Resolving the Negative Potential Side (n-side) Water-accessible Proton Pathway of F-type ATP Synthase by Molecular Dynamics Simulations*

Background: ATP synthase converts a transmembrane gradient of protons or Na+ into chemical energy. Results: Molecular dynamics simulations mapped a water half-channel as part of the proton path. Conclusion: The half-channel is inclined against the sense of rotation of the c-ring, minimizing futile proton flow. Significance: The geometry of the proton path is crucial in understanding the energy coupling of ATP synthase. The rotation of F1Fo-ATP synthase is powered by the proton motive force across the energy-transducing membrane. The protein complex functions like a turbine; the proton flow drives the rotation of the c-ring of the transmembrane Fo domain, which is coupled to the ATP-producing F1 domain. The hairpin-structured c-protomers transport the protons by reversible protonation/deprotonation of a conserved Asp/Glu at the outer transmembrane helix (TMH). An open question is the proton transfer pathway through the membrane at atomic resolution. The protons are thought to be transferred via two half-channels to and from the conserved cAsp/Glu in the middle of the membrane. By molecular dynamics simulations of c-ring structures in a lipid bilayer, we mapped a water channel as one of the half-channels. We also analyzed the suppressor mutant cP24D/E61G in which the functional carboxylate is shifted to the inner TMH of the c-protomers. Current models concentrating on the “locked” and “open” conformations of the conserved carboxylate side chain are unable to explain the molecular function of this mutant. Our molecular dynamics simulations revealed an extended water channel with additional water molecules bridging the distance of the outer to the inner TMH. We suggest that the geometry of the water channel is an important feature for the molecular function of the membrane part of F1Fo-ATP synthase. The inclination of the proton pathway isolates the two half-channels and may contribute to a favorable clockwise rotation in ATP synthesis mode.

ATP synthases from bacteria, mitochondria, and chloroplasts produce ATP, the universal fuel in biological cells. To synthesize the high energy triphosphate from ADP and inorganic phosphate, the enzymes use a transmembrane proton gradient (1)(2)(3)(4) and convert electrochemical into chemical energy. F-type ATP synthases consist of a transmembrane F o domain and an extramembranous F 1 domain. F o from chloroplasts comprises a rotating c-ring and a stator domain formed by subunits a, b, and bЈ. Although the stoichiometry of the transmembrane stator domain is fixed in all organisms, the c-ring has 8 subunits in bovine mitochondria (5), 10 subunits in yeast mitochondria (1), 14 subunits in chloroplasts (6,7), and 10 -15 subunits in bacteria (8 -15). The chloroplast F 1 domain comprises subunits ␥ and ⑀, which rotate with the c-ring, and the nonrotating catalytic domain ␣ 3 ␤ 3 , which is connected to the stator domain of F o through the ␦ subunit (see Fig. 1A). Light-driven proton pumps in chloroplasts and phototropic bacteria or respiratory-chain enzymes in mitochondria and aerobic bacteria generate a proton gradient across the membrane, resulting in a positively charged p-side 3 and a negatively charged n-side (p-side and n-side refer to lumen and stroma in chloroplasts, intermembrane space and matrix in mitochondria, and periplasm and cytoplasm in bacteria, respectively). The resulting proton motive force drives the c 8 -15 ␥⑀ rotor, which induces conformational changes in the ␤ subunits of the stator and, hence, provides the energy to produce ATP. Some fermenting bacteria use sodium ions instead of protons to drive ATP synthesis (16).
Subunits a and c of the transmembrane F o domain are directly involved in proton transport. Transport from the p-side to the middle of the membrane where protons bind to a conserved carboxylate of subunit c (Glu 61 in chloroplasts) is facilitated by an intrinsic channel or proton wire in subunit a (p-side half-channel) (17)(18)(19)(20)(21)(22). Protonation of the essential carboxylate in the c-subunit drives clockwise rotation of the entire c-ring (as seen from the p-side) (23). The step size of the c-ring corresponds to one subunit of c, indicating a proton motive force-driven Brownian ratchet mechanism (24,25). The protons are then released from the proton-binding site on subunit c to the n-side (17,26). An aqueous path at the interface between subunits a and c was inferred by mutagenesis and labeling studies (18,19,21,(27)(28)(29), which may function as the n-side half of the proton channel. F 1 F o -ATP synthases can also catalyze the reverse reaction driven by ATP hydrolysis and pump protons from the n-side to the p-side of the membrane via the same half-channels, resulting in a counterclockwise rotation of the c-ring rotor (23).
Preventing a nonproductive shortcut between the p-side and the n-side half-channels is an essential prerequisite for the transformation of proton flux into mechanical energy (26). A conserved positively charged side chain in subunit a (Arg 193 in chloroplasts and Arg 210 in Escherichia coli) isolates the two half-channels by preventing any protonated (and thus uncharged) Glu 61 of subunit c to pass the positively charged guanidinium group of the aArg 193 side chain (30 -35). Thus, aArg 193 prevents the futile counterclockwise rotation in ATP synthesis mode.
However, the structure of the a/c interface and how the protein complex maintains isolation between the half-channels, i.e. how the nonrotating shortcut of protons is prevented, has remained elusive so far. Although it is established that the n-side half-channel is water-accessible, the pathway of the protons has not yet been identified. Here we resolve this pathway at atomic resolution by MD simulations of the structure of the c 14 rotor ring of the chloroplast ATP synthase (7) in a lipid bilayer surrounded by water. The carboxylate cGlu 61 of one of the protomers was deprotonated, thus mimicking the protomer facing the a subunit in the intact F o domain. We identified a water-accessible region inside the lipid bilayer at the periphery of the c-ring rotor. This water channel runs from the n-side of the membrane to the deprotonated carboxylate cGlu 61 in the middle of the membrane. The aqueous n-side half-channel is not parallel to the axis of the c-ring rotor but rather inclined, which may enhance the isolation between the half-channels and may contribute to a favorable clockwise rotation in ATP synthesis mode. The MD results also give a plausible explanation of the peculiar properties of the double mutant cP24D/ E61G, which, in E. coli, is fully functional despite lacking the essential Asp 61 side chain (36); the water channel now reaches deep between two c-protomers, thus enabling the repositioning of the reversibly protonated carboxyl group of the suppressor mutant.

Generation of c 14 Rotor Ring/Lipid Bilayer Starting
Structures-Starting structures for MD simulations of the c 14 rotor ring in a membrane environment were generated from the crystal structure of the c-ring from spinach chloroplasts (Protein Data Bank (PDB) code 2w5j) (7). It comprises 14 identical c-subunits, each of which forms a hairpin structure consisting of two TMHs. Missing amino acids and side chains were added manually in the most plausible rotamer conformations and refined as described (7). The c-ring structure was oriented with respect to a lipid bilayer according to information provided by the Orientations of Proteins in Membranes database (37). The protein structure was then embedded in a bilayer of 288 1,2dioleoyl-sn-glycero-3-phosphorylcholine lipids, which itself was embedded in 20 Å layers of water molecules above and below. The pre-equilibrated configuration of the lipid bilayer and the water layers was taken from Rosso and Gould (38). Ninety-nine lipids that were sterically overlapping with the protein structure were removed. The center of the c-ring was filled with 12 lipids to model the lipid plug seen in isolated rings (39).
From this system, the following starting structures were generated. I) All but one of the Glu 61 residues of the subunits c were protonated (hitherto referred to as "13protonated"). II) All Glu 61 residues were protonated ("14protonated"). III) All Glu 61 residues were deprotonated ("0protonated"). IV) The double mutant cP24D/E61G was generated by pruning the side chain of Glu 61 and replacing Pro 24 with the most plausible rotamer of Asp, as provided by Coot (40). In this system, all but one of the Asp 24 residues were then protonated (double mutant). To reach electroneutrality for each of these starting structures, sodium ions were added by the leap program of the Amber package (41) as required. This resulted in system sizes of ϳ78,000 atoms.
Molecular Dynamics Simulations-MD simulations were performed with the Amber 10 suite of programs (41). For the protein and ions, the force field by Cornell et al. (42) was used with modifications suggested by Simmerling et al. (43). TIP3P was used as a water model (44). For the lipids, force field parameters and charges derived by Rosso and Gould (38) were used, which are based on the General Amber Force Field (45) and the RESP procedure (46), respectively. These parameters have been shown to yield area per lipid values, peak distances, and lipid volumes that converge around values close to experimental ones when simulations were performed under the condition of an isobaric-isothermal ensemble (NPT) with anisotropic pressure control (38). That way, a restraining of the area as in a constant area isobaric-isothermal (NPAT) ensemble can be avoided, which may lead to simulation artifacts (47).
Accordingly, after minimization of the systems for 2000 steps with the protein atom positions restrained, canonical ensemble (NVT) MD was carried out for 50 ps, during which the system was heated from 100 to 300 K, applying harmonic restraints with force constants of 5 kcal mol Ϫ1 Å Ϫ2 to the protein atoms. Subsequent NPT-MD was used for 150 ps to adjust the density. After gradually reducing the force constants of the harmonic restraints on solute atom positions to zero during 250 ps of NPT-MD, the following 20 ns of NPT-MD at 300 K were used to further equilibrate the system. Equilibration times of this length were shown to be necessary for typical simulation studies involving lipid bilayers (48). Finally, the following trajectories of 30 ns length generated by NPT-MD were used as production runs with conformations extracted every 20 ps. This resulted in ϳ250 ns of total simulation time.
Throughout the simulations, the particle mesh Ewald method (49) was used to treat long-range electrostatic interactions, and bond lengths involving bonds to hydrogen atoms were constrained using SHAKE (50). The time step for all MD simulations was 2 fs, with a direct space, nonbonded cutoff of 8 Å. The temperature was controlled using the Berendsen thermostat (51) with a time constant of 10 ps, and the Berendsen barostat (51) was used for anisotropic pressure control with a time constant of 2 ps.
Analysis of the Production Runs-The production runs were analyzed with ptraj of the Amber suite of programs (41). The "watershell" command was used for analyzing the first and second solvation shells with respect to the oxygen atoms of Glu 61 (13protonated, 14protonated, 0protonated) or of Asp 24 (double mutant) using lower and upper distances of 3.4 and 5.0 Å, respectively. Water densities were determined with the "grid" command using a cubic grid that encompasses the whole simulation cell with a spacing of 1 Å in each direction. Depicted contour surfaces encompass regions with at least 50 counts of water molecules over the 1,500 snapshots analyzed. Prior to calculating root mean square deviations (r.m.s.d.) and an average structure, the c-ring structure was superimposed with respect to all C ␣ atoms. A snapshot with a minimal C ␣ atom r.m.s.d. to the average structure was then taken as the representative structure depicted in Figs. 1 and 2. 1 and 2 angles of residues Glu 61 were determined as the torsion angles involving atoms N-C ␣ -C ␤ -C ␥ and C ␣ -C ␤ -C ␥ -C ␦ , respectively. The kink of the outer helices of the c-ring was determined as the angle of the point triple (center of mass of C ␣ atoms of residues 48 -50; C ␣ atom of residue 61; center of mass of C ␣ atoms of residues 74 -76). The S.E. was determined as the S.D. divided by the square root of the number of independent snapshots. The number of independent snapshots was determined from the time correlation function for the respective analysis. The correlation time was 200 ps in the case of the watershell analysis and 2.2 ns in the case of the analysis of the 1 and 2 torsion angles of Glu 61 .
To analyze possible conformational changes of the c-ring, an elastic network model analysis was performed with the ElNémo webserver (52) using default parameters. Molecule figures were prepared with PyMOL (Schrödinger, New York, NY).

RESULTS
We performed MD simulations of 50 ns length of the chloroplast c 14 -ring in an aqueous lipid bilayer with Na ϩ counter ions, with the last 30 ns taken for analysis. Although the conserved Glu 61 side chain from 13 protomers was protonated, the carboxyl group of one Glu 61 was deprotonated and thus charged (13protonated). This situation reflects the physiological condition of the c-ring in active ATP synthase where aArg 189 serves as counter ion. The water density was determined by counting the presence of water molecules within a cubic grid of 1 Å spacing. Fig. 1 shows an MD snapshot with a minimal r.m.s.d. to the average protein structure together with the water density as seen from the top, front, and side ( Fig. 1, B, C, and D, respectively). Although no water was present inside the 1,2-dioleoyl-sn-glycero-3-phosphorylcholine membrane at the start of the calculation, a water channel developed during the simulation originating from the n-side and reaching to the middle of the membrane bilayer, enclosing the deprotonated Glu 61 with its tip (Table 1).
We mapped the wetted surface on the c-ring conformation with a minimal r.m.s.d. to the average structure ( Fig. 2A). The main amino acids in contact with the water channel are listed in Table 2. This water channel is in agreement with experimental data that implied a water-accessible region in the interface between subunits a and c (18,19,21,(27)(28)(29). We also per-formed an MD simulation of a c-ring with Glu 61 of all 14 protomers protonated (14protonated) and of a completely deprotonated c-ring (0protonated). No water channel developed with the protonated (uncharged) c-ring ( Fig. 2B; Table 1). In contrast, with the fully deprotonated and therefore charged c-ring, the water encompassed almost completely the n-side half of the c-ring ( Fig. 2C; Table 1).
Suppressor Mutant cP24D/E61G-We also performed an MD simulation of the suppressor mutant cP24D/E61G, in which the carboxyl group of one Asp 24 was deprotonated and thus charged. In this mutant, the function of the reversibly protonated carboxyl group of conserved Glu 61 (Asp 61 in E. coli) is transferred from the outer cTMH2 to Asp 24 , which is located on the adjacent inner cTMH1 (36) . Fig. 2, D-F, show the resulting water density maps over the last 30 ns of the trajectory together with the closest-to-average conformation of the c-ring. As with the wild type c-ring, a water channel developed and reached as far as the carboxylate group of Asp 24 (Table 1).
Structural Changes of the c-Ring-Regarding possible structural changes of subunit c upon deprotonation, we do not observe significant differences between the protomer with deprotonated Glu 61 and the other, protonated protomers for the 13protonated system. In fact, the overall C ␣ atom r.m.s.d. remains stable below 4 Å over the trajectory, and the C ␣ atom r.m.s.d. of almost all TMHs remains below 3 Å. Furthermore, the kink of TMH2 of subunit c, which gives the c-ring the characteristic hourglass shape, is essentially the same; the kink angle of the helix with deprotonated Glu 61 is 158.7°Ϯ 1.0°(S.E.), whereas the average of 11 protonated helices is 159.7°Ϯ 0.8°. In the latter case, the two protomers adjacent to the helix with the deprotonated Glu 61 have been excluded to prevent a possible influence of the deprotonated protomer. The circular shape of the c-ring got distorted during the MD simulations (Fig. 1B). The degree and direction of distortion are consistent with results from an elastic network model analysis. This analysis reveals that the first four low frequency modes lead to a deformation of the c-ring structure perpendicular to the axis of the c-ring rotor; the first two of these modes preferentially distort the p-side of the c-ring, the last two distort the n-side (data not shown). In each case, the distorted structures take on an oval shape. A similar distortion, although to a lesser degree, has been observed in electron micrographs from Ilyobacter tartaricus (53). Finally, visual inspection of the trajectories showed no swiveling motion of the helices (data not shown).
Glu 61 Side Chain Conformation-Proton binding of Glu 61 is linked to a conformational change of this amino acid (7,54,55). We analyzed the conformations of Glu 61 according to the torsion angles 1 and 2 and classified the conformations as plus, trans, and minus according to Lovell et al. (56). Torsion angle 1 of the deprotonated Glu 61 adopts values of around Ϫ65°( minus) and around Ϫ177°(trans), whereas torsion angle 2 is mostly trans (Fig. 3A). Overall, the all-trans rotamer is mostly populated where the side chain points to the solvent. This finding agrees with the observation of an extended conformation of Glu 59 in the deprotonated c-ring from yeast (55). The protonated side chains have similar 1 torsion angles with minus and trans conformations and similar 2 torsion angles in minus and trans conformations. However, the plus conformation of 2 is populated, too (Fig. 3B). Overall, this leads to the trans/plus  Blue, main amino acids in contact with the water channel (see Table 2

TABLE 2 Main amino acids in contact with the water channel
Listed are amino acids close to the water channel as mapped in Fig 2A. The c-protomer with the deprotonated Glu 61 is denoted n. During ATP synthesis, the protomers to the left (in the orientation of Fig. 1A) are the next to be rotated into proton release position and are thus numbered nϪ1 and nϪ2. Accordingly, the protomer to the right is denoted nϩ1.  (Fig. 3, C and D) for both the deprotonated Glu 61 and the protonated ones, showing a pronounced mobility of these side chains.

DISCUSSION
A major challenge in understanding the molecular mechanism of energy coupling in the F 1 F o -ATP synthase is to resolve the proton transfer pathway in the transmembrane F o domain at atomic resolution. Currently, there are only two cryo-electron microscopy structures of F o subcomplexes available. The first structure is from Thermus thermophilus ATP synthase with 10 Å resolution (57). Here, subunit I (the functional homolog of subunit a) forms two complete half-channels from eight TMHs that are thought to be offset laterally. The other available structure is the ab 2 c 11 F o subcomplex of the Na ϩ -translocating F-ATP synthase from I. tartaricus with 7 Å resolution (53). Here, subunit a forms one four-helix bundle that probably provides the p-side half-channel. Additionally, up to three more TMHs, which belong to subunit a and perhaps subunit b, can be seen in the electron density. Subunits a of ATP synthases from E. coli or chloroplasts comprise five TMHs (58,59), and a water channel serves as the n-side half-channel (18,19,21,(27)(28)(29). Hence, the F o subcomplexes from E. coli, chloroplasts, and I. tartaricus may share the same architecture. The current model of the active ATP synthase (3) predicts that one protomer of subunit c has the conserved acidic side chain (Glu 61 in chloroplast ATP synthase) deprotonated. Its charge is neutralized by aArg 189 . The remaining c-protomers (13 in chloroplasts) have protonated Glu 61 and are embedded in the hydrophobic part of the membrane. To reflect this situation in our MD calculations, our starting model of the spinach chloroplast c-ring rotor was set to an ensemble of 13 protonated c-protomers plus one deprotonated c-protomer. This complex was embedded in a lipid bilayer enclosed by water layers. Although c-protomers are unlikely to be charged outside the a/c interaction, the MD simulation of this setting provides an elegant way to investigate possible structural changes that result from deprotonation of one c-protomer. We predicted earlier that for proton translocating ATP synthases, the conformational differences between protonated and deprotonated protomers are mainly limited to side chain movements of Glu 61 (7), as is probably the case with Na ϩ -translocating ATP synthases (60). This model was later confirmed with a high resolution structure of ATP synthase from cyanobacterium Spirulina platensis (54).
Our MD simulations presented here do support the idea that the main conformational changes of a protomer upon deprotonation are limited to the side chain movements of Glu 61 . The trajectories show no indication of helix swiveling, nor other major changes in helix conformation. Transitions between rotamer states in our simulation reveal a certain mobility of the side chains of both the protonated Glu 61 and the deprotonated Glu 61 . This observed mobility of the side chain in the hydrophobic part of the membrane moderates the stringent implication of the description of a "locked conformation" for protonated Glu 61 (60). The most striking result from our MD simulations is the location of water molecules inside the hydrophobic part of the lipid bilayer forming a water channel. The existence of a water channel has been extensively shown by in vitro experiments using chemical modification of c-rings with Cys substitutions (28,29). Previous MD simulations also found water molecules drawn into the membrane. However, these simulations used c-rings with all Glu 61 deprotonated (equivalent to our 0protonated system) (55,61), and a complete water ring was formed. Other simulations used c-rings with a single protomer deprotonated (equivalent to our 13protonated system), but the authors missed to observe a distinct water channel (61,62). The form and direction of the water channel identified here are instructive because the channel is not perpendicular to the membrane but points at an angle of ϳ45°from the n-side of the membrane to cGlu 61 and beyond. Therefore, the resulting proton wire does not follow the shortest path from cGlu 61 to the n-side but points against the rotational sense of the c-ring (in ATP synthesis mode). This orientation isolates the protons arriving from the p-side at the c-ring and the protons leaving from the c-ring to the n-side. As a result, two laterally offset half-channels are formed as proposed (3,4) and seen in T. thermophilus ATP synthase (57), but with the difference that the n-side half-channel in the case of chloroplast ATP synthase is a hydration layer in the a/c interface as depicted in Fig. 2A. The orientation can also be expected to favor deprotonation of Glu 61 on rotating from the position at protomer nϪ1 to the one at protomer n due to the presence of a higher dielectric environment that better stabilizes the emerging charge than a pure lipid environment. Accordingly, one may speculate that the orientation of the water channel contributes to a favorable clockwise rotation in ATP synthesis mode.
Control Calculations-Two additional MD simulations were performed to test the notion that a water channel is drawn into the hydrophobic part of the membrane if cGlu 61 is deprotonated and thus charged. With a fully protonated c-ring, less than 0.6 waters are found within 5 Å of the carboxyl group of Glu 61 ( Table 1). The fully deprotonated c-ring provides the opposite setting, where nearly 11 waters are found in the 5 Å shell ( Table 1). Comparison of the water densities shows that the water did not integrate into the membrane with the fully protonated complex, whereas with the fully deprotonated complex, a complete water ring developed from the n-side to the center of the membrane (Fig. 2, B and C). Thus, these results confirm that the charged Glu 61 in the hydrophobic part of the membrane leads to water molecules hydrating the protein/lipid interface.
Implications on an a/c Interface Model-As was argued earlier (62), c-protomers are not likely to be deprotonated outside the a/c interface. Here, we take full advantage of the possibility to simulate a functional state of the c-ring with one deprotonated and 13 protonated protomers in MD calculations. We predict that the resulting water channel shows a negative imprint of the hydrated a/c interface. This prediction is in line with findings from the E. coli subunit a. I) aTMH4 is positioned next to the outer TMH of c (63). II) This helix is kinked at the conserved Gly 213 , as shown by NMR experiments on site-directed paramagnetic spin-labeled protein (64). III) According to the topology established with vesicles (58) and with a/c fusion proteins (65), aTMH4 is oriented with the N terminus on the n-side and the C terminus on the p-side. IV) The 15 amino acids preceding the conserved Arg 210 are inaccessible to surface labeling (66). Therefore, it is feasible that the p-side half of aTMH4, i.e. the first 15 amino acids up to the conserved Arg 210 (E. coli numbering), forms a straight helix from the p-side to the middle of the membrane. This part of the helix is probably parallel to the rotational axis of the c-ring. The helix is kinked at Gly 213 and probably follows the direction of the water channel to the n-side. A layer of water molecules at the protein-protein interface provides the n-sided proton half-channel.
The proton-providing half-channel on the p-side is thought to be within subunit a, presumably inside a four-helix bundle (19,20,53). We predict that the location of this four-helical bundle is offset by one c-protomer from the n-sided water channel. As a result of the model, two c-protomers may be deprotonated simultaneously: one protomer with Glu 61 in the water channel, releasing a proton, and one protomer with Glu 61 interacting with conserved aArg 189 (aArg 210 in E. coli). This prediction is in accordance with the observation that in T. thermophilus ATP synthase, two four-helix bundles of subunit I are seen in the interface with the rotor (57).
Double Mutant-We modeled the suppressor mutant cP24D/E61G. In this variant, the reversibly protonated carboxylate is transferred from Glu 61 in cTMH2 to position 24 in the adjacent helix of the c hairpin. E. coli cells with this double mutant grow via oxidative phosphorylation (36). The ATP synthase complex with this suppressor mutant actually produces ATP (although slightly less efficiently) and is resistant to dicyclohexylcarbodiimide, a potent inhibitor of wild type c-ring (36). Our MD simulation shows that water is drawn deep into the protein structure to the carboxylate group of deprotonated Asp 24 (Fig. 2, D-F). Presumably, here at the interface between cTMH1 and cTMH2 is where the protons are bound. Thus, the water channel allows the protons to be released from a different position than in wild type. Its location in the inner interface of the hairpin-structured c-protomer may render the mutant insensitive to dicyclohexylcarbodiimide. In this simulation, the water channel is oriented in an angle against the rotational sense, too (i.e. in Fig. 2D, from the upper left to the center). The water channel is narrower than with wild type protein (Table 1), which may explain why the mutant is less efficient. Our model may also explain why Gly 61 is essential for this double mutant; the side chain of Asp 24 is likely to form a hydrogen bond with the nitrogen atom of Gly 61 with an O 3 N distance of ϳ3 Å (Fig.  2E), which may be prevented by C ␤ of any other amino acid in position 61.
Conclusion-The MD simulations presented here give a convincing picture; the charges of aArg 198 and cGlu 61 provide a strong driving force for water to move into the a/c interface, which is known to be hydrated at the n-side (18,19,21,(27)(28)(29). The surface of subunit c, and presumably also of subunit a, orients this water channel at an angle to the membrane normal. This water channel forms a proton path that is partially directed against the rotational sense of the c-ring (in ATP synthesis mode). Therefore, the hydrated proton path on the n-side is laterally offset to the p-side path, which presumably is inside a four-helix bundle in subunit a. This geometry promotes isolation between half-channels minimizing futile proton flow and is suggested to contribute to a favorable clockwise rotation in ATP synthesis mode.