Rotation of F₁-ATPase and the hinge residues of the β subunit

F₁ F_o -ATP synthase, the enzyme responsible for most of the ATP synthesis in the biological world, couples H⁺ translocation across membranes to synthesis/hydrolysis of ATP. It reversibly dissociates into a membrane part, F_o, a proton channel in itself, and a water-soluble part, F₁, an ATPase that is hence often called F₁ -ATPase. The subunit composition of F₁ is α₃ β₃ γδε; the α₃ β₃ γ complex is the catalytic core that retains the major kinetic features of F₁ -ATPase.

More than 10 years before the structure of F₁ -ATPase was determined, Boyer proposed the binding change mechanism in which three conformations of the β subunit with different affinities for adenine nucleotide were assumed (Boyer and Kohlbrenner, 1981; Boyer, 1993). The input of energy changes the affinity of binding sites by cooperative conformational changes. The molecular model that best explained the binding change mechanism was the physical rotation of the γ subunit relative to a ring consisting of the α₃ β₃ part. The crystal structure of bovine heart mitochondrial F₁ (Abrahams et al., 1994) indeed showed that the β subunits are in three different states, as predicted by the binding change mechanism: the AMP–PNP bound form (β_TP), the ADP bound form (β_DP) and the empty form (β_E). Elaborate experiments by two groups indicated that the γ subunit rotates during catalysis (Duncan et al., 1995; Sabbert et al., 1996). However, other movements such as simple conformational changes could also explain the observations, and exclusive evidence for the rotation was sought. A decisive answer was provided by visualising the rotation through single-molecule imaging.

We have established a novel system to detect rotation of a single F₁ -ATPase molecule (Noji et al., 1997). A fluorescent actin filament was attached to the γ subunit of α₃ β₃ γ complex from thermophilic Bacillus PS3, and the α₃ β₃ part was immobilised on a Ni²⁺ -nitrilotriacetic acid (Ni-NTA)-coated polystyrene bead (Fig. 1). Rotation of the actin filament in an anticlockwise direction was observed under a fluorescent microscope. The rotation was ATP-dependent and azide-sensitive (Noji et al., 1997). An actin filament that was attached to the ε subunit of α₃ β₃ γε rotated in the same direction, and the ε subunit therefore also constitutes part of the rotor apparatus in F₁ -ATPase (Kato-Yamada et al., 1998).

At ATP concentrations below 600 nmol l⁻¹, the γ subunit rotated in steps of 120 °, pausing at three dwell positions (Fig. 2) (Yasuda et al., 1998). Taking into account the pseudo-threefold symmetry of α₃ β₃ γ structure and that the binding of ATP to a catalytic site is the rate-limiting step under this condition, binding of ATP to an empty β subunit seems to initiate the stepping motion, and the work done during a step is fuelled by one ATP molecule. The plot of dwell time versus number of events was consistently exponential and was explained better by a one-ATP-per-step model than by a two-ATPs-per-step model (Yasuda et al., 1998). Thus, consumption of three ATP molecules for one rotation was confirmed. The possibility of smaller substeps is not excluded because of the limit of time resolution of our current system.

Attachment of an actin filament to the γ subunit enabled not only visualisation of the rotation but also estimation of the torque generated by the rotation. Fig. 3 shows the dependence of the rotational speed of actin filaments on the length of the filaments. The mean torque was approximately 40 pN nm (4×10⁻²⁰ J) irrespective of the load or the actin length, under conditions in which the free energy of hydrolysis of one ATP molecule (ΔG) was 90 or 110 pN nm. The work done by a one-third rotation was approximately 84 pN nm (40 pN nm × 2π/3), indicating that the thermodynamic efficiency of conversion of the chemical energy of ATP hydrolysis to the kinetic energy of rotation is nearly 100 %. This value contrasts with the reported values of efficiency of myosin/actin (20 %) (Ishijima et al., 1995) and kinesin/microtubule (50 %) (Svoboda et al., 1993). It is interesting to consider how the γ subunit would behave if ΔG were :::; 80 pN nm. One possibility is that the rotation would slow down, keeping approximately 100 % efficiency. Another possibility is that the rotational speed would not slow down, but that the enzyme would frequently fail in rotation even though ATP was hydrolysed, resulting in the overall consumption of two or more ATP molecules per step. To test which is the case is a challenge.

Once rotation of the γ subunit had been established, the next problem was the mechanism of rotation. It is most likely that shifting the C-terminal domain of the β subunit towards the γ subunit triggers rotation of the γ subunit. For this kind of domain motion to occur, there should be a segment corresponding to a hinge connecting the two domains. For example, the crystal structures of myosin indicated that Gly457 and Ile455 (Dictyostelium myosin II) of the ‘switch II’ region constitute a hinge, and mutations of Gly457 caused the loss of both ATPase and motor activity (Fisher et al., 1995a,b; Smith and Rayment, 1996; Sasaki et al., 1998). Gly709 and Gly720 at the each end of the helix SH1 (chicken vertebrate smooth muscle myosin) are also identified as hinge residues. In the case of G-proteins, crystallisation and mutational analyses have revealed that Gly199 of the α subunit of bovine transducin is the hinge residue, and mutation of the residue resulted in the loss of the ability to activate the effector (Lee et al., 1992; Noel et al., 1993; Lambright et al., 1996). Thus, to describe the sequence of conformational changes of the β subunit of F₁ -ATPase, it is important to identify and characterise a hinge.

To identify hinge residues in the β subunit, we compared the main-chain dihedral angles, ψ and Ψ (formed by four main-chain atoms starting from the carbonyl carbon of the previous residue and the amide nitrogen of the residue in question, respectively), of each residue in β_DP and β_E (refer to Introduction) of the crystal structure of bovine mitochondrial F₁ (Abrahams et al., 1994). Only five residues, Val162, Gly163, His179, Gly180 and Gly181 (numbering of Bacillus PS3), change either ψ or Ψ by more than 100 ° when the structure of β_DP is superimposed onto that of β_E. Val162 and Gly163 are residues in the Walker A motif, which is found at nucleotide binding sites in a variety of proteins, and are components of the P-loop (GGAGVGKT). The P-loop changes its conformation in response to nucleotide binding and hydrolysis, but this change is not a hinge-like motion. In contrast, Gly180 and Gly181 are located in the loop connecting helix B and strand β₄ (Fig. 4) (Abrahams et al., 1994). In β_TP and β_DP, His179 is also a component of the loop, but it is involved in helix B in β_E. Helix B is preceded by the P-loop, and the strand β₄ is part of the ‘wall’ of parallel β sheet that surrounds the P-loop and helix B. The catalytic glutamic acid exists at the end of strand β₄ that is followed by helix C. As a whole, the catalytic site of the β subunit consists mostly the P-loop, helix B, strand β₄, helix C and the wall of β sheet (Figs 4, 5) (Abrahams et al., 1994). When a nucleotide binds to the catalytic site, the N-terminal region of helix B is pulled up (Fig. 5), and this motion probably triggers swinging of the whole C-terminal helical domain. These three residues (βHis179, βGly180 and βGly181) appear to allow the C-terminal domain to swing by changing their dihedral angles drastically. A large movement of His179 was already indicated by the report that the ¹ H nuclear magnetic resonance peak of His179 of the free β subunit shifted downfield with increasing concentration of Mg-AMP-PNP (Tozawa et al., 1995). We replaced these residues with alanine and studied the characteristics of mutant α₃ β₃ γ complexes containing His179Ala, Gly180Ala and Gly181Ala.

A brief scheme of ATP hydrolysis catalysed by F₁ -ATPase is shown in Fig. 6. It has been known that, when the product ADP is held in a stable manner at a catalytic site in the β subunit during ATP hydrolysis, the F₁ molecule becomes inactive. This inactive form is called the Mg-ADP inhibited form (Jault et al., 1995; Matsui et al., 1997). When the ATPase reaction is started by addition of ATP, the α₃ β₃ γ complex gets trapped in the Mg-ADP inhibited form in some probability and accumulates as turnover of the catalysis repeats, and the initial rate of ATP hydrolysis (initial ATPase activity) decreases towards an equilibrium rate (steady-state ATPase activity). The transition from initial activity to steady-state activity can be simulated by exponential fitting; the half-time of the transition for the wild-type α₃ β₃ γ complex was approximately 19 s under the conditions tested. The ratio of steady-state to initial activities was approximately 0.4. Because the precursor of the Mg-ADP inhibited form is thought to be an intermediate of the catalytic cycle, F₁ ADP, these values reflect the relative population of F₁ ADP form during catalytic turnover.

Initial and steady-state ATPase activities of the mutants are shown in Fig. 7. In the case of the α₃ β₃ γ complex containing His179Ala, the initial ATPase activity did not change much, but the ratio of steady-state to initial activities decreased to 0.16. The half-time of the initial-to-steady-state transition was approximately 3.1 s. Thus, the propensity to fall into the Mg-ADP inhibited form increased. Interpretation of these changes is not easy because His179 becomes a part of helix B in the transition from β_DP (or β_TP) to β_E, as described above, and the replacement of His by Ala includes the loss of a hydrogen-bonding site for side chains of other residues. Therefore, His179Ala is not discussed in the next section. For the Gly180Ala mutant, the initial ATPase activity decreased to half that of the wild-type complex, but the decrease in the steady-state activity was more pronounced, to 7 % of the wild-type value. The ratio of steady-state to initial activities was approximately 0.05, and the half-time of the initial-to-steady-state transition was approximately 3.4 s. In contrast, the α₃ β₃ γ complex containing Gly181Ala showed only very little time-dependent transition of activity; the initial ATPase activity was not reduced, and no initial-to-steady-state transition was observed. Therefore, the Gly180Ala mutant demonstrates an enhanced propensity to fall into the Mg-ADP inhibited form, whereas the Gly181A mutant becomes resistant to Mg-ADP inhibition.

These observations regarding Gly180Ala and Gly181Ala can be explained by postulating a change in the stability of the closed form of the mutants (Fig. 8). At a saturating ATP concentration, the transition from open form 2 with an empty catalytic site to open form 1 with a bound ATP is very fast. In the case of Gly180Ala, the rapid initial-to-steady-state transition can be caused by the increased population of F₁ ADP, which results from stabilisation of the closed form. The stabilisation of the closed form leads to an increased activation energy for the transition from the closed form to open form 2, which may result in the decreased initial ATPase activity of this mutant. In contrast, the closed form of the Gly181Ala mutant appears to be less stable, and [F₁ ADP] may be lower than that in the wild-type complex. As a consequence, little or no Mg-ADP inhibited form is produced during catalysis, and the apparent rate of inactivation (k_app) decreases. The rate-limiting step in the catalytic cycle of this mutant may be the transition from open form 1 to the closed form, and the transition energy for this mutant is the same as that for the wild type, so that the initial ATPase activity is unchanged.

The stability changes in the closed form of the mutant complexes can be partly attributed to the steric stability of the mutated residues. The change in steric stability caused by the mutations discussed above was estimated in a Ramachandran map. The map was calculated for Gly-Gly, Ala-Gly and Gly-Ala peptides with various ψ–Ψ combinations (Fig. 9). The ψ–Ψ combination of Gly180 observed in the crystal structure of β_E is plotted in the area close to ‘outer limit’ of the energetically allowed region (Fig. 9A). Mutation of Gly180 to alanine seems to further destabilise this ψ–Ψ combination (Fig. 9B). Consequently, the open form (empty form) becomes less stable in this mutation, and [F₁ ADP] increases. More Mg-ADP inhibited form is then generated. The situation is the reverse for the Gly181Ala mutant; the Ramachandran plot shows that Gly181 in the closed form (ADP form) takes the ψ–Ψ combination that is only allowed for glycine residue and that, when Gly181 is replaced by alanine, this ψ–Ψ combination becomes impossible (Fig. 9C,D). Thus, the Gly181Ala mutation destabilises the closed form (ADP form), the population of the closed form [F₁ ADP] decreases, and the mutant complex becomes resistant to Mg-ADP inhibition (see legend to Fig. 8).

The two single alanine mutations had opposite effects on the stability of the closed form. To investigate these effects in one molecule, the ATPase activity of a double Ala mutant (Gly180Ala/Gly181Ala, termed the HAA mutant) was measured (Fig. 7). This mutation resulted in approximately the same initial ATPase activity as that of the single alanine mutant Gly180Ala, but the steady-state ATPase activity was 65 % of that of the wild type, contrasting with that of the single Gly180Ala mutant (7 %). For this double mutant, there was an additional intermediate phase in the time course of ATPase activity besides the initial and steady-state phases. Therefore, resistance to Mg-ADP inhibition could not be defined by half-time of the initial-to-steady-state transition. However, the steady-state ATPase activity indicated that resistance to Mg-ADP inhibition was regained when the second mutation βGly181Ala was introduced to the βGly180Ala mutant. It is likely that destabilisation and stabilisation of F₁ ADP occurred simultaneously, and the effects of both mutations seemed to be cancelled.

When we replaced all His-Gly-Gly residues with alanines (termed the AAA mutant), both initial and steady-state ATPase activity decreased drastically, to approximately 1 % of those of the wild type (because of limits to the detection of the fast transition from initial state to steady state, the initial ATPase activity may include an error; Yasuda et al., 1998). In addition, the AAA mutant had a much faster apparent rate of inactivation than that of the wild type and even of the Gly180Ala mutant (data not shown). From these results, we can assume that stabilisation of the closed form was considerably enhanced, further impairing the opening–closing motion of the β subunit. Taken together, these results strongly indicate that the His-Gly-Gly loop contributes to the opening–closing motion of the β subunit.

Rotation of these mutants was observed. As described in the previous section, the relative stability of the closed and open forms is changed by mutations at the hinge loop. Because we assume that the rotational work exerted by the hinge mutants corresponds to the potential difference between open form 1 and the closed form (Fig. 8), a change in the potential difference could lead to a change in the rotational torque. However, the γ subunit, for all the single alanine mutants, rotated with the same mean torque as that of the wild-type, irrespective of the actin length (Fig. 10). This is probably because the change in [F₁ ADP] affects the initial-to-steady-state transition rate in a more pronounced manner than it does the free energy of rotation. For example, a 10-fold increase in [F₁ ADP] will result in a 10-fold increase in the initial-to-steady-state transition rate, while it will cause only a 10 % increase in the potential difference (the work). In this case, torque should change by 10 %, a value far smaller than the experimental error range of the present rotary assay system.

Our experimental system using an actin filament has enabled us to perform a direct ‘seeing is believing’ experiment. However, it is still to be improved. The population of actively rotating actin filaments is only a few per cent of the total filaments. This may be due to interference by other neighbouring F₁ molecules, to surface obstruction or to denaturation of F₁ molecules by immobilisation. Therefore, without extensive statistical analysis, it is difficult to distinguish pauses and stops that arise as natural characteristics of this enzyme from those originating from the experimental manipulations.

A simultaneous observation of rotation and ATP hydrolysis by a single molecule has not yet been successfully achieved. Therefore, we are obliged to estimate the rate of ATP consumption during rotation from measurements using unimmobilised F₁ molecules in a solution without load. At low ATP concentrations, because the rate-limiting step of the reaction is ATP binding, hydrolysis of one ATP molecule measured in a bulk solution roughly corresponds to one-third of a rotation of the filaments (Yasuda et al., 1998). However, at high ATP concentrations, the frictional load of filament rotation limits the speed of rotation, and the rate of ATP hydrolysis is probably suppressed accordingly. Measurement of the ATPase activity of a single F₁ molecule or observation of the rotation of the unloaded γ subunit is necessary to relate the timing of rotation directly to that of ATP hydrolysis. Experimental methods to measure both these in the same system are awaited.

Single-molecule experiments using F₁ F_o -ATP synthase must be performed to extend our understanding of its function as a holoenzyme. The ATP synthesis reaction catalysed by F₁ F_o -ATP synthase may not be merely the reverse of ATP hydrolysis by F₁ -ATPase. In the process of ATP synthesis, the source of energy for rotation is not the chemical energy of ATP hydrolysis, but H⁺ flow through F_o. In addition, the generation of the Mg-ADP inhibited form appears to be absent during ATP synthesis (Bald et al., 1998). Therefore, the rotation of the γ subunit during ATP synthesis needs to be confirmed. Moreover, classification of F_o subunits into rotors and stators using the rotary assay will clarify the mechanism of F₁ F_o, as was achieved for F₁.

A recent report that F₁ F_o undergoes contraction (Syroeshkin et al., 1998) is worth considering in exploring other movements required for ATP synthesis, although we have been unable to repeat the ‘sonic wave-driven ATP synthesis’ (N. Mitome, unpublished results). If this movement is essential for ATP synthesis, real-time detection of the up–down motion of a F₁ F_o molecule would be the next challenge.

We acknowledge Dr Tazashi Matsui for constructing the overexpression system of TF₁ α₃ β₃ γ and Dr Takayuki Nishizaka for providing carefully prepared actin filaments.