Mitochondrial ATP Synthase

ATP synthesis from ADP, Pi, and Mg2+ takes place in mitochondria on the catalytic F1 unit (α3β3γδϵ) of the ATP synthase complex (F0F1), a remarkable nanomachine that interconverts electrochemical and mechanical energy, producing the high energy terminal bond of ATP. In currently available structural models of F1, the P-loop (amino acid residues 156GGAGVGKT163) contributes to substrate binding at the β subunit catalytic sites. Here, we report the first transition state-like structure of F1 (ADP·Vi·Mg·F1) from rat liver that was crystallized with the phosphate (Pi) analog vanadate (VO3-4 or Vi). Compared with earlier “ground state” structures, this new F1 structure reveals that the active site region has undergone significant remodeling. P-loop residue alanine 158 is located much closer to Vi than it is to Pi in a previous structural model. No significant movements of P-loop residues of the α subunit were observed at its analogous but noncatalytic sites. Under physiological conditions, such active site remodeling involving the small hydrophobic alanine residue may promote ATP synthesis by lowering the local dielectric constant, thus facilitating the dehydration of ADP and Pi. This new crystallographic study provides strong support for the catalytic mechanism of ATP synthesis deduced from earlier biochemical studies of liver F1 conducted in the presence of Vi (Ko, Y. H., Bianchet, M., Amzel, L. M., and Pedersen, P. L. (1997) J. Biol. Chem. 272, 18875-18881; Ko, Y. H., Hong, S., and Pedersen, P. L. (1999) J. Biol. Chem. 274, 28853-28856).

The mammalian mitochondrial ATP synthase (F 0 F 1 ) is a large protein complex (Fig. 1A) located in the inner membrane, where it catalyzes ATP synthesis from ADP, P i , and Mg 2ϩ at the expense of an electrochemical gradient of protons generated by the electron transport chain (reviewed in Refs. 1 and 2). Excluding regulators, the mammalian ATP synthase consists of 15 subunit types (3,4), of which five in the stoichiometric ratio ␣ 3 ␤ 3 ␥␦⑀ comprise the F 1 catalytic unit (5), an ATP hydrolysis-driven motor, whereas an additional 10 comprise F 0 . One part of F 0 , containing subunits a and c, is embedded in the inner membrane to form a proton-driven motor, whereas a second part composed of subunits b, oligomycin sensitivity conferring protein (OSCP), and F 6 forms a side stalk (4,6,7) or "stator" extending from the membrane-embedded segment to the top of F 1 (4,8).
During ATP synthesis in intact mitochondria, it is strongly believed, based on three-dimensional structures of F 1 (9,10) and single molecule technology applied to the simpler bacterial enzymes (11,12), that the proton-driven motor contained within F 0 drives the ATP hydrolysisdriven motor (i.e. F 1 ) in the reverse direction to make ATP (Fig. 1A). In addition, more recent studies show that in mitochondria, the entire "double motor" ATP synthase complex is itself in complex formation with the phosphate carrier and adenine nucleotide carrier, forming an ATP synthase-phosphate carrier-adenine nucleotide carrier supercomplex named the "ATP synthasome" (13,14).
To fully understand how ATP synthases work, we must elucidate three different levels of mechanism (Fig. 1A), these being (a) the mechanism (electrochemical) of proton gradient-driven rotation of the centrally located ␥ subunit; (b) the mechanism (mechanical) of coupling ␥ subunit rotation at every 120°to conformational/binding changes in each ␣␤ subunit pair; and finally (c) the mechanism (chemical) of ATP synthesis from ADP, P i , and Mg 2ϩ on each ␣␤ subunit pair followed by release of the bound ATP (2). It is this last step of the overall mechanism of ATP synthesis that is the focus of this study. Specifically, we wanted to know to what extent active site remodeling occurs when the substrate-bound state ("ground state") of F 1 moves into a more transition-like state (i.e. that state where ADP and P i are dehydrated in the presence of Mg 2ϩ and ATP is formed).
Both of our earlier biochemical studies (15,16) designed to obtain information about the transition state formed during the mitochondrial ATP synthase-catalyzed reaction and the crystallographic study reported here were patterned after the landmark biochemical (17) and crystallographic (18) studies on the ATPase myosin. Significantly, the studies with myosin employed vanadate (V i ) 5 (VO 4 3Ϫ ), a phosphate analog that in the presence of ADP and Mg 2ϩ resulted in an ADP⅐V i ⅐Mg⅐myosin transition-like state where V i substituted for the ␥-phosphate of ATP. Since V i is photoreactive, it was possible in the presence of UV light plus O 2 to cleave the myosin backbone (17), an oxidative event that occurred at the conserved serine in the third position within the P-loop (GESGAGKT). This finding was subsequently found to be consistent with a crystal structure of the ADP⅐V i ⅐Mg⅐myosin complex (18). Thus, studies on myosin provided strong support for the view that in the transition state, the ␥-phosphate of ATP and the third residue (serine) within the P-loop consensus region (GESGAGKT) reside close to one another. Although these same investigators used phosphate analogs other than V i (e.g. AlF 4 Ϫ to give an ADP⅐AlF 4 ⅐Mg⅐myosin complex, the crystal structure of which was solved also at atomic resolution (19), they concluded that the ADP⅐ V i ⅐Mg⅐myosin structure resembled most closely that expected for a true transition state (18).
With the above in mind, our earlier biochemical studies (15,16) with the catalytic F 1 moiety of the ATP synthase demonstrated first that in the presence of ADP, Mg 2ϩ , and V i , the ATPase activity of the F 1 catalytic unit is markedly inhibited as expected if an ADP⅐V i ⅐Mg⅐F 1 transition-like state is formed (15) and, second, that upon exposure to UV light ϩ O 2 , cleavage of the ␤ subunit occurs also at the third position (Ala 158 ) of the P-loop (GGAGVGKT). In control studies, prior incubation of F 1 with V i alone, ADP ϩ V i , and ADP ϩ Mg 2ϩ had little or no effect on ATPase activity. Taken together, these findings suggested, consistent with experiments that had been observed earlier for myosin (17,18), that an ADP⅐V i ⅐Mg⅐F 1 transition-like state had formed in which the ␥-phosphate of ATP and the third residue of the P-loop, in this case conserved Ala 158 , reside close to one another (Fig. 1B).
Here, we report the first crystal structure of an ATP synthase F 1 moiety (catalytic unit) crystallized in the presence of ATP, Mg 2ϩ , and V i and discuss its apparent validation of the conclusions made from our earlier biochemical studies (15,16). In addition, we discuss the possible direct relevance of this new structure to the reaction pathway for ATP synthesis in mitochondria and compare it with crystal structures (20,21) obtained for bovine heart F 1 in complex formation with aluminum fluoride, a potential transition state analog (22).

Materials
Rats (Harlan Sprague-Dawley CD , white males, retired breeders) were obtained from Charles River Breeding Laboratories and cared for and used FIGURE 1. A, illustration depicting the mammalian mitochondrial ATP synthase and the different levels of mechanism involved in making ATP from ADP, P i , and Mg 2ϩ . Excluding regulators, the ATP synthase from mammalian cells consists of 15 subunit types, ␣, ␤, ␥, ␦, ⑀, a, b, c, OSCP, F 6 , d, e, f, g, and A6L. The first five (underlined) in the stoichiometry ␣ 3 ␤ 3 ␥␦⑀ (5) form the F 1 headpiece, where the synthesis (or hydrolysis) of ATP takes place predominantly on ␤ subunits. The a, b, c, OSCP, and F 6 subunits comprise the proton motor/stator machinery essential for driving ATP synthesis on F 1 , whereas the functions of the remaining five subunits are unclear. The figure shows only those subunits that comprise the F 1 motor (catalytic headpiece), F 0 motor (proton channel), and the stator. The different levels of mechanism involved in ATP synthesis are noted to the right. B, simplified version of steps involved in the reaction pathway of ATP synthesis as deduced from previous biochemical work with V i (15,16). This previous work implicated remodeling of the active site of F 1 in proceeding from the substrate-bound ground state to the transition state with the methyl group of P-loop Ala 158 and the transition state complex becoming close neighbors. experimentally according to guidelines approved by the Animal Care and Use Committee of the Johns Hopkins University School of Medicine. ATP, ADP, MgCl 2 , ammonium sulfate, and sodium orthovanadate were obtained from Sigma. Crystallization plates were from Hampton Research. All other reagents were of the highest purity commercially available.

Methods
Preparation and Crystallization of F 1 -Rat liver F 1 was purified by the method of Catterall and Pedersen (23) with the modification described by Pedersen et al. (24). It was then lyophilized to dryness in P i and stored at Ϫ20°C until use. For crystallization, the lyophilized enzyme was dissolved at 25°C in 100 l of water and precipitated twice in 3 M ammonium sulfate, 5 mM EDTA. It was then dissolved in 5 mM ATP, 5 mM MgCl 2 , 5 mM V i , 25 mM sodium azide, and 50 mM MOPS, pH 8.8, at 23°C and crystallized by the sitting drop vapor diffusion method. The V i stock solution was carefully prepared as described previously (15), a very important series of steps, and the concentration was determined by measuring the optical density at 265 nm using the extinction coefficient 2925 M Ϫ1 cm Ϫ1 . Single crystals (prepared by YHK) were selected for analysis, equilibrated in cryoprotectant (10% glycerol), cooled in liquid nitrogen, and then transferred into a gaseous nitrogen stream at 100 K for x-ray data collection. After screening numerous crystals ranging in size from 0.2 to 1 mm at the National Institutes of Health facility (Rockville, MD), it was concluded that a relatively small crystal size (0.2-0.3 mm) and 10% glycerol are critical factors in obtaining the best diffraction data.
Crystal Properties and Structure Determination-Data on selected crystals (0.2-0.3 mm) were collected in 1°oscillations at 100 K on beamline 19ID of the Structural Biology Center (Argonne National Laboratory). A single data set composed of 120 frames was used for the study reported here. Data were corrected for image distortion and nonuniformity by D*TREK (25), processed by MOSFLM (26) or HKL2000 (27), and scaled by SCALA (28). These new crystals, designated here as "ADP⅐V i ⅐Mg⅐F 1 crystals," belong to space group R32 as originally reported by Amzel and Pedersen (29) for rat liver F 1 crystallized in the presence of ATP and P i and designated here as "ADP⅐P i ⅐F 1 crystals," and in the Protein Data Bank as 1MAB. In an earlier study (10), the ADP⅐P i ⅐F 1 crystals resulted in a 2.8-Å structure of F 1 containing ADP and P i in catalytic sites located predominantly on ␤ subunits and MgATP in noncatalytic sites located predominantly on ␣ subunits. The MgATP is endogenously and tightly bound to the noncatalytic sites on ␣ subunits of rat liver F 1 as isolated and is not derived from the crystallization medium.
Structure factor amplitudes were produced with TRUNCATE (28). A random set of the reflections containing 5% of the data were chosen as a test set that were excluded in refinement and used to calculate R free . The structure was solved by molecular replacement with the program MOL-REP (28) with 1MAB (see above) as the search model. An initial model was obtained with R work /R free ϭ 0.357/0.356 in the 20 -3.0 Å range using all of the data. The model was further improved by rounds of refinement with the program CNS (30) and rebuilding with xfit, a viewer in the Xtalview package (31). R free was used throughout the process as a parameter for quality control. The final R work /R free is 0.309/0.321 (20 -3.0 Å, no cut-off) with no water molecules built into the model. The positions of vanadate (V i ) and Mg 2ϩ were located at sites of 3 density at the active site in the A weighted F o Ϫ F c . omit map, where V i and Mg 2ϩ were omitted in the phase calculation. The trigonal-bipyramidal model of V i was obtained from the ADP⅐V i ⅐Mg 2ϩ ⅐myosin structure (18) and manually docked into the 3 density distal to ADP. The Mg 2ϩ was modeled into the 3 density proximal to ADP. Electron density maps were calculated using the program CNS (30).
Figures-Figures were prepared and rendered with Pymol (available on the World Wide Web at www.pymol.org), Xtalview, and Raster3D (32). Structural alignments for comparison were made using LSQKAB (27) or the LSQ fit function implemented in xfit. Schematic interactions were made using LIGPLOT (33).

RESULTS AND DISCUSSION
Rationale for Crystallization Conditions-P i added to the medium to obtain the earlier ADP⅐P i ⅐F 1 crystals (1MAB) of rat liver (10) was omitted from the crystallization medium designed to obtain ADP⅐V i ⅐Mg⅐F 1 crystals in order to minimize competition between P i and V i for the active site of F 1 . A final crystallization mixture consisting of ATP, MgCl 2 , V i , and azide was selected (see "Experimental Procedures"). Azide, an inhibitor of ATP hydrolysis catalyzed by F 1 , was included, since it is known that this inhibitor does not immediately prevent the cleavage of the ␥-phosphate of ATP (34). Rather, azide inhibits after some ATP hydrolysis has occurred, implicating its stabilization of intermediate states (34). Therefore, it was rationalized that subsequent to F 1 -catalyzed hydrolysis of ATP in the presence of V i , the transition-like state may be stabilized. In fact, ADP⅐V i ⅐Mg⅐F 1 -containing crystals were obtained.
Description of ADP⅐V i ⅐Mg⅐F 1 Crystals- Table 1 provides a summary of the x-ray data collected and refinement statistics. Like the earlier ADP⅐P i ⅐F 1 crystals (10), the ADP⅐V i ⅐Mg⅐F 1 crystals ( Fig. 2A, inset) obtained in the presence of ATP, MgCl 2 , and V i belong to space group R32, indicating that the three ␣/␤ subunit pairs of the functional unit have 3-fold crystallographic symmetry and that the single ␥ subunit in the crystal lattice resides in one of three possible orientations. Therefore, the occupancy of the ␥ subunit is 0.33. Space groups other than R32 were considered and excluded. These included three other possible C2 space groups, which are subgroups of R32, and space group P1. There is no difference in the R sym among these space groups even to the highest resolution shell. Finally, twinning was excluded based on the ratio ͗I 2 ͘/ ͗I 2 ͘ calculated by the program CNS (30). The R free was carefully checked throughout the refinement, which included many trials of simulated annealing and energy minimization that were performed until no further improvement could be achieved. R free is 5% of the reflections randomly chosen to be excluded from the refinement.
Description of the Overall ADP⅐V i ⅐Mg⅐F 1 Structure Relative to the ADP⅐P i ⅐F 1 Structure (1MAB)-The ADP⅐V i ⅐Mg⅐F 1 structure (Fig. 2) was solved by molecular replacement with the program MOLREP (28) with the ADP⅐P i ⅐F 1 structure (1MAB) as the search model (see "Experimental Procedures"). The coordinates of the ADP⅐V i ⅐Mg⅐F 1 structure described here (red) were aligned with those (blue) of the original ADP⅐P i ⅐F 1 structure (10). Fig. 2A depicts the alignment of the ␣␤␥ parts of each structure, and Fig. 2B shows the alignment of the ␣ 3 ␤ 3 ␥ catalytic unit. The alignment was based on amino acid residues 24 -91 of the ␣ subunit that have been reported to undergo little conformational change during catalysis (21). Overall, the two structures almost overlap as the r.m.s. difference of the peptide atoms is only 0.52 Å. Thus, the presence of V i ⅐Mg at the active site of rat liver F 1 induces little global conformational movement. (Although present, the small ␦ and ⑀ subunits are not resolved in either the 2.8 Å ADP⅐P i ⅐F 1 structure (10) or in the ADP⅐V i ⅐Mg⅐F 1 structure reported here but have been visualized in one F 1 structure (35) in 1:1 complex formation associated with the bottom of the ␥ subunit.) The ␣␤␥ part of the ADP⅐V i ⅐Mg⅐F 1 structure ( Fig. 2A) contains one ␣ subunit (residues 23-502), one ␤ subunit (residues 1-399 and 406 -477), one ␥ subunit (residues 1-45, 74 -87, and 206 -270, occupancy 0.33), one ATP, one ADP, two Mg 2ϩ , and one V i . The ATP is complexed with Mg 2ϩ to give ATPMg (yellow) in the ␣ subunit site, and the ADP is complexed with both Mg 2ϩ and V i to give ADP⅐V i ⅐Mg (green) in the ␤ subunit "catalytic" site (Fig. 2, A and B). Within the ␤ subunit, the V i ⅐Mg part of the ADP⅐V i ⅐Mg was located as described under "Experimental Procedures" and shown to interact with the P-loop (Fig. 2C).
The Active Site- Fig. 3A shows the atoms around the active site of the ADP⅐V i ⅐Mg⅐F 1 structure relative to the P-loop, 156 GGAGVGKT 163 . The Mg 2ϩ (dark blue) is coordinated by P-loop Thr 163 that plays the same role in myosin (18), and the vanadate (red) lies in a charged pocket surrounded by Lys 162 , Glu 188 , the catalytic base, and also Arg 189 and Arg 260 that are not shown. The stick model depicted without nucleotide FIGURE 2. A, ribbon diagram of the asymmetric unit of the ADP⅐V i ⅐Mg⅐F 1 transition statelike structure obtained from crystals (inset) prepared from rat liver F 1 (see "Experimental Procedures"). This structure is aligned with the asymmetric unit of the ADP⅐P i ⅐F 1 structure (Protein Data Bank code 1MAB) solved previously (10). The alignment is based on residues 24 -91 of the ␣ subunit, which undergo little conformational change during catalysis (21). The ADP⅐V i ⅐Mg⅐F 1 structure is colored red, the ADP⅐P i ⅐F 1 structure is blue, and the bound ATPMg and ADP⅐V i ⅐Mg, represented as ball models, are colored yellow and green, respectively. B, ribbon diagram of an ␣ 3 ␤ 3 ␥ unit of the ADP⅐V i ⅐Mg⅐F 1 structure (red; this study) aligned with the ␣ 3 ␤ 3 ␥ unit of the ADP⅐P i ⅐F 1 structure (blue). Both side and top views are shown with ATPMg and ADP⅐V i ⅐Mg, represented as yellow and green ball models, respectively. The basis for the alignment is as above. The two structures almost overlap, since the r.m.s. difference of all peptide atoms is only 0.52 Å, indicating that V i ⅐Mg at the active site induces little global conformational change in F 1 . C, location of V i and Mg 2ϩ near the P-loop ( 156 GGAGVGKT 163 ) of a ␤ subunit in the ADP⅐V i ⅐Mg⅐F 1 structure. The positions of V i and Mg 2ϩ were located by two pieces of 3 density at the active site in the A weighted F o Ϫ F c . omit map, where V i and Mg 2ϩ were omitted in the phase calculation. The trigonal-bipyramidal model of V i was obtained from the ADP⅐V i ⅐Mg⅐myosin structure (18) and manually docked into the density distal to ADP. The Mg 2ϩ was modeled into the density proximal to ADP. The ADPMg is dark blue, and the V i is red. The carbon atoms of the side chain of Glu 188 , the putative general acid, and the carbon atoms of Arg 337 are yellow. B, a schematic planar representation of the hydrogen bonds that are important for stabilizing the V i ⅐Mg-bound conformation of the P-loop. The carbonyl oxygens of Gly 157 and Ala 158 interact with the side chain of Arg 337 , and the carbonyl oxygen of Gly 156 interacts with imino nitrogen of Val 312 . All residues are from the ␤ subunit. Peptide carbon atoms are gray, nitrogen atoms are blue, and oxygen atoms are red. The distance of 3.8 Å from the ␤ carbon (gray) of alanine 158 to the nearest oxygen atom of vanadate (red) in this structure is significantly less than the distance of 5.2 Å from the same ␤ carbon to the nearest oxygen atom of phosphate in the ADP⅐P i ⅐F 1 structure (10). Units of distance are in angstroms.
in Fig. 3B shows that the P-loop of the ADP⅐V i ⅐Mg⅐F 1 structure is stabilized also by hydrogen bonds between Arg 337 and the carbonyl oxygens of Gly 157 and Ala 158 and by a hydrogen bond between the carbonyl oxygen of Gly 156 and the imino nitrogen of Val 312 . Finally, both A and B of Fig. 3 show that within the active site of the ADP⅐V i ⅐Mg⅐F 1 structure, the distance between the ␤ carbon of P-loop Ala 158 and the nearest oxygen atom of vanadate (red) is 3.8 Å, a value significantly less than the distance of 5.2 Å between the ␤ carbon of Ala 158 in the ADP⅐P i ⅐F 1 structure and the nearest oxygen atom of phosphate. This finding indicated that, whereas V i ⅐Mg had induced little global change in the F 1 structure as noted above, significant changes within the active site had occurred.
Structural Differences between the Active Sites of ADP⅐V i ⅐Mg⅐F 1 , and ADP⅐P i ⅐F 1 -As indicated above, a distinct difference in the active sites of the ADP⅐V i ⅐Mg⅐F 1 structure and the ADP⅐P i ⅐F 1 structure (1MAB) was observed as it relates to the distance from the ␤ carbon of Ala 158 to the V i and P i oxygen atoms, respectively. To determine whether structural differences in ␤ subunits were confined to the active site, the ␤ subunits of the two structures were aligned, the average distance between all amino acid atoms within residues 1-399 and 406 -477 in each structure was determined, and the differences were calculated. A plot was then made of the differences versus residue number. The results presented in Fig. 4A show clearly that a distinct difference exists between the locations of ␤ subunit atoms in the two structures within that part of the active site that includes the P-loop. Thus, the overall average difference in distance among all ␤ subunit atoms is only 0.36 Å, whereas the average difference among atoms in the P-loop region is 1.0 Å. Some positional differences also occur between residues 406 and 440 near the C terminus that are unlikely to be involved with function. These differences result from our use of different side chain rotamers from those used in 1MAB to fit the electron density encountered in this region.
Consistent with a change in the P-loop regions induced by V i ⅐Mg as inferred by our earlier biochemical studies (15,16), Fig. 4B shows an overlay of the P-loop region of the ␤ subunit of the ADP⅐V i ⅐Mg⅐F 1 structure (black) with that of the earlier ADP⅐P i ⅐F 1 structure (gray). From this overlay, it is clear that a significant difference exists between the position of Ala 158 in the P-loop in the two structures and also a significant difference in the position of V i and P i . Thus, the ␣-carbon of Ala 158 in the ADP⅐V i ⅐Mg⅐F 1 structure (black) has moved 1.1 Å in the N-terminal direction of the main chain, away from the ␣-carbon of Ala 158 in the ADP⅐P i ⅐F 1 structure (gray). In addition, the V i (red) is much closer to the P-loop in the ADP⅐V i ⅐Mg⅐F 1 structure than is the P i (green) in the ADP⅐P i ⅐F 1 structure. These conformational changes bring Ala 158 in the ADP⅐V i ⅐Mg⅐F 1 structure much closer to the V i such that this phosphate analog resides in a more hydrophobic environment than does the phosphate in the ADP⅐P i ⅐F 1 structure. Thus, the distance from the C␤ atom of Ala 158 (black) to the vanadium atom of V i (red) in the ADP⅐V i ⅐Mg⅐F 1 structure is 4.7 Å, compared with 6.2 Å between the C␤ atom of Ala 158 (gray) and the phosphorus atom of P i (green) in the ADP⅐P i ⅐F 1 structure. Table 2 provides additional information about distances between V i and P i atoms to nearby amino acid residues in the two structures.
Absence of Structural Differences in the P-loop Regions of Nucleotide Binding Sites Located within the ␣-Subunits of the ADP⅐V i ⅐Mg⅐F 1 and ADP⅐P i ⅐F 1 Structures-Although the ␣ subunit of the F 1 unit of the ATP synthase is noncatalytic, it also has a P-loop region ( 169 GDRQTGKT 176 ) that helps bind MgATP. Relative to the P-loop in the catalytic ␤ subunit, the P-loop located in an ␣ subunit nucleotide binding site would be predicted to remain stationary in the ADP⅐V i ⅐Mg⅐F 1 structure relative to the ADP⅐P i ⅐F 1 structure. To determine whether this is the case, the ␣ subunits of the two structures were aligned, the average distance between all amino acid atoms was determined, and the differences were plotted versus residue number. The results presented in Fig. 5A show that no obvious differences between the two ␣-subunit structures occur FIGURE 4. A, plot of the difference in distance between ␤-subunit atoms in the ADP⅐V i ⅐Mg⅐F 1 transition state-like structure reported here aligned with the corresponding ␤-subunit atoms of the ADP⅐P i ⅐F 1 structure (10). The two ␤-subunit structures were aligned and the average distance between corresponding amino acid atoms at each position throughout the sequences (residues 1-399 and 406 -477) were calculated and plotted against residue number. The average distance that includes all difference calculations between all corresponding atoms in the two structures is only 0.36 Å. In contrast, difference calculations between corresponding atoms in the two structures that include the P-loop ( 156 GGAGVGKT 163 ) gave an average value of 1.0 Å (red line). B, overlay of a stick representation of the P-loop region of the ␤ subunit of the ADP⅐V i ⅐Mg⅐F 1 transition statelike structure reported here with that of the ␤ subunit of the ADP⅐P i ⅐F 1 structure (10). The conformational differences in the P-loops of the two structures are clearly delineated as are the relative positions of the ␤-carbon atom of Ala 158 . In addition, the overlay shows that V i (red) is much nearer the P-loop in the ADP⅐V i ⅐Mg⅐F 1 structure than is P i (green) in the ADP⅐P i ⅐F 1 structure. in the P-loop region. This is shown more clearly in Fig. 5B, which depicts an overlay of the P-loop regions in the two cases. Ϫ /AlF 3 , and Mg 2ϩ induce changes in the P-loop/active site region within these bovine F 1 structures relative to that induced by ADP, V i , and Mg 2ϩ in the rat liver F 1 structure. With regard to the first reported structure, (ADP⅐AlF 3 )F 1 , that contained ADP, AlF 3 , and Mg 2ϩ in a sin-gle ␤-subunit (20), the authors state that there is little change in the structure of this subunit relative to that of the same ␤-subunit (␤ DP ) in the original bovine F 1 structure that contained ADPMg (9) (Protein Data Bank code 1BMF). With regard to the second reported bovine F 1 structure, (ADP⅐AlF 4 Ϫ ) 2 ⅐F 1 , in which two ␤-subunits are filled with ADP, AlF 4 Ϫ , and Mg 2ϩ , no obvious change is observed (Fig. 6A) in the P-loops relative to the ␤-subunit (␤ DP ) of the original bovine F 1 structure that contained only ADPMg (9). Therefore, both aluminum fluoride-containing bovine F 1 structures appear to be very near "ground state," a conclusion reached earlier by Allison et al. (37) for the (ADP⅐AlF 3 )F 1 structure. This is in sharp contrast to that of the rat liver ADP⅐V i ⅐Mg⅐F 1 structure (Fig. 6B) reported here in which the P-loop region has undergone a significant conformational change, moving the ␤-carbon atom of Ala 158 and the vanadium atom nearly 1.5 Å closer than the ␤-carbon atom of Ala 158 to phosphate atom distance in the original ADP⅐P i ⅐F 1 structure (10).

Comparison of Changes Observed in the P-loop Active Site Region
Conclusions-For the past 3 decades, a major objective related to work on the mitochondrial ATP synthase (F 0 F 1 ) has been to understand the mechanism by which ATPMg is made from ADP, P i , and Mg 2ϩ , a process that takes place primarily on ␤ subunits. The study reported here, in which the F 1 catalytic moiety has been crystallized for the first time in the presence of V i , a known transition state analog, has resulted in a structure (ADP⅐V i ⅐Mg⅐F 1 ) that meets two expectations of a transition state-like structure. One expectation is that one or more residues at FIGURE 5. A, plot of the differences in distance between ␣-subunit atoms in the ADP⅐V i ⅐Mg⅐F 1 transition state-like structure reported here aligned with ␣-subunit atoms in the ADP⅐P i ⅐F 1 structure (10). The two ␣-subunit structures (residues 23-502) were aligned, and the average distances between corresponding amino acid atoms at each position throughout the sequences were calculated and plotted against residue number. The plot shows that there is no major difference in the distance between corresponding atoms in the two ␣-subunit sequences including the ␣-subunit P-loop region ( 169 GDRQTGKT 176 ). Specifically, the overall average distance difference between all corresponding amino acid atoms in the two ␣-subunit structures is 0.32 Å, a value essentially the same as the difference of 0.33 Å found between corresponding amino acid atoms in the P-loop region. B, overlay of a stick representation of the P-loop region of the ␣-subunit of the ADP⅐V i ⅐Mg⅐F 1 transition state-like structure (black) with that of the ␣-subunit of the ADP⅐P i ⅐F 1 structure (gray). Both structures contain tightly bound ATPMg depicted as light blue in the ADP⅐V i ⅐Mg⅐F 1 structure and dark blue in the ADP⅐P i ⅐F 1 structure. Consistent with the calculations above and the plot in A, the binding of ADP⅐V i ⅐Mg to a ␤ subunit catalytic site to induce a transition state-like structure is essentially without effect on the noncatalytic ␣ subunit sites containing tightly bound ATPMg.  A and B). The overlays show that a significant conformational change is induced in the P-loop region in the presence of ADP, V i , and Mg 2ϩ in work reported here on rat liver F 1 relative to the earlier structural study in the presence of ADP and P i and without V i and Mg 2ϩ (10) but not in the P-loop region of bovine heart F 1 in the presence of ADP, AlF 4 Ϫ , and Mg 2ϩ (21) relative to the earlier structural study (9) with ADP and Mg 2ϩ (␤ DP subunit) and without AlF 4 Ϫ .
the active site that surround the reacting species, in this case ADP and P i (represented here by V i ), will undergo a positional change. This expectation is met, since the third amino acid within the P-loop ( 156 GGAGVGKT 163 ) (i.e. Ala 158 (␤-carbon atom)) and the nearest oxygen atom of V i move much closer (1.4 Å), consistent with interpretations of our earlier biochemical studies (15,16). In fact, other amino acids within the P-loop region also undergo positional changes as deduced from Table 2 and as illustrated in Fig. 6B. A second expectation of a transition state-like structure is that the reacting substrates, in this case ADP and P i in the presence of Mg 2ϩ , will be closer together in the transition state than in the ground state. The distance of 3.9 Å between the ␤-phosphate (phosphorus atom) of ADP and the vanadium of V i in the ␤ subunit in the ADP⅐V i ⅐Mg⅐F 1 structure described here is less than in the closest representative ground state structure, the ␤ E subunit of the (ADP⅐AlF 4 Ϫ ) 2 F 1 structure (21). This ␤ subunit structure, containing bound ADP, Mg 2ϩ , and the phosphate analog SO 4 2Ϫ , has a distance between the ␤-phosphate of ADP and the sulfur atom of SO 4 2Ϫ of 5.4 Å. Significantly, the 3.9-Å distance between ␤-phosphate of ADP and the vanadium atom of V i in our ADP⅐V i ⅐Mg⅐F 1 structure lies between the ground state distance (5.4 Å) and the distance of 2.9 Å between the ␤and ␥-phosphates of the ATP product analog AMP-PNP (Mg 2ϩ present) in the bovine heart F 1 structure (9). Therefore, the ADP⅐V i ⅐Mg⅐F 1 structure reported here satisfies the two expectations of a transition state-like structure noted above.
Finally, Fig. 7 emphasizes the potentially important roles of Mg 2ϩ and the methyl group of Ala 158 in the final events of ATP synthesis in mitochondria catalyzed by the F 1 moiety of the ATP synthase (F 0 F 1 ). Mg 2ϩ is suggested to help bring the ␤-phosphate of ADP and P i in the ground state (Fig. 7, top) closer together in the transition state (Fig. 7, center), whereas the methyl group of Ala 158 that has moved into the active site helps lower the dielectric constant and facilitate release of water (Fig. 7, center) and ATP formation (Fig. 7, bottom). These suggestions are consistent with our earlier biochemical studies with vanadate (15,16) and with a study (38) demonstrating that low dielectric media (i.e. organic solvents) facilitate F 1 -catalyzed ATP synthesis.