PIP2-dependent coupling of voltage sensor and pore domains in Kv7.2 channel

Phosphatidylinositol-4,5-bisphosphate (PIP2) is a signaling lipid which regulates voltage-gated Kv7/KCNQ potassium channels. Altered PIP2 sensitivity of neuronal Kv7.2 channel is involved in KCNQ2 epileptic encephalopathy. However, the molecular action of PIP2 on Kv7.2 gating remains largely elusive. Here, we use molecular dynamics simulations and electrophysiology to characterize PIP2 binding sites in a human Kv7.2 channel. In the closed state, PIP2 localizes to the periphery of the voltage-sensing domain (VSD). In the open state, PIP2 binds to 4 distinct interfaces formed by the cytoplasmic ends of the VSD, the gate, intracellular helices A and B and their linkers. PIP2 binding induces bilayer-interacting conformation of helices A and B and the correlated motion of the VSD and the pore domain, whereas charge-neutralizing mutations block this coupling and reduce PIP2 sensitivity of Kv7.2 channels by disrupting PIP2 binding. These findings reveal the allosteric role of PIP2 in Kv7.2 channel activation.

P hosphoinositides are major constituents of biological membranes and key regulators of fundamental biological processes including signal transduction, membrane trafficking, and cytoskeletal dynamics 1 . Among the phosphoinositides, phosphatidylinositol-4,5-bisphosphate (PIP 2 ) in the plasma membrane serves as a critical cofactor for many ion channels despite its low abundance (~1% of total acidic lipids) 2,3 . The affected channels include inward rectifier potassium (K + ) channels 4 , voltage-gated calcium channels, transient receptor potential channels, hyperpolarizationactivated cyclic nucleotide-gated channels, and voltage-gated potassium (K v ) channels 2,5 .
PIP 2 activates all five members of the K v channel subfamily Q (K v 7.1-K v 7.5) which control excitability of neuronal, sensory, and muscle cells 6,7 . Encoded by KCNQ1-KCNQ5 genes 7 , each K v 7 subunit has six transmembrane segments 8,9 . The first four segments (S1-S4) comprise a voltage-sensing domain (VSD) with the S4 being the main voltage-sensor 8,9 . The pore domain consists of the last two segments (S5-S6) flanking the pore loop which contains a highly conserved sequence and structure for K + selectivity and permeability 8,9 . The C-terminal intersection of four S6 segments constitutes the main gate 9,10 . Each subunit also has a long intracellular C-terminal tail that harbors four α-helices (A-D) 11 . Helix-A and Helix-B interact with calmodulin (CaM) 11 , Helix-C mediates inter-subunit interaction, while Helix-D specifies the subunit assembly as a homotetramer or a heterotetramer 11 .
Despite the common core structure, each K v 7 subunit follows a distinct, cell-specific distribution that dictates its physiological roles in different tissues 7,8 . In the heart, K v 7.1 assembles with an auxiliary β subunit KCNE1 to produce the slow K + current critical for repolarizing cardiac action potentials (APs) 7,8 . K v 7.4 is primarily found in cochlear hair cells of the inner ear 8 . In the central nervous system, K v 7 channels are mostly heterotetramers of K v 7.2 and K v 7.3, and to a lesser extent heterotetramers of K v 7.3 and K v 7.5 and homomeric K v 7.2 channels 12,13 . Neuronal K v 7 channels produce slowly activating and non-inactivating K + current (I M ) that suppresses repetitive firing of APs 12,13 , and dominant mutations in their subunits cause neonatal epilepsies including benign familial neonatal epilepsy (BFNE) and epileptic encephalopathy (EE) (rikee.org) [14][15][16][17] . EE is a collection of epileptic syndromes accompanied by profound neurodevelopmental delay and psychomotor retardation 18,19 .
K v 7 channels are inhibited by membrane PIP 2 depletion upon activation of G q -coupled receptors 2,6,13,20 . The underlying mechanism has been extensively investigated in K v 7.1 3 . Voltageclamp fluorometry studies have demonstrated that depolarization can activate the VSD of K v 7.1 but fails to open the pore upon PIP 2 depletion 21 , suggesting that PIP 2 is crucial for coupling the VSD to the pore domain. In the cryo-EM structure of K v 7.1 channel in complex with KCNE3 and CaM, PIP 2 interacts with the S2-S3 and S4-S5 linkers, and this interaction may facilitate the channel opening by converting the unstructured loop between the S6 to Helix-A (pre-Helix-A) to a helix 22 . In addition to the S2-S3 and S4-S5 linkers and S6 as potential PIP 2 binding sites in K v 7.1 21 , in vitro binding studies with helices A-D of K v 7.1 have also identified basic residues in distal Helix-B that interact with PIP 2 23 .
Despite the accumulating mechanistic insights into PIP 2dependent modulation of K v 7.1 3 , it remains unclear whether neuronal K v 7 channels are regulated by the same PIP 2 binding residues and mechanism as K v 7.1. There are several differences in PIP 2 -dependent modulation between K v 7.1 and neuronal K v 7 channels. First, PIP 2 sensitivity of K v 7.1 is regulated by KCNE1 24 , whereas PIP 2 directly modulates neuronal K v 7 channels without auxiliary β subunits 13 . Second, previous electrophysiology studies with site-directed mutagenesis have suggested potential PIP 2 binding sites unique to K v 7.2 and K v 7.3 including the regions between Helix-A and Helix-B (AB linker) and between Helix-B and Helix-C (BC linker) 14,[25][26][27] . Third, the AB linker of K v 7.2 is much longer than that of K v 7.1. Importantly, some epilepsy variants in K v 7.2 and K v 7.3 disrupt the channel sensitivity to the changes in cellular PIP 2 level 14,25,28,29 . Therefore, detailed investigation of how PIP 2 regulates neuronal K v 7 channels can increase our understanding of their physiological function in neurons and facilitate the development of new therapeutic strategies against epilepsy.
The first step toward understanding the molecular action of PIP 2 on neuronal K v 7 channels is to identify PIP 2 binding sites. To achieve this, we employed all-atom molecular dynamics (MD) simulations. This technique has been successfully employed to provide atomic-level structural insights on lipid-protein interactions in membrane proteins with high spatiotemporal resolution [30][31][32][33][34] , in close agreement with the experimental data [30][31][32][33][34][35] . We chose to identify PIP 2 binding sites in homomeric K v 7.2 channels for several reasons. First, they produce robust K + currents upon depolarization whereas homomeric K v 7.3 channels are nonfunctional [36][37][38][39] . Second, conditional deletion of K v 7.2 but not K v 7.3 results in cortical hyperexcitability, spontaneous seizures, and high mortality in mice 40 . Third, there are significantly more epilepsy mutations found in KCNQ2 than KCNQ3 (ClinVar Database, NCBI) 41,42 , and current suppression of homomeric K v 7.2 channels is a common feature of EE variants of KCNQ2 43 .
Here our MD simulations reveal multiple PIP 2 binding sites in homomeric K v 7.2 channels with more sites in the open state than the closed state. These sites include the S2-S3 linker, S4, S4-S5 linker, S6, pre-Helix-A, AB linker, Helix-B, and BC linker. Chargeneutralizing mutations of four PIP 2 -binding residues (R214Q in the S4, K219N in the S4-S5 linker, R325Q in pre-Helix-A, and R353Q in the AB linker) disrupt PIP 2 binding to the mutated residues and decrease current densities of K v 7.2 channels. Importantly, R214Q, K219N, and R325Q mutations reduce channel sensitivity to PIP 2 depletion, with the triple R214Q/K219N/R353Q mutation inducing the largest effect. Our simulations further show that R214Q, K219N and R325Q mutations decouple the VSD activation from the pore domain of K v 7.2 channels, while the R353Q mutation blocks the PIP 2 -induced increase in the propensity of helices A and B to interact with the inner leaf of the bilayer. These findings offer detailed mechanistic insights into PIP 2 -dependent modulation of K v 7.2 channels.

Results
Differential PIP 2 binding in closed and open K v 7.2 channels. To identify PIP 2 binding sites, we performed all-atom MD simulations on human K v 7.2 channels in explicit lipid bilayers composed of phosphatidylcholine and PIP 2 . By adopting an integrative structural modeling approach using X-ray and cryo-EM data, we first constructed the open and closed states of K v 7.2 channels within explicit lipid bilayers (Fig. 1a). The stability of the resulting models was investigated and reported in our previous publication 14 . At the beginning of the lipidbinding simulations, 8 PIP 2 lipid molecules (2.2% of the total lipid in each leaflet) were distributed around the channel with their starting positions randomized in each of the three independent, 500ns-long simulation replicates (Fig. 1b, Table 1). Differential binding of PIP 2 lipids to open and closed channels was captured and presented as the PIP 2 headgroup occupancy maps extracted from the entire simulation trajectory set for each state (Fig. 1c-d). Upon binding, PIP 2 remained stably bound throughout the rest of the simulation time in all identified binding sites in both states ( Fig. 1e-f).
We discover that the PIP 2 headgroup interacts with 3 distinct sites in the closed state (Fig. 1c): Site-C1 (intracellular N-terminal tail), Site-C2 (the S2-S3 linker), and Site-C3 (the interface formed by Helix-A, the S2-S3 and AB linkers) (Fig. 1c). Notably, repeating the lipid-binding simulations on the recently solved cryo-EM structure of a closed K v 7.2 channel 44 verified the formation of the same PIP 2 binding sites ( Supplementary Fig. S1). In the open state, we have identified four distinct PIP 2 -binding sites, all enriched with basic residues: Site-O1 (the interface formed by the S2-S3 and AB linkers), Site-O2 (distal Helix-B), Site-O3 (the interface formed by the N-terminal tail, the S2-S3 linker, distal Helix-B, and the BC linker), and Site-O4 (the interface formed by the ends of S4 and S6, the S4-S5 linker, and pre-Helix-A) (Fig. 1d).
Although PIP 2 interact with the N-terminal tail and the S2-S3 and AB linkers in both closed and open states, there are key differences in the PIP 2 binding sites of these two states. PIP 2 headgroup occupancies at the four sites in the open channel are more focused than those at the three sites in the closed channel (Fig. 1c, d). Importantly, PIP 2 localizes to the periphery of the VSD in the closed channel, whereas PIP 2 binding spreads to a larger area in the open channel including the cytoplasmic ends of the VSD and S6, pre-Helix-A, distal Helix-B, and the BC linker (Fig. 1c, d), suggesting that the opening of K v 7.2 channels involves PIP 2 interaction with VSD, the gate, and intracellular helices. independent simulation replicates were performed in a tetrameric K v 7.2 channel by randomly shuffling the initial positions of PIP 2 lipids, as shown by different colors (Runs 1-3). c-d Volumetric maps of PIP 2 headgroup occupancy extracted collectively from the simulation trajectories are shown as wireframes overlaid on the K v 7.2 channel structure for the closed state (c) or the open state (d). We label the PIP 2 binding sites according to their proximity to the pore region. In the closed state, PIP 2 headgroups bind to Site-C1 (N-terminal tail), Site-C2 (the S2-S3 linker), and Site-C3 (at the interface formed by Helix-A, the S2-S3 linker and the AB linker). In the open conformation, PIP 2 headgroups are observed to bind Site-O1 (the interface formed by the N-terminal tail, the S2-S3 linker, distal Helix-B, and the BC linker), Site-O2 (distal Helix-B), Site-O3 (the interface formed by the S2-S3 linker, distal Helix-B, and the BC linker), and Site-O4 (the interface formed by the distal ends of the S4 and S6, the S4-S5 linker, and pre-Helix-A). e-f Distance between the center of mass (COM) of PIP 2 headgroup and the COM of the binding site is plotted against simulation time. Initially all PIP 2 molecules were placed at least 15 Å away from any of the binding sites captured in K v 7.2 channel structure in the closed state (e) and the open state (f). Once PIP 2 lipids bind to any of the binding sites, they remained stably bound throughout the simulation time.
Characterizing specific lipid-protein interactions in K v 7.2 channels. To map PIP 2 -binding sites, we analyzed lipid-protein interactions during the last 200 ns of MD trajectories by quantifying the contact probability between each moiety of the PIP 2 headgroup and the PIP 2 -binding residues. In the closed channel, Site-C1 is formed by a cluster of basic residues (K76, R87, and R89) in the intracellular N-terminal tail. R87 and R89 exclusively interact with P5-phosphate, whereas K76 establishes contacts with all the hydroxyl and phosphate groups on the inositol ring in PIP 2 ( Supplementary Fig. S2a, d). In Site-C2, a cluster of basic residues in the S2-S3 linker (R153, R158, and K162) coordinate PIP 2 . R153 and K162 show high contact probabilities to P4phosphate and P5-phosphate, whereas R158 interacts with all groups in PIP 2 ( Supplementary Fig. S2b, e). Site-C3 is formed by R155 in the S2-S3 linker, Y347 in Helix-A, and R353 in the AB linker. R155 has high contact probabilities for P5-phosphate and the hydroxyl group at position 6 of the inositol ring, whereas R353 interacts preferentially with P4-and P5-phosphates ( Supplementary Fig. S2c, f).
Overall, PIP 2 binds to many more basic residues in the open state than the closed state. However, both states share 4 common PIP 2 binding residues including R87 at the N-terminal tail, R153 and K162 in the S2-S3 linker, and R353 in the AB linker. Interestingly, the simulation trajectories (Supplementary Movies 1-3) show that R353 comes in contact with PIP 2 first after which the lipid forms contacts with other residues in Site-O1, suggesting that R353 might act as an initial anchor point of PIP 2 in Site-O1.
Charge-neutralizing mutations of potential PIP 2 binding residues disrupt voltage-dependent activation of K v 7.2 channels. PIP 2 is required for activation of all K v 7 channels 3 . To test the functional impact of PIP 2 binding sites identified by our MD simulations, we introduced charge-neutralizing mutations in select residues that had high contact probability to PIP 2 headgroups. In Site-O4, we introduced R214Q in the distal S4, K219N in the S4-S5 linker, and R325Q in pre-Helix-A ( Fig. 3a-i). Since MD simulations identified R353 as an initial anchoring point for PIP 2 in Site-O1, we also made R353Q in the AB linker ( Fig. 3j-l). To test if charge-neutralizing mutations disrupt PIP 2 binding to mutated basic residues, we performed 500-ns MD simulations after introducing each mutation in the PIP 2 -bound conformation of wild-type (WT) channels and monitored the distance between the PIP 2 headgroup and the mutated residues.
Upon introducing R214Q, we observed that the bound PIP 2 dissociated from the mutated residue and Site-O4 and diffused to basic residues in distal Helix-B in 2 out of 3 simulations (Fig. 3a-c). The introduction of K219N or R325Q resulted in dissociation of PIP 2 from the mutated residue in all simulations ( Fig. 3d-i). Compared to the WT channel (Fig. 2d, h), these mutations also decreased PIP 2 binding to K319 and Q323 but not to other basic residues in Site-O4 ( Fig. 3e-f, h-i). Introduction of R353Q resulted in dissociation of PIP 2 from the mutated residue in all simulations, although PIP 2 remained bound to K162 and R165 in Site-O1 ( Fig. 3j-l).
To test if these mutations affect voltage-gated activation of K v 7.2 channels, we performed whole cell patch clamp recording in CHO hm1 cells 14,25,45 , which display depolarized resting membrane potential (V m ) due to a low level of endogenous K + channels 45,46 . Application of depolarizing voltage steps from −100 to +20 mV in cells transfected with GFP and WT K v 7.2 produced a slowly activating outward K + current with peak current density of 26.5 ± 1.4 pA/pF at +20 mV ( Fig. 4a-c, Supplementary Fig. S3). Current activation was sigmoidal, and full activation was reached from 0 mV step with half-maximal current activation potential (V 1/2 ) of −30.5 ± 0.5 mV ( Fig. 4a-d, Table 2). Due to this increase in outward K + currents, these cells also displayed hyperpolarized V m compared to cells transfected with GFP alone (Supplementary Table S1).
Compared to WT channels, K v 7.2-K219N channels produced K + currents with significantly smaller peak current density (8.2 ± 1.6 pA/pF at +20 mV), a large depolarizing shift in V 1/2 (−13.7 ± 1.7 mV), and a slower activation kinetic ( Fig. 4a-e, Table 2). The R325Q mutation significantly reduced peak current density by~90% (3.5 ± 0.4 pA/pF at +20 mV) ( Fig. 4a-c). At −20 to +20 mV, the R325Q-transfected cells produced currents that were larger than those in cells transfected with GFP alone ( Supplementary Fig. S3g), suggesting that K v 7.2-R325Q channels are functional. The peak current densities of K v 7.2-R214Q and K v 7.2-R353Q channels were also decreased by~30%, but their voltage dependence and activation time constant were unaffected ( Fig. 4a-e, Supplementary Figs. S3-S4). Similar surface expressions were observed for WT and all tested mutants except for K v 7.2-R214Q, which displayed lower surface expression than the WT (61.8 ± 11.0% of WT, Supplementary Fig. S5). The surface/total protein ratio of all tested mutants did not differ from WT K v 7.2 ( Supplementary Fig. S5e), suggesting that none of the mutations affected the proportion of K v 7.2 to express on the plasma membrane.
Charge-neutralizing mutations of potential PIP 2 binding residues alter PIP 2 sensitivity of K v 7.2 channels. To test if charge-neutralizing mutations alter gating modulation of K v 7 channels by PIP 2 , we increased cellular PIP 2 level by co-transfecting phosphatidylinositol-4-phosphate 5-kinase (PIP5K) which converts phosphatidylinositol 4-phosphate to PIP 2 47 . Since the endogenous membrane level of PIP 2 is not enough to saturate K v 7 channel  Table 2). However, the fold increases in their PIP5K-induced current potentiation were comparable to the WT channels (4.7 ± 0.4 fold for R214Q, 6.2 ± 0.5 fold for K219N, 4.0 ± 0.2 fold for R353Q) (  Table 2), suggesting that the chargeneutralizing mutation of a single residue may not be sufficient to fully dissociate PIP 2 from Site-O4 and Site-O1 (Fig. 3b,e,h,k).
Therefore, we next generated the R214Q/K219N double mutant (DM) and the R214Q/K219N/R353Q triple mutant (TM). Compared to K v 7.2-WT channels, both DM and TM channels displayed smaller peak current densities (DM = 12.5 ± 0.6 pA/pF, TM = 12.9 ± 1.3 pA/pF) and activated at more depolarized voltages with slower activation kinetics ( Fig. 4a- To further investigate PIP 2 sensitivity of mutant channels, we examined current decay upon membrane PIP 2 depletion. To achieve this, we coexpressed Danio rerio voltage-sensitive phosphatase (Dr-VSP) 14 . Upon activation of Dr-VSP, K + currents through WT channels reached a maximal decay of at +100 mV (0.52 ± 0.02, Fig. 5a-c). K v 7.2-R353Q channels displayed larger current decays at +40 and +60 mV, but showed similar decays to WT channels at +100 mV ( Fig. 5a-c). In contrast, the VSP-induced current decays of R214Q and K219N channels were smaller than WT channels and were minimal in K v 7.2-R325Q, DM, and TM channels (Fig. 5a-c), indicating these mutants have decreased sensitivity to PIP 2 depletion. PIP 2 -mediated correlated motions of the VSD and the pore domain. Voltage-dependent conformational changes of the VSD and the pore domain are critical for the gating of K v channels 49 . To examine PIP 2 -mediated allosteric interactions, we investigated the communities formed in the WT and mutant channels by employing dynamic network analysis and Pearson correlation. These communities correspond to sets of residues that move in a correlated manner during the MD simulations and thus represent strongly connected regions. We detected the presence of a pronounced community connecting the end of S4, the S4-S5 linker, S6, and pre-Helix-A in WT channels (Fig. 6a). The thickness of the edges connecting the amino acids corresponds to the strength of the correlation between them. All mutant channels show thinner edges within this community (Fig. 6b-d), indicating weaker correlated motions in this area compared to WT channels. Interestingly, we also detected the emergence of additional (sub) communities in all the studied mutants ( Fig. 6b-d), in line with the described weaker interactions in this region. The presence of 2 protein is shown in ribbon representation with S4 in red, helices A and B in brown, and the rest of the protein in ice blue. A PIP 2 lipid (carbon atoms in yellow, oxygen in red, and phosphorus atoms in tan) and the residues of the binding pocket at each site at the end of simulations are shown in sticks (basic residues in blue, polar in green, acidic in red, and hydrophobic in white). e-h The contact probability for each chemical moiety of PIP 2 headgroup with the key residues in the binding Site-O1 (e), Site-O2 (f), Site-O3 (g), and Site-O4 (h). The chemical moieties of PIP 2 headgroup include P1-phosphate (black), P4-phosphate (red), P5-phosphate (orange), 2-hydroxyl (cyan), 3-hydroxyl (brown), and 6-hydroxyl (pink) group of the inositol ring. A heavy-atom distance cutoff of 4 Å was used to define a contact between a protein residue and a phosphate group of PIP 2 , whereas a 3.5 Å cutoff was used to define a contact between a residue and a hydroxyl group of PIP 2 . Analysis of the contact probabilities was performed over the last 200 ns of the simulation trajectories. The number of PIP 2 interacting events: Site-O1 (n = 6), Site-O2 (n = 5), Site-O3 (n = 2), and Site-O4 (n = 2). Data represent mean ± SEM for analysis of each monomer in 3 independent trajectories. COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02729-3 ARTICLE COMMUNICATIONS BIOLOGY | (2021) 4:1189 | https://doi.org/10.1038/s42003-021-02729-3 | www.nature.com/commsbio disconnected communities suggests uncorrelated dynamics of this region (Site-O4) after decreased PIP 2 binding (Fig. 3). These data suggest that R214Q, K219N, and R325Q mutations disrupt the correlated motions of the VSD and the S6 gate in the pore domain of K v 7.2. The most significant PIP 2 -induced conformational change was observed in the cytoplasmic helices A and B (Fig. 7a-b). The effect was quantified by calculating the orientation of the helical pair with respect to the membrane normal (Fig. 7c). In a PIP 2 -free lipid bilayer, these helices fluctuate around their initial position (θ = 83.8 ± 13.2°) and remain in a largely solvent-exposed conformation (Fig. 7a,c). In the presence of PIP 2 , the helices adopt a conformation where they interact directly with the lipid bilayer (θ = 95.5 ± 3.7°) (Fig. 7b-c). This large-scale conformational change also forms a pathway for PIP 2 to move along Helix-B and ultimately bind to Site-O4 (Fig. 7d, Supplementary Movie 4). Since R353 in the AB linker acts as an initial anchor point for PIP 2 binding ( Supplementary Movies 1-3), we next tested if a charge-neutralizing  Cells were held at −80 mV. K + currents were evoked by depolarizing voltage steps for 1.5 s from −100 mV to +20 mV in 10-mV increments, followed by a step to 0 mV for 300 ms. a Representative traces after subtraction of leak currents. Leak current was defined as non-voltage-dependent current from GFP-transfected cells. Note different Y-axis scales for systems with and without PIP5K. b Average peak current densities (pA/pF) of K v 7.2 WT or mutant channels with or without PIP5K coexpression at all voltage steps. c Average peak current densities of WT or mutant K v 7.2 channels at +20 mV. p values are computed from one-way ANOVA post-hoc Fisher's test. The source data for Fig. 4b and Fig. 4d are available in Figshare 93 . Data represent the mean ± SEM. One-way ANOVA with post-hoc Fisher's multiple comparison test was used. GFP + selected variant: *p < 0.05 for K v 7.2 WT vs. mutant (**p < 0.01, ***p < 0.005); † p < 0.05 for K v 7.2 WT + PIP5K vs. mutants + PIP5K ( † † p < 0.01, † † † p < 0.005);^p < 0.05 for the difference between -PIP5K and +PIP5K within the same transfection (^^p < 0.01,^^^p < 0.005). mutation at this residue (R353Q) affects the PIP 2 -induced conformational change. We found that the R353Q mutation moves the helices back toward a more solvent-exposed conformation (θ = 86.7 ± 1.9°) (Fig. 7c).

Discussion
Neuronal K v 7 channels are known as the "M-channels" due to their inhibition by the activation of M1 and M3 muscarinic acetylcholine receptors 13 . PIP 2 hydrolysis underlies current inhibition of M-channels, bringing attention to K v 7-PIP 2 interaction 6,20 . Increasing the PIP 2 level enhances current density and the open probability of K v 7.2 channels and induces a hyperpolarized shift in their voltage-dependence 6,14,25,28,48 . However, the detailed mechanism underlying PIP 2 -dependent regulation of neuronal K v 7 channel remains unclear. In this study, we address this knowledge gap by identifying PIP 2 interaction sites in both open and closed K v 7.2 channels.
Our all-atom MD simulations have revealed PIP 2 localization to 3 sites at the periphery of the VSD in the closed K v 7.2 channel, whereas PIP 2 binds to 4 distinct sites at the VSD, the pore domain, and intracellular helices in the open channel (Fig. 1,  Supplementary Figs. S1-S2). The common PIP 2 binding domains in both open and closed channels are the intracellular N-terminal tail, the S2-S3 and AB linkers. Importantly, PIP 2 binding in the open channel is coordinated by multiple basic residues from different functional domains of K v 7.2.
Our identification of the S2-S3 and S4-S5 linkers as PIP 2 binding domains is consistent with a previous simulation study reported by Zhang et al. 50 . However, our study has identified many more PIP 2 binding regions in K v 7.2 channel compared to Zhang et al. These include the intracellular N-terminal tail, the distal ends of the S4 and S6 as well as pre-Helix-A, Helix-A, the AB linker, Helix-B, and the BC linker in the intracellular C-terminal tail. The difference could be attributed to structural templates used in each study. Our homology model of K v 7.2 was based on the cryo-EM structure of K v 7.1 containing cytoplasmic helices A-C and a part of intracellular N-terminal tail 10 . In contrast, Zhang et al modeled the K v 7.2 transmembrane domains only (residues 95-337) based on the crystal structures of K v 1.2 and an activated bacterial K + channel KcsA 50 . In the following sections, we discuss the functional implications of the additional PIP 2 binding sites, which we have identified in this study.
To the best of our knowledge, our MD simulations have identified a novel PIP 2 -interacting interface in the open K v 7.2 channel, Site-O4, which is comprised of the distal ends of S4 and S6, the S4-S5 linker, and pre-Helix-A. We show that R325 in pre-Helix-A interacts with PIP 2 (Fig. 2), and its charge-neutralizing mutation, R325Q, abolishes this interaction (Fig. 3), basal current, and the sensitivity to PIP 2 (Figs. [4][5]. Severe reduction in current expression and PIP 2 sensitivity has also been reported in K v 7.2 channels containing recurrent EE variant R325G 28 which causes drugresistant seizures, neurodevelopmental delay, and intellectual disability 18,51 . The R325 is the first residue in a "RQKH" motif conserved in all K v 7 subunits 3,22,28 . This motif in K v 7.1 undergoes a PIP 2 -induced conformational change from an unstructured loop to a helix 22 , further supporting that PIP 2 binding to R325 is involved in activation of K v 7.2 channels. In Site-O4, PIP 2 also interacts with R214 in distal S4 and K219 in the S4-S5 linker in the open channel ( Figs. 1 and 2). Although the S4-S5 linker was previously reported to mediate PIP 2 modulation of K v 7.1-K v 7.3 3,21,50,52 , to the best of our knowledge, the identification of R214 as a PIP 2 binding site in K v 7.2 is a novel finding. R214 is also the target of two epilepsy mutations 53-55 , supporting its functional importance. Although the chargeneutralizing mutations R214Q and K219N reduce PIP 2 interactions with the mutated residues (Fig. 3), their reduced sensitivity to PIP 2 depletion (Fig. 5) suggests increased PIP 2 affinity to other residues of the channel. K219N, but not R214Q, induces a depolarizing shift in voltage dependence (Fig. 4) similar to R243A and K248A in the corresponding residue in K v 7.3 27,52 . However, the double R214Q/K219N mutation further shifts the activation curve to a depolarized potential and induces less sensitivity to PIP 2 depletion compared to single mutants (Figs. 4-5), demonstrating the synergetic effects. These findings suggest that PIP 2 interaction with both the distal S4 and the S4-S5 linker is needed for robust voltage-dependent activation of K v 7.2 channels.
Previous studies proposed the role of PIP 2 in coupling of the VSD activation to pore opening in K v 7.1 and K v 7.3 10,21,22,52,56 . In K v 7.2 channels, we observe that PIP 2 binding to Site-O4 forms an allosteric network of interactions, leading to a correlated motion of the main voltage sensor S4, the S4-S5 linker, the gate S6, and pre-Helix-A (Fig. 6), suggestive of the coupling of the VSD to the pore domain. Charge neutralizing mutations (R214Q, K219N, or R325Q) induce PIP 2 dissociation from the mutated residues (Fig. 3) and disrupt the coordinated movements of the number, NA not applicable, Leak subtracted peak current (I) measured at +20 mV, V 1/2 half-activation potential, k the slope factor, τ activation time constant measured at +20 mV. All values are calculated from leak subtracted current. V 1/2 and k are calculated from normalized conductance G/G max . Mean ± SEM (GFP + selected variant: *p < 0.05 for K v 7.2 WT vs. mutant; † p < 0.05 for K v 7.2 WT + PIP5K vs. mutants + PIP5K;^p < 0.05 for the difference between -PIP5K and +PIP5K within the same transfection. The source data for Table 2  VSD and the pore domain (Fig. 6), indicative of their decoupling. To the best of our knowledge, this is the first study to provide an atomic-level structural basis for the allosteric role of PIP 2 in voltage-dependent activation of neuronal K v 7.2 channels.
Our MD simulations have identified Site-O2 comprised of the K552-R553-K554 motif in Helix-B and with R560 in the BC linker as a PIP 2 binding site in the open state (Fig. 2). We have previously shown that current potentiation induced by increasing PIP 2 level is abolished by EE mutations K552T, R553L, and R560W 14,25 but enhanced by K554N 25 , highlighting the role of these basic residues in PIP 2 modulation of K v 7.2 channels 14,25 . The corresponding basic residues in Helix-B of K v 7.1 bind to PIP 2 in vitro 23 but those in K v 7.3 do not affect its sensitivity to PIP 2 depletion 27 , suggesting the subunit-specific difference of Helix-B in mediating PIP 2 binding.
Our simulations also demonstrate that PIP 2 interacts with Site-O1 formed by F346, Y347, and R353 at the beginning of the AB linker and with K162 and R165 in the S2-S3 linker (Fig. 2). Due to the presence of a lysine residue (K162) and two aromatic residues (F346 and Y347), this site fulfills the requirement for a canonical PIP 2 binding site 57 . Consistent with the conserved sequence of the S2-S3 linker among K v 7 subunits 3 , PIP 2 binding to this linker has also been reported in K v 7.1 3,21,22,58 and K v 7.3 27,52 , suggesting the role of this linker in PIP 2 modulation of K v 7 channels.
Although a "cationic cluster" (K452/R459/K461) in the distal region of the AB linker is implicated in PIP 2 modulation of K v 7.2 current 26 , the charge-neutralizing mutation of R353 in the proximal region of this linker reduces basal current expression of K v 7.2 channels and increases current inhibition upon PIP 2 depletion, suggestive of decreased PIP 2 affinity (Figs. 4-5). Combination of R353Q with R214Q and K219N in Site-O4 further reduces basal current and the PIP5K-induced current enhancement (Fig. 4), suggesting that R353 modulates K v 7.2 channels by coordinating PIP 2 binding with other residues. Indeed, R353 serves as an initial anchor point for PIP 2 binding to Site-O1 and Site-O4 ( Supplementary Movies 1-4). Furthermore, PIP 2 induces the conformational change of helices A and B from the largely solvent-exposed conformation to a bilayer-interacting conformation, whereas the R353Q mutation attenuates this effect Fig. 5 Charge-neutralizing mutations of potential PIP 2 binding residues alter K v 7.2 current response to Dr-VSP activation. a Representative current traces showing Dr-VSP-mediated K v 7.2 current decay in CHO hm1 cells co-expressing Dr-VSP and K v 7.2 WT or mutants from −20 mV to +100 mV. CHO cells were held at -70 mV, and a brief voltage step to -60 mV was applied to calculate the linear leak. 10 s step depolarizations were applied in 20-mV steps from -20 to +100 mV with 2 min inter-step intervals to allow PIP 2 regeneration. Red trace shows current decay curve when cells were held at +40 mV. b Ratio of current decay in K v 7.2 WT or mutant channels at each voltage step. c K v 7.2 current decay ratio at +40 mV, +60 mV, +80 mV and +100 mV. The number of cotransfected cells that were recorded with EGFP-tagged Dr-VSP: K v 7.2 WT (n = 27), R214Q (n = 14), K219N (n = 13), R325Q (n = 12), R353Q (n = 15), R214Q/ K219N (n = 13) or R214Q/K219N/R353Q (n = 11). The source data for Fig. 5b is available in Figshare 93 . Data represent the mean ± SEM. One-way ANOVA Fisher's test results are shown (*p < 0.05, **p < 0.01 and ***p < 0.005).
( Fig. 7), suggesting that PIP 2 binding to R353 contributes to this conformational change.
The conformational change in K v 7.2 is different from that observed in the PIP 2 -bound K v 7.1 22,59 which lacks the analogous arginine residue and shows low sequence homology to the AB linker of K v 7.2. K v 7.3, on the other hand, has the analogous arginine residue in the AB linker with fairly well-conserved sequence, suggesting that K v 7.3 may adopt a similar conformation as K v 7.2 in the presence of PIP 2 . We propose that PIP 2 binding to R353 affects the opening of K v 7.2 channels by inducing a bilayer-interacting conformation of the helical pair and facilitating the movement of PIP 2 along Helix-B and its binding to distal Helix-B (Site-O2). Subsequent interaction with other basic residues in Site-O1 and Site-O4 will induce the transition from the closed to the open state by coordinating allosteric movement of the VSD and the pore domain. R353 is also the target of three epilepsy mutations (ClinVar Database, NCBI) 60-62 , further supporting the role of R353 in PIP 2 modulation of K v 7.2 channels.
Neuronal K v 7 channels are highly enriched in the axonal surface where they regulate the AP firing threshold, frequency and shape 13,63 , whereas dendritic K v 7 currents increase the threshold of Ca 2+ spike initiation 64 . Heterozygous deletion of KCNQ2 gene in mice leads to hippocampal hyperexcitability and increased seizure propensity 65,66 . Consistent with their critical role in inhibiting neuronal excitability, > 400 BFNE and EE Fig. 6 Charge-neutralizing mutations of potential PIP 2 binding residues in Site-O4 disrupt the correlated motions of the VSD and the S6 gate of K v 7.2 channels. Network-based community analysis in WT (a) and mutant K v 7.2 channels containing R214Q (b), K219N (c), and R325Q (d) mutations in Site-O4. A community represents a set of residues that move in a correlated manner during the MD simulations. The thickness of the edges between the amino acids within each community corresponds to the strength of correlation between them. The network analysis was performed after combining all 3 independent simulation trajectories. a Binding of PIP 2 at Site-O4 in the WT channel results in correlated motions of the VSD (the S4 and the S4-S5 linker) and the gate S6, highlighted by a single community in green. b Introduction of the R214Q mutation leads to uncorrelated motions of the VSD and the gate, highlighted by 3 separate communities colored in pink, green, and gray. c Introduction of the K219N mutation leads to uncorrelated motions of the VSD and the gate, highlighted by the smaller community in green and the emergence of an additional community colored in pink. d K v 7.2-R325Q channels displayed weaker correlated motions of the S4 and the S4-S5 linker, and the decoupling of the S6 of the pore domain from the VSD, highlighted by the much smaller community in green and the emergence of 2 additional communities colored in pink and gray. In PIP 2 -free membranes, the helical pair of the WT channel adopts a largely solvent-exposed conformation. Addition of PIP 2 induces a drastic conformational change in these helices from a solvent-exposed conformation to a bilayer-interacting conformation. However, the R353Q mutation in the AB linker moves the helices closer to a solventexposed conformation even in the presence of PIP 2 . d Large-scale conformational change in the cytoplasmic helices provides a pathway for PIP 2 movement along Helix-B to its ultimate localization at Site-O4. All the phosphorus atoms of the bilayer are shown in vdW (van der Waals) representation. K v 7.2 protein is shown in ribbon representation with the S4 in red, helices A and B in brown, and the rest of the protein in ice blue. mutations are found in KCNQ2 17 . Computational algorithms have identified the S4, the pore loop, the S6, pre-Helix-A, Helix B, and the BC linker as hotspots for EE mutations 14,67 , whereas BFNE mutations are enriched in the S2-S3 linker 67 . The overlap between epilepsy mutation hotspots and the PIP 2 binding domains identified in our study underscores the critical role of PIP 2 in the pathophysiological mechanism underlying KCNQ2related epilepsy.
Current anti-epileptic drugs are ineffective in treating many epilepsy patients with K v 7.2 EE variants 18,19,68 . M-current inhibition upon PIP 2 depletion results in neuronal hyperexcitability 69 , and impaired PIP 2 sensitivity of K v 7.2 channels is associated with EE variants 14,25,28 . Retigabine (INN; USAN ezogabine) is a selective agonist of K v 7.2-K v 7.5 channels, but not K v 7.1 channel 70,71 . Retigabine suppresses seizures in animal models and humans 70,71 , however, it has been discontinued as an antiepileptic drug due to adverse side effects 72 . However, it may be effective in opening EE mutant channels with impaired PIP 2 binding because it stabilizes the open state of K V 7.2 and K V 7.3 channels by binding to a hydrophobic pocket near the gate 73 . Alternatively, strengthening PIP 2 -K v 7.2 interaction may increase K v 7 current. For example, zinc pyrithione can rescue M-current in hippocampal neurons following PIP 2 depletion by competing with PIP 2 for K v 7.2 activation 74 . Recently, a compound, CP1, is shown to substitute PIP 2 for the VSD-pore coupling in K v 7.1 channel, and to a less extent, K v 7.2 and K v 7.2/K v 7.3 channels 75 , suggesting its potential to inhibit neuronal hyperexcitability. Our in-depth investigation of PIP 2 -K v 7.2 interaction may provide the foundation to explore a new class of therapeutics for epilepsy that can control PIP 2 modulation of neuronal K v 7 channels.

Methods
Structural models of open and closed states of Kv7.2. The open and closed conformations of K v 7.2 channel used in the molecular dynamics (MD) simulations were modeled following the procedure described in a recent study 14 . Briefly, the closed conformation of the channel was modeled based on the cryo-EM structure of K v 7.1 (PDB ID: 5VMS) 10 . Multiple sequence alignment of the target template and K v 7.2 was performed using TCoffee web server (https://www.ebi.ac.uk/Tools/ msa/tcoffee/). After the alignment, the homology model of the closed state was built with MODELLER 76 . Our final model contains residues from 74-363 and 537-594. The open conformation of K v 7.2 was then constructed by performing non-equilibrium, driven MD simulations. We performed a 20-ns targeted MD (TMD) 77 simulation during which the closed structure was driven towards an open form, while embedded in a lipid bilayer containing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine (POPC) and 2.2% 1-palmitoyl-2-oleoyl-sn-glycero-3phosphatidylinositol 4,5-bisphosphate (PIP 2 ) lipids. The target of the TMD simulation was selected to be the highly homologous K v 1.2/K v 2.1 channel in an open-state conformation (PDB ID: 2R9R) 78 . As major structural changes occur in the pore region of the channel, we applied a restraint (k = 250 kcal/mol/Å 2 ; only Cα atoms were driven) on the S4-S5 and S6 helices of each monomer to drive it towards the target open state. The structural stability of the open and closed conformations of K v 7.2 was evaluated by monitoring the degree of opening of the states, calculated by the number of water molecules in the pore helix and the selectivity filter during 400-ns MD runs in explicit lipid bilayers 14 . During these simulations we consistently observed stable and a higher number of water molecules in the open state of the channel, as compared to the closed state 14 .
PIP2-binding simulations. The open and closed conformations of K v 7.2 were embedded in multiple independently generated POPC membranes with or without 2.2% PIP 2 lipids (Table 1). All the membranes were constructed using the Membrane Builder module in CHARMM-GUI 79 , and the initial placement of PIP 2 lipids was intentionally varied in each membrane. All PIP 2 lipid molecules were initially at least 15 Å away from the protein. The systems were solvated using TIP3P water and buffered at 150 mM KCl to neutralize. The final simulation systems consisted of~300,000 atoms. All the simulations were performed in the absence of calmodulin. To further examine the specific lipid-protein interactions captured in the modeled closed K v 7.2, we also performed additional lipid-binding simulations of a recent cryo-EM structure of K v 7.2 (PDB ID: 7CR3) after removing calmodulin 44 . All the missing loops were constructed using MODELLER 76 and the entire tetrameric structure was embedded in an explicit lipid bilayer containing POPC and 2.2% of PIP 2 lipids. Three independent 500-ns simulations were performed on this system. Molecular dynamics simulation protocols. All the simulations were performed under periodic boundary conditions using NAMD2 80,81 and CHARMM36m force field parameters 82,83 for protein and lipid. During the initial equilibration, the protein's backbone atoms were harmonically restrained to their initial positions with a force constant of k = 1 kcal/mol/Å 2 . The restraints were released at the start of the production run. All the non-bonded forces were calculated with a cutoff of 12 Å and a switching distance of 10 Å. Long-range electrostatic forces were calculated using the particle mesh Ewald (PME) method 84 . A Langevin thermostat using γ = 1 ps −1 was used to maintain the system temperature at 310 K. The pressure was maintained at 1 bar using a Nosé Hoover Langevin piston method 85 . During pressure control, the simulation box was allowed to fluctuate in all the dimensions with constant ratio in the xy (lipid bilayer) plane. An integration time step of 2 fs was used in all the simulations.
Simulation analysis. To characterize lipid-protein interactions and potential lipid binding sites, occupancy maps of the PIP 2 headgroup were calculated using the VOLMAP plugin in VMD 86 . Based on our previous experience 35 , a 4-Å heavy-atom distance cutoff was chosen to define contacts between the phosphate groups of PIP 2 lipids and protein residues, while a 3.5-Å cutoff was used to define contacts with the hydroxyl groups of PIP 2 . The role of PIP 2 lipids in stabilizing the open conformation of K v 7.2 was determined by performing dynamical network analysis using the NETWORK-VIEW plugin 87 in VMD. In a network, all Cα atoms are defined as nodes connected by edges if they are within 4.5 Å of each other for at least 75% of the MD trajectory. Pearson correlation was used to define the communities (the set of residues that move in concert) in the network.
DNA construct and mutagenesis. Plasmid pcDNA3 carrying KCNQ2 cDNA (GenBank: Y15065.1) encoding K v 7.2 (GenBank: CAA 75348.1) was previously described 14,25,88,89 . This short isoform of K v 7.2 lacks 2 exons compared to the reference K v 7.2 sequence (GenBank: NP_742105.1). Our lab has previously shown that currents through this K v 7.2 variant can be potentiated by PIP 2 increase and are sensitive to Dr-VSP activation 14,25 . Plasmid pIRES-dsRed-PIPKIγ90 90 was a kind gift from Dr. Anastasios Tzingounis (University of Connecticut). Electrophysiology. Whole cell patch clamp recording was performed in Chinese ovary cells (CHO hm1) at room temperature (22-24°C) as described 14,25 . Cells were plated on 12 mm glass coverslips (Warner Instrument, 1 ×10 5 cells per coverslip) treated with poly D-lysine (0.1 mg/mL) (Sigma-Aldrich). To express K v 7.2 channels and PIP5K, cells were transfected with pEGFPN1 (0.2 μg), pIRES-dsRed-PIPKIγ90 (0.45 μg) and pcDNA3-K v 7. To record K + currents, cells were held at −80 mV. Currents were evoked by depolarization for 1.5 s from −100 mV to +20 mV in 10-mV increments, followed by a step to 0 mV for 300 ms. Leak-subtracted current densities (pA/pF), normalized conductance and channel biophysical properties were calculated as described 25 . Briefly, leak current was defined as non-voltage-dependent current through GFP-transfected CHO hm1 cells. Current density was calculated by dividing leak-subtracted current (pA) by capacitance (pF). V 1/2 and the slope factor (k) were calculated by fitting the points of G/G max to a Boltzmann equation in the following form: To examine the decline of K v 7.2 current upon activation of Dr-VSP, CHO hm1 cells were transfected with pDrVSP-IRES2-EGFP (0.5 μg) and pcDNA3-K v 7.2 WT or mutant (0.5 μg). The pDrVSP-IRES2-EGFP plasmid was a gift from Yasushi Okamura (Addgene plasmid # 80333). Voltage-clamp recording of K v 7.2 current upon depolarization-induced Dr-VSP activation was performed as described 91 with an external solution containing 144 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 0.5 mM MgCl 2 , 10 mM glucose and 10 mM HEPES (pH 7.4). Patch pipettes (3-4 MΩ) were filled with intracellular solution containing 135 mM potassium aspartate, 2 mM MgCl 2 , 1 mM EGTA, 0.1 mM CaCl 2 , 4 mM ATP, 0.1 mM GTP and 10 mM HEPES (pH 7.2). Cells were held at −70 mV and 10 s step depolarizations were applied in 20-mV steps from −20 to +100 mV with 2 min inter-step intervals to allow PIP 2 regeneration. The extent of K v 7.2 current decay upon Dr-VSP activation during 10 s depolarization was measured as the ratio of current at 10 s over peak current at each voltage step.
Surface biotinylation. CHO hm1 cells were plated on 60 mm culture dishes (Corning, 8 × 10 5 cells per well). Next day, cells were transfected with pcDNA3-K v 7.2 WT or mutant (0.8 μg) using FuGENE6 transfection reagent (Promega). At 24 h posttransfection, the cells were subjected to surface biotinylation as previously described 92 . The culture dishes containing transfected cells were placed on ice and washed with 1X PBS twice. To biotinylate surface proteins, the cells were then incubated with Sulfo-NHS-SS-Biotin (1 mg/mL, Pierce) in ice-cold PBS (3 mL) for 20 min. The cells were then washed with 1X PBS twice and 1X TBS once. Cells were collected using cell scraper in 400 μL ice-cold lysis buffer containing: (mM): 150 NaCl, 50 Tris, 2 EGTA, 1 EDTA, 1% Triton-X, 0.5% deoxycholic acid, supplemented with Halt protease inhibitor cocktail (Thermo Fisher Scientific). Lysates were harvested by 15 min incubation on ice, followed by 15 min centrifugation at 14,000 x g at 4°C. 40 μL lysates were saved for western blotting. The remaining 360 μL of lysates were incubated with 50% NeutraAvidin agarose beads (Pierce, 100 μL of 1:1 slurry) for overnight at 4°C to isolate biotinylated surface proteins from the lysate. After washing with the lysis buffer, biotinylated surface proteins were eluted by heating in 1x SDS sample buffer containing 50 mM TCEP at 75°C for 30 min. Eluted biotinylated surface proteins and lysates were examined by immunoblotting for K v 7.2 and αtubulin or β-tubulin as described in the previous section. ImageJ software (NIH) was used to measure background-subtracted intensity of each immunoblot band, and surface or total K v 7.2 intensity was normalized to total α-tubulin or β-tubulin band intensity. To confirm that cell membranes were intact and intracellular proteins were not biotinylated during surface biotinylation, immunoblot of biotinylated surface proteins was also performed with anti-α-tubulin or anti-β-tubulin.
Statistics and reproducibility. All the measurements were taken from distinct samples. All electrophysiology and immunoblotting data are reported as mean ± SEM. Origin 9.1 (OriginLab, Inc) was used for Student's t-test and one-way ANOVA with post-hoc Fisher's multiple comparison tests. Specifically, one-way ANOVA with post-hoc Fisher's test was used for surface biotinylation and electrophysiology figures with the exception that Student's unpaired t-test was used when comparing the results before and after PIP5K cotransfection in Supplementary Fig. 4a-b. Listed sample sizes of electrophysiology indicate number of cells successfully recorded, while the sample sizes of surface biotinylation studies represent the numbers of experimental replica. Statistical significance was assessed at a priori value (p) < 0.05.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The datasets generated and analyzed in the current study and presented as the main and supplementary figures are available in the Figshare repository with the identifier https://doi.org/10.6084/m9.figshare.15181038 93 . All other data that support the findings of this study will be available from the corresponding authors upon reasonable request.

Code availability
Simulation trajectories were collected using the simulation program NAMD. Visualization and analysis were performed using VMD and Python. All of these software packages are publicly available.