Glutamate acts on acid-sensing ion channels to worsen ischaemic brain injury

Ischaemic stroke remains the leading cause of death and disability in the world⁸. During stroke, ischaemia deprives brain infarct area of glucose and oxygen, elevating glutamate release and accumulation by 10–100-fold above physiological concentrations^1,9. This results in an overactivation of NMDARs, intracellular calcium overload and cell death. NMDAR-dependent excitotoxicity rationalizes NMDARs as a key target for stroke treatments². NMDAR antagonists in in vitro and in vivo models are highly neuroprotective^10,11, but have failed at different stages of clinical trials³, raising the possibility that NMDARs are not solely accountable for glutamate-induced excitotoxicity in stroke. Acid-sensing ion channels (ASICs) are regarded as a strong candidate for mediating neurotoxicity as local acidosis occurs in ischaemic areas during stroke^4,12,13,14. NMDARs and ASICs may be coincidently activated by excessive glutamate and H⁺, respectively, and these channels may cross-talk through intracellular signalling to aggravate neuronal death⁵. Glutamate was found to enhance recombinant ASIC1a currents¹⁵. However, it is unclear whether glutamate directly or indirectly interacts with native ASICs in neurons, and whether this interaction mediates and/or aggravates ischaemic injury.

To investigate whether the function of ASICs is regulated by glutamate (that is, l-glutamic acid, unless otherwise stated hereafter), we first performed patch-clamp recordings of acid-evoked currents (I_ASICs) from CHO cells expressing human ASIC1a (hASIC1a) channels. Glutamate potentiated I_ASICs evoked by extracellular fluid (ECF) with pH values ranging from 6.55 to 7.4 (that is, 0.04 to 0.28 μmol l⁻¹ proton concentrations) and left-shifted proton dose–response curve, with the half-maximum effective concentration (EC₅₀) being decreased from 189 nM to 152 nM, while the maximal currents were identical at pH values of less than 6.5 (Fig. 1a and Extended Data Fig. 1a–d). When the test pulses of pH 6.6 solution were preceded by perfusate with increasing acidity (from 7.6 to 6.8), the steady-state desensitization of I_ASICs was reduced by co-application of glutamate, producing a rightward shift of EC₅₀ from 51 to 61 nM (Fig. 1b), suggesting that glutamate reduces closed-state desensitization of ASIC1a channels. The potentiation of I_ASICs by glutamate was highly dependent on the acidity of test solutions as exemplified by EC₅₀ values, that is, 463.6 μM at pH 7.0 versus 382.3 μM at pH 6.8 (Fig. 1c). To test whether a direct interaction exists between glutamate and ASIC1a at different pH values, we used the microscale thermophoresis (MST) in vitro fluorescence binding assay, which detects ligand–receptor binding directly in cell lysates^16,17. This assay confirmed that glutamate binds to green fluorescent protein (GFP)-tagged hASIC1a in lysates of transfected HEK293T cells with a lower dissociation constant (K_d) value at pH 6.8 (113.3 μM) compared with at pH 7.0 (392.5 μM), but does not bind to GFP alone (Fig. 1d and Extended Data Fig. 1e). These results implicated glutamate as a positive allosteric modulator (PAM) for ASIC1a by direct binding. As ASICs can be homomeric ASIC1a channels or heteromeric channels with the ASIC2a or ASIC2b subunit in native central neurons¹⁸, we tested the effect of glutamate on these heteromeric channels. The potentiation by glutamate was preserved (Fig. 1e), indicating that ASIC1a is an integral component of ASICs for glutamate to exert its action. Finally, we performed a series of experiments that ruled out the possibility that the potentiation of I_ASICs by glutamate is due to unmasking of the inhibition of I_ASICs by Ca²⁺ or Zn²⁺ that is known to block ASIC1a^19,20 (Extended Data Fig. 1f–n).

a, Examples of I_ASICs evoked in an hASIC1a-transfected CHO cell in the absence and presence of 500 μM glutamate for pooled dose–response curves. n = 14. b, An example recording showing steady-state desensitization of ASIC1a currents with and without coapplied glutamate for pooled dose–response curves (n = 11) for each group. c, Dose–response curves for glutamate to potentiate I_ASICs at pH 7.0 (n = 14) and 6.8 (n = 8). d, MST imaging traces and dose–response curves showing direct binding between glutamate and GFP-tagged ASIC1a at pH 7.0 (n = 5) and 6.8 (n = 6). The vertical bars show the cold fluorescence detected at −1–0 s (blue); hot fluorescence detected at 4–5 s (red). e, Dose–response curves for ASIC1a/2a and ASIC1a/2b currents. n = 13 and 14 cells, respectively. f, Outside-out patch recordings of ASIC single-channel currents evoked by pH 7.0. g, All-points amplitude histogram of ASIC single-channel currents from f; curves were fitted by double Gaussian components. Bin = 0.05 pA. h, The P_o and amplitude of ASIC1a unitary currents evoked at pH 7.0 and 7.2. n = 7–9 patches. i, The effects of l-, dl- and d-isomers of AP5 (400 μM) on ASIC1a at pH 7.0. n = 11 cells. Three to six replicated cultures for patch recordings were tested. Data are mean ± s.e.m. Statistical analysis was performed using two-way analysis of variance (ANOVA) (h) and one-way ANOVA (i) with Tukey post hoc correction for multiple comparisons.

Source Data

Full size image

We next performed single-channel recordings in outside-out patches from ASIC1a-transfected CHO cells and found that glutamate increased the open probability (P_o) of ASIC1a at pH 7.0 and even a milder pH 7.2, without affecting the amplitude of ASIC1a unitary currents (Fig. 1f–h). The current–voltage relationship with and without glutamate showed the same slope values (conductance: 17.3 ± 0.9 pS (control) and 18.4 ± 0.6 pS (glutamate)) and reversal potential (33.15 mV (control) and 28.56 mV (glutamate)) (Extended Data Fig. 1o,p), indicating that the potentiation of I_ASICs by glutamate is solely attributed to P_o increase without affecting the single-channel conductance and ion selectivity of ASICs.

Having established the actions of glutamate on I_ASICs, we tested widely used agonists for glutamate receptors, namely NMDA, AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid), aspartic acid (Asp) and kainic acid, all of which showed similar effects, except for kainic acid (Extended Data Fig. 1q,r). We replicated these results in primary cultured cortical neurons and confirmed that glutamate potentiated I_ASICs at pH 7.0 from wild-type mice (Asic1a^+/+) but not from Asic1a-knockout mice (Asic1a^−/−), indicating that ASIC1a uniquely contains the site for glutamate interaction (Extended Data Fig. 1s). These results prompted us to investigate whether this potentiation could be affected by the classical NMDAR competitive blocker 2-amino-5-phosphonovaleric acid (AP5). Notably, we found no block of I_ASICs by three AP5 isomers (l, racemic mixture dl and d; 200 μM) (Extended Data Fig. 2a,b). Instead, the l- and dl-, but not d-isomer of AP5, enhanced I_ASICs (Fig. 1i). dl-AP5 increased the P_o of ASIC1a at both pH 7.0 and 7.2 (Extended Data Fig. 2c–f). Furthermore, an MST assay detected strong binding between l-AP5 and ASIC1a (Extended Data Fig. 2g,h). In contrast to l-glutamate and its sodium or potassium salts, d-glutamate did not potentiate I_ASICs (Extended Data Fig. 3). These results showed that the influence of AP5 and glutamate on ASIC1a depends on their specific stereochemistry, with effectiveness observed only in the l-isomer, but not the d-isomer, consistent with their optical activity and rotation properties (Supplementary Table 1). These observations suggested that glutamate and its structural analogues must share a common binding site, most likely on the extracellular domain of ASIC1a.

Excessive glutamate release and acidosis are believed to cause ischaemic neuronal death by activating NMDARs and ASICs, respectively^4,21. Our observations led us to hypothesize that glutamate may cause NMDAR-independent cell death by directly acting on ASICs to drive Ca²⁺ overload. To this end, we performed calcium imaging in cultured cortical neurons from Asic1a^+/+ and Asic1a^−/− mice in the presence of the AMPA receptor blocker NBQX, NMDAR pore blocker MK801 and voltage-gated calcium channel blocker CdCl₂ (Extended Data Fig. 4a). In Asic1a^+/+ cortical neurons, we found that application of pH 7.0 solution alone led to slow elevation of intracellular Ca²⁺ ([Ca²⁺]_i), which was robustly potentiated by co-application of glutamate, consistent with the idea that glutamate can promote Ca²⁺ overload through ASIC1a, independent of other routes of Ca²⁺ influx. This was reinforced by the same experiments with Asic1a^−/− neurons, in which glutamate could elevate [Ca²⁺]_i only slightly at pH 7.0 (Fig. 2a–c). To rule in/out other sources of elevated [Ca²⁺]_i, we blocked intracellular Ca²⁺ release via/from the metabotropic glutamate receptors (mGluRs, known to trigger intracellular release of Ca²⁺), mitochondria or endoplasmic reticulum²² with LY341495, CGP37157 and RO2959—inhibitors of the mitochondrial Na⁺/Ca²⁺ exchanger and Ca²⁺-release-activated Ca²⁺ (CRAC) channel in the endoplasmic reticulum, respectively. Under such conditions, glutamate remained capable of generating long-lasting Ca²⁺ elevation in Asic1a^+/+ neurons, but not under conditions with zero-Ca²⁺ ECF, or psalmotoxin-1 (PcTX-1; a blocker of ASICs) (Extended Data Fig. 4b,c). These results indicated that the persistent Ca²⁺ elevation must originate from direct entry through ASICs. With a cocktail of glutamate receptor antagonists (GluR-B), I_ASICs fully desensitized in response to pH 7.0 perfusate, but co-application of glutamate significantly attenuated its desensitization to sustain a steady-state inward current and enlarge charge integrals (Fig. 2d,e), which probably accounts for the cumulative Ca²⁺ overload. In parallel, co-application of glutamate caused a much large decrease in the mitochondrial membrane potential (Ψm) compared with pH 7.0 perfusate alone in Asic1a^+/+ neurons, but not in Asic1a^−/− neurons (Extended Data Fig. 4d–g). Such a decrease in Ψm is regarded as a hallmark of early events en route to neuronal death^23,24.

a–c, Examples and summary of changes in [Ca²⁺]_i levels and time-course imaging in primary cultured cortical neurons from Asic1a^+/+ (n = 18 and 18 cells from 3 cultures per group) and Asic1a^−/− (n = 16 and 18 cells from 3 cultures per group) mice with and without addition of 200 μM glutamate to the pH 7.0 solution. AUC, area under the curve. d, I_ASICs evoked at pH 7.0 from cultured cortical neurons for 30 s with and without glutamate. e, The peak amplitude and integral area of I_ASICs from d were pooled from n = 10 cells from 3 replicated cultures. 50 μM d-AP5, 10 μM NBQX and 10 μM LY341495 were applied. f, Schematic of the cell death experiment. DIV, days in vitro; P0, postnatal day 0. The diagram was created using BioRender (https://biorender.com). g, Calcein–propidium iodide (PI) staining of cultured neurons from Asic1a^+/+ and Asic1a^−/− mice under different conditions: pH 7.4 or 7.0 with or without 100 ng ml⁻¹ PcTX-1 or GluR-B; glutamate receptor blockers consisted of 50 μM d-AP5, 10 μM NBQX and 10 μM LY341495. h, Summary of the percentage of cell death calculated by calcein (live cells) and propidium iodide (dead cells) counting. n = 10–13 images from 3 replicated cultures for each group. i, LDH release from cultured neurons. n = 8 cultures for each group. j, The relative glutamate release with or without 1 h OGD. n = 5 cultures per group. k, Histology images of brain slices from mice subjected to MCAO. l, Quantification of the infarct volume after MCAO mice were injected with physiological saline and memantine (mem; 1 mg per kg). n = 3, 11, 8 and 9 mice for each group. Data are mean ± s.e.m. Statistical analysis was performed using two-tailed paired t-tests (e), and two-way (c,h,i) and one-way ANOVA (j,l) with Tukey post hoc correction.

Source Data

Full size image

To address whether glutamate causes cell death by acting on ASIC1a, we subjected primary cultured cortical neurons from Asic1a^+/+ and Asic1a^−/− mice to the oxygen–glucose deprivation (OGD) paradigm for 1 h before performing calcein–propidium iodide staining of dead cells and assaying lactate dehydrogenase (LDH) release due to neuronal injury (Fig. 2f). We found a marked increase in cell death in Asic1a^+/+ neurons at pH 7.0 compared with Asic1a^−/− neurons (Fig. 2g,h). Notably, OGD at lower pH values (that is, pH 6.0 or 6.5) did not increase cell death compared with that at pH 7.0 (Extended Data Fig. 4h–j). To examine the mechanisms underlying these counterintuitive observations, we compared the effects of increasing ECF acidity (pH 7.0, 6.5 and 6.0) on I_ASICs with or without glutamate co-application to the same cells in the presence of GluR-B. We found that glutamate significantly attenuated I_ASIC desensitization and generated a steady-state tonic current at pH 7.0, which was diminished at pH 6.0/6.5 despite large instantaneous currents being generated (Extended Data Fig. 4k,l), suggesting that the differences in tonic current underlie these unexpected outcomes of cell mortality. Compared with neurons treated at pH 7.4, slight acidification of ECF at pH 7.0 induced a significant increase in LDH release in Asic1a^+/+ neurons but not in Asic1a^−/− neurons (Fig. 2i).

Notably, addition of GluR-B augmented cell death and LDH release in Asic1a^+/+ but not Asic1a^−/− neurons at pH 7.0 (Fig. 2h,i). As none of three reagents in GluR-B potentiated I_ASICs (Extended Data Fig. 4m), we postulated that OGD might increase the extracellular concentration of endogenous glutamate in cultured neurons to potentiate I_ASIC tonic current after applying glutamate receptor blockers. Indeed, GluR-B led to around a fivefold increase in extracellular glutamate concentration after 1 h OGD (Fig. 2j). Blocking postsynaptic glutamate receptors might also cause a homeostatic upregulation of presynaptic glutamate release and postsynaptic excitability, which in turn amplifies ASIC1a-mediated tonic current to exacerbate cell death at pH 7.0 (Fig. 2h,i). By contrast, the reduced cell death at pH 6.0 (Extended Data Fig. 4h,i) may be attributed to full desensitization of ASICs and minimized tonic current, despite elevated glutamate during OGD. Blocking ASICs with PcTX-1 suppressed cell death and LDH release in Asic1a^+/+ neurons (Fig. 2h,i), indicating that the OGD paradigm favours acidosis- but not glutamate-receptor-mediated cell death. In ASIC1a-transfected CHO cells without native glutamate receptors, application of glutamate for 1 h at pH 7.0 yielded prominent cell death, which was abolished by PcTX-1 (Extended Data Fig. 4n,o). Thus, glutamate has a vital role in amplifying the level of [Ca²⁺]_i, mitochondrial dysfunction and cell injury/death even under slightly acidic conditions (for example, pH 7.0), whereas blockade of ionotropic and metabotropic glutamate receptors does the opposite to the expected neuroprotection. Instead, glutamate might work through ASIC1a to cause Ca²⁺ overload and neurotoxicity, with mild or moderate acidity being of the most lethal to neurons in the penumbra surround the ischaemic core.

To test the roles of ASICs versus NMDARs in ischaemic brain damage in vivo, we used a mouse stroke model by performing transient middle cerebral artery occlusion (MCAO) for 30 min to induce excessive glutamate release and acidosis during ischaemia and reperfusion. The infarct volume was quantified by post hoc tissue staining with 2,3,5-triphenyltetrazolium chloride (TTC) 24 h later (Extended Data Fig. 4p). MCAO reliably diminished the relative cerebral blood flow (rCBF) and cerebral infarction during surgery. We found significantly smaller infarct volumes in Asic1a^−/− mice compared with those in Asic1a^+/+ mice, while the rCBF during MCAO between the two genotype mice showed no difference (Extended Data Fig. 4q,r and Fig. 2k,l). As a positive control, we administrated memantine²⁵ (NMDAR open channel blocker, 1 mg per kg, intraperitoneally) and confirmed that it reduced the infarct volume in Asic1a^+/+ mice to a level comparable to that in Asic1a^−/− mice (Fig. 2k,l and Extended Data Fig. 4s). These data indicate that targeting ASICs can be as effective as blocking NMDARs in reducing ischaemic brain damage in the MCAO mouse model.

To predict the binding pocket on ASIC1a for glutamate and its structural analogues to function as PAMs, we performed in-depth analyses of the properties and conservation of the accessible surface area in the ASIC1a structure as well as computational docking onto the trimeric extracellular domain of the available homomeric chicken ASIC1a (cASIC1a) structure (Protein Data Bank (PDB): 5WKU)²⁶. As only a low-resolution cryo-electron structure of the human protein was available at the onset of this work²⁷, a higher-resolution structure of cASIC1a was used. The structure is expected to closely resemble that of hASIC1a, given a strong conservation between the human and chicken ASIC1a sequences (90% identity; Extended Data Fig. 5). We identified six putative glutamate-binding sites around amino acid residues Arg161, Lys379, Lys383, Gln226, Lys391 and Lys387 (Fig. 3a). Subsequent molecular docking (by HADDOCK; PDB: 5WKU) was performed to examine the plausibility of the proposed binding sites, compute the binding energetics and rank the sites (Supplementary Table 2). These putative sites were mapped back onto the hASIC1a sequence (that is, Arg160, Lys380, Lys384, Gln225, Lys392 and Lys388).

a, A top view of cASIC1a (PBD: 5WKU). Top six scoring binding residues of glutamate by initial docking calculation are shown. For clarity, only chain A is shown here. b, I_ASICs activated in a human ASIC1a(K380A)-transfected CHO cell in the presence and absence of 500 μM glutamate. Dose–response curves (solid lines) were contrasted to those from wild-type ASIC1a channels as in Fig. 1a (dashed lines). n = 12 cells for ASIC1a(K380A). c, Dose–response curves for hASIC1a(K380A)/hASIC2a and hASIC1a(K380A)/hASIC2b currents. n = 10 cells for each group. d, Top view of glutamate-bound cASIC1a. A magnified view of the glutamate-binding pocket is shown at the bottom right. The chemical structure of glutamate is shown at the top right. The surface of cASIC1a and glutamate are coloured white and cyan, respectively. e, The putative glutamate-binding pocket near the outer vestibule of the channel pore. f, MST assay showing the dose–response curve for glutamate binding with wild-type ASIC1a (data from Fig. 1d) or K380A mutant at pH 7.0. n = 5 replicated tests per group. Three to six replicated cultures for patch recordings were tested. Data are mean ± s.e.m.

Source Data

Full size image

Site-directed mutagenesis was then performed to individually replace each of these six plausible interface residues with alanine or leucine in hASIC1a before these mutants were expressed in CHO cells for patch-clamp experiments. We found that currents mediated by hASIC1a(K380A) mutant or mASIC1a(K378A) mutant displayed diminished sensitivity to glutamate at pH 6.55–7.0 (Fig. 3b and Extended Data Fig. 6a). By contrast, no changes in the magnitude of potentiation were observed in the hASIC1a K384A, Q225L and K388A mutants as compared to the wild type. Glutamate lost its PAM effect on the hASIC1a K392A and R160A mutants (Extended Data Fig. 6b,c). The three-dimensional structural proximity of Lys380 and Lys392 rationalized these two residues orientated towards the same binding pocket. The Arg160 mutant was not pursued further because it generated very little currents with markedly altered kinetics, making it difficult to ascertain whether this mutation perturbed protein structural stability, glutamate binding, and/or proton binding or channel gating. Finally, potentiation of I_ASICs by glutamate in CHO cells co-expressing the hASIC1a(K380A) mutant subunit and ASIC2a or 2b subunit was significantly reduced (Fig. 3c). These data indicated that Lys380 in hASIC1a is necessary for glutamate binding, suggesting that it is a key residue within the glutamate-binding pocket that enables its role as a PAM for hASIC1a. The K380A mutant probably did not affect the stability of the protein as its current was comparable to that of the wild type. Predictions of the protein stability after K379A mutation using the program MAESTRO suggested no significant effect on either trimer or monomer stability²⁸ (Supplementary Table 3).

To gain molecular insights into glutamate interactions within the proposed binding pocket around Lys380 of hASIC1a using an independent method, we performed further molecular docking by placing glutamate near the chain A of cASIC1a Lys379 at the outer vestibule of the ion permeation pathway (Fig. 3d). In Fig. 3e, the optimal glutamate docking pose to the putative binding site is shown with optimized side-chain positions of five interacting residues in the vicinity. Glutamate appears to form four specific hydrogen bonds to Gln227, Pro381, Lys383 and Gln398, two salt bridges to Lys379 and Lys383 (Fig. 3e), giving a very stable glutamate–cASIC1a complex. The putative binding pocket for glutamate is a solvent-exposed and electrostatically positive cavity (Supplementary Table 4), which makes it similar to the proton-binding pocket on ASIC1²⁶ that resides in its close proximity. Notably, the glutamate pocket forms at subunit interfaces stabilized by hydrophobic and polar contacts across the finger, thumb and palm domains. We postulated that, after glutamate binding, the conformation of the thumb domain may undergo rearrangements, which could be transduced to the channel pore through the palm domain to enhance ion permeation.

To determine whether Lys379 of cASIC1a (or Lys380 of hASIC1a) could represent the core binding site for glutamate, we analysed the molecular mechanics/generalized Born surface area (MM/GBSA) binding free energy and docking scores²⁹ of the glutamate–cASIC1a(K379) and glutamate–cASIC1a(K379A) complex models. We found that both parameters for the wild type were significantly more negative for the wild type compared with for the cASIC1a(K379A)–glutamate complex (Extended Data Fig. 7a), which suggests a destabilization of the complex by Lys379 mutation. Given that glutamate is a natural agonist for NMDARs, we constructed a glutamate–NMDAR complex model to identify any differences in comparison to our glutamate–cASIC1a complex model (Extended Data Fig. 7b). Although no difference was found in the docking scores between glutamate–NMDAR and glutamate–cASIC1a(K379) (wild type) complexes, the MM/GBSA of the glutamate–NMDAR complex was higher than that of glutamate–cASIC1a(K379) complex (Extended Data Fig. 7c), suggestive of higher glutamate-binding affinity to NMDARs than to ASIC1a. Molecular dynamics simulations of the glutamate–cASIC1a(K379) (wild type) and glutamate–cASIC1a(K379A) mutant complexes provide additional support for Lys379 of cASIC1a (that is, homologous site Lys380 of hASIC1a) being important for glutamate binding to ASIC1a channels (Extended Data Fig. 7d–j).

The binding of glutamate to the ASIC1a channel was validated by MST assays in which increasing concentrations of glutamate induced dose-dependent changes in the fluorescence of ASIC1a–GFP in the absence or presence of amiloride (100 μM, an open channel blocker of ASIC1a, non-competitive for glutamate binding) at pH 7.0 (Fig. 3f and Extended Data Fig. 7k,l). By contrast, weaker and unstable binding was observed between glutamate and hASIC1a(K380A) mutant while no conspicuous binding was detectable between glutamate and ASIC1a–GFP at pH 7.4 (Fig. 3f and Extended Data Fig. 7m). Thus, glutamate directly binds to a bona fide cavity at Lys380 of the ASIC1a channel.

Our structural findings raised the possibility of developing therapeutics to alleviate neuronal injury by targeting the glutamate binding site on ASIC1a. We tested the effects of a series of candidate chemicals on glutamate-mediated potentiation of I_ASICs (Extended Data Fig. 8). Among these, l-aminoadipic acid (l-AA) did not block I_ASICs, but markedly abolished the potentiation of I_ASICs by glutamate (Extended Data Fig. 8g–i). However, l-AA may exert an agonist activity for NMDAR³⁰ and mGluRs³¹, making it unsuitable for ischaemic stroke therapy. CGS19755, a rigid analogue of AP5 with water solubility and blood–brain barrier permeability³², significantly attenuated glutamate-induced potentiation of I_ASICs in CHO cells and culture neurons without affecting I_ASICs itself (Fig. 4a–c and Extended Data Fig. 9a). The effect of CGS19755 was dose dependent with a half-maximal inhibition concentration (IC₅₀) of 7.7 ± 2.6 μM (Fig. 4d). Computational docking showed that the affinity of CGS19755 binding to the cASIC1a(K379) pocket was much higher than that of glutamate (best pose of docking scores: −6.05 kcal mol⁻¹ (CGS19755 to cASIC1a) and −3.605 kcal mol⁻¹ (glutamate to ASIC1a)) (Extended Data Fig. 9b and Supplementary Table 4). An MST assay revealed strong binding between CGS19755 and ASIC1a (K_d = 1.1 ± 0.2 μM) (Fig. 4e). These data suggest that CGS19755 might be a potent competitive blocker of the glutamate-binding site on both ASICs and NMDARs.

a, Representative traces showing the effects of glutamate and CGS19755 on I_ASICs from an ASIC1a-transfected CHO cell. Inset: the chemical structure of CGS19755. b,c, Summary plots of CGS19755 on glutamate-induced potentiation of I_ASICs from ASIC1a-transfected CHO cells (b; n = 24 cells) and cultured neurons (c; n = 12 cells). d, Dose–response curve showing that CGS19755 blocked the potentiation of ASIC1a currents by glutamate. n = 15 cells. e, MST assay showing the dose–response curve for binding between CGS19755 and ASIC1a at pH 7.0. n = 5 replicates. f, The chemical structures of CGS19755, LK-1 and LK-2. Green coded circles represent the scaffold for virtual screening. g, LK-1- and LK-2-bound pockets. h, Dose–response curves showing that glutamate-induced potentiation of I_ASICs was inhibited by LK-1 (n = 14 cells) and LK-2 (n = 12 cells) at pH 7.0. i, Dose–response curves showing that LK-1 (n = 9 and 17 cells for GluN2A and GluN2B) and LK-2 (n = 15 and 21 cells for GluN2A and GluN2B) inhibited NMDAR currents evoked by 100 μM NMDA and 10 μM glycine. NR1 (GluN1) plus NR2A (GluN2A) or NR2B (GluN2B) subunits were co-expressed in CHO cells. j, Representative MST traces and dose–response curve showing direct binding between LK-2 and GFP-tagged ASIC1a at pH 7.0. The vertical bars show cold fluorescence detected at around −1 to 0 s (blue) and hot fluorescence detected at 4–5 s (red). n = 6 replicated tests. Three to six replicated cultures for patch recordings were tested. Data are mean ± s.e.m. Statistical analysis was performed using one-way ANOVA with Tukey post hoc correction (b,c).

Source Data

Full size image

Our experiments in vitro validated the efficacy of CGS19755 in neuroprotection (Extended Data Fig. 9c–i), raising the hope for drug candidates that selectively target the glutamate-binding site on ASICs but not on NMDARs³². This is important because NMDARs are instrumental for brain function under physiological conditions and are even pro-survival during stroke recovery³³, and the inhibition of NMDAR by its high-efficacy antagonists such as CGS19755 (also known as selfotel) may lead to unexpected side effects, such as psychosis shown in previously terminated clinical trials^34,35. To search for small molecules that could largely block glutamate-dependent enhancement of ASIC activity with minimal effects on NMDAR activity, we performed a CGS19755 structure-based computational drug screen. Two candidate compounds, LK-1 and LK-2, were identified as prototypes of a new class of neuroprotective small molecules (Fig. 4f,g and Extended Data Fig. 10a) on the basis of their molecular docking scores (best pose of docking scores: −6.371 kcal mol⁻¹ (LK-1 to ASIC1a), −4.32 kcal mol⁻¹ (LK-1 to NMDAR), −7.039 kcal mol⁻¹ (LK-2 to ASIC1a) and −4.686 kcal mol⁻¹ (LK-2 to NMDAR)). To verify that LK-1 and LK-2 are pharmacologically effective, we recorded I_ASICs by co-applying glutamate and either compound, and found that both compounds reduced glutamate-dependent potentiation of I_ASICs in a dose-dependent manner without affecting the basal currents of ASIC1a themselves (Fig. 4h and Extended Data Fig. 10b–d). Importantly, LK-1 and LK-2 were much less effective in attenuating NMDAR currents in acute isolated cortical neurons and CHO cells that express recombinant NMDARs (Fig. 4i and Extended Data Fig. 10e). NMDARs comprising NR1 (GluN1) plus NR2A (GluN2A) subunits were more sensitive to LK-1 and LK-2 compared with those comprising NR1 plus NR2B (GluN2B) subunits (Fig. 4i). Moreover, LK-2 was found to be a weaker antagonist for NMDARs compared with LK-1 (Fig. 4i), with an IC₅₀ two orders of magnitude higher, implicating its potential to preferentially block the glutamate-binding site on ASICs over that on NMDARs. Differential pharmacological effects of LK-1 and LK-2 on ASIC1a versus NMDARs were consistent with in silico docking results showing higher affinity of LK-1 and LK-2 for ASIC1a than NMDAR, but LK-2 is more likely to spare NMDARs than LK-1. An MST assay showed that LK-2 could directly bind to ASIC1a with a K_d of 1.9 ± 0.4 μM (Fig. 4j). These results rationalized LK-2 as a potent competitive blocker that preferentially targets the glutamate-binding site on ASIC1a channels over that on NMDARs.

To test whether small-molecule LK-2 could provide neuroprotection, we performed cell death and LDH-release assays at pH 7.0 after 1 h OGD in vitro, and found that LK-2 yielded protective effects comparable to CGS19755. This implicated glutamate-mediated enhancement of ASIC activity rather than NMDAR overactivation as the main contributor to cell injury under ischaemic conditions (Fig. 5a–c). LK-2 can protect ASIC1a-transfected CHO cells after prolonged (1 h) exposure to glutamate at pH 7.0, further demonstrating its efficacy to alleviate cell injury (Extended Data Fig. 10f,g).

a, Calcein–propidium iodide staining of cultured neurons with different treatments at pH 7.0 after 1 h OGD with 10 μM CGS19755 or 10 μM LK-2. b, The percentage of cell death from the experiment in a was calculated by counting the number of calcein-stained (live cells) and propidium iodide-stained (dead cells) stained cells. n = 10–13 images from three replicated cultures for each group. c, LDH release of cultured neurons with different treatments after 1 h OGD. n = 8 cultures for each group. d,e, Images of brain slices (d) and quantification of the infarct volume (e) after MCAO in mice administrated with CGS19755 (1 mg per kg, intraperitoneally (i.p.)), PcTX-1 (100 ng per kg, i.n.) and LK-2 (10 mg per kg and 30 mg per kg, i.p.). n = 8, 12, 14, 6, 10, 5 and 11 mice for each group. f, The experimental timeline of the MCAO mouse modelling and behaviour tests. One dose of 30 mg per kg LK-2 was applied. The diagram was created using BioRender. g, Motor learning performance was assessed by plotting the time for mice to fall off a rotarod over 5 trials 24 h and 7 days after MCAO in the sham-, saline- and LK-2-treated groups. h, Summary of the time for mice to fall off the rotarod was assessed by 1 test trial 24 h and 7 days after MCAO. For the behaviour tests, n = 11, 18 and 19 (each group 24 h after MCAO), and n = 11, 12 and 12 (each group 7 days after MCAO) mice. Statistical analysis was performed using one-way ANOVA (e) and two-way ANOVA (b,c,g,h) with Tukey post hoc correction. NS, not significant.

Source Data

Full size image

To directly investigate the effects of LK-2 in a mouse stroke model in vivo, we first tested whether LK-2 could pass through the blood–brain barrier by measuring its pharmacokinetics in mice. Injection of LK-2 at a dose of 30 mg per kg (intraperitoneally) led to a maximal concentration of 92,660 μg l⁻¹ in the plasma and 236 μg l⁻¹ in the brain tissue, demonstrating that LK-2 could penetrate the blood–brain barrier and potentially reach a micromolar concentration in cerebrospinal fluid to block the glutamate-binding site on ASICs, if considering a dilution factor of around 50 for fluid extracted from the wet brain tissue³⁶ and a local increase in blood–brain barrier permeability around the infarct (Supplementary Table 5). Moreover, LK-2 had a terminal elimination half-life (t_1/2) of 2.297 h, presuming its neuroprotective effects within the critical window of intervention after the onset of stroke (Supplementary Table 5 and Extended Data Fig. 10h). Thus, LK-2 displays favourable properties for neuroprotection against ischaemic brain injury in vivo.

Using the mouse MCAO model, we assessed the infarct volume with and without injection of LK-2 at the time of reperfusion (that is, 30 min after artery occlusion). We observed a significant reduction in brain damage by LK-2 at 30 mg per kg (intraperitoneally) compared with in the saline group in Asic1a^+/+ mice (Fig. 5d,e). And the protective effect provided by LK-2 (30 mg per kg) for brain damage was comparable to that provided by CGS19755 (1 mg per kg, intraperitoneally) (P = 0.8288). Asic1a^−/− mice were resistant to brain injury by the MCAO paradigm and further administration of CGS19755 (1 mg per kg, intraperitoneally) added little to further neuroprotection (Fig. 5e and Supplementary Table 6), reinforcing the indispensable role of glutamate-dependent potentiation of ASICs in mediating ischaemic brain injury.

As the MCAO model mainly results in infarction of the ipsilateral frontal cortex (including the motor cortex) and part of the striatum in the middle cerebral artery territory³⁷, on which sensorimotor behaviours depend, we examined mice using the rotarod task, a commonly used rodent motor behaviour paradigm³⁸. The time for mice to fall off the rotating rod was regarded as the readout, 24 h and 7 days after MCAO, to evaluate the short-term and long-term effects of LK-2 on motor learning (a part of cognitive function) and movement coordination ability (Fig. 5f). We found that LK-2-treated mice showed only marginal improvement in their latency to stay on the rotarod 24 h after surgery. But, by day 7 after MCAO, they displayed substantially superior motor learning and coordination ability compared with the saline-treated mice, as reflected by the steep slope of motor learning plots over five test trials with their final latency comparable to that of the sham group (Fig. 5g,h). No differences in improvement of motor behaviour were found between LK-2-treated male and female mice (Extended Data Fig. 10i), indicating its sex-independent efficacy for facilitating long-term recovery. We conclude that LK-2 functions as a competitive antagonist for the glutamate-binding site on ASICs and provides neuroprotection independent of its action on NMDARs, ultimately improving neurobehavioural outcomes after stroke.

Here we demonstrate that glutamate and its structural analogues, including agonists for NMDARs and even competitive NMDAR antagonist l-AP5 function as PAMs to potentiate currents mediated by ASICs. NMDAR antagonists including CGS19755, which are effective in animal models of stroke, have failed in clinical trials due to severe side effects, including neuropsychiatric and tolerance issues^34,39,40. NMDAR antagonists not only block pro-survival functions of NMDARs during stroke recovery but also induce severe psychosis⁴¹, making NMDARs difficult to target. Previous studies showed that TRPM2, TRPM4 and ASIC1a channels directly interact with NMDARs^5,6,7. The mapping of their respective interfaces provided means to disrupt NMDAR signalling without direct perturbations to the multifaceted functions of NMDARs in synaptic transmission and plasticity that are critical for cognitive functions. These studies all converged onto decoupling NMDARs from cell-death signalling. Our study showed that glutamate at clinically relevant concentrations potentiated I_ASICs and largely mediated neurotoxicity in vitro and in vivo. Indeed, ASIC1a is responsible for acidosis-mediated ischaemic brain injury and cell death of ischaemic core at a low pH 6.5–6.0 (refs. ^4,42,43). The potentiation of ASIC1a currents by glutamate at a mild pH range (for example, 7.0–6.5) sensitizes neurons to acidosis relevant to clinical conditions of ischaemic cell injury and death in penumbra. Synergistic actions of protons and glutamate on ASICs at a pH range of around 7.0 may maximally boost Ca²⁺ influx, mitochondrial dysregulation, membrane excitability and downstream signalling, driving cell death to expand infarct volume after ischaemic insult and reperfusion. Previous research showed that the activation of NMDAR potentiates ASIC1a currents through intracellular signalling cascades (that is, after 5 min OGD), indicating that acidosis neurotoxicity is secondary to NMDAR activation⁵. Our study argues that glutamate and protons, co-released from synaptic vesicles⁴⁴, can act as the first messengers for both NMDARs and ASICs and their downstream cell-death signalling cascade at the very early onset of ischaemic insult. Neuroprotective effects of LK-2 against ischaemic injury favour the working model that glutamate binding to ASIC1a may be involved after the onset of ischaemia, rather than being secondary. Aside from this pathological link to neurotoxicity, our findings implicate physiological roles of glutamate binding to ASIC1a in synaptic transmission and plasticity underlying learning and memory and other behaviours^45,46,47,48.

We demonstrated that glutamate binds directly to a cavity near the outer vestibule of the ASIC1a channel pore, with Lys380 being a critical amino acid residue for glutamate binding. Potent neuroprotection by the classical competitive NMDAR antagonist CGS19755 is probably accounted for by its dual bindings to this cavity on ASICs and that on NMDARs. LK-2 stemmed from the structure of CGS19755 probably spares NMDARs if properly dosed. In contrast to classical ASIC blockers with poor selectivity or blood–brain barrier permeability in drug delivery^4,49, LK-2 at low micromolar concentrations preferentially decouples glutamate from ASICs, showing a promising neuroprotective efficacy and neurobehavioural recovery after stroke. Together, these findings build a conceptual paradigm for mechanistic understanding of neurotoxicity in ischaemic stroke and for strategizing new stroke therapeutics independent of NMDARs.

Animals

Wild-type C57BL/6J mice were purchased from Vital River company (China) and housed in groups, in the animal facility at Shanghai Sixth People’s Hospital, Shanghai Jiaotong University. Asic1a^−/− (with congenic C57BL/6J background) mice were provided by the T.-L. Xu laboratory. All mice were kept in standard cages (15 cm × 21 cm × 13.5 cm) under a 12 h–12 h light–dark cycle with ad libitum access to food, water and nesting material. The mice were checked daily and their weight was monitored during the experiments. Mice were randomly allocated to treatment groups.

Ethical approval

Throughout the study, all efforts were carried out to minimize animal suffering and reduce the number of animals. Experiments were conducted in conformity with the institutional guidelines for the care and use of animals, and experimental protocols that were approved by the Ethics Committee of Shanghai Sixth People’s Hospital.

Electrophysiology analysis of cultures

Whole-cell recordings

Whole-cell patch clamp recordings were taken from either neurons or CHO cells in a recording chamber (Corning) mounted onto a fixed-stage upright microscope (Nikon). Patch electrodes (4–6 MΩ) were made from 1.5 mm borosilicate glass (World Precision Instruments). Whole-cell currents were recorded using an EPC-10 patch-clamp amplifier (HEKA). Data were acquired at 10–20 kHz and filtered at 1–3 kHz using a computer equipped with the Pulse 6.0 software (HEKA, Lambrecht). Cells were recorded at a holding potential of −60 mV unless otherwise described. For NMDAR current recordings, the holding potential was −70 mV. A multi-barrel perfusion system (SF-77B, Warner Instruments) and pressure regulator system (ALA-VM8, Scientific Instrument) were used to achieve a rapid exchange of extracellular solutions. The perfusion protocol was set and controlled by the PatchMaster software. To avoid use-dependent desensitization or rundown, ASICs were repeatedly activated by acidic solution at minimal interpulse intervals of >1 min. During each experiment, a voltage step of −10 mV from the holding potential was applied periodically to monitor the cell capacitance and access resistance.

Dose–response curves were fitted to the Hill equation: \(a=I/{I}_{\max }\,=\)\(1/[1+{10}^{n(b-{{\rm{EC}}}_{50})}]\), where a is the normalized amplitude of the I_ASICs, b is the concentration of proton in external solution ([H⁺]), EC₅₀ is the proton concentration or [H⁺] yielding half of the maximal peak current amplitude and n is the Hill coefficient. The IC₅₀ values for blocker dose–response curves were fitted using the following equation: \(I/{I}_{\max }=1/\left\{1+{\left[{{\rm{IC}}}_{50}/({\rm{blocker\; concentration}})\right]}^{n}\right\}\), where n is the Hill coefficient and IC₅₀ is the concentration of blocker producing 50% of the maximal block (I_max).

Single-channel recordings

Unitary currents were recorded using the outside-out configuration of the patch-clamp technique in ASIC1a transfected CHO cells. Channels were activated by rapidly moving squared-glass tubes delivering solutions of desired pH in front of the tip of the patch pipette. The delivery device achieves complete solution changes within 20 ms (SF-77B). When filled with solutions, pipettes had resistances of 6–10 MΩ. Single-channel currents were recorded using the EPC-10 patch-clamp amplifier (HEKA). The data were collected at 10 kHz and gain value 200 mV pA⁻¹, filtered at 1 kHz, and stored on a computer for analysis. Data were filtered off-line with a digital Gaussian filter to 1 kHz. All points amplitude histograms of ASIC unitary currents are fitted to the two-exponential equation:

where a is the count of ASIC unitary currents, b is the amplitude of ASIC unitary currents. A1 and A2 are the heights of the centre of the distribution, and M1 and M2 are the amplitude of ASIC unitary currents at the centre of the two distributions. SD1 and SD2 are measures of the widths of the distributions, in the same units as b (pA).

Solutions and chemicals

The artificial cerebrospinal fluid (aCSF) contained 124 mM NaCl, 5 mM KCl, 1.2 mM KH₂PO₄, 1.3 mM MgCl₂, 2.4 mM CaCl₂, 24 mM NaHCO₃ and 10 mM glucose, with pH 7.4 (300–330 mOsm). The ECF for culture cells or neurons contained 140 mM NaCl, 5 mM KCl, 1 mM CaCl₂, 1 mM MgCl₂, 10 mM glucose and 25 mM HEPES. For solutions with pH ≤ 6.8, MES was used instead of HEPES for stronger pH buffering. For NMDAR current recordings, MgCl₂ was removed. The intracellular solution for voltage-clamp recordings contained 130 mM K-gluconate, 2 mM MgCl₂, 1 mM CaCl₂, 10 mM HEPES, 0.5 mM EGTA, 4 mM Mg-ATP, with pH 7.3, and the osmolality was adjusted to 290–300 mOsm. For isolated neuron and NMDAR current recordings, K-gluconate was replaced by CsCl. All experiments were performed at room temperature (21–26 °C).

d-AP5, CGS19755, amiloride and PcTX-1 were purchased from Alomone Labs; l-AP5 and CGP39551 were purchased from Tocris; NBQX, LY341495, CGP37157 and RO2959 were purchased from MedChemExpress; and other chemicals were purchased from Sigma-Aldrich or Macklin. Note that the pH of all ECF in this study with any chemicals added was adjusted to the indicated values (the pH changes less than 0.02 after adding chemicals).

Cell culture and transfection

Chinese hamster ovary (CHO) K1 and HEK293T cells (provided by Stem Cell Bank, Chinese Academy of Sciences) were cultured in F12 and DMEM medium with 10% FBS (Gibco), respectively. Penicillin–streptomycin (Invitrogen) was added to the medium for preventing bacteria contamination at a final concentration of 1%. For NMDAR expression, 500 μM d-AP5 was added in the cultured medium to avoid glutamate-induced toxicity through activating NMDAR⁵⁰. 0.25% Trypsin-EDTA (Gibco) was used for cell passage. Cells were incubated at 37 °C in a humidified CO₂ incubator. Cells were transfected with plasmids described in detail previously⁵¹. In brief, at a cell density of 50–70%, a total 3.0 μg cDNA mixed with the Lipofectamine 3000 transfection kit (Invitrogen) was added to 35 mm dishes. For co-expression of ASIC1a plus 2a or 2b, equal amounts of cDNA for both subunits were used. For NMDAR expression, plasmids of GluN1, GluNR2A or GluNR2B and PSD-95 at a ratio of 1:4:0.5 were used. cDNA of green fluorescent protein (GFP) was linked at the N terminus of those proteins. All experiments were performed 24–48 h after transfection, all dishes were washed three times using ECF before experiments, and GFP-positive CHO cells were viewed under a fluorescence microscope for patch-clamp recordings. Human and mouse ASIC subunits were used for constructing all wild types and mutants in this study.

Primary cortical neuron culture

Primary cortical neurons were prepared and maintained as previously described⁵². In brief, cerebral cortices from 24 h postnatal Asic1a^+/+ or Asic1a^−/− mice were dissected in DMEM high-glucose solution and dissociated by 0.05% trypsin for 15 min. Cells were plated (~2 × 10⁵ cells per 35 mm dish for electrophysiology and cell death experiments) on Matrigel-coated (Corning) cover glasses or dishes. Cultures were maintained in Neurobasal-A medium (Gibco) containing 2% B27 (Gibco) and 1% GlutaMax supplements (Gibco) at 37 °C under a 5% CO₂ humidified atmosphere.

Acute isolation of cortical neurons

Acute dissociation of mouse cortical neurons was performed as described previously⁵³. In brief, mice from postnatal day 15 to 18 were anaesthetized with isoflurane. Cortical tissues were dissected and incubated in oxygenated ice-cold aCSF (with half concentration of CaCl₂). Transverse cortical slices (500 µm) were cut with a microtome (Leica VT1200) followed by incubation in aCSF containing 3.5 mg ml⁻¹ papain (Sigma-Aldrich) at 37 °C for 30 min. The slices were then washed three times and incubated in enzyme-free ECF solution for at least 15 min before mechanical dissociation. For dissociation, slices were triturated using a series of fire-polished Pasteur pipettes with decreasing tip diameters. Recording began 15 min after the mechanical dissociation. Only the neurons that retained their pyramidal shape and dendrites were used for recordings.

Site-directed mutagenesis

All ASIC1a mutants were obtained through PCR mutagenesis using wild-type Asic1a gene, inserted into the pEGFPC3 vector (Clontech) using the Seamless Cloning kit (Sunbio). For single-site mutations, the PCR (50 μl) reaction contained 1 μl DNA template, 2 μl primer pair, 4 μl dNTPs and 0.5 μl of high-fidelity DNA polymerase (PrimeSTAR, Takara). The PCR cycles were initiated at 98 °C for 3 min, followed by 30 amplifications (98 °C for 10 s, 55 °C for 15 s and 72 °C for 1 min). The last step was performed at 72 °C for 10 min. Each mutant cDNA was transformed into Escherichia coli chemically competent cells. All mutants were sequenced by PCR sequence analysis to confirm the presence of the desired mutation and the absence of undesired mutations.

Ca²⁺ imaging

Primary cortical neurons were incubated in a confocal microscopy (Zeiss 710) specialized dish (35 × 35 mm, Cellvis) with ECF (saturated with 95% O₂) and 1 μM fluo-3 AM (Beyotime) for 20 min at room temperature, followed by three washes and additional incubation in normal ECF for 15 min. The dish was then transferred onto the stage of the confocal microscope. Neurons were illuminated using a xenon lamp and observed with a ×40 UV fluor objective lens. The shutter and filter wheel were set to allow for 488 nm excitation wavelength. Images were analysed every 5 s in circumscribed regions of cells in the field of view. Digital images were acquired, stored and analysed by the ZEN software (Zeiss). For quantification, intracellular calcium levels were plotted as ΔF/F₀ ratios over time, where F₀ is the initial fluorescence intensity of each cell.

Imaging of mitochondrial membrane potential

The mitochondrial membrane potential (Ψm) was measured using JC-1 (Beyotime). Primary cortical neurons were loaded with JC-1 working solution in for 20 min, then washed and incubated in JC-1 wash buffer for 5 min before recording. JC-1 was imaged with 488 nm and 546 nm excitation wavelengths using a 40× objective. For quantification, the Ψm was measured as the fluorescence intensity ratio of red and green (R/G), then the ratio was normalized to the initial fluorescence intensity ratio of each cell.

OGD analysis

Cultured neurons were washed three times and incubated with glucose-free ECF (glucose was replaced by an equal concentration of sucrose) at pH 7.4 or 7.0 in an anaerobic chamber with an atmosphere of 95% N₂ and 5% CO₂ at 35 °C (controlled by an OxyCycler oxygen profile, BioSpherix). OGD was terminated after 1 h by replacing the glucose-free ECF with normal ECF.

Calcein–propidium iodide cell death assay

For cultured neurons, the calcein–propidium iodide cell death assay was performed after 1 h OGD and neurons were exposed to 1 μM calcein-AM and 2 μM propidium iodide (Solarbio) for 15 min (ECF was saturated with 95% O_{2 and} 5% CO₂). For ASIC1a–GFP-transfected CHO cells, the assay was performed after 1 h different treatments and the cells were exposed only to 2 μM propidium iodide for 15 min. All cells were then washed three times with ECF. After fixation in 4% paraformaldehyde (Solarbio) for 30 min, images were acquired shortly after staining by confocal microscopy. The live cells and apoptotic nuclei were determined by 488 nm and 546 nm excitation wavelengths using microscopy examination at ×40 magnification. The number of staining nuclei was counted using the Image J software. The percentage of cell death was calculated as: cd% = [n_red/(n_red + n_green)] × 100, where cd% is the percentage of cell death, n_red and n_green are the number of PI-stained (dead cells) and calcein-stained (live cells) cells or GFP-labelled cells, respectively.

Cell injury assay with LDH measurement

LDH release for cultured neurons was measured after 1 h OGD using the LDH assay kit (Beyotime). After incubation, culture medium (60 μl) was transferred to 96-well plates and mixed with 30 μl reaction solution provided by the kit for 30 min (37 °C). The optical density was measured at 490 nm and the background absorbance was subtracted at 620 nm, using a microplate photometer (Multiskan MK3, Thermo Fisher Scientific). The maximal releasable LDH in each group was obtained by incubation with 1% Triton X-100 for 30 min.

Glutamate-release assay

Glutamate production was measured using the Glutamate Assay Kit (Sigma-Aldrich, MAK004). In brief, the supernatants of cultured neurons were deproteinized with a 10 kDa MWCO spin filter before addition to the reaction. Glutamate standard was prepared by diluting 0.1 M glutamate (provided in the kit) to a different concentration gradient. The glutamate standard and each sample were added into a 96-well plate (50 μl per well), and 100 μl reaction solution was added to each well and mixed thoroughly. The mixture was incubated at 37 °C for 60 min. The absorbance of each well was measured on a microplate reader using 450 nm as the primary wavelength. The glutamate concentration of each sample was calculated by glutamate standard curve.

MCAO

Transient focal ischaemia was induced by suture occlusion of the MCAO in Asic1a^+/+ and Asic1a^−/− mice (the same number of male and female mice beyond 6 weeks was used). Animals were anaesthetized using an animal mini anaesthesia machine (RWD). During anaesthesia induction, the mice were put into a chamber (15 cm × 25 cm × 30 cm) with 2% isoflurane; the mice were put on a mask with 1% isoflurane during MCAO operation. Adequate ischaemia was confirmed by continuous laser Doppler flowmetry (moor FLPI-2). Mice that did not have a significant decrease in blood flow to less than 50% baseline values during MCAO were excluded. Rectal and temporalis muscle temperature was maintained at 33 ± 0.5 °C with a thermostatically controlled heating pad (Supplementary Table 6). Intraperitoneal injection was performed immediately after removing the suture occlusion. Mice were killed with isoflurane overdose 24 h after ischaemia.

Brains were removed and dissected coronally at 1 mm intervals, and stained with the vital dye TTC and the normal area was stained with TTC. Infarct volume (%) was calculated by summing infarction areas of all sections and multiplying by slice thickness, then dividing the whole volume of the brain. Manipulations and analyses were performed by individuals blinded to treatment groups. Depending on the experimental design, 30 min MCAO was performed for moderate ischaemic model.

Laser speckle imaging

Mice were anaesthetized by 1% isoflurane and their head was restrained in a stereotaxic cylinder frame to minimize breathing motion. The scalp and the skull fascia were gently incised down the midline and peeled to the side. Saline was titrated onto the skull to keep it moist. Laser speckle images were recorded using a CMOS camera before MCAO, 15 min after occlusion and 15 min after reperfusion. For each animal, three sets of raw speckle images were acquired in <15 s (250 frames in each set; image width, 752 pixels; image height, 580 pixels; exposure time, 20 ms). A speckle contrast image was calculated from each raw speckle image using a sliding grid of 2.5 mm × 2.5 mm. A mean speckle contrast image was calculated for each set and used to calculate the rCBF. The rCBF in the ipsilateral (ischaemic) hemisphere was normalized to the mean rCBF in the contralateral (non-ischaemic) hemisphere. Speckle images were obtained and processed using mFLPI2Meas (v.2.0), rCBF data from all pooled hemispheres were obtained using moorFLPIReview (v.50). All analyses were randomized.

Molecular docking and Prime-MM/GBSA binding free-energy calculation

The structures of cASIC1a (PDB ID: 5WKU; resting state) and NMDAR (PDB: 5IOU) were obtained from the PDB. Structures of chemicals (that is, l-glutamate, AP5) were obtained from the PubChem compound or ChemBioDraw Ultra 14 software. Initial docking studies involved preparation of the cASIC1a and glutamate using the High-Ambiguity Driven protein-protein DOCKing (HADDOCK) software. The entire accessible extracellular surface area of the protein was considered as a plausible interface with glutamate. After docking calculations, the highest scoring docking poses were analysed. The results revealed several plausible glutamate-binding sites (Supplementary Table 2), which were subsequently validated by site-directed mutagenesis.

After identification of a most probable binding site, additional docking poses were generated using the Schrödinger Maestro software suite (Schrödinger, 2020-3). Before docking, protein was processed using the Protein Preparation Wizard to add missing residues, optimize side-chain positioning, remove bound waters, optimize H bonds and minimize energy (using OPLS3e force field). Ligands were optimized using the OPLS3e force field in LigPrep module. Protein and ligand protonation states at pH 7.0 ± 0.2 were sampled using Epik.

Ligands were docked to an identified residue (for example, Lys379) in a grid box with dimensions of 25 × 25 × 25 ų. Extra-precision docking (Glide XP) was performed with flexible ligand sampling, and post-docking minimization was performed to generate a maximum of ten poses per ligand within the Glide program. The docking conformation with a highest docking score was further analysed. The binding free energies of all different poses from XP docking outputs were carried out using the Prime-MM/GBSA module²⁹. The binding energy proxy was calculated by the software according to the following equation:

Molecular dynamics simulations

For molecular dynamics simulation, the pose with highest docking score of each ligand–protein complex was selected from docking results and the ligand-bound protein systems were built in 150 mM NaCl aqueous solution. To investigate the stability of the docked ligand–protein poses, 50 ns simulations were performed. After 25,000 steps of minimization, the systems were equilibrated using isothermal-isobaric (NPT) ensembles at a constant temperature of 303.15 K, followed by 50 ns production runs. All simulations used the program GROMACS 2020.3. CHARMM36m force field was used for the protein, GROMOS 54A7 force field for ligand⁵⁴ and the SPCE model for water. The simulation trajectories were analysed for structural fluctuations using root-mean-square deviation and root-mean-square fluctuation calculations. MM-PBSA calculations on molecular dynamics simulation trajectories were performed with a modified gmx_mmpbsa bash script using solvent-accessible surface area as the model for non-polar solvation energy.

CGS19755-binding pocket analysis and drug design

The pose with the highest docking score of the CGS19755–cASIC1a complex was selected from molecular docking for binding pocket analysis using Fpocket 2.0 software. A dpocket program analysis was performed to produce pocket parameters using the default settings. We used a scaffold-replacement method based on the CGS19755 structure to screen the ZINC20 database (fragment, lead-like and drug-like molecules) using the Molecular Operating Environment software (v.2015.10). The three-dimensional conformations of the remaining about 222 compounds were generated by the ligPrep module of Maestro (Schrödinger) with the OPLS3e force field. Possible ionization states of each compound were generated in the pH range of 7.0 ± 0.2 using Ionizer. Possible tautomer forms were also generated for each ligand. Compounds were screened using the high-throughput virtual screening module followed by the extra-precision docking module in Glide. The Glide docking score was used to rank the results list. Finally, six hits were selected for the electrophysiological assay.

MST assay for protein- and compound-binding assays

The MST measurement for the binding of compounds to ASIC1a was performed using Monolith NT.115 (NanoTemper)¹⁶. In brief, the HEK293T cell was collected after 48 h transfection with eGFP-tagged wild-type or mutant ASIC1a plasmid and lysed by M-PER mammalian protein extraction reagent (Thermo Fisher Scientific). Protease/phosphatase inhibitor (1%, Cell Signaling) was added to the lysate to avoid protein degradation. The level of ASIC1a was verified by measuring the total fluorescence. For binding studies, the lysate was diluted fourfold using ECF (with 0.1% Tween-20) to provide the optimal level of the fluorescent protein in the binding reaction. Compounds were titrated by ECF at a 1:1 ratio, diluted 16 times. Subsequently, 5 μl cell lysate was mixed with 5 μl compounds at different concentrations. After 5 min incubation at room temperature, all of the samples were loaded into MST NT.115 standard glass capillaries and measurement was performed at 70–100% excitation power to control the fluorescence value between 300 and 400 using the MO control software (v.1.6.1). The thermophoresis time (t) was 23 s, and the experimental temperature (T) was 25 °C. At least three independent experiments were repeated and then the data were imported into MO affinity analysis software (v.2.3) of NanoTemper was used to calculate the K_d value using the K_d model. The cold and hot fluorescence were measured at −1–0 s and 4–5 s to avoid thermally induced protein configuration change, respectively. The ECF pH was 7.0 unless otherwise described.

Chemical synthesis

LK-1, LK-2 and other screened compounds were custom synthesized by Simcere Pharmaceutical for this study (Supplementary information). For electrophysiology, stock solutions of two compounds were prepared in water and diluted in the ECF solutions before use for in vitro experiments or in saline for animal studies in vivo.

Pharmacokinetics study

Pharmacokinetics of LK-2 was analysed in male C57BL/6J mice (n = 22). Plasma and brain concentrations were determined using LC–MS/MS methods after a single intraperitoneal injection dose (30 mg per kg) of compound as a clear solution in 0.9% saline at a concentration of 1 mg ml⁻¹. Blood samples were collected into an EDTA-coated test tube at the time points of 0.083 h, 0.25 h, 0.5 h, 1.0 h, 2.0 h and 4.0 h, and then centrifuged at 2,000g for 15 min to generate plasma samples. Brain samples were collected after intraventricle perfusion with normal saline and prepared by homogenizing tissue with 5 volumes (w:v) of 0.9% NaCl. LC–MS/MS methods to quantify LK-2 in plasma and brain samples were developed with the LC-MS/MS-T API 4000 instrument. General sample processing procedure was performed as follows: (1) an aliquot of 30 µl plasma sample, calibration standard, quality control, single blank and double blank sample was added to the 1.5 ml tube. An aliquot of 40 µl brain homogenate, calibration standard, quality control, single blank and double blank sample was added to the 96-well plate respectively. (2) Each sample (except the double blank) was quenched with 150 µl (for plasma samples) or 200 µl (for brain homogenates) of IS 1 (6 in 1 internal standard in methanol (labetalol, tolbutamid verapamil, dexamethasone, glyburide and celecoxib, 100 ng ml⁻¹ for each) with 40 mM DBAA) (the double blank sample was quenched in methanol with 40 mM DBAA), and the mixture was then vortex-mixed well (at least 15 s) and centrifuged for 15 min at 12,000g (for plasma samples) or 3,220g (for brain homogenates) at 4 °C. (3) An aliquot of 65 µl supernatant was transferred to the 96-well plate and centrifuged for 5 min at 3,220g at 4 °C. All of the processes were performed on the wet ice. All of the supernatants were directly injected for LC–MS/MS analysis. The column used was an ACQUITY UPLC BEH C18 2.1 × 100 mm, 1.7 μm column. The column temperature was 40 °C. The flow rate was 0.4 ml min⁻¹. The mobile phase consisted of A: 0.001% NH₃·H₂O with 0.18 mM DBAA in water; and B: 10 mM DMHA and 3 mM NH₄OAc in ACN/Water (v:v, 50:50). Standard curves were prepared by spiking compounds into control plasma and brain and these were used to determine drug concentrations. Pharmacokinetic parameters were calculated by non-compartmental analysis using DAS v.2.0 with the mean concentration at each timepoint.

Rotarod test

Motor behaviours were tested using the rotarod test (RWD). The rotarod measures the ability of mice to maintain balance on a motor pressure-driven rotating rod. In this test, mice were put on a rotating rod, and the time to fall off was recorded. Mice were trained five times to learn moving on the rod before testing, in which the speed was accelerated from 4 to 40 rpm in 3 min. The rod was cleaned with alcohol before each test.

Statistics

All data are reported as mean ± s.e.m. The number of biological replicates has been reported as nested data. Two-tailed paired and unpaired Student’s t-tests were used where appropriate to examine the statistical significance of the difference between groups of data. Comparisons among multiple groups were analysed using one-way and two-way ANOVA followed by Tukey multiple-comparison tests for post hoc analysis. Prism 8 was used to analyse all data.

Reporting summary

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