We first examine the role of the gating glutamate (Glu_g) in proton coupling by asking, as with conventional CLCs^12,13, how point substitutions at this position affect F⁻: H⁺ transport-stoichiometry. Ion fluxes are quantified in Eca-reconstituted liposomes pre-loaded with high F⁻ and suspended in low F⁻ (Fig 1a). Equimolar F⁻ efflux and H⁺ influx in wildtype (WT) Eca confirms its 1:1 stoichiometry, and similar experiments with Glu_g mutant E118D (Fig 1b) establish a 10-fold lower transport rate and weaker coupling, with ~2:1 stoichiometry. However, non-dissociable mutants E118A,Q produce a completely unexpected, even alarming, result: inversion of ion selectivity. Neither mutant transports F⁻, and instead both pass Cl⁻ at over twice the WT rate for F⁻, without concomitant H⁺ movement (all rates reported in Table 1). Since this result has no analogous precedent among Cl⁻ CLCs, it raises suspicions that previous CLC crystal structures may be inadequate templates for interpreting functional behaviors of the phylogenetically distant CLC^F proteins.

Accordingly, a crystal structure at 3.0 Å resolution of homodimeric Eca was solved in complex with a “monobody” crystallization chaperone¹⁴ that does not affect transport activity (Fig 2a, b, Supplementary Fig 1, Table 2). As expected from known CLC structures, each Eca subunit shows extracellular and intracellular aqueous vestibules impinging upon a polar but largely anhydrous ion-coupling region in the center of the protein 6–8 Å in length (Fig 2c). The protein’s fold mirrors other CLCs (Supplementary Fig 2a), but several of its 16 helices are shorter than their counterparts, and some are imprecisely aligned to them. The αA and αR helices of the Cl⁻ CLCs are absent in Eca (we retain conventional CLC helix labels, starting with helix αB), and the equivalent residues of the C-terminal αR-helix trace a well-ordered but meandering extended backbone trajectory.

Two F⁻ ions, F⁻_ex and F⁻_cen, appear in each Eca subunit located at the “external” and “central” sites in the canonical anion-binding region observed in all CLCs, roughly halfway across the membrane. These densities may be confidently assigned as F⁻ ions rather than isoelectronic waters from their congruence with Cl⁻_ex and Cl⁻_cen in structural alignments (Supplementary Fig 2b), and since many of their coordinating residues match those known in the Cl⁻ CLCs (Fig 3). For instance, both F⁻_ex and Cl⁻_ex are mainly embraced by backbone amides in equivalent stretches of sequence (G116-G119 in Eca, G146-G149 in CLC-ec1), as well as by the positive dipoles of the N-termini of helices αF and αN in both. Moreover, the tyrosine hydroxyls coordinating the central anions in the F⁻ and Cl⁻ transporters coincide precisely (Y396 in Eca, Y445 in CLC-ec1). Although the resolution here is too low to quantify F⁻ occupancy values, precedent from other CLC structures⁷ and the clear difference densities of these light ions suggest simultaneous occupancy of both.

In Cl⁻-transporting CLCs, the Glu_g sidechain has been variously observed in three rotameric positions proposed as intermediates in the transport cycle^6,9,15: ‘Down’, wherein the deprotonated carboxylate group displaces Cl⁻ from the central anion site, ‘Middle’, where also deprotonated it occupies the external site, and ‘Up’, extended such that the carboxyl group reaches for the extracellular solution, with Cl⁻_ex and Cl⁻_cen simultaneously occupying both sites. The Up-configuration has been seen only with the protonated surrogate glutamine substituted at this position, never with glutamate, whose carboxyl group, when deprotonated, seeks the electropositive anion-binding sites. But in WT Eca, Glu_g adopts the Up-rotamer (Fig 3a), allowing both F⁻ ions to occupy the transport sites; its protonation state is unknown under these pH 6 crystallization conditions.

The profound reversal of ion selectivity in E118Q would seem to anticipate a corresponding structural alteration. However, a crystal structure of this mutant (Fig 3c, Supplementary Fig 1, Table 2) fails to show any significant differences with WT. The appearance of F⁻ ions at the same sites as in WT motivated flux experiments (Fig 4) showing that Cl⁻ transport in this mutant is strongly blocked by F⁻ (K_i = 20 μM in the presence of 140 mM Cl⁻). This effect provides a proxy for F⁻ binding, at least 10-fold stronger than in WT, as observed previously by isothermal titration calorimetry (ITC)⁵ and confirmed here (Supplementary Fig 3). While these blocking experiments demonstrate tight F⁻ binding to the mutant’s transport machinery, they provide no information about where the ion binds - to the external site, the central site, or both. Unfortunately, we could not obtain ITC heat signals with E118Q or diffraction-quality crystals formed in F⁻-free Cl⁻ solutions.

In conventional CLCs, the Up position of Glu_g - previously observed only with substituted Gln - is widely considered to represent an outward-open state of the transport cycle. In this view, binding sites for Cl⁻_cen, Cl⁻_ex, and H⁺ enjoy access to extracellular solvent, either unfettered or accompanied by subtle pH-dependent backbone movements^7,16,17, while intracellular access is prevented by a “gate” formed by Cl⁻-coordinating sidechains of conserved Ser and Tyr at the central site^15,18,19. This picture is completely inverted in the CLC^F structure. Now for the first time Glu_g appears in the Up-configuration, its carboxyl in direct exchange with extracellular protons; remarkably, the anion-binding sites are accessible only to intracellular solvent, via a wide aqueous vestibule leading to F⁻_cen bound at its apex and F⁻_ex ~6 Å behind it (Fig 2c). Extracellular access to both ions is blocked off primarily by a stretch of extended sequence in conserved residues immediately preceding Glu_g (S114-R117), unlike the more open pathway in conventional CLCs. The Up-configuration of Glu_g positions it far away from the intracellular side of the protein. This antiporter conformation is simultaneously accessible to extracellular H⁺ (but not F⁻) and to intracellular F⁻ (but not H⁺).

Since numerous F⁻-coordinating moieties in Eca match those in the Cl⁻ CLCs, the new structures provide no immediate explanation for the unusual F⁻ specificity of the CLC^Fs. The coordination of F⁻_ex, mostly by backbone amides, is similar to the surround of Cl⁻_ex in other CLCs, except for the sidechain of T320, whose hydroxyl group hovers ~4 Å away from F⁻_ex (Fig 3). It is F⁻_cen that is differently clothed. The strictly conserved Cl⁻_cen-coordinating central serine residue in the conventional CLCs is missing in the CLC^Fs, and in a roughly equivalent position lies a conserved Met residue (Met79), suggested from sequence-gazing and confirmed by mutagenesis as a F⁻ selectivity determinant^2,5. It is particularly notable that its terminal methyl and γ-methylene groups, rendered weakly electropositive by proximity to the δ-sulfur, are positioned near F⁻_cen, since this arrangement quite precisely reprises F⁻ coordination in a functional Met mutant of a F⁻-specific ion channel²⁰. In addition, the coordination of F⁻_cen by the hydroxyl of Y396 establishes this residue as equivalent to the inner-gate tyrosine that coordinates Cl⁻_cen in Cl⁻ CLCs (Y445 in CLC-ec1). This same Tyr does double-duty by also orienting its aromatic ring to approach this same F⁻ edge-on, an electropositive quadrupolar interaction previously observed in the only other known F⁻ transport protein family, the Fluc ion channels^21,22. An additional oddity of the F⁻_cen coordination sphere is a close contact (3.3 Å) with the carboxyl group of E318, which has no counterpart in conventional CLCs. Beyond its involvement in F⁻ coordination, it is conspicuous from its location at the top of the intracellular cavity, close to where the carboxylate of Glu_g would be in its Down position, suggestive of a critical H⁺-handoff function. We suppose that in this configuration E318 is a protonated H-bond donor to F⁻_cen. The position of F⁻_cen at the top of the intracellular vestibule implies that it is also partly hydrated by crystallographically invisible bulk waters.

The CLC^F structure thus identifies three previously unsuspected residues contributing to F⁻ coordination: Y396, E318, and T320, the latter two without equivalents in the Cl⁻ CLCs. Mutants at these positions were constructed to gauge their effects on F⁻ transport rate, proton coupling, and anion specificity using liposome flux experiments. The mutants (Y396A, E318Q/A, T320A) express similarly to WT and show dimeric, homogeneous size-exclusion profiles. All give qualitatively similar results: 4–20-fold inhibition of F⁻ transport rates, specificity for F⁻ with no Cl⁻ permeation, and retention of proton coupling with respectable stoichiometry between 1:1 and 2:1 (Table 1). The full preservation of proton coupling in Y396A is particularly surprising, since the equivalent mutant in a Cl⁻-transporting CLC is severely uncoupled¹⁸. These facts rule out a few obvious mechanistic suggestions irresistibly arising from the structure: that E318 might be a required way-station for H⁺ transit, that Y396 might determine F⁻ selectivity and H⁺ coupling, and that H-bond donation to F⁻ by T320 might be necessary for proper antiport behavior.

Having failed to eliminate F⁻/H⁺ coupling by the sidechain substitutions above, we wondered if a Hail-Mary maneuver - glycine substitutions of residues coordinating F⁻_ex through backbone amides - might enhance conformational flexibility in this region and thereby place a thermodynamic penalty on F⁻ binding. All such mutants severely impair function except for one, V319G, which only modestly inhibits transport rate (3-fold), weakens H⁺ coupling 2-fold, and preserves F⁻ specificity (Table 1). In addition, ITC experiments show 5-fold decreased F⁻ binding affinity of the mutant (K_D = 1 mM) compared to WT (Supplementary Fig 3). This mutant apparently allows an ‘extra’ F⁻ ion to sneak through the transport cycle unaccompanied by proton counter-transport, perhaps due to its weakened binding; however, we cannot distinguish such ‘slippage’ from a fundamental change in mechanism to a tight 2:1 F⁻/H⁺ coupling.

The altered F⁻ affinity of V319G combined with its overall maintenance of F⁻-specific antiport motivated us to crystallize this mutant. The resulting structure reveals an unexpected conformation (Fig 5a, Supplementary Fig 4, Table 2). Here, Glu_g adopts the Down position, replacing F⁻_cen in the central anion-binding site and thereby blocking off the intracellular access to F⁻_ex seen in WT, a configuration similar to that observed in a eukaryotic CLC antiporter¹⁵. The Down rotamer has not been previously observed in any bacterial CLC structures but has been proposed as an essential, universal intermediate in the CLC antiport cycle⁹. In this configuration, backbone atoms around the mutated residue remain unmoved from WT positions. F⁻_ex, the sole F⁻ ion in the structure, having moved ~1.5 Å toward the extracellular side, is now occluded. This small outward displacement changes the ion’s coordination; it loses the mutated residue’s backbone amide and gains a short (2.6-Å) H-bond from T320, as well as an edge-on aromatic quadrupole from F158, which moves in to fill the space vacated by Glu_g’s 7.5- Å excursion from its Up position (Fig 5a). Although the F⁻ ion appears shallowly buried, Glu_g is not; its carboxyl group now directly faces the intracellular vestibule, where it is free to exchange solvent protons, a proton accessibility that is flipped compared to the WT structure (Fig 5b).

All chemcals were purchased at the highest grade from Sigma-Aldrich or Thermo Fisher. n-decylmaltoside (DM), n-decylphosphocholine (fos-choline-10), and 3-[(3-cholamidopropyl)-dimethylammo-nio]-1-propane sulfonate] (CHAPS) were obtained from Anatrace (Maumee, OH), and E. coli polar phospholipids (EPL) were from Avanti Polar Lipids (Alabaster, AL). K-isethionate solutions were prepared by titrating isethionic acid (Wako Pure Chemical Industries, Osaka, Japan) with KOH.

The Eca CLC^F construct used here from Enterococcus casseliflavus (GenBank EEV30821.1) contains three N-terminal alanines arising from the cloning site, but residues are numbered according to the WT sequence, ignoring this insertion. The “WT” background construct, on which all mutants were made, contains a M4I mutation to remove a possible alternative start codon. All Eca proteins carry an uncleaved C-terminal hexahistidine tag preceded by a GSGG linker. Eca was expressed from vector pASK90 with a tetracycline promoter³¹, and all mutations were made via standard PCR techniques.

Eca was expressed in BL21(DE3) cells. In brief, transformed cells were grown at 37 °C to ~0.8 OD in terrific broth and induced with 0.2 μg/mL anhydrotetracycline for 3 hr. Cells were harvested by centrifugation, and the pellets were resuspended in 100 mM NaCl, 50 mM tris pH 7.5 with DNAse and lysozyme. Cells were sonicated, and Eca was extracted with 4% DM for 2 h at room temperature. Samples were then centrifuged to pellet insoluble material and loaded on Talon cobalt resin (Takara Bio, Kusatsu, Japan) equilibrated with 100 mM NaCl, 5 mM DM, 20 mM tris pH 7.5. Protein was washed and eluted sequentially with 40 mM and 400 mM imidazole in the above buffer. The protein was concentrated and purified over a Superdex 200 Increase size-exclusion column (GE Healthcare) equilibrated with either 25 mM 4-(2-hydroxyethyl)-1-piperazine ethanesulfonic acid (HEPES), 100 mM NaCl, 5 mM DM, pH 7 (functional assays), or 10 mM HEPES, 100 mM NaF, 5 mM DM, pH 7 (crystallography). Eca for isothermal titration calorimetry was prepared as above, except that the Talon equilibration buffer was 10 mM HEPES, 100 mM Na/K tartrate, 2 mM NaCl, 5 mM DM, pH 7.5, and the Superdex-ITC buffer was 10 mM HEPES, 100 mM Na/K tartrate, 5 mM DM, pH 7.0.

Proteoliposomes were formed by mixing purified Eca with 10–20 mg/mL stocks of EPL solubilized with 30–40 mM CHAPS. Protein/lipid mixtures were extensively dialyzed using 10 kDa MWCO Slide-A-Lyzer cassettes (Thermo Fisher) against 150 mM 3-(N-morpholino)propanesulfonic acid (MOPS) pH 7.5 with either 150 mM KF or 150 mM KCl. Proteoliposomes were freeze-thawed 3x prior to use. For F⁻ blocking experiments (Fig. 4) F⁻ at the reported concentration was added to KCl-dialyzed proteoliposomes prior to the freeze-thaw cycles. Final protein/lipid ratios were 3–15 μg protein/mg lipid.

The monobody crystallization chaperone “X1” (Supplementary Fig 6) was selected from a phage display library as described¹⁴, and mutated slightly to improve crystal contacts. It was expressed from a pHFT2 vector in BL21(DE3) cells. Transformed cells were grown to OD 0.8 in terrific broth at 37 °C and induced 3 hr at 30°C with 0.2 mM isopropyl β-D-1-thiogalactopyranoside. Cells were harvested and sonicated as above. Sonicated cells were clarified by centrifugation and loaded on Talon cobalt resin by batch binding for 3 h. The N-terminal hexahistidine tag was removed by on-column cleavage by overnight incubation with TEV protease (0.2 mg/L of cell culture). Cleaved monobody was rinsed off the column with equilibration buffer, concentrated, and purified over a Superdex 75 Increase column equilibrated with 10 mM HEPES, 100 mM NaF, pH 7.

Liposomes loaded with either 150 mM KF or KCl were extruded through 0.4 μm membranes (Whatman) and 100 μL were centrifuged over a 1.5 mL Sephadex G-50 column equilibrated with flux-assay buffer (1 mM MOPS pH 7.5, 150 mM K-isethionate, 123 mM Na-isethionate). Liposomes were diluted into a stirred beaker containing flux buffer supplemented with either 1 mM KF or 1 mM KCl for F⁻ or Cl⁻ conditions⁵. For F⁻ blocking experiments, KF at the reported concentration was present on both sides of the liposome membrane. Flux was initiated by the addition of 0.9 μM valinomycin. Appearance of ions in the liposome suspension was followed continuously with F⁻, Cl⁻, or H⁺-specific electrodes in a stirred cell. Reported initial flux rates were calculated 2–10 sec after adding valinomycin, and are corrected for the background leakage through protein-free liposomes.

Isothermal titration calorimetry was performed as previously described⁵, with both protein and F⁻ in the above Superdex-ITC buffer. Measurements were made in a TA Instruments (New Castle, DE) Nano-ITC. Protein (170 μL) at ~30–250 μM was titrated with 2 μL injections of 5–30 mM KF at 25 °C, and data were fit to single-site isotherms.

Purified Eca and monobody X1 were concentrated and mixed to a final protein stock of 10 mg/mL Eca and 2.8 mg/mL monobody, supplemented with 11 mM fos-choline-10. Crystals were formed via the sitting-drop vapor diffusion method, with drop mixtures containing between 1 and 3 μL of protein and well solution mixed in equal volume. Well solutions consisted of 100 mM K-formate, 100 mM 2-[(2-amino-2-oxoethyl)-(carboxymethyl)amino]acetic acid (ADA) pH 6.0 or 6.2, and 20–24% PEG600. Crystals were incubated for ~5 weeks at 22 °C. Serial additions of cryoprotectant were required, consisting of 27%, 29%, and 34% PEG600 in solutions matching the final crystal solution. The initial drop volume of 27% PEG600 was first slowly added, followed by 2x the initial volume of 29% PEG600, and 4x the initial volume of 34% PEG600. Crystals were then picked with cryo loops and frozen in liquid nitrogen.

Datasets were collected at the Advanced Light Source beamlines 5.0.1 and 5.0.2. Frames were integrated, scaled, and merged using Xia2/DIALS/Aimless^32–34. Molecular replacement of WT Eca was done in PHASER³⁵, using as search model CLC Cl⁻/H⁺ antiporter from Synechocystis sp. (PDB# 3Q17 chain B) trimmed with Chainsaw³⁶ based upon a sequence alignment from AlignMe³⁷. The two monobodies were then placed using chain E from PDB# 5FXB, manually trimmed to remove regions of differing sequence. Molecular replacement for the mutant structures was done using the refined WT structure. Structure were refined using phenix.refine³⁸ and Refmac5³⁹, with final refinement done in Phenix. TLS and torsion-angle NCS restraints were used for all structures. V319G and E118Q refinement used reference model restraints against the final WT monobody chains, but not the Eca chains. Real-space refinement was done in COOT⁴⁰. Extent of F⁻ accessible vestibules was calculated in HOLLOW⁴¹.

In plots and Table 1, rates report mean ± s.e.m. of 3–6 repeats and of at least two independent protein preps. ITC experiments (illustrated in Fig SI 3) were repeated three times.

Crystallographic data are deposited in the Protein Data Bank: WT Eca (PDB 6D0J), E118Q (PDB 6D0K), V319G (PDB 6D0N). Source data for Figs 3d, 4b and 4c and Supplementary Fig 3 are available on request.