The voltage-gated potassium (Kv) channel formed by Kv7.2/3 (KCNQ2/3) heteromers is a major target for anticonvulsant drug development. Here, we screened 1,444 extracts primarily from plants collected in California and the US Virgin Islands, for their ability to activate Kv7.2/3 but not inhibit Kv1.3, to select against tannic acid being the active component. We validated the 7 strongest hits, identified Thespesia populnea (miro, milo, portia tree) as the most promising, then discovered its primary active metabolite to be gentisic acid (GA). GA highly potently activated Kv7.2/3 (EC50, 2.8 nM). GA is, uniquely, 100% selective for Kv7.3 versus other Kv7 homomers; it requires S5 residue Kv7.3-W265 for Kv7.2/3 activation, and it ameliorates pentylenetetrazole-induced seizures in mice. Also an active aspirin metabolite, GA provides a molecular rationale for the use of T. populnea as an anticonvulsant in Polynesian indigenous medicine and presents novel pharmacological prospects for potent, isoform-selective, therapeutic Kv7 channel activation.

Collection of plant samples
We collected, between 2019 and 2022, aerial parts of plants under permit from Mojave National Preserve (study # MOJA-00321), Yosemite National Park (study # YOSE-00839), Santa Monica Mountains National Recreation Area (study # SAMO-00192), Muir Woods National Monument (study # MUWO-00035), Santa Cruz Island and Santa Rosa Island (study # CHIS-0023), and Boyd Deep Canyon (Indian Wells, CA) (permit through Philip L. Boyd Deep Canyon. Desert Research 376 Center, University of California, Riverside, CA) in California, US. In USVI, we collected from Virgin Islands National Park, St. John (study # VIIS-20001), Salt River Bay National Historical Park and Ecological Preserve, St. Croix (study # SARI-00056) and Buck Island Reef National Monument, St. Croix (study # BUIS-00103). Plant samples were collected in a manner designed to not kill the remaining plant, sealed in Ziploc bags (SC Johnson, Racine, WI, US), kept cold, and frozen as soon as possible. Other plant extracts were made from plants purchased from Crimson Sage Nursery (Orleans, CA) and grown in the senior author’s garden in Irvine CA, Mountain Rose Herbs (Eugene, OR) or Mother’s Market (Irvine, CA). Plant samples were stored in -20 oC freezers until extraction.

Preparation of plant extracts
Leaves and flowers were pulverized using a bead mill with porcelain beads in batches in 50 ml tubes (Omni International, Kennesaw, GA, United States), then the homogenates resuspended in 80% methanol/20% water (100 ml per 5 g solid) and incubated for 48 hours at room temperature, with occasional inversion to resuspend the particulate matter. We then filtered the extracts through Whatman filter paper #1 (Whatman, Maidstone, UK), removed the methanol using evaporation in a fume hood for 24-48 hours at room temperature, centrifuged extracts for 10 minutes at 15 ^oC, 4000 RCF to remove the remaining particulate matter, followed by storage (-20 ^oC). On the day of electrophysiological recording, we thawed the extracts and diluted them 1:50 in bath solution (see below), equivalent to 5 mg fresh plant matter starting material/ml, immediately before use.

High-throughput screening for Kv7.2/3 activation
Plant extracts were applied to human Kv7.2/3 channels expressed in HEK293 cells using a FLIPR potassium assay kit and a Fluorescence Imaging Plate Reader (FLIPRTETRA™) instrument. All chemicals used in this project were purchased from Sigma-Aldrich (St. Louis, MO) unless otherwise noted and were of ACS reagent grade purity or higher. Stock solutions of the positive controls were prepared in dimethyl sulfoxide (DMSO) or deionized water, aliquoted, and stored frozen. Plant extracts were prepared in buffer and frozen until dilution into the assay buffers on test day. The ability of each plant extract to act as a Kv7.2/3 channel opener was evaluated in a thallium (potassium ion surrogate, Molecular Devices) flux assay by Charles River Laboratories (Cleveland, OH, US). The assay was performed with the FLIPR potassium assay kit (Molecular Devices) according to the manufacturer’s instructions. For dye loading, the growth media was removed and replaced with 20 μl of dye loading buffer for 60 minutes at room temperature. For stimulation (agonist mode): 5x (5 μL) plant extract, vehicle, or control article solutions prepared in the stimulation buffer (K+-free buffer with 5 mM Tl+) was added to each well for ~5 minutes. The agonist effects of the plant extract or control articles on KCNQ2/3 channels were evaluated. The positive control was Flupirtine (9 concentrations). Data acquisition was performed via the ScreenWorks FLIPR control software that is supplied with the FLIPR System (MDS-AT). Data were analyzed using Microsoft Excel 2013 (Microsoft Corp., Redmond, WA). For each well, the raw kinetic data were reduced to the maximum or Area Under Curve fluorescence after subtracting bias and possibly applying the negative control correction. Reduced data were analyzed as follows:

For each assay plate, a Z’ factor and Signal Window (SW) were calculated:
Z’ factor = (([Agonist Control mean] – 3 x [Agonist Control STDEV]) – ([Vehicle Control mean] + 3 x [Vehicle Control STDEV])) / ([Agonist Control mean] – [Vehicle Control mean]) SW = (([Agonist Control mean] – 3 x [Agonist Control STDEV]) – ([Vehicle Control mean] + 3 x [Vehicle Control STDEV])) / [Agonist Control STDEV]
Where the stimulation buffer was dispensed to Vehicle Control wells and a high concentration of agonist positive control was dispensed to Agonist Control wells.
Concentration-response curves were fitted to the agonist positive control. Reduced data from test article wells were normalized to the vehicle and agonist control means on each plate and expressed as normalized percent activation: Normalized % Activation = ([individual we 429 ll RLU] – [Vehicle Control mean]) / ([Agonist Control mean] – [Vehicle Control mean]) where individual well RLU = the relative light units for each well to which test article is dispensed. A significance threshold of 3 standard deviations from the vehicle control mean was calculated: Significance Threshold = 3 x [Vehicle Control STDEV] / ([Agonist Control mean] – [Vehicle Control mean]) Concentration-response curves for positive agonist controls for each plate were also conducted. The positive control results confirmed the sensitivity of the test systems to agonists. The test and control samples were prepared in the stimulation buffer (a combination of low Cl439 buffer, 5 mM Tl2SO4 and water). The signal elicited in the presence of the positive agonist control (30 or 100 μM Flupirtine) was set to 100% activation and the signal from the vehicle (stimulation buffer) was set to 0% activation.

High-throughput screening for Kv1.3 inhibition
Chemicals used in solution preparation were purchased from Sigma-Aldrich unless otherwise noted and were of ACS reagent grade purity or higher. Stock solutions of plant extracts and the positive controls were prepared in water and stored frozen, unless otherwise specified. Reference compound concentrations were prepared fresh daily by diluting stock solutions into a HEPES-buffered physiological saline (HB-PS) (composition in mM): NaCl, 137; KCl, 4.0; CaCl₂, 4.8; MgCl₂, 1; HEPES, 10; Glucose, 10; pH adjusted to 7.4 with NaOH. To minimize run-down of the Kv1.3 channel currents 0.3% DMSO was added in all reference, plant extract and control solutions. The plant extracts (diluted to 2% and 0.2%, equivalent to 5 and 0.5 mg fresh plant matter starting material/ml, respectively, were loaded into 384-well polypropylene compound plates and placed in the plate well of an automated patch-clamp (APC) system, SyncroPatchTM 384PE (SP384PE; Nanion Technologies, Livingston, NJ) immediately before application to Chinese Hamster Ovary (CHO) cells (stain source, ATCC Manassas, VA; substrain source, Chan 456 Test Corporation, Cleveland, OH, US) expressing human Kv1.3. Screening was conducted by Charles River Laboratories. Extracellular buffer was loaded into the wells of the Nanion 384-well Patch Clamp (NPC-384) chips (60 μl per well). Then, cell suspension was pipetted into the wells (20 μL per well) of the NPC-384 chip. After establishment of a whole-cell patch-clamp configuration, membrane currents were recorded using the patch clamp amplifier in the SP384PE system. Plant extracts were applied to naïve cells (n = 3, where n = the number cells/concentration). Each application consisted of addition of 40 μl of 2x concentrated test article solution to the total 80 μl of final volume of the extracellular well of the NPC-384 chip. Duration of exposure to each test article concentration was five (5) minutes. The intracellular solution was (in mM): KCl, 70; KF, 70; MgCl₂, 2; EGTA, 2.5; HEPES, 10; pH adjusted to 7.2 with KOH. In preparation for a recording session, the intracellular solution was loaded into the intracellular compartment of the NPC-384 chip. The extracellular solution was the HB-PS solution described above. Kv1.3 channel currents were elicited using test pulses with fixed amplitudes: depolarization pulse to +20 mV amplitude, 200 ms duration from the holding potential of –90 mV. The test pulses were repeated with frequency 0.1 Hz: 3 min before (baseline) and 5 min after test articles addition. Kv1.3 channel current amplitudes were measured at the peak and at the end of the step to +20 mV. The positive control antagonist used was 4-aminopyridine, prepared as a 1 M stock in water; test concentrations were 1, 3, 10, 30, 100, 300, 1000 and 3000 μM.

Channel subunit cRNA preparation and Xenopus laevis oocyte injection for manual two electrode voltage-clamp (TEVC) electrophysiology
We generated cRNA transcripts encoding human Kv1.1 (KCNA1), Kv1.2 (KCNA2), Kv2.1 (KCNB1), Kv7.1, Kv7.2, Kv7.3, Kv7.4, Kv7.5 (KCNQ1-5), and KCNE3 (MiRP2) by in vitro transcription using the mMessage mMachine kit (Thermo Fisher Scientific, Waltham, MA, USA) according to manufacturer’s instructions, af 483 ter vector linearization, from cDNA sub-cloned into expression vectors (pTLNx, pXOOM and pMAX) incorporating Xenopus laevis β-globin 5’ and 3’ UTRs flanking the coding region to enhance translation and cRNA stability. We injected defolliculated stage V and VI Xenopus laevis oocytes (Xenoocyte, Dexter, MI, USA) with the channel cRNAs (0.3-10 ng) and incubated the oocytes at 16 oC in ND96 oocyte storage solution containing penicillin and streptomycin, with daily washing, for 1-4 days prior to two electrode voltage-clamp (TEVC) recording. Mutant channel cDNAs were generated by GenScript Biotech (Piscataway, NJ).

Two-electrode voltage clamp (TEVC)
We conducted TEVC at room temperature with an OC-725C amplifier (Warner Instruments, Hamden, CT, USA) and pClamp10 software (Molecular Devices, Sunnyvale, CA, USA) 1-4 days after cRNA injection. We visualized oocytes in a small-volume oocyte bath (Warner) using a dissection microscope for cellular electrophysiology. We studied the effects of plant extracts and constituents solubilized directly in bath solution (in mM): 96 NaCl, 4 KCl, 1 MgCl₂, 1 CaCl₂, 10 HEPES (pH 7.6). We introduced extracts or compounds into the oocyte recording bath by gravity perfusion at a constant flow of 1 ml per minute for 3 minutes prior to recording. Pipettes (1-2 MΩ resistance) were filled with 3 M KCl. We recorded currents in response to voltage pulses between -120 mV or -80 mV and +40 mV at 10 mV intervals from a holding potential of -80 mV, to yield current-voltage relationships. We analyzed data using Clampfit (Molecular Devices) and Graphpad Prism software (GraphPad, San Diego, CA, USA), stating values as mean ± SEM. We calculated the voltage dependence of activation (V_0.5) by measuring currents at a voltage pulse of -30 mV (Kv7) or -50 mV (Kv1) immediately following prepulse voltages between -80 mV and + 40 mV. We plotted raw or normalized tail currents versus prepulse voltage and fitted them with a single Boltzmann function.

Seizure studies
The mouse study was performed under an approved Institutional Animal Care and Use Committee protocol at the University of California, Irvine. Male C57Bl/6 mice at 2 months of
age were injected intraperitoneally (IP) with either vehicle (saline) or gentisic acid at 2, 10 or 20 mg/kg (pH7.5). After 30 minutes, mice were next injected IP with pentylene tetrazole (80 mg/kg) and then observed by a scorer blinded to the treatment, who measured the latency to first clonic seizure.

In silico docking
We plotted and viewed chemical structures and electrostatic surface potential using Jmol, an open-source Java viewer for chemical structures in 3D: http://jmol.org/. For in silico ligand docking predictions of binding to Kv7.3, we performed unguided docking to predict potential binding sites, using SwissDock with CHARMM forcefields and the AlphaFold-predicted human Kv7.3 monomer structure. We prepared channel structures for docking using DockPrep in UCSF Chimera (https://www.rbvi.ucsf.edu/chimera), with which we also generated docking figures.

Statistics and Reproducibility
All values are expressed as mean ± SEM. One-way ANOVA (with Dunnett correction for multiple comparisons in the case of seizure studies) or t-test was applied for all tests; all p values were two-sided. Electrophysiological data were confirmed in at least two batches of oocytes. Biological replicates are defined as numbers of oocytes; sample sizes are given in the figure legends.