The Trials and Tribulations of Structure Assisted Design of KCa Channel Activators

Calcium-activated K+ channels constitute attractive targets for the treatment of neurological and cardiovascular diseases. To explain why certain 2-aminobenzothiazole/oxazole-type KCa activators (SKAs) are KCa3.1 selective we previously generated homology models of the C-terminal calmodulin-binding domain (CaM-BD) of KCa3.1 and KCa2.3 in complex with CaM using Rosetta modeling software. We here attempted to employ this atomistic level understanding of KCa activator binding to switch selectivity around and design KCa2.2 selective activators as potential anticonvulsants. In this structure-based drug design approach we used RosettaLigand docking and carefully compared the binding poses of various SKA compounds in the KCa2.2 and KCa3.1 CaM-BD/CaM interface pocket. Based on differences between residues in the KCa2.2 and KCa.3.1 models we virtually designed 168 new SKA compounds. The compounds that were predicted to be both potent and KCa2.2 selective were synthesized, and their activity and selectivity tested by manual or automated electrophysiology. However, we failed to identify any KCa2.2 selective compounds. Based on the full-length KCa3.1 structure it was recently demonstrated that the C-terminal crystal dimer was an artefact and suggested that the “real” binding pocket for the KCa activators is located at the S4-S5 linker. We here confirmed this structural hypothesis through mutagenesis and now offer a new, corrected binding site model for the SKA-type KCa channel activators. SKA-111 (5-methylnaphtho[1,2-d]thiazol-2-amine) is binding in the interface between the CaM N-lobe and the S4-S5 linker where it makes van der Waals contacts with S181 and L185 in the S45A helix of KCa3.1.


INTRODUCTION
Small-and intermediate-conductance calcium-activated potassium channels (K Ca ) are voltage independent and are gated by the binding of calcium to calmodulin, which functions as their calciumsensing beta subunit (Xia et al., 1998;Fanger et al., 1999). There are four members in the small-and intermediate-conductance K Ca channel subfamily, the small-conductance K Ca 2.1, K Ca 2.2, and K Ca 2.3, collectively known as SK channels, and the intermediate-conductance K Ca 3.1, also known as IK (Kaczmarek et al., 2017). These tetrameric membrane proteins consist of four six-transmembrane domains with their N-and C-termini positioned intracellularly. Calmodulin (CaM) is constitutively bound with its C-lobe to the calmodulin binding domain (CaM-BD) in the C-terminus of each K Ca channel subunit and opens the channel when the N-lobe binds calcium following increases in intracellular calcium in the proximity of the channel (Xia et al., 1998;Lee and MacKinnon, 2018). K Ca 2/3 channels are differentially expressed in the human body with K Ca 2 channels primarily expressed in, but not limited to, the CNS and K Ca 3.1 primarily found in peripheral tissues, lymphocytes and red blood cells (Adelman et al., 2012;Wulff and Kohler, 2013). K Ca 2 channels mediate after hyperpolarization and regulate firing frequency in neurons (Adelman et al., 2012). K Ca 3.1, on the other hand, is responsible for generating the driving force for calcium influx in immune cells and contributes to the regulation of vascular tone via the vascular endothelium . These expression patterns make both K Ca 2 and K Ca 3.1 channels attractive pharmacological targets (Wulff and Zhorov, 2008). Specifically, activation of K Ca 2 channels has been proposed for the treatment of diseases characterized by increased neuronal excitability, like ataxia (Shakkottai et al., 2011;Kasumu et al., 2012) and epilepsy, whereas activation of K Ca 3.1 has been proposed as a treatment for hypertension  and as a possible way to pharmacologically enhance anti-tumor T cell responses (Chandy and Norton, 2016). The idea behind the later hypothesis is that by enhancing K + efflux and thus lowering intracellular K + it might be possible to reset the "ionic checkpoint" and boost anti-tumor T cell functions in tumor infiltrating T cells, which have been shown to have increased intracellular K + concentrations suppressing their ability to activate (Eil et al., 2016). In support of this exciting therapeutic postulate, it has recently been demonstrated that pharmacological K Ca 3.1 activation can restore the ability of cancer patient derived CD8 + T cells to chemotax (Chimote et al., 2018).
Identifying potent and selective K Ca channel activators has been challenging. The first generation, which includes 1-EBIO, NS309 (Strobaek et al., 2004), and SKA-31 (Sankaranarayanan et al., 2009), are relatively unselective and only display a 5-10-fold selectivity for K Ca 3.1 over K Ca 2 channels (Christophersen and Wulff, 2015). This lack of selectivity has led to CNS related side effects when trying to use K Ca 3.1 activation as a new, endothelial targeted antihypertensive approach (Radtke et al., 2013). The second generation of K Ca channel activators, as exemplified by SKA-121 and SKA-111, are 40-100-fold selective for K Ca 3.1 and were efficacious in lowering blood pressure in normotensive and hypertensive mice while avoiding K Ca 2 channel mediated side effects on the CNS and on heart rate because of the improvement in selectivity . For the development of our first and second generation K Ca activators (the "SKA" compounds), we used a classical medicinal chemistry approach with no structural input during the structure activity optimization (Sankaranarayanan et al., 2009;Coleman et al., 2014). However, more recently, following the publication (Zhang et al., 2013) of the crystal structure of the K Ca 2.2 CaM-BD in complex with CaM and containing NS309 (pdb: 4J9Z), we generated homology models of the K Ca 3.1 and K Ca 2.3 CaM-BD in complex with CaM using the Rosetta molecular modeling suite and RosettaLigand for compound docking (Brown et al., 2017). Combining structural modeling and site-directed mutagenesis we determined that S372 in K Ca 3.1 (or the corresponding S632 in K Ca 2.3) is crucial for the activity of all 2-aminobenzothiazole/ oxazole-type K Ca activators, which are further stabilized by an extensive hydrogen bond network including E295, N300, R362 in K Ca 3.1 and M51 and E54 in calmodulin. Based on our findings we suggested that R362, a residue which is at the center of this network in the K Ca 3.1 but not the K Ca 2.3 models, is responsible for the 5-10-fold selectivity of these compounds for K Ca 3.1 over K Ca 2.3 (Brown et al., 2017).
In the current work we set out to use our atomistic level understanding of K Ca activator binding to develop a third generation of K Ca channel activators; this time with selectivity for K Ca 2.2 over K Ca 3.1. In this structure-based drug design approach we again used Rosetta ligand docking and carefully compared the binding poses of 2-aminobenzothiazole/oxazole-type K Ca activators in the K Ca 2.2 and K Ca 3.1 CaM-BD/CaM interface pocket. Based on differences between residues in the K Ca 2.2 and K Ca .3.1 models we manually designed 168 new virtual SKA compounds and tried to predict whether the compounds would show selectivity for K Ca 2.2 based on the computational docking models. The most promising compounds were synthesized, and their potency and selectivity for K Ca 2.2, and K Ca 3.1 tested by manual or automated electrophysiology. However, despite all these efforts, we failed to identify any K Ca 2.2 selective compounds and recently learned from the full-length cryo-EM structure of K Ca 3.1 published by the MacKinnon group (Lee and MacKinnon, 2018), that the C-terminal crystal was an artefact. Based on the full-length structure it was suggested that the "real" binding pocket for the K Ca activators is located between the S4-S5 linker and the CaM N-lobe (Lee and MacKinnon, 2018). We here confirmed this structural hypothesis through mutagenesis and now offer a new, corrected binding site model for the SKAtype K Ca channel activators. We hope that our findings provide a cautionary tale for the field in warning against some of the pitfalls in structure-based drug design.

Molecular Modeling
We previously described the generation of models (Brown et al., 2017) of the K Ca 3.1 and K Ca 2.3 C-terminal CaM-binding domain in complex with CaM with Rosetta computational modeling software (Rohl et al., 2004;Bender et al., 2016;Alford et al., 2017) using the x-ray structure of the K Ca 2.2 channel CaM-binding domain in complex with CaM and NS309 (Zhang et al., 2013) (pdb id: 4J9Z) as a template. We also previously provided a detailed description (Brown et al., 2017) of the procedure for ligand docking using the RosettaLigand docking application (Meiler and Baker, 2006;Davis and Baker, 2009;Bender et al., 2016) in the Rosetta program suite, version 3.7. Briefly, for this study, ligand conformers of variously substituted benzothiazoles/oxazoles were generated using Open Eye OMEGA software version 2.5.1.4 (OpenEye Scientific Software, Santa Fe, NM; http://www.eyesopen.com) (Hawkins et al., 2010;Hawkins and Nicholls, 2012; OEChem v OpenEye Scientific Software, Inc.), randomly placed within the binding pocket and then taken through the three stages of the RosettaLigand modeling which progresses from low-resolution conformational sampling and scoring to full atom optimization using Rosetta's all-atom energy function. A total of 10,000 models were generated for each virtual compound, the top 1,000 models with the lowest total energy score were selected, and the top 10 models with the lowest binding energy were identified, manually inspected for ligand/channel interactions and convergence between the K Ca 2.2 and the K Ca 3.1 model compared.
A model was considered converged if the top 10 models overlaid.
Cryo-EM structures of the full length K Ca 3.1 channel (Lee and MacKinnon, 2018) in two open states and one closed state (pdb id: 6CNN, 6CNO, 6CNM) were refined using the Rosetta cryo-EM refinement protocol (Wang et al., 2016) (Cryo-EM density map, Rosetta version 3.8) and the models with the lowest energy scores were chosen for docking of SKA-111. All molecular graphics of ligand, K Ca channel C-terminal CaM-binding domain in complex with CaM, or full length K Ca 3.1 channels were rendered using the UCSF Chimera software (Resource for Biocomputing, Visualization, and Informatics, San Francisco, CA).
Potential druggable sites in the Rosetta refined K Ca 3.1 open state-1 structure were identified using the SiteMap function in Glide (Schrödinger, LLC, New York, NY, 2018). SiteMap uses a site localization method based on interaction energies between the protein and grid probes, which is analogous to the Goodford's GRID algorithm. Sites were kept if they comprised at least 50 site points. A restrictive hydrophobicity definition and a standard grid (1.0 Å) were used for identifying potential binding pockets.
Protein Data Bank (pdb) format files of the Rosetta models of K Ca 3.1 open state 1 and open state 2 with SKA-111 docked in the interface between S 45 A helix and the CaM N-lobe are provided in the Data Supplement 1 ; pdb files of all other models are available upon request.

Molecular Biology
The cloning of human K Ca 3.1 (#AF033021) has been reported in the late 1990s (Logsdon et al., 1997). The gene was subcloned in-frame downstream to green fluorescent protein in the pEGFP-C1 expression vector (CLONTECH) (Wulff et al., 2001). All clones were verified by sequencing. Mutations were introduced using QuikChange site directed mutagenesis kit (Stratagene, La Jolla, CA) and were verified by fluorescence sequencing. For amino acid numbering for K Ca 3.1 we used the gene: Homo sapiens potassium channel, calcium activated intermediate/small conductance subfamily N alpha, member 4 (KCNN4). NCBI Reference Sequence: NP_002241.1.

Electrophysiology
All experiments were performed in either the inside-out or the wholecell configuration of the patch-clamp technique on either transiently transfected CHO cells or Human Embryonic Kidney (HEK) cell lines stably expressing hK Ca 3.1 or hK Ca 2.2 (Sankaranarayanan et al., 2009). All cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. Wild-type (WT) and mutant hK Ca 3.1 channel constructs were transfected using FuGENE 6 transfection reagent (Promega, Madison, WI) in OptiMEM reduced-serum medium (Life Technologies, Benicia, CA) for manual patch-clamp experiments or via electroporation (Nucleofector 2b and Lonza Amaxa Cell Line Nucleofector Kit T) according to the manufacturer's instructions for automated electrophysiology.
Cells transfected using FuGENE 6 were cultured in six-well plates for 24-48 h and then detached by TrypLE Express (Gibco, Grand Island, NY) and plated on coverslips for 30 min to 1 h for whole-cell recordings. For inside-out recordings, cells were plated 2-3 h before the experiments to attach them more firmly. Coverslips were placed in a 15 ml recording chamber mounted on an inverted microscope (Olympus XI-70 equipped with a pE-300Lite LED UV light source and filters; Olympus, Tokyo, Japan), and only clearly green fluorescent cells were patch-clamped. CHO cells transfected via electroporation were cultured in T75 flasks until 70% confluency and lifted with TrypLE, spun down, and resuspended. 1X10 6 cells were used per transfection with 1 µg of plasmid DNA. The setting for the Nucleofector 2b was, cell type: CHO, high efficiency. Cells were afterwards cultured in a T25 flask for 24 h before they were lifted for electrophysiology.
For manual whole-cell experiments, the extracellular solution contained 160 mM NaCl, 4.5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , and 10 mM HEPES (pH 7.4, 300 mOsm). Solutions on the intracellular side contained 154 mM KCl, 10 mM HEPES, 10 mM EGTA, 1.75 mM MgCl 2 , and 5.9 mM CaCl 2 for a calculated free Ca 2+ concentration of 250 nM (pH 7.2, 290 mOsm). For insideout experiments in symmetrical K + , the extracellular solution contained 154 mM KCl, 10 mM HEPES, 1 mM MgCl 2 , and 2 mM CaCl 2 (pH 7.4, 300 mOsm). Intracellular solutions contained 154 mM KCl, 10 mM HEPES, 10 mM EGTA, 1.75 mM MgCl 2 , and varying amounts of CaCl 2 for calculated free Ca 2+ concentrations of 0.05, 0.1, 0.25, 0.5, 1, 10, and 30 mM (pH 7.2, 280-300 mOsm). Free Ca 2+ concentrations were calculated using the July 3, 2009 online version of MaxChelator (https://somapp.ucdmc.ucdavis. edu/pharmacology/bers/maxchelator/webmaxc/webmaxcS.htm) assuming a temperature of 25°C, a pH of 7.2, and an ionic strength of 160 mM. Patch pipettes were pulled from soda lime glass (microhematocrit tubes; Kimble Chase, Rochester, NY) and had resistances of 1.5-3 MΩ when submerged. Experiments were controlled with a HEKA EPC-10 amplifier and Patchmaster software (HEKA, Lambrecht/Pfalz, Germany). In whole-cell experiments, cells were clamped to a holding potential of -80 mV, and K Ca currents were elicited by 200-ms voltage ramps from -120 to +40 mV applied every 10 s. For the inside-out experiments, cells were clamped to a holding potential of -80 mV, macro-patches were pulled and K Ca currents were elicited by 200-ms voltage ramps from -80 to +80 mV applied every 5 s. Solutions with different free Ca 2+ concentrations were then perfused in rapid successions to avoid run down. Each Ca 2+ concentration was normalized to 10 µM of free Ca 2+ by perfusing 10 µM Ca 2+ before and after each Ca 2+ concentration. If the two 10 µM values differed more than 10% from each other, the data point was excluded from the Ca 2+ concentration response curve.
The procedure for performing automated whole-cell K Ca channel recordings on a QPatch-16 automated electrophysiology platform (Sophion Biosciences) was previously described in detail by our laboratory (Jenkins et al., 2013). For the current study we used the solutions described above with disposable 16-channel planar patch chip plates (QPlates; patch hole diameter 1 mm, resistance 2.00-0.02MΩ, Sophion Biosciences, Woburn, MA). Cell positioning and sealing parameters were set as follows: positioning pressure -70 mbar, resistance increase for success 750%, minimum seal resistance 0.1 GΩ, holding potential -0 mV, holding pressure -20 mbar. To avoid rejection of cells with large K Ca 3.1 currents, the minimum seal resistance for whole-cell requirement was lowered to 0.001 GΩ. Access was obtained with the following sequence: 1) suction pulses in 29 mbar increments from -250 mbar to -453 mbar; 2) a suction ramp of an amplitude of -450 mbar; 3) -400 mV voltage zaps of 1 ms duration. Following establishment of the whole-cell configuration, cells were held at -80 mV and K Ca currents elicited by a voltage protocol that held at -80 mV for 20 ms, stepped to -120 mV for 20 ms, ramped from -120 to + 40 mV in 200 ms, and then stepped back to -120 mV for 20 ms. This pulse protocol was applied every 10 s. K Ca activator dilutions were prepared freshly (within 5 min of the initiation of the QPatch) with Ringer's solution from 10 mM stock solutions in dry DMSO. Final DMSO concentrations never exceeded 1%. Glass vial inserts (to avoid adsorption) were filled with 400 µL of compound solution and placed into the insert base plate for use in the QPatch assay. Exemplary QPatch raw current traces and plots of slope conductance versus time are shown in Supplementary Figure 3. For each compound we typically used 12 consecutive liquid periods Break-in, 3 saline additions to stabilize the current, 2 additions of test compound at 10 µM, 2 saline washes, 2 additions of SKA-31 at 10 µM, followed by 2 more saline washes. In each liquid period 10 pulse protocols were run.

Calcium Sensitivity Testing
When screening mutants, we always first assessed whether mutations had altered Ca 2+ sensitivity in a two-step process. First, mutant channels were patched with an intracellular solution containing 10 µM of free Ca 2+ . If the mutant did not display currents with amplitudes in the nA range under these conditions, we then tested whether the channel's control current (at 250 nM free Ca 2+ ) could be increased 5-10-fold in the presence of 100 µM EBIO (this fold increase level is common in the wild type channel). If the mutant did not meet either threshold it was deemed to have altered Ca 2+ sensitivity and was excluded from subsequent experiments investigating sensitivity to SKA-111. All mutants shown in Figure 7 or mentioned in the text exhibited current densities at +40 mV that were comparable to the WT channel and produced currents that were large enough to perform experiments. WT: 19.5 ± 23.4 pA/ pF; T212F-V272F: 12.4 ± 10.4 pA/pF; S181A: 11.6 ± 16.2 pA/pF; A184F: 7.2 ± 5.4 pA/pF; L185A: 6.9 ± 6.5 pA/pF; S181A-L185A: 8.8 ± 5.5 pA/pF; S372R: 13.8 ± 12.5 pA/pF (n = 5).

Chemical Synthesis
Compounds that were not commercially available were synthesized in our laboratory according to the general methods described below. Compounds reported previously were characterized by melting point, proton nuclear magnetic resonance ( 1 H NMR) and 13 carbon ( 13 C) NMR. All NMR spectra were recorded on an 800 MHz Bruker Avance III spectrometer. Data are reported as follows: chemical shift (δ), multiplicity, integration, coupling constant (Hz). Signals are designated as follows: s (singlet), d (doublet), dd (doublet of doublets), ddd (doublet of doublet of doublets), dt (doublet of triplets), t (triplet), quint (quintet), m (multiplet). New chemical entities were additionally characterized by high-resolution mass spectrometry (HRMS). For compounds (SKA-218, SKA-339, SKA-340, SKA-343 and SKA-347) where it was difficult to unambiguously confirm the structure based on 1 H and 13 C NMR, we grew crystals and subjected them to X-ray analysis, which allowed us to see the exact position of the substituents in the compounds. All compounds used for electrophysiological experiments were at least 95-98% pure based on NMR and/or mass spectrometry.

General Method I. Preparation of nitro-substituted 2-metylnaphtho[1,2-d]thiazoles or 1-substituted 4-nitronaphthalenes
One milliliter of HNO 3 was added slowly to 1 g of 2-methyl-βnaphthothiazole or 1-substituted naphthalene in a round bottom flask at room temperature without solvent. The reaction mixture was stirred for 1 h and then 1.5 ml of H 2 SO 4 (95%) was added slowly. The mixture was stirred at room temperature for 30 min until completion of the reaction was indicated by the disappearance of starting material on thin-layer chromatography (TLC). Then the pH of the reaction mixture was adjusted to between 7 and 7.5 with 4N NaOH, the mixture extracted with ethyl acetate and the organic phase washed with water and brine. The extract was dried with anhydrous sodium sulfate and solvent was evaporated under reduced pressure. The crude product was purified via flashchromatography (petroleum ether: ethyl acetate, 8:2).

General Method II. Preparation of amino-substituted 2-metylnaphtho[1,2-d]thiazoles or 1-amino 4-substituted naphthalenes
One gram of the previously prepared nitro-substituted 2-metylnaphtho[1,2-d]thiazole or 1-substituted 4-nitronaphthalene was dissolved in 30 mL of 95% ethanol and 5% water, 20 mg of palladium charcoal was added and then 2 mL of hydrazine were dropped in slowly. After the addition was completed, the reaction mixture was stirred at 80°C for 12 h and the progress of the reaction was monitored by TLC with dichloromethane. After completion, the reaction mixture was filtered to remove the palladium charcoal and concentrated to dryness under vacuum. The residue was dissolved in ethyl acetate and was washed with water and brine. The organic September 2019 | Volume 10 | Article 972 Frontiers in Pharmacology | www.frontiersin.org phase was dried with anhydrous sodium sulfate and the solvent was removed under vacuum. The crude product was purified by flashchromatography with ethyl acetate/petroleum ether (3:7 v/v) as eluent to give the product.

General Method III. Preparation of methyl-substituted acetonaphthones
Acetylchloride (750 μl, 10 mmol) and AlCl 3 (1.33 g, 10 mmol) were added sequentially to 20 ml of chloroform at room temperature. Methyl substituted naphthalenes (1 ml, 7 mmol) in 5 ml of chloroform were then added, the reaction mixture was stirred for 2 h, and the progress of the reaction was monitored by TLC. The reaction mixture was quenched with sodium bicarbonate solution and washed with water and brine. Solvent was removed under reduced pressure to give the crude product. The crude product was purified by flash-chromatography using ethyl acetate/petroleum ether (2:8 v/v).

General Method IV. Preparation of substituted 2-bromo-1-(methyl substituted naphthalen-1-yl)ethan-1-one or 2-bromo-1-(methyl substituted cyclohex-1-yl)ethan-1-one
Appropriately substituted acetophenones were dissolved in chloroform (20 ml) at room temperature and the reaction mixture was stirred. Liquid bromine (1.2-1.5 equivalent) in chloroform (5 ml) was then added drop-wise. The progress of the reaction was monitored by TLC. The reaction mixture was quenched with a saturated aqueous solution of sodium hydrogen carbonate and the pH was adjusted to between 7 and 7.5. The organic phase was washed with water and brine and the solvent was removed under vacuum. The crude residue was purified via flash chromatography (petroleum ether: ethyl acetate, 7:3).

General Method V. Preparation of substituted 4-phenylthiazole or 4-naphthalenyl thiazoles
Substituted 2-bromo-1-(methyl substituted naphthalen-1-yl) ethan-1-one or 2-bromo-1-(methyl substituted cyclohex-1-yl) ethan-1-one was added to a solution of a substituted thiourea in 20 ml of absolute ethanol. The mixture was refluxed for 2 h. After completion of the reaction, the ethanol was evaporated under vacuum. The dried reaction mixture was dissolved in dichloromethane and neutralized with saturated NaHCO 3. The organic phase was washed with water and brine, and then dried with anhydrous Na 2 SO 4. Finally, the solvent was evaporated under vacuum. The crude product was reconstituted in a methanol, treated with charcoal and recrystallized.

General Method VI. Preparation of heterocyclic 1-substituted 4-nitronaphthalenes
1-Fluoro-4-nitronaphthalene and a secondary or heterocyclic amine (molar ratio 1:3) were dissolved in 10 mL of DMF at 90°C. Potassium carbonate (3 equivalents) was then added. The reaction mixture was stirred and monitored by TLC. After completion, the reaction mixture was washed with brine several times to remove the DMF and extracted with ethyl acetate. Solvent was evaporated to give the crude product. The crude product was purified by flashchromatography using ethyl acetate/petroleum ether (2:8 v/v).

General Method VII. Preparation of substituted 2-aminonaphthothiazoles
Liquid bromine (150 μL, 3 mmol) and KSCN (485 mg, 5 mmol) were added to a solution of the 1-amino 4-substituted naphthalenes (3 mmol) in 10 ml of acetic acid and the reaction mixture was stirred at room temperature for 30 min. The reaction was quenched by adding 4N NaOH. The reaction mixture was then washed with water and brine and extracted with ethyl acetate. Solvent was evaporated to give the crude product, which was purified by flash-chromatography using ethyl acetate/ petroleum ether (3:7 v/v).

General Method VIII. Preparation of N-substituted naphtho[1,2-d]thiazol-2-amine
1-Naphthylamine 500 mg (3.5 mmol) was dissolved in 25 mL chloroform and 0.3 mL of trimethylamine before adding substituted isothiocyanate (3.8 mmol). The resulting mixture was refluxed at 65°C for 48 h. Solvent was then evaporated, and the resulting solid was washed with diethyl ether to obtain N,N'-disubstituted thiourea. The N,N'-disubstituted thiourea (2.2 mmol) obtained in the previous step was suspended in 25 mL chloroform, and a solution of liquid bromine in chloroform (1 eq.) added over a period of 1 h, after which sodium thiosulfate (aq.) was added to the reaction mixture. The reaction mixture was washed with 4N NaOH (aq.) and the organic layer was dried over anhydrous sodium sulfate and evaporated under reduced pressure.

Crystal Structure Determination
The SKA-218, SKA-339, SKA-340, SKA-343 and SKA347 crystals selected for data collection were mounted and optically centered in a nitrogen low temperature stream -183°C (90K), on the Bruker diffractometer with an APEX2 CCD detector or a Bruker D8 Venture diffractometer equipped with a Photon100 CMOS detector (Bruker, Madison, WI). Data were collected with the use of Mo Ka radiation in all cases (λ = 0.71073 Å). The structures were solved by direct methods (SHELXT) and refined by full-matrix least-squares on F 2 (SHELXL-2018/3). All non-hydrogen atoms were refined with anisotropic displacement parameters. For a description of the method, see (Sheldrick, 2008).

RESULTS
Structure-Based Drug Design Using the Crystal Structure of the K Ca 2.2 CaM-BD/ CaM Interface We previously generated homology models of K Ca 3.1 and K Ca 2.3 (Brown et al., 2017) using the Rosetta membrane method (Rohl et al., 2004;Bender et al., 2016;Alford et al., 2017) and the x-ray crystal structure of the K Ca 2.2 CaM-BD/CaM (Zhang et al., 2013) as a template. We localized the binding site of the benzothiazoles/oxazoles to the CaM-BD/CaM interface and generate models of the K Ca 3.1 and K Ca 2.3 CaM-BD/CaM complexes with SKA-121 and SKA-111 using Rosetta Ligand docking (Meiler and Baker, 2006;Davis and Baker, 2009;Bender et al., 2016). The docking models of K Ca 3.1 showed that the amino groups of the benzoxazole ring of SKA-121 and of the benzothiazole ring of SKA-111 form hydrogen bonds with M51 and E54 in calmodulin (Brown et al., 2017). Moreover, E54 was further stabilized by an extensive hydrogen bond network with R362, E295 and N300 in the K Ca 3.1 channel, which we hypothesized to be responsible for the K Ca 3.1 selectivity of SKA-121 and SKA-111. In the K Ca 2.3 or K Ca 2.2 model, however, SKA-111 and SKA-121 formed only hydrogen bonds with M51 and E54, due to the shorter length of the sidechain of S622 in K Ca 2.3 or N474 in K Ca 2.2 than that of the corresponding R362 in K Ca 3.1 (Supplementary Figure 1).
Using the information from these computational docking models, we here intended to design new K Ca 2 selective activators by attempting to predict whether the compounds would show selectivity for K Ca 2 over K Ca 3.1 channels. However, instead of our previously generated K Ca 2.3 CaM-BD/CaM homology model, we here used a K Ca 2.2 CaM-BD/CaM model based on the K Ca 2.2 CaM-BD/CaM-NS309 crystal structure (pdb:4J9Z). We made this switch because K Ca 2.2 is the most abundantly expressed K Ca 2 channel in the mammalian CNS (Adelman et al., 2012) and therefore constitutes an attractive target for the treatment of ataxia and epilepsy. We additionally docked eight more 2-amino-naphthobenzothiazole derivatives (SKA-31, SKA-44, SKA-45, SKA-72, SKA-73, SKA-107, SKA-117, and SKA-120) into the K Ca 2.2 and K Ca 3.1 homology models and found that these compounds exhibited the same hydrogen bond network as SKA-121 and SKA-111 in K Ca 3.1 (Supplementary Figure 1). Based on these docking poses we hypothesized that disruption of the hydrogen bond between the -NH 2 group of the benzothiazole ring, and the CaM M51 and E54 residues, which are present in both the K Ca 3.1 and K Ca 2.2 models, might be a way to achieve K Ca 2.2 selectivity. Our goal here was to first "break" this hydrogen bond to ideally achieve K Ca 2.2 selectivity and then regain potency by adding substituents in other positions to pick up unique contacts in K Ca 2.2. We therefore virtually added various substituents in the C-4,5,6,7,8,9 positions of SKA-74 ( Figure 1A), a compound which contains a methyl group in C-2 position instead of an -NH 2 group and which we had previously found to activate K Ca 2 and K Ca 3.1 channels with a similar ~30 μM potency . To improve van der Waals contacts we introduced -CH 3 , -Br, -Cl, -CF 3 groups. The C-5 position was chosen for the first trial since the K Ca 2.2 docking model of SKA-74 showed that it is adjacent to A484 and in range for new interactions. We also virtually introduced larger substituents such as cyclohexyl, cyclopentyl, cyclopropyl and phenyl on the 2-position amino group because the K Ca 2.2 model showed more space in this region of the binding site than FIGURE 1 | Design scheme of K Ca 2.2 selective activators and Rosetta models of the top 10 binding poses with the lowest energy of template compounds in the interface between CaM (pink) and the CaM-BD (light green) of K Ca 3.1 and K Ca 2.2. The docking model of SKA-74 (A) and SKA-76 (B) showed that the ten lowest binding energy scored models exhibit good structural convergence in K Ca 2.2 but not in K Ca 3.1 suggesting selectivity for K Ca 2.2 over K Ca 3.1. the K Ca 3.1 model. In addition, we also virtually generated double substituted SKA-74 derivatives (a bulky substituent in the C-2 position and methyl in C-5 position). In parallel we designed a small focused library of 2-aminothiazoles (SKA-75 and SKA-76 derivatives; Figure 1B). Our reasoning for the choice of these two compounds as additional templates was that SKA-75, like SKA-74, had previously been found to be of similar potency (~30 μM) on both K Ca 2.3 and K Ca 3.1, and that SKA-76 was slightly more potent on K Ca 2.3 (~25 μM) than K Ca 3.1 (~50 μM) (Coleman et al., 2014). In addition, the molecular docking models of SKA-75 and SKA-76, which both lack the continuous conjugation between the naphthalene and the 2-aminothiazole ring (Figure 1  right), showed that the ten lowest energy scored models exhibited good structural convergence in K Ca 2.2 but not in K Ca 3.1 ( Figure  1B). In order to improve the potency and selectivity of SKA-75 and SKA-76 we virtually added a methyl group in the 2,3,4,5,6,7 or 8 positions of the naphthalene ring in SKA-75 and SKA-76. We then replaced the naphthalene ring with a bi-phenyl ring, a 2-aminophenylthiazole or a 2-aminobiphenylthiazole. We further replaced the naphthalene ring with a phenyl ring and added larger groups such as N-trifluorophenyl, N-pyridine to the -NH 2 group of the 2-aminothiazole ring.
The virtually proposed 63 SKA-74 derivatives, 87 SKA-75 and SKA-76 derivatives as well as another 18 SKA-31-related compounds (see Supplementary Figure 2 for all structures) were randomly placed into the K Ca 2.2 and K Ca 3.1 homology models of the CaM-BD/CaM interface pocket, energy minimized through the three stages of the RosettaLigand method, and the top 10 lowest binding energy scoring models were analyzed. SKA-74 derivatives with -CH 3 , -CF 3, -Br and -Cl in 5-position were predicted to show selectivity for K Ca 2.2 channels over K Ca 3.1 (Figures 2A, B) because they converged well in the K Ca 2.2 model and made van der Waals interactions (dark purple in Figure 2B) with A477, V481 in K Ca 2.2 and M72, F68, I63 and M51 in CaM. In contrast, the top 10 lowest energy models of all SKA-74 derivatives with bulky substituents in C-2 position did not converge in either K Ca 2.2 or K Ca 3.1, whereas all double substituted SKA 74 derivatives were predicted to be K Ca 2.2 selective with good structural convergence in K Ca 2.2. For the SKA-75 and SKA 76 derivatives, the docking model suggested that addition of a methyl group to the 6-position of the naphthalene ring in SKA 76 (SKA-198) created new hydrogen bonds and good structural convergence in the K Ca 2.2 model but not in K Ca 3.1. Replacement of the naphthalene ring of SKA-75 with a bi-phenyl ring, a 2-aminophenylthiazole or a 2-aminobiphenylthiazole (SKA-232, SKA-230 and SKA-255) created new hydrogen bonds and good structural convergence in both K Ca 2.2 and K Ca 3.1. The molecular docking model showed that while these SKA-75 derivatives (SKA-232, SKA-230 and SKA-255) only formed 2 hydrogen bonds in K Ca 3.1, they formed four hydrogen bonds in K Ca 2.2 (see Figures 2C, D for SKA-230) suggesting selectivity for K Ca 2.2 as well as relatively high potency.

Synthesis and Activity Testing of the Newly Designed K Ca Channel Activators
Based on the docking models we chose 26 (16 SKA-74 and 10 SKA-75/76 derivatives) of the 168 virtual compounds for synthesis including some that were not predicted to be selective in order to verify that the predicted selectivity in the model is consistent with experiments (see Supplementary Figure 2 where the chosen structures are highlighted in color). A general scheme of the compound synthesis is given in Figure 3. To obtain 5-CH 3 substituted SKA-74 derivatives the commercially available starting materials, 2-methylnaphtho[1,2-d]thiazole (SKA-74) and 2-methylnaphtho[1,2-d]oxazole (SKA-103), were first brominated using liquid bromine to obtain SKA-132 and SKA-133, which were then reacted with CuCN in a cross coupling reaction with a palladium catalyst to obtain SKA-126 and SKA-135. SKA-130 was synthesized by electrophilic aromatic   The Full-Length K Ca 3.1 Cryo-EM Structure Reveals That the C-Terminal CaM-BD/ CaM Dimer Crystal Is an Artefact In our structure-based drug design attempt described above, we had used the C-terminal CaM-BD/CaM crystal structure, which consist of two vertically orientated CaM molecules and two horizontal K Ca 2.2 C-terminal fragments in an antiparallel arrangement (Zhang et al., 2012). In this structure, the CaM N-lobe interacts with the C-terminal region of the CaM-BD, whereas the C-lobe is bound to N-terminal region of the CaM-BD. Several K Ca channel activators, EBIO (Zhang et al., 2012), NS309 (Zhang et al., 2013) and, most recently in a publication from the Structural Biology group at Pfizer , CyPPA and riluzole were shown by X-ray crystallography and solid-state NMR to be located at the interface between the CaM N-lobe and the C-terminal region of the CaM-BD, where we docked our compounds. For clarity, only half of this so-called "dimer of dimers" complex, one CaM and one C-terminal fragment, was used for our modeling (see Supplementary Figure 1).
Recently, the MacKinnon group (Lee and MacKinnon, 2018) determined the full-length cryo-EM structures of K Ca 3.1 in the closed and in two activated states (pdb: 6cnm, 6cnn, and 6cno) and revealed that the C-terminal CaM-BD/CaM dimer crystal is an artefact. The full-length structure showed four CaMs per channel tetramer, with the CaM C-lobe of each CaM tightly bound to the CaM-BD of each subunit in the closed and the two activated states ( Figure 6A). However, the N-lobes were only clearly visible in the open, Ca 2+ -bound states and poorly resolved in the closed, Ca 2+ -free structure suggesting that they are flexible in the absence of Ca 2+ . When Ca 2+ binds to the N-lobe it moves from the bottom of the S2 segment to the bottom of the S4-S5 linker (which in K Ca 3.1 consists of two helices), while the C-lobe maintains its interaction with the HA and HB helices in the C-terminus. The N-lobe then pulls part of the S4-S5 linker, namely the S 45 A helix downward and this displacement expands the S6 helices and opens the pore (Lee and MacKinnon, 2018). In their study the MacKinnon group also proposed a new binding pocket for the K Ca activator EBIO formed by the S 45 A helix and the CaM N-lobe, in which EBIO binds to L185 in the S 45 A linker (Lee and MacKinnon, 2018) instead of L480 in the C-terminal crystal complex (Zhang et al., 2012). However, this very plausible alternative binding site hypothesis was not experimentally tested.
Probing the "New" SKA Compound Binding Site at the Interface Between the S 45 A Helix and the CaM N-Lobe Based on this new binding site hypothesis we here docked and energy minimized SKA-111 in the interface pocket between the S 45 A helix and the CaM N-lobe ( Figure 6A) using the open state of the full length cryo-EM structure of K Ca 3.1. The two open cryo-EM structures (6cnn and 6cno) were refined using the Rosetta cryo-EM refinement protocol with cryo-EM density map (Wang et al., 2016). A total of 10,000 models were generated for each state and the model with the lowest energy among the largest clusters of the top 1,000 models was used for RosettaLigand docking of SKA-111 starting from random ligand positions at the interface between the S 45 A helix and the CaM N-lobe. Similar to our previous docking model,  is stabilized in open state 1 by hydrogen bonds involving M51 and E54 and multiple CaM N-lobe residues are located within a 5 Å radius sphere around SKA-111 (F19, I27, L32, M51, I52, E54 V55, I63, F68, M71, M72, R74, K75) ( Figure 6A, open state 1). However, on the channel side SKA-111 now makes van der Waals interactions with S181 and L185 in the S 45 A helix ( Figure 6A). Interestingly, in open state 2 the lowest energy docking pose of SKA-111 is "flipped" 180 degrees around the hydrogen bond with CaM M51 and the molecule now makes van der Waals contacts with S181 and A184 in the S 45 A helix and again multiple residues in the CaM N-lobe (Supplementary Figure 4). The parent compound SKA-31 and the benzoxazole SKA-121 take up similar low energy docking poses in which they make contacts with M51 and E54 in CAM and S181 and/or L185 in K Ca 3.1 (Supplementary Figure 5).
In order to probe this "new" binding site we decided to mutate S181, L185 and the neighboring A184 to smaller, bulkier or charged residues to either disrupt contacts with SKA-111 or disturb the overall shape and size of the S 45 A helix/CaM N-lobe interface pocket. As described in the Materials and Methods, mutations were first tested for expression and Ca 2+ sensitivity and then used for evaluation of SKA-111 sensitivity. Mutating serine 181 to a shorter alanine, reduced SKA-111 potency 7-fold by removing the van der Waals contact to the 5-position -CH 3 group of SKA-111 (Figures 6B and 7). Mutating S181 to a charged arginine or large phenylalanine also reduced SKA-111 potency presumably by "pushing" SKA-111 forward. Introducing even larger tryptophan or tyrosine residues had an interesting effect on K Ca 3.1 gating and resulted in channels that already produced large nA currents in the presence of only 250 nM free Ca 2+ , suggesting that these mutants are more sensitive to Ca 2+ then the wild-type channel. We therefore did not use these two mutants for testing SKA-111 sensitivity. Mutating the other S 45 A helix residue, L185, that is in direct van der Waals contact with SKA-111 had a similar effect as mutating position 181. Replacing leucine 185 with a smaller alanine reduced SKA-111 potency by 7-fold, while substitution of a larger phenylalanine in 185-position produced an 8-fold reduction in potency (Figures 6B and 7). Introduction of even larger (W, Y) or charged residues unfortunately resulted in non-functional K Ca 3.1 channels that no longer respond to free Ca 2+ concentrations as high as 10 µM (Figure 6B). Although it is not in direct contact with SKA-111 we also mutated the neighboring A184 position to F because our model suggested that a phenylalanine in this position would push the CaM-K75 residue, that is contacting SKA-111 upwards. In keeping with this hypothesis, the A184F mutant was 9-fold less sensitive to SKA-111 than the WT K Ca 3.1 channel (Figures 6B and 7).
In order to more dramatically reduce SKA-111 sensitivity we next generated the double S181A-L185A mutant reasoning that it would disrupt both the van der Waals interactions SKA-111 makes with the S 45 A helix. In keeping with this idea, the S181A-L185A mutant produced a roughly additive effect when compare with the two single A mutations and reduced SKA-111 potency by 18-fold (Figures 6B and 7). In order to determine that this change in SKA-111 sensitivity is not caused by a reduced Ca 2+ sensitivity we performed in-side out recordings comparing the Ca 2+ sensitivity of the S181A-L185A mutant with the WT K Ca 3.1 channel and found that mutant did not differ in its Ca 2+ sensitivity ( Figure 7C). As a further control experiment, we tested the sensitivity of a double "fenestration" mutation, T212F-V272F, to SKA-111. We had previously generated this mutation, FIGURE 6 | Docking model of SKA-111 in the full-length K Ca 3.1 structure and mutational strategy. (A) Bottom and side view of the full-length K Ca 3.1 cryo-EM structure following Rosetta refinement in the Ca 2+ free closed state (pdb: 6cnm) and Open state I (pdb: 6cnn). The channel is shown in gray, the CaM C-lobe in purple and the CaM N-lobe in yellow. Next to open state 1 we show a zoom out of the lowest energy docking pose of SKA-111 in the interface between the CaM N-lobe and the S45A helix interface. Hydrogen bonds are shown in black, van der Waals interactions are visualized in purple. For clarity, not all side chains of CaM residues within contact range of SKA-111 are explicitly shown. Please note the channel residues S181, A184 and L185. (B) Alignment of the S 45 A helix sequence in K Ca 2.1, K Ca 2.2, K Ca 2.3 and K Ca 3.1. Residues that were mutated are highlighted in yellow and the EC 50 values for SKA-111 shown next to each mutant (for confidence intervals see Figures 7 and 8).
which closes the K Ca 3.1 fenestration between S5 and S6 with two bulky aromatic residues without changing the biophysical properties of the channel when identifying the binding site of the dihydropyridine nifedipine in K Ca 3.1 (Nguyen et al., 2017). As expected, the double "fenestration" mutant was as sensitive to SKA-111 as the WT channel (Figure 7).

Why Did the S372R Mutation in Our Previous Study Have Such a Large Effect of K Ca Activator Potency?
Taken together the above presented mutagenesis data suggest that the K Ca channel activator SKA-111 is indeed binding in the interface between the S 45 A helix and the CaM N-lobe as hypothesized by Lee et al. (Lee and MacKinnon, 2018). However, we were still puzzled by the fact that we had previously seen such a strong reduction in potency for multiple K Ca channel activators including SKA-31, SKA-111, EBIO and NS309, when mutating S372 in K Ca 3.1 or the corresponding S632 residue in K Ca 2.3 to arginine (Brown et al., 2017). Based on the C-terminal crystal fragment we had believed this residue to be located at the back of the interfacial binding site pocket. In the full-length K Ca 3.1 structure S372 is located at the top of the C-terminal HC helix, which forms a coiled coil located at the center of the channel. As shown in Figure 8A, S372 is facing outward from this coil towards N42 in the CaM N-lobe and accepting a hydrogen bond from its NH 2 group. Mutating S372 to a larger, charged arginine ( Figure 8B) in our Rosetta model of open state I "pushes" the outer edge of the CaM N-lobe 6.4 Å downwards, now resulting in a hydrogen bond with L39 in the CaM N-lobe and changing the shape and the volume of the interface pocket (see Figure 8C for an overlay of the WT and mutant channel). In keeping with this "distortion" of the interface pocket SKA-111 does not converge in the S372R mutant, while all top 50 lowest energy-binding poses virtually overlay in the WT channel pocket (Supplementary Figure 6). Interestingly, introducing the double AA mutation (S181A-L185A) into the S372R mutant, resulted only in small, but not statistically significant additional shift in the concentration-response curve of SKA-111 ( Figure 8D).

DISCUSSION
We here attempted to perform structure assisted design of K Ca 2.2 selective small molecule activators using the K Ca 2.2 CaM-BD/CaM FIGURE 7 | Mutations of S181 and L185 in the S 45 A helix disturb SKA-111 activity but not calcium gating. (A) Representative whole-cell WT and mutant K Ca 3.1 currents with an intracellular free calcium concentration of 250 nM in the presence and absence of SKA-111. (B) Concentration-response for SKA-111 induced current activation: WT (EC 50 = 146 nM, 95% CI: 99-193 nM), T212F-V272F (EC 50 = 153 nM, 95% CI: 114-182 nM, P = 0.1017), S181A (EC 50 = 1.012 µM, 95% CI: 0.780-1.244 µM, P < 0.0001), A184F (EC 50 = 1.326 µM, 95% CI: 1.205-1.447 µM, P < 0.0001), L185A (EC 50 = 0.993 µM, 95% CI: 0.903-1.083 µM, P < 0.0001), S181A-L185A (EC 50 = 2.654 µM, 95% CI: 2.619-2.624 µM, P < 0.0001). Whole-cell K Ca 3.1 currents were elicited by voltage-ramps from -120 to + 40 mV with an intracellular free calcium concentration of 250 nM. Data points are mean ± S.D. from 3-5 independent cells/recordings. The reported P values are for an extra sum-of-squares F test (GraphPad Prism5; GraphPad Software, La Jolla, CA) to compare the curves of K Ca 3.1 mutants to WT. (C) Inside-out calcium concentrationresponse curves for WT K Ca 3.1 (EC 50 = 437 nM, 95% CI: 353-521 nM, n H = 1.98) and the S181A-L185A double mutant (EC 50 = 392 nM, 95% CI: 287-497 nM, n H = 1.39, P = 0.4951). Data points are the mean ± S.D. from 3-5 independent recordings. The calcium-sensitivity of the mutant is statistically not different from the WT K Ca 3.1 channel (P = 0.4951 in extra sum-of-squares F test). crystal structure. This structure seemed extremely attractive for a structure based approach because it had been repeatedly crystalized by multiple groups. The first structure of the K Ca 2.2 channel C-terminal calmodulin binding domain (CaM-BD) in complex with CaM was reported in 2001 at 1.6 Å resolution and showed an elongated, anti-parallel dimer of two K Ca 2.2 C-terminal 76 amino acid long fragments. On each of these channel pieces CaM was tightly bound with its C-lobe to two alpha helices connected by a turn from the same channel subunit and "grabbed" the free end of the fragment from the other subunit in the dimer with its N-lobe suggesting that CaM-BD dimerization might gate K Ca 2 channels (Schumacher et al., 2001). The same dimeric crystal orientation was afterwards repeatedly observed in structural studies addressing the mechanism of action of small molecule K Ca channel activators and of PIP 2 on K Ca 2.2 channel function (Zhang et al., 2012;Zhang et al., 2013;Zhang et al., 2014). Zhang et al. showed in several highresolution X-ray structures that the K Ca activator EBIO, the more potent NS309, as well as several NS309 derivatives bind in the interface between the CaM N-lobe and the K Ca 2.2 CaM-BD (Zhang et al., 2012;Zhang et al., 2013;Nam et al., 2017). Independently of this work a group at Pfizer also crystallized the K Ca 2.2 CaM-BD/ CaM complex and showed the presence of two other K Ca activators, riluzole and a CyPPA derivatives, in the interface between the CaM N-lobe and the channel CaM-BD . This interaction Concentration-response curve for SKA-111 induced current activation: WT (EC 50 = 146 nM, 95% CI: 99-193 nM), S372R (EC 50 = 6.860 µM, 95% CI: 6.788-6.932 µM, P < 0.0001), S181A-L185A-S372R (EC 50 = 8.876 µM, 95% CI: 8.652-9.100 µM, P < 0.0001). Data points are mean ± S.D. from 3-5 independent cells/recordings. binding pocket (Figure 9B and Supplementary Figure 1). In both the "old" and the new binding pocket the amino group of the benzothiazole ring of SKA-111 forms two hydrogen bonds with M51 and E54 in the CaM N-lobe and makes nine van der Waals contacts with additional CaM residues. We suspect that these multiple, strong interactions with CaM are responsible for the fact that the benzothiazole riluzole and the structurally related EBIO and NS309 could be soaked into the C-terminal crystal where the highly flexible N-lobe of CaM had "grabbed" the only available part of the C-terminal helix that was not already occupied by the C-lobe. And as chance would have it the sequence in this part of the K Ca 2.2 C-terminus also contains a serine and leucin, similarly spaced apart, creating a very similar environment on the channel side for additional van der Waals contacts. The fact, that the majority of the contacts that SKA-111 is making in the interface pocket between the S 45 A helix and the CaM N-lobe are with CaM are also in line with our findings that we could significantly alter SKA-111 potency by mutating K Ca 3.1 residues in the S 45 A helix but not render K Ca 3.1 completely insensitive to its activating activity without rendering the channel non-functional through more drastic mutations.
We believe we have now identified the "correct" binding site for SKA-111 and related SKA compounds. However, since we only used mutagenesis in combination with functional electrophysiological recordings to confirm the binding site, there remains the caveat that we could have observed allosteric effects. It will therefore be important to perform additional experiments in future that demonstrate binding to the S 45 A helix for example with a photoaffinity probe and to directly measure ligand binding affinities using surface plasma resonance, isothermal titration calorimetry or differential scanning fluorimetry. These experiments should be performed with physiological ligand concentrations and not with saturating concentrations as was the case in the above described crystallography and solution state NMR studies in order to avoid the identification of "non-physiological relevant" binding sites.
Based on the K Ca 3.1 structural model we would further like to hypothesize here that the cytoplasmic facing surface of K Ca 3.1 and the closely related K Ca 2 channels offer multiple binding pockets for positive and potentially also negative gating modulators. As visualized in Figure 9C in dark blue, the S 45 A helix/CaM N-lobe interface is present four times in the complex consisting of four K Ca 3.1 α-subunits and four CaM molecules. How many of these pockets can be occupied by a small molecule activator is currently unknown and we would like to posit that it is possible that the occupancy of these four pockets might differ between different activators molecules and different K Ca channel subtypes. In addition to the S 45 A helix/CaM N-lobe interface, the SiteMap function of the Schrödinger Glide software identifies two other potential druggable sites in the cytoplasmic surface of K Ca 3.1 that could accommodate small molecules: one adjacent site located between the S 45 B helix and the HA helix in the C-terminus (orange circle in Figure 9C) and another site in the space between the CaM C-lobe and the HB helix in the C-terminus (sky blue circle in Figure 9C). Additional experimental and structural work will be necessary to determine if any, and how many, of these sites are targeted by positive or negative K Ca channel gating modulators.

DATA AVAILABILITY
The physical data of all synthesized compounds are provided in the Method section of this article. NMR spectra are available on request. Protein Data Bank (pdb) format files of the Rosetta models of K