Charge Transport in the ClC-type Chloride-Proton Anti-porter from Escherichia coli*

The first chloride transporter identified in the superfamily of ClC chloride channels was from Escherichia coli (EClC) (Accardi, A., and Miller, C. (2004) Nature 427, 803–807). Pathways, energetics, and mechanism of proton and chloride translocation and their coupling are up to now unclear. To bridge the hydrophobic gap of proton transport, we modeled four stable buried waters into both subunits of the WT EClC structure. Together they form a “water wire” connecting Glu-203 with the chloride at the central site, which in turn connects to Glu-148, the hypothetical proton exit site. Assuming the transient production of hydrochloride in the central chloride binding site of EClC, the water wire could establish a transmembrane proton transport pathway starting from Glu-203 all the way downstream onto Glu-148. We demonstrated by electrostatic and quantum chemical computations that protonation of the central chloride is energetically feasible. We characterized all chloride occupancies and protonation states possibly relevant for the proton-chloride transport cycle in EClC and constructed a working model. Accordingly, EClC evolves through states involving up to two excess protons and between one and three chlorides, which was required to fulfill the experimentally observed 2:1 stoichiometry. We show that the Y445F and E203H mutants of EClC can operate similarly, thus explaining why they exhibit almost WT activity levels. The proposed mechanism of coupled chloride-proton transport in EClC is consistent with available experimental data and allows predictions on the importance of specific amino acids, which may be probed by mutation experiments.

As has been established only recently, the superfamily of ClC-type chloride channels contains both channel and secondary-active transporter subtypes (2). Systems of the latter group exchange protons with chlorides in a fixed 2:1 stoichiometry and may function either as proton or chloride pumps (1,(3)(4)(5)(6)(7)(8)(9). The functional aspect of these chloride channels is fascinating; they manage to separate charges by transferring charge carriers of opposite signs in opposing directions across a membrane. The first ClC transporter identified was from Escherichia coli (EClC) 3 (1). In the mean time, more such transporters have been identified also in human cells (10 -12). Quite a few crystal structures of wild-type (WT) and mutant forms of the bacterial homologues are available (4,6,8,9,13,14). EClC is a homo-dimeric membrane protein ( Fig. 1) in which each chain provides up to three chloride anion binding sites, one at the intracellular surface called the internal chloride binding site (Cl (3) ) and another buried in the center of the protein, referred to as the central chloride binding site (Cl (2) ) where Glu-148 is found in immediate proximity. The crystal structure of the E148Q mutant of EClC (14), in which glutamate is replaced by glutamine, presumably mimicking the protonated form of glutamate, reveals a third chloride binding site (Cl (1) ) at the periplasmic surface. This chloride anion is right at the position where the Glu-148 side chain binds in the WT structure occluding this site, thus, blocking access for chloride from the periplasmic lumen. It is assumed (14) that this conformational change of Glu-148 is triggered by protonation and is ultimately responsible for the fast-gating behavior observed in homologous channel systems, i.e. the rapid transition between conducting and not-conducting states.
Although detailed experimental characterization of the transport mechanism remains elusive, limited information has been inferred from various mutants of EClC. Two conserved glutamates, Glu-148 and Glu-203, are essential for PT, and converting them into glutamines or other non-titratable residues completely abolishes active proton transport but retains passive chloride transport (1,3,9). Glu-203 is considered to be the proton entry site close to the intracellular lumen and Glu-148, correspondingly, the proton exit site, which in the "opened" channel state is exposed to the periplasmic space (3,9). Other point mutations at Glu-203 have varying impact on PT. Aspartate or histidine at this position preserve WT behavior, and lysine or arginine do this partially, whereas with non-titratable residues no significant PT is detectable (9).
The available crystal structures provide no explicit information about the PT pathway between the glutamates Glu-203 and Glu-148 except for a potentially titratable tyrosine (Tyr-445) and a crystal water between Glu-203 and Tyr-445 in both chains of EClC (354 and 121 in chain A and B, respectively). The hydroxyl-group of Tyr-445 is 7 and 6 Å away from Glu-203 and Glu-148, respectively. The crystal water, although forming a hydrogen bond with the backbone nitrogen * This work was supported by the German National Academic Foundation and the Volkswagen Foundation. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental material including Figs. 1-3. This paper is dedicated to Prof. Dale John Benos who died suddenly on Oct. 7,2010. We offer our heartfelt condolences to his family, friends, and colleagues. 1 Supported by the German National Academic Foundation and the Volkswagen Foundation. 2 To whom correspondence should be addressed. of Tyr-445 and the carboxyl group of Glu-203 (Fig. 2), is located about 7 Å away from the Tyr-445 hydroxyl-group and 12 Å from the Glu-148 acidic oxygens. Such distances are not appropriate to foster efficient PT. The tyrosine aromatic ring is surrounded by hydrophobic residues, whereas its hydroxyl hydrogen is bound strongly to the chloride anion in the Cl (2) binding pocket and is not capable of larger structural reorientations. It can be mutated to similar hydrophobic residues like phenylalanine or tryptophan with no apparent effect on the coupling between proton and chloride translocation (4,7). For these mutants direct PT between Glu-203 and Glu-148 would have to bridge a gap of about 15 Å. This may be within range of long distance electron tunneling processes (15) such as those in photosynthetic proteins (16,17) but is inconceivable for protons with their much larger mass. A pore-searching algorithm on EClC led to identification of two potential PT pathways, namely P1 and P2 (18). Based on a pore radius of about 1.5 Å, it was presumed that besides the aforementioned crystal water two more buried waters (W1 and W3) were located along pathway P1. But, that pathway ended "some distance from the Glu-148 gate" and was, therefore, considered unlikely (18). The second pathway (P2) would lead protons through a hydrophobic pore located at the opposite side of the Tyr-445 phenol ring plane to the central chloride binding site Cl (2) . In contrast to P1, protons on the P2 track could eventually reach the acidic oxygens of Glu-148 once they were located at Cl (2) at a distance of 4.1 Å. On the other hand, they are unlikely to tunnel all the distance between Glu-203 and Glu-148 without support by other titratable groups. The pore is mostly too narrow to accommodate buried waters, except for one potential site close to the proton entrance (18). Correspondingly, no crystal waters were detected along the pathway P2 (14). EClC is a homo-dimeric transmembrane protein with two chains, A and B, related by a 2-fold rotational symmetry. Each chain contains two chloride binding sites labeled Cl (2) and Cl (3) . Although the first is buried deeply in the hydrophobic core of the protein, the latter is located on the intracellular protein surface and, therefore, solvent-accessible. A third binding site Cl (1) (not shown) is occupied by the glutamate 148 in the WT crystal structure. A small conformational change of Glu-148 (see Fig. 2) opens the binding pocket for a third chloride. Cl (1) is connected with the periplasmic space via a funnel-shaped crevice that probably functions as the chloride entry site. Correspondingly, Cl (3) is the exit site for chlorides moving toward the intracellular lumen. Several other conserved residues (Tyr-445, Glu-203, Arg-28, and Glu-113) are found in the immediate environment of the chlorides, likely associated with EClC proton transport activity. The proton transport pathway is currently unknown. For EClC being an anti-porter, the protons must somehow proceed (blue arrow with question mark) in a direction opposite to the chloride trajectory (green arrow). This  Superimposed are the positions of nearby crystal waters in cyan as observed in the 1OTS crystal structure. An inset also shows the location of the additional water W5 predicted for the Y445F mutant. The waters bridge the gap between Glu-203 and chloride at Cl (2) establishing a new potential PT pathway between the entry and exit sites, Glu-203 and Glu-148, respectively, highlighted by a magenta-colored line labeled P3. For comparison, we added the alternative PT pathway proposed by two other groups (18,19) using a light blue colored line labeled P2. We superimposed yet another pathway (P1) discarded by the same investigators because it ends only with some distance to the putative proton exit site Glu-148. Although Glu-203 is assumed to be the proton entry site on the intracellular side, we believe P3 is more likely to extend upstream to the solvent-accessible residue Glu-117 (dashed magenta line). In this scenario, protons would enter EClC at Glu-117 and pass through an internal, water-filled cavity visible in chain B of the WT crystal structure. We connected the corresponding crystal waters 194, 214, 265, and 326 by red lines.
P2 was explored theoretically already by means of a hybrid quantum-classical molecular dynamics (MD) simulations (19) and found to be capable of transporting an excess proton placed at Glu-203 all the way downstream onto Glu-148. Despite the small diameter of the P2 pathway, the authors were able to place up to seven buried water molecules inside of P2 that also turned out to be absolutely necessary to enable fast proton transfer. This, from our view unexpected result implies considerable structural modifications not necessarily supported by EClC crystal structures. A more detailed discussion on this issue is given toward the end of this paper.
MD simulation methods of molecular systems provide many details on atomic level of description. The downside of this high resolution is their extensive computational cost rendering an energetic study of the numerous different combinations of chloride and proton occupancies of EClC ambitious if not impossible. It is, however, exactly this type of information that can serve as a solid basis for speculating on mechanistic details of the exchange cycle between chlorides and protons fostered by this molecular machine. Therefore, we applied a continuum-electrostatics approach that is less detailed but much faster and provides with current parameterization very precise energetics (20) if the employed structures are correct. It allowed us to explore all possible reaction schemes, which helps elucidating the underlying transport mechanism.
In a preparation mode we tried to add as much water molecules as possible inside the larger of the two pores, P1, by means of constrained geometry optimizations preserving most of the crystal structure. This way we were able to stabilize a single-filed water wire consisting of four buried waters connecting Glu-203 with Glu-148 via Cl (2) inside EClC. Although similar to P1 one of the buried waters (W4) of the water wire is outside of P1, thus establishing a new potential PT pathway if we also assume transient formation of hydrochloride. We identified a homologous pathway in both the Y445F and E203H mutants of EClC that could explain their almost WT-like activity levels. Using electrostatic energy computations, we tested the plausibility of this new PT pathway by assessing the protonation energies of the involved titratable groups. Moreover, based on characterization of a large number of possible chloride and proton occupancies, we propose a transport cycle fulfilling the observed 2:1 chloride/proton stoichiometry, consistent with the observed experimental transport rates. Our computations provide insight into how chloride translocation is tied to proton transfer by a dramatic shift of the Glu-148 pK A between its two conformations and predict a functional role of the conserved titratable residues Glu-113 and Glu-117 so far ignored in mutational studies.

EXPERIMENTAL PROCEDURES
Bookkeeping of EClC Microstates-The transport mechanism in both monomers of the homo-dimeric EClC is likely not directly coupled (21). Therefore, we used a simplified description of EClC function assuming that both monomers adopt the same state. In the following we characterize the possible chloride occupation states using a string of three symbols providing the status of the three crystallographically detected chloride binding sites (Cl (1) -Cl (3) ) per EClC mono-mer. Ordered from left to right, they refer to the periplasmic (chloride entrance; proton exit), central, and intracellular (chloride exit; proton entrance) site of the EClC channel. We place the symbol "Cl" in the ith position of this string, if the corresponding chloride binding site Cl (i) is occupied by chloride. Empty chloride binding sites are denoted by zero 0 at the corresponding string position. The chloride binding site Cl (1) can be empty or blocked by Glu-148 indicated by 0 or E at the first string position, respectively. According to our notation the chloride occupation state of the 1OTS crystal structure is represented by E Cl Cl. A possible transient protonation of the chloride at the central site (Cl (2) ) is denoted by the string Cl HCl Cl. The protonation states of the PT chain are denoted by a second string of characters appended to the first string, separated by semicolon. The characters of this string denote only those residues participating in the PT pathway that carry an excess proton starting with Glu-203 (proton entrance site) and ending with Glu-148 (proton exit site) with a linear chain of four (five in case of the Y445F mutant) buried waters W1-W4 (W1-W5). If in chloride occupation state Cl Cl Cl EClC carries a proton, e.g. at Glu-148 and W4 of the PT pathway, the complete string reads Cl Cl Cl; W4 E148. Any such string refers to an EClC structure with geometry-optimized hydrogen and buried water positions in the presence of the corresponding chlorides and protons in the transport pathways. This is why we use the notation "chloride occupancy," "protonation state," and "conformation" interchangeably.
Atomic Coordinates of EClC-All conformations considered in the present study are ultimately based on the 1OTS crystal structure (14). We removed all antibody chains and crystal waters except for 354 and 121 in chain A and B, respectively, and added hydrogen positions with CHARMM34 (22) followed by geometry optimization fixing all heavy atoms. Next, we modeled three additional buried waters in between Glu-203 and Glu-148 by translating copies of the crystal waters 354 and 121 parallel and anti-parallel to the Tyr-445-CB-Tyr-445-OH axes forming a linear water chain (W1-W4). First we optimized solely the positions of the four waters. A second optimization step also involved the protein environment of the four buried waters within a distance of 10 Å using the all-hydrogen CHARMM22 force field (23). The atoms of residues 28, 203, 445, 54, 55, and 148 as well as the chloride ions were held fixed. Other protein atoms were subject to a harmonic constraint using mass weighting and a small force constant of 0.1 kcal mol Ϫ1 Å Ϫ2 . For the Y445F mutant we simply deleted the hydroxyl group of Tyr-445 in 1OTS and performed the same protocol but added a fourth buried water (W5) at that position. The resulting coordinates correspond to the so-called "closed" conformation E Cl Cl in which Glu-148 occupies the external chloride binding site Cl (1) . A second opened conformation where Glu-148 is removed from Cl (1) and three chlorides are bound to EClC corresponds to Cl Cl Cl. It was modeled by means of constrained energy minimizations of the 1OTS EClC structure using the 1OTU crystal structure of the E148Q mutant (14) as a template because the latter had a significantly lower resolution. Details are described in the supporting information. Starting coordinates corresponding to EClC occupancy states other than E Cl Cl and Cl Cl Cl were prepared by deleting one, two, or all of the three chlorides per EClC monomer. For every EClC state considered we re-optimized hydrogen coordinates and the positions of the buried water molecules without constraints.
Computing EClC Energetics-This project was inspired by an elegant MD study of the KcsA potassium channel (24) in which the investigators determined the most likely ion permeation pathway by evaluating the free energies of all 2 4 ϭ 16 possible (potassium) occupancy states. In the case of EClC, the situation is more complex as there are at least 12 ϫ 7 ϩ 6 ϭ 90 possible states of the putative PT pathway per monomer. There are 12 ϭ 2 3 ϩ 2 2 different chloride occupancies (2 3 for the opened state with 3 chlorides and 2 2 for the closed state with 2 chlorides) combined with the protonation states of 6 titratable groups (Glu-203, W1-4, Glu-148) containing none or a single excess proton making up 7 states and 6 additional protonation states if the central chloride, that can carry a proton transiently, is present. This number explodes theoretically to 2304 possible combinations if one also considers states with two excess protons that we did, albeit only for selected chloride occupancies. Obviously, combinatorics sets a severe limit on the level of detail of the simulations. Therefore, we performed somewhat less detailed but much faster continuum electrostatic energy computations to characterize the different transporter states. In the past we demonstrated how useful electrostatic theory is in characterizing electron and proton transfer reactions in systems that were even larger than EClC (16,25,26). The supporting material explains the methodology in all details. Here, we give only a short description; electrostatic potentials are obtained by numerical solution of the linearized Poisson-Boltzmann equation using the program APBS (27,28) as in previous applications (20,29), except that the presence of the membrane was approximated by a slab volume around the protein as thick as 18 Å. From the potentials one can calculate the electrostatic component G E of the total free energy of EClC as a function of its protonation state S ជ H ϩ and chloride occupancy S ជ Cl Ϫ: The free energies G T, G C , G Cl Ϫ refer to ionization of titratable groups, residue conformations, and Cl Ϫ translocation from solvent to protein as explained below. The components represent the protonation states of the N T ϭ 78 -81 titratable groups per EClC monomer (depending on the chloride occupancy). In analogy, the three-component vector S ជ Cl Ϫ describes the chloride occupancy. G T measures the ionization energies of the protein titratable groups (30,31), In the above equation ␦ ϭ s H ϩ Ϫ r H ϩ equals zero if the titratable group is in its charge neutral reference state (r H ϩ ϭ 1 for acids and r H ϩ ϭ 1 for bases). Hence, G T becomes zero if all titratable groups are in their neutral reference state (i.e. s H ϩ ϭ 0, ϭ 1, … N T ). The intrinsic pK A (pK A ) is the pK A of the titratable group if all other titratable groups are in the reference protonation state. The pair-wise interaction between two titratable groups (, ) in absence of all other charges in the protein is given by the matrix elements W,. Note, that the intrinsic pK A values are themselves functions of the chloride occupancy state of EClC, a feature that will be explained below. Because the total number of possible protonation pattern (described by S ជ H ϩ) is very large in the case of EClC, we applied a Metropolis Monte-Carlo sampling method (31,32). Furthermore, we replaced G T in Equation 1 by mean field values G T (S ជ H ϩ red ,S ជ Cl Ϫ) created by averaging over the protonation states of the residues outside of the PT pathway, i.e. the average free energies obtained from an Metropolis Monte-Carlo sampling where the protonation of the titratable residues involved in the PT pathway were fixed in a specific state. Hence, the components of the reduced state vector S ជ H ϩ red correspond to the residues Glu-203, W1-W4, (W5), Cl (2) , and Glu-148. The energy required to protonate any group in the PT pathway of EClC can be calculated as follows where or 0 in the argument of G T and G C denote there is an excess proton at residue or none, respectively. In our multiconformational approach (20, 32) ⌬G prot accounts for the structural relaxation of EClC in a limited fashion; upon protonation of any group in the PT pathway and binding or removal of chloride, we re-optimized all hydrogen atoms and buried waters of EClC using CHARMM34 (22). Correspondingly, the quantities pK A and W, were computed from numerical solutions of the linearized Poisson-Boltzmann equation for every possible conformation {S ជ H ϩ, S ជ Cl Ϫ} (involving 141 conformations for WT EClC and 153 in case of the Y445F mutant). A detailed explanation of how this was done can be found in the supplemental material. We also needed to calculate the conformational energies G C characterizing the interaction between the reference states r H ϩ of the titratable and non-titratable residues in every conformation (31). Details on this procedure are given in the supplemental material. Protein flexibility not explicitly accounted for by different conformations is implicitly included in the dielectric constants (33): ⑀ p ϭ 4 for the protein and ⑀ m ϭ 4 for the membrane slab, and ⑀ w ϭ 80 for water. The pH was fixed at 4.5, where EClC operates optimally (34).
If we want to rank the various states of EClC according to energy based on Equation 1, we also needed to account for the energy G Cl Ϫ to move chlorides from solvent to protein. For simplicity, we evaluated G E , Equation 1, relative to the EClC reference state 0 0 0 corresponding to S ជ H ϩ Ј ϭ 0 ជ and S Cl Ϫ ϭ 0 ជ in which no chlorides are present in EClC and no excess protons rest on the residues establishing the PT pathway, (Eq. 4) where G T , G C , and G Cl Ϫ vanish at (S ជ H ϩ red ,S ជ Cl Ϫ) ϭ 0 ជ by definition. ⌬G E can now be calculated from the thermodynamic cycle depicted in supplemental Fig. S1. Accordingly, G Cl Ϫ (S ជ Cl Ϫ) is given by, with G (Cl Ϫ 3 HCl) ϭ ϪRT(pH Ϫ pK A HCl ) being the protonation energy of chloride at pH 4.5 in aqueous solution, G solv (HCl) the solvation energy of hydrochloride in aqueous solution, and G self (HCl) the electrostatic self-energy of hydrochloride scaled by ⑀ p ϭ 4. The symbol k simply counts the number of chlorides bound to EClC, i.e. k ϭ ͉S ជ Cl Ϫ͉ 2 . Using pK A HCl Ϸ Ϫ6, we obtained G (Cl Ϫ 3 HCl) ϭ 60.3 kJ mol Ϫ1 , G solv (HCl) ϭ Ϫ3.2 kJ mol Ϫ1 , and G self (HCl) ϭ Ϫ17.1 kJ mol Ϫ1 . Therefore, Equation 5 can written simply as

RESULTS AND DISCUSSION
We were successful in stabilizing in total four buried waters (W1-W4, where W1 and W2 replace the crystal water) yielding a single-filed water wire between Glu-203 and Cl (2) . Because of the additional buried water W4, the water wire establishes a new potential PT pathway (P3, Fig. 2) allowing protons to move from the previously unpredicted water W4 onto Glu-148 via Cl (2) assuming transient formation of hydrochloride. We identified a homologous pathway in both the Y445F and E203H mutants of EClC explaining their almost WT-like activity levels. Using electrostatic energy computations we tested the plausibility of P3 by assessing the protonation energies of the involved titratable groups. Moreover, based on characterization of a large number of possible chloride and proton occupancies, we propose a transport cycle fulfilling the observed 2:1 chloride/proton stoichiometry and being consistent with the observed experimental transport rates.
Buried Water-Water W1 forms five H-bonds, and the next neighboring water W2 forms four H-bonds. Four of the Hbonds of W1 involve surrounding polar protein groups; one is with W2 (details are given in the supplemental material). Despite these favorable protein-water interactions, neither W1 nor W2 is observed in the 1OTS protein crystal. Instead, the crystal water 121 of chain B lies in the middle between W1 and W2 (at a distance of 1.4 Å) almost like the average of their coordinates (Fig. 2). In chain A, however, our modeled water W2 is located significantly closer to the original position of the crystal water 354 at a distance of 0.85 Å. This discrepancy between the two polypeptide chains might be related to a subtle asymmetry of the EClC dimer structure that also unravels in a nearby water cluster present in chain B but not in chain A as discussed later. Compared with W1-2 our prediction of the waters, W3-4 is based on more shaky grounds; W3 is stabilized solely by the two H-bonds with its neighboring water molecules W2 and W4. The last water W4 forms an H-bond with W3 and the hydroxyl-group of Tyr-445.
The buried waters are arranged in a single-filed water wire located on the Cl (2) side of the Tyr-445 phenol ring plane (Fig.  2). Together with the hydroxyl group of Tyr-445, the four waters in WT EClC establish a continuous H-bond network that bridges the 11 Å distance between the closest acidic oxygen of Glu-203 and the central chloride in Cl (2) . Thus, the water wire forms a functional PT pathway between Glu-203 and the proton exit site Glu-148. In bulk, water protons proceed by the Grotthuss mechanism (35). In a similar fashion (36,37), protons might be conducted along the water wire in EClC. Further downstream the protons reach the central chloride ion and could tunnel the remaining distance of about 4 Å toward the acidic oxygens of Glu-148 in the closed state.
In the Y445F mutant we were able to place an additional fifth buried water (W5) that forms an H-bond directly with chloride in Cl (2) (Fig. 2, inlay). Interestingly, we could not stabilize W5 in the cavity opening after deleting the Tyr-445 hydroxyl group. Instead, following geometry optimization, W5 was always displaced by 2.24 Å with respect to the center of this cavity. As a consequence, W4 was also displaced as compared with its place in the WT EClC structure by 1.63 Å. The positions of water W1-3 remained unchanged.
Because of the constraints used in the geometry optimizations, the structural perturbations necessary to place the waters into 1OTS were rather small; the root mean square deviation of the protein heavy atom positions from the original coordinates was 0.22 Å. Also note that the coordinates of residues known to be involved in PT events like Glu-148 and those that are potentially involved like Arg-28 and Tyr-445 were rigidly fixed at their crystal structure positions. Our conservative approach to preserve the crystal structure coordinates might be the reason for the average water-water H-bond length being 2.69 Å and, therefore, shorter than the value of 2.8 -2.9 Å typically found in bulk water. In tests, we also performed completely unrestrained geometry optimizations of the water wire, which stayed intact, and the average H-bond length increased slightly to 2.75 Å. In this case, the overall root mean square deviation increased significantly to a value of about 1.4 Å, which is, however, not uncommon for geometry optimizations of a protein crystal structure. However, electrostatic computations that we perform are quite sensitive to such structural variations. In the past, we were quite successful by staying as close as possible to the crystal structure coordinates. Therefore, we used the constrained structures after making sure no steric clashes are present in those.
We also tried to model the buried waters W1-W5 with Dowser (38), a program well established in predicting buried water positions. However, it is not suitable to detect water wires in hydrophobic environment as the program does not account for water-water interactions, creating a problem in the detection of the waters W3 and W4. Instead, we computed binding energies for all buried waters using CHARMM34 and compared them with Dowser's threshold energy of Ϫ12 kcal/mol. This threshold is used in Dowser's own force field (39) to decide whether a water is sufficiently stabilized in a specific position. The binding energies obtained were in the range between Ϫ25 and Ϫ19 kcal/mol using the constrained structures and between Ϫ37 and Ϫ24 kcal/mol using completely unconstrained, relaxed structures. These values were safely below the Dowser threshold. We also computed the electrostatic component of the binding free energy of the individual waters that included the structural reorganization of the neighboring water molecules upon binding. Here, we obtained values between Ϫ18 and Ϫ8 kJ/mol, which are considerably higher but still below zero. A more detailed discussion of these results can be found in the supporting information.
Relevant Chloride Binding States-Using Equation 4 we ranked all possible combinations of chloride occupancies and single protonation states of EClC according to their electrostatic energies ⌬G E . We compiled the results in Fig. 3. We use a shorthand notation described in detail in the methods section to denote the various proton and chloride occupation states in EClC; the symbol "Cl" indicates bound chloride in the periplasmic, internal, and intracellular binding site (from left to right, respectively), E means Glu-148 is blocking the external site, and 0 indicates an empty site.
If no excess proton is present in the putative PT chain, the lowest energy state was E Cl Cl in which two chlorides per chain are bound in the sites Cl (2) and Cl (3) , and residue Glu-148 resides in the Cl (1) binding pocket blocking access of chlorides from the periplasmic lumen. This happens to be the same configuration observed in the WT 1OTS crystal structures of EClC (13,14). Therefore, we assume the E Cl Cl conformation to play the role of a ground state of the inactive transporter. Under the same conditions the computed electrostatic energy of Cl Cl Cl exceeds that of E Cl Cl by 34 kJ/ mol. However, with protonated Glu-148 the conformation Cl Cl Cl; E148 was more stable than E Cl Cl; E148 by about Ϫ20 kJ/mol. How does protonation of Glu-148 affect the chloride occupancy? We observed a large pK A up-shift of the gating residue Glu-148 with a magnitude of 9 pK units upon binding of chloride in Cl (1) ; in E Cl Cl we computed the pK A of Glu-148 to be about 0.5, whereas it was 9.9 in Cl Cl Cl. This is due to the strong, repulsive charge-charge interaction between Glu-148 and the chloride bound at Cl (1) , which are separated by only 5 Å (O-Cl distance) in this conformation. A slightly smaller pK A shift of 7-8 pK units was observed in the conformations Cl Cl 0, Cl 0 Cl, and Cl 0 0. Upon binding of chloride in the Cl (1) site, the gating residue Glu-148 is converted from a strong acid into a base. This helps explaining the observed pH dependence of the chloride transport through EClC (34,40); chloride uptake at Cl (1) requires protonation of Glu-148. The same results were obtained for the Y445F mutant EClC.
Proton Transport-We propose a PT pathway that guides the proton starting at Glu-203 via a Grothuss-like mechanism along four buried waters to the central chloride bound in Cl (2) from where it would have to hop onto Glu-148, the putative proton exit site. In E Cl Cl, where the external gate is closed, the distance between the nearest Glu-148 acidic oxygen and Cl (2) amounts to 4.1 Å only, whereas in Cl Cl Cl this distance is doubled. In the latter case the proton could reach Glu-148 only via a detour using the chloride in Cl (1) , where it needs to overcome also a larger distance of 5 Å between Cl (1) and Glu-148. Therefore, we conclude the PT transfer pathway is only functional when the external gate is closed and a chloride is present in the central binding site Cl (2) . This leaves only two conformations, namely E Cl Cl or E Cl 0 (being isoenergetic according to our computations), in which the proton could reach the exit on Glu-148.
The putative proton entry site Glu-203 has been shown to be essential for PT; in the E203Q mutant of EClC both proton-chloride coupling (3) and proton turnover (9) were completely disrupted. Surprisingly, replacing glutamate with the base histidine shows essentially WT activity levels (9). Based on the crystal structure of the E203H mutant, we re-modeled the water wire in the same manner as for the WT EClC. We found the structure of the water wire in this mutant essentially unperturbed compared with WT EClC. Similar to Glu-203 the corresponding residue His-203 is involved in an Hbond with W1. Next, we computed the pK A of His-203 and obtained an unusually low value of Ϫ0.5 that can be explained by the strong repulsive interaction between the protonated  (2) , dashed line with filled circles. The chloride site occupancies were sorted with ascending electrostatic energy for the case of no excess proton setting the fully deprotonated state 0 0 0 with no chlorides to zero energy. This explains the staircase form of the thick solid line without symbols for the deprotonated state. The energy difference between two points on the same curve is required for chloride translocation and/or conformational switch of Glu-148 between its open and closed conformer. The vertical energy gap between two points on different curves is the energy required for the corresponding PT process. More specifically, differences between two points in the diagram where one lies on the staircase curve and the second on any other curve provide the energy to protonate the corresponding residue. A similar plot corresponding to the Y445F mutant can be found in the supplemental Fig. S2.

Charge Transport in the ClC-type Chloride-Proton Anti-porter
state of His-203 and the nearby Arg-28, which donates an H-bond to His-203 as it does in the WT structure to E203. With a value of Ϫ0.9 the pK A of Glu-203 is qualitatively the same in the WT structure. In conclusion, His-203 effectively mimics all structural and chemical properties of Glu-203 relevant for efficient PT up-take and also release. We expect a similar behavior for the mutant residues E203K and E203R for which, too, significant PT rates were reported (9).
Based on the experimental evidence, it is only natural to suspect Glu-203 to be responsible for the proton up-take in EClC. Subtle structural aspects, however, do not seem to support this notion. First of all, this residue, although certainly located closely under the protein surface, is practically not solvent-accessible (total solvent-accessible surface area of 0.2 Å 2 in chain A and vanishing in chain B). Second, the surface area around this residue is characterized by a positive electrostatic potential, which is caused by the near-by residue Arg-28 being partially solvent-exposed with a total solvent-accessible surface area of 12 Å 2 . This would rather repulse than attract intracellular protons entering the protein. Correspondingly, in R28L mutants proton translocation does not change relative to WT EClC (3). Interestingly, in the environment of Glu-203 there is a second glutamate Glu-113 within H-bond distance. As has been noticed before (41), Glu-113 has an unusually elevated pK A (we computed pK A ϭ 7) and is, hence, protonated at pH 4.5. Like Glu-203, Glu-113 is not accessible from the intracellular lumen. However, its total solvent-accessible surface area amounts to about 2 Å 2 due to an internal cavity (Fig. 4). In the 1OTS crystal structure this cavity is empty in chain A but filled with four crystal waters (194, 214, 265, and 326) in chain B (Fig. 2). The water-filled cavity connects Glu-113 with a third glutamate Glu-117 located at the intracellular protein surface. The cavity is characterized by a negative electrostatic potential (Fig. 4). Therefore, it is tempting to speculate that protons at the protein surface are "sucked in" through this cavity, driven by the apparent, high proton affinity of Glu-113. This view is supported by the fact that Glu-113 is a conserved residue within prokaryotic homologues of EClC, whereas in many eukaryotes, like e.g. in the human transporters hClC4, hClC5, and hClC7 it is replaced by the protonated residue lysine, which may act in a similar fashion. Glu-117 is less strictly conserved but was replaced by the very similar aspartate in all prokaryotic anti-porters checked. In our energy computations we forced Glu-113 to deprotonate and found Glu-203 to become protonated in response, indicating a strong coupling between these titratable residues. Hence, Glu-113 might serve as a proton sink on the intracellular side of EClC that, on demand, injects a proton into the proton conducting pathway. Anyway, this residue has unusual properties and deserves closer attention. Constructing knock-out mutants at this position and assessing how it affects proton transport would be worthwhile, which to our knowledge has not been done yet.
We postulate transient protonation of the central chloride at Cl (2) and, hence, transient production of HCl in the PT process. This sounds implausible at first, as HCl is a classical example of a strong acid in aqueous solution. However, this mechanism was proposed before (42) based on several experimental facts that are hard to rationalize otherwise. In a combined crystallographic and electrophysiological study (4), the degree of coupling between chloride and proton translocation was found to strongly correlate with halide occupancy at Cl (2) , suggesting that the presence of an anion in Cl (2) is required for PT. Furthermore, EClC is capable of conducting polyatomic pseudo-halides like SCN Ϫ or SeCN Ϫ but cannot transport protons under these conditions (6). Anomalous x-ray scattering experiments accompanying this study revealed that the central Cl (2) site was empty in EClC co-crystallized with SeCN Ϫ , probably because its ion radius is too large to fit into the Cl (2) binding site. Again, the results implied that the presence of an anion in Cl (2) is a precondition of PT.
We calculated the protonation energy of chloride in Cl (2) according to Equation 3. It turned out to vary with the chloride occupancy in the range between ϩ67 kJ/mol in Cl Cl Cl and ϩ107 kJ/mol in 0 Cl 0 lacking chlorides at the peripheral sites. Obviously, the lower protonation energy of the central chloride is due to attractive interactions of the proton with the peripheral chlorides. The protonation energies correspond to pK A values in the range between Ϫ7 and Ϫ14 for the central chloride, which are values even below pK A Ϸ Ϫ6 for HCl in aqueous solution (43). However, the protonation energy of Cl (2) exceeds that of the nearby waters W3 and W4 by typically not more than 20 kJ/mol (Fig. 3). The relatively high protonation energies of these waters (their pK A values between Ϫ10.5 and Ϫ3.8) are due to the local hydrophobic protein environment that does a poor job of stabilizing an excess FIGURE 4. Intracellular proton uptake by EClC; the protein is represented as a color-coded isoelectrostatic potential surface indicating the boundary between water and protein dielectric lumen (the electrostatic potential was calculated including Cl (2) and Cl (3) , although the latter is not visible in this picture). The PT pathway between Glu-203 and Glu-148 that we proposed is indicated by a thick black line and labeled P3. Glu-203 is not directly solvent-accessible, and hence, protons have to enter the protein interior via other sites. The structure provides two possible routes for bulk protons from the intracellular side: The protons could enter from a crevice in the interface region of the homo-dimeric complex via Arg-28 as indicated by the yellow dotted line; alternatively, protons could be provided by Glu-113, which is protonated in equilibrium and connected with Glu-117 on the protein surface via a water-filled cavity (colored in red according to its negative potential). The figure was prepared using VMD (51).
proton. Consequently, protons that arrive on W4 being closest to Cl (2) are highly energized and because of that would be capable to move on to chloride bound in Cl (2) . In light of the physiological role of EClC (1,40), i.e. to pump protons against an osmotic gradient, it makes sense to position very acidic groups at the end of the PT chain to enable protons arriving at the exit site Glu-148 to be released into the periplasmic lumen even at very low pH.
Role of Tyr-445-Protonation energies obtained for the Y445F mutant are qualitatively the same as in WT EClC. However, the Y445F mutant EClC structure implies differences in the PT mechanism. In this mutant the additional terminal water W5 is directly hydrogen-bonded to chloride in the Cl (2) site. In WT EClC, on the other hand, the terminal water W4 is not. Hence, Tyr-445 connecting W4 with Cl (2) must be involved somehow in the proton transfer reaction. As discussed in the supporting information, Tyr-445 is not a titratable residue. According to our electrostatic computations, creation of an ionized intermediate would require much more energy than the roughly 70 kJ/mol required to protonate the central chloride and, thus, would hamper efficient PT between Tyr-445 and Cl (2) . To explore the possible protonation states of this system, we set up a reduced quantum chemical model and performed a vacuum geometry optimization with an excess proton localized on W4. The geometry-converged structure contained hydrochloride and a neutral Tyr-445 donating a H-bond to the initially protonated and now deprotonated water W4 (Fig. 5). The total electronic energy decreased continuously during geometry optimization, indicating that there is no metastable ionized intermediate of Tyr-445. Rather, PT between W4 and the central chloride proceeds in a concerted fashion by an electrophilic attack of the proton residing on W4 triggering the deprotonation of the hydroxylic group of Tyr-445.
Other residues can replace tyrosine in this position and maintain proper function of the transporter (4,7,8). However, the PT activity depends on the size of the mutant residue in position 445. Correspondingly, the mutant Y445F, where tyrosine is replaced with phenylalanine of almost equal size, essentially retains WT activity levels both in terms of chloride and proton turnover (4). On the other hand, replacing Tyr-445 with alanine causes a complete breakdown of PT, whereas the chloride transport is not impaired (4, 7). We infer from these facts that Tyr-445 serves two fundamental functions in WT EClC; (i) it defines the exact geometry of the central chloride binding site Cl (2) and facilitates chloride binding by providing a strong H-bond, and (ii) it creates a narrow, well defined hydrophobic pore for the linear water wire keeping the protonation energy of the waters at a high level to allow the exit of protons against large proton concentration. These conditions are presumably also fulfilled when substituting Tyr-445 by phenylalanine or lysine. In the former case we showed explicitly that extra water (W5) can replace the hydroxyl-group of Tyr-445. For smaller substituents like alanine with seven non-hydrogen atoms less than tyrosine, the linear water wire degenerates in a bulky water cluster that may contain up to seven waters more than WT EClC. The positions of these waters in the larger hydrophobic cavity are likely disor-dered, such that the position of a potential central chloride is less well defined or may not bind at all.
A Model for the Chloride-Proton Exchange Mechanism-For EClC to transport chloride across the membrane it must evolve through a series of chloride occupation states. According to our energy computations the most basic transport cycle would consist of the following states: E Cl Cl, 0 Cl Cl, Cl Cl Cl, Cl Cl 0, Cl 0 Cl, 0 Cl Cl, E Cl Cl. The net result of the complete cycle is translocation of a single chloride per cycle from the periplasmic medium into the intracellular space. Based on this simple process, we constructed an extended transport cycle incorporating both chloride and proton transfer steps that effectively describe the exchange of two chlorides with one proton (see the reaction scheme in Fig. 6). The cycle has two interesting features. (i) We needed to assume that the proton-conducting chains are doubly protonated during a specific phase of the whole cycle where one proton resides on W4 and a second on Glu-148, which is leftover from the previous cycle. (ii) EClC evolves through states with one, two, Optimization was performed with the B3LYP functional and basis set 6 -31g**ϩϩ as implemented in Jaguar 7.6 (52). As starting coordinates, we took the conformation corresponding to EClC state E Cl Cl; W4 from chain A and removed all atoms except of parts of Tyr-445, W2-W4, Glu-148, and the protein ligands around the central chloride binding site Cl (2) (side chain of Ser-107 as well as backbone atoms of Ile-356 and -357). After 40 iterations, the excess proton on W4 was localized on the hydroxyl-group of Tyr-445, and the central chloride was protonated. The figure was generated using MOLDEN (53). and three chlorides, and as a consequence, all three chloride binding sites are involved in the exchange cycle although it has been postulated that the Cl (3) site is not essential (42).
For clarity, we start our discussion of the transport cycle in a state with opened external gate 0 Cl Cl; E148, where Glu-148 is protonated. Under these conditions, binding of chloride in Cl (1) lowers the electrostatic energy ⌬G E according to Equation 4 such that the system proceeds to state Cl Cl Cl; E148. Because of Glu-148 being strongly basic (pK A ϭ 9.9) under these conditions, the proton is stabilized at Glu-148 and, thus, stays attached to the protein. Because the corresponding protonation energies are smallest in Cl Cl Cl; E148 among all other states considered, it is very likely that intracellular proton uptake occurs in this phase of the transport cycle. From now on the PT chain carries two excess protons where one is situated on Glu-148. We assumed that the newly up-taken proton goes directly to W4 where it has to wait for several reasons. The energy required for the subsequent PT between W4 and Cl (2) is considerably higher in state Cl Cl Cl; W4 E148 than later in state E Cl Cl; W4, where Glu-148 is deprotonated. Furthermore, an earlier PT would be unproductive because the proton could not reach the exit site Glu-148 in state Cl HCl Cl; E148 and would be transported back into the intracellular lumen by the chloride transfer steps required to move Glu-148 back into site Cl (1) . Alternatively, leakage into the periplasmic space might occur starting from Cl HCl Cl; E148 via PT to the periplasmic chloride in Cl (1) as the states Cl HCl Cl; E148 and HCl Cl Cl; E148 are roughly isoenergetic.
In state Cl Cl Cl; W4 E148, the chloride at Cl (3) is released into the intracellular lumen (Cl Cl 0; W4 E148) followed by two chloride transfer steps, effectively shifting the two chlorides in Cl (1) and Cl (2) by one position toward the internal lumen into the sites Cl (2) and Cl (3) , respectively, thus reaching state 0 Cl Cl; W4 E148. With removal of a chloride at Cl (1) , Glu-148 is converted back into a strong acid and can release its proton into the extracellular lumen. From now on, the proton-conducting chains of EClC contain only a single excess proton. Deprotonation of Glu-148 triggers a conformational change closing the external gate. In the resulting state, E Cl Cl; W4, PT to the central chloride in Cl (2) requires virtually no energy, producing the EClC state E HCl Cl. If the cycle is to fulfill the observed 2:1 stoichiometry, we have to assume the release of a second chloride at Cl (3) in this state, thus reaching E HCl 0. Because the hydrochloride proton comes from the hydroxyl group of Tyr-445, it initially forms a strong H-bond with that group (O-H distance 1.7 Å). To reach the proximal acidic oxygen of Glu-148, the proton needs to rotate around the chorine atom at Cl (2) by more than 100°to form a rather weak H-bond with that oxygen (O-H distance 3.0 Å). Therefore, we believe that this process is kinetically hindered, and it is more likely that the chloride at Cl (3) exits into the intracellular lumen before the proton reaches Glu-148. This difference in kinetics is at the crossroads of 1:1 versus 2:1 Cl-H stoichiometry. This kinetic argument leaves the open question, if besides a scenario of 2:1 stoichiometry, there may be other conditions allowing also 1:1 stoichiometry. ; open squares, excess proton at W4; green filled circles, excess proton at Cl (2) . Vertical blue arrows indicate PT steps, whereas green arrows represent chloride transfer steps. The y-components of the arrows provide the energies (given by the numbers in kJ/mol at the upward-pointing arrows) required to move protons or chlorides in the corresponding transfer steps. ful to have crystal structures at higher resolution and at physiological pH values.