphate) is bound to the catalytic site with the highest affinity. The assignment of βDP as the catalytic subunit was supported by the structure of AlF3‐inhibited F1 (Braig et al, 2000), where MgADP.AlF3 was bound to the βDP subunit but not to the βTP subunit.
In the structure of (ADP.AlF4−)2F1, both the βTP and the βDP catalytic sites bind MgADP.AlF4−, which is believed to mimic a pentacovalent intermediate in the hydrolysis reaction. As discussed previously (Menz et al, 2001b), the arrangement of catalytic residues in the two sites is very similar. The principal difference is in the position of the guanidinium group of αArg373, which is ca.1 Å further from the γ‐phosphate site in the βTP subunit (the γ‐phosphate site is the location of the γ‐phosphate of AMP‐PNP bound to the βTP subunit in the reference structure). Comparison of the βTP and βDP catalytic sites in BeF3−‐F1 (Figure 2A) also shows that the two sites are almost identical, except that αArg373 has shifted, as in the (ADP.AlF4−)2F1 structure. The spatial arrangement of the catalytic residues in the βDP subunit of the (ADP.AlF4−)2F1 structure is also very similar to that in the βDP subunit of the BeF3−‐F1 structure (representing the catalytic enzyme–ATP complex), with differences in side‐chain positions of less than 0.5 Å (Figure 2B). This is consistent with the proposed catalytic states of these two sites: a significant conformational change occurs only after ATP hydrolysis, to a state represented by the ‘half‐closed’ βADP+Pi subunit of the (ADP.AlF4−)2F1 structure.
All available structural and biochemical data support a catalytic mechanism in which the carboxylate group of βGlu188 polarizes or extracts a proton from a water molecule, promoting an in‐line attack on the γ‐phosphate of ATP. Other side chains (βLys162, βArg189, αArg373) and main‐chain atoms in the P‐loop play roles in orienting the γ‐phosphate and stabilizing negative charges that develop on the γ‐phosphate during catalysis. Either ADP or ADP.AlFx was bound to the βDP subunit in all previously determined structures of the bovine enzyme, and the solvent structure in the region of the γ‐phosphate must differ when ATP is bound. Only in the present structure, with ADP.BeF3− mimicking bound ATP in both the βTP and the βDP catalytic sites, has it been possible to examine which site is more likely to represent the catalytically active conformation on the basis of the location of the attacking water. The relatively high resolution of the BeF3−‐F1 structure (2.2 Å) also allows greater confidence in the observed positions of well‐ordered water molecules. The distinction between the two sites is very clear (Figure 4). In the βDP subunit, the water molecule (B‐factor 33 Å2) is 2.6 Å from the beryllium, and it also interacts with both carboxylate oxygens of βGlu188 (2.8 and 3.5 Å away). In the βTP subunit, the corresponding water molecule (B‐factor 33 Å2) is 3.8 Å from the beryllium, and it is not favourably placed for in‐line attack on the γ‐phosphate of an ATP molecule. This observation strongly supports the view that the βDP subunit represents the catalytically active conformation. The overall similarity of the two catalytic sites (Figure 2A) suggests that the shift in the position of the attacking water is an indirect result of a small shift of the side chain of αArg373. In the βTP subunit, the guanidinium group of αArg373 is positioned about 1 Å further away from the γ‐phosphate site than in the βDP subunit. This change is presumably responsible for the observed shift in the BeF3− group, which effectively pivots around the fluorine atom bound to the magnesium ion, producing a shift of 0.7 Å in the beryllium and 0.8 Å/1.0 Å in the other two fluorine atoms (Figure 5). The combined shifts of αArg373 and the BeF3− group result in a shift of the attacking water by 0.6 Å in the opposite direction, towards the carboxylate group of βGlu188, so that the distance to the beryllium changes from 2.6 to 3.8 Å. There are two further water molecules in the vicinity of the BeF3− group, but their positions change by less than 0.3 Å between the two sites. This assignment of catalytic activity to the βDP subunit agrees with a similar suggestion based on molecular dynamics simulations (Gao et al, 2003).
Structural basis for the difference between the βDP and βTP active sites
Because the catalytic sites are located at the interfaces between the α‐ and β‐subunits, the conformation of the active site depends on both the conformation of the individual α‐ and β‐subunits (their tertiary structures) and on the relative positioning of the subunits (the quaternary structure). The βDP and βTP subunits have very similar tertiary structures, particularly in the region of the nucleotide‐binding site, as have the αDP and αTP subunits. Therefore, the observed differences between the βDP and βTP catalytic sites are primarily the result of differences in the quaternary structure. This conclusion is apparent when the βDP/αDP subunits are superimposed (as a pair) onto the βTP/αTP subunits using only atoms from the β‐subunits to define the superposition. As expected, the β‐subunits superimpose well (r.m.s.d. 0.66 Å for all Cα atoms, 0.47 Å for Cα atoms in the nucleotide‐binding domain), but the α‐subunits do not (r.m.s.d. 3.56 Å for all Cα atoms, 2.09 Å for Cα atoms in the nucleotide‐binding domain). A rigid‐body rotation of 6.7° is required to superimpose the nucleotide‐binding domains of the α‐subunits (r.m.s.d. 0.63 Å after superposition). The rotation axis runs through the upper part of the nucleotide‐binding domain and is oriented approximately at right angles to the axis of the F1 assembly and at an angle of 45° to the αTP/βTP interface (the axis lies very close to the α‐carbons of αGly135 and αIle206). This rigid‐body rotation is primarily responsible for the shift in the position of αArg373 relative to the catalytic residues of the β‐subunits. Additional small rotations are required to superimpose the N‐ or C‐terminal domains, as the relative orientation of these domains differs in the αDP and αTP subunits. The positioning of the αTP subunit in the F1 complex is determined by its interactions with the neighbouring β‐subunits (βTP and βE) and with the centrally located γ‐subunit. These interactions are quite extensive, involving numerous segments of polypeptide, suggesting that no single interaction is responsible for the observed rotation of the nucleotide‐binding domain of the αTP subunit. However, it is clear that the asymmetry of the (αβ)3 subassembly is fundamentally due to the inherent lack of three‐fold symmetry in the γ‐subunit. Each α‐ and β‐subunit interacts with the γ‐subunit in a unique manner that results in the asymmetric quaternary structure of the complex.
The arginine finger αTPArg373
The ADP‐F1 structure is very similar to the reference structure, but there is a significant difference in the conformation of the side chain of αTPArg373. This side chain is believed to play a critical role in catalysis (Futai et al, 1989; Abrahams et al, 1994; Nadanaciva et al, 1999), but in the ADP‐F1 structure it points away from the active site rather than towards it, resulting in a shift of almost 10 Å in the position of the guanidinium group. This conformation has been observed previously in the αTP subunit of the DCCD‐inhibited enzyme (Gibbons et al, 2000), and the common feature of the two structures is the absence of a γ‐phosphate (or an analogue such as BeFx or AlFx) in the βTP site. There are several basic residues in the vicinity of the γ‐phosphate site, including βLys162, βArg189 and αArg373. It is possible that in the absence of favourable interactions between the γ‐phosphate (or analogue) and the guanidinium group of αArg373, this arrangement of basic residues becomes energetically unfavourable. Of these basic side chains, αArg373 would seem to have the greatest conformational freedom, as it is not involved with as many packing interactions as the other residues, and it has the opportunity for a radical change in conformation.
Comparison of the βTP/αTP interface in the ADP‐F1 structure and the reference structure (after superposing the βTP subunits) shows that the change in the conformation of αArg373 is accompanied by a rigid‐body rotation (1.5°) of the entire αTP subunit relative to the βTP subunit. The rotation results in a slight opening of the αTP/βTP catalytic interface, and in shifts of up to 1.5 Å for residues in the C‐terminal domain of the αTP subunit. A rotation of a similar magnitude (1.3°) is observed in the βE subunit, which forms the other interface with the αTP subunit, whereas the remaining α‐ and β‐subunits show much smaller changes (<0.5° rotation). These changes in the relative positions of the α‐ and β‐subunits (quaternary structure) are found in both ADP‐F1 and DCCD‐F1 (both have ADP bound to the βTP subunit) when compared with either the reference structure or BeF3‐F1 (both contain a γ‐phosphate or analogue in the βTP subunit). The position of the segment of the γ‐subunit that interacts with the βTP and αTP subunits is very similar in all these structures, which suggests that the change in the quaternary structure is a direct result of the presence or absence of a γ‐phosphate group in the catalytic site of the βTP subunit.
On the other hand, αArg373 in the αDP subunit of ADP‐F1 (or the reference structure) does not have the same conformation as in the αTP subunit of ADP.F1, even though ADP is bound to the β‐subunit in both cases. A possible explanation is that differences in the quaternary structure between the βTP/αTP and the βDP/αDP catalytic interfaces (for example, the more extensive interface between βDP and αDP subunits, or interactions with the γ‐subunit) make it energetically unfavourable for the βDP/αDP interface to open to accommodate the alternative conformation of αArg373 in the same way as observed in the βTP/αTP catalytic interface of the F1‐ADP or F1‐DCCD structures.
If the conformational flexibility of αArg373 is indeed associated with the type of nucleotide (ADP or ATP) bound to the catalytic subunit, and if this flexibility is coupled to the changes in the quaternary structure (as suggested by the ADP‐F1 and F1‐DCCD structures), then this residue may play an active role in catalysis by providing discrimination between ADP and ATP binding to the βE site (which transforms into the βTP site) during ATP hydrolysis.
Stoichiometry of beryllium fluoride binding
In BeF3−‐F1, all three α‐subunits bind ADP rather than ADP.BeF3−. This observation probably reflects the rate of formation of an ADP.BeF3− complex from free ADP and BeF3− being likely to be very slow in solution, and not being enhanced significantly when the nucleotide is bound to the α‐subunits. In contrast, the catalytic residues in the β‐subunits that promote ATP hydrolysis/synthesis are also very effective in catalysing the formation of ADP.BeF3−. Once formed, ADP.BeF3− binds very tightly to the β‐subunits, and so the concentration of free ADP.BeF3− in solution (which could then bind to the α‐subunits) will always be very low.
The presence of ADP.BeF3− bound to both the βDP and βTP subunits does not necessarily imply that there is a state in the catalytic cycle where both subunits bind ATP. The presence of ADP.BeF3− in these two sites could arise from a step that is not part of the normal catalytic cycle, for example a 120° rotation of the γ‐subunit without ATP hydrolysis. Similar considerations apply to the aluminium fluoride‐inhibited form of the enzyme, where ADP.AlF4−, mimicking the pentacovalent intermediate, is found in both the βDP and βTP subunits, although all current models suggest that only one of the three β‐subunits is catalytically active at any given time.
Two possible reaction schemes for ATP hydrolysis
The crystal structures of bovine F1‐ATPase complexed with aluminium and beryllium fluoride, together with the reference structure, provide detailed information on distinct conformational states of the enzyme. A remaining challenge is to relate these structures to an overall catalytic scheme, which also takes account of the results of both single‐molecule studies (Nishizaka et al, 2004) and kinetic data including tryptophan fluorescence experiments (Weber and Senior, 2000).
Two possible schemes that relate the rotation of the γ‐subunit to ATP binding and hydrolysis events in F1‐ATPase are shown in Figure 6 (adapted from Leslie and Walker, 2000). It is assumed that the nucleotide occupancy of the catalytic sites alternates between two and three, corresponding to tri‐site catalysis. This proposal is favoured by the results of tryptophan fluorescence experiments that show that on average all three catalytic sites are occupied during ATP hydrolysis at physiological rates (Weber and Senior, 2000) and by single‐molecule studies (Nishizaka et al, 2004), although the original bi‐site model (with catalytic site occupancies alternating between one and two) has also been defended recently (Boyer, 2002). The two schemes differ in the assignment of the β‐subunit where ATP hydrolysis occurs. In scheme (A), ATP binding to the βE subunit (state I) induces a large conformational change in that subunit, generating a rotation of the γ‐subunit to an intermediate position as shown in state II. More subtle conformational changes are simultaneously transmitted to the active site of the βTP subunit, promoting the hydrolysis of bound ATP to ADP+Pi. The combination of the conformational changes due to ATP binding to the βE subunit and ATP hydrolysis on the βTP subunit generates a 120° rotation of the γ‐subunit, converting the βDP subunit in state I to an open conformation (βE) in state III, releasing ADP and Pi. State III is equivalent to state I, but there has been a cyclic interconversion of the three catalytic sites. In the next step (not shown), ATP binds to the newly formed βE subunit of state III, resulting in hydrolysis of the ATP that had bound to the βE subunit in the previous step (which is transformed into a βTP subunit in state III). This produces a further rotation of the γ‐subunit by 120° and release of ADP+Pi from the newly created βDP subunit (repeating states I–III). A third round of ATP binding, ATP hydrolysis and product release brings the enzyme back to the original position.
In the alternative scheme (B), ATP binding to the βE subunit promotes hydrolysis in the βDP subunit (rather than βTP). This corresponds to hydrolysis of the nucleotide that bound two steps earlier in the sequence (where a step corresponds to a 120° rotation of the γ‐subunit). As in scheme (A), the combination of ATP binding and ATP hydrolysis (on different subunits) results in a 120° rotation of the γ‐subunit, and ADP+Pi release from the βDP subunit. Two further steps (each consisting of substeps of ATP binding followed by ATP hydrolysis and product release) bring the system back to its original position.
Both schemes are consistent with the direction of rotation of the γ‐subunit observed in single‐molecule studies (Noji et al, 1997). However, they differ in the average nucleotide occupancy of the catalytic sites, which is two ADP molecules and one ATP for scheme (A), and one ADP and two ATP molecules for scheme (B). On the basis of free energy simulations, Gao et al (2003) have described a similar scheme to (A). Tryptophan fluorescence measurements that distinguish between bound ADP and ATP support scheme (A) (Weber et al, 1996).
Recent single‐molecule experiments in which the rotation of the γ‐subunit and the binding of a fluorescent ATP analogue 2′‐0‐Cy3‐EDA‐ATP were monitored simultaneously have provided new insights into the catalytic cycle (Nishizaka et al, 2004). These results have been interpreted as showing that ATP binding to a nucleotide free β‐subunit generates an 80° rotation of the γ‐subunit, which triggers ATP hydrolysis and/or Pi release in the β‐subunit that bound ATP one step earlier in the sequence, resulting in a further rotation of 40° in the γ‐subunit. With the significant assumption that the use of a fluorescent ATP analogue while greatly reducing the rate of hydrolysis does not alter the basic reaction sequence, these results also support scheme (A).
Can the two proposed schemes be distinguished on the basis of the available crystal structures? Previously, the βDP subunit has been assigned as being closest to a catalytically active conformation on the basis of the arrangement of the catalytic residues in the structures of the aluminium fluoride‐inhibited forms of the enzyme (Braig et al, 2000; Menz et al, 2001b). On this basis, a detailed catalytic sequence based on scheme (B) was proposed (Menz et al, 2001b). Although the BeF3−‐F1 structure supports the assignment of βDP as being closer to the catalytic conformation, it is now apparent that this does not necessarily imply support for scheme (B). The recently reported single‐molecule experiments (Nishizaka et al, 2004) have been interpreted as suggesting that ATP hydrolysis probably occurs only after the 80–90° rotation of the γ‐subunit induced by ATP binding to an empty site (i.e. state II in Figure 6). There are no structural data available for an intermediate position of the γ‐subunit such as that shown in state II, but such a large rotation is likely to have a significant effect on the conformations at the catalytic sites of the βDP and βTP subunits. In particular, the βDP subunit will change towards the βE conformation, while the βTP subunit will change towards the βDP conformation. If ATP hydrolysis does indeed occur in state II, at this point the βTP subunit may adopt a conformation that is actually closest to that of the βDP subunit in the crystal structures (the catalytically active conformation), in which case the structural data also support scheme (A) in preference to scheme (B).
Because none of the crystal structures correspond exactly to the states depicted in Figure 6 in terms of the occupancy of the catalytic sites, they do not allow a clear distinction to be made between the alternative schemes. State I is best represented by the (ADP.AlF3)F1 structure (Braig et al, 2000), where AMPPNP is bound to the βTP subunit and ADP.AlF3 to the βDP subunit, but with relatively long distances between the aluminium and both the oxygen of the β‐phosphate (2.4 Å) and the nucleophilic water (3.1 Å). In terms of the overall structure of the enzyme, this arrangement is very similar to both the reference structure and BeF3−‐F1. The (ADP.AlF4−)2F1 structure, which, by analogy with other NTPases complexed with aluminium fluoride, is assumed to mimic the pentacovalent intermediate in the hydrolysis reaction, to some extent represents state II. However, the position of the γ‐subunit in this structure is closer to that in state III and the presence of ADP+Pi bound to the half‐closed βE subunit suggests that this structure represents a posthydrolysis, preproduct release state. Thus, it is best interpreted as representing an intermediate between states II and III in scheme (A).
The limited range of movement of the γ‐subunit seen in the crystal structures is to be expected if the larger rotational substep (80–90°) is associated with nucleotide binding rather than a catalytic event. All of the current structures represent conformations of the enzyme between states II and III in scheme (A) (note that states I and III are equivalent), which corresponds to a maximum rotation of 30–40° in the γ‐subunit. The possibility that lattice contacts influence the position of the γ‐subunit also cannot be ruled out, particularly since a significant fraction of the γ‐subunit (corresponding to the foot of the central stalk) was sufficiently mobile that it could not be modelled in the reference structure. However, it is unlikely that lattice contacts have a significant effect on the catalytic sites in the β‐subunits because the most flexible and structurally variable region of the γ‐subunit is below the point at which it interacts with the (αβ)3 subassembly.
The crystal structures demonstrate that the βDP and βTP catalytic sites are remarkably similar in their arrangements of catalytic residues. The most significant difference between them is the position of the guanidinium of αArg373, which is determined by the interactions between the relevant α‐ and β‐subunits, and the interactions of both α‐ and β‐subunits with the γ‐subunit. Furthermore, the BeF3−‐F1 structure illustrates how a small (ca. 1 Å) shift in the position of the guanidinium can produce a more radical difference in the position of the attacking water molecule, clearly differentiating the two sites in terms of their potential for catalysing ATP hydrolysis.
Materials and methods
Enzyme purification and inhibition
Bovine F1‐ATPase was purified from heart mitochondria as described previously (Lutter et al, 1993) with minor changes: a Sephacryl S‐300 column was replaced by an XK26/60 Superdex 200pg column, amastatin has been added to prevent proteolysis of the δ‐subunit and ATP in buffers has been replaced by ADP, 5 mM 2‐mercaptoethanol was replaced with 5 mM dithiothreitol. Glycerol treatment to remove endogenous bound nucleotides was not carried out. The purified F1‐ATPase was stored as an ammonium sulphate precipitate. It was collected by centrifugation (30 600 g, 20 min). The enzyme was exchanged on a spincolumn (Bio‐Gel P‐6 Medium, Bio‐Rad) into a buffer containing 100 mM Tris‐SO4, pH 7.8, 4 mM EDTA, 1 mM ADP and 10 mM MgSO4. To prepare the beryllium fluoride‐inhibited enzyme, the protein sample (1.3 ml, protein concentration 18.5 mg/ml) was mixed with a solution (440 μl) containing 40 mM MgSO4 and 10 mM ADP. After 20 min, 0.5 M NaF (100 μl) was added to the mixture, and 20 min later 250 mM BeCl2 (50 μl). At this point, the ATP hydrolase activity was inhibited by 98.5% (Pullman et al, 1960).
Bovine F1‐ATPase used for growing crystals of ADP‐F1 was purified as described above except that amastatin was replaced by bestatin and diprotin A. The purified enzyme was then subjected to nucleotide stripping as described previously (Lutter et al, 1993).
Crystallization and data collection
Crystals of BeF3−‐F1 were grown in microdialysis buttons (50 μl) with SpectraPor dialysis membranes (3500 MW cutoff). An equal volume of inside buffer (100 mM Tris–HCl, pH 7.2, 400 mM sodium chloride, 4 mM magnesium chloride, 0.04% (w/v) sodium azide, 2 mM EDTA, 0.004% (w/v) phenylmethylsulphonyl fluoride, 14% (w/v) polyethylene glycol (PEG) 6000, 660 μM ADP, 10 mM dithiothreitol, 1 mM beryllium chloride, 5 mM sodium fluoride) was added to the protein (final concentration 6 mg/ml). The samples were dialysed for 3 days against outside buffer (50 mM Tris–HCl, pH 8.2, 200 mM sodium chloride, 20 mM magnesium sulphate, 0.02% (w/v) sodium azide, 1 mM EDTA, 0.004% (w/v) phenylmethylsulphonyl fluoride, 9% (w/v) PEG 6000, 660 μM ADP, 5 mM dithiothreitol, 1 mM beryllium chloride, 5 mM sodium fluoride). Then, the outside buffer was replaced and the PEG 6000 concentration was increased to between 9.5 and 12.0% (in 0.25% steps). After a week, crystals began to appear and they were fully grown 2 weeks later. Typical maximum dimensions were 0.2–0.3 mm. In order to cryoprotect the crystals, the outside buffer was exchanged to one containing 13.75% PEG 6000 and 5% glycerol. Then, the glycerol concentration was increased stepwise to 10, 20 and 22% (30 min at each concentration). The crystals were harvested with a cryoloop (0.2–0.3 mm, Hampton Research, Laguna Niguel, CA), plunged into liquid nitrogen and stored at 100 K. ADP‐F1 crystals were grown at the final protein concentration of 5 mg/ml under the same conditions as the native F1 crystals (Lutter et al, 1993) except that AMP‐PNP was omitted from both the inside and outside buffers, and in the outside buffer the concentration of ADP was increased from 5 μM as in the native F1 crystallization conditions to 2 mM. The crystals were grown at a PEG 6000 concentration of 11.5%. They were cryoprotected with 25% glycerol (overnight soak at 10%, then 2 h each at 15, 20 and 25%). The crystals were frozen in a cryostream just before data collection.
Diffraction data for the BeF3−‐F1 crystal were collected to 2.2 Å resolution using an ADSC Q4R CCD detector on beamline ID14‐4 (λ=0.94 Å) at the ESRF (Grenoble, France). The ADP‐F1 diffraction data were collected using a Mar Research image plate detector on beamline ID2 (λ=0.91 Å) also at the ESRF. Both the BeF3−‐F1 and ADP‐F1 crystals belonged to the space group P212121 with unit cell dimensions a=283.5 Å, b=107.4 Å, c=137.9 Å and a=283.0 Å, b=107.6 Å, c=139.6 Å, respectively. There is one F1 molecule in the crystallographic asymmetric unit. The diffraction data were processed with MOSFLM (Leslie, 1992) and with programs from the Collaborative Computational Project Number 4 (CCP4) suite (CCP4, 1994).
Structure solution and refinement
The structure of BeF3−‐F1 was solved by molecular replacement with AMoRe (Navaza, 1994). The starting model was the reference structure refined against data collected from a single cryocooled crystal (PDB accession code 1E1Q) (Braig et al, 2000) but with the γ‐phosphate deleted from the nucleotide bound to the βTP subunit. After rigid‐body refinement with AMoRe, the R‐factor and correlation coefficient were 30.2 and 78.8%, respectively, for all data from 20.0–4.0 Å resolution. Further refinement was carried out using REFMAC5 (CCP4, 1994; Murshudov et al, 1997) alternating with manual rebuilding using O (Jones et al, 1991). The ADP‐F1 structure was solved by molecular replacement with AMoRe (Navaza, 1994) using the reference structure (PDB accession code 1BMF) as a starting model. The model was refined initially with X‐PLOR (Brünger, 1996) and then with REFMAC5 (CCP4, 1994; Murshudov et al, 1997) alternating rounds of refinement with manual rebuilding using O (Jones et al, 1991).
For the calculations of the Rfree value of the BeF3−‐F1 and ADP‐F1 structures, respectively 5 and 4.8% of the observed diffraction data were set aside and excluded from the refinement including the initial rigid‐body refinement. The final model for the BeF3−‐F1 structure includes four glycerol molecules and 1019 water molecules, and that of the ADP‐F1 structure includes three glycerol molecules and 266 water molecules.
The stereochemistry was assessed with PROCHECK (Laskowski et al, 1993). For the BeF3−‐F1 structure, 91.0% of the residues were assigned to the most favoured region of the Ramachandran plot, 8.6% to allowed regions and 0.3% to generously allowed regions. There are no residues in disallowed regions. For the ADP‐F1 structure, 90.0% of the residues are in the most favoured region, 9.6% in the allowed regions and 0.3% are in generously allowed regions.
The structures reported in this paper have been deposited in the Protein Data Bank with accession codes 1W0J (BeF3−‐F1) and 1W0K (ADP‐F1).
We thank beamline staff at the ESRF, Grenoble for assistance with data collection.
This paper is dedicated to Professor Yasuo Kagawa
- Copyright © 2004 European Molecular Biology Organization