Introduction

Acid-sensing ion channels (ASICs) are proton-gated cation channels widely expressed in nervous systems1,2. They are trimeric channels with two-transmembrane domains per subunit and are encoded by at least four genes resulting in homomeric (ASIC1a, ASIC1b, ASIC2a, ASIC3, and ASIC4) and heteromeric channels3,4. Mammalian ASIC1a is the most studied ASIC to date, serves as a primary sensor of acidosis in the brain and is involved in a variety of biological processes, including synaptic function and plasticity5,6,7,8, pain sensation9,10,11, seizure12, and neuronal injury30,33. Fourth, the second transmembrane domain (TM2) helix of ΔASIC1/JNJ–799760 undergoes a (TM2b) domain swap that allows the Gly443–Ala444–Ser445 motif (a.k.a. the GAS belt) to adopt an extended conformation, as that adopted by closed-state structures (Supplementary Table 1 and Supplementary Fig. 5). In contrast, the TM2 domain of 3S3W is continuous with no GAS belt extension or TM2b domain swap (Fig. 9e). Finally, the channel gate of ΔASIC1/JNJ–799760 is shut (as in closed-state structures), in contrast to the (paradoxical) “open” gate in 3S3W (Supplementary Table 1 and Supplementary Fig. 5).

Fig. 9: Key conformations of the ΔASIC1/JNJ-799760 complex.
figure 9

a Superposition of ΔASIC1/JNJ-799760 (colored by domain) and 3S3W apo (gray/transparent) structures. Binding of JNJ-799760 causes outward pivots of α4 and α5 (split into α5a and α5b) and displacement of the acidic loop. b Superposition of ΔASIC1/JNJ-799760 (green) and 3S3W (gray/transparent) near the acidic pocket. Note the increased side-chain distance of D238–D350 and of E239–D346. Numbers are in Å. Note that E239 and Y341 side chains in 3S3W clash with JNJ-799760. c Superposition of ΔASIC1/JNJ-799760 (green) and 3S3W (gray) for a portion of the β1–β2 linker. Note the ~180° flip of the T84–R85 peptide bond orientation. d Superposition of ΔASIC1/JNJ-799760 (green) and 3S3W (gray) for a portion of the β11–β12 linker. Note the swap of L414–N415 side-chain orientations and displacements of L414 and β11–β12 linker due to JNJ-799760 binding. e Comparison of chain A TM domains of ΔASIC1/JNJ-799760 (green) and 3S3W (gray). Note the extended GAS belt conformation and TM2b domain swap in ΔASIC1/JNJ-799760.

Full size image

The closed gate in ΔASIC1/JNJ-799760 rules out the possibility of an open or desensitized-like channel. Remarkably, the conformational features of ΔASIC1/JNJ-799760 described above are adopted invariably by all closed-state structures (Supplementary Table 1). Only two (closed gate and TM2b domain swap) of the five conformational features are shared between closed and desensitized structures, whereas the other three (expanded acidic pocket, flipped T84–R85 bond and non-swapped L414–N415 side chains) are characteristic only of closed channels, demonstrating that ΔASIC1/JNJ-799760 is a closed-state structure.

Additionally, ion binding data further support this contention. In structures of cASIC1 channels in the open20 and desensitized3,21,25, but not closed18,32, states, a Cl− ion is bound to a site coordinated by R310 and E314 from one subunit and by K212 from a neighboring subunit. There is no evidence of Cl− binding to this site in the structure of ΔASIC1/JNJ-799760.

Taken together, JNJ-799760 stabilizes the closed state of the channel, as demonstrated by both functional and structural results.

JNJ-67869386 and NJ799760 occupy the same binding pocket

Though structurally distinct, JNJ-67869386 and JNJ-799760 interact functionally in a way consistent with their binding competitively to the same site. As shown in Fig. 10a, the time course of current recovery from co-inhibition by JNJ-799760 and JNJ-67869386 lies intermediate between the kinetics of recovery individually from each of the molecules and is biphasic with characteristic time constants for the two molecules. This profile is exactly what would be expected if binding of the two molecules is mutually exclusive. If JNJ-799760 and JNJ-67869386 bound independently to distinct sites, current recovery would be dominated by the kinetics of the molecule that dissociates from the channel more slowly, JNJ-67869386, contrary to what is observed here.

Fig. 10: JNJ-67869386 and JNJ-799760 bind to the same pocket.
figure 10

a Kinetics of current recovery from inhibition by 100 nM JNJ-67869386, 1 µM JNJ-799760 or 100 nM JNJ-67869386 + 1 µM JNJ-799760. The steady-state current in the presence of compound is subtracted before normalization. Data are fitted to a single or double exponential function (dashed lines) with time constants of 9.1 ± 0.4 s (n = 11; solid circles), 3.1 ± 0.2 s/14.9 ± 1.6 s (n = 5; half-filled circles) and 3.9 ± 0.2 s (n = 5; open circles), respectively, for JNJ-67869386, JNJ-67869386 + JNJ-799760, and JNJ-799760. The three groups of data are significantly different from one another (p < 0.001; Two-way ANOVA). The holding pH is 8.2. b Docking of JNJ-67869386 (salmon carbons) to the binding site of JNJ-799760 on ΔASIC1 (green carbons). Stick representation of JNJ-67869386 and channel residues with which it interacts. Specific interactions between JNJ-67869386 and ΔASIC1 are shown as yellow dotted lines labeled with distances in Å. c Concentration-dependent inhibition of pH 6.0-induced currents of rASIC1a WT, F98A, and Y340A channels by JNJ-67869386 (solid symbols) and JNJ-799760 (open symbols). Data are fitted to a logistic function (dashed and dotted lines). For JNJ-67869386, IC50 = 28.9 ± 3.6 nM (n = 7), 626.3 ± 11.0 nM (n = 4) and 571.8 ± 176.9 nM (n = 5), respectively. For JNJ-799760, IC50 = 24.9 ± 1.3 nM (n = 4), 725.9 ± 356.1 nM (n = 4) and 260.0 ± 89.8 nM (n = 4), respectively. All mutant channel data are significantly different from the corresponding WT (p < 0.001; two-way ANOVA).

Full size image

To understand the molecular interactions of JNJ-67869386 with the channel, we developed a binding model based on the crystal structure of ΔASIC1/JNJ-799760. We hypothesized that both compounds occupy the same site (based on the data in Fig. 9a and their similar functional profiles). Molecular docking of JNJ-67869386 to the JNJ-799760-binding site in ΔASIC1/JNJ-799760 provides a model consistent with the molecules binding at the same site (Fig. 10b). The main hydrogen bond is formed between JNJ-67869386 and the backbone nitrogen of V327, an interaction also made by JNJ-799760. The amino moiety of the amino-indazole in JNJ-67869386 makes an additional hydrogen bond to the backbone carbonyl of C336. An equivalent interaction is not present in ΔASIC1/JNJ-799760. The core of JNJ-67869386 makes aromatic and hydrophobic interactions with the side chains of F99, L115, L116, Y159, M326, V327, and Y341, similar to those that JNJ-799760 makes with these residues in the crystal structure. Due to its shorter length, JNJ-67869386 does not form direct hydrogen bonds with L116 or P232 as JNJ-799760 does, although there may be a water-bridged interaction with P232. Overall, JNJ-67869386 makes interactions with many of the same residues that JNJ-799760 interacts with. Importantly, all the residues that JNJ-678969386 and JNJ-799760 interact with in the model or the crystal structure are conserved between cASIC1 and rASIC1a.

The docking model and co-crystal structure produce predictions regarding the interactions of JNJ-67869386 and JNJ-799760 with the surrounding channel residues that are testable by site-directed mutagenesis. Based on these interactions, we studied the effect of these molecules on two rASIC1a mutants, F98A and Y340A (corresponding to F99A and Y341A in cASIC1, respectively). The side chains of both F99 and Y341 of cASIC1 make aromatic interactions with JNJ-799760 in the crystal structure and with JNJ-678969386 in our binding model. Removing the aromatic side chain by mutating to alanine is predicted to decrease the potency of the compounds. Indeed, the potency of inhibition of pH 6.0-induced current by JNJ-799760 and JNJ-67869386 is significantly decreased for both mutants as predicted (Fig. 10c), providing further functional validation of the co-crystal structure and docking model. The pH dependence of activation and steady-state desensitization is similar for the mutant and wild-type channels (Supplementary Fig. 6), making the pharmacological comparisons straightforward as the same holding and test pHs can be (and are) used for all three channels. Taken together, these results indicate that the two molecules interact similarly and predictably with the two amino acid residues, strongly supporting the contention that they bind to the same pocket.

Discussion

Small molecules have been shown to variously modulate ASIC1a. For example, spermine decreases desensitization and current inhibition by PcTx116,34; compound 5b, a PcTx1-inspired small synthetic molecule that is modeled to bind in the acidic pocket, is an allosteric inhibitor of channel activation28; Daurisoline and 2-guanidine-4-methyl-quinazoline (GMQ, a non-proton agonist of ASIC3) both cause an acidic shift of the pH dependence of ASIC1a activation and desensitization35,36; and histamine shifts the pH dependence of activation toward more basic pHs and of desensitization toward more acidic pHs37,38. However, these and other studies of small molecule modulation of ASIC1a to date have generally lacked substantive evidence necessary for elucidating the molecular and structural basis of modulation. As of this study, amiloride was the only small molecule co-crystalized with ASIC1. Aside from binding to and plugging the channel pore, it also binds to the acidic pocket, to a site distinct from JNJ-799760 but overlap** with PcTx120. However, the functional relevance of this binding is unknown.

In this study, we combine functional, structural, computational, and mutational approaches to elucidate a mechanism by which two chemically distinct small molecules modulate the gating of ASIC1a. We show that structurally, both molecules bind to the same and previously unrecognized allosteric site at the acidic pocket. The crystal structure of ΔASIC1/JNJ-799760 reveals that JNJ-799760 keeps the acidic pocket in an expanded conformation as well as the channel in an overall conformation commensurate with the closed state. Functionally, these molecules serve as both negative and positive modulators. As negative modulators, they inhibit H+-evoked currents by shifting the pH dependence of activation toward more acidic pHs. As positive modulators, they impede channel desensitization and tachyphylaxis by causing an acidic shift of the pH dependence of steady-state desensitization. As such, they are gating modifiers that stabilize the closed state, corroborating the conclusion from the structural studies. Our results identify a previously unknown drug-binding site and represent a direct structural demonstration that binding of a small molecule modulates the gating of an ASIC channel.

JNJ-67869386 and JNJ-799760 bind to a previously unrecognized site at the acidic pocket, in a conformation distinct from what PcTx1 or amiloride interacts with19,20,25. Binding of different chemotypes of small molecules to this site highlights the significance of the locus for pharmacological modulation.

The acidic pocket contains three pairs of conserved acidic amino-acid residues. In the closed state of the channel, these pairs are deprotonated. The resulting electrostatic repulsion is thought to be responsible for maintaining the acidic pocket in an expanded conformation18. Protonation of these residues in the open and desensitized states removes the electrostatic repulsion, causing the acidic pocket to collapse. JNJ-799760 helps to keep these pairs apart via extensive interactions with nearby residues. The resulting expanded acidic pocket requires higher H+ concentrations to collapse, leading to an acidic shift in the pH dependence of channel activation.

Functional studies indicate that JNJ-799760 and JNJ-67869386 do not bind to desensitized channels at equilibrium. First, JNJ-799760 and JNJ-67869386 cannot bind to pre-desensitized channels, consistent with the notion that the compound binding site is either inaccessible, distorted or non-existent with the acidic pocket being in a collapsed conformation. Second, closed channels do not desensitize with compound bound. Instead, compound must first dissociate from the channel before desensitization can occur. Third, although open channels can initially undergo desensitization with compound bound, the compound-bound desensitized states (OBD in Fig. 6a) are unstable and channels irreversibly transition to compound-unbound desensitized states (OD).

Our X-ray data provide a structural basis for this mechanism. Structural overlays show that the side chain positions of E239 and Y341 in both 3S3W (Fig. 9b) and 2QTS (Supplementary Fig. 7) are in direct steric clash with JNJ-799760 in our co-crystal structure. This mutual exclusivity/occlusion indicates that the conformation of the binding pocket in the desensitized state is incompatible with JNJ-799760 binding. A similar comparison with 3S3X, a structure of the ΔASIC1/PcTx1 complex at pH 5.525, shows the existence of the same incompatibility (Supplementary Fig. 8). In addition, binding of JNJ-799760 pushes the α4 and α5 helices, with which PcTx1 makes extensive contacts, away from the acidic loop compared to the PcTx1-bound conformation (Supplementary Fig. 8). These structural conflicts are also evident in an overlay of 3S3X with the JNJ-67869386-docked pose (Supplementary Fig. 8), in agreement with the kinetic and steady-state data on the functional antagonism of the PcTx1 effect by JNJ-67869386.

Beyond the acidic pocket, binding of JNJ-799760 also produces large, global changes in the channel conformation, including (1) causing an about-face flip of the T84–R85 peptide bond in the β1–β2 linker of the palm domain, (2) un-swap** the side-chain positions of L414–N415 in the β11–β12 linker of the palm domain, (3) inducing the TM2b domain swap and extended conformation of the GAS belt, and (4) shutting the channel gate. These are also conformations invariably adopted by closed-state structures (Supplementary Table 1), suggesting that binding of JNJ-799760 destabilizes the desensitized-like state of the apo structure in favor of the closed state, which corroborates the conclusions based on our functional studies. These results, along with the fact that the predictions of our structural analysis are borne out by the functional studies of mutant channels, provide strong evidence of structure–function correlation.

As with JNJ-799760 and JNJ-67869386, divalent cations, such as Ca2+, Ba2+, and Mg2+, also stabilize the closed state of ASIC1a34. Yoder et al. 32 showed that binding of these ions at certain extracellular sites in ASIC1 is state dependent—occupancy is observed in the closed, but not in the desensitized state. These findings suggest a degree of similarity between modulation by divalent cations and our molecules. It is worth noting, however, that binding of divalent cations to these sites does not corelate with functional stabilization of the closed state, in contrast to what we observe for JNJ-799760 and JNJ-67869386.

Although a crystal structure of the ASIC1/JNJ-67869386 complex is lacking, several lines of evidence strongly indicate that JNJ-67869386 binds to the same pocket as JNJ-799760. First, current recovery from simultaneous inhibition by the two compounds takes on an intermediate, biphasic time course with characteristic time constants for the two individual molecules. This indicates that binding of these molecules is mutually exclusive rather than independent, the latter of which would, contrary to observation, result in the time course being dominated by the slower recovery from JNJ-67869386. Second, docking of JNJ-67869386 to the JNJ-799760 site produces many of the same interactions with the surrounding channel residues as JNJ-799760 does in ΔASIC1/JNJ-799760. Third, mutations of residues expected to interact with both JNJ-799760 (based on the crystal structure) and JNJ-67869386 (based on the docking model) significantly and similarly change the potency of JNJ-799760 and JNJ-67869386 and in the same direction as predicted. Lastly, effects of JNJ-67869386 and JNJ-799760 on ASIC1a function are qualitatively identical.

Given that the acidic pocket also adopts a collapsed conformation in the open state20, it is possible that at equilibrium, our molecules do not bind to the open state either. Consistent with this scenario, open-channel desensitization is faster in the presence of these compounds, suggesting compound-induced destabilization of channel opening. However, we cannot address this question directly in the current study because (1) structural information on an open channel in complex with our compounds is lacking (Indeed, whether our compounds can be co-crystalized with an open channel would itself constitute a direct test of this hypothesis), and (2) the rate of dissociation of these molecules from wild-type ASIC1a is too slow relative to the duration of channel openings to observe their dissociation directly from the open channel. Experiments using a mutant channel with long open durations should help test this hypothesis functionally.

Of the amino acid residues in ASIC1a that interact with JNJ-799760, only three are different in ASIC2a. Amino acids E97, Y158, and Y340 in rASIC1a are Gly, Leu, and His, respectively, at the corresponding positions in rat ASIC2a. However, JNJ-799760 and JNJ-67869386 are inactive at ASIC2a (IC50 > 50 µM), suggesting that one or more of these residues may be critical for the ASIC1a selectivity. Alternatively, the apparent selectivity of these compounds may be a consequence of diminished affinity and/or efficacy at high proton concentrations necessary to activate ASIC2a. Additional experiments with ASIC2a mutated to the corresponding ASIC1a amino acids at these positions may help to distinguish between these hypotheses.

In this study, we identified a small molecule binding site in ASIC1 and presented direct structural evidence that small molecule binding modulates the gating of an ASIC channel. Furthermore, we elucidated the molecular mechanism and structural basis of this modulation. Our findings provide important mechanistic and structural insight into the modulation of ASIC channels and contribute to the understanding of structure, function, and therapeutic targeting of this class of ion channels.

Methods

Cell culture and transient transfections

Chinese Hamster Ovary (CHO) cells stably expressing rat ASIC1a were cultured in Ham’s F12, supplemented with 10% FBS, 1% penicillin–streptomycin and 500 µg/mL G418 and incubated at 37 °C with 5% CO2. Transient transfections of rASIC1a and cASIC1 wild-type or mutant channels in CHO cells were performed with Lipofectamine 3000 (Thermo Fisher Scientific) according to the manufacturer’s recommendations. The green fluorescent protein (GFP) cDNA (at 1/10 the amount of channel cDNA) was co-transfected with channel cDNA to aid the identification of transfected cells under the microscope in patch clamp experiments. All cDNAs were sequence verified.

Electrophysiology

Cells were freshly dissociated with CellStripper (Corning) and dispersed in a chamber on the stage of an inverted microscope. Upon formation of the whole-cell conformation, the cell was lifted from the bottom of the chamber and placed at the tip of a tubing perfusing an extracellular solution containing (in mM): 149 NaCl, 2 CaCl2, 4 KCl, 1 MgCl2, 5 glucose, 10 HEPES, pH 7.4, 310 mOsm/L. Extracellular solutions at more basic or acidic pHs were made by titrating the above pH 7.4 solution with NaOH or HCl (5 mM MES was added to solutions at pH 6.0 and lower). Pipette electrodes were filled with an intracellular solution containing (in mM): 135 KCl, 4 MgATP, 0.3 Na2GTP, 10 EGTA, and 20 HEPES, pH 7.2, 290 mOsm/L.

All recordings were performed at room temperature using an Axopatch 200B amplifier and pClamp 11 software (Molecular Devices). Currents were measured by whole-cell patch clamp, digitized at 10 kHz and lowpass filtered at 2 kHz. Series resistance was 75% compensated. Responses were elicited by rapid perfusion of acidic solutions using the SF-77B Fast-Step Perfusion device (Warner Instruments) for 40 ms once every 30 s (unless indicated otherwise) and recorded till steady state was reached. The holding potential was 0 mV unless indicated otherwise.

PcTx1 was purchased from Peptides International (Louisville, Kentucky, USA) and solubilized in aqueous buffer as 1 mM stock. JNJ-67869386 and JNJ-799760 were synthesized in house and solubilized in DMSO as 10 mM stocks. All other chemicals were from Tocris (Minneapolis, USA).

Protein expression and purification

To form well-ordered crystals, ΔASIC1 with a N-terminal FLAG tag, a 10 × His tag and TEV protease cleavage site was expressed in Sf9 cells using the Bac-to-Bac baculovirus expression system. To prepare crude cell membranes, the cell pellet was first resuspended in a lysis buffer (20 mM KCl, 10 mM MgCl2, 10 mM HEPES, pH 7.5) containing the cOmplete™ EDTA-free protease inhibitor cocktail (1 tablet/50 mL; SigmaAldrich). After passing through a microfluidizer three times at 600 kPa, the lysate was centrifuged at 45,000 rpm (45Ti rotor) for 30 min. The pellet was then homogenized in a high salt buffer (lysis buffer containing 1 M NaCl) and centrifuged again at 45,000 rpm for 30 min. Finally, the crude membrane pellet was re-homogenized in a freezing buffer (lysis buffer containing 40% glycerol) and stored at −80 °C. For protein purification, crude membranes were first solubilized on ice for 2 h in a buffer containing 20 mM Tris (pH 7.5), 20 mM imidazole (pH 7.5), 150 mM NaCl, 10 mM MgCl2, 5 mM beta-mercaptoethanol, 2% DDM, 10% glycerol, and the cOmplete™ EDTA-free protease inhibitor cocktail. Unsolubilized elements were removed by centrifugation at 45,000 rpm for 30 min. The supernatant was incubated at 4 °C overnight with washed Talon resin. The resin was then thoroughly washed first with washing buffer (20 mM pH 7.5 Tris, 35 mM pH 7.5 imidazole, 150 mM NaCl, 10 mM MgCl2, 5 mM beta-mercaptoethanol, 0.05% DDM, 10% glycerol), and then again with washing buffer containing 100 µM JNJ-799760 (diluted from a 50 mM stock in 100% DMSO). The compound-bound ΔASIC1 protein was eluted with elution buffer (washing buffer containing 250 mM imidazole and 100 µM JNJ-799760). To remove the FLAG and His tags, TEV protease was added to the eluted protein (TEV protease:ΔASIC1 = 1:10, w–w). After incubation at 4 °C overnight, the sample was centrifuged at 20,000×g for 15 min to remove any precipitation and then applied through a Superdex 200 column in a solution containing 10 mM Tris (pH 7.5), 150 mM NaCl, 0.05% DDM, 5 mM beta-mercaptoethanol, 1 mM EDTA, and 25 µM JNJ-799760. The major peak fraction containing highly purified ΔASIC1 was collected, concentrated to 4 mg/mL using a Vivaspin® 6 concentrator (MWCO 100 kDa; Sartorius) at 2000×g, frozen in liquid nitrogen and stored at −80 °C.

Crystallization

Before crystallization, spermine and JNJ-799760 were added to protein samples to a final concentration of 10 mM and 100 µM, respectively. Crystals were obtained at 13 °C using the hanging drop vapor diffusion method. Drops were set up by mixing 1 µL protein sample and 1 µL reservoir solution (100 mM MgCl2, 100 mM HEPES, 28–30% PEG 400). Crystals were frozen and stored in liquid nitrogen for subsequent data collection.

X-ray data collection, processing, and structure determination

X-ray diffraction data were collected at 100 K and 1.0 Å wavelength on LS-CAT beamline 21-ID-D with a MAR 300 CCD detector (Argonne National Laboratory, Illinois, USA). Data frames were indexed and diffraction spots were integrated and scaled using the HKL-2000 program package39. The structure of the ΔASIC1/JNJ-799760 complex was determined by molecular replacement using 3S3W as a search model. Models were built with iterative rounds of manual model building in Coot40 and refined in REFMAC41 from the CCP4 program suite42 and Phenix.refine in the Phenix software43 until satisfactory statistics were achieved. The diffraction data processing and structure refinement statistics are summarized in Table 1. The crystal structure displays Ramachandran statistics with 95.78% of residues in the most favored regions and 4.06% of residues in the allowed regions of the Ramachandran diagram. The final refinement statistics and geometry of the crystal structure are shown in Table 1.

Modeling of JNJ-67869386 binding to ΔASIC1

We developed a binding model for JNJ-67869386 based on the crystal structure of ΔASIC1/JNJ-799760 and the hypothesis that both molecules occupy the same site. JNJ-67869386 was subject to automated docking using Schrödinger’s Glide module (v2017-1)44,45,46,47. The binding site grid was generated with default parameters, centered on the ligand in chain B. JNJ-67869386 was prepared with LigPrep and docked using Glide SP, all with default parameters.

Kinetic modeling

Kinetic simulations were performed using ChanneLab (Synaptosoft, Decatur, GA), with the following assumptions: (1) subunits are independent/non-cooperative, (2) channels with at least one desensitized subunit do not conduct current33 ; (3) partially and fully compound-bound open states are activated with the same probability and conductance; and (4) kinetics of compound binding/unbinding are the same at pH 8.2 and pH 7.1. Kinetics were simulated with the rate constants δ, θ, α, and β adjusted to produce the best visual global (simultaneous) fits to the experimental data at holding pHs of 8.2 (compound binding/unbinding and recovery from desensitization) and 7.1 (onset of closed-state desensitization). Since the channel opening/closing kinetics are much faster than that of compound binding/unbinding, the fractional current amplitude (simulating the experimental current response elicited by a pH 6.0 test pulse) was calculated as follows: I(t) = γO × QO × PC(t) + γOB1 × QOB1 × PCB1(t) + γOB2 × QOB2 × PCB2(t) + γOB3 × QOB3 × PCB3(t), where t denotes time; the subscripts denote open (O, OB1, OB2, and OB3) or closed (C, CB1, CB2, and CB3) states with zero, one, two, or three molecules bound; γ denotes conductance relative to the compound-unbound open state (thus, γO ≡ 1); P(t), the simulated response, represents the channel occupancy in each closed state as denoted by the subscript; and Q represents the conditional probability of channel opening (by a pH 6.0 test pulse) given that the channel is in the corresponding closed state (from which it is activated). Partly based on estimations from pH responses, the following values were used in the simulations: γOB1 = γOB2 = γOB3 = 0.77, QO = 0.7, QOB1 = QOB2 = QOB3 = 0.15.

Electrophysiology data analysis

Baseline values (i.e., current amplitudes at the conditioning pH) were subtracted to obtain responses evoked by the test pH. Responses were normalized for each cell before averaging (see figure legends for more detail on normalization for each type of experiment). Concentration–response data were fitted to a logistic function of the form: R = (A1–A2)/(1 + (C/C0)h)+A2, where R is the normalized response, C is either pH or compound concentration, C0 is the pH/concentration at which half-maximal response occurs (pH50 or IC50), h is the Hill coefficient, and A1 and A2 are constants. Kinetic parameters were obtained by fitting the data with either a single or double exponential function. Fitted data are shown as solid, dashed, or dotted curves.

Statistics and reproducibility

Statistical analyses were performed using two-tailed Student’s t-test, or one- or two-way ANOVA with post-hoc Tukey test as described in the text. Experimental results are reproducible and reported as mean ± SEM over independent measurements on n different cells. Data fitting and statistical analyses were performed using Origin (Northampton, MA, USA).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.