Development of an Automated Screen for Kv7.2 Potassium Channels and Discovery of a New Agonist Chemotype

Abstract

To identify pore domain ligands on Kv7.2 potassium ion channels, we compared wild-type (WT) and W236L mutant Kv7.2 channels in a series of assays with previously validated and novel agonist chemotypes. Positive controls were retigabine, flupirtine, and RL-81; i.e. Kv7.2 channel activators that significantly shift voltage-dependent activation to more negative potentials (ΔV₅₀) at 5 μM. We identified 6 new compounds that exhibited differential enhancing activity between WT and W236L mutant channels. Whole cell patch-clamp electrophysiology studies were conducted to identify Kv7.2. Kv7.2/3, Kv7.4, and Kv7.5 selectivity. Our results validate the SyncroPatch platform and establish new structure activity relationships (SAR). Specifically, in addition to selective Kv7.2, Kv7.2/3, Kv7.4. and Kv7.5 agonists, we identified a novel chemotype, ZK-21, a 4-aminotetrahydroquinoline that is distinct from any of the previously described Kv7 channel modifiers. Using flexible receptor docking, ZK-21 was predicted to be stabilized by W236 and bind perpendicular to retigabine, burying the benzyl carbamate group into a tunnel reaching the core of the pore domain.

Graphical Abstract

The Kv7 (also referred to as KCNQ) potassium channel family of transmembrane proteins, encoded by KCNQ genes, plays a critical role in cell excitability. They belong to the voltage-gated ion channel superfamily, are non-inactivating channels, and contribute both to the maintenance of the resting membrane potential and the control of excitability in cells.¹ The KCNQ gene family is comprised of 5 isoforms, Kv7.1–5, and assembles as either homo- or heterotetramers (i.e. Kv7.2/3, Kv7.3/5, or Kv7.4/5) (Figure 1). Each isoform has a specific distribution within the body, and specific mutations within the gene family have been implicated in human diseases.² Accordingly, the search for isoform-specific, potent chemical probes aimed at the mechanistic dissection of the function of Kv7 channels in various tissues remains an area of high interest.^3,4,5

Figure 1.

(A) Topology of single voltage-gated potassium ion channel subunit (Kv); the retigabine/W236 site is labeled in the S5 helix of the pore domain. (B) A heterotetrameric channel assembly.

Kv7.2 channels are located both in the central and peripheral nervous system,⁶ showing a high density in the neocortex and hippocampus. High-current amplitude Kv7.2/3 heterotetramers are the most abundant Kv7.2 channel assembly compared to the less abundant low-current amplitude homomeric channels and Kv7.4 or Kv7.5 heterotetramers.⁷ Particular attention has been given to Kv7.2/3 channels for their M-current contributions in neuronal tissue. Among other functions, the homomeric Kv7.2 isoform has been implicated in the pathophysiology of epilepsy,⁸ while the heteromeric Kv7.2/3 isoform has been associated with the induction and prolongation of tinnitus.⁹

To date, only two drugs have been approved by regulatory agencies with Kv7 agonism as a main mechanism of action for the treatment of seizures: retigabine (USA) and a closely related pyridine analog, flupirtine (EU). Despite its efficacy, retigabine was withdrawn from the market in 2017 due to off-target effects such as blue skin discoloration, eye abnormalities, and urine crystallization.¹⁰ Flupirtine is readily oxidized into toxic quinone diimines,¹¹ and in 2018, the European Medicines Agency recommended discontinuing this drug because of the risk for serious liver injury. Altogether, the withdrawal of both of these pan Kv7.2–5 agonists created a substantial void in the treatment of drug-resistant epilepsies and other hyperexcitability disorders. The therapeutic potential of Kv7 agonists for many channelopathies, including epilepsy, pain, tinnitus, cancer, cardiovascular disease, and neurodegeneration, is well validated. In fact, recent clinical studies have reenforced interest in retigabine as a target for precision medicine for the treatment of seizures associated with Kv7.2 developmental and epileptic encephalopathy.

Retigabine is a pan-Kv7.2–5 channel opener. Mutagenesis studies implicated a binding site near the interface between the voltage-sensing domain (VSD) and pore domain (PD) near W236, a region crucial for electromechanical coupling.¹² Recently, cryo-EM structures of Kv7.2 and Kv7.4 in complex with small molecule ligands including retigabine have confirmed this binding site and shed additional light on channel dynamics.^13,14

In prior work, we have reported the synthesis and characterization of a series of fluorinated analogs of retigabine such as RL-81, RL-24, and RL-56 that demonstrate superior selectivity and potency as agonists in Kv7.2–5 channel assays.^15,16 Other promising lead structures under recent development include P-retigabine, acrylamide (S)-1, BMS-204352, GRT-X, ICA-27243, PF-05020182, ML-213, ztz240, and compounds from Xenon and Knopp Pharmaceuticals (Figure 2).^17,18

Figure 2.

Structures of representative Kv7.2–5 channel activators.

To complement the RL-81 lead series, we sought to establish a high-content automated screen to identify structurally novel Kv7.2 modulators and assess their activity. Toward this goal, we developed a standard procedure for the Syncropatch 384PE platform to screen an in-house collection of 22 compounds at a concentration of 5 μM on wild-type (WT) and W236L mutant Kv7.2 channels. In addition to using flupirtine, retigabine and RL-81 as positive standards, we screened 5 diverse chemotypes, e.g. anilines RL-15, RL-25, RL-56 and RL-71,¹⁶ thiazine 1,1-dioxide CH-02, 3-aminoisoquinolinone DP-19,¹⁹ 4-aminotetrahydroquinolines ZK-16, ZK-21, ZK-29, ZK-94 and MS-56,²⁰ and 2-arylindoles PM-17, PM-24, PM-28, PM-29, PM-44, PM-57, PM-63 and PM-91 (Figure 3). The syntheses of CH-02 and the 2-phenylindole series are summarized in Schemes 1 and 2, respectively.

Figure 3.

Structures of the aniline (“RL”), thiazine 1,1-dioxide (“CH”), 3-aminoisoquinolinone (“DP”), 4-aminotetrahydroquinoline (“ZK”, “MS”), and 2-phenylindole (“PM”) series of screening samples.

Scheme 1.

Synthesis of CH-02.

Scheme 2.

Synthesis of PM-17, PM-24, PM-28, PM-29, PM-44, PM-57, PM-63, and PM-91.

S_NAr reaction of aminoalcohol 1 with commercially available trifluoronitroaniline 2 in the presence of triethylamine and iodine in DMSO generated the benzylic aniline 3 in 92% yield (Scheme 1). A Mitsunobu reaction with thioacetic acid converted the alcohol to the thioester 4 in 93% yield. After considerable optimization, we identified a two-step protocol of reductive ester cleavage followed by a cesium carbonate mediated S_NAr reaction under microwave (MW) conditions to access thiazine 6 in 91% overall yield. Zinc-reduction of the nitro group in 6 generated diamine 7, and N-carbamoylation followed by selective oxidation²¹ of the thioether with sodium tungstate provided thiazine 1,1-dioxide CH-02. While the yield for the reduction was high (98%), the last two steps suffered from lower efficiency (35–40%) due to challenges in the purification of intermediate 8 and product CH-02.

As an alternative to the benzothiazine and tetrahydroquinoline scaffolds, we also explored indole analogs⁴ in the PM-series (Scheme 2). Sonogashira coupling of alkynes 9a-c with 2-iodo-4-nitroaniline 10 followed by base-mediated cyclization provided 5-nitro-2-phenylindoles 11a-c in 36–66% overall yield.²² The nitro group in 11a-c was reduced with Zn/NH₄Cl to give the amine intermediate that was further converted to carbamates PM-63, PM-24, PM-44, PM-17 and PM-91 in 21–66% yield over two steps. Alternatively, aniline 12 generated from 11a was coupled with 4-trifluoromethoxy phenylacetic acid 13 and acid chlorides 14 and 15 to form amides PM-57, PM-29, and PM-28, respectively, in 30–38% yield.

SyncroPatch assay data.

According to the optimized procedure for the SyncroPatch 384PE platform described in the SI, the collection of 22 compounds was first screened at a single concentration of 5 μM at Kv7.2 wild type (WT) and W236L mutant channels (Figure 4A). Active test samples that bind in the channel pore region show a substantial negative ΔV₅₀ shift. Table 1 summarizes numerically how much samples affect the voltage dependence of these channels compared to the reference (V₅₀ is set as zero for reference (vehicle) for WT or W236L, respectively). As expected, the positive controls flupirtine, retigabine, and RL-81 increased the half-maximal voltage-dependent Kv7.2 activation (ΔV₅₀) up to three-fold at 5 μM and thereby provided large shifts in ΔV₅₀ values. In the RL-series, only RL-15 was significantly less potent in the WT channel assay. The activity of these analogs was critically dependent on the presence of W236. Among the remaining 15 compounds, we identified one, ZK-21, that also exhibited strongly differential enhancing activity between WT and W236L channels. One indole, PM-57, showed a small preference for WT binding. Furthermore, in a dose-response format, we observed EC₅₀ values between 50 nM and 8.6 μM in the WT channel, while none of these compounds showed measurable activity with the mutant channel (Figure 4B and Table 2). The most potent analog, RL-71, which has a substantially higher number of fluorine substituents compared to retigabine, is ca. 100-fold more potent on Kv7.2 than retigabine. A monodeuterated analog of RL-81, i.e. RL-25, was also analyzed in a concentration-response manner and found to be 4-times more active than retigabine but 20-times less active than RL-71, illustrating the important contribution of the 5-position fluorine atom to Kv7.2 channel activation (Figure 3 and Table 2). In contrast, the conformationally restricted analog CH-02 was completely inactive, suggesting that the correct positioning of a flexible benzylamine side chain is critical for channel binding, or, alternatively, the presence of the SO₂ group interferes with binding to the channel.

Figure 4.

Functional characterization of activators on human wild type (WT) and mutant W236L Kv7.2 channels. (A) Differential effect on Kv7.2 ΔV₅₀ of 22 analogs measured at 5 μM. Variation is expressed as ±SD, and values are listed in Table 1. (B) Concentration–response curves for six Kv7.2 channel hits. The ΔV₅₀ represent the half-activation voltage (V₅₀) of G-V curve shifts in the presence and in the absence of the compound. ΔV₅₀ (mV) = V₅₀ in control − V₅₀ in the presence of analogs was plotted against the compound concentration. Values are the mean ± SEM.

Table 1.

ΔV₅₀ values of 22 compounds measured at 5 μM on Kv7.2 and Kv7.2W236L from automated patch-clamp studies.^a Values are the mean ± SD.

Compound	ΔV50±SD (mV)
	Kv7.2 (N)	Kv7.2W236L (N)
Flupirtine	−23.8±7.2 (36)	1.2±5.3 (108)
Retigabine	−34.3±4.3 (108)	4.2±4.8 (135)
RL-81	−48.1±8.1 (90)	1.9±9.6 (144)
RL-25	−45.2±3.5 (63)	−4.4±4.5 (72)
RL-15	−24.6±12.7 (54)	−6.9±5.7 (126)
RL-56	−44.9±4.0 (45)	−9.3±5.5 (135)
RL-71	−41.2±15.5 (36)	−6.9±1.8 (117)
PM-17	−0.66±6.7 (72)	2.5±5.6 (198)
PM-24	−3.2±8.5 (81)	−0.31±8.7 (153)
PM-28	−4.1±8.0 (72)	3.2±4.3 (180)
PM-29	−1.1±6.4 (45)	2.4±4.7 (117)
PM-44	6.2±5.6 (72)	4.7±4.4 (126)
PM-57	−8.9±3.5 (36)	3.9±8.3 (99)
PM-63	−4.8±10.6 (90)	−6.0±9.0 (81)
PM-91	−8.3±12.3 (81)	−6.9±5.2 (108)
ZK-16	10.0±10.5 (63)	−2.7±11.7 (81)
ZK-21	−27.3±11.5 (162)	−6.9±1.6 (81)
ZK-29	−3.0±8.1 (45)	2.8±10.8 (108)
ZK-94	6.2±10.9 (45)	4.9±7.5 (99)
CH-02	4.4±6.8 (54)	2.8±7.1 (144)
DP-19	−3.0±4.7 (72)	1.3±5.6 (90)
MS-56	2.5±5.4 (63)	0.26±5.1 (108)

^aMeasured on the SyncroPatch 384PE using whole cell patch-clamp and multi-hole chips (4 holes per well).

Table 2.

EC₅₀ values for RL-15, RL-25, RL-56, RL-71, PM-91, and ZK-21 on Kv7.2 and Kv7.2W236L from automated patch-clamp concentration-response curves.^a

Compound	Kv7.2		Kv7.2W236L
	EC50 (μM)	EMAX (mV)	EMAX (mV)
Retigabine	4.21±0.85	54.4±3.6	1.11±0.99
RL-15	1.13±0.95	31.8±2.7	8.25±1.74
RL-25	0.98±0.69	52.3±1.6	5.75±1.66
RL-56	3.03±0.93	56.2±1.7	9.13±1.31
RL-71	0.05±0.51	40.9±1.3	7.92±1.05
PM-91	8.59±0.37	14.1±0.05	8.93±3.56
ZK-21	2.23±0.99	26.5±2.9	7.14±0.68

^aSyncroPatch 384PE with whole cell patch-clamp and multi-hole chips (4 holes per well). For Kv7.2W236L channels, EC₅₀ values could not be calculated. Kv7.2, N=488; Kv7.2W236L, N=514. Values are the mean±SD.

Overall, these findings demonstrate that the SyncroPatch is a robust and high signal-to-noise assay platform for Kv7 channels. We were able to identify a promising new agonist chemotype, exemplified by ZK-21, with an EC₅₀ value of 2.2 μM. Since ZK-21 also lost activity with the Kv7.2W236L mutant, this new lead structure is likely to occupy the same pore region S5 binding site as retigabine and the RL-series of compounds. In comparison, the PM series 2-phenylindole scaffold was only moderately active, with an EC₅₀ value of 8.6 μM for the most potent analog, PM-91. These results are in contrast to N-benzylated indole analogs such as compound 24a (Figure 2), which were previously found to be submicromolar and selective Kv7.2–4 agonists.⁴

Patch-clamp electrophysiology studies.

Whole cell patch-clamp electrophysiology studies were conducted on Kv7.2/7.3 heterotetramers, as well as Kv7.4 and Kv7.5 ion channels to further determine the selectivity of representative analogs in the potent RL-series, RL-15, RL-24, RL-56, and RL-71. In this assay, RL-56 proved to be an exceedingly potent (EC₅₀ ca. 20 nM) Kv7.2/3 agonist, with 50- to 200-fold selectivity over Kv7.4 and Kv7.5 channels (Table 3, Entry 3). In general, we consider a 5-fold difference in EC₅₀ between different channel subtypes to reflect a selectivity suitable for further lead development. Introduction of an additional fluorine group ortho to the benzylic amine and replacement of the cyclopropyl carbamate with an ethyl carbamate in RL-71 maintains submicromolar potency on Kv7.2/3 heterotetramers and introduces potential selectivity for Kv7.4 vs Kv7.2/3 and Kv7.5 (Entry 4). However, given the relatively small sample size for the experiments with RL-71, further work will be needed to verify this selectivity. Interestingly, while the pyridine-containing RL-24 is much less active as a Kv7.2/3 agonist, it has a respectably potency (EC₅₀ = 0.54 μM, Entry 3) in the Kv7.5 assay and, therefore, might serve as a potential lead structure for more selective Kv7.5 channel agonists. The weakly active RL-15 confirms our previous data showing that fluorination next to the carbamate nitrogen interferes with binding at the W236 site and negatively effects Kv7.2/3 channel agonism, but this compound also retains remarkable potency and selectivity for Kv7.4 (Entry 1).¹⁶ It is also noteworthy that our data suggest a much stronger activity of RL-56 on heteromeric Kv7.2/3 than homomeric Kv7.2 channels, implying that this analog binds preferentially to Kv7.3. A related trend with ca. 4-fold lower EC₅₀ values for Kv7.3 over Kv7.2 has previously been noted for retigabine.²³ Overall, the patch-clamp electrophysiology studies suggest that the RL-series of analogs remains a very attractive scaffold to generate highly potent and selective potassium channel modulators that could be targeted for specific channelopathies.⁸

Table 3.

EC₅₀ values for RL-15, RL-24, RL-56, and RL-71 on Kv7.2/7.3, Kv7.4 and Kv7.5 from patch-clamp electrophysiology studies.^a

Entry	Compound	EC50±SEM (μM)
		Kv7.2/3	Kv7.4	Kv7.5
1	RL-15	1.64±5.57	0.32±1.49	3.78±1.49
2	RL-24	3.36±5.54	0.66±0.33	0.54±0.14
3	RL-56	0.018±0.01	0.91±0.39	3.78±1.48
4	RL-71	0.79±0.24	0.087±0.13	1.06±2.02

^aUsing an established molecular transfection protocol,²⁴ ion channel and GFP DNA were transfected in a 2:1 ratio into Chinese Hamster Ovary (CHO) cells. These ion channels transiently expressed either heterologous K_v7.2/7.3 ion channels or homologous Kv7.4 and Kv7.5 ion channels. Five different concentrations were tested: 1 nM, 10 nM, 100 nM, 1 μM and 10 μM. N = 4–14. RL-71 was tested in a smaller sample size (N = 3–4) compared to the other compounds. Values are the mean ± SEM.

Docking studies.

To develop a structure-activity relationship model, interpret the observed activities, and assess the suitability of the Kv7.2 retigabine site for a virtual screening campaign, we docked the 22 compounds into recently resolved CryoEM structures for Kv7.2 at the retigabine site between helices 5 and 6 of the pore domain (pdb: 7CR2) and the ztz240 site in the voltage sensor domain (pdb: 7CR1) as a control. To begin, we used UCSF Dock 3.7, which is suitable for ultra-high throughput virtual screening and leverages a fixed receptor conformation and reference ligand to expedite ligand sampling at an average rate of 1 sec/cpd.²⁵ Typically, a docking site is prepared by optimizing the discrimination of active ligands from known or likely inactive molecules. To recapitulate the binding mode of retigabine at its site, we increased the polarity of the backbone carbonyl group of the R305 residue that interacts with the aniline moiety from −0.5 to −0.9 and the polarity of backbone H from 0.248 to 0.648 to maintain the overall charge.²⁶ Interestingly, we found that RL-56, RL-81, retigabine, and RL-15 were the top scoring ligands, and, overall, we detected a significant spearman rank correlation of 0.46 (p<0.05) between the measured ΔV₅₀ value and the dock energy. In contrast, we found no significant correlation at the ztz240 site with a rank correlation of 0.19 (Figure 5).

Figure 5.

Docking studies at the K_v7.2 retigabine (A) and ztz240 (B) binding sites. Dots represent experimentally tested ligands in shown in Figure 3 with Docking Energy <−10, where those with experimental Kv7.2 ΔV₅₀ <−20 mV at 5 μM concentration are labeled and shown in red.

Since ZK-21 showed a W236-sensitive activation (we consider < −10 mV to be the threshold of activity to overcome the measurement error) but did not bind at either site with the retigabine templated Dock setup. Therefore, we hypothesized that ZK-21 may bind at the retigabine site but adopt an alternative binding mode. To investigate further, we docked the ZK-series ligands sampling 10,000 trajectories of the flexible-receptor/flexible-ligand RosettaLigand protocol,²⁷ taking an average of 177 h per target ligand complex. Interestingly, we identified a binding mode for ZK-21 where the tetrahydroquinoline packs against W236, supporting the benzyl carbamate to extend perpendicular to retigabine into the core of the pore (Figure 6A–D). In this binding mode, the ZK-21 carbamate is analogous to the retigabine aniline in interacting with the S303 hydroxyl group, and the ZK-21 methoxy aligns with the retigabine carbamate. Initially, we used OpenEye Omega to identify ligand conformations, which sampled the tetrahydroquinoline of ZK-21 in an equatorial half chair and the other ZK-series compounds into a half-boat ring pucker (Figure 6E–H). For ZK-94, the phenyl substituent that replaces the isopropyl group in ZK-21 can form an intramolecular cation-π interaction with the carbamate nitrogen, corkscrewing it further into the channel and rotating the carbamate away from engaging S303. As a result, ZK-21 achieved the lowest energy among all ligands tested. All ZK-series compounds showed well-formed folding funnels in our simulations, with ZK-21 being the best overall when quantified by the Pnear score. To test if ZK-94 and the other ZK-compounds could adopt more favorable binding conformations with alternate ring puckers, we used CORINA to explicitly sample equatorial and axial half-chair ring pucker conformations in separate RosettaLigand docking runs. Accordingly, we found that both ZK-21 and ZK-94 could adopt similar binding modes in the equatorial half-chair conformation, with ZK-94 more energetically favorable. However, compared with ZK-21 sampled by Omega, ZK-94 does not directly engage the hydroxyl group of S303 with its benzyl carbamate and we did not observe canonical ring-stacking geometries with W236. Since Rosetta does not account for ligand strain, we evaluated the empirical maximum bond strain derived from the Cambridge Structural Database.²⁸ We found that for the CORINA equatorial half-chair top scoring poses, ZK-21 was more stable than ZK-94, consistent with our assessment that ZK-94 prefers the half-boat conformation originally sampled by Omega. Next, we used RosettaLigand to dock retigabine, ZK-21, and ZK-94 into the W236L mutant conformation, by exercising the mutation and using Rosetta to repack and minimize the starting conformation. We found that for retigabine this yields slightly improved interface energy scores, but without the tryptophan to stack against the central ring system, RosettaLigand oriented the aniline nitrogen between S5 and S6, leaving it uncoordinated rather than engaging S303. Both ZK-21 and ZK-94 scored worse in the mutant conformation.

Figure 6.

ZK-21 binding mode predicted by RosettaLigand. (A) Top-down view of the Kv7.2 homotetramer clipped at the retigabine binding site between the S5 and S6 helices of the pore domain. (B) Overlay of the experimental retigabine and predicted ZK-21 binding modes and (C,D) schematics of the receptor contacts. ZK-21 is predicted to bind perpendicular to retigabine allowing the benzyl carbamate to fill the tunnel towards the pore core. Binding modes for (E) ZK-21, (F) ZK-94, (G) ZK-29, and (H) ZK-16.

In conclusion, our synthetic chemistry and screening campaign identified 6 new compounds that exhibited differential enhancing activity between WT and W236L mutant channels as pore domain ligands on Kv7.2 potassium ion channels. Whole cell patch-clamp electrophysiology studies provided Kv7.2. Kv7.2/3, Kv7.4, and Kv7.5 selectivity data for selected analogs. Our results validate the SyncroPatch platform for high-throughput screening and establish new SAR trends. Specifically, in addition to selective Kv7.2, Kv7.2/3, Kv7.4. and Kv7.5 agonists, we identified a distinct new Kv7 channel agonist chemotype, ZK-21. Furthermore, docking studies identified a novel binding mode for ZK-21 that positions its benzyl carbamate moiety perpendicular to retigabine towards the pore core of the channel and provides specific residue interactions that may explain its improved activity over the structurally related ZK-94. Finally, the ability for Dock to discriminate retigabine, the RL-series, and the ZK-series compounds suggests this is a promising site for interrogation in future virtual screening campaigns.