Structure–function relationships of A-, F- and V-ATPases

Adenosine 5′-triphosphate (ATP) synthesis by oxidative phosphorylation or photophosphorylation is a multi-step, membrane-located process that provides the bulk of cellular energy in eukaryotes and many prokaryotes. Most of the ATP synthesis in these cells is catalyzed by the enzyme, F₁F_o-ATP synthase, also called F₁F_o-ATPase (F-ATPase), which in its simplest, bacterial, form is composed of eight subunits (α₃:β₃:γ:δ:ε:a:b₂:c_9–12). The archaeal A₁A_o-ATP synthase (A-ATPase) has ten subunits (A₃:B₃:C:D:E:F:G:H:I:K_x), the actual subunit stoichiometry being unknown. The term ATPase reflects the fact that the F- and A-enzymes are reversible and can act as proton (or Na⁺)-pu>mping complexes. The F- and the A-ATPases transform energy from a gradient of ions across the membrane to synthesize ATP (Mitchell, 1961; Dimroth, 1997; Müller et al., 1999). Conversely, the free energy of ATP hydrolysis can be coupled to proton (or Na⁺) translocation and generate an ion-motive force (IMF), as in the genetically related vacuolar-type, H⁺-translocating ATPases (V-ATPases). The V-ATPases, consisting of at least twelve distinct subunits (A₃:B₃:C:D:E:F:G_y:H_z:a:d:e:c₆), generate IMFs that are used for ligand trafficking, signaling, nutrient uptake and diverse activities in endomembranes and plasma membranes of animal cells (Wieczorek et al., 1999).

A-, V- and F-ATPases consist of a mosaic of globular structural units, including domain and secondary structures, which also serve as functional units. Morphologically each of these enzymes has three components: a membrane-bound sector, A_o/F_o/V_o, which contains the ion channel, a central connecting stalk, and an approximately spherical assembly, A₁/F₁/V₁, which contains the catalytic sites (Schäfer et al., 1999; Leslie and Walker, 2000; Forgac, 2000). Side-view projections of the F₁F_o- (Wilkens and Capaldi, 1998) and V₁V_o-ATPases (Boekema et al., 1997) show a second stalk (stator) as a fourth distinct feature extending from the F_o or V_o portion. In the case of the Escherichia coli F₁ moiety the central stalk is composed of γ_ec and ε_ec, which are the equivalent of δ_m in mitochondrial F₁F_o, and the stator is formed by the δ_ec and b subunits (Pedersen et al., 2000). The bacterial δ subunit (δ_ec) bears homology to one of the mitochondrial F_o subunits called OSCP (Table1). The mitochondrial F₁ ε subunit (ε_m) has no counterpart in the bacterial F₁F_o enzyme. The central element of the F₁ complex, subunit γ, has been shown to move relative to the α₃β₃ complex during ATP hydrolysis (Capaldi et al., 1996; Junge et al., 1997; Masaike et al., 2000). This rearrangement is proposed to drive the motion of a ring of c_9–14 subunits (Fillingame, 1996; Stock et al., 1999; Seelert et al., 2000; Stahlberg et al., 2001) in the F_o domain (Sambongi et al., 1999; Pänke et al., 2000; Tsunoda et al., 2001), each containing two transmembrane helices (Rastogi and Girvin, 1999).

Based on their subunit composition and primary sequences, the A-type (archaeal) enzymes are more closely related through evolution to V-type than to F-type ATPases (Iwabe et al., 1989; Ihara et al., 1992; Müller et al., 1999). Three interdigitating copies of the nucleotide-binding subunits A and B of the A₁/V₁-ATPases and subunits β and α of the F₁-ATPase, respectively, exhibit more than 25% primary sequence identity (Nelson, 1992). The minor subunits C, D, E, F, G and C, D, E, F, G, H of the A₁ and V₁-ATPases, respectively, form the stalk and are proposed, by analogy to F-ATPases, to be involved, either directly or indirectly, in conversion of energy into controlled motion (Müller et al., 1999; Grüber et al., 2000a). However, the minor stalk subunits of A₁, F₁ and V₁-ATPases show much less similarity than the headpiece subunits, suggesting that there are differences between the three classes of enzyme (Müller et al., 1999). One fundamental distinction is the reversible dissociation of the V₁ from the V_o complex as an in vivo regulatory mechanism for the control of V-ATPase activity (Wieczorek et al., 2000). By contrast, the A₁/F₁ and A_o/F_o sectors form stable associated complexes in the cell. Moreover, A- and F-ATPases, unlike V-ATPases, share the ability to synthesize ATP. Nevertheless, lineage profiles based on primary sequence reveal that A- and V-ATPases are more closely related to each other than to F-ATPases (Schäfer et al., 1999). Despite the fact that A-ATPases display chimeric properties of V- and F-ATPases the structure/function relationships of these enzymes remain a mystery. This review will focus on recent advances in elucidating structural and functional relationships of A₁/V₁- and F₁-ATPases.

Over the past decade, a tremendous amount of structural information about the F₁F_o-ATPase has been obtained using electron microscopy (reviewed in Gogol, 1994; Böttcher and Gräber, 2000), macromolecular crystallography (Abrahams et al., 1994; Shirakihara et al., 1997; Bianchet et al., 1998; Hausrath et al., 1999; Stock et al., 1999; Groth and Pohl, 2001) and nuclear magnetic resonance (NMR) spectroscopy (Wilkens et al., 1995; Wilkens et al., 1997; Rastogi and Girvin, 2000). Significant insights into the molecular mechanism of ATP hydrolysis came from the X-ray structure of the bovine heart α₃β₃γ complex of the F₁-ATPase (MF₁; Abrahams et al., 1994). The crystallographic model describes the three alternating α and β subunits as being arranged hexagonally, surrounding a solvent-filled cavity which is traversed by a part of the γ subunit. This coiled-coil structure of the γ subunit is asymmetrically located in the shaft relative to the axis of the α₃β₃ complex and protrudes from it by about 30Å into the stalk region. A third short α-helix of the γ subunit is inclined at a 45° angle to the coiled-coil domain at the bottom of the F₁ as it merges with the stalk that connects the F₁ and F_o parts. A key feature of the structural model is its asymmetry, particularly in the nucleotide occupancy and conformations of the catalytic β subunits. One β subunit, designated β_TP, is ligated with MgAMP-PNP and linked to the short α-helix of the γ subunit via the C-terminal domain. A second β subunit, β_DP, has MgADP bound and the third catalytic site, β_E, is free of nucleotides and Mg²⁺. The three non-catalytic α subunits are ligated with MgAMP·PNP and adopt similar conformations, although one α subunit, which contributes to β_TP and is therefore denoted α_TP, displays a small rigid body rotation of the top domain relative to the other two domains (Abrahams et al., 1994).

Neither the structural model of the bovine F₁-ATPase (Abrahams et al., 1994) nor the subsequently determined structure of rat liver F₁-ATPase (Bianchet et al., 1998), which includes almost all of the residues of the α and β subunits, include either the small subunits δ_m and ε_m, or approximately half of the γ subunit residues. Subsequently, crystals containing all five subunits of the Escherichia coli F₁-ATPase (α₃:β₃:γ:δ_ec:ε_ec) and the α₃β₃γε_ec complex of the same organism have been obtained and diffracted to a resolution of 6.4Å and 4.4Å, respectively (Grüber et al., 1997; Hausrath et al., 1999). Besides the α and β subunits and the known part of the coiled-coil α-helices of the MF₁ the electron-density map at 4.4Å extends 12 and 20 residues, respectively, thereby adding 15Å to the length of the N-terminal α-helix and 23Å to the C-terminal helix of the γ subunit (Hausrath et al., 1999). This structure reveals that γ extends from the α₃β₃ hexagon far enough to traverse the full length of the central stalk, in agreement with the refined crystal structure at 2.4Å resolution of the γ_ec-ε_ec complex from E. coli (Rodgers and Wilce, 2000) and the complete bovine F₁-ATPase (Fig.1; Gibbons et al., 2000). In these structural models the γ subunit is arranged in six α-helices and five β-stranded β-sheets (Rodgers and Wilce, 2000; Gibbons et al., 2000). The bottom of γ is in contact with the external loops of c subunits (Watts et al., 1995; Watts et al., 1996); the entire γ subunit makes up a coupling domain, which couples ATP hydrolysis with ion pumping (Capaldi et al., 1996; Junge et al., 1997). Adjacent to the ‘bottom’ part of subunit γ an additional density has been observed in the map obtained from crystals of the α₃β₃γδεc₁₀ subcomplex of yeast mitochondrial ATP synthase at a 3.9Å resolution (Stock et al., 1999) and the bovine F₁ (Gibbons et al., 2000). In these densities the structure of the ε_ec subunit (Wilkens et al., 1995; Uhlin et al., 1997), the counterpart of the yeast (δ_ye; Giraud and Velours, 1994) and bovine δ subunit (Fig.1; δ_m), has been modeled. Like subunit ε_ec, the subunits δ_ye and δ_m are composed of a C-terminal helix–loop–helix structure and an N-terminal 10-stranded β-sandwich structure. However, the modeling of the ε_ec subunit indicates that the C terminus of the polypeptide is turned away from the bottom domain of the catalytic β subunit. This domain is believed to be involved in the coupling of catalytic-site events (Capaldi et al., 1996; Grüber and Capaldi, 1996) along with γ and ε_ec acting as a rotor (reviewed in Junge et al., 1997; Masaike et al., 2000; Tsunoda et al., 2001). This structural feature is in conflict with the model of the γ_ec-ε_ec subcomplex (Rodgers and Wilce, 2000), in which the ε_ec is located in close proximity to β via its C-terminal α-helix and with its β-sandwich barrel turned toward the bottom of γ. As shown by cryo-electron microscopy (Gogol, 1994) and biochemical studies (Mendel-Hartwig and Capaldi, 1991; Wilkens and Capaldi, 1998), the ε_ec subunit can exist in different states in the complex depending upon whether ATP, MgATP or MgADP is bound to the enzyme (Mendel-Hartwig and Capaldi, 1991; Wilkens and Capaldi, 1998). Using E. coli F₁ mutants with cysteine substitutions in the C termini of the α, β and ε_ec subunits it has been shown that in the ATP-conformation, when the γ and ε_ec subunits are linked to α subunits, the high-affinity site is completely closed with nucleotide unable to get in or out. In contrast, in the ADP-conformation, when the small subunits are linked to β subunits, there is nucleotide exchange in and out from the high catalytic site (Grüber and Capaldi, 1996). Therefore, the question arises as to whether the arrangement of the equivalent to the bacterial ε subunit in the yeast and bovine F₁F_o complexes reflects a trapped state during its nucleotide-dependent movement.

A model-independent approach, based upon the multipole expansion method using spherical harmonics (Stuhrmann, 1970), has been developed to complement crystallographic studies of the quaternary structure of macromolecules such as the F₁-ATPase from E. coli from solution X-ray scattering data. Application of this approach has led to a low-resolution (32Å) structure of the F₁ complex under nearly physiological and saturating nucleotide conditions (Svergun et al., 1998a; Svergun et al., 1998b). The hydrated F₁-ATPase (Fig.1, Fig.5) is a compact molecule with a headpiece of approximately 108Å from top to bottom and 110Å wide. However, the overall structure is asymmetric due to the stem (stalk) that is approximately 42Å in length and 53Å in cross section (Svergun et al., 1998b; Grüber, 2000). These dimensions are consistent with recent data regarding the central stalk in the 2.4Å resolution structure of the bovine F₁-ATPase (α₃β₃γδ_mε_m) with a length of 47Å and 51Å × 41Å in cross section (Gibbons et al., 2000), showing the regions of the γ, δ_m and ε_m subunits are exposed at the foot of the stalk, and thereby in close contact with F_o subunits, which will consequently facilitate the mechanistic linkage of ATP hydrolysis to ion pumping (Fig.1).

The idea that molecules now known to be V-ATPases may structurally resemble F-ATPases was suggested by early micrographs of insect plasma membranes (Gupta and Berridge, 1966; Anderson and Harvey, 1966), which showed repeating, spike-like units supporting globular structures. The spike-with-globule structures are widely distributed on transporting plasma membranes and were designated ‘portasomes’ (Harvey et al., 1981). Meanwhile, negatively stained membranes from bovine chromaffin granules (Schmidt et al., 1982) and the vacuolar membranes of Neurospora crassa (Bowman et al., 1989) were shown to contain similar structures, and sequencing of the genomic DNAs encoding V-ATPase subunits (Bowman et al., 1988) demonstrated beyond doubt that the particles are V₁-ATPases. The regulatory mechanism of reversible disassembly of the V₁ and V_o complex, first shown in the Manduca sexta midgut (Sumner et al., 1995) and also later in yeast (Kane, 1995), suggested that a study of the dissociated V₁ complex could provide valuable information about the structural features of this enzyme. The recovery of disassembled V₁ particles from the cytoplasm in high yield and purity (Gräf et al., 1996) made the structural description possible.

Two major advances have been made toward elucidating the quaternary structure of V₁ during the past two years. First, the gross structure of the M. sexta midgut V₁-ATPase was investigated by SAXS (Svergun et al., 1998b). The enzyme is highly elongated with a maximal length of about 220Å. The solution scattering data define a hydrated complex with a headpiece approximately 145Å in diameter and a stalk approximately 110Å in length (Fig.5). Second, image processing of electron micrographs of negatively stained V-ATPases from Clostridium fervidus (Boekema et al., 1998) and V₁-ATPase from M. sexta (Radermacher et al., 1999) yielded two-dimensional structures at a resolution of 18Å and 24Å, respectively. A comparison of the independently identified structures (Boekema et al., 1998; Svergun et al., 1998b; Radermacher et al., 1999) revealed that the headpiece consists of a pseudo-hexagonal arrangement of six masses, surrounding a seventh mass. These six masses, which are assumed to consist of the major subunits A and B, are arranged in an alternating manner (Boekema et al., 1998; Svergun et al., 1998b). The first three-dimensional reconstruction of the V₁ complex was determined at 32Å resolution from negatively stained preparations of the M. sexta V₁-ATPase (Fig.2) (Grüber et al., 2000a). A striking feature of the reconstruction is the presence of six elongated lobes, approximately 20Å in diameter and 90Å in length, which are parallel to the threefold axis (Fig.2B). These lobes, which represent the alternating three copies each of subunits A and B, can be traced for most of the length of the V₁-ATPase. The hexagonal barrel of subunits A and B encloses a core of approximately 40Å. In this model the V₁ complex is barrel-shaped, being approximately 110Å high and 135Å wide. At both ends of the hexagonal barrel extensions can be observed. The extensions on one side (Fig.2A) are consistent with published two-dimensional average images of the V₁V_o-ATPase from bovine brain clathrin-coated vesicles, where elongated features (Fig.2F–H; as, ce) can be seen at the very top of the V₁ domain (Wilkens et al., 1999). The extensions on the opposite side can be attributed to traces of the stalk, e.g. the extension visible in Fig.2B,C. The correspondence of dimensions of the hexagonal domain as determined by SAXS (see above) and electron microscopy indicate that the stalk is not completely resolved in the three-dimensional reconstruction, presumably due to absorption and drying of the V₁ particle on the carbon film. However, a striking fact is that the shape and interdigitation of the A₃B₃ subunits, located around the periphery of the barrel, are in agreement with the three-dimensional model of the related F₁-ATPase, derived from two- (Capaldi et al., 1992; Gogol, 1994) and three-dimensional crystals of this complex (Abrahams et al., 1994; Bianchet et al., 1998; Hausrath et al., 1999), in which the alternating subunits α and β interdigitate for the full length surrounding the γ subunit.

Further insights into the topology of the V₁-ATPase from M. sexta were obtained by tryptic digestion (Grüber et al., 2000a). Proteolytic digestion revealed that the nucleotide-binding subunit B was cleaved rapidly, yielding two fragments with apparent molecular masses of 25 and 34kDa, whereas trypsin had no obvious effect on the major catalytic subunit A. Among the stalk subunits, C–H, subunit D was cleaved most rapidly by trypsin, followed by subunits F, G, C and H, with subunit E being modified most slowly. The rapid and slow cleavage of subunits D and E, respectively, is an important finding since both have been proposed as structural and functional homologues of the γ subunit of F-ATPases (Bowman et al., 1995; Nelson et al., 1995; Xu and Forgac, 2000). The observation that the D subunit is cleaved immediately into small peptides by trypsin is surprising; it implies that this polypeptide is exposed in the enzyme and does not support its putative role as a γ homologue (Grüber et al., 2000a).

Copper chloride-mediated disulfide formation yielded further insight into the proximity of the M. sexta V₁ subunits to each other and into their functional relationships. When the enzyme was incubated with 2mmoll⁻¹ CaADP on ice before Cu²⁺ treatment, two bands with apparent molecular masses of 120 and 110kDa, consisting of the subunits A,E,F and B,H, respectively, were obtained. A B,H product did not occur when cross-linking was conducted in 2mmoll⁻¹ CaATP on ice to slow down ATP hydrolysis, implying that subunit H moves away from B to the A,B interface (Fig.3). This interpretation is consistent with the model of the yeast V₁-ATPase, in which subunit H was located at an interface of the nucleotide-binding subunits A and B (Tomashek et al., 1997). Moreover, in the presence of CaATP, two new bands with apparent molecular masses of 42 and 44kDa, and composed of the subunits E,G and E,F, respectively, were observed. A homologous cross-linked product consisting of subunits E and G was also been generated using dimethyl sulfoxide (Thomashek et al., 1997) and disuccinimidyl glutarate (Xu et al., 1999) as cross-linking reagents.

Taken together, the trypsin cleavage and the cross-linking data imply that one region of subunit E is shielded by the smaller subunits F and G (Fig.3). In addition, the disulfide bonding of the catalytic A subunits with subunit E in the presence of CaADP indicates that these subunits are near neighbors. The close proximity of subunit E to subunits A, F and G would allow stalk subunit E to couple events in catalytic subunit A via stalk subunits F and G (Grüber et al., 2000a) in the V₁ portion of the V₁V_o-ATPase to events in the ion-translocating V_o portion (Fig.3; Tomashek et al., 1997).

There is abundant evidence that V-ATPase activity is modulated by disulfide-bond formation (Feng and Forgac, 1992; Forgac, 2000). A mechanism of reversible disulfide-bond formation between cysteine residues (Cys₂₅₄ and Cys₅₃₂) of the catalytic A subunit was proposed to regulate the V-ATPase in vivo (Oluwatosin and Kane, 1995; Forgac, 2000). A mechanism was proposed in which disulfide bond formation is quickly followed by dissociation of the V₁ and V_o complexes (Dschida and Bowman, 1995), implying that nucleotide-binding and hydrolysis in the A subunit of the V₁ domain have to be linked by the stalk region to ion translocation in the membrane-bound V_o domain. Recently SAXS experiments showed that reducing the V₁-ATPase of M. sexta leads to significant changes in the overall dimensions of the complex. The radius of gyration of the oxidized and reduced enzyme are 62±6Å and 58±6Å, respectively, whereas the maximum dimension of both complexes remains constant at 220±10Å (Grüber et al., 2000b). The shapes of both complexes (Fig.4) were determined ab initio at a resolution of 27Å by a simulated annealing procedure, based on a representation of the structure in terms of dummy atoms (Svergun, 1999). Both low-resolution structures have a characteristic mushroom-like shape with a central stalk of significant length, similar to the identified structures of the V₁-ATPase from Clostridium fervidus (Boekema et al., 1998) and M. sexta (Svergun et al., 1998), using electron microscopy and SAXS, respectively. Comparison of the oxidized and reduced models indicates that the main conformational changes upon reduction take place in both the crown-like region at the very top of the globular headpiece, where the major subunits A and B are located, and in the elongated stalk. Both regions evolve into an arrow-like shape after reduction (Grüber et al., 2000b). Based on homology of the subunits A and B to the related F-ATPase subunits β and α, respectively (Nelson, 1992), whose N termini form a β-barrel domain in a crown-like fashion (Bakhtiari et al., 1999), the conformational changes at the top of the V₁-ATPase are presumably due to rearrangements in the N termini of the A and B subunits. As shown more recently by three-dimensional reconstructions of the related F₁F_o-ATPase from E. coli, a crown-like shape, which is missing in the absence of the nucleotide (Böttcher et al., 2000), evolves upon binding of the non-cleavable nucleotide analogue AMP-PNP into the catalytic β subunit. The appearance of the crown has been attributed to rearrangements in the N-terminal domains of the α and β subunits, located at the very top of F₁. Moreover, when AMP-PNP or ADP are bound to the catalytic site, subunit β assumes its closed conformation, in which the adenine-binding pocket moves into close proximity with the phosphate-binding domain, the P-loop; it moves away when the binding-site is empty (open conformation), as shown by the crystallographic model of the α₃β₃γ subcomplex of bovine heart mitochondrial F₁-ATPase (Abrahams et al., 1994). There is a striking similarity between the crown structure of the E. coli F₁F_o-ATPase (Böttcher et al., 2000) that evolves after binding of AMP-PNP (closed conformation) and the crown-like feature that is observed in the oxidized V₁-ATPase (Grüber et al., 2000b). In this state, the catalytic A subunit is proposed to be in a closed conformation (Forgac, 2000), and alters into a wedge-like shape after reduction of V₁. In summary, the structural changes in the headpiece, upon reduction of the enzyme, correspond with alterations of the protuberance of the stalk into a wedge-like feature, which enables the enzyme to transmit the activating movements that take place in the V₁ headpiece to the V_o complex.

Like F₁ and V₁ complexes of F- and V-ATPases, the A₁ complex of archaeal ATPases possesses a pseudo-hexagonal arrangement of the major subunits A and B, as proposed from two-dimensional images of the thermoacidophilic archaea Sulfolobus acidocaldarius and Methanosarcina mazei Gö1 (Lübben et al., 1988; Wilms et al., 1996). However, in contrast to the related F₁- and V₁-ATPases described above, little is known about the overall structure of the enzyme. This information gap is largely due to the instability of the isolated complexes (Wilms et al., 1996; V. Müller, unpublished). A new avenue of research was opened by the cloning and sequencing of the A₁A_o-ATPase encoding genes from methanogenic archaea. In M. mazei Gö1, the genes encoding the ATPase are clustered on the chromosome and comprise an operon. A fragment containing ahaE, ahaC, ahaF, ahaB, ahaA and ahaG was cloned in an overexpression vector and transformed into the F₁F_o-ATPase negative mutant E. coli DK8, which produced an A₁-ATPase upon induction of gene expression. This A₁ complex is made up of the five different subunits A, B, C, D and F, with apparent molecular masses of 64, 51, 41, 24 and 11kDa, respectively, as estimated from the amino acid sequences (Wilms et al., 1996); subunit E was not produced in E. coli (T. Lemker and V. Müller, unpublished). Based on SAXS data, the A₃B₃CDF-complex (Fig.5) comprises a headpiece approximately 94Å long and 92Å wide (Grüber et al., 2001). Superposition of the low-resolution structure of the A₁ complex with the atomic model of the α₃β₃γ complex of the related F₁-ATPase from E. coli (Hausrath et al., 1999) reveals a striking similarity, especially with respect to the disposition of the nucleotide binding subunits α and β, the homologs of subunits B and A, respectively (Fig.5B; Grüber et al., 2001). This structural similarity lends support to the view that A- and F-ATPases share a common catalytic mechanism for ATP synthesis. The overall structure of the hydrated particle is asymmetric because of the stalk, which is approximately 84Å long and 60Å in diameter. The shape and length of the stalk strongly resemble those in the stalk domain of the related V₁-ATPase from M. sexta (see Fig.5) (Svergun et al., 1998b; Grüber et al., 2000b). A comparison of the overall structures of the A₁- and V₁-ATPases (Svergun et al., 1998b; Grüber et al., 2001) with the low resolution (Svergun et al., 1998a; Svergun et al., 1998b) and atomic models of the E. coli (see Fig.5; Hausrath et al., 1999) and bovine heart F₁-ATPase (Gibbons et al., 2000) identified major differences between these molecules. In particular, the F₁-ATPase differs in shape from the other two enzymes and has a significantly shorter stalk, being approximately 40–45Å long and 50–53Å wide (Gibbons et al., 2000; Grüber, 2000). These differences are consistent with the proposed evolutionary linkage of A₁- and V₁-ATPases, which are thought to have evolved from common ancestral genes (Iwabe et al., 1989; Ihara et al., 1992). Tryptic digestion studies of the A₃B₃CDF complex have shown that the subunits C and F are exposed in the complex, whereas subunit D is well protected from the effect of trypsin (Grüber et al., 2001). The shielding of subunit D from trypsin is an important finding since this subunit has been proposed as the structural and functional homolog of the γ subunit of F-ATPases (Müller et al., 1999; Grüber et al., 2001). In experiments where CuCl₂ was added after preincubation of MgATP, the cross-linked product A-D was formed, which was absent in the presence of MgADP+P_i (Ü. Coskun, J. Godovac-Zimmermann, T. Lemker, V. Müller and G. Grüber, unpublished data). The disulfide bond that forms between catalytic subunit A and subunit D in the presence of MgATP indicates that these subunits are near neighbors. Furthermore, the absence of an A-D product, when MgADP+P_i are bound, also indicates a rearrangement of these subunits due to nucleotide binding. Taken together, the shielding of A₁ subunit D from trypsin and the nucleotide-dependent cross-linking of A to D are reminiscent of the shielding of the V₁ subunit E from the protease and the substrate-dependent proximity of E to the catalytic A subunits (Grüber et al., 2000a), which suggests in turn that A₁ subunit D and V₁ subunit E may have similar functional roles.

Phylogenetic studies show that A- and V-ATPases evolved from a common ancestor (Iwabe et al., 1989; Ihara et al., 1992). The evolutionary relationship of both enzymes, and their relationship to F-ATP synthases, were confirmed by comparing the low-resolution structure of the A₁- F₁- and V₁-ATPases, which revealed that a knob-and-stalk-like shape is common to all three complexes. The stalk domains of the more closely related A₁ and V₁ are remarkably similar in shape and dimensions, and are different in these respects from the F₁-ATPase. Despite the differences in structure, A₁A_o and F₁F_o enzymes function as ATP synthases in cells whereas the V₁V_o-ATPase works as an ATP-driven ion pump. The elucidation of the structural basis for this functional difference is a challenge for future studies.

Further evidence for a closer relationship between A- and V-ATPases than either has to F-ATPase is found by comparing A_o and V_o to F_o domains. Unlike the F_o domain (see above) the V_o and A_o domains contain only two membrane integral subunits, a (I in A₁A_o-ATPases) and c (the so called proteolipid). There is a surprising variation in the number of proteolipid subunits: two additional proteolipid subunits, c′ and c′′, both of which have homology to subunit c (Stevens and Forgac, 1997) are found in some V-ATPases, the F-ATPase from the bacterium Acetobacterium woodii contains two 8-kDa proteolipid subunits, c₂ and c₃, and the A-ATPase from the archaeon Archeoglobus fulgidus contains two 8kDa proteolipids (Müller et al., 1999; Müller et al., 2001). Moreover, there is an astonishing variability in the size of the proteolipids in archaeal A-ATPases with two, four or six transmembrane helices and a variable number of conserved ionizable groups per monomer (Müller et al., 1999). A proteolipid with four transmembrane helices but only one ionizable group has recently been described in the bacterial F₁F_o-ATPase from A. woodii (Aufurth et al., 2000). Regarding these remarkably complex machines there still remains the question of how the energy coupling between the proposed ion-gradient-driven motion within the A_o and V_o domains is coupled by their stalks to the reversal of the ATP-driven motion in A₁ and V₁ domains.

This research was supported by grants from the Deutsche Forschungsgemeinschaft to V.M. (Mu801/10), H.W. (Wi 698 and SFB 431) and G.G. (GR 1475/6-1), respectively, and from the National Institutes of Health to W.R.H. (AI 22444).