A Unique Resting Position of the ATP-synthase from Chloroplasts*

The chloroplast ATP-synthase catalyzes ATP synthesis coupled to transmembrane proton transport. The enzyme consists of two parts, a membrane-embedded F0part and an extrinsic F1 part, which are linked by two connectors. One of these rotates during catalysis and the other remains static. Although the atomic structures of various sub-complexes and individual subunits have been reported, only limited structural information on the complex, as a whole, is available. In particular, information on the static connector is lacking. We contribute a three-dimensional map at about 20-Å resolution, derived from electron cryomicroscopy of enzymes embedded in vitrified buffer followed by single particle image analysis. In the three-dimensional map both connectors, between the F1 part and the F0 part, are clearly visible. The static connector is tightly attached to an α subunit and faces the side of the neighboring β subunit. The three-dimensional map provides a scaffold for fitting in the known atomic structures of various subunits and sub-complexes, and suggests that the oxidized, non-activated ATP-synthase from chloroplasts adopts a unique resting position.

F-type ATP-synthases are found in bacteria, chloroplasts, and mitochondria. They catalyze ATP synthesis/hydrolysis coupled to transmembrane proton transport. All ATP-synthases consist of two parts, a membrane-embedded F 0 part and a membrane extrinsic F 1 part, which are connected by a thinner connecting region.
In the ATP-synthase from chloroplasts, the F 1 part is composed of five different subunits with the stoichiometry (␣␤) 3 ␥␦⑀. The ␣ and ␤ subunits surround the central ␥ subunit, which, together with the ⑀ subunit, forms the central stalk. The F 0 part is involved in proton translocation and consists of four different subunits I, II, III, and IV. Subunit III is the major component, 14 copies of the subunit III form a ring (1). A segment of this ring, together with subunit IV, forms the proton channel (2). Subunits I and II have an amphiphilic character, with a single transmembrane helix serving as a membrane anchor. The homologous b subunits in Escherichia coli form a dimer with inner-dimer contacts in the transmembrane Nterminal region (3) as well as between amino acids 53 and 122 (4). The b-dimer connects the F 0 part to the F 1 part by interacting with the ␦ subunit (5) at the top of F 1 (6) and forms a peripheral stalk.
According to a current functional model for F-type ATPases (2,7), the subunit III-ring in the membrane and the ␥ and ⑀ subunit in the stalk region form a rotor, which rotates during proton translocation. This rotation induces conformational changes in the catalytic nucleotide binding sites during ATP synthesis/hydrolysis. A second static connection (stator) formed by subunits I and II, prevents co-rotation of the (␣␤) 3 core complex. The orientation of the rotor determines the occupancy of the three catalytic nucleotide binding sites. There are three equivalent orientations of the central rotor relative to the individual catalytic binding sites as shown by micro videograms (8). Because rotational catalysis requires functional equivalence of the catalytic binding sites, all three conformations should occur with the same probability. In the ATP-synthases from chloroplasts, this type of rotational multisite catalysis requires activation of the enzyme by a transmembrane potential difference of protons, ⌬ Hϩ (9,10). Up to now it is not understood if the inactivate chloroplast ATP-synthase also adopts each of the three conformations with the same probability or takes up a unique resting conformation.
Although various high resolution structures of sub-complexes from different F-type ATPases are known (for example, bovine MF 1 (␣␤) 3 ␥␦⑀ (11); chloroplast CF 1 (␣␤) 3 ␥ (12); E. coli EF 1 (␣␤) 3 ␥ (13); yeast F 1 c 10 (␣␤) 3 ␥␦c 10 (14)), none of these sub-complexes include homologues of the smaller subunits that form the stator in the chloroplast ATP-synthase. Because the small static subunits are missing, which are asymmetrically attached to the (␣␤) 3 core, the individual nucleotide binding sites are only defined in respect to the rotor and not in respect to the stator, which is insufficient to distinguish between the three possible conformations. Information on this issue can be obtained by electron microscopy and image reconstruction of a complete ATP-synthase. Up to now, projection maps of the ATP-synthase from mitochondria (15), chloroplasts (16), and E. coli (17), which show rotor and stator in the connecting region, have been obtained. However, two-dimensional projection maps alone, without prior knowledge of the three-dimensional shape, are inadequate to discern between different conformational states and projections of the same object in different directions. Such a discrimination can only be done, if the three-dimensional volumes are also known. Nevertheless, three-dimensional maps at 30-to 35-Å resolution, of the negatively stained E. coli (18) and chloroplast enzymes (19), show the stator either weakly or not at all. Therefore, three-dimensional maps are still insufficient to decide whether or not the complex adopts different conformations. Here we present a three-dimensional map of the ATP-synthase from chloroplasts embedded in vitrified buffer, in which the stator and rotor are clearly visible and only a single conformation, as expected for a unique resting position, is observed. mM KCl, 150 mM NaCl, 10 mM NaH 2 PO 4 , pH 7.2. Grids coated with a perforated carbon film were glow-discharged and used within 30 min. After adding 2 mM AMP-PNP, 1 2 l of the sample was applied to the pretreated grid. The grid was mounted in a modified controlled environment freezing apparatus (20). After blotting for 15 s with filter paper (Whatman No. 1), the samples were plunged into liquid ethane.
Electron Microscopy and Image Processing-The vitrified samples were transferred with a Gatan 626 cryo-holder into a Philips CM-200-FEG electron microscope. The microscope was operated under low dose conditions at 200 kV accelerating voltage. Micrographs were taken on Kodak SO-163 film, at a nominal magnification of ϫ50,000 with an underfocus of 3-5.1 m. The micrographs were developed for 10 min in full-strength Kodak D-19 developer at room temperature.
For image processing, suitable micrographs were scanned with a Zeiss SCAI scanner with a pixel size of 14 m corresponding to 2.8 Å at specimen level. The particle images were selected interactively and were boxed off from the micrographs using the MRC image processing programs (21). Further image processing was carried out using the IMAGIC-5 software package (22) as described previously (23) except for the following modifications. The phases of the particle images were corrected for the contrast transfer function. The particle images were band-pass filtered, including information between 1/16 and 1/110 Å Ϫ1 .
The band-pass filtering and the contrast transfer function of the microscope lead to an underestimation of the low resolution amplitudes, causing dark fringes and an overestimation of holes in the complex. To minimize this effect in the final three-dimensional map, the low resolution amplitudes of the raw three-dimensional map (A raw ) were scaled relative to the amplitudes of a binarized three-dimensional map (A bin ), in which gray values inside the particles were set to 1 and outside to 0. The threshold for binarizing was chosen so that the total number of voxels set to 1 accounted for the expected molecular mass of the ATPsynthase surrounded by the detergent micelle (ϳ750 kDa). Raw map and binarized map were Fourier-transformed. For each band of spatial frequencies (R 1 Յ R Յ R 2 ) the ratio of the average amplitudes A was calculated, The ratio varied at low spatial frequencies R and reached constant values for a small band of frequencies at about RR ϭ 1/45 Å Ϫ1 . At frequencies below this band, the ratio was used as a scaling factor and at frequencies above this band the constant ratio in the band was used for scaling of the amplitudes A raw in the raw map as follows. For band Ͻ RR, and for band Ն RR, The scaled three-dimensional Fourier transform was transformed to real space to give the final three-dimensional map (Fig. 5), which showed the same gross features as the raw map but smaller holes and reduced dark fringes, which facilitated fitting of the high resolution structures. In this approach, the absolute threshold for binarization was not too critical, however, the absolute size and shape of the binarized reference map were crucial. Therefore, scaling the electron microscopic reconstructions with atomic models of smaller sub-complexes, or models lacking the detergent micelle, yields poorer results than using the binarized reconstruction for scaling (data not shown).
To demonstrate the effect of the scaling approach, a model experiment was performed: from the yeast F 1 c 10 sub-complex (Protein Data Bank 1QO1 (14)) a density map at 20-Å resolution was calculated (Fig.  1A). This map was convoluted with the contrast transfer function as calculated for a defocus of 3000 nm using a Philips CM-200-FEG operating at 200 kV. The resulting map showed dark fringes and holes at places where holes were not observed in the original map. After correcting phases for the contrast transfer function, the holes were observed at the correct positions (Fig. 1C). However, they were darker and larger than in the original data. In addition, dark fringes surrounded the particle. Only little improvement was observed, when various reconstructions were calculated using different defocus values and had phases corrected for the contrast transfer functions were combined (Fig.  1D). We assume, that the dark fringes and holes are predominantly effects of uncorrected amplitudes at low spatial frequencies. In the "true" reconstruction, we do not expect any negative values (dark fringes and dark holes). This situation was approximated by a binarized map (Fig. 1E), the threshold of which was chosen so the total number of voxels set to 1 accounted for the expected molecular mass. The binarized map gave an estimate for the average amplitudes of the low spatial frequencies of a reconstruction without dark fringes. The map was used for scaling the amplitudes of the low spatial frequencies of the combined reconstruction ( Fig. 1C) as described above. Indeed, scaling reduced the dark fringes and decreased the size of the holes (Fig. 1F). Although the scaled map (Fig. 1F) was still not identical to the starting map (Fig. 1A), it was more similar in its overall appearance than the map (Fig. 1D), where only phases were corrected for the contrast transfer function.
To estimate the resolution of the three-dimensional map, two independent maps of half of the class averages were calculated, and their Fourier shell correlation was determined. The atomic models were fitted manually to the three-dimensional map using the program O (24). For the difference map, the fitted atomic models were filtered to the resolution of the three-dimensional reconstruction. After binarizing both maps with a threshold value of 0.3, a difference volume was calculated. Fig. 2 shows a typical micrograph of the ATP-synthase from chloroplasts embedded in a thin layer of vitrified buffer. The protein formed a uniform monodisperse particle distribution. For image processing about 10,000 particle images were selected from 150 micrographs. The particle images were aligned, classified according to their similarity, and averaged. The class averages (Fig. 3A) showed projections of particles with a larger and a smaller domain, which were linked by one or two connectors. The larger  (14) (1QO1), a density map at 20-Å resolution was calculated. A central slice parallel to the long molecule axis is shown. B, the density determined in A was modulated by a contrast transfer function calculated for a nominal defocus of 3000 nm. C, the phase information of the density shown in B was corrected for the contrast transfer function. D, four maps similar to the one in C, but calculated and phase-corrected for different defocus values (3000, 3300, 3600, and 3900 nm) were combined. E, the map shown in D was binarized with a suitable threshold, describing a volume that would approximately fit the molecular mass of the complex. F, the amplitude information of the binarized map shown in E was used to scale the phase-corrected image shown in D as described.

Electron Microscopy and Image Processing-
domain corresponded to the F 1 part and the smaller domain to the F 0 part, shielded by the detergent micelle.
The class averages were combined into a three-dimensional map by exact filtered back-projection reconstruction (25) (Fig.  3B). All class averages could be matched by projections of the three-dimensional map (Fig. 3C). The Euler angles of the class averages covered most of the asymmetric unit (Fig. 4), leaving only small areas unaccounted for. To estimate the overall resolution, we calculated the Fourier shell correlation between two maps, each representing half of the data. The correlation dropped to 0.5 at (1/21) Å Ϫ1 and cut the three-times noise correlation curve at (1/18) Å Ϫ1 , indicating an overall resolution of about 20 Å.
The F 1 Part-In our three-dimensional map the F 1 part was formed by six elongated elements arranged roughly hexagonally around a central rod (see map in Fig. 5B, labeled 1-6). These elements could also be recognized in the surface representations (Fig. 3B) and most likely correspond to the ␣ and ␤ subunits. To decide which of the elements were ␣ and which were ␤ subunits, we compared this region of the map to the atomic structures of different F 1 sub-complexes (PS3-(␣␤) 3 (26); chloroplasts-(␣␤) 3 (12); mitochondrial-(␣␤) 3 ␥ (27)). Although these sub-complexes had different nucleotide occupancies and varied significantly in the region nearest to the connecting region, the ␣ subunits always extended further from the center than the ␤ subunits. We observed a similar pattern in our map (Fig. 5B, 2), where the elements 1, 3, and 5 extended further from the center than 2, 4, and 6. For this reason, we conclude that the ␣ subunits form elements 1, 3, and 5 and that the ␤ subunits are represented by elements 2, 4, and 6. The ␣␤dimers in our map were not related by strict 3-fold symmetry. Therefore, we fitted the asymmetric (␣␤) 3 sub-complex of the mitochondrial F 1 (PDB 1E79 (11)), rather than the structure of the symmetric (␣␤) 3 sub-complex of the F 1 , from chloroplasts (PDB 1FX0 (12)), into our three-dimensional map (Fig. 6). The match between the mitochondrial F 1 sub-complex and our map was best when the ␤ subunit with the empty nucleotide binding site was superimposed to the elongated element 6 in our map. We propose that the rod in the center of the six elongated elements (Fig. 5A, elements 6 and 7, and 5B, element 2) correspond to the two long helices of the ␥ subunit in the atomic structure of the mitochondrial F 1 part.
Fitting the ␣ and ␤ subunits and the two long helices of the ␥ subunit to the map, as described above, accounted for most of the observed density in the F 1 part. In the F 1 part only a peanut-shaped density at the top and an elongated density, running parallel to the ␣ subunit corresponding to element 1, was unaccounted. The peanut-shaped feature at the top of F 1 part consisted of two domains (Fig. 5B, 1). One of the domains was located in the center of a crown-like region at the top of the ␣ and ␤ subunits, formed by the ␤-barrels, as seen in the x-ray structure (PDB 1E79 (11)). The other domain created a bridge to the ␣ subunit in element 1. We assume that most of the peanut-shaped feature was occupied by subunit ␦, because its central domain coincided with the area that is recognized by a monoclonal antibody against the C-terminal region of subunit ␦ in E. coli (6). Consequently, in our three-dimensional map the N-terminal domain of subunit ␦ must occupy the remaining peripheral domain into which the structure of the N-terminal domain of E. coli ␦ subunit (28) fitted nicely (Fig. 6).
The Connecting Region-The F 1 and F 0 part were joined by two connecting elements, a thin peripheral connector and one that was larger, and more centrally localized. The latter was formed by subunits ␥ and ⑀. In the past, two different conformations have been observed for the central connection. One was derived from crystals of a ␥Ј⑀ sub-complex of the E. coli enzyme (29) and the other from a mitochondrial F 1 (11), and it has been argued that both conformations occur during catalysis (13). The conformations vary in the orientation of the ⑀ subunit (or related ␦ subunit in mitochondria F 1 ) relative to the ␥ subunit with the ⑀ subunit rotating 81°and undergoing a net translation of 23 Å. We fitted both types of central connections into our map (not shown) and found the conformation observed in the mitochondrial F 1 was the best match (Fig. 6).
In contrast to an earlier reconstruction of the negatively stained ATP-synthase from E. coli, where two peripheral connectors have been identified (18), only one peripheral connection was observed. Whether or not the two connections observed in the ATP-synthase of E. coli are a specialty of the bacterial enzyme or some artifact probably caused by the staining procedure remains open.
In the chloroplast ATP-synthase the thinner, peripheral connection was attached at the periphery of the ␣ subunit of the F 1 part (Fig. 3B, element 1). The connection was wider at the sites of interaction with the F 0 and F 1 part and thinner between these points. It remains uncertain if this slimming is a genuine feature of the connection or only appears in our model due to increased flexibility in this area. Whichever the case, we think that this peripheral link is the stator, formed by subunits I and II, in the ATP-synthase from chloroplasts. For the related b-subunits of E. coli, secondary structure determination indicates a helix content of 80% (30), suggesting that the observed stator might contain only a little more than two extended helices, which is also supported by x-ray analysis (31). In the F 1 part, the stator does not account for an individual feature and is presumably tightly attached to the ␣ subunit (element 1), pointing toward the ␤ subunit (element 6).
To follow the path of the stator through the F 1 structure, we calculated the difference between the volume occupied by the manually fitted x-ray structures and the volume occupied by our three-dimensional map (Fig. 7A). In the F 1 part and in the connecting region, the fitted x-ray structures did not account for subunits I and II, the C-terminal domain of the ␦ subunit, and the disordered N-terminal residues of the ␣ and ␤ subunits. According to the difference, the stator contacted the ␣ subunit (Fig. 5B, element 1), involved in forming the tight catalytic nucleotide binding site in the center, at the "bottom." Then the stator lined the side of the ␣ subunit, which faced the ␤ subunit that forms part of the empty catalytic nucleotide binding site (Fig. 5B, element 6). At the top of the F 1 part, the stator ended with a contact to the ␦ subunit at the uppermost part of the N-terminal domain. The observed position of the stator is in agreement with earlier cross-linking data for the stator-forming b and ␣ subunits in E. coli (32,33).
The F 0 Part-The F 0 part had an elliptical cross-section in the plane of the membrane (Fig. 5B, 4). An outer oval belt, probably formed by detergent, surrounded an inner circular ring of similar diameter, as the subunit III complex observed by atomic force microscopy (1). Therefore, we conclude that the subunit III complex forms the inner ring. To get some impression of its space requirements in the context of our map, we generated a ring of 14 copies of the related E. coli c-subunit. There are two structures available from NMR measurements that vary in the deprotonation of Asp-61, which in E. coli has a pK a value of 7.1 (34). Accordingly, we generated a ring of the protonated (35) and the deprotonated (36) conformation of the related c-subunit (36) using MolMol (37) and placed both in the observed density. The manual placement was guided by the hole in the center of the F 0 part in our map, which we superimposed on the central channel of the modeled ring. A belt of unaccounted density surrounded both modeled rings. This belt was smaller due to the modeled ring of the deprotonated conformation. However, at the level of observed detail, both fits were somewhat arbitrary. In Fig. 6 the modeled ring of the deprotonated form is shown, because at pH 7.2 this conformation should be adopted by the majority subunits. The ring was sealed from one side by the central stalk and from the other side by a plug. The nature of the plug was unclear, but it was probably formed by either detergent or by tightly bound lipids. The latter was observed in two-dimensional crystals of the related c complex of I. tartaricus (38). When the volume occupied by the modeled ring was subtracted from the observed density, a belt with a bulge was left at the side where the stator emerged. We conclude that the membrane domains of subunits I, II, and IV formed this bulge. DISCUSSION ATP-synthases work with a rotational mechanism, which requires the catalytic nucleotide binding sites to be functionally equivalent. Accordingly, there should be three equivalent conformations in which the catalytic nucleotide binding sites have the same overall occupancy, but vary in the occupancy of the individual binding sites as depicted in Fig. 8. In all three conformations, the core of (␣␤) 3 ␥⑀ III 14 should have the same structural organization. Only the presence of the small peripheral subunits (I, II, IV, and ␦) provides the means to distinguish between the three conformations. In the three conformations, the small peripheral subunits also have to adapt to spatially different environments, because the invariable core does not have a strict 3-fold symmetry. In our samples, we have the small static subunits present, and therefore, we are in principle, able to discern between the three conformations. Nevertheless, at the given resolution, our data does not show any indication of the presence of multiple conformational states. All observed projections can be matched by projections calculated from the same three-dimensional map, which is one indication that a conformationally homogeneous particle population was explored (19). We can also exclude the possibility that particle images representing different conformations were accidentally combined in the three-dimensional map for the following reason: if particles with the described variation in conformations would have been averaged in our three-dimensional map, we would either expect the occurrence of three weak peripheral connectors or of a central connector with a 3-fold symmetryaxis approximately parallel to the long molecule axis and/or with a fuzzy outline. However, we observe only a single peripheral connector and a central stalk with a well-defined asymmetric shape. This stalk accommodates the atomic model of the mitochondrial central stalk in only one orientation. The two alternative orientations of the central stalk, which could be expected for the other two equivalent conformations (Fig. 8), do not match the observed density (not shown). Therefore, we conclude that the majority of the isolated, inactive ATP-synthases adopts a single unique resting position (equivalent to the one depicted in the center of Fig. 8). At present we do not know if this is a specialty of the ATP-synthase from chloroplasts, which requires activation by ⌬ Hϩ for rotational multisite catalysis, or whether it is a genuine feature for all ATP-synthases.
A prerequisite for a unique resting position would be functional none-equivalent catalytic nucleotide binding sites. Somehow, the ATP-synthase must "know" in which position to stop when catalysis halts, due to inactivation by an insufficient ⌬ Hϩ . This could be achieved by a change of the binding constants for nucleotides in one of the three catalytic nucleotide binding sites. We think that in the ATP-synthase from chloroplasts, the change in binding constants might be realized by the stator, which is anchored in the membrane on one hand, where it could sense changes in the ⌬ Hϩ , and on the other hand is also tightly attached to one of the ␣ subunits where it could influence the properties of the adjacent nucleotide binding sites. Therefore, the stator could "communicate" changes in ⌬ Hϩ to the binding sites by dislocating the ␣ subunit slightly and thus changing the binding constant of the adjacent nucleotide binding sites.
According to our fit, in the resting position, the stator is attached to the tight catalytic nucleotide binding site (Fig. 8,  center), which carries an ADP in the dicyclohexylcarbodiimide (DCCD) inhibited mitochondrial F 1 sub-complex (11). An ADP on a tight catalytic nucleotide binding site, which cannot be removed, is also found in the purified ATP-synthase from chloroplasts (39). It is conceivable that this peculiar property of one of the binding sites is caused by the stator, which in the resting position could force the adjacent site in a closed conformation and thus trapping the bound ADP. Because rotational catalysis requires functional equivalence of the catalytic nucleotide binding sites, we speculate that activation by ⌬ Hϩ , shifts the stator, which in turn relocates the attached ␣ subunit rendering the binding sites functionally equivalent and thus enabling multisite catalysis again. FIG. 7. Slices through a difference map between the volume derived from electron microscopic data and the volume occupied by the fitted atomic structures (Fig. 6), which was reduced to the same resolution as the electron microscopic map. For comparison, both maps were set to one inside the volume and to zero outside the volume. The slices are gray where the map and the fitted structure occupied the same volume. Additional density of the fitted x-ray structures is white, and additional density in our map is black. A, slices from the center of the complex parallel to the long axis; B, slices perpendicular to the long axis, which are equivalent to those in Fig. 5B but rotated in plane to match those in A. The scale bar corresponds to 100 Å.
FIG. 8. Schematic representations of the three-functionally equivalent conformations of the ATP-synthase. The ATP-synthase is depicted from the top, perpendicular to the plane of membrane. The conformationally invariant core is marked by a black outline and consists of ␣ subunits (white), ␤ subunits (medium gray), the subunit III ring (light gray), and the ␥ subunit (black). The subunits forming the catalytic binding sites are labeled according to the nomenclature introduced by Abrahams et al. (27) (E for empty, D for ADP, and T for ATP or ADP). This naming convention does not necessarily reflect different occupancies in the catalytic binding sites but is also used in other structures (40) where the identity of the sites is determined by the orientation of the ␥ subunit. At the periphery of F 1 the stator (outlined by the dashed lines) is formed by subunits ␦, I, II, and IV and adapts to the spatially different environments of the three conformations. The putative, unique resting position is equivalent to the conformation shown in the center.