Three-dimensional Map of a Plant V-ATPase Based on Electron Microscopy*

V-ATPases pump protons into the interior of various subcellular compartments at the expense of ATP. Previous studies have shown that these pumps comprise a membrane-integrated, proton-translocating (V0), and a soluble catalytic (V1) subcomplex connected to one another by a thin stalk region. We present two three-dimensional maps derived from electron microscopic images of the complete V-ATPase complex from the plantKalanchoë daigremontiana at a resolution of 2.2 nm. In the presence of a non-hydrolyzable ATP analogue, the details of the stalk region between V0 and V1 were revealed for the first time in their three-dimensional organization. A central stalk was surrounded by three peripheral stalks of different sizes and shapes. In the absence of the ATP analogue, the tilt of V0changed with respect to V1, and the stalk region was less clearly defined, perhaps due to increased flexibility and partial detachment of some of the peripheral stalks. These structural changes corresponded to decreased stability of the complex and might be the initial step in a controlled disassembly.

The catalytic A-subunits of V-ATPase are homologous to the catalytic ␤-subunits in F-ATPase (5) and the B-subunits of V-ATPase to the non-catalytic ␣-subunits in F-ATPase (6). The membrane-integrated V-ATPase c-subunit has probably emerged by gene duplication (7) from a common ancestor of Fand V-ATPases. The G-subunit of V-ATPases has sequence similarity to the hydrophilic part of the membrane-anchored F-ATPase b-subunit (8). Other components of the F-ATPase machinery do not have any homologues in V-ATPase. Furthermore, V-ATPases contain various subunits (C, F, H, a, and d) whose functions and relationships to F-ATPases still need to be elucidated. This divergence might reflect an adaptation to the different physiological requirements of F-and V-ATPases. Unlike F-ATPases, V-ATPases usually do not synthesize ATP but hydrolyze ATP to generate proton motive force. When resources become scarce, V-ATPases are required to be shut down to save the diminishing ATP levels for more vital cellular processes. This shutdown is achieved by reversible disassembly of the V-ATPase complex, a process that is unknown in the regulation of F-ATPases. In yeast, disassembly is initiated in response to glucose deprivation (for review see Ref. 9). Although the underlying molecular mechanism is not completely understood, it is speculated that the disassembly might be initiated by a brief drop in cellular ATP levels caused by decreasing glucose levels (10). This drop in ATP levels possibly causes conformational changes in the V-ATPase that start the disassembly process.
Although we have detailed knowledge of the structural organization of various subcomplexes of F-ATPases (11,12), only little is known about the structure of V-ATPases. The V-ATPase topology was investigated by a number of biochemical experiments, particularly contacts and proximities between subunits were explored by cross-linking studies (for review see Refs. 13 and 14). These topological data were complemented by low resolution three-dimensional information about the isolated V 1 subcomplex (15)(16)(17) and the isolated V 0 subcomplex (18).
Up to now, three-dimensional information for a complete V-ATPase complex has not been available. Two-dimensional projection maps derived from electron micrographs of two different V-ATPases (19,20) have given an impression of the overall architecture of the enzyme and revealed a complicated connecting region, composed of at least three stalks, between V 0 and V 1 (21,22). However, the lack of three-dimensional information has made it impossible to come to any conclusions on the exact number of connecting elements or to determine whether these peripheral connections are formed by structurally identical components.
To shed some light on the architecture of V-ATPases, we have calculated three-dimensional maps from electron micrographs of the gold thioglucose-stained plant V-ATPase from Kalanchoë daigremontiana (Mother-of-Thousands) in the pres-ence and absence of the non-hydrolyzable ATP analogue AMP-PNP. AMP-PNP served to mimic nucleotide concentrations comparable with those found in cells under normal growth conditions, whereas the V-ATPase without AMP-PNP added reflects a complex under complete nucleotide-deprived surroundings. With these studies we were able to get detailed information of the three-dimensional organization of the V-ATPase complex as well as of its response to changing nucleotide concentrations.

EXPERIMENTAL PROCEDURES
Purification-The V-ATPase from leaves of the crassulacean acid metabolism plant K. daigremontiana Hamet et Perrier de la Bâ thie was, with slight modifications, isolated as described previously (23). Tonoplast vesicles were prepared from leaf tissue homogenates by sucrose density ultracentrifugation. For solubilization of the membranebound proteins, the tonoplast vesicle suspension was incubated with 1% (w/v) Triton X-100 for 30 min on ice and subsequently centrifuged for 30 min at 300,000 ϫ g and 4°C (70Ti rotor; Beckman). The supernatant was immediately subjected to size-exclusion chromatography (23) with minor modifications. A volume of 500 l was applied to a Superose 6 column (Amersham Biosciences) that was equilibrated with the elution buffer (10 mM MgSO 4 , 2 mM dithiothreitol, 1 mM EGTA, and 1 mM dodecyl maltoside, 10 mM Tris, pH 8), at a flow rate of 0.5 ml min Ϫ1 . Fractions of 1 min (0.5 ml) were collected.
Protein Determination, SDS-PAGE, Activity Assay, and Ion-exchange Chromatography-The protein content was quantified spectrophotometrically at a wavelength of 620 nm using the dye Amido Black 10B (Merck) and bovine serum albumin as protein standard (24).
The subunit composition of the preparations was determined by PAGE in the presence of 0.1% SDS on 13.5% polyacrylamide gels (25). Proteins in the gels were visualized by silver staining (26).
Rates of ATP hydrolysis activity were calculated from the amount of inorganic phosphate released during an incubation period of 30 min at 37°C as described previously (27). Inorganic phosphate was assayed using the method of Lin and Morales (28). ATP hydrolysis activity of the V-ATPase was determined from measurements in the absence and presence of 5 nM concanamycin A 1 , a specific V-ATPase inhibitor (29).
To investigate further the stability of the complex, purified V-ATPase fractions from the Superose 6 column were reapplied to small ionexchange columns of DEAE-Sepharose Fast Flow (Amersham Biosciences) equilibrated in loading buffer (4 mM MgSO 4 , 2 mM dithiothreitol, 1 mM EGTA, 1 mM dodecyl maltoside, 10 mM Tris-HCl, pH 6.5) with and without 2 mM AMP-PNP. The columns were eluted with increasing concentrations of NaCl (50, 100, 250, and 500 mM) again with and without 2 mM AMP-PNP. Fractions were analyzed by SDS-PAGE and visualized by silver staining. The expected elution fractions (250 mM NaCl) were imaged by electron microscopy exclusively to check for intact particles but not to collect and process any data.
Electron Microscopy-Carbon-coated 400 mesh copper grids were glow discharged in air for 1 min and used within 1 h for sample preparation. The protein was applied to the grids either undiluted or with the non-hydrolyzable ATP analogue AMP-PNP (100 mM) added to a final concentration of 2 mM. After about 30 s of incubation the sample was removed by blotting the edge of the grid with kitchen paper. The grid was then washed several times with staining solution by applying a drop of staining solution to the grid and then removing it with kitchen paper. As staining solution either 1% gold thioglucose for the samples without added AMP-PNP or 1% gold thioglucose, 2 mM AMP-PNP for samples with added AMP-PNP was used.
Untilted and 20 -30°tilted micrographs were taken under "low dose" conditions with an accumulated total dose of less than 2000 e/nm 2 . The data were collected on a Philips CM120 Biotwin electron microscope operating at 100 kV and equipped with a tungsten filament. The images were taken with a defocus of 500 -700 nm and were recorded on Kodak SO-163 film that was developed for 10 min in Kodak D-19 developer at room temperature. The magnification of the microscope was calibrated with catalase (30) and was ϫ50,000.
Image Processing-Electron micrographs were scanned with 21 m/ pixel on a Zeiss SCAI scanner corresponding to a pixel size of 0.42 nm at the specimen level. In total 11,745 particle images of V-ATPase with added AMP-PNP and 10,139 particle images of V-ATPase without added AMP-PNP were selected and boxed off from the micrographs using the MRC software package (31). All subsequent image analysis steps were performed within the IMAGIC 5 software package (32). Each set of particle images was processed separately following the standard procedures outlined in the IMAGIC 5 manual. In brief, the particle images were normalized in their gray-value distribution, bandpassfiltered using a low frequency cut-off of 1/16 nm Ϫ1 and a high frequency cut-off of 1/1.6 nm Ϫ1 , and mass-centered. The centered particle images were grouped into classes using the multivariate statistical analysis module of IMAGIC 5, and the relative orientations of the class averages were calculated using sinogram correlation. The class averages were combined to a three-dimensional map using the weighted back-projection algorithm. The spatial orientations of the class averages were refined by sinogram correlation between projections of the three-dimensional map and the class averages. After several rounds of refinement the projections of the current best map were used as a set of references for multireference alignment followed by classification, sinogram correlation, and three-dimensional image reconstruction. This was repeated until no further improvement was observed.
For final refinement of the three-dimensional map the spatial orientations of the particle images were determined by projection matching. The reference projections were generated from the three-dimensional map by projection and were spaced by 12°uniformly across the asymmetric unit. All particle images that aligned best to a certain reference were averaged. The projection angles of the matching references were assigned to the averages, which were then used for the calculation of the three-dimensional map. The quality of the averages was evaluated by comparison to the matching projections of the three-dimensional map. Usually averages with a high error included only few particle images and were consequently noisier. These averages were excluded. From the remaining averages an improved three-dimensional map was calculated.
For estimation of resolution the particle images were divided into two halves. From each half an independent three-dimensional map was calculated. The agreement of the two maps was measured by Fourier shell correlation. The correlation dropped to 0.5 at a Fourier spacing of 1/2.2 nm Ϫ1 for the complex with and without added AMP-PNP and cut the 3 curve (noise correlation) at 1/1.6 nm Ϫ1 (Fig. 3), which was the limit introduced by the cut off chosen for the initial bandpass filtering of the raw data.

RESULTS
The K. daigremontiana V-ATPase samples used in our analysis by electron microscopy were characterized for purity and activity by SDS-PAGE and activity assays, respectively. Fig.  1A shows the typical polypeptide pattern of the V-ATPase after purification that corresponded to what has been reported previously (23,33). During recent work the D-subunit (34 kDa) and two E-subunit isoforms (32 and 33 kDa) have been identified in the K. daigremontiana V-ATPase by matrix-assisted laser desorption ionization-mass spectroscopy (34). All other subunits of V 1 (A-C and F-H) and subunits a, c, and d of V 0 were identified according to their molecular mass in comparison to the molecular mass of V-ATPase subunits of other species (for review see Ref. 3). The activity of the preparation was 1.7 mol of ATP mg Ϫ1 protein min Ϫ1 and was completely inhibited by the specific V-ATPase inhibitor concanamycin A 1 . Samples for electron microscopy were prepared from as little as 40 ng. Electron micrographs of V-ATPase negatively stained with gold thioglucose showed a homogeneous particle distribution (Fig. 1B). For image processing, only particle images of dumbbell-like shape were chosen to ensure that all of the selected complexes consisted of V 0 and V 1 .
V-ATPase with AMP-PNP Added-For imaging the V-ATPase in the presence of AMP-PNP, AMP-PNP (final concentration 2 mM) was added to both the purified enzyme and to the staining solution. We provided the substrate analogue AMP-PNP both to mimic nucleotide concentrations comparable with those found in cells under normal growth conditions and to ensure a stable, active conformation of the enzyme complex from a structural point of view. From a total of 134 micrographs (of which 59 were tilted by either 20, 25, or 30°) of V-ATPase, almost 12,000 single particle images were extracted and subjected to objective alignment procedures followed by multivariate statistical analysis. The class averages that were obtained represented different projections of the V-ATPase ( Fig. 2A). The lower V 0 subcomplex was bean-shaped, and the upper V 1 subcomplex appeared to be asymmetric, with a prominent "spike" density on top ( Fig. 2A, white arrowheads). In addition, small "knob"-like densities were visible at the periphery of V 1 ( Fig. 2A, 1-3, white arrows). V 1 and V 0 were connected by a central stalk from which a strong peripheral density extended in most of the class averages ( Fig. 2A, asterisks). Some projections revealed an additional elongated second peripheral density ( Fig. 2A, 1, 2, and 4, black arrows) that stretched out from V 0 and reached all the way up to the top of V 1 .
The class averages were combined into a three-dimensional map shown as surface representations in Fig. 2B. Projections of the three-dimensional map (Fig. 2C) were calculated in the same directions as determined for the class averages to validate the map. The projections matched the corresponding class averages ( Fig. 2A), illustrating that the three-dimensional map described the original data accurately. This was further supported by the Fourier shell correlation computed from two separate three-dimensional maps each calculated from half of the collected data (Fig. 3, curve 1). The Fourier shell correlation showed a smooth fall off with increasing Fourier spacing, dropping to a correlation of 0.5 at 1/2.2 nm Ϫ1 and traversing the 3 curve at 1/1.6 nm Ϫ1 (limit introduced by bandpass filtering of the original data).
The density distribution inside the complex can be better explored in slices through the map (Fig. 4A) in an orientation approximately perpendicular to the supposed plane of the membrane. These slices revealed a V 1 with distinct structures, a V 0 with an oval cross-section, and a thin stalk region. The V 1 had an overall diameter of 11.7 nm and an elongated narrow cavity in its center. At the top of V 1 , the spike density (  (Fig. 4A, slices 6 -8, black arrowheads). In the following, the described peripheral stalks are referred to as "prominent," "intermediate," and "faint" stalk.
In Fig. 4B, a selection of slices through the three-dimensional map parallel to the supposed plane of the membrane is shown. The first section revealed the spike at the crest of V 1 to have the form of a circular arc fragment of about 100°. Section two showed a cross-section through the upper part of V 1 . The likely positions of the three A-and three B-subunits are indicated by black and white arrows, respectively (see "Discussion"). Section three described the transition from V 1 to the stalk region, with a rather round outline. Section four showed cross-sections of the central and the three peripheral stalks (see arrowheads). The positions of the peripheral stalks could be described by the angles enclosed between the axes connecting the centers of the peripheral stalks to the midpoint of the central stalk. These angles were 60°between the prominent and the intermediate stalks, 120°between the prominent and the faint stalks, and 180°between the faint and intermediate stalks. Section five represented a cross-section of the V 0 . It showed a pronounced outer ring of density and a diffuse density distribution inside.
V-ATPase without AMP-PNP Added-During purification loosely bound nucleotides were removed from the V-ATPase. To test the stability of the complex within different nucleotide contexts, in a separate experiment the purified V-ATPase was reapplied to a DEAE-cellulose column in the presence or absence of the substrate analogue AMP-PNP. When AMP-PNP was not added to the complex, subunits came off in wash fractions before the elution (250 mM NaCl) as seen on SDS-PAGE, and almost no intact particles were observed by electron microscopy (data not shown). In contrast, when AMP-PNP was added to the sample and buffers, the majority of the protein that was eluted from the column represented intact V-ATPase holoenzymes as seen on SDS-PAGE and as visualized by electron microscopy (data not shown).
To investigate what structural changes were linked to the observed lability of the complex in the absence of loosely bound nucleotides, i.e. without added nucleotides in the surrounding buffer, we analyzed another data set of about 10,000 individual images of the same purified V-ATPase preparation as used in the electron microscopic investigation described above, but this time without adding AMP-PNP. The data were collected and processed in the same way as the first data set. A surface representation of the three-dimensional map (Fig. 5B) shows the same gross architecture as already observed in the threedimensional map of the V-ATPase with AMP-PNP added (Fig.  5A, hereafter referred to as AMP-PNP map), although some of the structural details were different. Both V 1 and V 0 occupied the same volumes as in the AMP-PNP map. Again, V 1 had a spike at its very top (see white arrowhead), albeit less prominent than in the AMP-PNP map. V 1 and V 0 were also joined by a central stalk that was thinner than that in the AMP-PNP map. Furthermore, a peripheral density extended from the base of the central stalk like a little arm. When the threedimensional map was aligned to the AMP-PNP map with the spike and the direction of elongation of the V 0 subcomplex in the plane of the membrane matching, this small stalk lined up with the faint stalk in the AMP-PNP map. In the class averages and in slices through the three-dimensional map (not shown), the typical features of the three peripheral stalks were visible but were indistinct and lower in density than in the AMP-PNP map and were therefore not seen in the surface representation of the three-dimensional map. The most obvious difference between the two maps, however, was the tilt of V 0 with respect to V 1 which changed by about 30°in comparison to the AMP-PNP map.

DISCUSSION
Validation of the Three-dimensional Map-A prerequisite for the combination of two-dimensional projection maps into a meaningful three-dimensional map is the consistency of all projections with a unique three-dimensional volume. For negatively stained samples, this is only fulfilled when particles are completely embedded in stain and do not suffer flattening and other distortions that alter the shape of the particles depending on their orientation on the support film (35). Therefore, rather than staining with the commonly used uranyl acetate, which often causes such complications, we chose to use gold thioglucose. In the past gold thioglucose has occasionally been used for staining, e.g. two-dimensional crystals (36,37), whole centrioles (38), or individual complexes (39,40). In most of these investigations a better particle preservation was observed by embedding in gold thioglucose than by conventional negative staining. It is likely that under low dose conditions gold thioglucose preserved structures similarly well as the chemically related glucose does, which can maintain structures to better than 1.0 nm resolution even at room temperature (41). In our hands gold thioglucose provided good projection consistency, as was supported by two observations. 1) In general, all of the calculated class averages contained particle images that were extracted from both tilted and untilted micrographs. All particles that were averaged into one class were supposed to have the same orientation with respect to the electron beam but different orientations with respect to the support film. We found that within one class average, averages of tilted particles did not vary significantly from those of untilted particles. Thus, projections were consistent and did not depend on the orientation of the particles on the support film. 2) All class averages could be matched by projections generated from the threedimensional map, demonstrating that the map is sufficient to explain all the observed data (Fig. 2C).
The V 1 Subcomplex-The three-dimensional structure of the isolated V 1 subcomplex of Manduca sexta midgut V-ATPase had been studied previously (16,17) in detail by electron microscopy and image processing. The V 1 subcomplex shows a pseudo-hexagonal symmetry, as expected from biochemical studies (42) and from the similarity between the V-ATPase Aand B-subunits and the hexagonally arranged F-ATPase ␣and ␤-subunits (12). However, none of the peripheral stalks are resolved in the map of the isolated V 1 (17), although most of the subunits supposed to form the peripheral connections are present in the sample. In the context of the holoenzyme in K. daigremontiana where the peripheral connections were visible, the V 1 showed a hexagonal arrangement of subunits only in the upper part of V 1 (Fig. 4B, slice 2), where the three peripheral stalks did not interfere with the subunit arrangement of the Aand B-subunits as in the lower part of V 1 (Fig. 4B, slice 3).
From the outer surface of the upper third of V 1 , three knoblike densities extended (Fig. 2B, 1-3, white arrows) and were also revealed in the class averages ( Fig. 2A, 1-3, white arrows). Similar knobs had been found in the class averages of V-ATPases of bovine clathrin-coated vesicles (20) and of Caloramator fervidus (19). It had been suggested that compared with the related F-ATPase those knobs are formed by the N-terminal, non-homologous inserts of the A-subunits (5,20). Because there are three A-subunits alternating with three B-subunits in a hexagonal arrangement, the three-dimensional map should show three A-subunit-associated knob extensions at the periphery of V 1 that were related by 120°rotation around the long axis of the molecule. Indeed, knobs were observed at the expected positions (Fig. 2B, 1-3, white arrows) and were missing at the opposite sites corresponding to the equivalent parts of the B-subunits that do not have the inserts. Thus, the positions of the A-and B-subunits could be assigned to our three-dimensional map as indicated by black (A-subunits) and white arrows (B-subunits) in Fig. 4B (slice 2).
The spike above the prominent and intermediate stalks was a unique feature that formed a curved segment and spanned roughly a third of a circle, probably contacting one A-subunit and one B-subunit. We suggest that this spike at the top of V 1 was formed by part of the N-terminal region of the a-subunit, as discussed in detail below.
The V 0 Subcomplex-As in the case of the F-ATPase, the membrane-integrated V 0 subcomplex of the V-ATPase is thought to consist primarily of a ring of multiple copies of c-subunits. However, each of these V-ATPase c-subunits has four transmembrane helices, twice as many as a c-subunit in the F-ATPase. Sequence comparisons support the hypothesis that the c-subunits of both ATPase-types originated from a common ancestral gene that underwent gene duplication (7). So far, for F-ATPases the observed c-subunit stoichiometries range from 10 copies in yeast (11) to 14 copies in chloroplasts (43) corresponding to 20 -28 transmembrane helices. For V-ATPase, a similar number of transmembrane helices would be achieved by placing five to seven c-subunits into a ring. Recent biochemical studies support this stoichiometry (44). Thus, we assumed a high structural similarity between the c-subunits in F-and V-ATPase and placed a theoretical atomic model of the Escherichia coli F-ATPase c 12-ring 2 (45) into the V 0 density of our three-dimensional map (Fig. 6). Apparently, the V 0 subcomplex was sufficiently large to accommodate the c 12 -ring with additional space left in the plane of the membrane and above the c-ring.
Most of the extra space of V 0 in the plane of the membrane was probably occupied by the detergent micelle that shields the hydrophobic side chains of the amino acids, usually buried in the membrane, from the aqueous environment surrounding the isolated complex. This detergent belt probably accounted for the dense outer ring that is visible in cross-sections through the V 0 (Fig. 4B, slice 5), because its width (2.5 nm) is similar to the radius of a dodecyl maltoside micelle (ϳ3 nm) (46). In the plane of the membrane, the V 0 was slightly elongated (Fig. 4B, slice 5) which would be consistent with another subunit being attached to the periphery of the c-ring. This peripheral subunit was most likely the a-subunit, the only other subunit traversing the membrane and being predicted to form nine transmembrane helices in its C-terminal region (47). These helices might cover part of the c-ring with a single layer and would contribute a rim of about 1 nm width, which is consistent with the observed elongation in the plane of the membrane. The remaining space in V 0 above the modeled c-ring might be partly occupied by the d-subunit that is a hydrophilic, V 0 -associated subunit (48). Moreover, several stalk subunits (C and E; see below) have been found to form crosslinked products with the c-subunit (49,50) suggesting that they could also contribute to that density.
V 0 in the context of the holoenzyme as observed in our map resembles very much the structure of the isolated V 0 from clathrin-coated vesicles (18), with some discrepancies. In clathrin-coated vesicles the subunit Ac45 plugs the lumenal side of V 0 . This subunit and consequently the plug is not present in plant V-ATPases. At the cytosolic side some rearrangements of the smaller subunits occurred. These rearrangements were probably induced by the attachment of the V 1 to the V 0 and the formation of the stalk region in the holoenzyme.
The Stalk Region-Among the most frequently discussed questions about V-ATPase structure is what the number and positions of the stalks are. In our three-dimensional map, we identified one central and three peripheral stalks, where stalks were considered as connectors between V 1 and V 0 . Interestingly, none of the peripheral stalks seemed to connect directly to the periphery of the V 0 , although this had been observed in F-ATPases (51)(52)(53)(54). However, all of the peripheral stalks attached to the membrane integrated V 0 subcomplex close to the base of the central stalk.
Existing models of the stalk region propose a large range of subunit topologies, which is partly due to a lack of structural information. Therefore, we tried to combine the current knowledge about the organization of the V-ATPase stalk region with the observations from our three-dimensional map to find the most likely subunit topology. We assumed, from the current literature available, the stalk region to consist of the subunits C-H and the hydrophilic part of the a-subunit.
The central stalk was proposed to compose the D-or E-subunit or a D-E-subunit pair, respectively (55)(56)(57). However, recent cross-linking experiments have suggested that the Esubunit occupies a peripheral position (58). Such an exposed position agrees with several regulatory protein-protein interactions described for the E-subunit (59,60). Accordingly, the central stalk is probably formed by the D-subunit, which is essential for coupling proton transport and ATP hydrolysis (61), and the F-subunit, which is in close proximity to the 2 The atomic coordinates of the A1C12 subcomplex of F 1 F 0 -ATP synthase are available in the Research Collaboratory for Structural Bioinformatics Protein Data Bank under code 1C17 (45). D-subunit (62,63). Both subunits together could fit into the observed volume of the central stalk in our map.
The faint peripheral stalk showed some resemblance to the stator of various F-ATPases observed earlier by electron microscopy (52)(53)(54). In F-ATPases this stator is mainly formed by a dimer of b-subunits whose soluble parts are homologues to the G-subunit in V-ATPase (8). Therefore, we reasoned that the faint stalk in V-ATPases was probably partly formed by one or two (63) copies of the G-subunit and the E-subunit. The E-subunit was shown to be associated with the G-subunit (62)(63)(64) and is located at the periphery of V 1 (58).
The intermediate stalk matched unambiguously the recently published atomic model of the H-subunit of yeast V-ATPase 3 (65). Interestingly, the H-subunit has a characteristic twist that exactly followed the observed density of the intermediate stalk in our map (Fig. 6) and allowed us to determine the absolute handedness of the map. The fit showed that the large N-terminal domain of the H-subunit was attached to V 0 and the short C-terminal domain joined V 1 . The yeast H-subunit strikingly resembles the structure of importins (65), which have a shallow groove that binds to the nuclear localization signal, thereby mediating the recognition of diverse proteins (66). In the crystal structure of the yeast H-subunit, this groove is occupied by its own N-terminal end. It was reasoned that this N-terminal end might have an autoinhibitory function. However, the intermediate stalk in our three-dimensional map did not provide density for the N-terminal end of the H-subunit (Fig. 6), suggesting that the groove was accessible in the context of a complete, functional V-ATPase.
The volume of the prominent peripheral stalk was significantly larger than the one of the intermediate stalk, indicating that it was probably composed of more than one of the putative stalk subunits. Apart from the hydrophilic N-terminal half of the a-subunit, which was proposed to form a peripheral V 0 -V 1 connection (67), a good candidate for a second subunit would be the C-subunit. In yeast, during glucose-mediated disassembly this subunit is the only one that detaches from both the V 1 and V 0 (68), which would be consistent with it being in a peripheral, accessible position.
V-ATPase without AMP-PNP-The V-ATPase of K. daigremontiana without added AMP-PNP showed a significantly different conformation as the complex with added AMP-PNP (Fig.  5). The differences were characterized by the "disappearance" of the prominent and intermediate stalks and a different tilt of V 0 with respect to V 1 . Because the same protein sample was used for all of our investigations, it was plausible that the disappearance of the peripheral connections did not reflect a genuine loss of subunits but rather an increased flexibility in the absence of AMP-PNP. Indeed, cross-sections through the three-dimensional map of V-ATPase without AMP-PNP showed weak, diffuse densities in the regions where the prominent and intermediate stalks should have been (not shown), indicating a flexible, probably partly detached, arrangement of these elements. Only the faint stalk, which we think was related to the stator forming b-subunit in F-ATPases, remained clearly visible. In the biochemical experiment the purified intact V-ATPase complex (as it was used for both electron microscopic investigations) could not be eluted successfully from an ion-exchange column in the absence of AMP-PNP, thus giving a further indication of the increased lability of the complex.
The Model-In combining our three-dimensional maps with previous studies (5,16,49,50,58,(62)(63)(64)(65)68), we propose the following tentative model of the V-ATPase, as depicted in Fig.  7. With our results we were able to obtain a much closer perception of the overall architecture of the enzyme. The V 1 and V 0 subcomplexes were found to be connected by a central stalk and three peripheral stalks. Interestingly, the three peripheral stalks were each found to have individual sizes and shapes and were located in different positions with respect to the AB-subunits. Three A-and three B-subunits alternate to form most of the V 1 subcomplex. The A-subunits contain the characteristic knobs, which correlate to the N-terminal extensions that are non-homologous to the ␤-subunits in F-ATPase (5). We propose that the central shaft is composed of the D-and F-subunits (63) and is connected tightly to the membrane integrated c-ring of V 0 . The faint stalk consists of a G-E heterodimer (63) and is the only connection that is not destabilized in the absence of AMP-PNP highlighting its role as a stator similar to the one formed by the b-subunits in F-ATPases. The G-E heterodimer interacts tightly with other subunits in the stalk region, like d and C, which themselves contact the a-subunit. This whole assembly is in close proximity to the c-ring; however, it does not form any strong interactions. The intermediate stalk is composed of the H-subunit, as suggested by fitting the atomic model of the yeast H-subunit (65) into our three-dimensional map. The amphiphilic a-subunit is anchored with its hydrophobic C-terminal half in the membrane proximal to the c-ring. With its hydrophilic N-terminal region the a-subunit accounts for part of the prominent stalk and stretches all the way to the top of V 1 . We suggest that the spike at the top of V 1 is formed by the N-terminal part of the a-subunit contacting an AB-heterodimer in V 1. Thus bridging V 1 and V 0 , the a-subunit could play an important role in the association of both subcomplexes. The rest of the prominent stalk is formed by the C-subunit, which easily detaches from V 1 and V 0 during disassembly (68), supporting the idea of its peripheral, exposed location. The d-subunit and parts of the a-, C-, and E-subunits surround the central shaft. This assembly possibly provides a static bearing for the central shaft and also forms the putative contact means for the peripheral stalks to the membrane anchor of the a-subunit.
Beyond fitting the previous observations as described above, this model is consistent with other findings for V-ATPases. First, it agrees with cross-linking data for certain V-ATPases (Fig. 7, pairs of triangles) (16,49,62,63) as well as with co-immunoprecipitation experiments (50). Second, it is consistent with the pre-assembly complexes (62).
Functional Implications-Based on the structural model that we proposed for the V-ATPase, we attempted to draw conclusions regarding the functionality of the complex. Because of several subunit homologies and the similar architectures of F-and V-ATPases, it is thought that V-ATPases may function through a rotational mechanism similar to that of F-ATPases (69). According to this hypothesis and our V-ATPase model, ATP hydrolysis at the catalytic A-subunits would cause rotation of the central shaft formed by the D-F-subunit pair. This rotation would be transmitted to the c-ring, which would rotate relative to the a-subunit, pumping protons along the subunit interface into the lumen (70). In F-ATPases, the co-rotation of the other subunits is prevented by a single thin peripheral connection between F 1 and F 0 . According to our model, in V-ATPases, three peripheral stalks would be connected to the membrane-integrated part of the a-subunit, forming a whole network of static connections between V 1 and V 0 . During rotation, this network could dissipate the forces efficiently across the whole V 1 subcomplex. Furthermore, it is conceivable that such a network might form a bearing for the central rotating shaft in the sense that it would have not only one anchoring point in the V 1 (analogous to F-ATPases) but also a second fixed point close to the membrane (Fig. 7, see color code). Thus, the whole assembly might be more robust. On the other hand, it might also be more efficiently shut down by the removal of a single subunit from this static network of interacting subunits. This loss could displace or disrupt the bearing and thereby impair smooth rotation of the central shaft. In the three-dimensional map of the V-ATPase without the substrate analogue AMP-PNP, i.e. at conditions mimicking ATP deprivation, we probably see such an inactive, unstable conformation with a displaced bearing causing the tilt of V 0 with respect to V 1 . The prominent and intermediate stalks are probably already partly detached, which could be the first step in a controlled disassembly of the complex under depleting ATP levels.