Structural evidence of a new catalytic intermediate in the pathway of ATP hydrolysis by F₁–ATPase from bovine heart mitochondria

Abstract

The molecular description of the mechanism of F₁–ATPase is based mainly on high-resolution structures of the enzyme from mitochondria, coupled with direct observations of rotation in bacterial enzymes. During hydrolysis of ATP, the rotor turns counterclockwise (as viewed from the membrane domain of the intact enzyme) in 120° steps. Because the rotor is asymmetric, at any moment the three catalytic sites are at different points in the catalytic cycle. In a “ground-state” structure of the bovine enzyme, one site (β_E) is devoid of nucleotide and represents a state that has released the products of ATP hydrolysis. A second site (β_TP) has bound the substrate, magnesium. ATP, in a precatalytic state, and in the third site (β_DP), the substrate is about to undergo hydrolysis. Three successive 120° turns of the rotor interconvert the sites through these three states, hydrolyzing three ATP molecules, releasing the products and leaving the enzyme with two bound nucleotides. A transition-state analog structure, F₁–TS, displays intermediate states between those observed in the ground state. For example, in the β_DP-site of F₁–TS, the terminal phosphate of an ATP molecule is undergoing in-line nucleophilic attack by a water molecule. As described here, we have captured another intermediate in the catalytic cycle, which helps to define the order of substrate release. In this structure, the β_E-site is occupied by the product ADP, but without a magnesium ion or phosphate, providing evidence that the nucleotide is the last of the products of ATP hydrolysis to be released.

High-resolution structures of F₁–ATPase from bovine heart and yeast mitochondria have provided the molecular basis for the catalytic mechanism of ATP synthesis and hydrolysis by F₁F_o–ATPase (1–10). During ATP synthesis, a mechanical rotation of the γ-subunit driven by the transmembrane proton-motive force, exerted on a ring of C-subunits in the F_o membrane domain of the intact enzyme, takes each of the three β-subunits in its catalytic domain through the states represented by the β_TP-, β_DP-, and β_E-subunits, thereby synthesizing three ATP molecules during each 360° rotation (1, 11). Hydrolysis of ATP reverses the direction of rotation, and protons are ejected from the mitochondrial matrix into the intermembrane space through the F_o domain of the enzyme (12, 13). The rotation of the γ-subunit is not continuous, but it proceeds in 120° steps consisting of 90° and 30° substeps (14). The ground-state structure of F₁–ATPase (denoted F₁–GS) probably represents the state of the enzyme at the end of the complete 120° rotary step.

Attempts have been made to access structures that represent conformations of β-subunits that are intermediate between the three conformations in the ground-state structure by inhibiting the enzyme in various ways. However, many of the inhibitors arrest the catalytic cycle in the ground state (15–17), except that the natural inhibitor protein IF₁ arrests bovine F₁–ATPase in either a posthydrolysis preproduct-release state, or, in a “dead-end” state of the enzyme (18). Another structure of bovine F₁–ATPase inhibited by magnesium.ADP and AlF₄⁻, designated F₁–TS, represents a “transition state” intermediate in ATP hydrolysis (5), with magnesium.ADP–fluoroaluminate bound to both β_DP- and β_TP-subunits, and magnesium.ADP and sulfate (or phosphate) to the β_E-subunit. In this structure, the β_E-subunit had adopted a “half-closed” conformation with bound Mg.ADP that represents a posthydrolysis, preproduct release-conformation in the catalytic cycle. It is more difficult to relate this transition-state structure to the rotary cycle of the enzyme, but it might correspond to the state of the enzyme after a 90° rotary substep.

As described here, we have captured the structure of a catalytic state of the enzyme that appears to represent a posthydrolysis, preproduct-release substep, intermediate between the transition-state and the ground-state structures in the catalytic pathway.

Results and Discussion

Structure Determination.

Crystals of bovine F₁–ATPase were grown in the presence of ADP and magnesium ions, as described previously (10), except that the concentration of free magnesium ions in the mother liquor was diminished by the inclusion of phosphonate, a chelator of divalent metal cations. The structure of the resulting bovine F₁–ATPase complex, known as F₁–PH, was solved by molecular replacement with data to 2.6 Å. The asymmetric unit contains one copy of the complex. Data processing and refinement statistics are summarized in Table S1. In the initial model, each nucleotide-binding site in the catalytic β_DP- and β_TP-subunits and in the three noncatalytic α-subunits contains an ADP molecule and an associated magnesium ion. After one round of refinement, there was also positive difference electron density (>4σ) corresponding to an ADP molecule bound with high occupancy in the nucleotide-binding site in the β_E-subunit (Fig. S1); but there was no evidence for the presence of an associated magnesium ion in this site. Nor was there evidence of a bound phosphonate ion (or phosphate or sulfate) either in its vicinity, or elsewhere in the structure.

The final model of the complex contains the following residues: α_E, 24–510; α_TP, 23–401 and 410–510; α_DP, 24–404 and 407–510; β_E, 9–387 and 396–474; β_TP, 9–474; β_DP, 9–475; γ, 1–47, 67–86, 105–116, 127–148, 159–173, and 206–270; δ, 18–100; and ε, 10–25 and 41–47 (Fig. 1).

Fig. 1.

The structure of the bovine F₁–ATPase complex, known as F₁–PH. (A) The overall view of the complex from the side. The α-, β-, γ-, δ-, and ε-subunits are red, yellow, blue, magenta, and green, respectively. Bound nucleotides in catalytic and noncatalytic subunits are black. With the exception of the ADP molecule bound to the β_E-subunit, each bound nucleotide is accompanied by a magnesium ion. (B) Cross-sectional view from above toward the membrane domain of the intact F₁F_o–ATPase, showing the C-terminal domains of α- and β-subunits (residues 380–510 and 364–474, respectively) arranged alternatively around the γ-subunit (residues 1–47 and 206–270).

Structure of Bovine F₁–PH.

In many respects, the structure of F₁–PH shown in Fig. 1 is similar to previous structures of F₁–ATPase. This similarity is demonstrated by the superimposition of the structure of F₁–PH and the ground-state structure of bovine F₁–ATPase (2JDI) (2). The α_TP-, α_DP-, and β_TP-subunits are essentially identical in the two structures; the rmsd values for the structures are 2.54 and 1.38 Å in comparisons made with and without the central stalk, respectively. The main difference between the two structures is that the β_E-subunit of the ground-state structure has no bound nucleotide; whereas, the nucleotide-binding site of the β_E-subunit in F₁–PH is occupied by an ADP molecule, but that ADP molecule has no accompanying magnesium ion. The structures of F₁–PH and the ground state (Figs. S2A and S3) also differ in other details. For example, in F₁–PH, the C-terminal domains of the neighboring α_E- and β_DP-subunits (residues 380–510 and 364–474, respectively) have adopted a more open conformation than in the ground-state structure (rmsd values 3.35 and 4.69 Å, respectively) (Fig. S3A). Another difference is in the position adopted by the part of the central stalk that protrudes from beneath the α₃β₃-subdomain (Fig. S3B; see below).

Nucleotide Binding to the β_E-Subunit.

In the nucleotide-binding site of the β_E-subunit of F₁–PH, the adenosine moiety of ADP is sandwiched between the side chains of residues βY345 and βF424, which is found in α-helix C3 (residues 418–426), as in the β_DP- and β_TP-subunits of the same structure. However, as shown in Fig. 2, the nucleotide is bound more weakly in the β_E-subunit of F₁–PH where it has lost most of the direct interactions between its phosphate groups and residues in the P-loop of the catalytic site, and also indirect interactions with the protein involving the magnesium ion that are present in the β_DP- and β_TP-subunits (see below). The crystallographic B-factors of the bound ADP molecules support this interpretation. Assuming full occupancy of the nucleotide, the value for ADP in the β_E-subunit is 69.9 Ų, and those for the ADP molecules in the β_DP- and β_TP-subunits are 41.2 and 39.9 Ų, respectively.

Fig. 2.

Interactions between ADP and P-loop regions in the catalytic sites of the β_DP- and β_E-subunits in bovine F₁-PH. The P-loop region is bold. Also shown is the arginine-finger residue αR373. A and B show bonding interactions of ADP plus a magnesium ion in the β_DP-subunits, and of ADP only in the β_E-subunit. Distances are given in Ångströms. In the transition from A to B brought about by counterclockwise rotation of the central stalk (viewed from the membrane domain of the intact F₁F_o–ATPase), the nucleotide is carried to the right (away from the central stalk), disrupting the coordinating water shell around the magnesium ion (Fig. S4). The nucleotide remains bound in the hydrophobic pocket provided by residues βF424 and βY345. Subsequently the ADP will be released, producing the β_E-state observed in ground-state structures of bovine F₁–ATPase. Release of ADP is probably brought about by the movement of α-helix C3 (residues 418–426) in the β-subunit.

As the hydrolysis of ATP proceeds, α-helix C3 hinges around residue βF418 away from the nucleotide, until it reaches the position observed in the F₁–PH structure in which the α-helix is rotated approximately 18° outward, relative to the same α-helix in the β_TP-subunit. In the catalytic sites where nucleotide is bound, residue βF424 is close to the ribose moiety; but in the β_E-subunit of the ground-state structure, where no nucleotide is bound, α-helix C3 has moved to its most distant position from the adenosine-binding pocket, placing βF424 too distant from βY345 to provide a binding site for the adenosine moiety. In the structure of F₁–PH, as observed in previous structures, the positions of βY345 and βF418, and the P-loop remain in a fixed relative geometry, and the position of α-helix C3, and especially of residue βF424, are critical for binding and release of nucleotide. The release of nucleotide bound to the β_E-subunit of F₁–PH requires a further (small) counterclockwise rotation of the central stalk, as viewed from the foot of the central stalk. This rotation could be transmitted to the β_E-subunit via the “catch” region (residues 399–412) of the α_E-subunit, which is in van der Waal’s contact with the N-terminal α-helix in the coiled-coil region of the γ-subunit.

Release of the Magnesium Ion.

In the β_DP- and β_TP-subunits of the ground-state structure of F₁–ATPase (2), magnesium ions in catalytic sites are hexa-coordinated by three ordered water molecules, by the β-hydroxyl of residue βT163 in the P-loop, and by oxygen atoms βO2 and γO2 of the ATP analog AMP–PNP (Fig. S4A). The coordinating water molecules, W₁–W₃, are hydrogen-bonded directly to side chains of the protein and are also supported by an outer shell of water molecules that themselves are hydrogen-bonded to the protein. During catalysis, the conversion of the β_DP–catalytic site to the β_E-site, brought about by rotation of the central stalk, carries the adenosine-binding pocket in the nucleotide-binding domain and its bound nucleotide away from the P-loop and other catalytic-site residues. This outward movement of the region containing the adenosine-binding pocket takes the site sequentially through the states represented in Fig. S4 A–C, where both the nucleotide and the magnesium ion are retained, and finally to the state observed in the β_E-subunit of current structure depicted in Fig. S4D. In this last panel, the water-coordination sphere of the magnesium ion has been disrupted, and the ion is not observed in the structure but has been modeled to show the loss of the coordinating environment. This disruption of the water-coordinating sphere removes most of the enthalpically favorable interactions of the metallic cation with the protein. Therefore, the energetic penalty increases for the entropically unfavorable ordered water shell surrounding the magnesium ion, and so the ion is released (Fig. S4). Modeling of the magnesium ion into the β_E-subunit of the F₁–PH structure depicted in Fig. S4D shows that the only residue within coordinating distance of the water molecules that coordinate the magnesium ion is residue βT163.

A similar mechanism of release of the magnesium ion before the nucleotide has been proposed based on structures of other nucleotide hydrolases that were crystallized in the presence of the magnesium ion chelator, EDTA. In the kinesin superfamily member, protein-1A (19), a “Mg²⁺-water cap” is released first, followed by the magnesium ion. Then, the bound ADP is exchanged by unbound Mg–ATP. In the guanine nucleotide-binding proteins, such as Ras, release of the magnesium ion decreases the affinity of the nucleotide. The magnesium ion is pushed out of its position by elements of the guanine nucleotide-binding protein itself, for example, residue A59 in Ras, or from residues of the guanine nucleotide exchange factor, which accelerate the release of nucleotide by several orders of magnitude (20).

Position of the Central Stalk.

Although the α₃β₃ domains of F₁–PH and the bovine ground-state structure are very similar, there is considerable divergence in the central stalks, especially in regions that extend beyond the α₃β₃ domains. The superimposition of the central stalks alone helps to illustrate this point (Fig. S3B). Significant differences in the structures and positions of the central stalk in various structures of F₁–ATPase have been noted previously (3, 5, 6, 18). Attempts have been made to interpret these positions in terms of the catalytic cycle of the enzyme, but it has been noted that the positions could be influenced by contacts between adjacent complexes in the lattice of the crystals used in the structure determination. Therefore, any interpretation of the position of the central stalk would have to be interpreted cautiously. To investigate systematically the possible influence of crystal lattice contacts on conformations of subunits of F₁–ATPase, we compared all of the available crystal structures with each other, including the structure of F₁–PH, and we have analyzed crystal lattice contacts in each case (see Fig. S5 for crystal lattice contacts in F₁–PH). The main conclusions from this study are that, with the possible exception of yeast F₁–ATPase inhibited with the yeast inhibitor protein, IF₁, the structures of the catalytic sites are not influenced by lattice contacts, and that therefore the observed structures describe the cycle of catalytic events in the active sites of the enzyme. The second conclusion is that in many structures of F₁–ATPase, the “foot” regions of the central stalk are displaced by lattice contacts. A wide range of positions has been observed with rotations in the range −15° to +32°. In F₁–PH, the central stalk is rotated by +32° relative to the ground-state structure, the most extreme displacement yet observed. Its position is determined by crystal lattice contacts with the crown region of adjacent F₁ complexes (Fig. S5). Because a wide range of rotations of the foot of the central stalk is observed in the crystal structures, even when the enzyme is believed to be in the same catalytic state, it is very difficult to relate the crystallographic results to the discrete states observed in the “single molecule” biophysical experiments (21, 22).

Mechanism of ATP Hydrolysis by F₁–ATPase.

The hydrolysis of an ATP molecule bound to the β_DP-subunit proceeds via the in-line nucleophilic attack of the γ-phosphate by a water molecule activated by residue βE188. The arginine-finger residue, αR373, plays a critical role in this step by correctly positioning the γ-phosphate of the nucleotide. The reaction advances via formation of a transition-state intermediate, as observed in the transition analog structure of bovine F₁–ATPase, F₁–TS, followed by scission of the terminal phosphate and release of products. The order of release of products has proved to be difficult to establish in such a complex enzyme as F₁–ATPase, where there are three catalytic sites with different nucleotide affinities at any one time. In the crystal structure of yeast F₁–ATPase, phosphate is found bound to the β_E-subunit in one of the three independent copies of the enzyme in the asymmetric unit, about 8 Å from the position occupied by the γ-phosphate of ATP in other sites. This observation suggests that phosphate is released after ADP and the magnesium ion (6). However, no phosphate is bound in this location in the other two copies of the enzyme in the asymmetric unit. In many ground-state structures of bovine F₁–ATPase, density for phosphate (or its analog sulfate) has been observed in a different position to where it has been observed in the yeast enzyme (18), close to the P-loop region. However, when this electron density is modeled as a phosphate, the temperature factors are very high, and there are only a few interactions with the protein. Furthermore, this site lies between the positions of the α- and β-phosphates of a nucleotide is bound to the P-loop and cannot be occupied simultaneously by both phosphate and nucleotide. It is likely that this site is an anion-binding pocket that is occupied fortuitously in the crystal structures, and that it is not a catalytically relevant phosphate-binding site. In addition, in the β_E-subunit of the current structure (where sulfate had been excluded from the crystallization buffer), there is no evidence for bound phosphate (or its analog phosphonate), which tends to suggest that phosphate is released before the nucleotide. On the available evidence it appears that, in the bovine enzyme at least, both phosphate and the magnesium ion are released before ADP, but the order of their release is not known.

Despite these remaining uncertainties, it is possible to place the conformation of the catalytic site of the β_E-subunit in F₁–PH in a series of structural transitions observed in various structures of bovine F₁–ATPase, which together describe the binding of substrates, the hydrolysis of ATP, and the release of products. In this scheme, the β_E-subunit of F₁–PH occupies a position immediately before the conformation of the catalytic site observed in the β_E-subunit in ground-state structures of bovine F₁–ATPase, and following the conformation of the catalytic site of the β_E-subunit observed in the structure of the transition analog complex of bovine F₁–ATPase. Thus, it is best interpreted as representing a posthydrolysis, pre-nucleotide-release complex in the hydrolysis of ATP. An alternative possible interpretation that it represents a post-nucleotide-binding premagnesium-ion-binding state in ATP synthesis is implausible, as it is unlikely that free nucleotide (lacking an associated magnesium ion) exists in the matrix of mitochondria.

Rotary Catalysis.

During ATP hydrolysis, the substrate, Mg.ATP, enters the unoccupied β_E-subunit, inducing a 120° rotation of the central stalk, converting it to the β_TP-conformation and followed by rearrangement of the active site residues. A hydrophobic pocket is formed in which the adenosine moiety is sandwiched between residues βF424 and βY345. The magnesium ion is hexa-coordinated by residue βT163 and the β- and γ-phosphates of the nucleotide, and three ordered water molecules. Residue αR373 reorients its position from pointing away from the nucleotide-binding site to a new position, where it points toward the γ-phosphate of the nucleotide. An additional 120° rotation of the central stalk forms the β_DP-conformation where the arginine finger is 0.7–1.5 Å closer to the γ-phosphate of ATP, and the nucleophilic water has moved 0.7–0.8 Å closer to the γ-phosphate than in the β_TP-catalytic site, allowing nucleophilic attack to proceed. An intermediate rotation of the central stalk results in the formation of a half-closed β_E-subunit containing Mg.ADP and the released inorganic phosphate. A further rotation of the central stalk completes the 360° rotary cycle, releasing the phosphate first, then the ordered water molecules, and finally the magnesium ion. The sequence in which these product-release events occur can be inferred from thermodynamic considerations, as well as from related mechanisms established in other NTPases. However, further biochemical and structural data will be needed to define this sequence experimentally in F₁–ATPase. A final outward hinging of α-helix C3 leads to the release of the weakly associated ADP, and its replacement with an incoming Mg.ATP molecule (Fig. 3 and Fig. S6).

Fig. 3.

Conformational changes occurring in one of the three catalytic sites of bovine F₁–ATPase. In Upper, the conformations of the β_E-, β_TP- and β_DP-subunits are taken from the ground-state (GS) structure of bovine F₁–ATPase. They are placed in the order described before (4, 18), each conversion requiring a 120° rotary step of the central stalk of the enzyme. In the conversion of β_E–GS to β_TP–GS, the substrate, magnesium.ATP, is bound to the catalytic site. The next 120° rotation converts β_TP–GS to β_DP–GS, the catalytically active site of the enzyme. In the next 90° substep of rotation (probably), the transition state forms. It is depicted in the figure as the transition-state analog of ATP hydrolysis, ADP–AlF₄, in the β_DP-site of F₁–TS. In the next 30° rotational substep (probably), scission of the γ-phosphate of ATP occurs (depicted as β_E–TS in the structure F₁–TS) followed by release of the magnesium ion and phosphate, producing the state β_E–PH found in the structure F₁–PH, regenerating β_E–GS. The magnesium ion, its hydrating water molecules, and bound nucleotide are shown in green, red, and gray, respectively. In β_TP–GS, the bound magnesium ion is obscured by the β- and γ-phosphates of the bound ATP molecule.

Methods

Purification and Crystallization of F₁–ATPase.

Bovine mitochondrial F₁–ATPase was purified as described previously, except that 0.02% azide, and 5 mM β-mercaptoethanol were omitted from buffers, and a Sephacryl S-300 column was replaced by a HiLoad 26/60 Superdex 200-pg column (GE Healthcare Life Sciences) (1). The purified F₁–ATPase was precipitated with ammonium sulfate, redissolved in buffer made in D₂O and consisting of 200 mM Tris-DCl (pH 7.2), 400 mM NaCl, and 1 mM ADP (final concentration 10 mg/mL). An equal volume of buffer made in D₂O and composed of 100 mM Tris-DCl (pH 7.2), 400 mM NaCl, 4 mM MgCl₂, 1 mM ADP, and 14% (wt/vol) PEG 6000 was added. The solution was centrifuged (33,000 g, 10 min) and the supernatant was transferred to microdialysis buttons (50 μL, Cambridge Repetition Engineers) and covered with dialysis membrane (SpectraPor, 3,500 molecular weight cutoff, Spectrum Medical Industries). The button was dialyzed at room temperature against 3 mL of buffer consisting of 100 mM Tris-DCl (pH 8.2), 400 mM NaCl, 4 mM MgCl₂, 1 mM ADP, 5 mM phosphorous acid, and 9% (wt/vol) PEG 6000. After 24 h, the external buffer was replaced with the same buffer containing concentrations of PEG 6000 from 10 to 14% (wt/vol) in 0.25% steps. Crystals formed after 1 wk and were fully grown after 4 wk.

Data Collection and Structure Determination.

Crystals were harvested in a MicroMesh loop (MiTeGen), and their diffraction properties monitored during dehydration in a free mounting system (Proteros Biostructures GmbH) as described previously (23). Then the crystals were vitrified in liquid nitrogen. X-ray diffraction data were collected from the vitrified crystals on beamline ID23eh1 (wavelength 0.9790 Å) with a Q315r charged-coupled detector (Area Detector Systems Corporation) at the European Synchrotron Radiation Facility, Grenoble, France. Diffraction images were integrated with MOSFLM (24) and data were reduced with SCALA (25) and TRUNCATE (26). The structure was solved by molecular replacement with the structure of azide-inhibited F₁–ATPase (PDB ID code: 2CK3) with PHASER (27). It was remodeled manually with COOT (28). Alternate rounds of rebuilding and refinement, were carried out with REFMAC5 (29). Stereochemistry was assessed with COOT and MolProbity (30). Images of 3D structures and electron-density maps were generated with PyMOL (31). The structure was compared with others using the SUPER alignment tool in PyMol.