Biaryl sulfonamide motifs up- or down-regulate ion channel activity by activating voltage sensors

Many types of disease caused by increased electrical excitability, such as epilepsy, cardiac arrhythmia, and pain, are sometimes caused by ion-channel mutations and can, in some cases, be treated by ion-channel modulators (Catterall, 2014; Zamponi et al., 2015). The majority of existing pharmaceutical drugs that function as ion-channel modulators block the ion-conducting pore and thus decrease the channel current (Hille, 1977; Ragsdale et al., 1994). It is mechanically simple for a compound to block a channel by plugging the pore and thereby preventing ion flux (Hille, 1977; Ragsdale et al., 1994; Zhou et al., 2001). In contrast, a few drugs and drug-like compounds increase the current (Wuttke et al., 2005; Peretz et al., 2010; Li et al., 2013; Ottosson et al., 2015; Salari et al., 2018). However, to increase a current, another mechanism must be used to keep the channel open. For instance, retigabine, an antiepileptic drug that opens the voltage-gated potassium channel K_V7.2, binds near the gate that opens and closes the channel and bends the gate open (Wuttke et al., 2005). Alternatively, some compounds act on the channel’s voltage-sensing mechanism to activate the voltage sensor of the channel’s voltage-sensor domains (VSD) and thereby indirectly pull the gate open.

VSD-acting compounds can enter the VSD of K_V7-type channels from the extracellular solution and bind at the center, between the four transmembrane segments, S1–S4, to keep the voltage sensor S4 in an activated, up, state (Peretz et al., 2010; Li et al., 2013). Alternatively, hydrophobic and negatively charged polyunsaturated fatty acids (PUFAs) or resin acids act from the lipid-facing part of the VSD of K_V1- or K_V7-type channels to electrostatically attract the positively charged voltage sensor S4 to activate and open the channel (Börjesson et al., 2008, 2010; Börjesson and Elinder, 2011; Ottosson et al., 2014, 2015, 2017; Liin et al., 2015, 2016a; Yazdi et al., 2016; Elinder and Liin, 2017; Salari et al., 2018; Silverå Ejneby et al., 2018). To specifically control excitability in different cell types, new types of ion-channel modulators are needed; new modulators based on new scaffolds can potentially be developed into either openers or closers and can potentially be designed to act on specific channels.

To search for new ion-channel modulator scaffolds, we performed an electrophysiological screen of almost 18,000 small-molecule compounds from a diverse range of compound libraries. We found that sulfonamides, a sulfonyl group connected to a nitrogen (Fig. 1 A), formed a family of K_V-channel–opening compounds. The minimal requirement for the active compounds was a biaryl-sulfonamide motif (e.g., Fig. 1 B, which we will refer to as the short biaryl-sulfonamide motif). Our data suggest that these molecules bind to the VSD of the voltage-gated Shaker K_V channel to keep the voltage sensor S4 in an activated, up, state, and the pore gate open.

The Shaker H4 channel (accession no. NM_167595.3 in the Bluescript II KS[+] plasmid; Kamb et al., 1987), with removed N-type inactivation (ShH4IR; Hoshi et al., 1990), referred to as the WT Shaker K_V channel, was used throughout this work. In addition, a modified, so-called 3R Shaker K_V channel was used, in which two introduced positively charged arginines (M356R and A359R), in addition to one native arginine (R362), made the channel more sensitive to PUFAs and resin acids (Ottosson et al., 2014). The 3R Shaker K_V channel was also expressed in a CHO-K1 stable cell line (Ottosson et al., 2015). To measure gating currents, we used the W434F mutation in the WT Shaker K_V channel background (Perozo et al., 1993). Point mutations in the Shaker K_V channel were introduced using the QuikChange Site-Directed Mutagenesis kit (Stratagene) and verified by sequencing, as previously described (Börjesson and Elinder, 2011).

Compounds were selected from several chemical libraries: Acids (Vitas-M Laboratory); Consortium set (Specs)*; Drug Like (Enamine)*; Hit Finder (Maybridge)*; Ion Channel Ligand and Nuclear Receptor Ligand (Enzo Life Sciences); Known Drugs (Prestwick Chemical); LCBKI Primary Screening Set (Chemical Biology Consortium Sweden)*; Natural derivatives (TimTec); NIH Clinical Collection (NIH); Peptidomimetics (ChemDiv Inc.)*; and Tocriscreen Plus (Tocris Bioscience). Apart from LCBKI, provided by the Chemical Biology Consortium Sweden, and the NIH clinical collection, provided by the Libraries Roadmap Initiative, all compound libraries were commercially available. Whenever a selection was made (*), a chemist was instructed to search for small, hydrophobic, and partially charged compounds.

CHO-K1 cells were grown to semiconfluence (maximum 75%) in T-150 Costar flasks (Corning) at 37°C (final 12–15 h before experiment at 30°C) in a humidified environment (5% CO₂), in DMEM/F12+ Glutamax, supplemented with 10% FBS, G418 sulfate, and 1% MEM nonessential amino acids. Just before use, the cells were washed twice with prewarmed Mg- and Ca-free PBS (Invitrogen) and detached with 3 ml TrypLE (Gibco) for 4 min at 30°C. Cells were then detached from the bottom of the flask by gentle tapping, and 7 ml (Mg- and Ca-containing) Dulbecco’s PBS (D-PBS) was added and aspirated into a centrifuge tube before centrifugation (50 g for 3 min). The resulting supernatant was discarded, and the cell pellet was gently resuspended in 5 ml D-PBS. A 0.1-ml aliquot of cell suspension was removed, and the cell concentration was determined using a Burker chamber. After the cells had been counted, the cell suspension was adjusted with PBS to reach a cell concentration of 1.3 million/ml and finally added to the IonWorks cell boat.

Assay-ready plates were prepared at the High Throughput Center at Karolinska Institutet. All compounds were dissolved in 100% DMSO to a final stock concentration of 10 mM. 100 nl of the 10-mM DMSO solutions of compounds and controls were transferred to low-volume 384-well polypropylene Greiner plates by acoustic dispensing (Echo 550; Labcyte). For single-concentration measurements, compounds were placed in columns 3–22, and DMSO controls and a known modulator were placed in columns 1, 2, 23, and 24. The assay-ready plates were heat-sealed using a thermal microplate sealer (PlateLoc; Agilent) and stored overnight at 4°C. On the day of the experiment, the compound plates were diluted by adding 30 µl PBS to each well and mixed using a Biomek Fx^P Lab Automation workstation (Beckman Coulter), giving a compound plate concentration of 30.3 µM (0.33% DMSO; three times their final test concentration). Before the experiment, all plates were centrifuged (1,000 g for 2 min) to remove air bubbles. In one library, two compounds were tested per well, and if the double compound effect was significant, the two compounds were retested separately.

Recordings were performed in CHO-K1 cells using the PPC mode of the IonWorks Quattro automated patch-clamp system (Molecular Devices). The operation protocol for this system has been previously described (Schroeder et al., 2003; Bridgland-Taylor et al., 2006; Harmer et al., 2008). The intracellular solution contained (in mM) 100 potassium-glutamate, 40 KCl, 3.2 MgCl₂, 3 EGTA, and 5 HEPES, pH 7.25–7.3 with KOH, and the extracellular solution was D-PBS (Gibco) and contained (in mM) 138 NaCl, 2.7 KCl, 1.5 KH₂PO₄, 8 Na₂HPO₄, 0.9 CaCl₂, 0.5 MgCl₂, and 5.5 glucose. The tested compounds were added to the extracellular solution. To establish the whole-cell perforated patch configuration, 25 mg/ml amphotericin B was added to the intracellular solution, resulting in a final concentration of 150 µg/ml. After attaining the whole-cell configuration, the voltage protocol was applied once to obtain a control recording (prescan) and again after compound application (post-scan) to measure the potentiation of potassium currents. During the compound incubation (∼2 min), that is, between the pre- and post-scans, there was no clamping of the membrane potential. Before the test voltage protocol, the cells were held at ‒80 mV for 20 s. Currents were evoked by a three-pulse voltage train (Fig. 1 C): pulse 1, a 150-ms square-shaped step to +10 mV; pulse 2, a 150-ms square-shaped step to +30 mV; and pulse 3, a 200-ms linear ramp from −80 mV to +50 mV directly followed by a 100-ms step to +50 mV. The cells were kept at ‒80 mV for 300 ms between each step. Wells with a seal resistance of <20 MΩ or with a current <0.05 nA were not included in the analysis. The current signal was sampled at 10 kHz. All recordings were made at room temperature (∼21°C). To minimize compound well carryover, we included a wash step in the IonWorks protocol with ethanol and PBS in between the addition of compounds. Moreover, in between the experiments, we also ran an IonWorks cleaning protocol with ethanol and PBS.

All animal experiments were approved by the local Animal Care and Use Committee and followed international guidelines. Generally, 50 nl WT or mutated Shaker K_V channel cRNA (50 pg) were injected into each oocyte using a Nanoject injector (Drummond Scientific). X. laevis surgery, oocyte isolation, and oocyte storage after RNA injection followed our previously described protocol (Börjesson et al., 2010). Injected oocytes were kept at 8°C for 1–6 d before electrophysiological recordings were performed. All chemicals were supplied by Sigma-Aldrich if not stated otherwise.

All manual electrophysiological recordings were performed 1–6 d after cRNA injection by the two-electrode voltage clamp technique performed at room temperature (20–23°C), using a CA-1B amplifier (Dagan). Signals were lowpass-filtered at 5 kHz and digitized by a Digidata 1440A converter (Molecular Devices). The amplifier’s leak compensation was used, and Clampex 10.5 software (Molecular Devices) was used to obtain data and create voltage protocols. The extracellular control solution (1K) contained (in mM) 88 NaCl, 1 KCl, 15 HEPES, 0.4 CaCl₂, and 0.8 MgCl₂, pH adjusted to 7.4 by NaOH (reaching a sodium concentration of ∼100 mM). Compounds were dissolved to 100 mM in 99.5% EtOH and stored at ‒20°C. Compounds were diluted in the extracellular control solution to the desired test concentration immediately before experiments. To measure closing kinetics, we used a high potassium–containing solution (100K) in which NaCl in the 1K control solution was replaced with KCl and pH set with KOH, yielding a potassium concentration of ∼100 mM.

The extracellular control solution was added to the bath with a gravity-driven perfusion system. Sulfonamides were added to the bath using a Mini-Peristaltic Pump II (Harvard Apparatus). Glass microelectrodes were pulled from borosilicate glass, filled with 3 M KCl, and had a resistance of 0.5–2 MΩ. The holding voltage was set to ‒80 mV. Currents were evoked by 100-ms-long square-like steps ranging from −80 up to +50 mV (WT) and +70 mV (3R) in steps of 5 mV. The pulses occurred at a frequency of 0.2 Hz. To study gating currents, the linear membrane capacitance was compensated for by the use of the amplifier’s capacitance compensation for a 10-ms voltage pulse between 0 and ‒10 mV.

The manual electrophysiological data were processed with Clampfit 10.4 (Molecular Devices) and GraphPad Prism 5 (GraphPad Software).

Modulation of the K_V-mediated current was assessed by dividing the post-scan steady-state K_V current (for the last 10 ms of each pulse) by the respective prescan K_V current for each well. For the quantitative analysis presented in this paper, we used the data from +10 mV. In control solution (at +10 mV) during the experiment, there was a consistent rundown of the relative current by −20 ± 7% (mean ± SD). Therefore, all experimental data (post-scan/prescan) between 0.6 and 1.0 (mean ± 3 SD) were regarded as without effect.

Conductance G(V) was calculated as

where I_K is the mean current from the steady-state phase at the end of each 100-ms pulse, V is the membrane voltage, and V_rev is the reversal potential for K⁺ (set to ‒80 mV). To analyze gating currents, a leakage current correction was performed offline. This was achieved by using Clampfit software to eliminate the steady-state leakage current at the end of the ON pulse or the OFF pulse, after the current had reached steady state. The gating charge was calculated as the area under the curve from 1.5 ms after onset of the OFF pulse (to exclude the remaining capacitive current) to 100 ms after onset of the OFF pulse. These data were fitted to a Boltzmann-inspired equation:

where A is the amplitude of the curve, V is the absolute membrane voltage, V_1/2 is the midpoint, s is the slope (Börjesson et al., 2008), and n is an exponent. For most of the ion-current analyses, n was set to 4 to allow a good fit to the steep foot of the G(V) curve. For the gating current analyses, n was set to 1. The compound-induced G(V) shift (ΔV_G(V)) was determined at the 10% level of the maximum conductance in control solution (Börjesson et al., 2008).

Data for concentration- and pH-dependent G(V) shifts (ΔV_G(V)) in oocytes were fitted to a concentration–response equation:

where ΔV_G(V),max is the maximal shift, c_1/2 is half-maximal effective concentration (or pK_a value in the pH experiments), and c is the concentration.

Marvin was used to draw chemical structures (Marvin 16.12.9, 2016, ChemAxon). The logarithm of the acid dissociation constants, pK_a, for the ionic forms of the compounds were calculated using the Marvin Calculation plugin. The pH range for calculating microspecies distribution (%) was set to 0–14 and temperature to 298°K, and the pK_a was obtained from the global mass and charge conservation law (Macro mode). The ionic strength was considered to be 0.1 M. The octanol-water partitioning coefficient, log P, for uncharged compounds was calculated using the Marvin Calculation plugin. The consensus model in ChemAxon was used for calculations, and electrolyte concentrations were set to 0.1 M for Cl⁻, Na⁺, and K⁺.

Mean values are expressed as mean ± SEM. One-way ANOVAs, followed by Dunnett’s multiple comparison test, were used to compare the effects of sulfonamides on each mutant to the corresponding effects on the WT Shaker K_V channel. Two-tailed t tests were used to compare the effects of sulfonamides on the WT Shaker K_V channel to a hypothetical value. P < 0.05 was considered significant for all tests.

The RosettaLigand docking (Lemmon and Meiler, 2012) software (v2016.16) was used to dock the Cmpd 5 and Cmpd 6 ligands to the Shaker K_V channel VSD. A homology model for the structure of the Shaker K_V channel was constructed using HHpred (Hildebrand et al., 2009) and Modeller 9v13 (Sali and Blundell, 1993) with the K_V1.2–K_V2.1 paddle chimera channel (PDB ID: 2R9R) as template. Alignment for the homology modeling is provided as a supplemental file. Ligand conformers were constructed using OMEGA from OpenEye Scientific (Hawkins et al., 2010), with default parameters resulting in 201 and 30 conformers for Cmpd 5 and Cmpd 6, respectively. The ligand was placed in an approximate binding pocket suggested by a recent study on Na_V1.7 channels (Ahuja et al., 2015); before each docking, the ligand was randomly rotated and translated between 0 and 2 Å uniformly. The docking protocol samples all the different conformers of the ligand, allowing the side-chains and backbone of the protein receptor close to the ligand (<6 Å) to adapt to the ligand. In total, 49,277 and 42,004 docking poses were generated for Cmpd 5 and Cmpd 6, respectively. In addition, 40,000 docking poses were generated for Cmpd 6 in the deprotonated state. The 5% lowest-energy docking poses were ranked by the interface delta energy, defined as the total energy of the complex with ligand bound minus the total energy of the complex and ligand alone.

Files, scripts, and instructions on how to reproduce the docking are provided in the datasets, as well as a pymol session of the best scoring docking poses.

To search for compounds modulating a voltage-gated K channel, we screened almost 18,000 compounds on a specially designed Shaker K_V channel expressed in Chinese hamster ovary (CHO) cells (Ottosson et al., 2015). This channel was developed to be sensitive to negatively charged hydrophobic compounds that electrostatically activate the voltage sensor and thereby open the ion channel. Two extra (positively charged) arginines were introduced at the top of the voltage sensor S4 (M356R/A359R, called the 3R channel because there are three critical arginines in positions 356, 359, and 362; Ottosson et al. [2014]). The rationale for using this channel was to identify compounds for further investigation and refinement. However, this mutant does not exclude the detection of compounds acting on other sites via other mechanisms. The compounds were selected from several compound libraries (see Materials and methods), and the selection criteria were partly hydrophobic compounds with a negative charge. The compounds’ effects (at a concentration of 10 µM) were explored using a high-throughput electrophysiological patch-clamp technique (IonWorks). As a readout, we used the steady-state current in compound solution relative control solution, measured during the last 10 ms of a 150-ms voltage step to +10 mV (from a holding voltage of −80 mV; pulse 1 in Fig. 1 C).

The current in the compound solution relative to the current in the control solution for all tested compounds is shown in Fig. 1 D. Note that the number of experiments is slightly lower than 18,000 because some compounds were tested two at a time; if the effect was significant, both compounds were rerun individually. Because of the experimental rundown during the experiment (see Materials and methods), all data between 0.6 and 1.0 was regarded as without effect (shaded area in Fig. 1 D). Almost 20% of the compounds caused a significant current reduction (<0.6), and 6.5% of all data showed a significant current increase (>1.0), but it should be noted that the actual fraction is slightly lower because this data also include reruns and confirmatory runs. To reduce the number of false hits, we considered only compounds with a relative current of 1.2 as clear K-channel openers (dashed line in Fig. 1 D) and 0.4 as clear K-channel blockers (dotted line in Fig. 1 D). In the screen, 243 compounds (1.3% of all tested compounds) were classified as clear K-channel openers, and 140 as clear blockers (0.8% of all tested compounds). 26 of the clear openers (11% of the hits; Table 1) were sulfonamides characterized by the motif in Fig. 1 A, whereas none of the clear blockers had this motif.

The sulfonamide family was selected for closer analysis. Most of the sulfonamide compounds (19 of 26) were tested in repeated screens, and seven of them in triplicate. Libraries, names, structures, and data from the screen of identified sulfonamides are listed in Table 1. Most of the compounds have one ring structure at each end of the sulfonamide motif. One compound has one atom between the two rings (the nitrogen trapped in the ring), 16 have two atoms (O and N), one has three atoms, three have four atoms, one has five atoms, one has six atoms, and three have a free end with just one ring connected to the sulfonamide motif. Thus, 16 of the 26 compounds have a characteristic short biaryl-sulfonamide motif with just the two-atom sulfonamide motif between the two rings (Fig. 1 B). Interestingly, of the ten best compounds (mean channel-opening effect of >1.25; Table 1), nine had a biaryl motif and one had just an additional atom in between. In the less potent group (mean channel-opening effect <1.25), 7 of 16 compounds had a short biaryl-sulfonamide motif, whereas 9 had other types of sulfonamide structures. Thus, the short biaryl-sulfonamide motif could be the key to the channel-opening effect. However, it should be noted that there was variability between the different runs, and some of the compounds were run only once.

In the screen described above, compounds were tested on the 3R Shaker K_V channel in an attempt to find compounds electrostatically acting on the outer end of S4, as has been described for PUFAs and resin acids (Börjesson and Elinder, 2011; Ottosson et al., 2014, 2015, 2017; Liin et al., 2015; Yazdi et al., 2016; Silverå Ejneby et al., 2018). In the next step, we explored 11 selected sulfonamide compounds on the WT Shaker K_V channel expressed in X. laevis oocytes at 10 and 100 µM at pH 7.4. The compound selection was designed to choose diverse types of compounds, with respect to both molecular structures and effects in the screen.

Cmpd 5 at 100 µM clearly increased the WT Shaker K_V channel current at −30 mV (Fig. 2 A). The effect reached saturation within 2 min. After washout, the effect recovered to control level within ∼5 min. This time course of the effect was typical for the compounds, and we made sure to reach saturation in all experiments. At −35 mV, 100 µM of Cmpd 5 increased the current fivefold (Fig. 2 B) by shifting the conductance-versus-voltage curve, G(V), by −10.8 ± 0.6 mV (n = 4), with only minor effects on the maximum conductance (Fig. 2 C). Other typical examples of the effect of compounds on the G(V) curve are shown in Fig. 2 (D–F). Of the 11 tested compounds, 5 displayed clear effects of approximately equal magnitude (blue symbols in Fig. 2 G). One compound had small but significant effects at 100 µM (black symbols in Fig. 2 G). Five of the compounds displayed no significant effects on the WT Shaker K_V channel (red symbols in Fig. 2 G). The reason for this lack of effect, despite effects in the primary screen, could be (a) the differing expression system (X. laevis oocytes vs. CHO cells), (b) differing channels (WT vs. 3R channel), or (c) false-positive effects in the initial screen. However, there is a clear pattern on the WT Shaker K_V channel that resembles the pattern from the initial screen on the 3R Shaker K_V channel; all effective compounds share the short biaryl-sulfonamide motif (Fig. 1 B), whereas all compounds lacking this motif (Cmpds 9–11) had no effect. This suggests that the WT Shaker K_V channel has a distinct binding site for the short biaryl-sulfonamide motif. To explore the mechanism of effective sulfonamides further, we selected two of the active compounds for a more detailed characterization. Cmpd 5 (Fig. 2 H) had the largest effect at 100 µM, and Cmpd 6 (Fig. 2 H) had the largest effect at 10 µM.

To further test the hypothesis that a short biaryl-sulfonamide motif is required to open the Shaker K_V channel, we tested the antidiabetic drug glibenclamide (also called glyburide), which closes the ATP-sensitive K channel, depolarizes pancreatic β cells, and increases the release of insulin (Ashcroft and Ashcroft, 1990). This drug was in fact one of the 26 compounds identified in the original screen (Table 1), but because of its elongated biaryl-sulfonamide motif (Fig. 3 A), we did not predict it to open the WT Shaker K_V channel. Consistent with this hypothesis, 30 µM of the drug had no effect on the Shaker K_V channel (ΔV_G(V) = −0.3 ± 0.2 mV, n = 3; Fig. 3, A and B). To test if it is the two aromatic rings in combination with the amine group and not the sulfonic group that opens the channel, we tested another ion-channel modulating drug in clinical use, the antiepileptic drug retigabine, which opens the voltage-gated K_V7.2/7.3 M-channel and dampens excitability (Main et al., 2000; Rundfeldt and Netzer, 2000; Wickenden et al., 2000; Wuttke et al., 2005). Retigabine, at 30 µM, had no effect on the WT Shaker K_V channel (ΔV_G(V) = 0.0 ± 0.0 mV, n = 3; Fig. 3, C and D).

Cmpd 5 had no effect on the channel-opening kinetics at +60 mV (Fig. 4 A); the opening time constant in compound solution relative control solution was 0.95 ± 0.02 (n = 5; ns). In contrast, Cmpd 5 clearly slowed down the channel-closing kinetics at −60 mV (Fig. 4 B) by a factor of 4.0 ± 0.7 (n = 4; P < 0.05). The selective effect on channel closing by Cmpd 5 is shown in Fig. 4 C. Cmpd 6 had a very similar profile to Cmpd 5. Cmpd 6 had no effect on the channel-opening kinetics at +60 mV (Fig. 4 D); the opening time constant in compound solution relative control solution was 1.17 ± 0.11 (n = 5; ns). In contrast, Cmpd 6 clearly slowed down the channel-closing kinetics at −60 mV (Fig. 4 E) by a factor of 6.2 ± 1.5 (n = 4; P < 0.05). The selective effect on channel closing by Cmpd 6 is shown in Fig. 4 F.

To test if the compound also affected the voltage sensor movement (directly or indirectly), we measured gating currents. The W434F Shaker K_V channel mutant does not conduct ions, so isolated gating currents can easily be measured (Perozo et al., 1993). The ON-gating currents, generated when the oocyte is pulsed from a negative holding voltage of −80 mV to more positive voltages, showed a characteristic amplitude and voltage dependence. Cmpd 5 had no effect on the ON-gating currents (Fig. 5 A). The small effects during the first milliseconds are because of alterations in the capacitance compensations during the experiment. In contrast, Cmpd 5 clearly slowed down the OFF-gating currents at −80 mV (from different prepulse voltages; Fig. 5 B) by a factor of 4.8 ± 0.1 (n = 3; P = 0.013). The steady-state gating-charge movement at different voltages, Q(V), was measured by integrating the OFF-gating current (as shown in Fig. 5 B) and plotted versus the prepulse voltage (Fig. 5 C). Cmpd 5 shifted the Q(V) curve by −7.6 mV ± 1.1 mV (n = 3, P = 0.022) but did not affect the maximum gating charge (−3 ± 2%, n = 3, ns). Cmpd 6 had similar effects to Cmpd 5. The ON-gating currents were not affected (Fig. 5 D). The OFF-gating currents were clearly slowed down (Fig. 5 E; by a factor of 2.7 ± 0.2 at −80 mV; n = 7; P < 0.0001). The Q(V) curve (Fig. 5 F) was shifted by −3.4 mV ± 0.4 mV (n = 7, P = 0.0003), but the maximum gating charge was not affected (+1 ± 4%, n = 4, ns).

In summary, neither Cmpd 5 nor Cmpd 6 had any effects on the ion-channel opening kinetics or the ON-gating charge movements (Fig. 5 G). Both Cmpds 5 and 6 clearly slowed down channel closure and return of gating charges (Fig. 5 H). Both Cmpds 5 and 6 clearly shifted G(V) and Q(V) in a negative direction along the voltage axis. Altogether, this suggests that the short biaryl-sulfonamide compounds trap the ion channel in an open state with the voltage sensor in an activated state.

Cmpd 5 dose-dependently increased the current of the WT Shaker K_V channel by shifting the G(V) curve in a negative direction along the voltage axis (Fig. 6 A). As shown by Eq. 3 in Materials and methods, the maximum shift (ΔV_G(V),max) was −14.8 mV, and the concentration causing 50% shift (c_1/2) was 35 µM. The nitrogen of the compound is protonable, and the molecule is expected to be uncharged at low pH and negatively charged at high pH. The calculated pK_a value is 6.85, and an alteration in extracellular pH is expected to alter the charge of the compound around neutral pH if the compound experiences a watery surrounding. To test whether the charge is important for the G(V) shift, we therefore studied the effect of different pHs. Cmpd 5 had no effect at pH 5.0 or 6. At pH 7.4 the effect peaked, but surprisingly, it decreased at higher pH (Fig. 6 B). We do not know the reason for this decreased effect at high pH, but the data are consistent with the idea that the negative charge of the compound is important for the effect. Like Cmpd 5, Cmpd 6 dose-dependently increased the current of the WT Shaker K_V channel by shifting the G(V) curve in a negative direction along the voltage axis (Fig. 6 C). At a concentration of 30 µM, the compound showed a saturated voltage shift. The maximum shift (ΔV_G(V),max) was −10.0 mV, and the concentration causing 50% shift (c_1/2) was 9.4 µM. The calculated pKa value was 7.69, and the experimentally obtained pK_a value was 6.7 (Fig. 6 D; Eq. 3), which suggests that the molecule (or the binding site) is almost fully deprotonated at a neutral pH value. To summarize, short biaryl sulfonamides bound to the channel caused G(V) shifts up to ‒15 mV, 10–30 µM of the compounds caused 50% of the maximum shift, and the negatively charged form of the compound (or a deprotonated binding site) is possibly required for these effects.

Where is the site of action for the short biaryl-sulfonamide compounds? Negatively charged PUFAs and resin acids bind in the lipid bilayer, close to the gating charges of the VSD, and, from this position, they electrostatically attract the gating charges to pull the channel open (Börjesson et al., 2008; Börjesson and Elinder, 2011; Ottosson et al., 2014, 2015, 2017; Liin et al., 2015, 2016b; Yazdi et al., 2016; Silverå Ejneby et al., 2018). Because a negative charge is also probably required for the short biaryl sulfonamides to open the channel, we hypothesized that they share a binding site and act via the same mechanism as PUFAs and resin acids. A characteristic feature for compounds acting via this site and mechanism is that the introduction of two positive charges at the extracellular end of the voltage sensor S4 (facing the lipid bilayer, i.e., the 3R, or M356R/A359R, channel) increases the drug-induced G(V) shift by a factor of ∼3 (Ottosson et al., 2014, 2015). However, the 3R mutation (M356R/A359R/R362R; Table 2) did not alter the shift for either Cmpd 5 or Cmpd 6 (Fig. 7, A–C), thus refuting this hypothesis.

Instead, we hypothesized that the compounds bind to the top of the VSD in the pocket formed at the center between the transmembrane segments S1, S2, S3, and S4 of the VSD. This site accommodates other types of ion-channel modulators (Peretz et al., 2010; Li et al., 2013; Ahuja et al., 2015). To test whether the biaryl sulfonamides could bind to this site, we performed in silico docking of Cmpd 5 and 6. Both molecules bound with favorable binding energies (Fig. 8, B and D), with an energy gap similar to that found for known binders of similar size (Meiler and Baker, 2006). Cmpd 5 had a well-defined binding pose, with all of the top 10 binding poses more or less identical, differing only slightly in their binding energies (Fig. 8 B). In this binding pose, Cmpd 5 is interacting with residues Y323, F324, and T326 at the extracellular end of S3 and the gating charges R365, R368, and R371 of S4 (Fig. 8 A). The Cmpd 6 binding pose formed two distinct clusters, one with the difluorophenyl ring pointing outward and interacting with F324 or P273 (Cluster 1), and one with the difluorophenyl ring pointing inward with the fluorines interacting with the gating charges R365 and R368 (Cluster 2); in both clusters, the benzene group was interacting with Y323 (Fig. 8 C). Because Cmpd 6 is more active at high pH and its nitrogen is most likely deprotonated at neutral pH (Fig. 6 D), dockings with a deprotonated Cmpd 6 were also performed. These dockings generated similar binding poses as with the neutral nitrogen, protonated state, but with slightly more poses in cluster 1 compared with cluster 2 among the top 10. The deprotonated Cmpd 6 also exhibits more favorable binding energies compared with the protonated Cmpd 6 (Fig. 8 D), in agreement with the higher activity of Cmpd 6 at higher pH (Fig. 6 D).

To verify the suggested binding poses of Cmpd 5 and Cmpd 6, we mutated the residues suggested to be important for the binding. Table 2 shows the electrophysiological characteristics of these mutations. The Cmpd 5–induced G(V) shift was decreased by ∼50% by the Y323C, F324C, and T326C mutations (Fig. 7, C and D). The Cmpd 6–induced G(V) shift was also reduced by ∼50% by the Y323C mutation (Fig. 7 C). Both the Cmpd 5– and Cmpd 6–induced G(V) shifts were completely abolished when all three residues were mutated at the same time, Y323C/F324C/T326C (Fig. 7, C and E), thus supporting the hypothesis that these residues are important for the Cmpd 5 and 6 effects. To test if the triple mutation Y323C/F324C/T326C affected the affinity to a binding pocket, we also measured the effect at different concentrations. Data are consistent with a decreased affinity by a factor of 9 for Cmpd 5 (Fig. 7 F) and by a factor of 15 for Cmpd 6 (Fig. 7 G).

To test if the (probably) negatively charged biaryl sulfonamide electrostatically attracts the positive gating charges in the gating canal to pull S4 into an up position and keep the channel open, we tested the effect of neutralization of a single gating charge, R365C. However, this single mutation was not enough to prevent the G(V)-shifting effect of the compounds (Fig. 7 C).

In the present study, we have identified the short biaryl-sulfonamide motif as a common motif for several openers of the Shaker K_V channel. No sulfonamide compounds with longer (or shorter) motifs between the two rings opened the Shaker K_V channel. The pH experiments suggested that the negatively charged form of the compound is the active variant to facilitate opening of the channel (or alternatively that the compound is more likely to bind to a deprotonated form of the binding site). Our data suggest that the biaryl sulfonamides bind to the extracellular central pocket of the Shaker K_V channel VSD. The bound compounds keep the channel in an open state, most likely by keeping S4 in an activated up state, preventing downward movement.

This central VSD site is different from what has been found for PUFAs and resin acids, but the chemical properties of the compounds are also significantly different. However, in contrast with the proposed binding site for the biaryl sulfonamides, in the extracellular water-facing crevice of the VSD, these compounds bind at the interface between the lipid bilayer and the VSD. These different binding sites are reflected in the different chemical properties of the compounds. The lipid-facing PUFAs and resin acids have LogP values between 5.5 and 6.5 (Ottosson et al., 2017), whereas the supposedly water-facing biaryl sulfonamides have lower LogP values (Cmpd 5, LogP = 4.35; Cmpd 6, LogP = 4.13). Similarly, there is a large difference between calculated and experimentally obtained pK_a values. PUFAs and resin acids have calculated pK_a values ∼3.5–4.5, but their functional pK_a values are ∼3 pH-steps higher (Ottosson et al., 2017), reflecting a binding site in an environment different from water. In contrast, the biaryl-sulfonamide compounds studied in the present investigation have pK_a values around 7, similar to their functional pK_a values, reflecting a binding site in a water environment.

The negatively charged variant of PUFAs and resin acids open both the WT Shaker K_V channel and the 3R Shaker K_V channel, but the 3R channel to a larger extent, and the positively charged residues at the top of S4 are critical for the effect of the compounds (Börjesson et al., 2008, 2010; Börjesson and Elinder, 2011; Ottosson et al., 2014, 2015, 2017; Silverå Ejneby et al., 2018). The G(V)-shifting effect caused by the biaryl sulfonamide compound bound to its site is not sensitive to positively charged residues added to the extracellular end of S4, as tested using the 3R Shaker K_V channel. This reinforces the view that the biaryl sulfonamide and resin acid/PUFA sites are different from each other. In the molecular docking, the charge of the biaryl-sulfonamide compound electrostatically interacts with residue R365 (the second arginine) of S4. However, when we neutralized R365 to a cysteine (R365C), the effect of the biaryl sulfonamides was not reduced. One possible explanation is that the biaryl-sulfonamide compound instead interacts with R368 or R371 when R365 is neutralized. Both R368 and R371 pass the hydrophobic barrier and are exposed on the extracellular side, when the channel switches from a resting state to the activated (open) state (Tao et al., 2010; Henrion et al., 2012).

The binding site proposed in this investigation, centrally in the VSD between S3 and S4, is close, but not identical, to the site found for other channel-opening compounds that bind within the VSD. NH29 opens the K_v7.2 channel by binding in the space between S2 and S4 (Peretz et al., 2010), and ztz240 and other compounds open the K_V7.2 channel by binding deeper into the VSD (Li et al., 2013). Thus, the VSDs of K_V channels seem to have a cleft at the top, which can harbor different small-molecule compounds to keep the channel activated.

The sulfonamide motif, a sulfonyl group connected to a nitrogen (Fig. 1 A), is widespread in medical drugs and covers a wide range of therapeutic uses. The motif even led to the naming of sulfa antibiotics, which preceded penicillin. Other sulfonamide compounds are anticonvulsant (e.g., sultiame), antihypertensive (e.g., thiazide diuretics), or antidiabetic (e.g., glibenclamide). The sulfonamide compounds have diverse sites and mechanisms of action. Many sulfonamides act by inhibiting different types of enzymes (Winum et al., 2006), but some block ion channels; for instance, the antidiabetic drug glibenclamide blocks the ATP-sensitive potassium channel in pancreatic β cells (Ashcroft and Ashcroft, 1990). Different types of aryl sulfonamides, including aminopiperidine sulfonamides, have been reported to block different types of voltage-gated calcium and sodium channels: Ca_V2.2 (Shao et al., 2012), Ca_V3.2 (Jarvis et al., 2014), and Na_V1.7 (Focken et al., 2016). However, to our knowledge, there are no sulfonamide pharmaceutical drugs on the market acting on voltage-gated ion channels.

Recently, it was shown that some sulfonamides bind to the VSD of domain IV in a voltage-gated Na channel to lock the channel in a nonconducting state (McCormack et al., 2013; Ahuja et al., 2015). However, the voltage sensor S4 was locked in an activated up state, but the channel was probably inactivated and thus nonconducting. Thus, short biaryl sulfonamides are able to open K channels and close Na channels by similar mechanisms. This suggests that short biaryl sulfonamides can potentially reduce neuronal excitability by targeting multiple ion channels (a potentially powerful strategy (Tigerholm et al., 2012; Salari et al., 2018). It also suggests that short biaryl sulfonamides can potentially be designed to specifically open or close different types of voltage-gated ion channels.

We thank Gunnar Nordvall for chemistry advice and compound selection, Anders B. Eriksson (SciLifeLab) for help with compound handling, Andreas Nolting for construction of the cell line, Anders Fridberger for comments on the text, and Alexis Reisch for linguistic advice. We thank the Chemical Biology Consortium Sweden at the Science for Life Laboratory for providing the chemical libraries tested in the screen. The NIH Clinical Collection was provided through the NIH Molecular Libraries Roadmap Initiative. The LCBKI primary screening set was provided by CBCS. Part of this work was assisted by Karolinska High Throughput Center (KHTC), a core facility at Karolinska Institutet with affiliation to Science for Life Laboratory (www.scilifelab.se/facilities/khtc/).

This work was supported by grants from the Swedish Research Council (2016-02615), the Swedish Heart-Lung Foundation (20150672), the Swedish Brain Foundation (2016-0326), and the Swedish Society for Medical Research.

The authors declare no competing financial interests.

Author contributions: F. Elinder conceived the study. S.I. Liin, P-E. Lund, J.E. Larsson, J. Brask, and F. Elinder designed experiments and analyzed data. S.I. Liin and J.E. Larsson performed manual electrophysiological experiments. P-E. Lund performed automated electrophysiological experiments. B. Wallner performed molecular docking simulations. S.I. Liin, P-E. Lund, J.E. Larsson, J. Brask, B. Wallner, and F. Elinder wrote the paper.

Kenton J. Swartz served as editor.