Crystal Structure of the Mg·ADP-inhibited State of the Yeast F1c10-ATP Synthase*

The F1c10 subcomplex of the yeast F1F0-ATP synthase includes the membrane rotor part c10-ring linked to a catalytic head, (αβ)3, by a central stalk, γδϵ. The Saccharomyces cerevisiae yF1c10·ADP subcomplex was crystallized in the presence of Mg·ADP, dicyclohexylcarbodiimide (DCCD), and azide. The structure was solved by molecular replacement using a high resolution model of the yeast F1 and a bacterial c-ring model with 10 copies of the c-subunit. The structure refined to 3.43-Å resolution displays new features compared with the original yF1c10 and with the yF1 inhibited by adenylyl imidodiphosphate (AMP-PNP) (yF1(I–III)). An ADP molecule was bound in both βDP and βTP catalytic sites. The αDP-βDP pair is slightly open and resembles the novel conformation identified in yF1, whereas the αTP-βTP pair is very closed and resembles more a DP pair. yF1c10·ADP provides a model of a new Mg·ADP-inhibited state of the yeast F1. As for the original yF1 and yF1c10 structures, the foot of the central stalk is rotated by ∼40 ° with respect to bovine structures. The assembly of the F1 central stalk with the F0 c-ring rotor is mainly provided by electrostatic interactions. On the rotor ring, the essential cGlu59 carboxylate group is surrounded by hydrophobic residues and is not involved in hydrogen bonding.

The F 1 F 0 -ATP synthase is an essential membrane rotary motor that uses transmembrane electrochemical ion gradients to synthesize ATP. To date, only cryo-electron microscopy has provided a complete view of the yeast ATP synthase (1). The structure of the Saccharomyces cerevisiae F 1 c 10 -ATP synthase subcomplex provided the first model at 3.9-Å resolution of the molecular assembly between the membrane rotor ring and the central stalk of a F 1 F 0 -ATP synthase (2). Both adenylyl imidodiphosphate (AMP-PNP) 3 and ADP were added to the crystallization medium, a structure that will be referred to as yF 1 c 10 . It was solved by molecular replacement using the C␣ coordi-nates of the crystal structures of the AMP-PNP-inhibited bovine F 1 -ATPase containing (␣␤) 3 ␥-subunits, which is considered as the bovine reference structure bF 1 ⅐AMP-PNP (3), and of the Escherichia coli ⑀-subunit, the homologous of the mitochondrial ␦-subunit (4), as models. A decameric ring of c-subunits (also named subunit 9 in yeast) was afterward found in the residual electron densities and the NMR solution structure of the E. coli c-monomer (5), which shares only 18% of identity with the yeast subunit, was used to build the c 10 -ring in the maps. The smallest subunit (⑀-subunit) was indistinguishable in electron densities but was present as shown by a SDS-PAGE analysis of the crystal (2). Unfortunately and unexpectedly, the peripheral stalk of the enzyme dissociated during the crystallization process. Owing to the lack of crystals that diffracted at high enough resolution and lack of crystal structures of yeast subunits or subcomplexes, refinement of the yF 1 c 10 was impossible a few years ago. Although the model deposited in the Protein Data Bank is a mix of x-ray and NMR unrefined C␣ atom coordinates of bovine and E. coli homologous proteins, the structure was fitted successfully in the envelope of the electron microscopy map of F 1 F 0 -ATP synthase to obtain the envelope of the peripheral stalk (1,6).
The first F 1 -ATPase x-ray structure (3) supports Boyer's binding change mechanism for catalysis (7). It was proposed that (i) the empty and open catalytic site ␤ E was the open site with low affinity for nucleotides, (ii) the ␤ TP site filled by ATP (or AMP-PNP) was the loose conformation, and (iii) the ␤ DP site filled by ADP was the tight conformation where synthesis occurs (8). The opening of catalytic sites can be estimated by means of the buried area at the ␣-␤ interface. The loose and tight conformations were further identified in the structures of the ground state (9) and of an intermediate state analog (10), but they remained debatable because the ␤ TP site has also been proposed as the high affinity catalytic site (11).
The inconsistent presence of ADP in the ␤ DP site of the bovine reference structure (3) has been explained by the Mg⅐ADP-inhibited state of the enzyme. Dicyclohexylcarbodiimide (DCCD) inhibits both the H ϩ transport and the ATPase activity (12). It has been used to block the enzyme for crystallization (13). It reacts preferentially with a single c-subunit of the F 0 rather than a single ␤-subunit of the F 1 . During the past decade, the detailed structures of the bovine F 1 -sector inhibited by Mg⅐ADP (bF 1 ⅐ADP, Protein Data Bank (PDB) code 1w0k) (10) or covalently inhibited on ␤ DP Glu 199 by DCCD (bF 1 ⅐DCCD) (13) and of the yeast enzyme in the presence of AMP-PNP and ADP at 2.8-Å resolution (yF 1 ) (14) have been refined. There are three independent copies of the enzyme in the yF 1 asymmetric unit, yF 1 (I) and yF 1 (III) were similar, and yF 1 (II) was different. In the nucleotide-free yeast F 1 structure (yF 1 apo(I-III)), the DP and TP pairs adopt "closed" conformations despite the absence of bound nucleotides (15).
Concerning the membrane rotors, the structure solved at 2.4-Å resolution of the Na ϩ -F-ATPase c-ring from Ilyobacter tartaricus contains 11 monomers (16). The K-subunit (NtpK) of the Na ϩ -V-ATPase of Enterococcus hirae with four transmembrane helices forms decameric rings (17). In the E. hirae DCCD-inhibited K 10 -ring structure (PDB code 2DB4), dicyclohexyl-N-acylurea (DCU) was found to be linked to Glu 139 equivalent to the yeast cGlu 59 , with DCU groups protruding outside the ring.
Here, we report the x-ray structure of the F 1 c 10 subcomplex from S. cerevisiae inhibited by Mg⅐ADP and DCCD (yF 1 c 10 ⅐ADP) refined to a resolution of 3.43 Å. However, DCU modifications on both sectors could not be ascertained on the electron density maps. The ␤ DP and ␤ TP catalytic sites adopt closed conformations and are both occupied by ADP. This structure represents the first view of the Mg⅐ADP-inhibited state of the yeast enzyme. The structure of the F 1 -sector was compared with the structures of the bovine enzyme and with all independent copies found in the yeast F 1 with bound AMP-PNP and without bound nucleotides. Compared with the bovine enzyme, the yeast central stalk is twisted. In the F 0 rotor ring, the essential cGlu 59 carboxylate group is only surrounded by apolar residues. Its closest hydrogen bond acceptor, the cLeu 57 carbonyl oxygen of the adjacent c-subunit, is too far away to make a direct hydrogen bond. The proton binding has specific features compared with the bacterial Na ϩ transporting (16) or the cyanobacterial (19) and chloroplastic (20) H ϩ -transporting F-type ATP synthase rotor structures.

EXPERIMENTAL PROCEDURES
yF 1 F 0 Enzyme Purification and yF 1 c 10 Crystallization-All purification steps were performed as described previously (21) except that the nickel-nitrilotriacetic acid elution fraction was concentrated and subjected to a purification step on a Superdex 200 gel filtration column (HR 10/30). The gel filtration buffer contained 0.64 mM n-dodecyl ␤-D-maltoside, 100 mM NaCl, 2 mM MgCl 2 , 25 mM trehalose, 0.5 mM EDTA, 3 mM sodium azide, 20 mM Tris-HCl, pH 8.0. Fractions corresponding to the major peak were analyzed by SDS-PAGE electrophoresis. These fractions contained all the known subunits of the monomeric F 1 F 0 (subunits ␣, ␤, ␥, 4, 6, OSCP, d, ␦, h, f, ⑀, i, 8, and 9) (supplemental Fig. S1). Assembly states were asserted by nondenaturing Blue native PAGE. The purified F 1 F 0 -ATP synthase is active, migrated as a fully assembled monomeric form (21). Neither ADP nor ATP was added during the purification. The protein solution was concentrated with a Centricon concentrator unit (YM-100, Millipore). The final enzyme concentration (9 mg/ml) was assessed by the Lowry method. The enzyme was incubated for 1 h at 20°C with 660 M ADP and then inhibited for 1 h with 100 M DCCD in the presence of 2.5 mM DTT and 0.5 mM PMSF.
Crystallizations were carried out at 20°C, by the sitting drop vapor diffusion method, after mixing equal volumes of protein and reservoir solutions (100 mM NaCl, 12% PEG MME 5000, 3 mM sodium azide, 100 mM Hepes, pH 7.5). Crystals grew in a few days as flattened rods of typically 150 to 400 m in length and 20 m in thickness. Crystals were cryo-protected with mother liquor containing 15% (v/v) glycerol and flash-frozen in liquid nitrogen.
Structure Determination-X-ray diffraction data were collected on beamline ID29 (European Synchrotron Radiation Facility). Many crystals mounted from the same drop diffracted at an ϳ4-Å resolution, with the best one diffracting at 3.2-Å resolution. Severe anisotropic diffraction and radiation damage limited the quality of the data sets. The best data set (77% complete at 3.43 Å) was merged with one of lower resolution at 3.67 Å. All data were processed with IMOSFLM (22) and the CCP4 suite (23). The cell parameters of yF 1 c 10 ⅐ADP are close to those of the original yF 1 c 10 structure (2). The current structure was solved by molecular replacement with the PHASER program (24). The three independent conformations yF 1 (I-III) in which nucleotides were removed (PDB code 2HLD) were used successively as search model (14). The solutions obtained with yF 1 (I) and yF 1 (III) were better (Z-score, 50; R work , 0.44) than with yF 1 (II) (Z-score, 30; R work , 0.47). Because the central stalk of yF 1 (I) was more complete than that of yF 1 (III), the yF 1 (I) solution was retained. The asymmetric unit contains one molecule resulting in a solvent content of 67% (V m ϭ 3.7 Å 3 /Da). Rigid body refinement of individual subunits resulted in R work and R free equal to 0.38 using 3.7 Å data. In the F 1 -sector, the electron density maps are mainly continuous along the backbones and are good enough to rule out the possibility of statistical disorder.
A c-monomer template was built using the coordinates of the I. tartaricus rotor c 11 -ring (PDB code 1YCE) (16) by removing six and seven residues at the N and C termini, respectively. To ascertain the number of c-subunits in the crystal, ring models containing from 8 to 12 copies (N) of the c-subunit were constructed as described below. The main axis of the I. tartaricus ring was oriented along the Z-axis and then rotated to place the center of the template on the X-axis. Rings were finally generated by successive (NϪ1) rotations of (2/N) degrees of the template about the Z-axis. Optimization of the ring radii was based on an analysis of intersubunit contacts. With the fixed current F 1 model as input, molecular replacement searches were run using each ring model with PHASER (24). The best solution was clearly obtained for the decameric ring, i.e. a Z-score of 24, instead of a Z-score Ͻ15 with the other ring models. Rigid body and overall temperature factor refinement of individual subunits resulted in R work and R free equal to 0.33. In the c-ring, the side chains were added to the model with the help of the I. tartaricus model (32% of identity, 59% of similarity with S. cerevisiae). The backbone electron density was better defined in a third and in the lower half of the ring where the sequence could be assigned with certainty. When no side chain information was present in the template or no density visible in the map, the most favored rotamer with acceptable van der Waals exclusion test was added to the model. Strong noncrystallographic symmetry restraints were applied separately on helices H1 and H2 but not on the loop. The chains of the c-ring were labeled from J to S in the direction of the hydrolysis (25), starting with the chain having the lowest overall temperature factor. The assignment was checked on 2F o Ϫ F c unaveraged and Non-crystallographic Symmetry-averaged (calculated using the DM program) electron density maps on chains with the lowest temperature factors ( Fig. 1) using COOT (26).
Recently, modifications of standard refinement protocols were developed to refine low resolution structures for which a structural knowledge of components is available at higher resolution (27). F 1 -and c 10 -chains were divided into several groups according to the structural domains: the N-terminal ␤-barrel domain (␣26 -95 or ␤6 -82), the central nucleotide binding domain (␣96 -381 or ␤83-357) and the C-terminal helical domain (␣382-509 or ␤358 -475) of each catalytic head subunit, the coiled-coil helical domain (␥1-59, ␥201-278) and the globular domain (␥71-200) of the ␥-subunit, the ␤-sandwich domain (␦9 -90) and the two C-terminal helices (␦92-97, ␦99 -138) of the ␦-subunit and finally the outer (c1-41) and inner (c42-76) helices of each c-subunit. Rigid body, individual coordinates and atomic displacement parameter refinements were further performed with PHENIX (28). Extensive torsion angle simulated annealing refinement with harmonic restraints applied on atomic coordinates was carried out using the maximum likelihood amplitude target in CNS (version 1.2) (29). The final model refined to 3.43 Å resolution had R work and R free of 0.312 and 0.319, respectively. Data and refinement statistics are supplied in Table 1. The refined structure was subjected to validation using PROCHECK (30). The secondary structures were assigned using STRIDE (31). The r.m.s.d. between equivalent C␣ atoms and the calculation of rotation angles were obtained with LSQKAB from the CCP4 package (23). Protein structures were illustrated by using the PyMOL program (32). The electrostatic potentials were calculated with APBS (33). supplemental Fig. S2 was drawn with ESPRIPT (34). To facilitate comparison with the bovine sequence, the bovine numbering is indicated in parentheses.
Intermolecular Lattice Contacts-The buried surface areas (Å 2 ) of interfaces in yF 1 c 10 ⅐ADP, yF 1 , and bF 1 -ATPase were calculated with the PDBePISA server.

RESULTS
Overall Description of the Structure-The structure was solved by molecular replacement using the coordinates of the yeast yF 1 (14) and of the I. tartaricus rotor ring (16). These high resolution structures provided accurate models to phase and then refine the yeast F 1 c 10 subcomplex structure. According to the functional ␣and ␤-subunits reference scheme, as in Ref. 3, chains A, B, and C were named ␣ E -, ␣ TP -, ␣ DP -subunits, and chains D, E, and F were named ␤ DP -, ␤ E -, and ␤ TP -subunits, respectively (supplemental Fig. S2). The final model contains all subunits of both the F 1 -sector (␣ E 26 -510, ␣ TP 25-509, ␣ DP 26 -510, ␤ E 6 -478, ␤ TP 7-478, ␤ DP 6 -475, and ␥1-59 and ␥71-278, ␦9 -138, and ⑀1-59) and the F 0 membrane rotor composed of a ring of 10 c-subunits (c1-76) (Fig. 2). No densities that could be attributed to any peripheral stalk subunits were visible. The overall structure of the F 1 -sector of yF 1 c 10 ⅐ADP is more similar to the three yeast yF 1 (I-III) structures (root mean square deviation (r.m.s.d.) Ͻ 1.7 Å) than to those of bovine bF 1 structures (r.m.s.d. Ͼ 4.5 Å) ( Table 2). yF 1 (I), the most complete model, and yF 1 (III) were found to be more similar (r.m.s.d. Ͻ 1.47 Å). Thus, yF 1 (I) will be considered as the yeast reference structure.
Structure of the F 0 Rotor Ring-The c-subunit folds as a helical hairpin with the N-terminal helix c2-38 (helix H1) connected by a polar loop c39 -45 containing an R(N/Q)P motif to the C-terminal helix c46 -74 (helix H2) (Fig. 3A). A decameric c-ring displays an hourglass shape with a central pore. The ring is 60 Å in height and has a diameter of 57 Å at the top, 51 Å at the bottom, and 43 Å in the middle. The above-mentioned distances are between local helical axes on the cGly 21 -C␣ atoms plane. As shown in Fig. 3B, the neighboring helices H1 are tightly packed with 7.5 Å between their local axes. One outer helix H2 N interacts with two inner helices, H1 N at 7.4 Å and where h is the index for unique reflections, and j is the index for symmetry redundant reflections. R sym and R merge are the reliability factors for single crystal (1 and 2) and merged data sets, respectively. I h is the mean weighted intensity after rejection of outliers. c R work and R free were calculated by ⌺ʈF observed ͉ Ϫ k͉F calculated ʈ/⌺͉F observed ͉. R free was calculated using 5% random data (3767 reflections) omitted from refinement. d Percentage of residues in most favored regions, allowed regions and disfavored regions, respectively. H1 Nϩ1 at 9.0 Å to form a tightly packed three-helix bundle (open angle ϳ50°) turned inwards. Furthermore, one inner helix H1 N interacts with two outer helices, H2 N and H2 NϪ1 , to form a more open three-helix bundle (open angle ϳ80°) turned outwards that harbors the proton binding site. The neighboring helices H2 are further away at 11.5 Å (Fig. 3B). In helix H1, a 14°k ink in the cLeu 20 region follows a two-turn long 3 10 helix (c16 -20) (Fig. 3A). A conserved GxGxGx(G/A) motif in the middle of the helix H1 (c21-27) (supplemental Fig. S3) participates in this close packing (50). In helix H2, a more pronounced 29°kink is located below cGlu 59 in the region opposite the glycine-rich motif (Fig. 3A). The central constriction is caused by the bend of both inner and outer helices. The pore has a diameter of 14 -16 Å in the middle and 23 Å at the bottom and is lined by Ile 10 , Ile 14 , Ile 17 , Ile 24 , Ile 28 , and Leu 20 residues (Fig. 3A). There are some narrow clefts between the outer helices, and one of these clefts (Fig. 3B) is lined by Leu 19 , Gly 21 , Ala 22 , Gly 23 , and Ile 26 from H1 N , Glu 59 , and Gly 62 from H2 N and Ser 58 , Ala 60 , Thr 61 , Phe 64 , from H2 NϪ1 . There is no intrinsic channel lined by polar groups in the upper and lower parts of the c 10 -ring. The cGlu 59 side chains with the lowest temperature factors (J-M) display well defined electron densities (Fig.  1). Nearby, small residual densities are not sufficiently high to model likely DCU modifications. One of oxygen atoms of the cGlu 59 side chain is accessible from the outer surface, whereas the second atom is turned toward the center of the outer threehelix bundle and is not hydrogen bonded (Fig. 3, B and C). Large rotations of the cGlu 59 side chains are sterically hindered by cIle 26 , cPhe 55 , cAla 56 , and cLeu 63 of the same c-subunit.
In yF 1 c 10 ⅐ADP, TP Pair Resembles the DP Pair of Yeast and Bovine F 1 -Despite the limited data resolution, residual electron densities appeared clearly in five nucleotide binding sites. These densities were sufficiently defined to be able to distinguish between di-and triphosphate nucleotides (supplemental Fig. S4). Mg⅐ATP molecules were modeled in the three noncatalytic sites, whereas Mg⅐ADP molecules were modeled in both the ␤ DP and ␤ TP catalytic sites. The refined temperature factors of nucleotides were compatible with fully occupied binding sites. No densities were observed in the ␤ E catalytic site. The mean overall temperature factor of the ␣ E -, ␣ DP -, ␤ DP -, and ␤ TP -subunits (95 Å 2 ) is lower than that of the ␣ TP -and ␤ E -subunits (125 Å 2 ).
When ␣and ␤-subunits were individually compared with those of the yF 1 (I) model (14) Fig. S5). The change can be described as a rigid body rotation of 7.7°of subdomain II. When E, DP, and TP pairs are compared with yF 1 (I) pairs (    (Fig. 4B) and bF 1 ⅐DCCD (2940 Å 2 , r.m.s.d. 1.14 Å). It is definitely more closed than the TP interface of yF 1 (I) (2280 Å 2 , r.m.s.d. 1.64 Å) (Fig. 5C). This closure can be described as an overall 5.7°rotation of the ␣ TP -toward the ␤ TP -subunit and results in the ϳ8 Å displacement of the tip (␣ TP 407-411) of the helix-turn-helix structure of the C-terminal domain toward the adjacent ␤ TP -subunit, tangentially to the C-terminal helix of the ␥-subunit (Fig. 5C and supplemental Figs. S5 and S6C). As a probable consequence, the tip is visible in the 2F obs Ϫ F calc electron density maps contoured at 1.5, whereas it is absent in the F obs Ϫ F calc electron density maps of yF 1 contoured at 2.0. In the crystal, neighboring molecules are arranged to form infinite columns along the [101 ] direction of the unit cell by a coaxial association with the collar protruding from c 10 -ring that plugs the cavity lined by the six N-terminal ␤-barrels of (␣␤) 3 . In the orthogonal plane, neighboring columns are in the opposite direction and display a V-shaped arrangement (supplemental Fig. S2). They interact through the central nucleotide-binding domains. Four connecting loops (␣103-108, ␣120 -126, ␣224 -229, and ␣193-195) of ␣ E -and ␣ DP -subunits interact tightly, whereas the outer surfaces of ␤ DP -and ␤ TP -subunits face each other more loosely. The outer surfaces of ␣ TP -and ␤ E -subunits are widely solventaccessible. The TP closure cannot be attributed to direct lattice contacts of the ␣-subunit.
Faint residual densities were observed in the hydrophobic binding pocket around the ␤ DP -and not ␤ TP -Glu 200 side chain without structural changes in the neighborhood when the ␤ DP -subunit was compared with yF 1 (I) and ␤ TP -subunit to yF 1 apo(III). Attempts to model DCU modification were unsuccessful. Indeed, no structural changes were observed previously between bF 1 ⅐DCCD (13) and bF 1 ⅐ADP (11) structures. Finally, the yF 1 c 10 ⅐ADP structure represents an Mg⅐ADP-inhibited state of the yeast F 1 enzyme, which is relatively different from those observed in bovine structures.
Central Stalk-A whole ␥␦⑀ central stalk was observed in yeast F 1 structures (2,14,15) and in bF 1 ⅐DCCD (13). The ␥-, ␦and ⑀-subunits of yF 1 c 10 ⅐ADP are individually similar to those of yF 1 (I) (r.m.s.d. Ͻ 1.05 Å, Table 2). Overall, the ␥␦⑀ central stalks of the yeast F 1 structures (yF 1 c 10 ⅐ADP, yF 1 (I), yF 1 (II), yF 1 c 10 ) are also similar (r.m.s.d. Ͻ 1.7 Å), whereas they differ significantly from the bovine one (r.m.s.d. ϭ 3.05 Å, Table 2). The overall temperature factors of ␥-, ␦and ⑀-subunits are 120 Å 2 , 147 Å 2 , and 149 Å 2 , respectively. In F 1 complexes, the orientation of the ␥-subunit relative to the (␣␤) 3 component Hydrophobic residues lining the pore and significant residues are labeled. B, section of the c 10 -ring viewed from F 1 at the level of proton binding sites. The molecular surface of a half-ring is colored in gray and red for apolar and acidic residues, respectively. One cleft located slightly below the proton binding site is indicated by the red arrow. The inner and outer three-helix bundles are outlined by blue and red triangles, respectively. The rotation direction of the ATP synthesis is indicated. C, the essential cGlu 59 carboxylate group is surrounded by hydrophobic residues and is not involved in hydrogen bonding. The conserved residues of the GxGxGx(G/A) motif are labeled in orange. determines the catalytic state of the enzyme and, according to the binding change mechanism, a 120°rotation of the ␥-subunit during the ATP hydrolysis cycle results in E 3 TP, TP 3 DP, and DP 3 E interconversions. The three ␣-subunits were used to superimpose the central pseudo 3-fold axes. The upper part (␥256 -278) of the C-terminal helix of the ␥-subunit are well superimposed (Fig. 5A). This helical segment is constrained to lie inside the narrowest part of the ␣ 3 ␤ 3 inner channel lined by the proline-rich collar (␣Pro 290(288) , ␣Pro 291(289) , ␣Gly 292 and ␤Pro 276(276) ) (3). A conserved ␥Gly 273(268) allows this helix to nestle against the ␣ DP -and ␤ DP -subunit interface. Below, the N-and C-terminal helices interact to form a coiled coil (Fig. 5B). From a tight turn of the yeast backbone at residue ␥Tyr 255(250) , the stalk deviates increasingly as it becomes closer to the c-ring rotor, with ϳ18-Å shifts relative to bF 1 ⅐DCCD at the periphery of the foot (Fig. 5D). Accordingly, the long C-terminal helix is more curved in yeast (75 Å of curvature radius) than in bovine structures (100 Å of curvature radius). The ␥-subunit twist is definitely an intrinsic property of the yeast enzyme. In yF 1 c 10 ⅐ADP, the region ␥235-242 of the C-terminal helix interacts with the helical domains of ␣ TP -and ␤ TP -subunits (Fig. 5C), and the TP closure results in a significant rotation/translation of this helix. In yeast, the region ␥18 -25 of the N-terminal helix interacts with the C-terminal helical domains of ␣ Eand ␤ DP -subunits, and the partial opening of the DP interface in the yeast structures could result from these interactions. The ␣ DP -subunit interacts with the globular domain of the ␥-subunit, whereas the helical domain of ␤ E -subunit does not interact directly with the ␥-subunit (Fig. 5C). The closure of the TP pair and the subtle differences in the relative orientation and position of the ␥-subunit result from the Mg⅐ADP inhibition. At the F 1 -F 0 rotor interface, the foot of the yeast central stalk is rotated by ϳ40°(36°for yF 1 (I), 41°for yF 1 c 10 or 43°for yF 1 c 10 ⅐ADP) relative to the bF 1 ⅐DCCD in the direction of the hydrolysis (Fig. 5D and supplemental Fig. S6D). As a consequence, the presence of the c 10 -ring and the crystal packing have no significant influence on the ␥-subunit twist. The ␦-subunit interacts extensively with the ␥and ⑀-subunits along with the c-ring (supplemental Tables S2  and S3). In the ␦-subunit, the second helix of the C-terminal domain folds back between the first helix and the N-terminal ␤-sandwich. The N-terminal helix of the small ⑀-subunit is inserted between the ␥-subunit and the two domains of the ␦-subunit. Comparison with bF 1 ⅐DCCD shows that the ␥and ␦-subunits have r.m.s.d. values of 2.60 Å and 2.16 Å, respectively. The overall r.m.s.d. of the ␦-subunit decreases to 0.8 Å only if the ␤-sandwich is considered, whereas the helical domain is more variable in sequence and conformation.
Interface between the F 0 Rotor Ring and Central Stalk-The foot sole of the central stalk is roughly flat and interacts with the upper surface of the membrane rotor, which adopts a complementary shape with the four C␣ atoms of the residues c39 -42 lying in a plane slightly tilted down the ring axis. The membrane rotor axis deviates from the pseudo 3-fold axis of F 1 by 8.5°with a minimum displacement of 2.8 Å (Fig. 2). The c-subunits that interact the most with the ␦or ␥-subunit have well defined  Fig. S6). The nucleotides of yF 1 c 10 ⅐ADP are drawn as a ball-and-stick diagram, and Mg 2ϩ is drawn as a sphere. The central stalks are shown using a schematic except for the coiled-coil of ␥-subunits (␥1-50 and ␥205-276), which are drawn as linked C␣. Blue, bF 1 ⅐DCCD (12); green, yF 1 (I) (13); and red, yF 1 c 10 ⅐ADP. The yF 1 c 10 ⅐ADP ring (multicolored surface) and the rotor axis (yellow decagon) are drawn. A, the C-terminal helices ␥256 -278 are represented as sticks in the proline-rich collar (␣Pro 290 , ␣Pro 291 , ␣Gly 292 , and ␤Pro 276 ). The conserved ␥Gly 273(268) is near the ␣ DP -and ␤ DP -subunits. B, at the end of the coiled-coil ␥1-17 and ␥230 -259, the Tyr 255(250) backbone is more twisted in yeast. C, near the hinge residues, ␥18 -25 and ␥235-242, the yeast and bovine N-terminal helices clearly deviate. The arrow marks the deviation of ϳ8 Å of the tip ␣ TP 407-411 toward the ␤ TP -subunit. D, the foot of the yeast yF 1 c 10 ⅐ADPcentral stalk is rotated 43°relative to the bF 1 ⅐DCCD stalk in the ATP direction of the hydrolysis. electron densities and the lowest temperature factors (130 Å 2 ), as reported previously in yF 1 c 10 (2). Only three c-subunits interact with the ␦-subunit, four with the ␥-subunit, two with both subunits, and one c-subunit does not interact. The buried area with the ␦-subunit (660 Å 2 ) is more than twice the area with the ␥-subunit (200 Å 2 ) (supplemental Table S3), and the interface area represents only a quarter of the accessible surface of the top side of the rotor ring.
In the ␥-subunit, the loop ␥195-202 linking the globular domain to the C-terminal helix is almost planar and orthogonal to the rotor axis (Fig. 6A). This loop goes down and crosses over the ring pore with ␥Gly 195 and ␥Lys 196 on top of the inner ring and ␥Asp 200 nearly in line with the rotor axis (Fig. 6C). The loop contains three acidic residues (␥Glu 198 , ␥Asp 200 , and ␥Asp 202 , the first being highly conserved Glu/Asp), so the negatively charged stalk surface could interact with the positive rotor ring surface (Fig. 6, B and C).
In the ␦-subunit, the region ␦36 -46 containing the fourth strand and the loop connecting the two sheets of the ␤-sandwich is again almost planar and orthogonal to the rotor axis (Fig. 6A). It interacts tightly with three c-subunits (with ␦Ser 36 , ␦Gly 40 , and ␦His 45 and cArg 39 , cAsn 40 , and cSer 42 ) (Fig. 6, B and  C). The last helix of the helical domain runs far from, but quite parallel to, the c-ring surface.
In the I. tartaricus Na ϩ F-ATPase, the Na ϩ is bound by Glu 65 , two polar side chains (Gln 32 and Ser 66 ), one backbone carbonyl (Val 63 ), and a water molecule (35). In chloroplastic and cyanobacterial H ϩ F-ATPase, the carboxyl oxygen atoms of the essential Glu 61/62 are bound by Gln 28/29 , Tyr 66/67 , and the backbone carbonyl Phe 59/60 . In S. cerevisiae, the polar residues are replaced by hydrophobic ones (Ile 26 , Ala 60 , and Phe 64 ) (supplemental Fig. S3 and Fig. 7) and the kink in outer helix H2 disrupts the intrahelical hydrogen bond networks, so cAla 56 and cLeu 57 backbone carbonyl groups are not hydrogenbonded. The cLeu 57 carbonyl is 5.1 Å apart from the cGlu 59 carboxyl oxygen atoms (Fig. 3C); therefore, a nonbonded interaction would require a molecular bridge. As a result, a water molecule could be inserted at a position close to that of the Na ϩ in I. tartaricus c 11 -ring and could establish hydrogen bonds with cGlu 59 carboxyl and cLeu 57 carbonyl groups. The equivalent distance is 4.5 Å in the Na ϩ binding site (16) and 2.7 Å in the proton binding site with a direct hydrogen bond (16,19). Hence, the current structure does not rule out the involvement of a hydronium ion in the proton translocation mechanism in yeast. Two types of proton binding sites could exist, with or without hydronium ion requirement. Recently, the observed bell-shaped pH profile for DCCD labeling of the acidic c-ring residues of H ϩ -transporting F-ATP synthase (36) were taken to be the involvement of a hydronium ion in the binding site, as proposed earlier by Boyer (37). A putative hydronium binding scheme for yeast H ϩ F-ATP synthase could be built on that of the sodium binding of Na ϩ F-ATPases (38), with a working mode similar to the "push and pull" functional model (39).
The outer surface of the c-ring is essentially hydrophobic with the exception of the essential cGlu 59 embedded at 24 Å from the two sides of the membrane, if cLys 44 /cAsp 45 and cLys 8 /carboxyl-terminal groups, at a distance of ϳ43 Å, mark the inner and outer surfaces of the mitochondrial membrane, respectively (Fig. 3A). On the matrix side, the pore is closed partly by the cArg 39 side chains that could act as a socket to receive the foot of the central stalk. An enlargement to 25 Å near a collar of cAsn 35 could indicate the top of a lipid plug. At the pore openings, a few residual density peaks cannot be attributed with certainty to ordered lipids or detergent molecules. However, there is enough space to accommodate, in vivo, two opposite plugs of a few lipids. It has been shown in E. coli that the pore is occupied by phospholipids (40).
The ␤ DP and ␤ TP binding sites contain an Mg⅐ADP molecule with the "arginine finger" ␣Arg 375(373) pointing in toward the site (supplemental Fig. S4). The structure reported here is probably equivalent to the ADP-inhibited state of the enzyme because our crystallization conditions leading to bound Mg⅐ADP and no free phosphate (41). This form is probably not an intermediate in either ATP hydrolysis or ATP synthesis but rather a pause state from which reactivation occurs upon the presence of protonmotive force in mitochondrial ATP synthase (7,42). The yF 1 c 10 ⅐ADP crystals were grown in the presence of ADP and azide without ATP analogs like various bovine F 1 -ATPase crystals. Azide is known to enhance Mg⅐ADP inhibition (43,44). The bF 1 ⅐ADP⅐N 3 Ϫ structure at 1.95-Å resolution revealed how azide enhances the binding of the ADP molecule in the ␤ DP -subunits (45). Afterward, the hypothetical presence of an azide molecule was proposed in bovine structures in the ␤ DP catalytic sites (3,10,13). Similarly, it cannot be ruled out that an azide molecule is bound to the ␤ DP and ␤ TP sites of yF 1 c 10 ⅐ADP.
Whatever the crystallization method, the detergent concentration or the inhibitor used to crystallize the yeast F 1 F 0 -ATP synthase, all subunits of the peripheral stalk were lost as indicated by SDS-PAGE analysis of crystals. SDS-PAGE analysis of three-dimensional crystals of the bacterial F 1 F 0 -ATP synthase from Chloroflexus aurantiacus obtained by slow detergent removal also indicates the loss of the peripheral stalk during crystallization (46). The crystallization conditions may induce the destabilization of the a-subunit/c-ring and the peripheral stalk/F 1 interfaces, leading to the F 1 -c-ring subcomplex.
In E. coli (47) and I. tartaricus (48), three acidic residues have been identified in the ␥-subunit as good candidates for interaction with the c-ring. However, we cannot exclude that the positive charges of conserved cArg 39 could be partially neutralized by the polar heads of a phospholipid plug. The loss of the peripheral stalk and the large crystal lattice interface (980 Å 2 ) between the c-ring and the ␤-barrels domains of the (␣␤) 3 could disturb the interface of a similar area (890 Å 2 ) between the central stalk and the c-ring (supplemental Table S3), so the densities of cArg 39 and ␥Glu 198 side chain residues are very poorly defined.
The winding of the central stalk observed in different conformational states of yeast F 1 complexes, with or without the c-ring, appears specific to the yeast F 1 . The structure of a complex between bovine F 1 and a truncated stator containing OSCP-, b-, d-, and F 6 -subunits reveals the position of the peripheral stalk along the cleft of a noncatalytic interface (49) and suggests a low degree of freedom for the location of the membrane domain of the stator relative to the membrane rotor. The peripheral stalk prevents rotation of the (␣␤) 3 head but does not prevent its essential conformational changes. The predictions of the transmembrane regions and the secondary structures of the yeast subunit b (subunit 4) suggest that the peripheral stalks are highly similar. In yF 1 c 10 ⅐ADP, the stoichiometry of 10 monomers per c-ring and the rotation of ϳ40°of the central stalk foot with respect to bovine are compatible with a step-by-step rotation of 36°of the rotor, without involving any shift of the membrane stator and hence modification of the overall conformation of the peripheral stalk. Finally, when comparing the yF 1 (I) and yF 1 c 10 ⅐ADP structures, it appears that the presence of the c-ring does not modify the position of the ␥-subunit and of the foot sole of the central stalk (Fig. 5).
Refinement of the crystal structure of the yeast F 1 c 10 inhibited by AMP-PNP should unveil some unknown features of this complex. High resolution structures of rotor rings of bacterial or mitochondrial H ϩ -transporting F-ATP synthase are now required to clarify the nature of the proton binding site in mitochondrial ATP synthases.