Two KCNQ2 Encephalopathy Variants in the Calmodulin-Binding Helix A Exhibit Dominant-Negative Effects and Altered PIP2 Interaction

Heterozygous missense variants in KCNQ2, which encodes the potassium channel subunit Kv7.2, are among the most common genetic causes of severe neonatal-onset epileptic encephalopathy. Because about 20% of known severe Kv7.2 missense changes lie within the intracellular C-terminal region, improving understanding of the underlying pathogenic mechanisms is important. We analyzed the basis for the severe phenotypes of Kv7.2 A337T and A337G, variants in the C-terminal’s calmodulin (CaM)-binding Helix A. When expressed heterologously in mammalian cells, alone or in combination with wild type Kv7.2 or with wild type Kv7.2 and Kv7.3, both variants strongly suppressed channel currents. A337T channels expressed alone exhibited significantly reduced protein half-life and surface trafficking and co-immunoprecipitated less CaM. For both variants, increasing cellular phosphatidylinositol 4,5-bisphosphate (PIP2) by overexpression of PI(4)P5-kinase restored current densities. For both variants, the fraction of current suppressed by activation of M1 muscarinic receptors with 10 μM oxotremorine methiodide, which depletes PIP2, was less than for controls. During voltage-sensitive phosphatase-induced transient PIP2 depletion and resynthesize, potassium current inhibition and recovery kinetics were both markedly slowed. These results suggest that these variants may reduce currents by a mechanism not previously described: slowing of PIP2 migration between the bulk membrane and binding sites mediating channel electromechanical coupling. A novel Kv7.2/3-selective opener, SF0034, rescued current amplitudes. Our findings show that these two Helix A variants suppress channel current density strongly, consistent with their severe heterozygous phenotypes, implicate impairment of CaM and PIP2 regulation in KCNQ2 encephalopathy pathogenesis, and highlight the potential usefulness of selective Kv7 openers for this distinctive pathogenic mechanism and patient subgroup.


INTRODUCTION
Variants in KCNQ2 can underlie a severe syndrome consisting of intractable neonatal onset seizures and profound global developmental delay called KCNQ2 encephalopathy (Weckhuysen et al., 2012;Millichap et al., 2016). Indeed, recent cohort studies have shown that 10-30% of previously undiagnosed patients with early infantile epileptic encephalopathy (EIEE) have de novo heterozygous missense or single codon deletion variants in KCNQ2 (Saitsu et al., 2012;Weckhuysen et al., 2012;Kato et al., 2013;Milh et al., 2013;Olson et al., 2017). KCNQ2 was first discovered at the locus for a mild, self-limiting neonatal syndrome, benign familial neonatal epilepsy (BFNE) (Biervert et al., 1998;Charlier et al., 1998;Singh et al., 1998). KCNQ2 variants causing BFNE are usually inherited as an autosomal dominant trait with high penetrance but sometimes appears de novo. It is important to understand KCNQ2 genotypephenotype relationships so that individuals with BFNE or with rare non-pathogenic variants can be rapidly distinguished from those with EIEE-causing variants. BFNE infants have a good prognosis, and ill infants with non-pathogenic KCNQ2 variants need further diagnostic efforts. Individuals with pathogenic KCNQ2 encephalopathy variants have ended their diagnostic odyssey, have a poor long-term prognosis, and are appropriate candidates for novel therapeutic trials.
Considerable evidence indicates that CaM and PIP 2 interactions are allosterically coupled (Kosenko et al., 2012;Kosenko and Hoshi, 2013;Zaydman and Cui, 2014;Alberdi et al., 2015;Chen et al., 2015). Although Kv7.1 atomic structures have recently been obtained in the presence of CaM, with and without bound PIP 2 MacKinnon, 2017, 2020), dynamic mechanisms of Ca 2+ -CaM and PIP 2 regulation remain poorly understood. CaM binding at two α-helical segments of the Kv7.2 CT domain, Helix A and B, has been proposed to be required for ER exit and targeting to the axon (Etxeberria et al., 2008;Alaimo et al., 2009;Cavaretta et al., 2014;Liu and Devaux, 2014). PIP 2 binding is required for the voltage-dependent opening of Kv7 channels and appears to contribute directly to coupling voltage sensor movement and pore gate opening (Zaydman et al., 2013;Zaydman and Cui, 2014;MacKinnon, 2017, 2020). However, sites functionally implicated in Kv7 channel PIP 2 modulation extend beyond the binding pocket seen by cryoelectron microscopy and include conserved H, K, and R residues from the S2-3 linker, S4-5 linker, S6, the CT region upstream of the Helix A, and on linkers between the A, B, and C helices (Hernandez et al., 2008;Telezhkin et al., 2013;Zaydman and Cui, 2014;Hille et al., 2015).
Most BFNE missense variants cause mild loss-of-function when co-expressed with wild type (WT) subunits, whereas most EIEE variants cause severe dominant-negative suppression of activity. However, only a small proportion of known variants have been studied to date. Previous studies of human Kv7.2 variants from the CT domain have mostly focused on those causing BFNE (Richards et al., 2004;Alaimo et al., 2009;Orhan et al., 2014;Alberdi et al., 2015;Ambrosino et al., 2015). Recently, a heterozygous Kv7.2 EIEE variant in the proximal CT, R325G, was shown to suppress currents by directly impairing PIP 2 binding (Soldovieri et al., 2016). Here, we analyze the functional consequences of A337T and A337G, two KCNQ2 encephalopathy variants (Saitsu et al., 2012;Millichap et al., 2016), at a conserved residue in the CaM-binding Helix A ( Figure 1A). The pathogenicity of these variants has remained less certain due to rarity: To date, only one individual has been described with each. We found that both variants cause loss of current amplitude sufficient to be considered as dominant-negative, supporting the clinical diagnoses of KCNQ2 encephalopathy. We obtained evidence that the suppression of Kv7.2 + Kv7.3 channel function involves a novel pattern of disruption of PIP 2 and (for A337T) CaM interaction. Ezogabine was previously used successfully to control seizures and improve EEG in the infant with A337T (Millichap et al., 2016). We show that SF0034, a close ezogabine analog with improved potency and KCNQ2/3 selectivity (Kalappa et al., 2015), rescued current loss. Our findings highlight the importance of PIP 2 interaction in causation and potential therapy for KCNQ2 encephalopathy.

Ethics
Animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at Baylor College of Medicine, in accordance with guidelines of the Association for Assessment and Accreditation of Laboratory and Animal Care (AAALAC) and the U.S. National Institutes of Health (NIH). Clinical data were collected according to procedures reviewed and approved by the Institutional Review Board of Baylor College of Medicine. The study protocol allows inclusion of information on registered patients enrolled after informed written consent, as well as deidentified data regarding anonymous patients entered after verbal parental consent to the treating physician.  (Biervert et al., 1998;Schroeder et al., 1998) clones were provided by Thomas Jentsch and subcloned into pcDNA3.1 (Thermo Fisher Scientific). Point mutations were introduced using QuikChange (Agilent Technologies). C. intestinalis voltage sensitive phosphatase (ciVSP) cDNA (Murata and Okamura, 2007)

Cell Culture and cDNA Expression
Chinese hamster ovary (CHO) cells and human embryonic kidney (HEK) 293 T cells were maintained in Dulbecco's modified Eagle's medium (DMEM, Gibco) at 5% CO 2 and 37°C. The side chain changes resulting from the two clinical variants at A337 are shown. Alignment of human KCNQ sequences around KCNQ2 A337 (red and asterisk) shows high conservation across the gene family. Basic residues previously implicated in PIP 2 binding (Zhang et al., 2013;Kim et al., 2016a;Soldovieri et al., 2016;purple) and the canonical CaM binding IQ motif (blue), are highlighted. A337 (highlighted in red and asterisk) is near the proximal end of Helix A. (B) Representative currents from Chinese hamster ovary (CHO) cells transfected with WT Kv7.2, A337G, and A337T in response to depolarization steps from −80 mV to between −70 and +40 mV. (C) Current-voltage plots show significant current reductions in cells expressing either A337T or A337G (**p < 0.01, ***p < 0.001, Two-way ANOVA, Bonferroni test, n = 14-31, see Table 1). (D) Conductance-voltage plots show no significant differences of WT and mutant-containing channels (see Table 1). (E) Western blots of cell lysates (left lanes, **p = 0.003, One-way ANOVA, Tukey test) and surface biotinylated proteins (right lanes, ***p < 0.001, One-way ANOVA, Tukey test) from CHO cells electroporated with empty pcDNA3 vector (negative control), Kv7.2, A337G or A337T (n = 4). In plots, quantification of total channel protein normalized to GAPDH (left) and surface channel protein normalized to total channel protein are shown. Values are normalized to WT. (F) Western blot of cell lysates (left lanes) and surface membrane biotinylated proteins (right lanes) from CHO cells electroporated with WT Kv7.2 alone or with 1:1 ratio of Kv7.2 and A337G or A337T cDNAs (n = 4). The cropped blot images are from full-length gels shown as supplementary information (Supplementary Figure S2).
The extracellular solution consisted of (in mM): 138 NaCl, 5.4 KCl, 2 CaCl 2 , 1 MgCl 2 , 10 glucose, 10 HEPES, pH 7.4 with NaOH (Miceli et al., 2013). Pipette solution contained (in mM): 140 KCl, 2 MgCl 2 , 10 EGTA, 10 HEPES, and 5 Mg-ATP, pH 7.4 with KOH. Series resistance was compensated by 70% after compensation using Axopatch 200B fast and slow capacitance controls. Currents were digitally sampled at 5-10 kHz, depending on the protocol, and filtered at 5 kHz using a low-pass Bessel filter. For voltage-activation experiments, cells were held at −80 mV and depolarized in 10 mV incremental steps from −80 to +40 mV for 1 s, then stepped to 0 mV for 60 ms. Tail currents at 0 mV were fitted using the Boltzmann function: to obtain the half-maximum activation voltage (V 1/2 ) and the slope factor (k). Activation and deactivation rates were measured in SigmaPlot or Clampfit using a single exponential function: I = Ae −t/τ + C. TEA sensitivity experiments were performed as previously described (Soldovieri et al., 2016). Currents were elicited by 3 s ramp depolarizations from −80 to 40 mV. To monitor inhibition by Oxo-M, cells were pulsed every 2 s for 200 ms to +40 mV from a holding potential of −80 mV before, during and after perfusion with drug solution. In experiments using ciVSP to study recovery, cells were held at −80 mV, then pre-pulsed for 2 s at −20 mV before stepping to +100 mV for 10 s, and then held at −20 mV for 30 s to monitor recovery. Kinetic parameters for current depletion and recovery were determined by non-linear regression models in SigmaPlot. Rates of current decline after ciVSP activation were as described previously (Falkenburger et al., 2010;Alberdi et al., 2015). Current recovery for channels containing variants were fit poorly using this squared exponential but could be better fit using another sigmoidal equation, the 4-parameter logistic: y ymax y ymax t 2 ). A calculated liquid junction potential (−3.62 mV) was not corrected.
Biotinylation, Immunoprecipitation, and Western Blotting CHO cell membrane protein biotinylation was performed 48 h after electroporation, using Sulfo-NHS-LC-Biotin (Pierce) as described previously (Kosenko et al., 2012). Excess biotin was quenched using 100 mM glycine and cells were lysed. After clearing by centrifugation, NeutrAvidin agarose beads (Pierce) were used to collect biotinylated proteins. For CaM co-IP, lysates were incubated with Dynabead protein A magnetic beads (Thermo Fisher) pre-coated with guinea pig anti-KCNQ2-C2 antibodies (Surti et al., 2005). Proteins were separated on 8% (Kv7) or 12% (CaM) SDS-PAGE gels, transferred onto nitrocellulose membranes, and analyzed by western blotting. For CaM blots, PVDF membranes were used, and protein gels were transferred in 0.1 M potassium phosphate buffer (pH 7.4) overnight and then fixed in 0.2% glutaraldehyde for 45 min. Membranes were blocked and then blotted with antibodies, imaged using enhance chemiluminescence (ThermoFisher Pierce Supersignal) and film, and bands were quantified as described previously (Cooper et al., 2001;Pan et al., 2006;Jin et al., 2009). Primary blotting antibodies were guinea pig-anti-KCNQ2-C2 and anti-KCNQ3-N1, mouseanti-calmodulin (Pierce), and chicken-anti-GAPDH as loading control (Cell Signal) in 2% IgG-Free BSA (Jackson ImmunoResearch), blocking buffer was tris-buffered saline with 0.1% Tween-20 (TBS-T, Bio-Rad), and secondary antibodies were HRP-conjugated anti-guinea pig, anti-mouse or anti-chicken (Jackson ImmunoResearch). For quantification of channel or calmodulin expression in whole cell lysates, each channel band was normalized against the intensity of the anti-GAPDH band in the same lane. Bar graphs show results of normalized versus WT control samples. Band intensities were averaged across the indicated numbers of independent transfections and blots.

Cycloheximide Treatment
Forty-eight hours after transfection, HEK cells were treated with 100 μg/ml of CHX (Sigma-Aldrich) or dimethyl sulfoxide (DMSO) (Thermo Fisher) at the same start time and then processed for cell lysis at subsequent time points as shown.
In each experiment, acquisition settings for each color channel were identical for all samples. Quantification of the immunofluorescence signals along the AIS was performed as described previously (Battefeld et al., 2014). The NIS-Elements intensity profile tool was used to mark the AIS trajectory, using abrupt increases and decreases in labeling for ankyrin-G to mark the AIS origin and distal tip. For each color channel in each set of images acquired in parallel, the black (0 intensity) level was specified by measuring a minimal intensity value at a non-AIS location; this quantity was subtracted from all measurements for all images to be compared. Intensity values along each AIS trajectory were binned in 16 segments, and these values were averaged to determine mean segment intensities for each condition and fluorophore. These values were normalized to the mean maximal values measured in images of control neurons electroporated with WT Kv7.2 and Kv7.3 cDNAs to obtain relative intensity values at each distance along the AIS (Battefeld et al., 2014).

Statistics
Statistical analyses were performed in SigmaPlot (v.13). Oneor two-way ANOVA was followed by Tukey's or Bonferroni's post-hoc test for multiple comparisons. For drug treatment studies, one-or two-way repeated measures (RM) ANOVA was performed. On figure panels: *p < 0.05, **p < 0.01 and ***p < 0.001. All fluorescence intensities, band intensities, and electrophysiology analyses are reported as mean ± SEM.

Clinical Descriptions
Description of patients with EIEE and the KCNQ2 de novo variants c.1009G>A; p.Ala337Thr (Millichap et al., 2016) and c.1010C>G; p.Ala337Gly (Saitsu et al., 2012) appeared previously. The A337T individual is now age 8 years of age and has been followed clinically since birth. Electroencephalograms (EEGs) at ages 6 days, 5 months (prior to ezogabine use), and 6 months (with high dose ezogabine) were presented previously. Focal tonic seizures and clusters of flexor spasms, hypotonia, and encephalopathy were noted at age 2 days. The neonatal EEG background had primarily a burst-suppression appearance, except for brief intervals of continuity (high-amplitude delta activity with multifocal epileptiform discharges) during wakefulness. The EEG evolved by age 4 months to a modified hypsarrhythmia pattern with multifocal epileptiform activity and frequent electroclinical focal tonic seizures (several per day) continued despite serial anti-seizure medication trials including phenobarbital, topiramate, oxcarbazepine, and lamotrigine. Ezogabine treatment at age 5 months was followed closely with serum levels, EEG, and clinical observation. Ezogabine caused urinary retention at 23 mg/kg/day but this resolved easily and the treatment was resumed at 14 mg/kg/day and later increased to 21 mg/kg/day. Ezogabine exposure was associated with improvement in seizures, which was reflected in diminished epileptiform activity. The patient remained on ezogabine until it was removed from the market by the manufacturer almost 4.5 years later. Despite ezogabine, however, the patient remains globally developmentally delayed (nonverbal but can choose objects by pressing a button on an assistive device, non-ambulatory), although breakthrough seizures became rare. The A337G individual was described at 7 years old in 2012 (Saitsu et al., 2012). Very briefly, the patient was diagnosed with Ohtahara syndrome as a neonate, due to tonic seizure onset at age 7 days, burst-suppression EEG, and encephalopathy. The patient remained globally impaired at age 7 years, without the ability to speak meaningful words or walk.
To analyze the losses of total and membrane protein when A337T was expressed alone, we compared the protein stability of WT Kv7.2 and A337T channels in transfected HEK cells using the translation inhibitor, cycloheximide. After CHX treatment, A337T protein degradation was more rapid than WT (Supplementary Figure S4). Treatment with 0.1% DMSO (the CHX vehicle) gave no significant effects over 24 h, but A337T protein levels were lower than the WT, as shown above using CHO cells (see Supplementary Figure S5 for full blots).
Next, we used co-immunoprecipitation to compare effects of A337T on the binding of CaM by homomers and heteromers. Expressed alone, A337T exhibited significantly decreased ability to co-precipitate CaM, whereas binding by G279S, a missense change at the GYG selectivity filter with strongly dominantnegative effects on conduction , was similar to WT (Kv7.2: 1.00 ± 0.16; G279S: 0.71 ± 0.22; A337T: 0.24 ± 0.12; Figure 3A). However, in heteromeric channels containing the A337T variant subunit, CaM-bound levels were comparable to WT and G279S controls ( Figure 3B). Thus, for A337T, reduced protein levels were correlated with reduced half-life and CaM binding. However, the near-absence of current observed when A337T or A337G were expressed alone and the partial loss of current seen when the mutant subunits were co-expressed to mimic heterozygosity could not be explained by these mechanisms (see Supplementary Figure S6 for full blots).
In Cultured Hippocampal Neurons, Three Pore Domain Variants (A274M, A294V, and G279S) Reduce AIS Trafficking, but A337T Localizes Normally to the AIS Localization of Kv7.2/Kv7.3 heteromeric channels at the AIS is critical for their roles in regulating neuronal excitability (Battefeld et al., 2014). To assess for potential AIS trafficking effects of variants, we electroporated cDNAs into freshly dissociated hippocampal neurons at embryonic day 18-19 (E18-19) and analyzed their ability to become concentrated along the AIS as axons develop in culture. At 7 DIV, ankyrin-G is highly concentrated at the AIS but endogenous Kv7.2 and Kv7.3 remain weakly expressed and are undetectable at the AIS under our imaging conditions (Nakada et al., 2003;Rasmussen et al., 2007;Abidi et al., 2015). To maximize sensitivity for observing an effect of the pathogenic Kv7.2 variants, each was co-expressed with Kv7.3 in a 1:1 ratio, i.e., a subunit composition that produced 80% current loss in CHO cells. Neurons co-expressing WT Kv7.2 and Kv7.3 exhibited concentrated staining for both subunits in the AIS (Figure 4A). Quantification of the relative staining intensity along the AIS revealed a length profile similar to that seen in vivo (Battefeld et al., 2014), with light labeling proximally, and strong labeling in the middle and distal AIS ( Figure 4B). By contrast, three previously studied pore domain variants, G279S, A294V, and T274M (Peters et al., 2005;Orhan et al., 2014;Abidi et al., 2015), did not concentrate at any portion of the AIS and were instead retained at the soma (Figures 4C-H). Unlike these pore variants, A337T proteins were concentrated in the AIS in a pattern that was similar to that seen in neurons co-expressing Kv7.2 and Kv7.3 only (Figures 4I,J). Thus, even as homozygous heteromers, A337T-containing channels traffic to the AIS similarly to WT Kv7.2 + Kv7.3 heteromers, while the three different pore domain variants appear to strongly suppress AIS trafficking.
To analyze these mechanisms further, we co-expressed the Ciona intestinalis voltage-sensitive phosphatase (ciVSP) with channel subunits. Previous studies have shown that strong membrane depolarization of cells expressing ciVSP rapidly depletes membrane PIP 2 , and M-current is inhibited at rates reflecting net unbinding of PIP 2 from the channels (Murata and Okamura, 2007;Falkenburger et al., 2010). In CHO cells co-expressing ciVSP and WT channels, 10 s long step depolarizations induced currents that activated and then declined ( Figure 6A; see Supplementary Figure S8A). The onset and rate of current decline increased progressively with membrane depolarizations above +30 mV ( Figure 6A). Because the M1R and PIP5K experiments indicated that A337T and A337G may reduce basal PIP 2 binding, we expected ciVSP co-expression to reveal a hastening of unbinding for those variants, as seen in studies of R325G, a site 12 residues closer to the S6 membrane helix which may interact directly with PIP 2 (Soldovieri et al., 2016). Unexpectedly, in cells co-expressing ciVSP and A337T + Kv7.2 + Kv7.3, at steps to +30 mV and above, the rate of current decline was slower and the fraction of current depleted after 10 s was less than for WT (Figure 6B). At +100 mV, we found that inclusion of A337T or A337G significantly delayed ciVSP-induced current declines compared to controls (Figure 6C). The decay rate constants were 3.29 s (Kv7.2 + Kv7.3), 14.80 s (A337T + Kv7.2 + Kv7.3), and 9.38 s  Figure S8B). It is unlikely that co-expression of A337G or A337T was slowing the rate at which ciVSP activation was depleting unbound, bulk membrane PIP 2 . Instead, inclusion of the variants seemed to slow the effective rate of dissociation from the channels (see Discussion). Reduced equilibrium binding of a ligand to a single site can result from slowing of both "on" and "off " rates, if the change in "on" rate is greater. To investigate whether the variants very strongly slowed rates of PIP 2 binding, we used an established depletion-recovery protocol (Falkenburger et al., 2010). PIP 2 was extensively depleted by strong ciVSP activation (10 s at +100 mV), and then currents were monitored for 30 s at −20 mV, as PIP 2 was resynthesized ( Figure 6D). As observed previously (Falkenburger et al., 2010), recovery followed an S-shaped time course, and the rate of recovery appeared independent of the extent of prior PIP 2 depletion (see Supplementary Figure S8C). Compared to WT channels, currents recovered more slowly in channels including the A337 variants than in WT-only channels, and this appeared due to an increase in the latency in initiation of recovery (Figures 6D-F; see Supplementary Figure S9), potentially contributing to reduced apparent affinity. However, the approximately 2-fold slowing in recovery was similar to the 3-to 5-fold change in the rate of current reduction seen in the depletion experiments. To further probe PIP 2 interactions with the gating machinery, we measured activation and deactivation kinetics, since mutations near some PIP 2 binding sites can affect the rate of deactivation (Chen et al., 2015). A337T did not influence voltage-dependent activation and deactivation rates (Figures 2D,E). Together, these data show that incorporation of A337T or A337G into heteromeric Kv7.2 + Kv7.3 channels impaired PIP 2 interaction in a novel pattern, appearing to decrease occupancy at sites required for electromechanical coupling at basal conditions of whole cell patch recording, without hastening the unidirectional rate of unbinding if membrane PIP 2 was strongly depleted.

SF0034 Rescued Subthreshold Current Density in Heterozygous A337T Channels
Ezogabine, an FDA-approved drug that increases Kv7.2/7.3 currents by shifting voltage activation to more hyperpolarized voltages and increasing maximal current density, has been proposed as a targeted therapeutic in KCNQ2 encephalopathy and use with some apparent benefit has been described in small clinical case series, including in the A337T proband (Gunthorpe et al., 2012;Millichap and Cooper, 2012;Weckhuysen et al., 2013;Millichap et al., 2016). However, ezogabine exhibits modest selectivity and potency and has side effects, including acute urinary retention and blue skin discoloration, which limit its clinical usefulness (Brickel et al., 2012;Clark et al., 2015;Food and Drug Administration, U.S., 2015;Kalappa et al., 2015). The infant with neonatal onset epileptic encephalopathy and a heterozygous A337T variant was given ezogabine at 4.5 months of age, resulting in improved seizure control and interictal electroencephalographic pattern, but urinary retention limited dosing to 21 mg/kg/day (Millichap et al., 2016), which achieves a brain concentration estimated at approximately 1 μM (Gunthorpe et al., 2012). SF0034 is an ezogabine analogue with improved potency and selectivity in in vitro and in vivo (Kalappa et al., 2015). In HEK cells, SF0034 was ~5-fold more potent than ezogabine in shifting the voltage-dependence of activation of Kv7.2 + Kv7.3 heteromers but was less efficacious than ezogabine against Kv7.4 or Kv7.5 homomers. The effects of SF0034, like for ezogabine, were abolished by point mutation of Kv7.3 Trp265. SF0034 treatment was shown to slow spike frequency in mouse CA1 pyramidal neurons, but this effect was markedly reduced in littermate mice with Kv7.2 conditionally deleted from these neurons. To begin to assess whether such "second generation" openers merit exploration as candidate therapeutic agents for KCNQ2 encephalopathy, we studied the effects of SF0034 on channels including the A337T pathogenic variant.

DISCUSSION
Our analysis of human KCNQ2 variants from severely affected individuals, with particular attention to effects of co-expression with WT subunits to mimic heterozygosity as is observed in people, serves two important purposes. First, finding a markedly abnormal functional profile supports a variant's classification as pathogenic for a severe phenotype (Miceli et al., 2013;Orhan et al., 2014;Richards et al., 2015;Soldovieri et al., 2016). Second, characterization of novel variants giving severe phenotypes, even when heterozygous (i.e., expressed along with wild type subunits), builds an "intolerance map" of functionally critical sites within the protein structure and may highlight contributions to allosteric mechanisms that have been overlooked by hypothesis-driven experiments. Although heterozygous missense variants clustered near the CaM-binding Helix A and B of the intracellular CT domain account for approximately 21% of cases of KCNQ2 encephalopathy (Millichap et al., 2016), studies to date have only begun to explain why very severe phenotypes result from some variants, while others cause benign familial neonatal epilepsy (Orhan et al., 2014;Soldovieri et al., 2016;Zhang et al., 2020). Understanding how these variants confer severe outcomes may aid development of targeted novel therapies. Our analysis of Kv7.2 A337T and A337G, de novo missense variants within Helix A from patients with refractory seizures and developmental impairment (Saitsu et al., 2012;Millichap et al., 2016), yielded three main findings. First, these variants generated minimal or no detectable current when expressed alone and induced dominant-negative suppression when co-expressed with WT Kv7.2 (A337G) or with Kv7.2 and Kv7.3 (both A337T and A337G). The fractional reduction in conductance seen here was similar to that previously observed for T274M and G290D, two severe pore domain variants near the selectivity filter, and for R325G, a missense change at a proximal CT arginine residue that may contribute directly to PIP 2 binding (Orhan et al., 2014;Soldovieri et al., 2016). Thus, our findings support the conclusion that these rare variants are pathogenic and provide further support for the idea that dominant-negative current suppression is an informative in vitro signature of severe EIEE phenotypes. Second, although A337G was not studied, trafficking of A337T to the AIS was normal in hippocampal neurons, as also shown recently for R325G and I205V (Soldovieri et al., 2016;Niday et al., 2017).  (D) Current-voltage relationships for channels co-expressed with PIP5K compared to without PIP5K (***p < 0.001, two-way ANOVA, Bonferroni test, n = 5-16, see Table 1). (E) Conductance-voltage relationships from cells transfected with PIP5K compared to WT Kv7.2 without PIP5K. Activation of WT Kv7.2 expressed alone, WT + A337T, and WT + A337G are significantly shifted to hyperpolarized potentials (***p < 0.001, two-way ANOVA, Bonferroni test, n = 5-16, see Table 1). (F) Representative currents produced by the indicated combinations of heteromeric WT and mutant Kv7.2 subunits in cells co-transfected with PIP5K. (G) Current densities in cells co-expressing PIP5K compared to WT Kv7.2 + Kv7.3 (***p < 0.001, Two-way ANOVA, Bonferroni test, n = 10-20). (H) Conductance-voltage relationships from cells transfected with PIP5K compared to WT Kv7.2 + Kv7.3 expressed without PIP5K. PIP5K overexpression shifts activation significantly to hyperpolarized potentials (***p < 0.001, two-way ANOVA, Bonferroni test, n = 10-20, see Table 1).
In contrast, we showed that three pore domain mutations, a well-studied experimental mutation of the selectivity filter, G279S, and two of the most highly recurrent human KCNQ2 encephalopathy variants (Millichap et al., 2016), T274M and A294V, strongly prevent AIS trafficking. These findings indicate that severe variants fall into at least two classes with respect to trafficking, a major difference in disease mechanism that has implications for therapy. It is unknown why mutations buried within the transmembrane ion pore would prevent AIS surface expression. It is intriguing to imagine that neuronal quality control mechanisms could distinguish between available and pore-blocked channels, and this deserves more study. Third, we showed that PIP 2 -dependent modulation was altered by both A337 variants, even though this residue, unlike the R325 studied by Soldovieri et al. 2016, lies within the CaM binding Helix A at a distance from any basic residue previously implicated in direct PIP 2 interaction. Our PIP5K experiments suggest that treatments increasing PIP 2 or making channels more responsive to PIP 2 represent potential therapeutic approaches for these variants. Although the above listed conclusions are robust, our studies also exposed questions that are highly intriguing, interrelated, and for the present time, incompletely resolved, in spite of new related information on the structure of Kv7 channels obtained by cryoelectron microscopy (Sun and MacKinnon, 2020). These questions concern the impact of the variants on CaM binding, the kinetics of PIP 2 association and dissociation from the channel, and on allosteric interaction between these two modulators.

A337T CaM Binding Changes and Relevance to Pathogenicity
We found that protein half-life, total and surface protein abundance, and CaM binding were all reduced by the A337T mutation, but only when A337T subunits were expressed alone, i.e., without WT subunits. Protein stability effects were not seen for A337G, even though A337G strongly reduced current amplitudes. These rare variants, like nearly all of those found in KCNQ2 encephalopathy, cause clinical symptoms in heterozygous individuals, so the observed effects on CaM binding and protein expression may not contribute to clinical severity. Alternatively, important effects on CaM binding may be insensitively detected in the transient expression systems used here and may be informatively brought to light by expressing the mutated subunit alone. Another example suggesting the potential predictive utility of data from variants expressed alone experiments is provided by a recent analysis of R144Q, R198Q, R201C, and R201H, KCNQ2 voltage-sensor domain variants associated with severe encephalopathy phenotypes. These variants gave strong hyperpolarizing shifts in gating voltage-dependence if expressed alone but much smaller effects when co-expressed with WT subunits (Miceli et al., 2015;Millichap et al., 2016;Mulkey et al., 2017). Nonetheless, the markedly altered voltagedependence exhibited by the variants when expressed alone correlated with the delayed age of onset of seizures and impaired development, distinctive aspects of the otherwise severe clinical phenotypes. Analysis of a larger number of pathogenic variants in vitro, and, potentially, studies of mouse models where the mutant alleles are driven by native promoters (Ihara et al., 2016), V 1/2 = −23.8 ± 0.6 mV, k = 10.7 ± 0.7 mV; SF0034: V 1/2 = −34.8 ± 0.6 mV, k = 10.5 ± 0.7 mV (n = 16). (C) Conductance-voltage relationship of A337T + Kv7.2 + Kv7.3 before and after SF0034 treatment. Ctrl: V 1/2 = 23.1 ± 0.4 mV, k = 8.9 ± 0.4 mV; SF0034: V 1/2 = −32.5 ± 0.4 mV, k = 9.6 ± 0.5 mV (n = 16).
will clarify parameters for interpreting results obtained from variant-alone expression as part of efforts to assess pathogenicity and predict clinical severity. . Also, recovery from ciVSP-induced current inhibition followed a sigmoid time course well-fitted by a squared exponential function, as seen previously-an indication that binding of more than one PIP 2 molecule is required for a channel to become active, as predicted by functional analysis (Falkenburger et al., 2010) and apparent in the new atomic structure. The ability of PIP5K co-expression to shift voltage-gating to the left, for both WT homomers and heteromers (Figure 6) suggests that elevated PIP 2 concentration increased binding stoichiometry above levels minimally sufficient for voltage-gating. In sum, these results indicate that, under our recording conditions without PIP5K co-expression, PIP 2 binding is not saturated and that Kv7.2 + Kv7.3 heteromers have higher affinity than Kv7.2 homomers, consistent with earlier single channel recordings (Hernandez et al., 2008). Channels including A337T and A337G subunits gave reduced current amplitudes, whether the mutated subunits were expressed alone or co-expressed with WT subunits. The ability of M1 receptor or ciVSP activation to inhibit currents was reduced for heteromers containing mutant subunits, as if the mutations were occluding the PIP 2 depleting effects of these treatments. In addition, when either A337T or A337G was expressed alone or with WT subunits, PIP5K co-expression strongly enhanced currents to levels equal to WT-only channels, and shifted gating to hyperpolarized potentials, as also seen for WT channels. These findings suggest that the channels including the A337 variants carry out electromechanical coupling similarly to WT channels, but bind less PIP 2 under the basal conditions of our experiments. Further experiments, including recording of excised patches using application of PIP 2 analogues to assess effects on single channel open probability and kinetics, could be used to test this interpretation.
The striking observations that A337T and A337G slow the kinetics of both current inhibition and current recovery during and after a ciVSP activation step (Figure 6;  Supplementary Figures S8, S9) were unexpected. Those results could be explained if A337T and A337G exert their main effects not by altering PIP 2 binding at the channel sites directly mediating electromechanical coupling, but instead, by delaying the migration of PIP 2 between those sites and the bulk membrane. The Kv7.1 atomic structure shows conspicuous water-filled portals between the bulk cytoplasm and the intracellular mouth of the ion pore which includes patches lined by a number of basic residues MacKinnon, 2017, 2020). It is noteworthy that additional basic residue sites between Helix A and B have been shown to functionally contribute to PIP 2 interaction (Hernandez et al., 2008;Telezhkin et al., 2013;Kim et al., 2016a). These sites between Helix A and B are likely more peripherally located, but are not yet included in the known atomic structure. These basic sites may contribute to channel availability and modulation kinetics by maintaining an annular microdomain with an elevated local PIP 2 concentration around the channel and helping to orchestrate migration bidirectionally.
Structures of the Helix A from Kv7.1, Kv7.3, and Kv7.4, co-crystallized with Helix B and bound CaM, show the conserved Ala residue (homologous to KCNQ2 A337) buried under and interacting with sidechains from the CaM C-lobe (Sachyani et al., 2014;Strulovich et al., 2016;Chang et al., 2018). In the intact channel, the four subunits' Helix A-B/CaM modules form a ring below the transmembrane domain (Mruk et al., 2012;Sachyani et al., 2014;Alberdi et al., 2015;MacKinnon, 2017, 2020). CaM binding is highly dynamic (Mruk et al., 2012) and has been proposed to include multiple potential modes: binding to a Helix A alone, a Helix B alone, an A-B pair contributed by a single subunit, or crosslinked helices from two subunits (Alaimo et al., 2014;Sachyani et al., 2014;Alberdi et al., 2015;Chang et al., 2018;Sun and MacKinnon, 2020). Importantly, the Kv7.1 structure obtained with and without PIP 2 bound reveals a large, nearly 180° flipping of the entire Helix A-B/CaM complex between the two states. How the large segment between Helix A and B participates in this state change is unknown. A337T and A337G are centrally located near the pivot point for this remarkable conformational flip. They may impair PIP 2 's migration between the bulk membrane and the S4-linker binding site by altering the preferred CaM binding configuration, or by slowing the rate of the c-terminal conformational flip that follows PIP 2 binding and unbinding. Evidence for dual impact on CaM and PIP 2 was developed for another pathogenic variant, L637R, located much more distally in Helix D (Alberdi et al., 2015). There is already functional evidence that several modulatory signals, including elevated [Ca 2+ ]i and phosphorylation by PKC and protein kinase CK2, may coordinately regulate the CaM configuration and PIP 2 binding and thus, available current (Suh and Hille, 2008;Kosenko et al., 2012;Kosenko and Hoshi, 2013;Kang et al., 2014;Chang et al., 2018). It will be exciting to see a more integrated understanding of this refined and ancient mechanism emerge through future experiments.
Although ezogabine has also shown some benefits in seizure control in several KCNQ2 encephalopathy infants, including an individual heterozygous for A337T (Weckhuysen et al., 2013;Millichap et al., 2016), side effects have limited its use. SF0034 is an ezogabine analog showing greater potency and selectivity in both cell-based and in vivo models (Kalappa et al., 2015). We found that SF0034 can increase macroscopic currents of A337T-containing channels, by some measures restoring current density levels to that of WT channels in the absence of drug. Novel Kv7 openers merit further study as a candidate therapy for KCNQ2 encephalopathy due to loss-of-function variants, including those in the CT region that impair PIP 2 modulation. Our A337T patient received ezogabine in the 6th month of life, with benefits for seizures but continued severe developmental impairment. Very early diagnosis and intervention may represent the best opportunity for favorable impact on development.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

AUTHOR CONTRIBUTIONS
BT and Z-GJ performed all biochemistry, neuronal transfections, imaging, and electrophysiology experiments. BT and MX designed biochemical experiments. BT and Z-GJ designed electrophysiology