Divergent Cl- and H+ pathways underlie transport coupling and gating in CLC exchangers and channels

The CLC family comprises H+-coupled exchangers and Cl- channels, and mutations causing their dysfunction lead to genetic disorders. The CLC exchangers, unlike canonical 'ping-pong' antiporters, simultaneously bind and translocate substrates through partially congruent pathways. How ions of opposite charge bypass each other while moving through a shared pathway remains unknown. Here, we use MD simulations, biochemical and electrophysiological measurements to identify two conserved phenylalanine residues that form an aromatic pathway whose dynamic rearrangements enable H+ movement outside the Cl- pore. These residues are important for H+ transport and voltage-dependent gating in the CLC exchangers. The aromatic pathway residues are evolutionarily conserved in CLC channels where their electrostatic properties and conformational flexibility determine gating. We propose that Cl- and H+ move through physically distinct and evolutionarily conserved routes through the CLC channels and transporters and suggest a unifying mechanism that describes the gating mechanism of both CLC subtypes.


Introduction
The CLC (ChLoride Channel) family is comprised of Clchannels and H + -coupled exchangers whose primary physiological task is to mediate anion transport across biological membranes (Accardi, 2015;Jentsch and Pusch, 2018). The human genome encodes nine CLC homologues, four (CLC-1,-2, -Ka, -Kb) are Clchannels that reside in the plasma membrane and five (CLC-3 through À7) are 2 Cl -:1 H + antiporters that localize to intracellular compartments along the endo-lysosomal pathway. Mutations in at least five of the human CLC genes lead to genetically inherited disorders of muscle (Thomsen and Becker type Myotonia congenita), kidney (Bartter Syndrome Type III and IV, Dent's disease), bone (Osteopetrosis) and central nervous system (neuronal ceroid lipofuscinosis, retinal degeneration, deafness, syndromic intellectual disability, seizure disorders), highlighting the fundamental roles of CLC channels and transporters in human physiology (Jentsch, 2008;Stauber et al., 2012;Hu et al., 2016;Jentsch and Pusch, 2018;Palmer et al., 2018). Several disease-causing mutations occurring in CLC channels and transporters impair the response to the physiological stimuli regulating their activity, such as voltage, pH and nucleotide concentration (Accardi, 2015;Alekov, 2015;Bignon et al., 2018;Jentsch and Pusch, 2018).
High-resolution structural information on the CLC-ec1 and cmCLC exchangers (Dutzler et al., 2002;Feng et al., 2010) as well as CLC-K and CLC-1 channels (Park et al., 2017;Park and MacKinnon, 2018;Wang et al., 2019) revealed that both CLC subtypes share a common dimeric architecture, where each monomer forms physically distinct and functionally independent ion permeation pathways. This pathway is defined by three anionic binding sites (Dutzler et al., 2002;Dutzler et al., 2003;Accardi and Picollo, 2010;Feng et al., 2010;Park et al., 2017; Figure 1A). The external and central sites, S ext and S cen , are alternatively occupied by the permeant anions or by the negatively charged side chain of a conserved glutamic acid, Glu ex ( Figure 1A). The internal site, S int , binds anions weakly and likely serves as a recruitment site of intracellular permeating ions Picollo et al., 2009). The presence of a fourth binding site, S ext* , in direct contact with the extracellular solution has been proposed based on electrostatic calculations and is supported by electrophysiological (Lin and Chen, 2000), structural (Park et al., 2019), and molecular dynamics simulation studies (Faraldo-Gó mez and Roux, 2004;Mayes et al., 2018).
The CLC Clpore can adopt at least three conformations, differentiated by the position and protonation state of Glu ex (Dutzler et al., 2002;Dutzler et al., 2003;Feng et al., 2010; Figure 1A). Extensive functional work suggests that cycling through these conformations underlies the Cl -/H + exchange cycle in the CLC transporters and gating in the CLC channels (Lísal and Maduke, 2008;Feng et al., 2010;Feng et al., 2012;Basilio et al., 2014;Khantwal et al., 2016;Vien et al., 2017). All proposals accounting for the CLC exchange cycle suggest that these transporters do not utilize a classical 'ping-pong' mechanism of antiport, where the transporter sequentially interacts with one substrate at a time (Picollo et al., 2012). Rather, the CLCs were proposed to simultaneously bind (Picollo et al., 2012) and translocate Cland H + through partially congruent pathways (Accardi et al., 2005;Zdebik et al., 2008;Figure 1B). The H + pathway is delimited by two glutamic acids that are conserved in the CLC transporters: Glu in serves as the intracellular proton acceptor and is distal from the Clpermeation pathway, while Glu ex is the extracellular intersection between the H + and Clpores ( Figure 1B; Accardi et al., 2005;Zdebik et al., 2008). Several models have been proposed for the exchange cycle of the CLCs (Miller and Nguitragool, 2009;Feng et al., 2010;Basilio et al., 2014;Khantwal et al., 2016). However, none provided a molecular mechanism describing how the Cland H + ions bypass each other while moving in opposite directions through the permeation pathway. These proposals share the critical assumption that protonation of Glu ex within the pathway destabilizes its binding to S cen and/or S ext , favoring its exit from the pathway. While this mechanism readily explains outward H + transfer, it requires that during H + influx a protonated and neutral Glu 0 ex outcompetes the negatively charged Clions bound to the anion-selective S ext and S cen sites. This is an energetically unfavorable transition, which should result in intrinsic rectification of transport. Indeed, free-energy calculations show that a protonated Glu ex encounters a high energy barrier to enter the Clpermeation pathway (Kuang et al., 2007). In contrast, the CLC exchangers function with comparable efficiency in the forward and reverse directions (Matulef and Maduke, 2005;Leisle et al., 2011;De Stefano et al., 2013), suggesting that the entry and exit of ; center, protonated Glu ex (mimicked by E148Q mutation) reaching out of the ion pathway (PDBID: 1OTU); right, Glu ex bound to S cen (cmCLC WT, PDBID: 3ORG). Selected residues are shown as sticks and Clions as green spheres. (B) The partially congruent Cland H + pathways are shown in CLC-ec1 WT structure (Glu ex and Glu in are shown in yellow, Clions as green spheres). the protonated Glu ex from the Clpathway are equally favorable. Another free-energy calculation study shows that the reaction rates of deprotonation of Glu 0 ex in both directions of the H + pathway are comparable to each other, while both reactions are coupled to Clat S cen (Lee et al., 2016). Further, the mechanisms regulating the release of ions from S ext and S cen and their coupling to the movement of Glu ex and of H + through the protein are unknown. While biochemical evidence suggests that ion release is rate-limited by a conformational step (Picollo et al., 2009), no release mechanism has been identified. Therefore, two key mechanistic features at the heart of the H + :Clexchange mechanism of the CLCs, the pathways and the coupling mechanism of the substrates, remain unknown.
Here we combined molecular dynamics simulations with biochemical and electrophysiological measurements, and atomic mutagenesis to investigate the mechanism of H + /Clexchange. We find that, contrary to previously proposed models, a protonated Glu ex does not move through the Clpore. Rather, we identify two highly conserved phenylalanine residues that form an aromatic slide which allows the protonated (neutral) Glu ex to move to and from S cen without directly competing with Clions for passage through the anion-selective pathway. Further, we show that the rotational movement of the central phenylalanine residue, that enables the formation of the aromatic slide, is an important regulator of ion movement within the pathway, providing the molecular mechanism for the coupled exchange of H + and Clby the CLC transporters. Mutating these residues in prokaryotic and mammalian CLC exchangers severely impairs transport indicating that the role of these aromatic side chains is evolutionarily conserved. Since these phenylalanine residues are highly conserved throughout the CLC family, we hypothesized the role of the aromatic slide might be evolutionary conserved also between CLC channels and exchangers. Indeed, we found that they play important roles in the gating of the prototypical CLC-0 channel. Using atomic-scale mutagenesis, we probed how the aromatic slide residues interact with Glu ex in CLC-0, and found that the central phenylalanine interacts electrostatically with the gating glutamate, and that its conformational rotation is necessary for channel gating. We propose a novel mechanism for CLC mediated H + :Clexchange, where the Cland H + pathways are distinct and intersect only near the central ion binding site.

Results
Molecular dynamic simulations suggest F357 controls entry and release of Clfrom S cen/title We used molecular dynamics simulations to probe the energetic landscape of ion movement through the permeation pathway of CLC-ec1 to ask whether it is regulated by conformational rearrangements of the pore. The first state we considered is one with Glu ex protonated and out of the pathway, an E148Q-like state (Dutzler et al., 2003;Figure 1A), so that all binding sites are accessible to ions. The first Clion to reach the pore preferably binds to S cen or S ext , this remains true when another ion is bound to S extx002A * ( Figure 2A, dashed line (i), 2B). When a Cloccupies S int , the second ion can occupy S cen or S ext , though binding to S cen is less favorable by 2 kCal mol À1 (Figure 2A, dashed line (ii), 2B). Unexpectedly, our calculations suggest that simultaneous binding of Clto S cen and S ext does not correspond to a local free energy minimum, i.e. to a (meta-)stable state. The doubly occupied state, identified by an asterisk on Figure 2A, is unfavorable by 4 to 6 kCal mol À1 in comparison to the binding of a single ion to S ext or to the doubly occupied configuration with ions in S int and S ext .
Since current models postulate a state with simultaneous occupancy of S cen and S ext , we set out to investigate what gives rise to this energetic barrier. Analyzing the conformational sampling underlying the PMF calculations, we noted that simultaneous occupancy of S cen and S ext correlated with fluctuations of the rotameric state of F357 ( Figure 2-figure supplement 1A,B). This residue forms part of the Clpermeation pathway of the CLC channels and transporters by coordinating ions in S cen and S ext with its backbone amide (Park et al., 2017). In our simulations, we find that the F357 side chain exists in equilibrium between two rotameric states with c1 angles of À160˚('up'), as seen in the crystal structures of WT and mutant CLC-ec1 (Dutzler et al., 2002;Dutzler et al., 2003), and of À70˚('down') ( Figure 2E). To test whether this conformational rearrangement affects Clpermeation, we restrained c1 (F357) to the 'up' or 'down' rotamers and determined the energetic landscape of ion binding in these conformations ( Figure 2C,D). We find that when F357 is in the 'down' state, a free energy barrier of~12 kCal mol À1 opposes ion movement within the pore, and the S cen / S ext double occupancy state remains unstable ( Figure 2C). In contrast, when F357 is constrained to the 'up' conformer, two Clions can simultaneously bind to S cen and S ext , and the different stable states along the permeation pathways are separated by free energy barriers of~2-4 kCal mol À1 ( Figure 2D). Reciprocally, the ion occupancy state directly impacts the conformation of F357: the 'down' state is favored when no ions occupy the pathway or when only S ext is occupied (Figure 2 The position of the different binding sites along the Z-axis is indicated. The Z = 0 coordinate is an arbitrary point defined as the center of mass of backbone atoms around S cen . The asterisk indicates the position of the doubly occupied state S cen /S ext . Each color of the iso-contoured map represents a dG of 2 kCal mol À1 , as indicated by the color scale. (B) The scheme illustrates the key ion occupancy states. Circles represent Clions identified with numbers (1, 2) that correspond to the axis labels (Z1, Z2) of the PMF plots; the labels at the top (i, ii) refer to the transitions indicated in panel A. The doubly occupied state S cen /S ext (*) is essential for ion transport but is not observed as a stable state in the PMF calculation of panel A. (C-D) The PMF calculation was repeated with a harmonic restraint applied to the side chain of residue F357 to maintain c1 around À70˚('down' conformer), (C) or À160˚('up' conformer), (D). Simultaneous Clbinding to S ext and S cen becomes energetically favorable under the 'up' conformer. Color scale is equivalent to (A). (E) F357 side chain exists in an equilibrium between two rotameric states ('up', c1=-160˚; 'down', c1=-70˚). The online version of this article includes the following figure supplement(s) for figure 2:  whether the dynamics of F357 are affected by the introduction of a crosslink between A399 and A432 ( Figure 2-figure supplement 2A), which inhibits transport of CLC-ec1 by preventing a conformational rearrangement involved in coupling between the intra-and extra-cellular gates (Basilio et al., 2014). We found that constraining the relative movements of A399 and A432 inhibits the transition of F357 between its rotamers by adding a free energy barrier of~6 kCal mol À1 (Figure 2-figure supplement 2), suggesting that this transition is part of the transport cycle. Taken together, these results suggest that the rearrangement of F357 is a critical determinant of the energy barrier height for ion movement within the CLC pore: the 'up' conformer of F357 is compatible with ion transport while its 'down' conformer disfavors ion transitions.

Rotation of F357 enables the formation of an aromatic slide
We next asked how Cland Glu ex interact along the transport cycle. As in the previous section, we first considered states in which Glu ex is protonated (Glu 0 ex ) and outside the Clpathway in an E148Q-like conformation ( Figure 3A). Analysis of the conformational sampling of the PMFs presented in Figure 2 reveals that, in this configuration, Glu 0 ex is stabilized by the carboxylate group of D54 via a water molecule ( Figure 3A) or by a hydrogen bond with the backbone of A189 ( Figure 3B). The simulations also reveal that the carboxylate group of Glu 0 ex rarely visits S ext and rather interacts with the aromatic ring of F190, even if S ext is free of Cl -( Figure 3C).
We then considered states in which Glu ex occupies S cen . We calculated the PMF describing the binding of Clto S ext for both the charged and protonated forms of Glu ex (Figure 3-figure supplement 1). The PMFs show that the binding of Clto S ext requires the protonation of Glu ex , in agreement with our previous work suggesting synergistic binding of Cland H + (Picollo et al., 2012). Interestingly, in the case of the protonated Glu ex , a free energy well is also observed at the level of S cen , initially occupied by the carboxylate group of Glu 0 ex . Inspection of the sampled structures reveals that, when S ext is occupied by a Cl -, the Glu 0 ex only partially occupies S cen . Its carboxylate group moves sideway toward F357 and away from the Clpathway. Two key conformations are observed. A first one in which a proton wire composed of two water molecules forms spontaneously between Glu 0 ex (E148) and Glu in (E203) ( Figure 3D), potentially allowing the deprotonation of Glu 0 ex . In this conformation, the carboxylate group of Glu 0 ex also forms a hydrogen bond with Y445. A second conformation reveals the possibility for Glu 0 ex to form a p-dipole interaction with the aromatic ring of F357 ( Figure 3E). The displacement of Glu 0 ex toward F357 allows the bound Clto reach S cen ( Figure 3F). The upward movement of F357 would bring the carboxylate group of Glu 0 ex in the vicinity of F190. These calculations suggest that the conformational rearrangement of F357 enables the formation of an aromatic slide through which a protonated Glu 0 ex can move to and from S cen without having to compete with the bound ions in the ion pathway.
It is important to note that in these calculations do not capture the full complexity of the energetic landscape of the interactions between ions in the CLC-ec1 pathway and F357, a task that would at minimum require a multidimensional PMF calculation involving at least three reaction coordinates: the position of the two Clions and c1(F357) angle. Additional reaction coordinates would likely be required such as the conformation and protonation state of E148, which interacts with both the ions and aromatic ring of F357, or that of conformational rearrangements taking place in CLC-ec1 during transport (Chavan et al., 2019). Thus, our calculations focus on the two key degrees of freedom of the system: (1) how the conformation of F357 affects ion occupancy of the CLC-ec1 pathway ( Figure  , so that that the transition along this degree of freedom is not fast enough to be thoroughly sampled in absence of a biasing potential, especially when ions enter or exit the S cen and S ext binding sites. Thus, special care must be taken in interpreting the sampling of the conformations of F357 ( While our finding that an inhibitory crosslink (Basilio et al., 2014) prevents the rearrangement of F357 (Figure 2-figure supplement 2), provides an initial validation of the mechanistic inferences of our MD simulations, the hypothesis that F190 and F357 play a critical role in determining Cl -/H + coupling and control a rate-limiting barrier for ion transport in CLC-ec1 requires experimental probing.
The aromatic slide residues are essential for Cl -:H + coupling and exchange in CLC-ec1 To test these hypotheses, we mutated them to alanine and determined the unitary transport rate and stoichiometry of the Cl -/H + exchange cycle. Both mutations slow the turnover rate and degrade the exchange stoichiometry ( Figure 4). The F190A mutant slows transport~4 fold ( Figure 4B Figure 2A, in which Glu 0 ex is initially placed on the extracellular side of the pore. Glu 0 ex is part of a hydrogen bond network involving D54 (A) or A189 (B). Glu 0 ex can also interact with the side chain of F190 (C). (D-F) Conformations of the pore extracted from a 1D PMF describing the binding of a Clto the pore when Glu 0 ex is initially bound to S cen (see Figure 3-figure supplement 1). A proton wire is spontaneously formed between Glu ex (E148) and Glu in (E203), which are bridged by two water molecules (D). Glu 0 ex can form a dipole-p interaction with F357 aromatic side chain, leaving S cen empty (E). Clmoves from S ext to S cen , while the side chain of Glu 0 ex is stabilized outside the ion pathway by its interaction with F357, and in proximity of F190 (F). The snapshots shown here appear spontaneously in PMF calculations in which the position of ions was restrained. The illustrated side chain and water wire conformations are observed in at least three sampling windows and remain stable for most of the 500 ps sampling time of a given window. The online version of this article includes the following figure supplement(s) for figure 3: As the F357 backbone lines S cen and S ext (Dutzler et al., 2002), we tested whether an alanine substitution affects the integrity of the binding sites in two ways. First, we used isothermal titration   calorimetry (ITC) to measure Clbinding to the F357A mutant and found that it has a WT-like affinity of~0.7 mM (Figure 4-figure supplement 1A). Second, we introduced the F357L substitution, which preserves side chain hydrophobicity and volume but removes the aromatic ring. We found that F357L affects the turnover rate and exchange stoichiometry like F357A (Figure 4-figure supplement 1B,C). Thus, the effects of the F357A mutant reflect the absence of the aromatic side chain rather than a structural disruption of the ion pathway. Taken together our functional results are consistent with the insights from our MD simulations and indicate that F190 and F357 play a key role in coupling and transport of CLC-ec1 by modulating the movement of Glu ex in and out of the pathway. To test this conclusion, we reasoned that the F190A and F357A mutations should not impair transport when introduced on the background of the H + -uncoupled E148A mutant (Accardi and Miller, 2004). The E148A mutant alone reduces Cl -turnover~3 fold compared to WT to~974 ion s À1 (Walden et al., 2007; Figure 4F). The E148A/F357A mutant transports Clat a similar rate of the parent E148A construct,~582 ion s À1 ( Figure 4F). Remarkably, the double E148A/F190A mutant 'rescues' the transport defect of the E148A parent single mutants and has a WT-like transport rate of 3,494 ions s À1 ( Figure 4F). The non-additivity of the effects of the F357A and F190A mutations with that of the E148A mutant qualitatively supports the idea that these Phe residues may interact with Glu ex during transport. Notably, F190 and F357 are among the highest conserved residues throughout the CLC family, respectively at~94% and~76% (Figure 4-figure supplement 2), suggesting that their functional role might be evolutionarily conserved. In the remainder of this work we will refer to these residues across different CLC homologues as Phe ex (F190 in CLC-ec1) and Phe cen (F357 in CLC-ec1).

The role of Phe ex and Phe cen is conserved in mammalian transporters
We asked whether the role of Phe ex and Phe cen is conserved in the mammalian CLC exchangers CLC-7 and CLC-5. Alanine substitutions of CLC-7 Phe ex (F301A) and Phe cen (F514A) reduced the amplitude of outward currents; the current at +90 mV is reduced by~25% for F301A and by~50% for F514A compared to WT ( Figure 5D). These effects could reflect either a reduction in turnover, as caused in CLC-ec1 by the corresponding mutations, or a reduced plasma membrane expression of the mutant transporters. Both mutations also left-shifted the voltage dependence of CLC-7 ( Figure 5E); with the F301A mutant decreasing rectification so that measurable currents could be seen at negative potentials ( Conversely, a 10-fold decrease in [H + ] ex (from pH 7.5 to pH 8.5) only mildly affected the reversal potential (< À10 mV; Figure 5-figure supplement 1B), consistent with the degraded coupling stoichiometry seen in the corresponding F190A mutant in CLC-ec1 ( Figure 4D). The inherent difficulties in controlling intracellular Cland H + concentration in oocytes prevented us from determining the exchange stoichiometry for WT and mutant CLC-7. While CLC-7 WT currents do not reach steadystate even after a 2 s activation pulse ( Figure 5A), activation of the F301A and F514A mutants is accelerated, enabling a more thorough characterization of their voltage dependence (Figure 5B,C; Figure 5-figure supplement 1C,D). The slow activation kinetics of WT CLC-7 prevent a precise determination of its V 0.5 , an osteopetrosis-causing mutant with accelerated kinetics was used to estimate the V 0.5 of CLC-7 to around +80 mV (Leisle et al., 2011). The F301A mutant shows a reduced voltage dependence, with an apparent P minx00A0 of~0.8 ( Figure 5F, Suppl. File 1). In contrast, the F514A shows a sigmoidal voltage dependent G-V which could be well fit with a Boltzmann function, with z~1 and V 0.5~+ 41 mV ( Figure 5F, Suppl. File 1), consistent with the left-shift seen in the normalized I-V curves ( Figure 5E). The corresponding mutations in the ClC-5 exchanger, F255A (Phe ex ) and F455A (Phe cen ), have qualitatively similar effects to those seen in CLC-7: the F255A mutant reduces voltage-dependent gating, resulting in measurable currents at negative voltages, and the activation threshold of the F455A mutant is left-shifted ( Figure 5-figure supplement 2). However, the fast gating kinetics and strong rectification of this homologue prevent a reliable determination of the G-V relationship and thus a quantification of the effects. Taken together, our results are consistent with the idea that Phe ex and Phe cen play a key role in regulating the movement of Clions and Glu ex in and out of the anion pathway. This hypothesis is also consistent with the finding that the voltage-dependence of mammalian CLC exchangers arises from the movement of permeating ions and of Glu ex through the electric field of the transport pathway (Smith and Lippiat, 2010;Zifarelli et al., 2012).

Phe ex and Phe cen are important for CLC channel gating
To probe whether the aromatic slide also plays a similar role in determining the voltage dependent gating of the CLC channels, we mutated Phe ex (F214) or Phe cen (F418) to alanine in the prototypical CLC-0 channel. Opening of CLC-0 channels is regulated by two processes: individual pores open and close independently in a process called single-pore gating and cooperatively during commonpore gating (Accardi, 2015;Jentsch and Pusch, 2018). Single-pore gating is thought to entail rearrangements of Glu ex similar to those underlying the exchange cycle of the transporters, while the mechanistic underpinnings of common-pore gating remain poorly understood (Accardi, 2015;Jentsch and Pusch, 2018). In the CLC-0 channel, single-pore gating is activated by depolarizing voltages ( Figure 6A,D), while common-pore gating is hyperpolarization activated and occurs on much     slower time scales ( Figure 6E,H). In the F214A and F418A mutants, the single-pore gate exhibits nearly voltage-independent open probabilities over the voltage range tested ( Figure 6B-D), suggesting that these mutations favor the open conformation of Glu ex (E166 in CLC-0). The voltage dependence of the common-pore gate is also affected in these mutants, as the minimal open probability increases, the gating charge decreases and V 0.5 shifts to more positive values ( Figure 6E-H). These results suggest that Phe ex and Phe cen are shared determinants of the voltage dependence of the common-and single-pore gating processes of CLC-0. These findings are consistent with the results of the MD simulations on the CLC-ec1 exchanger that the carboxyl group of Glu 0 ex interacts with the aromatic-slide residues to enter the pathway (Figure 3). Lastly, we note that the qualitative effects of the Phe ex mutants are conserved between CLC-7, CLC-5 and CLC-0, as in all three homologues the voltage dependence of the currents at negative voltages is drastically reduced ( Figure 5B,E,F; Figure 5-figure supplement 2B,D; Figure 6B,D). However, we cannot exclude that these observed effects are due to indirect effects of the alanine substitutions on the CLC proteins. Indeed, the interpretation of the effects of any mutation is often complicated by the multiple simultaneous changes in side-chain properties introduced by the substitution.

Atomic mutagenesis of the aromatic slide in CLC-0
To circumvent the limitations of alanine substitutions, we used non-canonical amino acid (ncAA) mutagenesis to selectively manipulate specific properties of the phenylalanines comprising the aromatic slide (Figure 7) using three derivatives: Cyclohexylalanine (Cha), 2,6diFluoro-Phenylalanine (2,6F 2 -Phe) and 2,6diMethyl-Phenylalanine (2,6diMeth-Phe; Figure 7A). In Cha, the benzene ring is replaced by the non-aromatic cyclohexane ring, rendering the side chain electroneutral, while minimally altering size and hydrophobicity (Mecozzi et al., 1996;Ahern et al., 2006; Figure 7A). In 2,6F 2 -Phe, the fluorine atoms at positions C2 and C6 withdraw p electrons from the face of the benzene ring which makes the edges close to the backbone electronegative, the distal edges electropositive and neutralizes the negative face of the aromatic ring (Mecozzi et al., 1996;Ahern et al., 2006; Figure 7A). Importantly, the hydrophobicity of the residue is not affected as benzene and hexa-fluoro-benzene have similar logP values (Leo et al., 1971). Substitutions with these two non-canonical amino acids specifically alter the p-electron distribution of Phe cen and Phe ex , thus testing the role of their electrostatic interactions with Glu ex . The methyl groups at positions C2 and C6 in 2,6diMeth-Phe ( Figure 7A Cha, Cyclohexylalanine; 2,6F 2 -Phe, 2,6Fluoro-Phenylalanine; 2,6diMeth-Phe, 2,6diMethyl-Phenylalanine. Right panel: stick representation of the amino acids, left panel: surface electrostatic potential of benzene and its derivatives with red and blue corresponding to À20 and +20 kCal mol À1 , respectively (Mecozzi et al., 1996). The surface electrostatic potential of 2,6diMeth-Phe is assumed similar to Phe because methyl group substitutions do not withdraw electrons from the benzene ring. (B, C) Normalized G-V relationships of the single-(B, C) and common-pore (D, E) gating processes for CLC-0 with the following replacements at Phe ex (B, D) and Phe cen (C, E): Phe (black circles), Cha (green squares), 2,6F 2 -Phe (pink triangles) and 2,6diMeth-Phe (cyan diamonds). WT G-V curves (from Figure 6) are shown as gray dashed lines for reference. Solid lines represent fits to a Boltzmann function with an offset (see Materials and methods, Equation 1). Note that the G-V data for F418X+2,6diMeth-Phe was obtained on the background of the C212S mutant (2,6diMeth-Phe*) to isolate its effects on the single-pore gating process. The effects of F418X+2,6diMeth-Phe substitution on the WT background are shown in  around the c1 angle (Harrison et al., 2003;Li et al., 2007), allowing us to explicitly test the role of the rotation of Phe cen in CLC gating. Neutralization of electrostatics at Phe cen with Cha right-shifts the single-pore G-V by~50 mV, while the redistribution of the electrons in 2,6F 2 -Phe left-shifts it by~50 mV ( Figure 7C, Supplementary file 1). These replacements have smaller effects at Phe ex as both the F214Cha and the F214-2,6F 2 -Phe substitutions cause an~20-25 mV right-shift in the G-V ( Figure 7B,  Supplementary file 1). Similarly, substitutions at Phe cen also affect common-pore gating ( Figure 7E, Supplementary file 1) while those at Phe ex cause only minor alterations ( Figure 7D,  Supplementary file 1). Notably, the effect of F418Cha is comparable to that of F418A, suggesting that the electrostatics dominate the interactions of Phe cen with Glu ex during common-pore gating ( Figures 6H and 7E, Supplementary file 1). These substitutions suggest that the p-electron distribution of Phe cen plays a major role in channel gating, while the steric and hydrophobic nature of the phenylalanine side chain appears to play a bigger role at Phe ex (Figure 7, Figure 7-figure supplement 1).
Finally, to test whether the interconversion between the 'up' and 'down' rotamers of Phe cen plays a role in channel gating we used the rotationally restricted isostere 2,6diMeth-Phe ( Figure 7A). The gating kinetics of the common-pore of F418-2,6diMeth-Phe are accelerated so that they become comparable to those of the single-pore gate (Figure 7-figure supplement 2A,B). To separate the effects of this non-canonical substitution on the single-and common-pore gating processes we used the C212S mutant which locks the common gate open (Lin et al., 1999;Figure 7-figure supplement 2C,D). The V 0.5 of the single-pore gate of the double mutant C212S F418-2,6diMeth-Phe is shifted by~+40 mV, while the V 0.5 of the common-pore gate of the single mutant is shifted by~À25 mV ( Figure 7C,E, Suppl. File 1). Thus, restricting the conformational rearrangement of Phe cen between the 'up' and 'down' conformers impairs gating of the CLC-0, consistent with the finding of our MD simulations on the CLC-ec1 transporter (Figure 3). Taken together, our results suggest that the aromatic slide formed by Phe cen and Phe ex forms an evolutionarily conserved structural motif that enables movement of Glu ex in and out of the Clpore during the exchange cycle and gating of CLC exchangers and opening of CLC channels.

Discussion
Despite extensive structural and functional investigations, the mechanisms underlying the exchange cycle of the CLC transporters and opening of the CLC channels remain poorly understood. These processes are evolutionarily and mechanistically related (Miller, 2006;Accardi, 2015;Jentsch and Pusch, 2018): in both channels and transporters Glu ex moves in and out of the Clpathway in a protonationdependent manner. However, the molecular steps that underlie these rearrangements are unclear. The CLCs exchangers do not follow the 'ping-pong' or sequential kinetics that characterize most conventional transporters. In contrast, CLCs simultaneously bind H + and Cl - (Picollo et al., 2012), and substrate movement occurs along two partially congruent translocation pathways (Accardi et al., 2005;Zdebik et al., 2008). Yet, how the two substrates bypass each other in the protein remains unknown. The recent structures of the CLC F exchanger and of the CLC-1 Clchannel suggested that Glu ex might interact with Phe cen (Last et al., 2018;Park and MacKinnon, 2018), but the functional implications of these interactions are not clear. Current transport mechanisms for CLC exchangers are not reversible, as they all postulate a step where a protonated and neutral Glu ex displaces Clions bound to S ext and S cen sites to transfer its proton to the internal solution (Accardi et al., 2005;Zdebik et al., 2008). This transition is energetically unfavorable (Kuang et al., 2007), and would result in highly asymmetric transport rates. However, the CLC exchangers function with comparable efficiency in both directions (Matulef and Maduke, 2005;Leisle et al., 2011;De Stefano et al., 2013), arguing against the existence of an asymmetric rate-limiting step. Here, a combination of MD simulations with biochemical and electrophysiological experiments suggest that Glu ex takes different routes depending on its protonation state, allowing for full reversibility of the transport mechanism.
Formation of an aromatic slide is essential to Glu ex movement Our data suggest that while the deprotonated and negatively charged Glu ex moves through the Clpathway, the protonated and neutral Glu 0 ex interacts with two highly conserved phenylalanines, Phecen and Phe ex . These residues can form an aromatic slide that enables movement of protons bound to Glu ex in and out of the pathway, connecting the central binding site and the extracellular solution. In the available structures of CLC channels and exchangers (Dutzler et al., 2002;Dutzler et al., 2003;Accardi et al., 2005;Accardi et al., 2006;Lobet and Dutzler, 2006;Feng et al., 2010;Jayaram et al., 2011;Basilio et al., 2014;Park et al., 2017;Last et al., 2018;Park and MacKinnon, 2018;Wang et al., 2019), Phe cen is in the 'up' rotamer suggesting this conformation is likely the most stable. However, our combined simulation and functional data suggest that Phe cen can adopt two distinct rotamers around its c1 angle (Figure 2), and that this rearrangement determines the energy barrier for Clmovement within the pore (Figure 2, Figure 2-figure supplement 2) and enables Glu 0 ex to reach or leave S cen via the aromatic slide pathway (Figure 3). Further, we show that the aromatic slide residues play an evolutionarily conserved key role in the CLC exchange cycle and in CLC channel gating. Our conventional mutagenesis underscores the conserved importance of Phe cen and Phe ex in transporter and channel gating: alanine substitutions at these positions impair the absolute transport rate, the voltage dependence and the exchange stoichiometry of the transporters (Figures 4 and 5, Figure 5-figure supplement 2), and nearly abolish the voltage dependence of channel gating ( Figure 6). We used atomic scale mutagenesis to test the prediction of our MD simulations that formation of the aromatic slide entails a rotation of the side chain of Phe cen around its C a -C b bond. When we replace Phe cen with 2,6diMet-Phe, a Phe derivative that specifically constrains this rotational rearrangement, we find that opening of the single-and common-pore gates in the CLC-0 channel is impaired (Figure 7, Figure 7-figure supplement 2). Thus, the aromatic slide forms an evolutionarily conserved structural motif that might enable movement of a protonated Glu 0 ex in and out of the pathway in both CLC channels and transporters, consistent with the finding that some CLC channels mediate some residual H + transport (Picollo and Pusch, 2005;Lísal and Maduke, 2008). The low functional expression of the CLC-5 and À7 transporters encoding the non-canonical amino acid prevented a similar test in these homologues. Moreover, for CLC-ec1, the technology for incorporation of these particular unnatural amino acids is currently not available for prokaryotic expression systems.
Our results also suggest that Phe ex and Phe cen interact differently with Glu ex . The aromatic ring of Phe ex appears to play a structural role in helping position Glu ex where it can interact with Phe cen in the 'up' conformation ( Figure 3). Indeed, removal of the aromatic ring of Phe ex severely affects voltage dependent gating of CLC channels and transporters ( Figures 5 and 6), while selective manipulations of its electrostatic properties have only minor effects on channel gating ( Figure 7B,D), possibly because this residue lines a water-accessible extracellular vestibule in the CLCs. In contrast, the electrostatic properties of the aromatic ring of Phe cen are essential determinants for Glu ex movement in the CLC-0 channel: elimination of the p-electrons of Phe cen favors the closed state of the single-pore gate ( Figure 7C), while their re-localization to the proximal edge promotes opening ( Figure 7C). These findings are consistent with the location of Phe cen within the core of the protein where changes in electrostatics are likely to have more impact, and with our MD simulations suggesting that Glu ex forms a p-dipole interaction with the aromatic ring of Phe cen (Figure 3). Indeed, the interaction between the buried aromatic Phe cen and Glu ex could account for the observed shifts in the pKa of Glu ex (Robinson et al., 2017) in CLC channels and transporters to keep Glu ex protonated during its transition through the protein (Hanke and Miller, 1983;Niemeyer et al., 2009;Picollo et al., 2010;Picollo et al., 2012). Our results are in harmony with the recent proposal that Glu ex can adopt a conformation where it directly interacts with Phe cen in a fluoride selective CLC antiporter (Last et al., 2018), in a mutant mimicking a fully-protonated CLC-ec1 (Chavan et al., 2019) and in the CLC-1 channel (Park and MacKinnon, 2018; Figure 4-figure supplement 2C).

A mechanism for Cl -/H + exchange
Our finding that two Phe residues form an evolutionarily conserved secondary pathway that enables movement of the protonated Glu ex in and out of the ion transport pathway allows us to propose a 7state mechanism for the CLC transporters that explains the stoichiometry of 2 Cl -:1 H + , is fully reversible, and accounts for previous results. Our findings provide structural and mechanistic grounds to explain how Glu ex and Clions can swap places in the CLC permeation pathway (Jentsch and Pusch, 2018). For simplicity of representation, we consider that an ion in the S ext * site is out of the pathway and as such, in our scheme we do not show ions bound to this site. As a starting configuration, we consider a state where Glu ex and Glu in are de-protonated, Glu ex occupies S cen , Phe cen is in the 'up' position and no Clions are bound to the pathway (Figure 8, I). After a Clion binds to S int (Figure 8, II), A B Figure 8. Proposed mechanism for the 2 Cl -: 1 H + CLC exchangers. (A) Schematic representation of the 2 Cl -: 1 H + CLC exchange cycle. The Clions are shown as green circles and H + as blue circles. The Cland H + pathways are shown in pale green and cyan respectively. For clarity, the residues that undergo conformational rearrangements in each step are highlighted in red. The H + pathway is shown as discontinuous when it is in a non-conductive Figure 8 continued on next page opening of the intracellular gate, formed by Ser cen and Tyr cen (Basilio et al., 2014), allows the ion to move into S cen , displacing Glu ex to S ext (Figure 8, III). Binding of a second ion to S ext favors protonation of Glu ex , and is accompanied by the displacement of Glu 0 ex out of the pathway and closure of the internal gate (Figure 8, IV). The protonated Glu 0 ex diffuses toward the aromatic slide and interacts with Phe ex (Figure 8, V). Movement of Glu 0 ex along the aromatic slide allows the interdependent rearrangement of Phe cen to the 'down' state and release of a Clion, i.e. release from S ext to the extracellular milieu and the transfer of the second ion from S cen to S ext (Figure 8, VI). This conformation, where Glu 0 ex interacts with Phe cen on the side of the Clpathway and S ext is occupied, favors the formation of a water wire (Figure 8, VI), which enables proton transfer from Glu 0 ex to Glu in (Figure 8, VII). Deprotonation of Glu ex allows it to move into S cen , favoring the release of the second Clfrom S ext to the outside, and of the proton from Glu 0 in to the intracellular solution, returning to the starting configuration (Figure 8, I). We note that the Clbinding affinity to S int is low as there is no free energy barrier preventing the release of the ion (Accardi et al., 2006;Lobet and Dutzler, 2006;Picollo et al., 2009). Thus, when the transporter is in states IV through VII it is likely that Clions will bind to and unbind from S int . However, as long as the intracellular gate remains closed these events would not participate in the permeation process. Thus, for simplicity we have omitted them from the gating scheme. In sum, we propose that the Cland H + ions bypass each other while moving in opposite directions through a CLC transporter by taking physically distinct routes: the Clions move through the anion selective pore while the H + moves along a pathway comprised of a water-wire and the aromatic slide ( Figure 8B). The distinct routes for Cland H + allow for a complete reversibility of the cycle.
Remarkably, our findings show that the role of the aromatic slide residues Phe cen and Phe ex is evolutionarily conserved between CLC exchangers and channels. We propose that the aromatic slide in CLC-0 could provide a conserved pathway that enables the residual H + transport associated with the common-pore gating process of the CLC-0 channel to occur (Lísal and Maduke, 2008). Indeed, the role of Phe cen in CLC function is also supported by the finding that mutations at this position severely affect CLC-1 channel gating and conductance (Estévez et al., 2003) and cause dominant myotonia (Imbrici et al., 2015). Thus, our proposed exchange mechanism for the CLC transporters also captures the key rearrangements that underlie gating of the CLC channels. In this framework, rapid ion conduction would be enabled by the disruption of the intracellular gate, a hypothesis justified by the widened intracellular vestibule seen in the recent structures of the bCLC-K and hCLC-1 channels (Park et al., 2017;Park and MacKinnon, 2018;Wang et al., 2019).

Molecular system
The molecular systems, based on the CLC-ec1 crystal structure (PDB: 1OTS) (Dutzler et al., 2003), were assembled using the CHARMM-GUI web service (Jo et al., 2008). As it was shown that the monomeric form of CLC-ec1 is functional (Robertson et al., 2010), a single subunit was used. The x-ray structure includes residues 18 to 458. Residues 18 to 31 are normally interacting with the opposite subunit and were removed here to prevent any unexpected interaction with the pore. All residues are in their natural protonation state at pH seven except Glu148, which is tested in both its unprotonated and protonated forms. The orthorhombic periodic simulation cell contains a ClC-ec1 monomer inserted in a bilayer composed of 262 dimystoylphosphatidylcholine (DMPC) lipids, about The inner gate opens, a second intracellular Clbinds so that both S int and S cen are occupied and Glu ex moves to S ext . (IV) The Clions move to S cen and S ext , the inner gate closes, Glu ex moves out of the pathway and becomes protonated. (V) Glu 0 ex interacts with Phe ex . (VI) Glu 0 ex interacts with Phe cen which rotates to the down conformation enabling the formation of a water wire connecting Glu 0 ex and Glu in ; one Clion is released from S ext and the other one moves from S cen to S ext . (VII) The H + is transferred to Glu in favoring movement of Glu ex into S cen and the Clion in S ext is released to the extracellular solution, returning to the initial state (I). (B) Physically distinct Cland H + pathways are indicated in CLC-ec1 WT structure (Glu ex and Glu in are shown in yellow, Phe ex in blue, Phe cen in red, Clions as green spheres). Both pathways converge at Glu ex .
13 600 TIP3P water molecules, and potassium/chloride ions to neutralize the systems at a concentration of about 150 mM KCl.

Potential function and simulations
All simulations were run using the CHARMM simulation software version c36 (Brooks et al., 2009) using the CHARMM36 force field (Best et al., 2012). Simulations were performed in an isother-malÀisobaric ensemble with a pressure of 1 atm and a temperature of 323 K. Particle-mesh Ewald method (Essmann et al., 1995) was applied for the calculation of electrostatic interactions with a grid spacing of 1 Å . The cutoff distance for van der Waals interactions was taken at 12 Å with a switching function starting at 10 Å . Time step for the integration of the motion was set to two fs. The membrane systems were equilibrated following a standard protocol offered by CHARMM-GUI (Wu et al., 2014). After equilibration, a simulation of 100 ns was run for each molecular system prior to the free energy calculations. All MD simulations were performed at 0 mV.
Using the CHARMM36 force field, we compared the strength of a p-dipole interaction and that of a typical H-bond. We calculated that the interaction between a protonated glutamate side chain and the aromatic ring of a phenylalanine is À5.0 kCal mol À1 (for an optimized conformation in vacuum). For comparison, the interaction between a protonated glutamate and a serine side chain is À5.8 and À7.2 kCal mol À1 , depending if the unprotonated or the protonated oxygen of the carboxylate group is involved. These number are in line with ab initio calculations that reported interactions on the order of À2.4 to À7.1 kCal mol À1 (Du et al., 2013).

Potential of mean force calculations
The potential of mean forces (PMF) were calculated using the self-learning adaptive umbrella sampling method (Wojtas-Niziurski et al., 2013). In the case of PMFs describing the displacement of one or two ions, the reaction coordinate is the position of a given ion in the pore, as projected on the normal to the membrane (Z axis). The reference point is the center of mass of the backbone of residues 107, 356, 357, and 445. Independent simulations of 500 ps were performed every 0.5 Å along the reaction coordinate using a biasing harmonic potential of 20 kCal/mol . Å 2 . For the 2D PMFs of Figure 2, the ions were initially placed in S cen and S ext . For the 1D PMF of Figure 3-figure supplement 1, the ion was placed in S ext . In the case of PMFs describing the rotation of the F357 side chain, the reaction coordinate is the c1 angle. Simulations were performed every 10 degrees using a biasing harmonic potential of 20 kCal/mol . rad 2 . In all cases the first 50 ps of every independent window simulation was considered as equilibration and thus excluded from the analysis. The independent simulations were combined and unbiased using the weighted histogram analysis method (WHAM) (Kumar et al., 1992). The conformations shown in Figure 3 are observed in at least three sampling windows and remain stable for most of the sampling time of a given window.

Cland H + flux recordings for CLC-ec1 and variants
Cland H + fluxes of CLC-ec1 wild-type and mutant proteins reconstituted into proteoliposomes were recorded simultaneously and the coupling stoichiometry was determined as described (Basilio et al., 2014).

Isothermal titration calorimetry (ITC)
Clbinding affinity was determined for purified CLC-ec1 F357A as described (Picollo et al., 2009;Basilio et al., 2014) using a nanoITC instrument (TA Instruments). For these experiments, the final purification step of the protein was purified over a gel filtration column pre-equilibrated in 100 mM Na-K-Tartrate, 20 mM Hepes, 50 mM DMNG, pH 7.5 (Buffer B0) and concentrated to 50-195 mM. The injection syringe was filled with buffer B0 with 50 mM KCl added, to achieve final Molar Ratios of 70-100. Each experiment consisted of 30-48 injections of 1 ml of the ligand solution at 3-4 min intervals into the experimental chamber kept under constant stirring at 350 rpm and at 25.0 ± 0.1˚C. All solutions were filtered and degassed prior to use. The ITC data was fit to a single site Wiseman isotherm using the NanoAnalyze program from TA instruments.

In vitro cRNA transcription
RNAs for all CLC-0, hCLC-7, hCLC-5 and mOstm1 wild-type and mutant constructs were transcribed from a pTLN vector using the mMessage mMachine SP6 Kit (Thermo Fisher Scientific, Grand Island, NY) (Pusch et al., 1995;Steinmeyer et al., 1995;Leisle et al., 2011). For all experiments in this paper a plasma-membrane localized version of hCLC-7 has been used as 'wild-type' which has been published earlier termed as CLC-7 PM (Leisle et al., 2011). For final purification of cRNA the RNeasy Mini Kit (Quiagen, Hilden, Germany) was employed. RNA concentrations were determined by absorbance measurements at 260 nm and quality was confirmed on a 1% agarose gel.

Protein expression in Xenopus laevis oocytes and two electrode voltage clamp (TEVC) recordings
Xenopus laevis oocytes were purchased from Ecocyte Bio Science (Austin, TX, USA) and Xenoocyte (Dexter, Michigan, USA) or kindly provided by Dr. Pablo Artigas (Texas Tech University, USA, protocol # 11024). For conventional CLC expression, following injection and expression conditions have been used: for CLC-7 (WT, F301A or F514A) and Ostm1, 25-75 ng of each cRNA were injected per oocytes and currents were recorded after~60-80 hr; for CLC-5 (WT, F255A or F445A), 50 ng cRNA were injected per oocyte and currents were recorded~48-72 hr; for CLC-0 (WT, F214A or F418A), 0.1-5 ng cRNA were injected and currents were measured~6-24 hr after injection. For nonsense suppression of CLC-0 constructs (F214X, F418X, C212S F418X), cRNA and misacylated tRNA were coinjected (up to 25 ng of cRNA and up to 250 ng of tRNA per oocyte) and currents were recorded 6-24 hr after injection. TEVC was performed as described (Picollo et al., 2009). In brief, voltage-clamped chloride currents were recorded in ND96 solution (in mM: 96 NaCl, 2 KCl, 1.8 CaCl2, 1 MgCl2, 5 HEPES, pH 7.5) using an OC-725C voltage clamp amplifier (Warner Instruments, Hamden, CT). The data was acquired with Patchmaster (HEKA Elektronik, Lambrecht, Germany) at 5 kHz (CLC-0) and 20 kHz (CLC-5, CLC-7) and filtered with Frequency Devices 8-pole Bessel filter at a corner frequency of 2 kHz. Analysis was performed using Ana (M. Pusch, Istituto di Biofisica, Genova) and Prism (Graph-Pad, San Diego, CA, USA). Standard voltage-clamp protocols have been applied for the three CLC proteins, the holding potential was constant at À30 mV. For CLC-0 two different recording protocols have been used to distinguish single-pore from common-pore gating. During the single-pore gating protocol the voltage was stepped to +80 mV for 50 ms and then a variable voltage from À160 mV to +80 mV increasing in 20 mV steps was applied for 200 ms, followed by a 50 ms pulse at À120 mV for tail current analysis. For CLC-0 common-pore gating, 7 s voltage steps from +20 mV to À140 mV have been applied in À20 mV increments followed by a 2.5 s +60 mV post pulse for tail current analysis. For CLC-5, a simple voltage step protocol was applied: 400 ms steps from À80 to + 80 mV in 20 mV increments. For CLC-7/Ostm1, voltage was clamped at variable values from À80 to +90 mV in 10 mV steps for 2 s, followed by a 0.5 s post pulse at À80 mV for tail current analysis.
To estimate the voltage dependence of CLC-0 and CLC-7/Ostm1 mutants gating, tail current analysis was performed and data was fit to a Boltzmann function of the form: P o ¼ P min þ ð1 À P min Þ 1 þ exp ½ðV 0:5 À VÞ=k (1) where P o/italic is the open probability as a function of voltage and is assumed to reach a value of unity at full activation. P min/italic is the residual open probability independent of voltage. V 0.5 is the voltage at which 50% activation occurs, and k=RT/zF is the slope factor, R is the universal gas constant, T is temperature in K, F is the Faraday constant, and z is the gating charge. For analysis of the activation kinetics of CLC-7/Ostm1 and its variants, activating voltage pulses (from +20 to +90 mV) were fit to a bi-exponential function of the following form: where I is the current as a function of time; A 1 , A 2 and A 0 are fractional amplitudes obtained by normalizing to the total current. While A 1 and A 2 are time-dependent components, A 0 is time-independent. t 1 and t 2 are the corresponding time constants.

Statistical analysis
All values are presented as mean ± s.e.m. To determine statistical significance Student's t-test (twotailed distribution; two-sample equal variance) was performed. The threshold for significance was set to p=0.05.