The crystal structure of yeast mitochondrial F1 ATPase contains three independent copies of the complex, two of which have similar conformations while the third differs in the position of the central stalk relative to the α3β3 sub‐assembly. All three copies display very similar asymmetric features to those observed for the bovine enzyme, but the yeast F1 ATPase structures provide novel information. In particular, the active site that binds ADP in bovine F1 ATPase has an ATP analog bound and therefore this structure does not represent the ADP‐inhibited form. In addition, one of the complexes binds phosphate in the nucleotide‐free catalytic site, and comparison with other structures provides a picture of the movement of the phosphate group during initial binding and subsequent catalysis. The shifts in position of the central stalk between two of the three copies of yeast F1 ATPase and when these structures are compared to those of the bovine enzyme give new insight into the conformational changes that take place during rotational catalysis.
Mitochondrial F1Fo ATP synthase (EC 184.108.40.206) is responsible for the synthesis of more than 90% of cellular ATP in the eukaryotic cell under aerobic conditions. The ATP synthase is a molecular motor that converts the energy of the proton motive force across the mitochondrial membrane to a more generally usable form of energy, namely ATP. The enzyme is comprised of two functional and separable units, F1 and Fo, both of which can function as molecular motors. F1 is a water‐soluble component, which contains the catalytic sites that effect the phosphorylation of ADP:Mg. Fo is the membrane portion of the enzyme and acts as a proton turbine, utilizing the proton motive force to generate the rotation of the central stalk of F1.
F1 is composed of five different subunits with stoichiometry α3β3γδε with a combined molecular weight of 360 000 Da. The three active sites are formed at the interfaces between the α‐ and β‐subunits (Abrahams et al, 1994). Part of the γ‐subunit consists of a coiled‐coil that is contained within the α/β core of the enzyme. The mitochondrial δ‐ and ε‐subunits are bound at the foot of the γ‐subunit and the three subunits form the central stalk of the enzyme (Gibbons et al, 2000). The Fo component contains a membrane rotor made up of 10 copies of the c‐subunit in the yeast enzyme. The c‐subunit contains an essential carboxylate group that forms part of the proton translocation pathway, which is believed to involve two half channels at the interface of the membrane‐bound a‐subunit and the ring of c‐subunits. A peripheral stalk formed minimally by subunits b, d, h, and subunit 5 (OSCP) links the a‐subunit to the periphery of the α/β core and acts as a stator, preventing the core from following the rotation of the central stalk (Kane Dickson et al, 2006; Walker and Kane Dickson, 2006). Additional subunits (subunits 8, e, f, g, i, and k) whose roles are currently uncertain and two regulatory proteins (Inh1, Stf1) complete the composition of F1Fo. Overall, the yeast mitochondrial ATP synthase is composed of 18 polypeptides and has a molecular weight of over 550 000 Da.
The binding change mechanism provides a framework for explaining how energy derived from the proton motive force is used to synthesize ATP (Boyer et al, 1973; Mitchell, 1977). According to this scheme, the three active sites adopt different conformations, which interconvert sequentially during catalysis. ATP synthesis occurs in the site with highest affinity for ATP, referred to as the tight site, and this step is proposed to be isoenergetic between bound ADP:Pi and ATP. Energy derived from the proton motive force is required to alter the conformation of the tight site to one with lower affinity, allowing the release of newly synthesized ATP. The crystal structure of bovine F1 (Abrahams et al, 1994) shows several features that are consistent with the binding change mechanism. In this structure, two of the catalytic sites are in a closed conformation. One, at the interface of the αDP‐ and βDP‐subunits, contains ADP (DP site), while the second, at the interface of the αTP‐ and βTP‐subunits, contains AMP‐PNP (TP site). The third catalytic site, at the interface of the αE‐ and βE‐subunits, is in an open conformation and devoid of bound nucleotide (E site). It was proposed that DP represents the high‐affinity (tight) catalytic site, TP represents the intermediate affinity nucleotide‐binding site, while newly formed product will be released from the low affinity E site (Abrahams et al, 1994). The structure also suggested that rotation of the γ‐subunit within the α3β3 subcomplex was responsible for the interconversion of the active sites (Abrahams et al, 1994). ATP‐dependent rotation of the γ‐subunit was dramatically demonstrated by single molecule experiments showing rotation of a fluorescently labeled actin filament covalently bound to the γ‐subunit (Noji et al, 1997).
The electron density map of the yeast F1c10 subcomplex revealed a ring of 10 c‐subunits, which is in intimate contact with the foot of the central stalk (Stock et al, 1999). One complete rotation of the ring results in the translocation of 10 protons (one per subunit) across the membrane, possibly by a mechanism involving two half‐channels located at the interface between the a‐subunit and the c‐ring as first proposed by W Junge in 1993 (Junge et al, 1997). In the F1 domain, a complete rotation of the γ‐subunit is accompanied by the synthesis of three molecules of ATP. The stoichiometry of the c‐ring dictates a nonintegral proton:ATP ratio, which requires that the synthesis of one ATP molecule is accompanied by the translocation of a mean value of 3.3 protons. This suggests that there must be some elasticity within the central stalk.
The crystal structure of yeast mitochondrial F1‐ATPase described here shows the same asymmetric features as those described for the bovine enzyme. An analysis of the differences in conformation between the three independent copies of yeast F1 and comparisons with structures of the bovine enzyme have provided new insights into the catalytic mechanism, and in particular the details of the phosphate (Pi)‐binding site.
Yeast F1 ATPase was crystallized in a monoclinic space group (P21) with three F1 complexes (over 9000 residues) in the crystallographic asymmetric unit (Mueller et al, 2004a). The structure was solved by molecular replacement (Materials and methods) and refined against X‐ray data to a resolution of 2.8 Å (Table I).
The quality of the electron density of the three F1 complexes was variable. The F1 complex with the best‐defined density is referred to as yF1I, while yF1II and yF1III have progressively less‐well‐defined density. Electron density for the central stalk (composed of the γ‐, δ‐, and ε‐subunits) was sufficiently good for yF1I to allow building a nearly complete model, while these subunits are less complete for yF1II and III. Details of the residues present in the final models for yF1I, II, and III are given in Supplementary data. yF1I and II lie ‘side by side’ with a similar orientation, but rotated along the axis of the γ‐subunit by about 180o, while yF1III is located above and between yF1I and yF1II. Thus, the crystal contacts are quite different for the three complexes (Supplementary Table 1) and this is presumably responsible for the difference in quality of the electron density.
The scheme used for naming the α‐ and β‐subunits of the bovine enzyme will be used here (Abrahams et al, 1994). Yeast residue numbering will be used, with the corresponding bovine residue number, if different, in parentheses. The insertions and deletions when comparing bovine and yeast subunits are detailed in Supplementary data.
Overall structure of the yeast F1‐ATPase
The structure of yF1I was compared to that of yF1II and yF1III and also to the bovine enzyme. The structure of bovine F1 ATPase inhibited by BeF3−:ADP (Kagawa et al, 2004) was used for comparison of the α‐ and β‐subunits. BeF3−:ADP mimics ATP and is the closest structural homolog to AMP‐PNP, which is bound to both the βDP‐ and βTP‐subunits in all three complexes in the yeast crystal structure. All noncatalytic nucleotide‐binding sites in yF1 contain AMP‐PNP. The structure of the bovine enzyme inhibited with dicyclohexylcarbodiimide (DCCD) (Gibbons et al, 2000) was used for comparison of the γ‐, δ‐, and ε‐subunits, because this structure has the most complete model of the central stalk.
The overall structures of all three complexes of the yeast F1 ATPase in the crystallographic asymmetric unit are very similar (Table II). However, it is clear that the structure of yF1I is more similar to yF1III than to yF1II. While the overall differences are relatively small, they lead to some significant differences in the active sites. Because yF1I and III adopt very similar conformations, and yF1I is more complete, the following discussion will focus on yF1I and II.
When yF1I is compared to bovine F1 ATPase (Table II), the differences in α‐carbon positions are less than 1 Å for individual α‐ and β‐subunits, and 1.2 Å for the α3β3 subcomplex, indicating that the overall structures are very similar. The differences are significantly larger for the γ‐, δ‐ and ε‐subunits, but in the case of the γ‐subunit decrease dramatically if the subunit is broken down into subdomains. For instance, for residues 1–30, 106–151, and 206–276, the differences are 0.8, 1.0, and 1.3 Å, respectively.
The α/β pairs that form the catalytic interfaces in yF1I and II display larger differences than when individual subunits are compared (Table II). This is particularly true for the αDP/βDP pair, and reflects the more open αDP/βDP interface in yF1I compared to yF1II (Figure 1). The opening of this interface can be described as a relative rotation of 5.5° of the αDP‐subunit relative to the βDP‐subunit, and this rotation has the greatest effect on the relative positions of the C‐terminal helical domains. The αDP/βDP interface in bovine F1 is very similar to that in yeast yF1II (Figure 1), so that yeast yF1I represents the novel conformation.
Similar but smaller differences are seen when comparing the αE/βE interface. Again, the interface is more open in yeast yF1I than in yF1II, and in this case the relative rotation of the αE/βE subunits is 3.7°. As with the αDP/βDP interface, yF1II is very similar to the bovine structure. By contrast, the αTP/βTP interface is essentially the same in yeast yF1I and II, and both closely resemble the corresponding interface in bovine F1.
The DP and TP active sites
The catalytic sites of the yeast enzyme share many features found in structures of the bovine enzyme. The most significant difference is that the yeast βDP active site has AMP‐PNP:Mg bound in all three complexes rather than ADP:Mg (Supplementary Figure 2S). Based on the quality of the electron density and the refined B‐factors for the nucleotide, the βDP active site is fully occupied with AMP‐PNP:Mg.
The arrangement of catalytic residues in the yeast DP active site is similar to that of the corresponding site in BeF3−‐inhibited bovine F1 (Figure 2), although small shifts in side chain positions are observed. The side chain interactions within the active sites of the bovine and yeast enzymes are shown schematically in Supplementary Figure 3S. For simplicity, the βDP active sites of yeast yF1I, yF1II, and yF1III complexes will be referred to as DP‐I, DP‐II, and DP‐III in the following discussion. Yeast βGlu189(188) acts as a general base in the catalytic reaction. The distance of the phosphorous atom in the γ‐phosphate (or Be atom in BeF3−), from Cγ of the carboxylate of βGlu189(188) is 5.4, 5.6, and 5.2 Å in βDP of the bovine enzyme, and the DP‐I and DP‐II sites of yeast, respectively, and 5.7 Å in the corresponding TP sites. While the reactive water molecule is not observed in the yeast model in either the DP‐I or DP‐II sites, there is some weak electron density in the DP‐III site that may be due to the catalytic water. The apparent absence of the catalytic water in the yeast structure is probably due to the limited resolution of the X‐ray data.
The structure of bovine F1 ATPase with the transition state analog, ADP:AlF4−, bound to the βDP and βTP sites provides good evidence that αArg375(373) stabilizes the transition state of the pentacoordinate intermediate during the reaction (Menz et al, 2001). Biochemical evidence supports this role for αArg375(373) and also suggests that this residue is critical in the cooperativity of the enzyme (Le et al, 2000; Senior et al, 2002). When the structures are superimposed using the P‐loop as the reference point, the guanidinium group of αArg375(373) of the yeast DP‐II active site is nearly identical in position to αArg373 of the bovine enzyme, but in the yeast DP‐I active site residue αArg375(373) is 1 Å closer to the γ‐phosphate of the nucleotide. The change in position of αArg375(373) is due to a translation of the α‐carbon atom of αArg375(373) in the DP‐I active site of about 1.8 Å along an axis that is parallel to the axis formed by the α‐ and β‐phosphates of the nucleotide.
βArg190(189) is critical for both nucleotide binding and transition state stabilization (Le et al, 2000; Menz et al, 2001; Senior et al, 2002). In the yeast DP‐II active site, the guanidinium group forms a bidentate salt bridge interaction with one oxygen atom on the γ‐phosphate and a monodentate interaction with a second oxygen atom of the γ‐phosphate (Supplementary Figure 3S). In yeast DP‐I active site, this residue is displaced 0.5 Å away from the γ‐phosphate resulting in the loss of the single interacting salt bridge, but the bidentate interaction is retained (Supplementary Figure 3S). The position of βArg190(189) relative to the nucleotide in yeast DP‐II active site is very similar to that of the corresponding residue (βArg189) in the βDP‐subunit of bovine F1 with bound BeF3−:ADP (Kagawa et al, 2004).
The binding pocket for the ribose moiety of the nucleotide is more solvent accessible in the αDP/βDP interface in the yeast complex yF1I than in yF1II, as a result of the relative rotation of the α‐ and β‐subunits described above. This is reflected in the loss of a hydrogen bond between the carbonyl oxygen of αSer374(372) and the O2’ oxygen of the ribose which are 3.0 Å apart in the yF1II (3.1 Å in bovine F1) but 5.3 Å apart in yF1I.
Overall, the conformation of yeast DP‐II catalytic site is more similar to bovine BeF3−:ADP F1 than yeast DP‐I. In contrast, the βTP catalytic sites are very similar in yeast yF1I and II, and both closely resemble βTP site of the bovine BeF3−:ADP F1 structure.
Pi is bound to the βE‐subunit of yF1II
Following initial refinement of the model, a strong electron density peak (8σ) was found in the general region of the nucleotide‐binding site of the βE‐subunit of yF1II (Figure 3A), which was modeled as Pi. The identity of the molecule is not certain, as sulfate would fit the density equally well and 0.25 mM NiSO4 was present in the crystallization medium, but for simplicity, the bound ligand will be described as phosphate. There are multiple interactions between the phosphate oxygen atoms and side chain atoms of αArg375(373), βLys163(162), βArg190(189), βAsp256, βAsn257, and βArg260 (Figure 3B). βLys163(162) forms a salt bridge with the carboxylate of βAsp256, which may reduce the negative charge of the carboxylate allowing H‐bonding with the proton of the phosphate oxygen. Interestingly, the Pi‐binding site is far from the P‐loop. If the βDP‐subunit of yF1I is superimposed on the βE‐subunit of yF1II using the P‐loop as the reference, the phosphate is 7.7 Å from the γ‐phosphate of AMP‐PNP. A change in conformation of the βE‐subunit from the open to the closed state, presumably as a result of movement of the γ‐subunit and/or binding of ADP to βE, is required to move the phosphate group for phosphorylation of ADP during synthesis. The structure suggests that during ATP hydrolysis, the phosphate group is not released until the βE‐subunit has adopted a fully open conformation.
What are the structural differences, which account for binding of Pi to βE of yF1II but not yF1I? A superposition of the Pi‐binding site in yF1I and yF1II (Figure 3C) indicates only small differences in the positions of the residues involved in binding Pi. The biggest differences are shifts of the guanidinium groups of αArg375(373) and βArg190(189) by 1.3 and 0.8 Å, respectively. In both cases, the guanidinium group is located closer to the Pi site in yF1II than in yF1I. These small shifts are related to the slightly more open form of αE/βE interface in yF1I compared to yF1II, which in turn is associated with the different position of the γ‐subunit relative to the α3β3 core in these two complexes.
For all the reported bovine structures, with the possible exception of the enzyme inhibited with DCCD (Gibbons et al, 2000), there is no evidence for Pi binding at the equivalent site in the open βE‐subunit. In several structures, there was evidence that a disordered phosphate or sulfate group was bound to the P‐loop, but this was not then, and is not now, thought to be the biologically relevant Pi‐binding site (Abrahams et al, 1994, 1996; Kagawa et al, 2004). In order to understand why the yF1II complex, but not the bovine enzyme, binds Pi, the structure of this region of bovine βE was superimposed onto that of βE from yF1II (Figure 3D). The differences are larger than those seen when comparing yF1I and II. The guanidinium group of αArg375(373) is displaced 2.6 Å and is in a different rotamer conformation in the bovine structure. The α‐carbons of βLys163(162), βArg190(189), and βAsn257 superimpose quite well, but the side chains have different rotamer conformations. The result is that there is no well‐defined Pi‐binding pocket in bovine βE.
A surface representation of the αE/βE interface for yF1II reveals no obvious access to the Pi‐binding site from the external surface of the F1 complex. If ADP were bound, access would be even less favorable due to electrostatic repulsion from the nucleotide diphosphate in addition to the steric hindrance of the neighboring protein atoms. However, the phosphate site is readily accessible from the internal solvent‐filled cavity of F1. This internal cavity is connected to the external milieu by a large access channel between the C‐terminal domains of the αE‐ and βE‐subunits (Supplementary Figure 4S).
Conformation of the central stalk
Electron density for the central stalk region was visible in all three complexes, but the extent and quality of this density was variable. The most complete model was built for the stalk of yF1I (Figure 4). The overall fold of the γ‐, δ‐, and ε‐subunits are identical to those of the corresponding bovine subunits, with the exception of the C‐terminal extension present in yeast ε‐subunit. The three subunits also associate in a very similar way in the bovine and yeast structures. The γ‐subunit contains a coiled‐coil formed by an N‐terminal α‐helix (residues 1–56), a C‐terminal α‐helix (residues 208–277), and a foot composed of an α/β bundle (Figures 4A and B). The δ‐subunit consists of a β‐sandwich made up of two 5‐stranded β‐sheets (residues 13–86) followed by two α‐helixes (residues 99–137) (Figure 4D). The ε‐subunit has a hairpin‐like structure formed by two α‐helixes (residues 10–23 and 31–36) with extended regions of polypeptide at the N‐ and C‐termini. The yeast ε‐subunit has 10 extra amino acids at the C‐terminus compared to bovine ε‐subunit. These residues fold back and sit under the central core of the central stalk (Figure 4E).
There is a noticeable kink in the C‐terminal helix of the yeast γ‐subunit in the region of residue 250 (245) and a smaller kink in the N‐terminal helix at residue 16. These features are associated with rather different positions of the central stalk relative to the α3β3 subcomplex in the yeast and bovine structures. If the β‐barrel domains of the α‐ and β‐subunits are used to superimpose the bovine F1‐DCCD and yeast yF1I structures, the lower (membrane proximal) regions of the γ‐subunit are rotated counter‐clockwise by about 30o, when viewed from the foot of the stalk (Figure 4B). This is the direction that the central stalk rotates during the hydrolysis of ATP. However, the rotation axis is not parallel to the pseudo three‐fold axis of F1 (deviation ∼16°) nor does it pass through the center of the α3β3 subcomplex (the minimum displacement is about 5 Å), and therefore it is not clear how closely this 30° rotation models the rotation of the stalk that occurs during catalysis. The twisting of the central stalk is also not uniform: the C‐terminal residues (γ245–277) of the C‐terminal α‐helix superimpose very well with little or no evidence of rotation, but the twisting becomes progressively greater towards the N‐terminal region of the helix, reaching a maximum displacement of 14.6 Å in the foot of the γ‐subunit. The N‐terminal α‐helix does not overlap well from the very beginning of the helix, but like the C‐terminal α‐helix, the overlap is poorer as the comparison approaches the foot of the γ‐subunit. In addition, the N‐terminal α‐helix of the yeast enzyme is translated by about 2.5 Å parallel to the helix axis. Despite these differences, the side chain interactions between the N‐ and C‐terminal helices within the coiled‐coil are largely conserved. This conservation occurs by some changes in the rotamer conformations of side chains as well as smaller movements in the backbone of the N‐terminal α‐helix.
Rotation of the central stalk
The position of the central stalk relative to the α3β3 subcomplex determines the catalytic states of the three catalytic sites. In order to compare the different bovine structures with the two distinct conformations of the yeast enzyme present in yF1I and II, all the structures were superposed on their respective pseudo three‐fold axes (defined using the α‐subunits) so as to give the best superposition of the N‐terminal β‐barrel domains. The rotation of the central stalk was then defined as the angular rotation about the (common) pseudo three‐fold axis, which provided the best superposition of residues 18–25 of the γ‐subunit, as this region of the γ‐subunit is involved in critical contacts with the C‐terminal domain of the βE‐subunit, imposing the open conformation. The bovine reference structure (Abrahams et al, 1994), the ADP:AlF4−‐inhibited bovine structure, and the yF1I and II structures of yeast were compared in this way. Relative to the bovine reference structure, the γ‐subunit of yeast yF1I has rotated 12° in the direction of ATP hydrolysis (counter‐clockwise as viewed from the foot of the stalk), while the rotation is 4° and 16°, both in the direction of ATP synthesis, for the γ‐subunits of yF1II of yeast and ADP:AlF4−‐inhibited bovine F1. These rotations, and the resulting changes at the catalytic sites, are entirely consistent with the biochemical pathway for ATP synthesis or hydrolysis. The 16° rotation (yF1I to yF1II, Figure 4F) is associated with Pi binding to the βE‐subunit on the ATP synthesis pathway, and the additional 12° rotation (yF1II to ADP:AlF4− inhibited bovine) corresponds to binding of ADP and the partial closure of the catalytic site of βE. The 12o rotation in the direction of ATP hydrolysis (bovine reference to yF1I) results in partial opening of the αDP/βDP catalytic interface, in preparation for the release of the products of hydrolysis from this site.
The crystal structure of yeast F1 ATPase described here is the first high‐resolution example of an F1 ATPase from a source other than bovine mitochondria in which the asymmetric features of the enzyme have not been obscured by crystallographic symmetry. The presence of a crystallographic three‐fold axis in structures of the rat mitochondrial and chloroplast enzymes (Bianchet et al, 1998; Groth and Pohl, 2001) results in an electron density map in which any difference in conformation of the α/β subunits has been averaged, making interpretation difficult. It is therefore highly significant that the three independent yeast F1 structures show essentially the same asymmetric features as those observed in the bovine structures, despite a completely different crystal packing of the F1 complexes. This finding supports the view that the observed asymmetry is an essential feature of the structure of F1 ATPase and a crucial element of the rotary catalytic mechanism.
The initial structure of the bovine enzyme was believed to represent the ADP‐inhibited form (Abrahams et al, 1994), primarily based on the observation that Mg:ADP, and not Mg:AMP‐PNP, was bound to βDP. This led to speculation that the structure was of an inactive form of the enzyme that did not represent a conformation adopted on the catalytic pathway. There was also uncertainty whether the E site was competent for binding substrates. Many of the subsequent bovine structures support the view that the original structure is representative of an intermediate in the reaction pathway, most recently, the structure with ADP:BeF3− bound to two of the three active sites (Kagawa et al, 2004). The structures of the yeast enzyme provide two additional lines of evidence that also support this view. First, all three structures of the yeast enzyme have Mg:AMP‐PNP bound to both βDP and βTP. Thus, the conformations of the yeast enzyme cannot be that of the ADP‐inhibited form of the enzyme and yet they all have the same asymmetric features of the bovine enzyme. Secondly, βE of yF1II has Pi bound to a site that is consistent with biochemical data on Pi binding. Thus the structure of the enzyme represents an active intermediate in the catalytic pathway.
Why is Mg:AMP‐PNP bound to the βDP site of the yeast enzyme but Mg:ADP is bound to the bovine βDP site? Possibly, the Mg:ADP in the bovine βDP site is due to the slow hydrolysis of Mg:AMP‐PNP (Tomaszek and Schuster, 1986) and small differences in the structure of the yeast enzyme make the βDP site less reactive and unable to hydrolyze AMP‐PNP.
Analysis of the central stalk has provided additional insight into some features of the structural architecture of the enzyme. The foot of the central stalk in the bovine structure inhibited with DCCD was rotated 30° and displaced by about 14 Å relative to that of yF1I. Despite this large shift in the foot of the γ‐subunit, the final 31 residues of the C‐terminal α‐helix overlap well and residues 1–40 of the N‐terminal α‐helix, although displaced, remain close to the position of the corresponding yeast N‐terminal helix. The relative position of the final 31 residues of the C‐terminal α‐helix of the γ‐subunit is the same regardless which structures are compared. This suggests that the proline‐rich collar, first described in the native structure of the bovine enzyme, appears to be at least one structural constraint that holds this in position.
The Pi‐binding site
The analysis of the Pi‐binding site identified in the βE‐subunit of yF1II provides a new understanding of the structural changes that occur during catalysis. Bound phosphate, while not observed in other structures, is not unexpected as biochemical studies have indicated that bovine F1 ATPase has one or two high‐affinity phosphate‐binding sites (Penefsky, 1977). There are six amino‐acids residues, αArg375, βLys163, βArg190, βAsp256, βAsn257, and βArg260, which form the Pi‐binding site in the empty subunit. Of these, there is direct biochemical evidence that βLys163, βArg190, βArg260, and αArg375 are involved in binding Pi in the Escherichia coli enzyme (Ahmad and Senior, 2004a, 2004b, 2005). The ATPase activity of an E. coli F1 mutant with βAsn257 (yeast numbering) mutated to alanine was protected from inhibition by NBD‐Cl in the presence of 2.5 or 10 mM Pi (Ahmad and Senior, 2004b). This observation was interpreted as suggesting that βAsn257 is not directly involved in binding Pi, but can be explained equally well if, at these concentrations, Pi binding is not affected by the loss of the interaction with the asparagine.
A retrospective analysis was carried out on all published structures of bovine F1 ATPase focusing on the region of the Pi‐binding site identified in the yF1II complex. This revealed that in the structure of bovine F1 inhibited with DCCD (Gibbons et al, 2000), there is a ligand, modeled as a sulfate, that binds in the same region. However, the position of this sulfate differs by 2.0 Å, and only three residues (βArg189, βArg260, and αArg373) are involved in binding the sulfate group. It is possible that this represents an intermediate state of the Pi‐binding site, with a lower affinity than the site characterized in the yF1II complex.
The recent structure of the bovine F1 ATPase (Bowler et al, 2006) in which the inhibitor azide has been resolved suggests that azide, which was present in some of the earlier crystals of bovine F1, may affect Pi binding. Biochemical studies indicate that azide inhibits Pi binding to bovine F1 ATPase (Kasahara and Penefsky, 1978).
Rotation of the γ‐subunit
The γ‐subunit of yF1II complex is rotated 16o clockwise relative to that of the γ‐subunit of yF1I (as viewed from the foot of the γ‐subunit). The clockwise rotation is in the direction of ATP synthesis, which is expected if Pi is binding to the substrate‐binding site. Prior studies have suggested that a proton gradient across the mitochondrial membrane is required for the efficient binding of Pi to the enzyme (Boyer, 1989; Al‐Shawi and Senior, 1992; Al‐Shawi et al, 1997), implying that the γ‐subunit must be rotated in the direction of ATP synthesis. This is consistent with the structural results, although it is not clear if the differences in conformation of the yF1I and yF1II complexes are due to crystal contacts or whether there are two distinct conformations present in the crystallization medium (with and without bound Pi). Even in the former case, there is strong support from the biochemical data that the yF1II structure represents a physiologically relevant mode of Pi binding.
Many of the residues involved in the Pi‐binding site also provide ligands for binding nucleotide or Mg2+ in the βDP and βTP catalytic sites. Of the six residues that form the Pi‐binding site, αArg375, βLys163, and βArg190 interact directly with the nucleotide phosphates and in the bovine enzyme, βAsp256 forms a H‐bond with a water molecule, which in turn is bound to the Mg2+ ion. A Pi‐binding site is also present in the structure of the bovine enzyme inhibited with AlF4−:ADP (1H8E) (Menz et al, 2001). The binding of Mg:ADP and sulfate to the βE‐subunit is coincident with the rotation of the γ‐subunit, relative to the γ‐subunit of yF1I, by 28° in the direction of ATP synthesis and with the partial closing of βE. This structure is considered to be an intermediate in the reaction pathway.
A model for the conformational changes accompanying ATP synthesis
Comparison of the structures of the βE‐subunits from yF1II and the bovine enzyme inhibited with AlF4−:ADP, and the βDP site of yF1I, provides a model for the movement of phosphate during the catalytic reaction (Figure 5).The P‐loop motif overlaps very well in all three structures, as do the nucleotides bound to the βE site of the bovine enzyme and the βDP site of yF1I. Assuming that these three structures represent different steps in the reaction mechanism, this illustrates the relative movements of the catalytic side chains and the phosphate group during catalysis.
From a combination of the structural and biochemical data, the following model of ATP synthesis can be assembled. Rotation of the γ‐subunit (driven by proton translocation through Fo) results in phosphate‐binding to an unliganded βE‐subunit and a partial closure of the βE/αE interface. At this stage, the bound phosphate molecule is located almost 8 Å from its end point, marked by the γ‐phosphate of AMP‐PNP. The binding of Mg:ADP to the βE‐subunit is accompanied by further rotation of the γ‐subunit and a more dramatic closure of the βE/αE interface. During this stage, the phosphate has moved approximately 5 Å to the intermediate position observed in the AlF4−:ADP‐inhibited bovine F1 structure. Further rotation of the γ‐subunit results in the close approach of the phosphate group to the β‐phosphate of ADP and positions the catalytic residues (βLys163, βGlu189, βArg190, αArg375) appropriately for an in‐line attack that results in the formation of ATP.
Although the model as presented above implies that Pi binds before the nucleotide (i.e. ordered binding of substrates), this is not a requirement. The formation of the nucleotide‐binding site may be dependent on a small rotation of the γ‐subunit, but it is not dependent on the presence of Pi. Therefore, there is no reason why the nucleotide should not bind before Pi. This raises the question of why nucleotide was not bound to the βE‐subunit in spite of the presence of both AMP‐PNP and ADP in the crystallization medium. The concentration of ADP (0.025 mM) is close to the Kd for binding to the third site and the concentration of AMP‐PNP (0.5 mM) is well above the Kd (Corvest et al, 2005). At present there is no satisfactory explanation for this result. The presence of a channel leading to the Pi‐binding site from the internal solvent‐accessible cavity of F1 is consistent with the random order of binding, or release, of ADP and Pi (Kayalar et al, 1977; Perez and Ferguson, 1990; Milgrom and Cross, 1997). If access to the Pi site is only possible from the external surface of the α3β3 subcomplex, the binding of ADP or ATP would severely inhibit the subsequent binding of Pi, resulting in an ordered reaction. It should be noted that models of ordered binding, with either Pi or ADP binding first, have been suggested (Gao et al, 2005; Xing et al, 2005).
Mg:ADP and Mg:ATP or ADP and ATP bind with similar affinities to the lowest affinity site of the E. coli enzyme (Weber et al, 1993, 1994). This is consistent with the structural data, which show that the side chains that could interact with the γ‐phosphate are in fact far removed from the nucleotide‐binding site when the β‐subunit adopts the open conformation. Similarly, because the phosphate‐binding site is spatially distinct from the nucleotide‐binding site, it would be predicted, as observed, that phosphate does not inhibit the binding of either ATP or ADP (Weber and Senior, 1995).
This model also provides an explanation for how the enzyme is able to make ATP under cellular conditions where the ATP/ADP ratio is from 10 to 50. ATP cannot be accommodated in the half‐closed conformation of the βE‐subunit in the AlF4−‐inhibited bovine enzyme, as this would result in severe steric clashes of the γ‐phosphate with both the side chain of αArg373 and the bound phosphate. Consequently, regardless of the order of binding of phosphate and nucleotide, the required conformational change to the half‐closed state can only take place when ADP, and not ATP, is bound.
There has been some debate on whether the F1 ATPase follows bi‐site or tri‐site kinetics (Milgrom et al, 1998; Weber and Senior, 2001; Nishizaka et al, 2004; Milgrom and Cross, 2005). The structure of the yF1II complex is supportive of the tri‐site kinetic model, since there are two molecules of AMP‐PNP and a phosphate bound to the three catalytic sites that is a predicted intermediate during tri‐site, but not bi‐site catalysis. However, while supportive, this structure does not prove tri‐site kinetics.
Comparison with other molecular motors
F1‐ATPase is structurally homologous to the hexameric AAA+ family of proteins. Both contain a core nucleotide‐binding domain with a RecA fold and both undergo large conformational changes in response to nucleotide binding. Comparison of these molecular motors reveals both similarities and differences. The large‐scale domain motions are similar in nature and magnitude in F1‐ATPase and the heat shock protein HslU (Wang, 2004). However, the small local re‐arrangements that accompany the pseudo rigid‐body domain rotations, for example those involved in the formation of the phosphate‐binding site in yF1II, are equally important. These details are likely to be specific to each individual motor. The coupling between hydrolysis events in different subunits provides another example of the differences between F1‐ATPase and AAA+ proteins, and probably between different members of the AAA+ family. In F1‐ATPase, the asymmetric interactions with the γ‐subunit determine the catalytic state of the β‐subunits. Further insight into the relationship between the orientation of the γ‐subunit and the nature of the catalytic interfaces is revealed by comparison of the yF1I and yF1II complexes. Rotation of the γ‐subunit ensures strictly sequential hydrolysis of ATP in the β‐subunits. By contrast, recent biochemical data suggest that the ClpX protease undergoes a random order of hydrolysis (Martin et al, 2005), while a concerted mechanism has been proposed for the SV40 helicase (Gai et al, 2004). While following some common general principles, there are important differences in the mode of action of these superficially similar molecular motors.
Materials and methods
A genetically engineered yeast strain, DMY301[pβNH6] was used to express and purify the F1 ATPase. This strain has a null mutation in the gene encoding the β‐subunit of F1 ATPase and it was transformed with a vector that contains the gene encoding the β‐subunit with a His6 tag just after the mitochondrial leader peptide. The expression and purification of the His6 F1 ATPase is described elsewhere (Mueller et al, 2004b).
Crystallization, data reduction, and refinement
Crystals were obtained by vapor diffusion using the sitting drop method and frozen as described (Mueller et al, 2004a). Data were collected from a single crystal at ID5 of the Advanced Photon Source at 100 K and processed using MOSFLM (Leslie, 1992) and SCALA (Evans, 2006). Molecular replacement was performed using the bovine structure (1E1Q), minus the nucleotides as an initial model in AMoRE (Navaza, 2001). Model building was performed with XtalView (McRee, 1999), O (Jones et al, 1991), and COOT (Emsley and Cowtan, 2004), and refinement was done with CNS (Brunger et al, 1998) and REFMAC5 (Murshudov et al, 1997). Noncrystallographic symmetry (NCS) restraints for main chain and side chains were applied to improve the refinement. TLS refinement (Schomaker and Trueblood, 1968; Winn et al, 2001) was used to account for overall anisotropic motion of the molecules. Data processing and refinement statistics are shown in Table I. The stereochemistry was assessed using PROCHECK (Laskowski et al, 1993). The Ramachandran plot showed that 99.5, 99.3, and 99.6% of residues are in the most favored or additionally favored regions for yF1I, II, and III, respectively. Structure comparisons, axis rotations, and r.m.s. deviations were performed with SUPERPOSE, LSQKAB, and TOP3D, contained in the CCP4 (Collaborative_Computational_Project, 1994) package, EDPDB (Zhang and Matthews, 1995), and MOE (Group, 2004). Figures were generated using Pymol (DeLano, 2002), SPOCK (Christopher, 1998), GRASP (Nicholls et al, 1991), and O (Jones et al, 1991).
Coordinates and structure factors have been deposited in the Protein Data Bank (accession code 2HLD).
Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).
This work was supported by grants from the National Institute of Health (NIH, R01‐GM067091 and R01‐GM066223) to DM Mueller and Medical Research Council support to JE Walker and AGW Leslie. DM Mueller received a Fogarty Senior Fellowship from the NIH (1F06TW002379) for part of the tenure of this project. Data were collected in part at the DuPont‐Northwestern‐Dow Collaborative Access Team (DND‐CAT ID‐5A) beam lines at the Advanced Photon Source, Argonne National Laboratory with assistance from Zdzislaw Wawrzak. The DuPont‐Northwestern‐Dow Collaborative Access Team (DND‐CAT) sector of the APS is supported by EI Dupont de Nemours & Co., the Dow Chemical Company, and the National Science Foundation through Grant DMR‐9304725. Use of the APS was supported by the US Department of Energy, Office of Science of Basic Energy Sciences, contract No. W‐31‐109‐ENG‐38. This project was aided with the use of the X‐ray facility at Rosalind Franklin University.
- Copyright © 2006 European Molecular Biology Organization