Direct neurotransmitter activation of voltage-gated potassium channels

Voltage-gated potassium channels KCNQ2–5 generate the M-current, which controls neuronal excitability. KCNQ2–5 subunits each harbor a high-affinity anticonvulsant drug-binding pocket containing an essential tryptophan (W265 in human KCNQ3) conserved for >500 million years, yet lacking a known physiological function. Here, phylogenetic analysis, electrostatic potential mapping, in silico docking, electrophysiology, and radioligand binding assays reveal that the anticonvulsant binding pocket evolved to accommodate endogenous neurotransmitters including γ-aminobutyric acid (GABA), which directly activates KCNQ5 and KCNQ3 via W265. GABA, and endogenous metabolites β-hydroxybutyric acid (BHB) and γ-amino-β-hydroxybutyric acid (GABOB), competitively and differentially shift the voltage dependence of KCNQ3 activation. Our results uncover a novel paradigm: direct neurotransmitter activation of voltage-gated ion channels, enabling chemosensing of the neurotransmitter/metabolite landscape to regulate channel activity and cellular excitability. M-current is conveyed by voltage-sensitive KCNQ channels, which are enriched in GABAergic neurons and are activated by anticonvulsants such as retigabine. Here the authors show that GABA directly activates KCNQ3, at the residue required for its anticonvulsant activity.


Results
KCNQ2-5 channels contain GABA binding sites. Focusing on chemical properties required for anticonvulsant action of retigabine and related compounds on KCNQ2/3, including negative electrostatic surface potential localized close to a carbonyl oxygen 13 , we identified GABA as a candidate. GABA, the primary inhibitory neurotransmitter, bears strong negative electrostatic surface potential near its carbonyl oxygen, in contrast to the chemically related excitatory neurotransmitter, glutamate (Fig. 1d). Strikingly, in silico docking predicted binding of GABA to KCNQ3-W265, similar to retigabine and ML-213 (Fig. 1e). GABA does not readily cross the plasma membrane, but W265 lies in a crevice apparently accessible from the extracellular side (Fig. 1f); this portion of S5 also appeared accessible in the related Kv1.2-Kv2.1 "paddle-chimera" structure, which includes lipids 14 (Fig. 1g). Accordingly, Xenopus oocyte expressed homomeric wild-type human KCNQ2-5-bound extracellularly applied 3 H-GABA, while KCNQ1, which lacks W265, did not ( Fig. 1h; Supplementary Fig. 1). Further indicative of the specificity of 3 H-GABA binding to KCNQ2-5, non-injected oocytes bound minimal 3 H-GABA ( Supplementary Fig. 1) and 3 H-GABA binding to homomeric wild-type human KCNQ3 was outcompeted by coincubation with cold (unlabeled) GABA (1 mM) (Fig. 1i). Finally, results of saturation binding experiments conducted with oocyteexpressed homomeric wild-type human KCNQ3 using 10 different concentrations of 3 H-GABA were consistent with binding to a specific site in KCNQ3 with a K d of 126 nM and a B max of 27 fmol per oocyte ( Fig. 1j; see Supplementary Fig. 1d for an expanded view of the low [GABA] portion of the curve). The B max value was similar to results obtained from previous intact oocyte radioligand binding studies of other heterologously expressed membrane proteins 15 .
GABA activation of KCNQ2/3 channels. We next focused on heteromeric KCNQ2/3 channels, the primary molecular correlate of neuronal M-current 4 . GABA activated KCNQ2/3 with increasing efficacy at more negative membrane potentials, a property important for subduing neuronal activity; thus GABA (10 µM) negative-shifted the V 0. 5 Table 6). At −60 mV, 10 µM GABA increased KCNQ2/3 current fourfold (Fig. 2f). The EC 50 for KCNQ2/3 activation was 0.85 ± 0.1 µM GABA at −60 mV (Supplementary Table 9), while the excitatory neurotransmitter glutamate had no effect on KCNQ2/3 even at 10 mM ( Fig. 2g; Supplementary Fig. 8; Supplementary Table 7). Reflecting its greater efficacy at more negative membrane potentials and the efficiency of KCNQ channels in setting membrane potential (E M ), GABA exerted potent effects on KCNQ2/3-dependent membrane hyperpolarization (EC 50 , 120 nM) (Fig. 2h). Canonical pentameric GABA A receptors, which are triggered to open when GABA binds to an extracellularly located, inter-subunit GABA-binding site, participate in either phasic or tonic inhibition, depending on their location and subunit composition 17,18 . Tonic extracellular GABA concentration in mammalian brain is calculated to be~160 nM 19,20 , and transient peak GABA concentrations of several millimolar occur in the synaptic cleft 21 . Canonical α x β 3 γ 2 GABA A Rs exhibit GABA EC 50 values of 1.0-157 µM 22,23 . Thus, KCNQ2/3 channel GABA affinity compares to that of the most sensitive α x β 3 γ 2 GABA A Rs. GABA begins to activate KCNQ2/3 immediately on wash-in; the current augmentation takes~3 min to plateau, and persists during wash-out (Fig. 2i). Time to onset of functional effects upon wash-in did not appear voltage-dependent, as it was not altered by switching to +60 mV during wash-in; the GABAaugmented current was readily inhibited by washing in the KCNQ channel inhibitor, XE991 (50 µM) (Fig. 2j) GABA activates KCNQ2/3 channels by binding to KCNQ3-W265. Our radioligand binding data ( Fig. 1h-j) demonstrate direct binding of GABA to KCNQ2-5, but we nevertheless tested for other possibilities and artifacts. At 100 µM, GABA neither induced current in, nor shifted the resting membrane potential of, water-injected oocytes, demonstrating our observations did not arise from an endogenous ion channel (Fig. 3a), neither did increasing GABA to 1 mM produce any noticeable effects (Supplementary Fig. 10). Neither did GABA, even at 1 mM, affect activity of KCNA1 (Kv1.1), a Kv channel from a different subfamily ( Fig. 3b; Supplementary Fig. 11; Supplementary Table 10). GABA B receptors are metabotropic receptors that can activate some potassium channels via pertussis-sensitive G proteins 24 . In previous studies, it was concluded that Xenopus oocytes do not express endogenous GABA B receptors 25 , which can activate some potassium channel subtypes (not reported for KCNQs), but we nevertheless tested for their possible role in the observed effects, using four approaches. These studies ruled out a role for GABA B receptors in KCNQ activation by GABA, as follows. First, GABA activated KCNQ2/3 even after overnight treatment of oocytes with pertussis toxin, which inhibits G proteins associated with GABA B activation of some potassium channels ( Fig. 3c Tables 15,  16).
We next tested the effected of GABA on native M-current in PC12 cells, a rat pheochromocytoma cell line of neural crest origin that retains many of the properties of dopaminergic neurons and has previously been used to study M-current 27,28 . Culturing PC12 cells for 4-7 days with NGF resulted in extensive neurite outgrowth and increased expression of native voltagegated potassium currents (Fig. 4c), as previously reported 27,28 . We found that 8/10 PC12 cells studied expressed potassium current sensitive to the KCNQ channel blocker XE991 (10 µM). Importantly, in seven of these eight cells, GABA (100 µM) negative-shifted the voltage dependence of the PC12 cell tail current activation, in the presence of picrotoxin and CGP55845 to block native canonical GABA A and GABA B receptors, respectively. The mean shift of −13.5 mV in the midpoint of voltage-dependent activation in PC12 cells in response to 100 µM GABA, in the presence of canonical GABAR blockers (Supplementary Table 19), was similar to that for KCNQ2/3 channels expressed in oocytes and in CHO cells (−14 mV) with 100 µM GABA application. Application of KCNQ channel blocker XE991 (10 µM) removed the GABA-dependent negative-shift in the PC12 cell tail currents (Fig. 4d).
We also recorded the effects of GABA on dorsal root ganglion (DRG) neurons isolated from adult mice, using perforated patchclamp whole-cell recordings. Cells were held at −20 mV to inactivate other Kv channels and then pulsed to −60 mV, after wash-in of inhibitors of canonical GABA A and GABA B receptors (100-300 µM picrotoxin; 10 µM CGP55845, respectively), first in the absence, and then in the presence, of GABA (100 µM). XE991 (10 µM) was subsequently washed in, in the presence of the other drugs, to inhibit M-current. From a holding potential of −20 mV, pulses to −60 mV produced a partial deactivation and then stabilization characteristic of M-current; current at −60 mV was increased by GABA and this current was decreased by XE991 (Fig. 4e). GABA (100 µM) increased the relative proportion of −60 mV current to that of −20 mV current, consistent with greater GABA efficacy at more negative voltages (Fig. 4f). Accordingly, the mean GABA-dependent current increase was threefold at −60 mV, but there was no mean change in current at −20 mV (Fig. 4g). We also quantified the deactivating portion of the −60 mV current in isolation. GABA decreased this deactivating current by~25%, consistent with fewer channels closing upon the transition from −20 mV to −60 mV, in the presence of GABA. XE991 applied in combination with GABA inhibited this deactivating current >50%, consistent with the majority of the current being passed by KCNQ channels (Fig. 4h). Thus, GABA activation of M-channels is direct, independent of the expression system, and occurs with both exogenously and endogenously expressed channels.
GABA overcomes muscarinic inhibition of KCNQ2/3 channels. Muscarinic inhibition of KCNQ channels is achieved by disruption of KCNQ regulation by phosphatidylinositol 4,5-bisphosphate (PIP 2 ). PIP 2 facilitates KCNQ channel opening, thus hyperpolarizing KCNQ channel voltage-dependent activation, enabling opening at more negative membrane potentials. Muscarinic acetylcholine receptor (mAChR) activation disrupts this, by hydrolyzing PIP 2 and/or by reducing PIP 2 sensitivity, thus positive-shifting KCNQ channel voltage dependence of activation, inhibiting activity particularly at negative potentials 29 . The lack of GABA responsiveness of KCNQ1, 2, and 4 ( Fig. 2a) effectively ruled out the possibility that endogenous mAChRs/ changes in PIP 2 were mediating GABA effects on KCNQ3. However, we further tested this and also asked whether GABA could overcome muscarinic inhibition of KCNQ2/3. For these experiments, we reverted to the Xenopus oocyte expression system. As expected, application of acetylcholine (ACh) to activate endogenous oocyte mAChRs inhibited KCNQ2/3 activity, especially at negative voltages; maximal effect occurred at 1 µM ACh and higher (Fig. 5a (Fig. 2a). Importantly, GABA overcame ACh-dependent KCNQ2/3 inhibition ( Fig. 5g; Supplementary Fig. 25; Supplementary  Table 24), indicating that GABA can override the canonical muscarinic inhibition of KCNQ2/3 for which the M-current is named. To further test this conclusion, we pre-incubated KCNQ2/3-expressing oocytes with the phosphoinositide 3/4kinase inhibitor, wortmannin (30 µM for 3 h), to prevent PIP 2 synthesis and partially deplete intracellular PIP 2 . Under these conditions, KCNQ2/3 was still robustly activated by GABA, which now (at 10 µM) shifted the V 0.5 of KCNQ2/3 activation by −24 mV ( Fig. 5h; Supplementary Fig. 26; Supplementary  Table 25). Thus, GABA activation of KCNQ2/3 can overcome the inhibitory effects of mAChR activation or direct PIP 2 depletion/prevention of PIP 2 synthesis. As a final control for the independence of GABA from mAChR regulation, we examined effects of GABA on KCNQ1-KCNE1, a heteromeric channel with 100-fold higher PIP 2 sensitivity than homomeric KCNQ1 30 . We found no effect of GABA on KCNQ1-KCNE1 activity, nor on the resting membrane potential of KCNQ1-KCNE1-expressing oocytes (Fig. 5i, j; Supplementary Fig. 27; Supplementary Table 26). Thus, KCNQ activation by GABA does not involve mAChRs or PIP 2 , and indeed can overcome these factors in KCNQ2/3 channels.

Discussion
Prior phylogenetic analyses of KCNQ sequences showed their absence in Porifera (sponges) and ctenophores, and suggested that KCNQ emerged as a single gene in a cnidarian/bilaterian ancestor. The KCNQ encoded by this single gene evolved PIP 2 sensitivity even before the Bilaterian/Cnidarian divergence, indicating the importance of this feature as early as 600 million years ago 32 . This PIP 2 sensitivity developed before the duplication of KCNQ into genes orthologous to present-day KCNQ1 and KCNQ5, creating multiple KCNQ genes per genome. Some cnidarians, including Hydra (a Hydrozoan) have multiple KCNQ genes, whereas others, such as Nematostella (an Anthozoan) have only one. Previous phylogenetic analysis suggested that cnidarian KCNQs do not fit into either of the KCNQ clades present in bilaterian genomes, which are split into KCNQ1-like and KCNQ2-like 32 . As protostome KCNQs lack the W270 equivalent ( Fig. 1a, b), it is interesting to ponder whether cnidarian KCNQs evolved the S5 W independently to the deuterostomes, or whether protostomes lost a W already present in a cnidarian/bilaterian ancestor. Either way, this places emergence of KCNQ5-W270 or its equivalent much earlier than evolution of the anchor motif for interaction with Ankyrin G, required for location at axon initial segments, which appeared in KCNQs in the time period of divergence of extant jawless and jawed vertebrates 33 . For comparison, Hydra are also known to express functional canonical GABA A receptors with pharmacology similar in some respect to mammalian GABA A receptors 34 . GABA itself is found in all organisms; glutamate decarboxylase (GAD), which catalyzes conversion of glutamate into GABA, is found in bacteria (where . g Mean dose response of KCNQ2/3 channels at −60 mV for GABA (calculated EC 50 = 0.85 µM; n = 10) and glutamate (no effect; n = 5). h GABA hyperpolarizes resting membrane potential (E M ) of unclamped oocytes expressing KCNQ2/3. Left, effects of 10 µM GABA; right, dose response; n = 10, ****P < 0.0001. i Exemplar −60 mV KCNQ2/3 current before (left, black), during wash-in of GABA (purple) and after wash-out (right, black). j Exemplar −60 mV KCNQ2/3 current before (left, black), during wash-in of GABA (purple), wash-in of XE991 (red), and after wash-out (right, black). Membrane potential was clamped at −60 mV except for a 2-min pulse to +60 mV during the early phase of GABA wash-in. k Mean activation (left) and deactivation (right) rates for KCNQ2/3 before (control) and after wash-in of GABA (n = 10); **P < 0.01; ***P = 0.0007. Activation rate was quantified using voltage protocol as in c. Deactivation rate was quantified using voltage protocol shown above decarboxylation of amino acids including glutamate aids in acid resistance 35 ) and plants (e.g., in soybean 36 ) in addition to animals.
KCNQ5-W270/KCNQ3-W265 are necessary for binding but not sufficient for GABA-dependent activation; KCNQ2 and KCNQ4 each bear an equivalent residue and bind GABA but are not activated by GABA, analogous to but more extreme than the lower retigabine sensitivity of KCNQ2 versus KCNQ3 37 . Thus, other residues are important, for determining binding affinity/ precise position and/or communicating GABA binding to the KCNQ gating machinery. The finding that KCNQ5, the modern KCNQ most closely related to the progenitor KCNQ found in primitive deuterostome nervous systems, is GABA-sensitive, is consistent with the idea that the ability to activate in response to GABA was a positive selection pressure contributing to conservation of KCNQ5-W270.
This also suggests a plausible explanation for why modern, neuronal KCNQs bear the W265 equivalent while the modernday cardiac and epithelial KCNQ1 subunit does not. After the founder KCNQ gene was duplicated in early bilaterians, tens of   38 . Conversely, it is interesting to note that retigabine itself is a weak activator of GABA receptors 39 , perhaps a further hint of similarities between the W265-based binding site in KCNQ3 and the GABA-binding site of canonical GABA A receptors. It will be of interest to determine whether cnidarian KCNQs bearing the S5 W are also sensitive to GABA, and to compare their expression pattern to that of non-W cnidarian KCNQs.
Our principal finding, that M-channels are novel, unexpected targets for GABA, with sensitivity comparable to canonical GABA A Rs in vertebrate CNS, changes current understanding of the potential mechanisms underlying GABA-mediated inhibition, which up until now was considered to involve solely GABA A, B, C receptors. The EC 50 we observe for KCNQ2/3-dependent cellular hyperpolarization of 120 nM GABA (Fig. 2h), is similar both to the Kd we observe for 3 H-GABA binding to KCNQ3 (Fig. 1j), and to the tonic level of GABA (160 nM), consistent with KCNQ2/3 channels directly binding GABA to influence neuronal membrane potentials at rest. Our findings, together with the known ability of KCNQ2/3 to regulate GABA release 8,9 , and strong expression overlap of KCNQ2/3 with GABAergic neurons 7 , suggest the potential for an underlying feedback mechanism.
Additionally, our results demonstrate that KCNQ2/3 has the capacity to act as a chemosensor of the extracellular neurotransmitter/metabolite landscape, which could enable M-channels to respond to the balance of these molecules and by doing so regulate cellular excitability over time. Metabolic shifts favoring increased extracellular GABOB over GABA would be predicted, based on our data, to favor cellular excitability, as GABOB could compete out GABA from KCNQ3-W265, resulting in a positive shift in the voltage dependence of M-current activation compared to if GABA were bound (as in Fig. 6d). It is also interesting to speculate a role for BHB activation of KCNQ2/3 channels in the anticonvulsant effects of the ketogenic diet, given that BHB is the first ketone body produced during fasting or ketogenic diets, and also in diabetic ketoacidosis. However, all these ideas require further testing in native brain tissue preparations before any firm conclusions can be drawn as to the precise physiological roles of M-current GABA sensitivity. Finally, as GABA and analogs, in particular GABOB (EC 50 of 0.12 µM), activate KCNQ2/3 channels with a greater potency (albeit lower efficacy) than retigabine and analogs, the new findings could point the way to alternative chemical spaces for therapeutic neuronal KCNQ activation.

Methods
Channel subunit cRNA preparation and oocyte injection. cRNA transcripts encoding human KCNQ1, KCNQ2, KCNQ3, KCNQ4, KCNQ5, KCNA1, or KCNE1 were generated by in vitro transcription using the T7 polymerase mMessage mMachine kit (Thermo Fisher Scientific), after vector linearization, from complementary DNA (cDNA) sub-cloned into plasmids incorporating Xenopus laevis β-globin 5′ and 3′ untranslated regions flanking the coding region to enhance translation and cRNA stability. cRNA was quantified by spectrophotometry. Mutant KCNQ3 cDNAs were generated by site-directed mutagenesis using a QuikChange kit according to manufacturer's protocol (Stratagene, San Diego, CA) and corresponding cRNAs prepared as above. Defolliculated stage V and VI Xenopus laevis oocytes (Ecocyte Bioscience, Austin, TX) were injected with Kv channel α subunit cRNAs (5-10 ng). Oocytes were incubated at 16°C in Barth's saline solution (Ecocyte) containing penicillin and streptomycin, with daily washing, for 3-5 days prior to TEVC recording or 3 H-GABA binding assays.
Radioligand binding assays. Each group of oocytes was placed in a round-bottomed, 10-ml culture tube, washed with ND96, and then resuspended in ND96 containing 100 nM γ- [2,  quantified using a surface biotinylation approach similar to previously described for oocyte membrane proteins 40 as follows: in parallel experiments to the binding assays, at 5 days post-cRNA injection, Xenopus oocytes (5 per group) expressing KCNQ1, 2, 3, 4 or 5, and water-injected controls, were washed with ND96, and gently agitated for 1 h at 4°C in sulfo-NHS-SS-biotin (Pierce) (0.5 mg/ml in ND96). Oocytes were next washed in 5 ml per group quenching solution (192 mM glycine, 25 mM Tris-HCl, pH 7.4) for 10 min to quench unreacted biotin. Oocytes were next incubated in 1 ml per group homogenization buffer (in mM, 150 NaCl, 2 CaCl 2 , 20 Tris, 2% Nonidet P-40) with 2% protease inhibitor cocktail (Pierce) and then lysed with a 20 ml pipette tip, vortexed, centrifuged at 4000 r.p.m. (10 min, 4°C ). The supernatant was mixed in Eppendorf tubes end-over-end overnight at 4°C with high-capacity streptavidin agarose beads (Thermo Fisher, Huntington Beach, CA) (20 µl per group). The beads were gently centrifuged to precipitate surfacebiotinylated membrane proteins that were bound to the avidin-coated beads, washed four times in homogenization buffer, resuspended in SDS-PAGE loading buffer containing 5% dithiothreitol, heated to 70°C for 10 min with intermittent vortexing, and then separated on 4-12% ClearPage TEO-Tricine gradient gels (CBS Scientific, San Diego, CA). Gels were stained for 20 min in Coomassie Brilliant Blue G-250 (Sigma) in methanol/acetic acid, then destained for several hours in methanol/acetic acid. Coomassie blue was used, as its staining versus protein concentration is linear over a wide protein concentration range, membrane proteins are relatively scarce in non-injected oocyte membranes, and this approach avoids epitope tagging the different KCNQs, which could lead to other artifacts such as altered GABA binding or surface expression. There was one prominent protein band in the gel lanes of KCNQ-expressing groups, and a similar but much fainter band in non-injected oocytes (perhaps corresponding to endogenous Xenopus KCNQ1 41 ). Band densities were quantified using ImageJ (NIH, Bethesda, MD), averaged for four groups of five oocytes each, and then binding assay 3 H-GABA CPM were normalized to mean band density to correct for the different surface expression of the different KCNQ isoforms. This correction did not qualitatively alter the pattern of results, which in either corrected or non-corrected datasets showed a clear distinction between binding in water or KCNQ1-cRNAinjected oocytes versus the much higher binding in oocytes expressing KCNQ2-5 ( Supplementary Fig. 1).
For the binding competition assay, the control group was treated as before. The competition group was placed in a round-bottomed, 10-ml culture tube, washed with ND96 containing 1 mM cold γ-aminobutyric acid (Sigma, Carlsbad, CA) and resuspended in ND96 containing 100 nM γ-[2,3-3 H(N)]-aminobutyric acid and 1 mM cold GABA (Sigma), and incubated for 30 min at room temperature. The competition group was then washed four times in 16°C ND96 containing 1 mM γaminobutyric acid, then transferred to individual wells in a 96-well plate and lysed in 0.2% SDS in ND96. Each oocyte lysate was transferred to a scintillation vial containing 5 ml cytoscint scintillation cocktail fluid (MP Biomedicals). Vials were capped, shaken, and then allowed to sit at room temperature for at least 30 min before scintillation counting in a Beckmann Coulter LS6500 liquid scintillation counter. For the saturation binding assays, a similar approach was taken but a range of ten different γ-[2,3-3 H(N)]-aminobutyric acid concentrations were employed (0.1-1000 nM), with versus without 1 mM cold GABA. The groups with cold GABA in addition to γ-[2,3-3 H(N)]-aminobutyric acid were used to calculate non-specific binding, and subtracted from total binding to give specific binding. CPM were converted to fmol/oocyte after correcting for counter efficiency. B max and K d values were calculated using automated nonlinear regression analysis in GraphPad prism.
Two-electrode voltage clamp. TEVC recording was performed at room temperature with an OC-725C amplifier (Warner Instruments, Hamden, CT) and pClamp8 software (Molecular Devices, Sunnyvale, CA) 3-5 days after cRNA injection as described in section above. Oocytes were placed in a small-volume oocyte bath (Warner) and viewed with a dissection microscope. Chemicals were sourced from Sigma. Bath solution was (in mM): 96 NaCl, 4 KCl, 1 MgCl 2 , 1 CaCl 2, 10 HEPES (pH 7.6). GABA and GABOB were stored at −80°C as 1 M stocks in molecular grade H 2 O and diluted to working concentrations on each experimental day. Valeric acid was stored as a 5 mM stock in Ringer's solution at 4°C and diluted to working concentrations each experimental day. β-hydroxybutyric acid was stored at 4°C as a 480 mM stock in 100% ethanol and diluted to working concentrations each experimental day. All compounds were introduced to the recording bath via gravity perfusion at a constant flow of 1 ml per min for 3 min prior to recording. Pipettes were of 1-2 MΩ resistance when filled with 3 M KCl. Currents were recorded in response to pulses between −80 mV and +40 mV at 20 mV intervals, or a single pulse to +40 mV, from a holding potential of −80 mV, to yield current-voltage relationships, current magnitude, and for quantifying activation rate. Deactivation was recorded at −80 mV after a single activating pulse to +40 mV, from a holding potential of −80 mV. For wortmannin studies, oocytes were incubated in 30 µM wortmannin for 3 h at 16°C prior to recording. TEVC data analysis was performed with Clampfit (Molecular Devices) and Graphpad Prism software (GraphPad, San Diego, CA, USA); values are stated as mean ± SEM. Normalized tail currents were plotted versus prepulse voltage and fitted with a single Boltzmann function: where g is the normalized tail conductance, A 1 is the initial value at −∞, A 2 is the final value at +∞, V 1/2 is the half-maximal voltage of activation, and V S the slope factor. Activation and deactivation kinetics were fitted with single exponential functions.
Whole-cell patch clamp electrophysiology. CHO cells (ATCC) were seeded on poly-L-lysine-treated glass coverslips in 24-well plates and transfected the next day using 0.5 µg each of mouse KCNQ2 and KCNQ3 cDNA in a CMV vector per well in normal media (DMEM with 10% FBS and 1% penicillin/streptomycin). The transfection reagent used was TransIT-LT1 (Mirus Bio LLC, Madison, WI, USA) and cells were patched 48-72 h post transfection. PC12 cells (Sigma) were grown in a 95% O 2 /5% CO 2 and humidified environment at 37°C in RPMI 1640 medium supplemented with 10% heat inactivated horse serum, 5% fetal bovine serum, and 1% penicillin/streptomycin. Media was replenished every 3 days and cultures were split once a week. For differentiation, cells were plated on 13 mm plastic coverslips, pre-treated with poly-L-lysine and collagen, in RPMI 1640 medium supplemented with 1% horse serum and 50 ng/ml nerve growth factor (NGF) and cultured for up to 7 days. Neurite outgrowth was observed from days 3-4 of differentiation and cells were used for experiments from 5 days post differentiation. Cell culture plastic ware and reagents were purchased from VWR and Fisher Scientific unless otherwise stated. For both CHO cells and PC12 cells, currents were recorded in whole-cell mode at room temperature (22-25°C) with 3-6 MΩ borosilicate glass electrodes backfilled with solution containing (in mM): 90 K acetate, 20 KCl, 40 HEPES, 3 MgCl 2 , 1 CaCl 2 , 3 EGTA-KOH, 2 MgATP; pH 7.2. Cells were continuously perfused at 1-2 ml/min with extracellular solution (ECS) containing (in mM): 135 NaCl, 5 KCl, 5 HEPES, 1.2 MgCl 2 , 2.5 CaCl 2 , 10 glucose; pH 7.4. Drugs were added to the ECS where indicated, as follows: 100 µM picrotoxin, 10 µM CGP55845, 10 µM XE991, and 100 µM GABA. All chemicals were purchased from Fisher Scientific or Sigma-Millipore.
Cells were held at −80 mV in voltage clamp before applying the voltage step protocols and currents were recorded, in response to pulses between −80 mV and +40 or +60 mV at 20 mV intervals followed by a single pulse to −30 mV, using a CV −7A Headstage (Axon Instruments, Foster City, CA, USA). Currents were amplified using a Multi-clamp 700B (Axon Instruments), low-pass filtered at 2-10 kHz using an eight-pole Bessel filter and digitization was achieved (sampling at 10-40 kHz) through a DigiData 1322A interface (Molecular Devices). The pClamp8 (Molecular Devices) software package was used for data acquisition and Clampfit was used for analysis, along with Graphpad Prism 7.0 (Graphpad). Normalized tail currents were plotted versus prepulse voltage and fitted with a single Boltzmann function.
Neuronal isolation and recordings. Adult male mice were purchased from Jackson Labs and then housed and used according to a protocol approved by the Institutional Animal Care and Use Committee of University of California, Irvine, and consistent with recommendations in the NIH Guide for the Care and Use of Laboratory Animals (8th edition, 2011). To analyze the effects of GABA on native neuronal M-current, DRG were isolated from adult male mice (2-4 months of age), and dissociated using collagenase and trypsin. DRG neurons were then cultured on poly-L-lysine and laminin-coated plastic coverslips in RPMI 1640 with 10% horse serum and 1% penicillin/streptomycin at 37°C for 1-4 days before electrophysiological recordings were performed using perforated patch clamp. Amphotericin B (VWR, Radnor, PA) was dissolved in DMSO at 5 mg/ml (stock solution), then diluted to 150-200 µg/ml in intracellular solution and sonicated with a probe sonicator on ice for 5 s, kept at room temperature in the dark, and replaced every hour with a fresh solution from the stock. The working amphotericin solution was backfilled into borosilicate glass electrodes (3-5 MΩ) previously tip-filled to~0.5 mm with intracellular solution lacking amphotericin B, immediately prior to sealing onto individual neurons. After whole-cell configuration was achieved and capacitance stabilized, whole-cell recordings were performed essentially as described in the prior paragraph, with the following differences: neurons were held at −20 mV to inactivate other channels before the test pulses; offline leak subtraction was performed.
Sequence analysis and modeling. Chemical structures and electrostatic surface potentials (range, −0.1 to 0.1) were plotted using Jmol, an open-source Java viewer for chemical structures in 3D: http://jmol.org/. For docking, the X. laevis KCNQ1 cryoEM structure 42 was first altered to incorporate KCNQ3/KCNQ5 residues known to be important for retigabine and ML-213 binding, and their immediate neighbors, followed by energy minimization using the GROMOS 43B1 force field 43 , in DeepView 44 . Thus, X. laevis KCNQ1 amino acid sequence LITTLYIGF was converted to LITAWYIGF, the underlined W being W265 in human KCNQ3/ KCNQ5 and the italicized residues being the immediate neighbors in KCNQ3/ KCNQ5. In addition, X. laevis KCNQ1 sequence WWGVVTVTTIGYGD was converted to WWGLITLATIGYGD, the underlined L being Leu314 in human KCNQ3/KCNQ5 and the italicized residues being the immediate neighbors in KCNQ5 and/or KCNQ3. Surrounding non-mutated sequences are shown to illustrate the otherwise high sequence identity in these stretches. Unguided docking of GABA, retigabine, and ML-213, to predict native binding sites, was performed