Structure-Based Identification and Characterization of Inhibitors of the Epilepsy-Associated KNa1.1 (KCNT1) Potassium Channel.

Drug-resistant epileptic encephalopathies of infancy have been associated with KCNT1 gain-of-function mutations, which increase the activity of KNa1.1 sodium-activated potassium channels. Pharmacological inhibition of hyperactive KNa1.1 channels by quinidine has been proposed as a stratified treatment, but mostly this has not been successful, being linked to the low potency and lack of specificity of the drug. Here we describe the use of a previously determined cryo-electron microscopy-derived KNa1.1 structure and mutational analysis to identify how quinidine binds to the channel pore and, using computational methods, screened for compounds predicated to bind to this site. We describe six compounds that inhibited KNa1.1 channels with low- and sub-micromolar potencies, likely also through binding in the intracellular pore vestibule. In hERG inhibition and cytotoxicity assays, two compounds were ineffective. These may provide starting points for the development of new pharmacophores and could become tool compounds to study this channel further.

Introduction

Gain-of-function mutations in the KCNT1 gene are associated with severe, drug-resistant forms of childhood epilepsy (Barcia et al., 2012, Heron et al., 2012). Epilepsy of infancy with migrating focal seizures (EIMFS) and autosomal dominant nocturnal frontal lobe epilepsy (ADNFLE) were the first disorders found to be associated with KCNT1, with the more recently described Ohtahara syndrome (Martin et al., 2014), West syndrome (Ohba et al., 2015), Lennox-Gastaut syndrome (Jia et al., 2019), and sleep-related hypermotor epilepsy (Cataldi et al., 2019, Rubboli et al., 2019). In addition to frequent seizures, patients may also present developmental delay and psychiatric and intellectual disabilities (Gertler et al., 2018, Lim et al., 2016). KCNT1 encodes a sodium-activated potassium channel subunit, KNa1.1, which has previously been termed SLACK and Slo2.2 (Joiner et al., 1998, Yuan et al., 2003). Similar to other potassium channels, the functional proteins are formed by a tetramer of subunits, each of which possesses six transmembrane alpha helices, a re-entrant pore loop between the fifth and sixth helix that forms the selectivity filter, and two intracellular regulation of conductance of potassium (RCK) domains (Hite et al., 2015). KNa1.1-containing channels are expressed throughout the central nervous system (Bhattacharjee et al., 2002, Rizzi et al., 2016) and are believed to have a stabilizing effect on the membrane potential following sodium influx during neuronal excitation (Budelli et al., 2009, Cervantes et al., 2013, Hage and Salkoff, 2012, Liu and Stan Leung, 2004, Nanou et al., 2008). Virtually all epilepsy-associated KCNT1 mutations increase channel activity, although why epilepsy should arise is not understood, since potassium channel opening is usually associated with a decrease in neuronal excitability. One proposed mechanism, based on studies of human stem cell-derived neurons harboring one such mutation, is that hyperexcitability can be caused by an enhanced sodium-dependent after-hyperpolarization, facilitating an increase in the rate of action potential firing (Quraishi et al., 2019).

Quinidine is a class I antiarrhythmic agent, which exerts its effects by non-selectively inhibiting cardiac cation channels at micromolar concentrations. Notably, quinidine also inhibits KNa1.1 channels (Yang et al., 2006), including those harboring epilepsy-causing mutations, at similar concentrations, leading to the hypothesis that quinidine could reverse the gain of function and treat KCNT1-associated epilepsy (Mikati et al., 2015, Milligan et al., 2014). Limited improvement has been achieved in a small number of patients using quinidine therapy, but in the majority of cases it is ineffective (Chong et al., 2016, Fitzgerald et al., 2019, Madaan et al., 2018, McTague et al., 2018, Mikati et al., 2015, Mullen et al., 2018). The lack of selectivity and low potency of quinidine in inhibiting KNa1.1 channels, with IC50 values in the order of 0.1 mM (Rizzo et al., 2016, Yang et al., 2006), in the central nervous system without significantly affecting cardiac function is a concern and limits the dosing levels of quinidine (Mullen et al., 2018). Moreover, there is still a paucity of information on the binding site and the mode of action of KNa1.1 inhibitors. Other reported inhibitors of KNa1.1 are bepridil (Yang et al., 2006) and clofilium (de Los Angeles Tejada et al., 2012), both of which also have inhibitory effects on cardiac cation channels. The identification of alternative drugs that better target KNa1.1 are therefore desired (Mikati et al., 2015, Milligan et al., 2014). The mechanisms by which these three drugs inhibit potassium channels, hERG in particular, have been studied previously and involve direct block of the pore via the intracellular vestibule and are coordinated by aromatic side-chains, such as phenylalanine (Kamiya et al., 2006, Knape et al., 2011, Macdonald et al., 2018, Perry et al., 2004, Wrighton et al., 2015).

With the development of direct electron detectors, more powerful microscopes, and improved data processing software, there has been an expansion of the use of cryo-electron microscopy (cryo-EM) and single particle analysis to determine high-resolution structures of membrane proteins (Rawson et al., 2016). Bypassing the need to crystallize membrane protein samples makes this approach particularly attractive. One of the first high-resolution ion channel structures determined by cryo-EM was the chicken KNa1.1 channel, initially described at 4.5 Å resolution (Hite et al., 2015), with the inactive and activated conformations subsequently resolved to 4.3 and 3.8 Å, respectively (Hite and MacKinnon, 2017). The increase in resolution enables the generation of molecular models, which presents new opportunities that can be utilized in computer-aided drug discovery (CADD) or in silico drug design. To discover new therapies for rare disorders, such as those caused by KCNT1 mutations, efficiencies of drug development cost and time are required (Swinney and Xia, 2014). Computer-aided approaches, such as virtual screening, can help reduce the need for functional screening of large compound sets by predicating which chemicals are likely to occupy a defined binding site (Lin et al., 2020).

In this study, we used the cryo-EM-derived structures of the KNa1.1 channel to model quinidine binding in silico, having hypothesized that it blocks the channel pore, similar to its interaction with hERG. Having identified the intracellular pore vestibule as the likely site for known inhibitors to bind, we then conducted virtual high-throughput screening of a library of commercially available compounds with the view of identifying more potent and selective binders. We now report the identification of six diverse compounds that inhibit KNa1.1 more potently than the quinidine.

Results

Identification of Inhibitor Binding Site in the KNa1.1 Channel Pore

In the absence of potent and selective inhibitors of KNa1.1 potassium channels, we sought to use computational approaches to identify inhibitors with improved properties. To identify a region in the KNa1.1 structure to focus in silico screening, we first sought to identify how compounds known to inhibit the channel exert their effects. Hypothesizing that both quinidine and bepridil inhibit channels by occupying the inner pore vestibule we created a minimal structural model of the channel pore by removing the S1 to S4 and the cytosolic domains of the cryo-EM structures of the “closed” and “open” chicken KNa1.1 channel (Hite and MacKinnon, 2017). This left the S5, pore loop, and S6 of each subunit, which is highly conserved between chicken and human KNa1.1. Using automated procedures, both inhibitors were docked into the pore in its closed conformation at two distinct sites: in the vicinity of the equivalent positions of F346 and M354 in the S6 segment of the human isoform (Figure 1A). In contrast, using the model of the open conformation, quinidine and bepridil could only be docked to the site involving F346 (Figure 1A). To validate the docking, both residues (F346 and M354) of human KNa1.1 were mutated to isoleucine and serine (Figures 1B and 1C) and inhibition by quinidine and bepridil was evaluated further. Mutation of M354 had modest effects, with no significant effect on potency of quinidine, but the mutation of F346 to I346 reduced the potency approximately 10-fold (Figure 2). We found that mutating these pore residues, particularly F346, also increased channel activity, with respect to current amplitude and reduced current rectification (Figures 1B–1D). To exclude the possibility that increased channel activity was not the cause of the reduced efficacy of the inhibitors, for example, through preferential block of the closed state, their effects on a disease-causing gain-of-function mutation, Y796H (Heron et al., 2012), was explored. This mutation causes an amino acid substitution in a KNa1.1 intracellular region distal to the pore domain and caused a large increase in KNa1.1 channel activity (Figures 1B–1D). In contrast to the pore mutants, quinidine inhibited this mutant channel with a 3-fold increase in potency (Figure 2B).

Figure 1

Molecular Docking of Inhibitors and Functional Analysis of Mutant KNa1.1 Channels Used in This Study

(A) Docking of quinidine and bepridil (yellow) into the KNa1.1 pore domain, comprising the S5, P loop, and S6 of the closed and open conformational states (gray); side chains equivalent to F346 (magenta) and M354 (cyan) of the human KNa1.1 homolog are indicated.

(B and C) (B) Representative whole-cell currents and (C) mean (±SEM, n = 5–9 cells) current-voltage plots from non-transfected (NT) HEK 293 cells and cells transfected with wild-type (WT), pore mutant (F346S, F346I, M354S, M354I), or disease-causing mutant (Y796H) KNa1.1.

(D) Mean current amplitude at 0 mV from the data in (C); ∗p < 0.05, ∗∗p < 0.005 compared with WT, independent one-way ANOVA with Games-Howell post hoc test.

Figure 2

Concentration-Inhibition Analysis of Wild-Type and Mutant KNa1.1 Channels by Quinidine and Bepridil

(A) Representative currents evoked by voltage ramps from cells expressing WT or mutant KNa1.1 with increasing concentrations of quinidine as indicated.

(B) Mean (±SEM) concentration-inhibition plots for wild-type and mutant KNa1.1 channels in response to 3 μM to 1 mM quinidine. Mean (±SEM) IC50 for WT, 124.99 ± 34.52 μM (n = 5 cells); F346I, 736.08 ± 94.09 μM (n = 5); F346S, 1.23 ± 0.19 mM (n = 4); M354I, 99.23 ± 49.61 (n = 5); M354S, 247.16 ± 19.96 μM (n = 5); Y796H, 38.00 ± 12.89 μM (n = 5).

(C) Mean (±SEM) concentration-inhibition plots for wild-type and mutant KNa1.1 channels in response to 0.3–100 μM bepridil. IC50 for WT, 6.36 ± 2.12 μM (n = 5); F346I, 35.91 ± 11.01 μM (n = 4); F346S, 23.43 ± 5.17 μM (n = 5).

Virtual Screening and Validation of KNa1.1 Inhibitors

Having identified the internal vestibule, just below the selectivity filter, as the likely site for inhibitor binding, we used this region in the minimal pore structure of the open channel conformation to dock compounds using in silico screening and a diverse library of 100,000 drug-like molecules. This approach was complemented by computational identification of compounds with structural similarity to bepridil, the most potent of the known inhibitors of KNa1.1, and then also docking these compounds into the structure of the KNa1.1 channel pore. Both computational techniques resulted in a list of compounds, ranked by their docking score and predicted binding affinities. A selection of 17 compounds (details in Table S1), based on their ranking and commercial availability, were then obtained and evaluated for their ability to inhibit human KNa1.1 channels expressed in HEK cells. At 10 μM, six of the compounds inhibited KNa1.1 currents by at least 40% and were selected for further analysis (Figure 3A). Initially, and to validate inhibition in the inner pore vestibule, we tested the ability of these compounds to inhibit F346S KNa1.1, which had showed reduced sensitivity to both quinidine and bepridil. At 10 μM, the degree by which F346S KNa1.1 was inhibited by each compound was reduced, compared with WT KNa1.1 (Figure 3A). Concentration-inhibition analysis (Figures 3B–3D) yielded mean IC50 concentrations ranging from 0.6 to 7.4 μM with WT KNa1.1. In comparison, quinidine and bepridil inhibited WT KNa1.1 with IC50 values in the order of 125 μM and 6.4 μM, respectively (Figures 2 and 3D). Inspection of the inhibitors docked into the KNa1.1 pore domain suggests that binding involves both hydrophobic interactions with S6 pore-lining residues and hydrogen-bonding with P loop residues (Figure 4 and Table S1).

Figure 3

Functional Evaluation of Top-Scoring Molecules from In Silico Docking

(A) Mean (±SEM, n = 3–4 cells) WT KNa1.1 conductance measured as the slope of the current evoked by depolarizing voltage ramps, relative to baseline, in the presence of 10 μM test compound; with those that were active (right of dashed line) counter-tested with F346S KNa1.1 pore mutant (∗p < 0.05, ∗∗p < 0.005, ∗∗∗p < 0.0005, t test.).

(B and C) (B) Representative traces and (C) mean (±SEM, n = 5–7 cells) concentration-inhibition plots for active inhibitors.

(D) Summary table with mean (±SEM) potencies of compounds inhibiting WT and Y796H KNa1.1 (n = 5–7 cells). ∗p < 0.05 versus potency with WT KNa1.1 (Student t test).

See also Table S1, Figures S1 and S2.

Figure 4

Chemical Structures of Inhibitors (Magenta) and Their Docked Poses in the KNa1.1 Pore Domain

Side chains at positions equivalent to the human isoform are indicated (pink): T314 in the P loop and F346 and M354 in the S6 segment of human KNa1.1. For clarity, the compounds are shown above each panel for the inhibitors analyzed in Figure 3.

The nature of the inhibition by these compounds was explored further. To be used therapeutically, KNa1.1 inhibitors would be required to inhibit channels that have epilepsy-causing amino acid substitutions. We therefore explored the potency of the inhibitors with Y796H KNa1.1 gain-of-function mutation. Each of the six compounds inhibited the mutant channels, but in this experiment BC7 and BC14 were significantly less potent (Figures 3B–3D and S1). We also explored the importance of chemical groups in channel inhibition, using BC12 as an example, for which several analogues were commercially available. Each analogue failed to reduce WT KNa1.1 channel currents by more than approximately 10% at 10 μM and were deemed inactive (Figure S2).

Preliminary Toxicological Assessment of KNa1.1 Channel Inhibitors

Finally, to anticipate toxicological effects of these compounds, should they or derivatives be developed further, we studied their effects on hERG potassium currents and in a cellular toxicity assay. From measurements of tail current amplitudes, 10 μM BC5, BC6, and BC7 almost completely inhibited hERG channels expressed in HEK 293 cells (>80%, Figures 5A and 5B). In contrast, BC14 had a partial effect (approximately 45%) at the same concentration, whereas BC12 and BC13 were even less effective, reducing currents by approximately 10%–20%. In cytotoxicity assays, which involved exposing HEK 293 cells to compounds for 24 h, only BC7 exhibited a concentration-dependent reduction in cell viability, at concentrations of 1 μM and above (Figure 5C). With BC5, B6, and BC14, a reduction in cell viability was found at concentrations an order of magnitude higher than the IC50, whereas BC12 and BC14 had no effect at all concentrations tested, up to and including 100 μM (Figure 5C). Blasticidin (10 μg/mL) and DMSO (10% v/v) reduced cell viability in the order of 45% and 90%, respectively.

Figure 5

Preliminary Toxicological Assessment of KNa1.1 Channel Inhibitors

(A) Representative hERG whole-cell currents recorded from a transfected HEK 293 cell in the absence (control) and presence of 10 μM inhibitor, as indicated.

(B) Mean (±SEM, n = 3) tail current at −50 mV remaining in the presence of each inhibitor.

(C) Cytotoxicity assays indicating mean (±SEM, n = 3) viability of HEK 293 cells using WST reagent following overnight exposure to inhibitor at the indicated concentrations, with 10 μM blasticidin and 10% v/v DMSO as positive controls; ∗p < 0.05, ∗∗p < 0.005, ∗∗∗p < 0.0005, independent one-way ANOVA with Dunnett's post hoc test.

Discussion

Analysis of compounds identified as potential KNa1.1 inhibitors by in silico docking yielded six previously unknown inhibitors, each more potent than quinidine that has been trialed clinically as a stratified treatment for KCNT1-associated epilepsy. These inhibitors are structurally diverse and there is no clear pharmacophore, although BC6 and BC7 are among the scaffolds with structural similarity to bepridil. We propose that each of these compounds, in addition to quinidine and bepridil, inhibits the channel by blocking the pore via the intracellular vestibule. This is indicated by the reduced efficacy of each compound with the F346S pore mutation, and which is likely to be independent of the increased channel activity caused by the mutation, since with quinidine there was no similar loss of efficacy with Y796H, an epilepsy-causing mutation distal from the pore. The effects of mutations of F346 are somewhat modest, yielding a 10-fold decrease in potency, compared with WT KNa1.1, suggesting that other interactions are necessary, but are consistent with the hypothesis that inhibitors block the channel by occupying the inner vestibule of the pore. This residue is also at the equivalent position in the S6 segment as F656 in hERG that co-ordinates inhibitor binding of quinidine, bepridil, and clofilium (Kamiya et al., 2006, Macdonald et al., 2018, Perry et al., 2004). This mode of action requires the inhibiting compound to traverse the plasma membrane and enter the pore via the cytoplasm. It is noteworthy that each of the compounds tested that had a calculated logP (cLogP) value of 3.2 or less did not exhibit inhibition of KNa1.1, and this may have been a result of poor membrane permeability. By using a model of the ion channel pore that is putatively in the open conformation, we had anticipated that inhibitors identified through docking would achieve similar or even higher potency with the Y796H mutant KNa1.1, but this was not always the case. It is possible that the apparent decrease in potency of some of the compounds with Y796H may be attributed to the high level of basal channel activity with this mutant channel and incomplete series resistance compensation. This could underestimate channel inhibition at the lower concentrations and increase variability. The increase in potency of quinidine with Y796H, relative to WT KNa1.1, however, is consistent with previous studies of other gain-of-function mutations (Rizzo et al., 2016).

It may have been possible to obtain inhibitors by testing compounds identified through computational tools to have structures similar to the known inhibitors, quinidine, bepridil, and clofilium, but without using molecular docking. The risk with this approach is that the compounds may also have similar potency in inhibiting cardiac ion channels, which would preclude their further development. Indeed, this is what we found with BC6 and BC7, which both almost completely inhibited hERG currents at 10 μM. Instead, by docking a diverse library, we were able to identify compounds that were structurally distinct from the known inhibitors and that included some that had little effect on hERG at 10 μM. Together, this highlights the importance of screening a wide range of compounds, which in our approach employed virtual screening using computational approaches instead of a high-throughput cell-based functional screen. Other sites in the channel protein could potentially be used to target molecular docking and virtual screening, which may enhance KNa1.1 selectivity over other ion channels, but there is presently a lack of understanding of how other KNa1.1 domains could be targeted pharmacologically to modulate channel function.

Although the cryo-EM structures indicate that activation by sodium involves an expansion of the intracellular pore vestibule (Hite and MacKinnon, 2017), functional experiments with this and the closely related KNa1.2 and KCa1.1 (BKCa) channels point to the selectivity filter and proximal hydrophobic residues, rather than an S6 helix bundle, as the location of the channel gate (Garg et al., 2013, Giese et al., 2017, Jia et al., 2018, Suzuki et al., 2016). This means that the inhibitors described here block at the channel gate and this should be a mode of inhibition that is efficacious with virtually all clinical gain-of-function mutations, independent of the mechanism by which increased open probability is achieved, rather than an inhibitor that binds to modulatory sites. An exception, however, may be F346L that was identified in a patient with EIMFS (McTague et al., 2018), which, based on our findings with mutations generated at this site in the pore, may have reduced inhibitor sensitivity. Indeed, it was found that F346L KNa1.1 channels expressed in Xenopus oocytes were resistant to block by 300 μM quinidine (McTague et al., 2018), although the pharmacological effects of quinidine on channels containing a mixture of F346L and wild-type subunits, to mimic the heterozygous disorder, has not been explored.

Notwithstanding our preliminary efforts to characterize the potential toxicity of these inhibitors by studying inhibition of hERG channel currents and the effects in cell viability assays, we are unable to make any statement regarding their selectivity or safety ahead of any in vitro or in vivo investigation. Since the computational approach was based on the inhibition by quinidine and bepridil, one might expect a similar range of ion channels to be inhibited by the compounds described here. However, the low level of inhibition of hERG at 10 μM by BC12 and BC13 suggests that across the compounds there may be varying levels of selectivity, which could be further improved by synthesis and analysis of derivative compounds and testing their effects on a range of different cation channels. We do note, however, the characteristics of pan-assay interference (PAINS) in compounds BC6, BC7, and BC13, owing to the presence of the conjugated carbonyl group, meaning there may be non-specific effects in other functional assays. The lack of effects of the four analogues of BC12 tested demonstrates the potential for generating an inhibitor pharmacophore from this compound; thus, these may provide starting points for the development of more potent inhibitors.

The generation of high-resolution structural data by cryo-electron microscopy and single particle analysis has had a significant impact in structural biology. Importantly, circumventing the need to crystallize samples means that membrane proteins are more amenable to analysis, and also that human proteins, rather than homologs from prokaryotes and lower eukaryotes, now feature more prominently. This means that, for in silico analysis and drug discovery, target proteins very close to human, if not the human protein itself, can be utilized. The pore domain of the chicken KNa1.1 subunit, however, is virtually identical to that of the human homolog, and we demonstrate its suitability for in silico docking experiments to provide potential inhibitors to characterize using a functional assay. Our study, in addition to identifying KNa1.1 inhibitors from a compound library, provides a further example of the use of cryo-electron microscopy-generated membrane protein structural data in identifying new small molecule inhibitors through a structure-based discovery approach.

Limitations of the Study

A structure of the human KNa1.1 channel was not available, so we employed that of the chicken homolog. The channel domain used for the docking experiments is highly conserved between chicken and human KNa1.1, and the specific residues discussed are conserved. By restricting the docking experiments to the intracellular pore vestibule of the channel, to decrease computational complexity, we disregarded potential interactions of compounds with other parts of the protein.

Resource Availability

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr Jonathan Lippiat (j.d.lippiat@leeds.ac.uk).

Material Availability

Plasmids generated in this study will be made available on request, but we may require a payment and/or a completed Materials Transfer Agreement if there is potential for commercial application.

Data and Code Availability

Raw data will be shared upon receipt of a reasonable request.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.