Structural characterization of an intermediate reveals a unified mechanism for the CLC Cl−/H+ transport cycle

Among coupled exchangers, CLCs uniquely catalyze the exchange of oppositely charged ions (Cl− for H+). Transport-cycle models to describe and explain this unusual mechanism have been proposed based on known CLC structures. While the proposed models harmonize many experimental findings, there have remained gaps and inconsistencies in our understanding. One limitation has been that global conformational change – which occurs in all conventional transporter mechanisms – has not been observed in any high-resolution structure. Here, we describe the 2.6 Å structure of a CLC mutant designed to mimic the fully H+-loaded transporter. This structure reveals a global conformational change to a state that has improved accessibility for the Cl− substrate from the extracellular side and new conformations for two key glutamate residues. Based on this new structure, together with DEER measurements, MD simulations, and functional studies, we propose a unified model of the CLC transport mechanism that reconciles existing data on all CLC-type proteins.


INTRODUCTION 1
CLC transporter proteins are present in intracellular compartments throughout our bodies -in 2 our hearts, brains, kidneys, liver, muscles, and guts -where they catalyze coupled exchange of 3 chloride (Cl -) for protons (H + ) (Jentsch and Pusch, 2018). Their physiological importance is 4 underscored by phenotypes observed in knockout animals, including severe neurodegeneration and 5 osteopetrosis (Sobacchi et al., 1993;Stobrawa et al., 2001;Hoopes et al., 2005;Kasper et al., 2005), 6 and by their links to human disease including X-linked mental retardation, epileptic seizures, Dent's 7 disease, and osteopetrosis (Lloyd et al., 1996;Hoopes et al., 2005;Veeramah et al., 2013;Hu et al., 8 2016). 9 CLC-ec1 is a prokaryotic homolog that has served as a paradigm for the family (Estevez et 10 al., 2003;Lin and Chen, 2003;Engh and Maduke, 2005;Miller, 2006;Matulef and Maduke, 2007). Its 11 physiological function enables resistance to acidic conditions, such as those found in host stomachs 12 (Iyer et al., 2003). Like all CLC proteins, CLC-ec1 is a homodimer in which each subunit contains an 13 independent anion-permeation pathway (Miller and White, 1984;Ludewig et al., 1996;Middleton et 14 al., 1996;Dutzler et al., 2002). Studies of CLC-ec1 revealed the importance of two key glutamate 15 residues -"Glu ex " and "Glu in " (Figure 1A) in the transport mechanism. Glu ex is positioned at the 16 extracellular entryway to the Cl --permeation pathway, where it acts both as a "gate" for the transport 17 of Cland as a participant in the transport of H + (Dutzler et al., 2003;Accardi and Miller, 2004). Glu in 18 is located towards the intracellular side of the protein and away from the Cl --permeation pathway, 19 where it appears to act as a proton transfer site (Accardi et al., 2005;Lim and Miller, 2009). 20 Glu ex has been observed in three different positions relative to the Cl --permeation pathway in 21 CLC transporter crystal structures: "middle", "up", and "down". The "middle" conformation is observed 22 in the WT CLC-ec1 structure, where Glu ex occupies the extracellular anion-binding site, "S ext " (Dutzler 23 et al., 2002) (Figure 1B). The "up" conformation is seen when Glu ex is mutated to Gln, mimicking 24 protonation of Glu ex ; here, the side chain moves upward and away from the permeation pathway, 25 allowing a Clion to bind at S ext (Dutzler et al., 2003) (Figure 1B). The "down" conformation is seen 26 in the eukaryotic cmCLC structure, where Glu ex plunges downwards into the central anion-binding 27 site, "S cen " (Feng et al., 2010) (Figure 1B). The intracellular anion-binding site, "S int ", is a low-affinity 28 site (Picollo et al., 2009) and is not depicted. Of note, cmCLC differs from CLC-ec1 (and from the 29 human CLC transporters) in that it lacks the Glu in residue; a threonine is observed at the 30 corresponding position. In CLC-ec1, a titratable side chain at the Glu in position is required for coupled 31 transport (Lim and Miller, 2009). 32 The rotation of the Glu ex side chain is the only conformational change that has been detected 1 crystallographically in the CLC transporters. A central question therefore is whether and how other 2 protein conformational changes contribute to the CLC transport mechanism. In previous work, we 3 used a spectroscopic approach to evaluate conformational changes in CLC-ec1, and we found that 4 raising [H + ] (to protonate Glu ex ) caused conformational change in regions of the protein outside of the 5 permeation pathway, up to ~20 Å away from Glu ex (Elvington et al., 2009;Abraham et al., 2015). 6 Using a combination of biochemical crosslinking, double electron-electron resonance (DEER) 7 spectroscopy, functional assays, and molecular dynamics (MD) simulations, we concluded that this 8 H + -induced conformational state represents an "outward-facing open" state, an intermediate in the 9 transport cycle that facilitates anion transport to and from the extracellular side (Khantwal et al., 2016). 10 Here, to obtain a high-resolution structure of the H + -induced conformational state, we 11 crystallized a triple mutant, "QQQ", in which glutamines replace three glutamates: Glu ex , Glu in , and 12 E113; the latter is a residue within hydrogen bonding distance to Glu in and computationally predicted 13 to be protonated at neutral pH (Faraldo-Gomez and Roux, 2004). In contrast to the single-point 14 mutants of Glu in and Glu ex , which reveal either no conformational change (Gln in ) (Accardi et al., 2005) 15 or only a simple side chain rotation (Gln ex ) (Dutzler et al., 2003), the QQQ mutant structure reveals 16 global conformational change which generates the expected opening of the extracellular permeation 17 pathway. Unexpectedly, this structure additionally reveals new side chain conformations for both 18 Gln ex and Gln in , which bring the two residues to within kissing distance, ~5 Å. Based on this new 19 structure, together with MD simulations, DEER spectroscopy, and functional studies, we propose an 20 updated model of the Cltransport cycle. 21

Unique conformations of the H + -transfer glutamates: 24
The QQQ mutant (E148Q/E203Q/E113Q) was crystallized in the lipidic cubic phase, without 25 any antibody Fab fragment. The structure, determined at 2.6 Å resolution (Table 1), reveals entirely 26 unanticipated changes in the conformations of the Glu ex and Glu in residues, Q148 and Q203. Gln ex 27 residue is in a new position, not previously observed in the CLC transporters, away from the 28 permeation pathway and into the hydrophobic core of the protein, a conformation we designate as 29 "out" (Figure 1C,D). Gln in , which is positioned identically in all structures to date, is also in a new 30 position, towards the core of the protein. Together, these movements bring Gln ex and Gln in close 31 together, separated by only 5.6 Å ( Figure 1E). In comparison, this distance is 12.8 Å in the WT CLC-1 ec1 (1ots) (Dutzler et al., 2003). 2

Opening of the extracellular vestibule and correlation with increased Cltransport rates: 3
Analysis of the QQQ structure using HOLE, a program for analyzing the dimensions of 4 pathways through molecular structures (Smart et al., 1996), reveals an opening of the extracellular 5 vestibule, increasing accessibility from the extracellular solution to the anion-permeation pathway, in 6 contrast to previously described structures. In the WT protein, two sub-Angstrom bottlenecks occur 7 between S cen and the extracellular side of the protein (Figure 2A). In the QQQ protein, these 8 bottlenecks are relieved, widening the pathway to roughly the size of a Clion. In contrast, a single 9 point mutation at the Glu ex position (E148Q) relieves only one of the two bottlenecks (Figure 2A). 10 This observation is consistent with the QQQ structure representing the CLC-ec1 outward-facing open 11

state. 12
Mutating Glu ex to Gln renders CLC proteins unable to transport H + but still capable of 13 transporting Cl - (Accardi and Miller, 2004). The difference in the HOLE pore-radius profiles for the 14 QQQ versus E148Q (Glu ex to Gln ex ) mutants predicts that Cltransport through QQQ will be faster 15 than through E148Q, assuming no rate-limiting conformational changes intracellular to S cen and S ext . 16 Consistent with this prediction, we found Cltransport through the QQQ mutant to be ~2-fold faster 17 than through E148Q (Figure 2B,C). Since QQQ and E148Q differ at only two residues (glutamine 18 versus glutamate), we hypothesized that lowering the pH to protonate the glutamate residues would 19 allow the E148Q mutant to adopt the QQQ-like (outward-facing open) conformation. In support of this 20 hypothesis, at pH 4.5 there is no difference in transport rates between E148Q and QQQ ( Figure 2D). 21

Structural changes underlying the opening of the extracellular vestibule 22
The extracellular bottleneck to anion permeation is formed in part by Helix N, which together 23 with Helix F forms the anion-selectivity filter (Dutzler et al., 2002). Previously, we proposed that 24 generation of the outward-facing open state involves movement of Helix N in conjunction with its 25 neighbor Helix P (at the dimer interface) to widen this bottleneck (Khantwal et al., 2016). Structural 26 alignment of the QQQ mutant with either E148Q (Figure 3A-C) or WT ( Figure 3C) confirms the 27 movement of these helices. These structural changes involve shifts in highly conserved residues near 28 the anion-permeation pathway, including F190 (Helix G), F199 (Helix H), and F357 (Helix N). The 29 side chains of all repositioned residues show good electron density ( Figure 3D, Figure 3 -figure  30 supplement 1). Together, these motions widen the extracellular bottleneck ( Figure 3E,F). This 31 widening is accompanied by subtle changes in the S ext Cl --binding site (Figure 3 -figure  1   supplement 2A,B). However, we did not find any significant difference in the affinity of Clmeasured 2 by isothermal titration calorimetry (ITC) (Figure 3 -figure supplement 2C), To evaluate changes 3 independent of the structural alignment method, we generated difference distance matrices 4 (Nishikawa, 1972). Comparison of QQQ to WT CLC-ec1 confirms a hot spot of conformational change 5 at Helices M-N-O, as well as changes in Helices C, G, H, and Q (Figure 3 -figure supplement 3A). 6 In contrast, comparison of single-Glu mutant structures to WT reveals only minor (≤ 0.8 Å) changes 7 (Figure 3 -figure supplement 3B). 8 In addition to the conformational changes of the transmembrane helices, there are significant 9 rearrangements of two of the interhelical linkers (Figure 3 -figure supplement 4A). At the 10 extracellular side, there are small changes at the I-J linker, which interacts with the Fab fragment in 11 the other structures. At the intracellular side, the Helix H-I linker (residues 205-213) undergoes a 12 several-Angstrom displacement compared to WT. This movement, though relatively large, cannot be 13 of functional significance: previously it was shown that movement of the linker can be restricted via 14 an inter-subunit disulfide cross-link at residue 207, with no significant effect on function (Nguitragool 15 and Miller, 2007). The cross-link at residue 207 does not affect the positioning of Helices H and I 16 (Figure 3 -figure supplement 4B). 17

Validation of QQQ conformational change in WT CLC-ec1 using DEER spectroscopy 18
Our working hypothesis is that the QQQ mutant structure mimics the outward-facing open 19 intermediate in the WT CLC transport cycle. When working with a mutant, however, one always 20 wonders whether any conformational change observed is relevant to the WT protein. We therefore 21 used double electron-electron resonance (DEER) spectroscopy to evaluate conformational change 22 in WT CLC-ec1. DEER spectroscopy is advantageous because it can evaluate conformational 23 change by site-directed spin labeling, without the constraints of crystallization. Accurate distance 24 distributions can be obtained for spin labels separated by ~20-70 Å (Jeschke, 2012;Mishra et al., 25 2014;Stein et al., 2015). Since CLC-ec1 is a homodimer ~100 Å in diameter, a simple labeling 26 strategy with one spin label per subunit can provide a sample with optimally spaced probes for 27 distance-change measurements. For example, the extracellular side of Helix N ( Figure 3A) is 28 separated by ~50 Å from its correlate in the other subunit. To test the hypothesis that this helix moves, 29 we generated WT CLC-ec1 with spin labels at positions 373 and 374 on Helix N and performed DEER 30 measurements under two conditions, pH 7.5 and pH 4.5. The rationale for this experimental strategy 31 is that pH 4.5 will promote protonation of Glu ex and Glu in , thus favoring a global conformation 1 comparable to that observed in the QQQ structure ( Figure 4A). 2 Consistent with our hypothesis, spin labels on Helix N exhibited pH-dependent changes in 3 distance distributions, in the direction predicted by the QQQ structure ( Figure 4B,C). Similar results 4 were obtained for a spin label on Helix O ( Figure 4D). In all cases, spin-labeling did not have any 5 effect on CLC-ec1 activity (Figure 4 -figure supplement 1). At Helix P, our previous data provide 6 further support for the relevance of the conformational changes observed in QQQ. First, we showed 7 that CLC-ec1 activity is inhibited by cross-linking of D417C, on Helix P, across the dimer interface 8 (Khantwal et al., 2016). Thus, residues D417 must move apart from one another during the transport 9 cycle. We additionally used DEER to demonstrate that this outward motion of D417C occurs in 10 response to lowering the pH (Khantwal et al., 2016). Correspondingly, the intersubunit Cα distance 11 for D417 is 11.6 Å in the QQQ structure, compared to 8.8 Å in WT ( Figure 4E), thus supporting the 12 conclusion that the QQQ structure corresponds to a low-pH structure of WT CLC-ec1. 13

Analysis of water connections to Gln ex 14
Previous computational studies indicated that water wires can transiently bridge the Glu ex and 15 Glu in residues separated by 12.8 Å in the CLC-ec1 WT structure, which may serve as the pathway 16 for H + transfer (Wang and Voth, 2009;Han et al., 2014). The proximity of these residues in the QQQ 17 structure motivated us to re-evaluate this phenomenon. In our previous studies on WT 18 extended MD simulations revealed that water spontaneously enters the hydrophobic core of the 19 protein and transiently and repeatedly forms water wires connecting Glu ex and Glu in (Han et al., 2014). 20 Analogous simulation of the QQQ mutant revealed a dramatic and unanticipated result: water 21 penetration into the hydrophobic core of the protein is greatly increased, and water wires directly 22 connect bulk water in the intracellular solution to Gln ex , without requiring intermediate connection to 23 Gln in (Figure 5A-C). These water pathways were observed frequently during our 600-ns simulation 24 ( Figure 5D); in contrast, such water pathways were not observed in our previous 400-ns WT 25 simulation (Han et al., 2014). The number of water molecules needed to reach bulk water follows a 26 normal distribution, with chains of 5 or 6 water molecules predominating ( Figure 5D). In contrast, the 27 majority of water wires connecting Glu ex to Glu in in the WT simulation involved 7 or more water 28 molecules (Han et al., 2014). Moreover, the occurrence of water pathways in the QQQ simulation 29 (36.5%) is over an order of magnitude greater than the occurrence of water wires between Glu in and 30 Glu ex in the WT simulation (1.3%). 31 The absence of water pathways in the WT simulation is likely due to steric hindrance by Glu in , 1 E113, and bulky side chains in the vicinity, which together block direct access of intracellular bulk 2 water toward the protein interior (despite the conformational flexibility of Glu in (Wang and Voth, 2009)). 3 In the QQQ simulation, Gln in can equilibrate among five side chain conformations (Clusters 1-5), all 4 of which can support water pathways (Figure 5 -figure supplement 1A,B). Most of the water 5 pathways (96% of pathways observed) occur when Gln in is rotated away from its starting conformation 6 (Clusters 1-3), allowing water to flow along a pathway near Q113 (Figure 5 -figure supplement  7 1C,D). In these conformations, the Gln in side chain bends away from Q113 and from the bulky 8 residues F199 and I109, thus allowing intracellular bulk water to enter the protein interior without 9 encountering steric occlusion ( Figure 5 -figure supplement 2A). 10 The predominant water pathway observed in our simulations is roughly parallel to the Cl -11 permeation pathway ( Figure 5A). This pathway for water (and hence H + ) entry into the protein is 12 different from that previously suggested by us and others. Previously, it was proposed that H + access 13 to the interior of the protein occurs via an entry portal located near the interfacial side of the 14 homodimer Han et al., 2014) rather than on the "inner" pathway observed here. 15 While we do see some water pathways occurring along the interfacial route, on a pathway that is 16 lined by Gln in , these occur only rarely ( Figure 5 -figures supplement 1D). Importantly, the previous 17 mutagenesis studies supporting the interfacial route are also concordant with the inner water pathway 18 observed here. In the previous studies, mutations that add steric bulk at either E202 (Lim et al., 2012) 19 or the adjacent A404 (Han et al., 2014) were found to inhibit the H + branch of the CLC-ec1 transport 20 cycle. The observation that all water pathways involve rotation of E202 away from its starting position 21

Proton pumping without a titratable residue at the Glu in position 24
Glu in has long been modeled as a H + -transfer site in the CLC Cl -/H + mechanism (Accardi et 25 al., 2005;Miller, 2006;Lim and Miller, 2009;Basilio et al., 2014;Accardi, 2015;Khantwal et al., 2016). 26 However, this modeling is contradicted by the observation that several CLC transporters can pump 27 H + in the absence of a titratable residue at the Glu in position (Feng et al., 2010;Phillips et al., 2012;28 Stockbridge et al., 2012). In our analysis of water pathways in the MD simulations, we observed that 29 these pathways are not always lined by the Gln in side chain ( Figure 5 -figure supplement 1). This 30 finding strongly suggests that while Glu in facilitates water pathways, it is not required as a direct H + -31 transfer site. To test this hypothesis experimentally, we evaluated for Cl --coupled H + pumping in two 32 mutants lacking a titratable residue at Glu in . In the first experiment, we replaced Glu in (E203) and its 1 H-bonding partner, E113, with the residues found in cmCLC, a eukaryotic transporter that catalyzes 2 2:1 stoichiometric exchange of Clfor H + (Feng et al., 2010). In support of our hypothesis, we found 3 that the E203T/E113K mutant catalyzes Cl --driven H+ pumping ( Figure 6A-C). While the coupling 4 stoichiometry is substantially degraded compared to the WT protein ( Figure 6D), the thermodynamic 5 fact arising from this experiment is that H + pumping occurs with a non-titratable (Thr) residue at Glu in . 6 By comparison, in line with conventional modeling, the Glu ex mutant E148Q wholly fails to catalyze 7 proton pumping ( Figure 6A-C). To test whether H + pumping requires polar residues at Glu in and/or 8 E113 positions, we examined the double alanine mutant, E203A/E113A. Even with two alanines at 9 these positions, H + pumping is retained (Figure 6A-C). Thus, conventional thinking of Glu in as a H + -10 transport site must be re-evaluated. 11

A unifying model for the CLC transport mechanism 12
Based on the information gleaned from our study of the QQQ structural intermediate, we 13 propose an updated transport model for 2:1 Cl -/H + exchange by CLC transporters. This updated 14 model is inspired by four key findings. First, the outward-facing state has improved accessibility for 15 Clto exchange to the extracellular side (Figure 2). This state had been previously predicted 16 (Khantwal et al., 2016) but is now seen in molecular detail. Second, the protonated Glu ex can adopt 17 an "out" conformation, within the hydrophobic core of the protein ( Figure 1D). This novel conformation 18 allows us to eliminate a disconcerting step that was part of all previous models: movement of a 19 protonated (neutral) Glu ex -in competition with Cl --into the S cen anion-binding site (Miller and 20 Nguitragool, 2009;Feng et al., 2012;Basilio et al., 2014;Khantwal et al., 2016). Third, water pathways 21 can connect Glu ex directly to the intracellular solution ( Figure 5). Finally, H + pumping does not require 22 a titratable residue at Glu in (Figure 6). Together, these findings allow us to propose a revised Cl -/H + 23 exchange model that maintains consistency with previous studies and resolves lingering problems. 24 In our revised model (Figure 7A), the first three states are similar to those proposed previously 25 (Miller and Nguitragool, 2009;Feng et al., 2012;Basilio et al., 2014;Khantwal et al., 2016). State A 26 reflects the structure seen in WT CLC-ec1, with Glu ex in the "middle" conformation, occupying S ext , 27 and a Cloccupying S cen . Moving clockwise in the transport cycle, binding of Clfrom the intracellular 28 side displaces Glu ex by a "knock-on" mechanism (Miller and Nguitragool, 2009), pushing it to the "up" 29 position and making it available for protonation from the extracellular side (State B). Protonation 30 generates state C, which reflects the structure seen in E148Q CLC-ec1 where Gln ex mimics the 31 protonated Glu ex . This sequence of Clbinding and protonation is consistent with the experimental 32 finding that Cland H + can bind simultaneously to the protein (Picollo et al., 2012). Subsequently, a 1 protein conformational change generates an "outward-facing open" state (D). While this state had 2 previously been postulated (Khantwal et al., 2016), the QQQ structure presented here provides critical 3 molecular details. 4 State D involves a widening of the extracellular vestibule, which will facilitate Clbinding from 5 and release to the extracellular side. In the QQQ structure (our approximation of State D), the 6 reorientation of Helix N results in subtle changes in Cl --coordination at the S ext site (Figure 3 -figure  7 supplement 2), which suggests that binding at this site may be weakened, though we currently lack 8 direct evidence for this conjecture. Regardless of the affinity at S ext , the opening of the extracellular 9 permeation pathway in State D will promote Clexchange in both directions, which is essential to 10 achieving reversible transport. 11 In addition to involving a widening of the extracellular vestibule, state D has the protonated 12 Glu ex in an "out" conformation and within ~5 Å of Glu in (Figure 1C,D). At first glance, this positioning 13 suggested to us that Glu in might be participating in an almost direct hand-off of H + to and from Glu ex . 14 However, MD simulations revealed that Glu in is highly dynamic and most often is rotated away from 15 its starting position, allowing the robust formation of water pathways from the intracellular bulk water 16 directly to Glu ex (Figure 5) (State E). Once such transfer occurs, the deprotonated Glu ex will be 17 disfavored in the hydrophobic core, and it will compete with Clfor the S cen anion-binding site, 18 generating State F. Although this conformational state has not been observed crystallographically for 19 CLC-ec1, the computational studies of Piccolo et al. (Picollo et al., 2012) found that Glu ex favors the 20 S cen position when there are no Clions bound in the pathway (as in State F), and those of Mayes et 21 al. found that the "down" position is in general the preferred orientation for Glu ex (Mayes et al., 2018). 22 In addition, a recent structure of an Asp ex CLC-ec1 mutant supports that the carboxylate likes to reach 23 down towards S cen , in a "midlow" position, excluding the presence of Clat both S cen and S ext (Park et 24 al., 2019), as depicted in State F. From this state, binding of Clfrom the intracellular side (coordinated 25 with inner-gate opening (Basilio et al., 2014)) knocks Glu ex back up to S ext , generating the original 26 state A. This transport cycle is fully reversible, allowing efficient transport in both directions, as is 27 observed experimentally (Matulef and Maduke, 2005). 28

DISCUSSION 29
In this study, we aspired to determine the high-resolution structure of the CLC "outward-facing 30 open" conformational state. Our conclusion that the QQQ mutant structure represents such an 31 intermediate in the CLC transport mechanism is supported by several pieces of evidence. First, the 1 WT protein under low-pH conditions (glutamate residues protonated), adopts a conformation different 2 from the high-pH condition, as detected by DEER spectroscopy, with the conformational change in 3 the direction predicted by the QQQ structure ( Figure 4A-D). Second, the movement of helix P in the 4 QQQ structure is as predicted by the fact that cross-linking helix P inhibits transport (Khantwal et al.,5 2016) ( Figure 4E). Third, and perhaps most compellingly, the details observed in this conformational 6 state reconcile a multitude of findings in the literature. 7 Our proposed updated transport model (Figure 7A), in addition to retaining key features 8 based on previous models (Miller and Nguitragool, 2009;Feng et al., 2012;Basilio et al., 2014;9 Khantwal et al., 2016), unifies our picture of both CLC transporter and channel mechanisms. First, it 10 is compatible with transporters that have non-titratable residues at Glu in and E113. Our simulations 11 and experiments (Figures 5 and 6) lead to the conclusion that these residues play a key role in 12 regulating water pathways rather than in direct hand-off of H + . From this perspective, the evolution of 13 non-titratable residues in either (Feng et al., 2010) or both (Phillips et al., 2012;Stockbridge et al., 14 2012) of these positions is perfectly sensible. In addition, previous mutagenesis experiments on  ec1, which demonstrated a surprising tolerance for mutations at Glu in (Lim and Miller, 2009), now 16 make more sense. Strikingly, the structural positioning of T269 in cmCLC, located at the Glu in 17 sequence position, matches the structural positioning of Gln in in the QQQ mutant, such that side chain 18 dynamics could facilitate comparable water pathways ( Figure 7B). 19 The second unifying feature of our model is that it attests to Glu ex movements being conserved 20 amongst every known type of CLC: 2:1 Cl -/H + exchangers, 1:1 F -/H + exchangers, and uncoupled Cl -21 channels. Previously, an "out" position for Glu ex had been proposed to be essential to the mechanism 22 of F -/H + exchangers (Last et al., 2018), which allow bacteria to resist fluoride toxicity (Stockbridge et 23 al., 2012). However, such a conformation had not been directly observed, and it was postulated that 24 it may be only relevant to the F -/H + branch of the CLC family. Structurally, Glu ex in the "out" position 25 has previously only been observed in a CLC channel. This structure appears to represent an open-26 channel conformation, and thus was deduced that the "out" position is relevant only to CLC channels, 27 particularly because of residue clashes that were predicted based on transporter structures known at 28 the time (Park and MacKinnon, 2018). An overlay of the CLC-ec1 QQQ structure with the hCLC-1 29 channel structure reveals that the Glu ex "out" conformations are similar ( Figure 7C). Thus, this 30 conformation is a unifying feature of CLC channels and transporters. Moreover, this conclusion 31 connotes that all CLC proteins act via a "windmill" mechanism (Last et al., 2018), in which the 32 protonated Glu ex favors the core of the protein while the deprotonated Glu ex favors the anion-1 permeation pathway. Such a mechanism is preferable to previous "piston"-type mechanisms, with 2 Glu ex moving up and down within the anion-permeation pathway, which required a protonated 3 (neutral) Glu ex to compete with negatively charged Clions. 4 Elements of the transport cycle require future experiments to elucidate details. Prominently, 5 the nature of the inward-facing conformational state remains uncertain. In our model, we indicated 6 inward-opening with dotted lines (Figure 7, States F, A, B) to reflect this uncertainty. One proposal is 7 that the inner-gate area remains static and transport works via a kinetic barrier to Clmovement to 8 and from the intracellular side (Feng et al., 2010). Consistent with this proposal, multiscale kinetic 9 modeling revealed that 2:1 Cl -/H + exchange can arise from kinetic coupling alone, without the need 10 for large protein conformational change (Mayes et al., 2018). An alternative proposal is that CLCs 11 visit a conformationally distinct inward-open state, based on the finding that transport activity is 12 inhibited by cross-links that restrict motion of Helix O, located adjacent to the inner gate (Basilio et 13 al., 2014;Accardi, 2015). This putative inward-open state appears distinct from the conformational 14 change observed in the QQQ mutant, as the inter-residue distances for the cross-link pairs (399/432 15 and 399/259) are unchanged in QQQ relative to WT. The details of the kinetic-barrier and 16 conformational-change models, and the need for additional experiments on this aspect of transport, 17 have been clearly and comprehensively discussed (Accardi, 2015;Jentsch and Pusch, 2018). 18 Despite having an open extracellular vestibule, the QQQ mutant is a slow transporter -almost 19 an order of magnitude slower than the wild-type protein (at pH 4.5). This relative slowness may be 20 due to the QQQ's increased Clbinding affinity (140 µM, Figure 3 -Figure supplement 2, vs 600 21 µM for WT (Picollo et al., 2009;Khantwal et al., 2016)) and hence slower dissociation rate (Picollo et 22 al., 2009). Increased Cl --binding affinity (relative to WT) is observed in all Glu ex mutants (Picollo et 23 al., 2009;Picollo et al., 2012;Park et al., 2019), likely because no carboxylate side chain is available 24 to compete Clout of the binding sites. 25 Simulations of the QQQ conformational state with the glutamine residues reverted to the 26 native, protonatable glutamate side chains will be needed for full understanding of how protonation 27 and deprotonation of these residues affect the conformational dynamics of side chains and water 28 pathways. In our current model, we propose that Glu in needs to be in the protonated (neutral) state 29 to adopt the position that allows water pathways. This proposal appears harmonious with the 30 hydrophobic nature of the protein core explored by the Glu in side chain and the fact that other CLC 31 homologs use neutral residues at this position (Feng et al., 2010;Phillips et al., 2012;Stockbridge et 1 al., 2012). In addition, the proposal is consistent with MD simulations that show the Glu ex /Glu in doubly 2 protonated state is highly populated (Mayes et al., 2018) and can favor formation of water pathways 3 under certain conditions (Ko and Jo, 2010). Nevertheless, simulations with glutamate side chains in 4 the QQQ conformational state, together with explicit evaluation of H + transport (Wang et al., 2018;5 Duster et al., 2019), are needed to elaborate details of the H + -transfer steps. In addition, multiscale 6 modeling can expand the picture to include multiple pathways that are likely to occur (Mayes et al., 7 2018). Recognizing the importance of elaborating these details, the results reported here represent 8 an essential and pivotal step toward a complete, molecularly detailed description of mechanism in 9 the sui generis CLC transporters and channels. 10 11

Protein preparation and purification 13
Mutations were inserted in the wild type CLC-ec1 protein using Agilent QuikChange Lightning 14 kit and were confirmed by sequencing. Protein purification was carried out as described (Walden et 15 al., 2007), with a few changes to the protocol for ITC and crystallization experiments. For ITC, QQQ 16 or E148Q were purified in buffer A (150mM Na-isethionate, 10mM HEPES, 5mM anagrade decyl 17 maltoside (DM) at pH 7.5). For crystallization experiments, QQQ was extracted with DM. The 18 detergent was gradually exchanged for lauryl maltose neopentyl glycol (LMNG) during the cobalt-19 affinity chromatography step. The final size-exclusion chromatography step was performed in a buffer 20 containing LMNG. All detergents were purchased from Anatrace (Maumee, OH). 21

Crystallography 22
Purified QQQ protein was concentrated to at least 30 mg/mL. Concentrated protein was mixed 23 with 1.5 parts (w/w) of monolein containing 10% (w/w) cholesterol using the syringe reconstitution 24 method (Caffrey and Cherezov, 2009), to generate a lipidic cubic phase mixture. 25-nL droplets of 25 the mixture was dispensed on glass plates and overlaid with 600 nL of precipitant using a Gryphon 26 crystallization robot (Art Robbins Instruments, Sunnyvale, CA). Crystallization trials were performed 27 in 96-well glass sandwich plates incubated at 16 ºC. The best crystals were obtained using a 28 precipitant solution consisting of 100mM Tris (pH 8.5), 100 mM sodium malonate, 30% PEG 400 and 29 2.5% MPD Crystals were harvested after 3-4 weeks of incubation and flash-frozen in liquid nitrogen 30 without further additives. 31 Structure determination and refinement 1 X-ray diffraction data were collected at APS at GM/CA beamline 23ID-D and were processed 2 using XDS (Kabsch, 2010) and AIMLESS (Evans, 2006) from the CCP4 suite (Winn et al., 2011). 3 Owing to radiation damage, a complete dataset was collected by merging data from 3 different 4 crystals. Phases were obtained using PHASER (McCoy et al., 2007) with PDB ID 1ots as a search 5 model. Iterative refinement was performed manually in Coot (Emsley and Cowtan, 2004) and 6 REFMAC (Murshudov et al., 1997). The final model contained all residues except those of helix A 7 due to lack of density for this region of the protein. Helix A is observed in different conformations in 8 the monomeric versus dimeric CLC-ec1 structures, and has no impact on function (Robertson et al., 9 2010). 10

Reconstitution and chloride flux assay 11
Flux assay results presented in this paper required a variety of experimental conditions for 12 reconstitutions and flux assays, summarized in Table 2. For flux assays comparing activity at pH 7.5 13 and 4.5, purified CLC-ec1 were first reconstituted at pH 6. The samples were then aliquoted and pH-14 adjusted using a 9:1 ratio of sample and the adjustment buffer. This step was taken to eliminate 15 variability from separate reconstitutions. For experiments testing H + pumping in mutants, a pH 16 gradient was used to ensure any measured H + transport was from H + pumping and not H + leak. 17 To measure the rate of H + and Cltransport in flux assays, purified CLC-ec1 proteins were 18 reconstituted into phospholipid vesicles (Walden et al, 2007). E coli polar lipids (Avanti Polar Lipids, 19 Alabaster, AL) in chloroform were dried under argon in a round-bottomed flask. To ensure complete 20 removal of chloroform, the lipids were subsequently dissolved in pentane and dried under vacuum on 21 a rotator, followed by further drying (5 minutes) under argon. The lipids were then solubilized at 20 22 mg/mL in buffer R ( Table 2) with 35 mM CHAPS on the rotator for 1.5-2 hours. Purified proteins (0.4 23 -5 µg per mg lipids) were added to the prepared lipid-detergent mix. The detergent in the samples 24 was then gradually removed by dialysis over 2 nights. Each reconstitution sample was divided into 25 two for measurement in duplicate. Duplicates were averaged to obtain a turnover rate value. (In 26 reporting results of experiments, each "n" is an average of duplicates from a reconstitution sample.) 27 Reconstituted vesicles were subjected to 4 freeze-thaw cycles and were then extruded with 28 an Avanti Mini Extruder using a 0.4 µm-filter (GE Healthcare, Chicago, IL) 15 times. For each assay, 29 60 -120 µL of extruded sample were buffer-exchanged through 1.5-to 3.0-mL Sephadex G-50 Fine 30 resin (GE Healthcare, Chicago, IL) columns equilibrated with buffer F ( Table 2). Exchange was 31 accomplished by spinning the columns at ~1100 g for 90 seconds using a clinical centrifuge. The 1 collected sample (80 -200 µL) was then added to buffer F (500 -600 µL) for flux-assay measurement. 2

Extravesicular [Cl -] and [H + ] were monitored using a Ag·AgCl electrode and a pH electrode, 3
respectively. The electrodes were calibrated by known additions of KCl (in 20-136 nmol steps) and 4 NaOH (in 10-50 nmol steps). Sustained ion transport by CLC-ec1 was initiated by addition of 1.7-3.4 5 µg/mL of valinomycin (from 0.5 mg/mL stock solution in ethanol). 6

Isothermal titration calorimetry 7
Titration isotherms were obtained using a VP-ITC microcalorimeter (MicroCal LLC,8 Northampton, MA) at 25 ºC. For the experiment, QQQ or E148Q protein were purified in buffer A. 9 Titrant used in the experiment was 30 mM KCl in buffer A. The starting concentration of protein was 10 15-20 µM, in a volume of 1.5 mL. KCl (30 mM) was syringe-titrated into the sample cell in thirty 10-11 µL injections. The reference data were obtained by titrating buffer A into the protein-containing 12 solution. Data were analyzed using Origin 7.0 software, with fitting using the "one set of sites" model 13 (keeping n=1). The other thermodynamic parameters were obtained accordingly. Isethionate was 14 chosen as the anion of choice for purification of proteins for the ITC experiments since the QQQ 15 mutant shows aggregation upon purification in tartrate-containing solutions, which were previously 16 used (Picollo et al., 2009;Khantwal et al., 2016). The mutant is comparatively stable in isethionate 17 and continues to remain stable throughout the ITC experiment.

DEER spectroscopy 20
Protein purification and sample preparation for proteins used in DEER spectroscopy was 21 performed as described (Khantwal et al., 2016). DEER experiments were performed at 83 K on a 22 Bruker 580 pulsed EPR spectrometer at Q-band frequency (33.5 GHz) using either a standard four-23 pulse protocol (Jeschke and Polyhach, 2007) or a five-pulse protocol (Borbat et al., 2013). Analysis 24 of the DEER data to determine P(r) distance distributions was carried out in homemade software 25 running in MATLAB (Brandon et al., 2012;Stein et al., 2015). In the original five-pulse protocol paper 26 the pure five-pulse signal was obtained by subtracting the artefact four-pulse data (Borbat et al., 27 2013). This method requires the ability to discern clearly the extent of the artefact. For the data in 28 this study, we chose to simultaneously fit the four-and five-pulse data with a single Gaussian 29 component in order to improve accuracy of subtracting the four-pulse artefact (Figure 4 -figure  30 supplement 2). Confidence bands for the distance distributions were determined using the delta 31 method (Hustedt et al., 2018). The confidence bands define the 95% confidence interval that the best 1 fit distance distribution will have. In the case of a Gaussian distribution, the shape of the confidence 2 bands can be non-Gaussian. 3

Simulation system setup 4
The structure of the CLC-ec1 QQQ mutant crystallized in this work at 2.6-Å resolution was 5 used as the starting structure for the MD simulation. The 2 Clions bound at S cen and S int sites in each 6 of the two subunits were preserved for the simulation. In our initial refinement of the QQQ structure, 7 we had modeled water rather than Clat the S ext site, and therefore the simulation was performed 8 without Clat this site. The pKa of each ionizable residue was estimated using PROPKA (Olsson et 9 al., 2011;Rostkowski et al., 2011), and the protonation states were assigned based on the pKa 10 analysis at pH 4.5. Missing hydrogen atoms were added using PSFGEN in VMD (Humphrey et al., 11 1996). In addition to the crystallographically resolved water molecules, internal water molecules were 12 placed in energetically favorable positions within the protein using DOWSER (Zhang and Hermans, 13 1996;Morozenko et al., 2014). One of the energetically favorable water molecules was added right 14 between the side chains of Q148 and Q203, nicely bridging the two residues. The QQQ protein was 15 embedded in a POPE lipid bilayer using the CHARMM-GUI membrane builder (Wu et al., 2014). The 16 membrane/protein system was fully solvated with TIP3P water (Jorgensen et al., 1983) and buffered 17 in 150 mM NaCl to keep the system neutral. The resulting systems consisting of ~155,000 atoms 18 were contained in a 164 x 127 x 98-Å 3 simulation box. 19

Simulation protocols 20
MD simulation was carried out with NAMD2.12 (Phillips et al., 2005) using CHARMM36 force 21 field (Klauda et al., 2010;Huang and MacKerell, 2013) and a time step of 2 fs. Periodic boundary 22 conditions were used throughout the simulations. To evaluate long-range electrostatic interactions 23 without truncation, the particle mesh Ewald method (Darden et al., 1998)

was used. A smoothing 24
function was employed for short-range nonbonded van der Waals forces starting at a distance of 10 25 Å with a cutoff of 12 Å. Bonded interactions and short-range nonbonded interactions were calculated 26 every 2 fs. Pairs of atoms whose interactions were evaluated were searched and updated every 20 27 fs. A cutoff (13.5 Å) slightly longer than the nonbonded cutoff was applied to search for interacting 28 atom pairs. Simulation systems were subjected to Langevin dynamics and the Nosé-Hoover 29 Langevin piston method (Nose, 1984;Hoover, 1985) to maintain constant pressure (P = 1 atm) and 30 temperature (T = 310 K) (NPT ensemble). 31 The simulation system was energy-minimized for 10,000 steps, followed by two stages of 1-1 ns relaxation. Both the protein and the Clions in the binding sites were positionally restrained (k = 1 2 kcalmol -1 Å -2 ) in the first 1-ns simulation to allow the membrane to relax. In the second 1-ns 3 simulation, only the protein backbone and the bound Clions were positionally restrained (k = 1 4 kcalmol -1 Å -2 ) to allow the protein side chains to relax. Then a 600-ns equilibrium simulation was 5 performed for the system without any restraint applied. 6

Analysis of water pathways 7
The water pathways between Q148 (Gln ex ) and the intracellular bulk water was searched 8 using a breadth-first algorithm. In subunit 2 of the homodimer, Gln ex drifted up and away from the 9 "out" position at the beginning of the simulation (within 5 ns), and it did not return to the "out" position 10 during the simulation. In subunit 1, Gln ex remained near the "out" position for the first 400 ns; we 11 focused our analysis of water pathways on this subunit. The simple distance criterion of 2.5-Å for the 12 hydrogen bonds, which was found to be very useful and cheapest in computational terms in a 13 previous study (Matsumoto, 2007) was used to determine whether water molecules are connected 14 through continuous hydrogen-bonded network. The water pathway with the smallest number of O-H 15 bonds in each frame was considered as the shortest path. The first water molecule in each water 16 pathway is searched using a distance cutoff of 3.5-Å for any water oxygen atoms near the OE1/NE2 17 atoms of Q148. The water pathway is considered to reach the intracellular bulk once the oxygen atom 18 of the newly found water molecules is at z < -15 Å (membrane center is at z = 0). 19

ACKNOWLEDGMENTS 20
We thank Chris Miller and Martin Prieto for comments on the manuscript. We are grateful to 21 Brian Kobilka for use of the crystallization equipment and to K. Chris Garcia for use of the MicroCal 22   Comparing turnover rates at pH 7.5 and 4.5 (Figure 2) 333 mM KCl, 55 mM Nacitrate, 55 mM Na 2 HPO 4 , pH 6.0 pH adjustments by adjustment buffers (10x) (after reconstitution): (D) Overlay of Glu ex /Gln ex conformations seen in QQQ (purple), E148Q (blue), WT (grey) and cmCLC (pink). (E) Comparison of inter-residue distances and positioning. In WT CLCec1 (grey), the two glutamate residues face away from each other and the distance between them is 12.8 Å. In the QQQ structure, these residues reach inwards to the core of the protein, approaching within 5.6 Å of one another.   for samples with spin labels at residue 373 (B) or 385 (D) were acquired using the standard fourpulse protocol; data for the sample labeled at residue 374 (C) were acquired using the five-pulse protocol (Figure 4 -figure supplement 2). (E) The intersubunit distance for residue D417 on Helix P increases in QQQ compared to WT. pH-dependent changes in the intersubunit distance for spinlabeled D417C were shown previously (Khantwal et al., 2016).      (Lobet and Dutzler, 2006). We attempted to perform such experiments on QQQ; however, these experiments make use of the anomalous signal obtained from Brbinding, and we have not been successful in our attempts to obtain quality diffracting QQQ crystals in the presence of Br -.  In WT (grey) and QQQ (purple), residues Q207 on the two subunits are separated by 15 Å. In the Q207C structure (yellow), in which the Q207C residues are cross-linked, the H-I linker is not visible, and thus the exact positioning of the disulfide bond is uncertain; however, the observation that the H-I helices (including E203, Glu in ) are unaffected by the cross-link illustrates that H-I linker movement does not affect the main body of the protein.  (lower panels). The data shown in the left panels were collected with all 5 pulses (black) or with the amplitude of pulse 5 set to zero (green). The former experiment provides a combination of a pure five-pulse decay and a four-pulse artefact; the latter provides a measure of the four-pulse artefact.
These data were analyzed to obtain the pure 5-pulse traces, shown as the blue and orange traces in the right panels. The 5-pulse experimental results (black) are a scaled sum of the pure 5-pulse traces (blue for pH 7.5; orange for pH 4.5) and the scaled 4-pulse artefact (green). The scaling factors are obtained as variables in the fitting routine. (C) On the left are the pure five-pulse data obtained from removing the four-pulse artefact, as described in (B). On the right are the single Gaussian distributions and confidence bands for the two conditions. These data are the same as those shown in Fig. 4C.  shows that the inner water pathway is lined by F199 and I109; at the latter position, mutations were found to specifically inhibit the H + branch of the CLC transport cycle (Han et al., 2014). S107 and Y445, located at the inner gate of the Cl --permeation pathway, are shown for orientation. The middle and right panels show overlays to compare side chain positioning in the simulation snapshot to side chain position in the WT and QQQ crystal structures, respectively. Gln in and F199 side chains project into the water pathway and must rearrange to accommodate water entry into the cavity. (B) A rotated view of the inner water pathway shows tight packing of residues around Glu in /Gln in and the "interfacial pathway" residues E202, and A404 (left panel). The same view with overlay of side chains from the WT crystal structure (middle panel) and the QQQ crystal structure (right panel) show that water entry requires rotation of Glu in /Gln in away from crystallographic conformations. The tight packing explains how substitutions of large residues for E202 or A404 Han et al., 2014) would obstruct Glu in /Gln in from moving to accommodate the inner water pathway. The distribution of the side chain conformations of E202 over the trajectory (left panel) and water pathways (right panel, normalized to the total number of water pathways). Water pathways are only observed when E202 is in the cluster A or B conformation, rotated away from its starting position.
Video 1 Dynamics of water pathways and the Gln in sidechain in the QQQ mutant simulation.
Representative simulation trajectory (250.5-337.5 ns) showing the dramatic hydration of the hydrophobic lumen by water penetration from the intracellular bulk. The Gln ex (Q148) sidechain is accessible to the intracellular bulk through the continuous water pathways spontaneously and frequently formed during the simulation. The water pathway conformations undergo dynamical changes due to the sidechain orientations of Gln in . Protein helices are shown as orange ribbons. The Clion at the S cen site is shown as a green sphere. Key amino acids in proximity to the water pathways or involved in Cl --coordination are shown. See also Figure 5 and Figure 5 supplements.