Research Article

Molecular Architecture of the Rotary Motor in ATP Synthase

See allHide authors and affiliations

Science  26 Nov 1999:
Vol. 286, Issue 5445, pp. 1700-1705
DOI: 10.1126/science.286.5445.1700


Adenosine triphosphate (ATP) synthase contains a rotary motor involved in biological energy conversion. Its membrane-embedded F0 sector has a rotation generator fueled by the proton-motive force, which provides the energy required for the synthesis of ATP by the F1 domain. An electron density map obtained from crystals of a subcomplex of yeast mitochondrial ATP synthase shows a ring of 10 c subunits. Each c subunit forms an α-helical hairpin. The interhelical loops of six to seven of the c subunits are in close contact with the γ and δ subunits of the central stalk. The extensive contact between the c ring and the stalk suggests that they may rotate as an ensemble during catalysis.

ATP is the universal biological energy currency. ATP synthase produces ATP from adenosine diphosphate (ADP) and inorganic phosphate with the use of energy from a transmembrane proton-motive force generated by respiration or photosynthesis [for reviews, see (1–3)]. The enzyme consists of an extramembranous F1 catalytic domain linked by means of a central stalk to an intrinsic membrane domain called F0. The globular F1 domain is an assembly of five different subunits with the stoichiometry α3β3γ1δ1ɛ1. In the atomic structure of bovine F1, the α and β subunits are arranged alternately around a coiled coil of two antiparallel α helices in the γ subunit. The catalytic sites are in the β subunits at the α/β subunit interface (4). The remainder of the γ subunit protrudes from the α3β3 assembly and can be cross linked to the polar loop region of the c subunits in F0(5, 6). In mitochondria, the δ and ɛ subunits are associated with the γ subunit in the central stalk assembly (7–10), as are the bacterial and chloroplast ɛ subunits, the counterparts of mitochondrial δ. ATP-dependent rotation of γ and ɛ within an immobilized α3β3 complex from the thermophilic bacterium Bacillus PS3 has been observed directly (11, 12). The rotation of the γ subunit in ATP synthase is thought to be generated by the passage of protons through F0. Because there is only one intrinsically asymmetric γ subunit, it interacts differently with each of the three catalytic β subunits in F1, endowing them with different nucleotide affinities (4). Rotation of the central stalk is accompanied by conformational changes in the β subunits, which cycle sequentially through structural states corresponding to low, medium, and high nucleotide affinities. These three states are probably associated with release of product (ATP), binding of substrates (ADP and inorganic phosphate), and ATP formation, respectively. The cycle is known as the “binding change mechanism” of ATP synthesis (13). In eubacteria, the procedure is reversible, and hydrolysis of ATP in F1 generates rotation of γ, resulting in the pumping of protons back across the membrane through F0 (2).

In contrast to the detailed structural model established for most of the bovine F1 domain (4,14–16) and additional structural information in other species (17, 18), little is known about the structural details of F0. All species contain three common subunits known as a, b, and c. InEscherichia coli, the experimentally determined ratio of these subunits is a1b2c9-11(19, 20). Cross-linking and genetic experiments (21, 22), as well as evolutionary arguments (23), have been interpreted as showing the presence of 12 c subunits per F0. From biochemical studies and mutational analysis in bacteria (2, 24) and fungi (25), it is known that both a and c subunits contain functional groups that are essential for proton translocation through the membrane. A nuclear magnetic resonance (NMR) structure of a monomer of the E. coli c subunit in a chloroform:methanol:water mixture shows that the protein is folded into two α helices (presumed to be transmembrane in the intact enzyme), linked by a loop (presumed to be extramembranous) (26). The COOH-terminal α helix contains a conserved side chain carboxylate (Asp61 in E. coli and Glu59 in Saccharomyces cerevisiae) essential for proton translocation (2). Models have been proposed in which the a and b subunits are on the outside of a multimeric ring of c subunits within the membrane, with the interface between a and c subunits providing the pathway for proton translocation (Fig. 1) (27–31). Protonation and deprotonation events in this interface have been proposed as part of a mechanism for generating conformational changes in F1 (32), possibly through rotation of the c ring (33). Because the generation of each ATP requires rotation of the γ subunit through 120°, a link has been made between the number of protons that must be translocated per ATP synthesized [estimated to be three or four in different systems (34, 35)] and the number of c subunits in the ring. The required rotation of γ through 120° can be achieved with either 9 c subunits in the ring and 3 translocated protons or 12 c subunits in the ring and 4 translocated protons (36). It has also been proposed that a rotary mechanism would require a peripheral “stator” to counter the tendency of α3β3 to follow the rotation of γ (36). It has been suggested that the b subunits of F0 may form part of such a stator, and evidence from reconstitution and cross-linking studies provides support for this hypothesis (37, 38). A feature connecting the periphery of F0 to F1 (the peripheral stalk) that might represent the stator has been visualized by single-particle electron microscopy of ATP synthase from bacteria, chloroplasts, and mitochondria (39–41).

Figure 1

Hypothetical model for the generation of rotation by proton transport through the F0 domain of ATP synthase [according to Junge (33)]. The central cylinder (blue) consists of c subunits; the external part (green) corresponds to a single a subunit. The red line indicates the proton path. [Reproduced with kind permission of the Swedish Academy of Sciences.]

Structure analysis. In this study, we purified ATP synthase from S. cerevisiae mitochondria and carried out crystallization experiments (42). The enzyme from this source consists of at least 13 different types of subunits: the F1 subunits α, β, γ, δ, and ɛ; the F0subunits a, b, and c, plus f and d; OSCP and ATP8, which have homologs in other mitochondrial ATP synthases; and the unique yeast subunit h (43–46). Other possible yeast subunits have been discussed as being involved in dimer formation (47). According to analyses by SDS–polyacrylamide gel electrophoresis (SDS-PAGE), high-performance liquid chromatography (HPLC) analysis, and NH2-terminal sequencing, the purified complex used here for crystallization consists of subunits α, β, γ, δ, ɛ, b, OSCP, d, a, h, f, ATP8, and c (in diminishing apparent molecular weight order for F1 and F0 on SDS gels) and is similar to other preparations (46, 48, 49). Crystals were grown in the presence of both ADP and the nonhydrolyzable ATP analog 5′-adenylyl-imidodiphosphate (AMP-PNP) by the microbatch technique (50) with polyethylene glycol 6000 as precipitant (42). Subunits α, β, γ, δ, ɛ, and c were detected in the crystals by SDS-PAGE, and there was no evidence for the presence of the other subunits. It appears that they have dissociated from the complex during crystallization, leaving a subcomplex consisting of F1 and 10 copies of subunit c. A similar F1c complex has been isolated previously from spinach chloroplasts (51).

The structure was solved by molecular replacement with the “stand-alone” version of the program AMoRe (52) with data from 15 to 5 Å resolution and the bovine F1–adenosine triphosphatase Cα coordinates [Protein Data Bank (PDB) accession code 1bmf (4)] as a search model (Table 1). The complete bovine model, positioned according to the results of the molecular replacement, was used to calculate phases to 3.9 Å resolution. After solvent density modification with the program Solomon/CCP4 (53, 54) assuming a solvent content of 60%, the α helices of the c subunits and extensive additional density in the central stalk region could be seen in the electron density map (Fig. 2). Because of the limited resolution of the x-ray data, the side chain density is not clear, but the main chain density is generally unambiguous.

Figure 2

Stereo views of an electron density map of the yeast F1c10 complex. The solvent flattened map was calculated at 3.9 Å resolution and contoured at 1.5 σ. (A) Side view containing the bovine F1 Cα model (with α in orange, β in yellow, and γ in green). The density of symmetry-related molecules in the crystal is masked out. The inset indicates the location of the subunits within the complex. The location of the section shown in (C) is indicated by the white box; the direction of the view is indicated by the arrow. The presumed membrane region (M) (2) is marked by the two dotted lines. The c subunits are numbered 3, 2, 1, 10, and 9 (the best ordered c subunit was chosen as number 1). The overall height of the complex is ∼190 Å, of which the α3β3 subcomplex accounts for 83 Å, the stalk for 50 Å, and the c subunits for 58 Å. (B) Enlarged view of the δ/γ-c contact region with the model (and numbering) of the E. coli ɛ subunit (in red) and the E. coli c subunit (in white) fitted into the density, contoured at 1.0 σ. The white box in the inset indicates the location of the displayed section within the complex. (C) End-on view of the density of the c ring. The inset shows the location of the α, β, γ, and δ subunits in relation to the c subunits. The helices of the c subunit are drawn as blue circles, the larger outer circles accounting for the larger side chains in the COOH-terminal helix. The outer diameter of the c ring is 55 Å (top) to 42 Å (equator) to 45 Å (bottom), and the inner diameter is 27 Å (top) to 17 Å (equator) to 22 Å (bottom). The dimensions exclude consideration of unresolved regions of density, including amino acid side chains and detergent or lipid molecules. The two regions of density near subunit 10 are not extensive and are likely to be noise.

Table 1

Statistics of data collection and structure determination. For data collection, the crystals were harvested in a buffer containing 20% glycerol as cryo-protectant and then frozen rapidly in liquid nitrogen. The crystals are small (up to 150 μm in the largest dimension) and diffract x-rays weakly. They belong to the monoclinic space group P21 with unit cell dimensions a = 135.9 Å, b = 175.3 Å,c = 139.2 Å, and β = 91.6°. Assuming one F1c10 complex with 453.2 kD per asymmetric unit, the estimated solvent content is 66%. A data set was collected to 3.6 Å resolution at beamline ID02B (λ = 0.99 Å), at the ESRF, Grenoble, France, with a Mar Research image plate detector (1600 pixel mode). The data were processed with MOSFLM (73) and SCALA/CCP4 (54). Because of anisotropic diffraction and slight radiation damage, the final data set was restricted to 3.9 Å resolution.

View this table:

The crystal packing (Fig. 3) is atypical for membrane proteins and belongs to neither of the types described previously (55). The main crystal contacts are between the bottom of a ring of 10 c subunits and the pseudo–three-fold top of an adjacent F1 assembly. The hydrophobic external regions of the c oligomers, which are normally in contact with phospholipid, are not involved in any crystal contacts and are probably covered by unresolved molecules of detergent.

Figure 3

Stereo view of the crystal packing of the yeast F1c10 complex. A 45 Å thick section through the crystal perpendicular to the crystallographic y axis is shown. The electron density is contoured at 1.2 σ. The red lines mark the x and z axes of the crystal lattice. All figures were prepared with the program MAIN (72).

Molecular architecture of the F1c10complex. The α and β subunits (73 and 79% conserved, respectively, between cow and yeast) are well defined in the electron density and have similar conformations and nucleotide compositions to their bovine counterparts. The NH2- and COOH-terminal α helices of the γ subunit (40% sequence identity to the bovine subunit) extend ∼25 Å farther than in the bovine model toward the c subunits, where they make a sharp bend. In the bovine F1structure, this part of the γ subunit and also subunits δ and ɛ were disordered. There is additional density adjacent to the lower segment of the NH2-terminal α helix of the γ subunit that extends over the c subunits (Fig. 2B), which can be interpreted with the crystal structure of the E. coli ɛ subunit (26% sequence identity to the yeast δ subunit) (56) (PDB accession code 1aqt). The 10 strands of the β barrel are visible in the yeast electron density, with β strand 4 and the following loop (residues 31 to 39 in E. coli numbering) packing against the c subunits and β strands 5, 8, 9, and the loop between β strands 1 and 2 (residues 39 to 42, 68 to 73, 74 to 79, and 9 to 14, respectively) in contact with the γ subunit. The lower α helix (COOH-terminal) lies about 5 Å above the c subunits, and the upper α helix is closer to the α3β3 subcomplex, extending into the solvent. This interpretation of the yeast electron density is in general agreement with a previous model of the E. coli F1c complex (56) and with cross-linking studies (57), although it cannot explain cross linking between the E. coli ɛ subunit and the DELSEED region of a β subunit (9).

The electron densities of individual c subunits show that they consist of two α helices, linked by a loop. There is no interpretable side chain density, and the density could not be improved by modeling the subunits as polyalanine or by averaging. Therefore, it is not yet possible to build an unambiguous atomic model. The electron density is consistent with the NMR structure of the E. coli monomer determined in a chloroform:methanol:water mixture (26) (PDB accession code 1a91). Although the overall shape of the E. coli and the yeast c monomers is similar, the E. coli structure is not a perfect match to the yeast density (Fig. 2B). This could be due to the low sequence conservation (23% identity) between the two species, or it could, at least in part, arise from the contacts between neighboring protomers. The 10 c protomers are packed tightly against each other, forming two rings of single α helices (see Fig. 2C). The outer α helices are kinked in the middle, making the cylindrical outer surface of the oligomer slightly concave. Despite the absence of clear side chain density, it is possible to assign the α helices forming the inner and outer rings to the NH2-terminal and COOH-terminal regions, respectively, as follows. Cysteine residues introduced at positions 40, 42, and 44 of the E. coli c subunit (equivalent to 38, 40, and 42 in yeast) can be cross linked to the ɛ subunit in the E. coli enzyme and are therefore thought to lie in the polar loop between the two α helices (57). The density for the inner α helix is 58 Å long and would accommodate 39 residues in an α-helical conformation, consistent with residues 40 to 44 forming the polar loop. The density for the outer α helix is 47 Å long. This would accommodate 31 residues in an α-helical conformation, which is in excellent agreement with the total of 76 residues in the yeast c subunit. The kink of about 23° in the COOH-terminal α helix is around Gly62 (equivalent to Pro64 in E. coli). This arrangement would also place the glycine/alanine-rich NH2-terminal α helix in the inner ring and accommodate the larger side chains of the COOH-terminal α helix in the outer ring. Because of the lack of side chain density, the conserved carboxylate (Glu59) cannot be placed with certainty, but it would lie about halfway along the outer COOH-terminal α helix. The resulting model shares many features in common with that proposed for the E. coli ring of c subunits (58), except that the latter contains 12 rather than 10 copies. It differs from an alternative model (59, 60), which placed the NH2-terminal helices on the outside. A model for the c subunit of the sodium-driven enzyme from Propionigenium modestum based on NMR data consists of four distinct α-helical segments with the conserved carboxylate located near the cytoplasmic membrane surface (61). There is no evidence for four distinct α helices in the yeast density, and the conserved carboxylate would appear to lie more centrally within the membrane, although the position of the membrane boundary cannot be defined with certainty.

The 10 c subunits form an almost symmetrical ring, although the density for subunits 6, 7, and 8 is weaker, as seen from the asymmetrical appearance of the electron density (Fig. 2C). The central cavity of the c oligomer contains a region of elongated electron density, which is stronger in the upper and lower thirds of the cylinder than in the center (Fig. 2C). The origin of this density is not clear, but the need to maintain a semipermeable membrane in vivo suggests that the cavity might be filled with phospholipid.

The polar loop regions of five of the c subunits (c1 to c5) are covered by the δ subunit (Fig. 2B). Two loops (c2 and c3) are in intimate contact, probably involving hydrogen bonds between pairs of main chain atoms, and another three (c1, c4, and c5) could form side chain contacts to either the β barrel of the δ subunit or the lower of its two α helices. The γ subunit probably makes side chain contacts to one or two further c subunits (c9 and c10) and to one subunit (c1) that is also in contact with the β barrel of the δ subunit. Therefore, six or seven consecutive c subunits or about two-thirds of the top surface of the ring are in contact with the foot of the stalk. These c subunits are much better defined in the electron density than the exposed ones, in which the density is weaker and discontinuous. The 62 residues of the yeast ɛ subunit could not be assigned unambiguously to the density, and about 100 residues of the γ subunit remain unassigned, although there is extensive uninterpreted electron density both on top of the δ subunit and next to the COOH-terminal α helix of γ.

The extent of the interface between the c subunits and the γ and ɛ subunits apparent in the electron density map of the yeast F1c10 oligomer supports the hypothesis that the complex consisting of the γ, δ (E. coli ɛ), ɛ, and c subunits rotates as an ensemble during catalysis, but other interpretations are still possible.

Functional implications of the c ring stoichiometry. An unexpected feature of the electron density map is that the ring of c subunits contains 10 protomers and not, as widely anticipated, 12 (21–23, 58–60). At this stage of the analysis, it is possible that some c subunits have been lost from the ring during crystallization. However, the extent of the interaction between adjacent c subunits within the ring makes this unlikely, particularly as there is no evidence to suggest a strong association between subunit c and any of the other components of F0. In addition, it is likely that the crystal contact between the ring of c subunits and the pseudo–three-fold F1 domain would have selected for a population of complexes with 12 c subunits, if present. In consequence, the three-fold symmetry of the F1 sector and the 10-fold symmetry of the c ring within one complex do not match. The principle of symmetry mismatch in molecular rotary engines has been discussed in relation to the DNA translocating machinery of bacteriophages (62,63). During phage assembly, DNA is injected through the sixfold symmetrical tail into its head (fivefold symmetry axis of an icosahedron). The dodecameric head-tail connector is believed to rotate with respect to the head during DNA packaging (64). The symmetry mismatch between the head and the connector is thought to facilitate rotation by avoiding deeper energy minima that would accompany matching symmetries. Symmetry mismatch also appears to be important in the bacterial flagellar motor, which, similar to ATP synthase, is driven by the proton-motive force (65). A further example of symmetry mismatch is found in some ATP-dependent proteases (66), where a rotational mechanism has been suggested to contribute to protein unfolding. The arrangement of the components of the molecular motor in ATP synthase is rather different than that in either the bacteriophage or ATP-dependent proteases. In the latter cases, there are direct interactions between two rings of subunits with different symmetries. In the ATP synthase, there are two quite distinct regions of interaction. The first, in F1, involves the γ subunit and the α3β3 subcomplex, whereas the second, in F0, involves the ring of c subunits and the a subunit.

The number of c subunits in the ring has profound implications for the number of protons translocated by F0 for each ATP molecule synthesized in F1. Each ATP synthetic event requires that the γ subunit rotates through 120°. This rotation is generated in F0 either by rotation of the c ring with γ (as the F1c10 structure appears to indicate) or by some other mechanism in which the c subunits impart rotation to γ without themselves rotating. A c stoichiometry of 9 or 12 has been considered to require translocation of three or four protons per ATP synthesized to impart a 120° rotation to the c ring (see Fig. 1). A ring with 10 c subunits in it suggests that the H+/ATP ratio is probably nonintegral and that its value lies between 3 and 4. Many of the experimentally measured H+/ATP ratios lie in this range (34, 35).

Assuming that the c ring and γ subunit rotate as an ensemble, a nonintegral H+/ATP ratio also makes it more likely that there is some degree of elasticity in the γ subunit itself, so that the stepping of the c ring (10 steps per rotation) can be matched to the stepping in F1 (three steps per rotation). Elasticity in the γ subunit is an explicit feature of two models for ATP synthase (67, 68), whereas in a third model (69) it is not required.

It has been suggested that the number of c subunits in E. coli varies depending on the carbon source (70,71). More c subunits appear to be assembled into the complex if cells are grown in glucose than with succinate. Therefore, in the latter case, the c ring would be smaller with an accompanying reduced gearing ratio. Changing gears and therefore the membrane potential required for ATP synthesis could be a regulatory mechanism. A variable number of c subunits could also explain the observation that covalently linked trimers of E. coli c subunit can form an active complex (22).

The electron density shown here provides direct proof that c protomers form a ring that is in close contact with the γ and δ subunits. Our interpretation is consistent with previous ones, in which subunits γ, δ, and ɛ form a rotating ensemble. It strongly supports the idea that the c oligomer is part of a rotary motor, converting electrochemical energy into chemical energy stored in ATP. To understand the mechanism of the generation of rotation, it will be necessary to establish the nature of the interaction between the c subunit oligomer found here and the remaining subunits in F0.

  • * To whom correspondence should be addressed. E-mail: walker{at}


View Abstract

Stay Connected to Science

Editor's Blog

Navigate This Article