Dynamic role of the tether helix in PIP₂-dependent gating of a G protein–gated potassium channel

G protein–gated inwardly rectifying potassium (GIRK or Kir3) channels are expressed in various regions of the brain, where they control the resting membrane potential and excitability of neurons (Lüscher and Slesinger, 2010). Mouse and human molecular genetic studies have indicated a role for GIRK channels in a variety of human disorders, including addiction, alcoholism, Down’s syndrome, and depression (for a review, see Mayfield et al., 2015). Similar to other inwardly rectifying potassium channels, activation of GIRK channels hyperpolarizes the membrane potential, reducing neuronal excitability. A large number of neurotransmitters (e.g., GABA, dopamine, glutamate, serotonin, acetylcholine, and opioid peptides) stimulate G protein–coupled receptors that directly activate GIRK channels (Ehrengruber et al., 1997; Lüscher et al., 1997; Scanziani, 2000; Wiser et al., 2006; Luján et al., 2009) via G protein Gβγ subunits (Logothetis et al., 1987; Reuveny et al., 1994; Wickman et al., 1994). In addition to G proteins, alcohol has been shown to directly activate GIRK channels (Aryal et al., 2009). However, GIRK channels remain mostly closed in the absence of activators. The structural mechanism underlying their low probability of opening in the absence of such activators remains poorly understood.

Four different GIRK subunits (GIRK1, GIRK2, GIRK3, and GIRK4) have been identified in mammals (Lesage et al., 1995; Inanobe et al., 1999; Wickman et al., 2000; Lüscher and Slesinger, 2010), and each is regulated by the membrane phospholipid phosphatidylinositol 4,5 bisphosphate (PIP₂). Huang et al. (1998) showed that depletion of PIP₂ prevents GIRK activation by Gβγ subunits, indicating that PIP₂ is a cofactor for G protein regulation of GIRK channels. In fact, PIP₂ is an essential cofactor for many different types of ion channels (Hille et al., 2015). Interestingly, the strength of the PIP₂ interaction with the channel can determine the level of basal channel activity, which varies considerably among different inward rectifiers. For example, constitutively open inward rectifiers, such as Kir2 and Kir4 channels, bind PIP₂ with relatively high affinity and have an open probability of close to 1 (Hibino et al., 2010). In contrast, Kir channels that are gated by G proteins, ATP, or ethanol (e.g., Kir3/GIRK, Kir1, and Kir6) exhibit lower relative affinities for PIP₂ and a corresponding lower open probability, suggesting a correlation between the strength of the PIP₂ interaction and open channel probability (Hibino et al., 2010). GIRK channel activators, such as ethanol and Gβγ subunits, appear to increase the relative affinity for PIP₂, leading to an increase in the frequency and/or channel open time (Huang et al., 1998; Petit-Jacques et al., 1999; Bodhinathan and Slesinger, 2013). Conversely, ATP binding to Kir6/SUR channels leads to a reduction in PIP₂ relative affinity (Baukrowitz et al., 1998).

Several laboratories have investigated what determines the relative strength of PIP₂ association with Kir channels. Mutagenesis studies originally identified several conserved basic amino acids as well as some hydrophobic residues that influence the strength of PIP₂ interaction with the channel (Zhang et al., 1999; Soom et al., 2001; Lopes et al., 2002). Accordingly, the interaction of PIP₂ with GIRK channels could be converted from a weak to a strong interaction by introducing point mutations in the putative PIP₂-binding site, resulting in large agonist-independent currents similar to Kir2 channels (Zhang et al., 1999; Zhou et al., 2001). These studies led to the proposal that positively charged basic amino acids form an essential part of the PIP₂-binding site and that changes in the association of PIP₂ with this region of the channel underlie gating of different inward rectifiers. However, how the nature of the interactions with these positively charged amino acids changes during gating is unknown.

Atomic resolution structures of GIRK2 and Kir2.2 channels in the presence or absence of PIP₂ have provided structural details on the molecular interactions underlying PIP₂-mediated gating of Kir channels (Whorton and MacKinnon, 2011, 2013). Inward rectifiers possess two gating structures, a G loop gate located at the apex of the cytoplasmic domain and a hydrophobic gate formed by amino acids in the pore-facing M2 transmembrane helices (i.e., helix bundle-crossing [HBC] gate; Lüscher and Slesinger, 2010). Both gates must move sufficiently to support K⁺ conduction through the channel pore. A cluster of four basic amino acids in a helical structure, referred to as the “tether helix” (also called the C-linker), appears to coordinate PIP₂ (Whorton and MacKinnon, 2011, 2013). Curiously, these structural studies reveal only subtle differences in how PIP₂ interacts with Kir2.2 and GIRK2 channels (Hansen et al., 2011; Whorton and MacKinnon, 2011, 2013), with the positive charge appearing to have a dominant role in PIP₂ binding for both channels. Furthermore, the GIRK2 structure solved in the presence of PIP₂ and Gβγ subunits is partially closed, as is the Kir2.2/PIP₂ structure (Hansen et al., 2011; Whorton and MacKinnon, 2011, 2013), suggesting that the binding of PIP₂ to an open channel may look entirely different from the current set of atomic resolution structures. Another limitation with ion channel structures is they represent a static channel and lack the dynamic interactions that occur during gating. Thus, it remains to be determined how the molecular interactions of PIP₂ with the channel protein explain the large differences in resting basal channel activity and subsequent activation among different Kir channels. Here, using a combination of functional electrophysiological studies and molecular dynamics (MD) simulations, we sought to investigate how the highly conserved basic amino acids in the tether helix contribute to channel activation.

The following cDNAs, mouse GIRK2c (Kir3.2c), mouse Kir2.1, human m2R, enhanced yellow fluorescent protein (eYFP), and Danio rerio voltage-sensitive phosphatase (Dr-Vsp; gift from Y. Okamura, Osaka University, Osaka, Japan) were cloned in the mammalian expression vector pcDNA3.1 (ThermoFisher). Point mutations were introduced by site-directed mutagenesis (QuikChange II XL, Agilent Technology) and confirmed by automated DNA sequencing. GIRK2* was generated by replacing four native cysteines in GIRK2 (or Kir3.2) channels with thiol-unreactive amino acids (C65V, C190T, C221S, and C321V).

Whole-cell patch-clamp recordings were performed 24 to 48 h after transfection (Bodhinathan and Slesinger, 2013). Borosilicate glass electrodes (Warner Instruments) with access resistances ranging from 3 to 6 MΩ were filled with an intracellular solution containing 130 mM KCl, 20 mM NaCl, 5 mM EGTA, 5.46 mM MgCl₂ (1.5 mM free Mg²⁺), 2.56 mM K₂ATP, 0.3 mM Li₂GTP, and 10 mM HEPES (pH 7.4, ∼313 mOsm). The extracellular “20K” solution contained 20 mM KCl with 140 mM NaCl, 0.5 mM CaCl₂, 2 mM MgCl₂, and 10 mM HEPES (pH 7.4, ∼318 mOsm). 100 mM ethanol, 1-propanol, or 2-methyl-2,4-pentanediol (MPD) were added directly to the 20K solution. MTS-HE, MTS-Y, and MTS-F (Tocris) were dissolved in DMSO and diluted in 20K solution directly before application to cells. Whole-cell patch-clamp recordings were made as described previously (Bodhinathan and Slesinger, 2013).

The crystal structure of the GIRK2 channel bound to a PIP₂ head group was used for MD simulations (see supplemental Materials and methods). We used the Protein Data Bank (PDB): 4KFM structure for simulations, because this channel is in a preopen conformation (Whorton and MacKinnon, 2013). We compared the 3SYA and 4KFM structures and determined there is a small RMSD (Whorton and MacKinnon, 2011, 2013), with the largest movements in the LM loop and a 0.41 Å RMSD for the c-α in the PIP₂ pocket, formed by residues 88–92 (from the slide helix), 192–203 (HBC gate and tether helix), and 62–64 (N-terminal b-loop). Four 1-stearoyl-2-arachidonoyl-sn-glycero-3-phosphoinositol 4,5 bisphosphate (18:0, 22:4 PIP₂) molecules were aligned to the crystal structure of PIP₂, one at each binding site. Gβγ subunits were not included in the calculations. The GIRK2-PIP₂ system was embedded in a bilayer consisting of 202 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC; 16:0, 18:1 PC) lipids using the CHARMM-GUI membrane builder (Brooks et al., 2009; Wu et al., 2014). The protein–lipid system was then solvated with TIP3P water molecules and 150 mM KCl. The CHARMM36 force field was employed (Best et al., 2012). Additionally, five potassium ions were placed in the ion conduction pathway, and four sodium ions were placed in the sodium-binding site as seen in the crystal structures. A second system was set up with the same conditions but with a 6′Y mutation generated by in silico mutagenesis using CHARMM-GUI. The two systems (WT and 6′Y) were subjected to stepwise decreased restraint equilibration used in CHARMM-GUI membrane builder (Jo et al., 2007). The two systems (WT and 6′Y) were then equilibrated for 20 ns using positional restrain of 1,000 KJ/mol/nm² on all Cα atoms of the protein before the 400-ns unrestrained production simulations. We then repeated the 400-ns production simulations using different initial velocities for both systems. Simulations were conducted using GROMACS 4.6 (Hess et al., 2008) with a 2 fs integration time step and visualized using VMD, Pymol (Humphrey et al., 1996), or Discovery Visualizer 4.0 (Accelrys).

Mouse GIRK2 (containing amino acids 52–380) was expressed and purified in Pichia pastoris (a gift from R. Mackinnon, The Rockefeller University, New York, NY) as described previously (Whorton and MacKinnon, 2011; Glaaser and Slesinger, 2017). Purified GIRK2 channels were reconstituted into a lipid vesicle mixture containing POPE/POPG as described previously (Whorton and MacKinnon, 2011). For the K⁺ flux assay, vesicles were diluted 1:200 into flux buffer containing the H⁺-sensitive dye 9-amino-6-chloro-2-methoxyacridine (ACMA), as described previously (Whorton and MacKinnon, 2011; Glaaser and Slesinger, 2017). Fluorescence quenching was measured using a Flexstation 3 microplate reader (Molecular Devices). Changes in fluorescence were monitored after the addition of the H⁺ ionophore m-chlorophenyl hydrazone (CCCP), followed by the addition of 30 µM diC₈-lipids (PI, PI(4)P, PI(5)P, PI(3,4)P₂, PI(4,5)P₂, and PI(3,4,5)P₃; Echelon). The K⁺ gradient was collapsed with 50 nM of the K⁺ ionophore valinomycin, allowing for determination of vesicle capacity at the end of the experiment. The “relative K⁺ flux” was calculated by measuring the decrease in fluorescence (i.e., quenching) 900 s after the addition of a vehicle or phospholipid. This decrease in fluorescence was normalized to the basal fluorescence before adding CCCP and to the maximal decrease in fluorescence following the addition of valinomycin (F_v). The leakage flux for the proteoliposomes (no diC₈) was subtracted to give the “fractional activation.”

Pooled data are presented as mean ± SEM and evaluated for statistical significance (defined as P < 0.05 for all tests). An unpaired Student’s t test was used to compare data from two separate groups. One-way ANOVA was used for comparing three or more groups, followed by Bonferroni post hoc test for comparison between groups when applicable. Where indicated, n represents the number of cells from which the currents were recorded under each condition. All data were analyzed using Excel (Microsoft) and Prism (GraphPad).

Additional details regarding the methods, along with supplemental data and videos, are provided online. Figs. S1–S6 contain additional analyses of the MD simulations, and Videos 1–4 are 400-ns videos of the WT and KY MD simulations, focusing on the PIP₂ interaction in the PIP₂ pocket and movement of the -2′F (F192) in the helix bundle crossing gate.

To better understand the structural mechanism underlying PIP₂-dependent gating of GIRK channels, we evaluated the role of the positively charged, basic amino acids in the tether helix (Whorton and MacKinnon, 2011, 2013). To facilitate a comparison with the previous studies that have examined these highly conserved charged residues, we propose a numbering system whereby the first of these highly conserved residues (K194 in GIRK2) is defined as the 0′ position (0′K), K199 as 5′K, and K200 as 6′K (Figs. 1 a and S1 a).

In the GIRK2/PIP₂/Gβγ atomic structure (PDB: 4KFM), which is considered to be in a preopen conformation (Whorton and MacKinnon, 2013), 0′K, 5′K, and 6′K in the tether helix form hydrogen bonds with the 5′ phosphate (5′-PO₄) of PIP₂ (Fig. 1 a). In the Kir2.2/PIP₂ structure (PDB: 3SPI), the 0′K and 5′K form hydrogen bonds with the 5′-PO₄, but the 6′K forms a hydrogen bond with the 4′-PO₄ (Fig. S1 b; Hansen et al., 2011). Thus, the 6′K in the tether helix of GIRK2 and Kir2.2 interact with different phosphates on the inositol ring of PIP₂. Given these variations in how PIP₂ is coordinated in the GIRK2 and Kir2.2 structures, we hypothesized that this region of the tether helix may be important for regulating the PIP₂-dependent gating of GIRK2 channels.

To test this idea, we used a strategy of chemical modification in which a single cysteine is introduced at 5′K or 6′K and is then probed with a membrane permeant MTS compound. This technique has the advantage of enabling the study of a channel before and after modification and has been used previously to probe gating structures of inwardly rectifying potassium channels (Guo et al., 2002; Lopes et al., 2002; Xiao et al., 2003; Bodhinathan and Slesinger, 2013). Here, we focused on the effect of MTS hydroxyethyl (HE), which would covalently attach a CH₂-CH₂-OH side chain to the sulfhydryl of the engineered cysteine, mimicking a serine/threonine type of amino acid substitution, i.e., a polar, uncharged residue. We introduced the cysteine into a GIRK2 channel that was modified to contain no other internal cysteines (referred to as GIRK2*). GIRK2* was shown previously to retain both Gβγ- and alcohol-dependent activation (Bodhinathan and Slesinger, 2013). Using whole-cell patch-clamp recordings from HEK293 cells transiently transfected with GIRK2*-5′C or GIRK2*-6′C cDNA, along with the m2 muscarinic receptor cDNA, we measured the basal, m2 muscarinic receptor–activated and alcohol-activated GIRK2 currents before and after MTS modification (Bodhinathan and Slesinger, 2013). In HEK293 cells expressing GIRK2*-5′C, there was little activation by carbachol or alcohol, and transfected cells did not show any appreciable Ba²⁺-sensitive basal currents. Bath application of MTS-HE (0.1–1 mM) for ∼5 min did not appear to alter the basal currents (Fig. 1 b; n = 2). Expression of GIRK2*-6′C in HEK293 cells also produced currents that were small and relatively insensitive to 5 µM carbachol or 100 mM alcohols (Fig. 1 c). In contrast to 5′C, however, bath application of 1 mM MTS-HE dramatically increased the amplitude of the Ba²⁺-sensitive basal current for GIRK2*-6′C channels (Fig. 1, c and d). The neighboring 3′C in the tether helix also showed no increase in basal current with MTS-HE (Fig. 1 d). For 6′C, the basal current density increased from −3.4 ± 0.8 pA/pF to −232.2 ± 38.0 pA/pF (n = 7) and was now inhibited by ∼90% with MPD (−89.5% ± 6.9%, n = 7) after modification with MTS-HE.

We next examined the effect of two other membrane-permeant MTS reagents, 0.01 mM MTS-phenylalanine (F) and 0.1 mM MTS-tyrosine (Y). Like MTS-HE, MTS-Y and MTS-F dramatically increased the Ba²⁺-sensitive basal currents by ∼20-fold (Fig. 1, e and f). These results demonstrate that the positive charge of 6′K is not essential for MTS activation of GIRK2*-6′C channels. In addition, MPD inhibited the MTS-activated basal currents from 60% to 100% after modification with MTS-F and MTS-Y (Figs. 1 e and S2). A large agonist-independent basal current and inhibition with MPD are features of a constitutively active GIRK2 and Kir2.1 channels (Aryal et al., 2009). These results demonstrate the Cys at the 6′ position is unable to support channel activation (i.e., little Ba²⁺-sensitive current), but modification with a larger uncharged side chain (e.g., Y or F) greatly activates the channel.

With the MTS modification experiments, there are four cysteines per tetrameric channel that can be modified but the exact stoichiometry of modification is unknown. To study the effect of four substitutions per tetramer, we engineered a single point mutation at 6′K in GIRK2*. We predicted that a conserved charged substitution at this position (i.e., K200R [6′R]) would produce a channel with a small basal current and normal alcohol modulation, whereas a phenylalanine or tyrosine substitution would most closely mimic modification with MTS-F or MTS-Y (i.e., increase the basal current and change alcohol activation). Expression of GIRK2*-6′R produced channels with a small Ba²⁺-sensitive basal current, which was activated by ethanol, propanol, and MPD, similar to GIRK2* channels (Fig. 2, a–d). In contrast, mutation of this positively charged residue to a tyrosine (GIRK2*-6′Y) produced large Ba²⁺-sensitive basal currents that were relatively insensitive to alcohols, similar to MTS-Y modification of 6′C. Expression of GIRK2*-6′F produced Ba²⁺-sensitive basal currents that were inhibited by MPD, similar to MTS-F modification of 6′C, but the basal current was not significantly larger (Fig. 2, c and d). Collectively, these results suggest that a Tyr side chain at the 6′ position of the tether helix may strengthen the interaction with PIP₂, leading to a large basal current and small alcohol response (Aryal et al., 2009).

To probe for possible changes in the relative association of PIP₂ with the channel, we studied the effect of reversibly depleting plasma membrane PIP₂ on whole-cell Kir currents using the voltage-activated phosphatase Dr-VSP (Bodhinathan and Slesinger, 2013; Adney et al., 2015; Rjasanow et al., 2015). Previous studies showed that activation of Dr-Vsp by depolarizing the membrane to +100 mV for varying lengths of time produces a time-dependent, graded depletion of PIP₂ and a corresponding reduction in Kir current (Bodhinathan and Slesinger, 2013). To examine the relative association of PIP₂ with GIRK2*-6′Y channels, we coexpressed Dr-Vsp with 6′Y channels and studied the effect of varying length prepulses to +100 mV on the GIRK current measured at −120 mV (Fig. S3). Longer depolarization times led to more rapid and complete depletion of membrane-bound PIP₂. Both 300 and 500 ms prepulses reduced the steady-state basal current, reaching a maximal inhibition of ∼70% of the Ba²⁺-sensitive basal current (0.7 fractional inhibition; Fig. 3, a and d). The 30% residual current could reflect tightly bound PIP₂ or a change in PIP selectivity and was not examined further. In contrast, activation of Dr-Vsp with shorter prepulses (e.g., 50 or 100 ms) reduced ∼100% of the basal current for WT GIRK2*-6’K channels (Fig. 3, b and d). For comparison, we also measured the Dr-Vsp sensitivity of Kir2.1 channels, which are known to bind PIP₂ with higher relative affinity (Huang et al., 1998; Sui et al., 1998). Expression of Kir2.1 produced large Ba²⁺-sensitive basal currents that were partially inhibited by 100 mM propanol (Fig. 3, c and d), similar to previous results (Kobayashi et al., 1999; Lewohl et al., 1999). Activation of Dr-Vsp for 50 or 100 ms produced little change in the basal current, whereas a 500 ms prepulse significantly reduced the Ba²⁺-sensitive basal currents by ∼90% (Fig. 3, c and d).

We next analyzed the rates of current depletion and recovery after Dr-Vsp activation by fitting the decay of current with a single exponential and determining the time constant (tau). These rates correlate with the differences in PIP₂ affinity, where higher-affinity channels typically have a slow depletion and fast recovery (Bodhinathan and Slesinger, 2013; Adney et al., 2015; Rjasanow et al., 2015). The rate of recovery depends on both the relative affinity of the channel for PIP₂ and the rate of resynthesizing membrane-bound PIP₂. We assume the latter is similar when comparing different channels. We found that the tau for depletion was not significantly different for 6′K and 6′Y channels, but the tau for recovery was significantly faster for GIRK2*-6′Y (Fig. 3, e and f), 24.9 ± 3.2 s (n = 10) for 6′Y, and 35.8 ± 3.4 s (n = 9) for 6′K. This faster recovery for 6′Y supports the interpretation of a stronger PIP₂ interaction (Bodhinathan and Slesinger, 2013; Adney et al., 2015; Rjasanow et al., 2015; Dai et al., 2016).

To gain structural insights into how 6′Y might play a critical role in determining PIP₂ interactions, we conducted atomistic MD simulations to investigate possible differences in the PIP₂ pocket. GIRK2 (6′K; WT) or GIRK2 6′Y channels were embedded in a POPC bilayer containing four PIP₂ lipids with their head groups aligned onto their crystallographic positions. Using a starting condition defined as a preopen conformation, as depicted in the 4KFM structure (Whorton and MacKinnon, 2013), we examined the dynamics of PIP₂ interaction with the WT (6′K) or 6′Y channel in the absence of Gβγ subunits during 400-ns simulations (Fig. 4; see Materials and methods for details). During simulation of the WT channel, the side chain of 6′K appears flexible and adopts a downward orientation, away from the transmembrane domains (TMDs). Furthermore, the 5′ phosphate group of PIP₂ also moves downward away from 0’K in the pocket, forming a new set of interactions (Fig. 4 a, top; and Videos 1 and 2). In the 6′Y channel simulations, on the other hand, the 6′Y side chain maintains an upward orientation (i.e., points toward the lipid bilayer), and the 5′-PO₄ of PIP₂ appears to be more stably bound in its canonical site (Fig. 4 a, bottom; and Videos 3 and 4). To characterize the dynamics of these interactions, we examined the frequency of hydrogen bond formation with the 5′-PO₄ of PIP₂ of one chain in the 6′K or 6′Y channel over the 400-ns simulation (Fig. 4 b). The time-dependent plots show that the 0′K and 5′K form hydrogen bonds with the 5′-PO₄, and at ∼100 ns, the hydrogen bonds break as the PIP₂ molecule appears to move away from its canonical binding site (Fig. 4 b, blue trace). Coincident with this change, the negatively charged glutamate at 203 (E203; 9′E) forms a hydrogen bond with 6′K, whereas K64 forms a hydrogen bond with 5′-PO₄. In the 6′Y channel simulations, however, both 0′K and 5′K maintain stable hydrogen bonds with the 5′-PO₄ throughout the simulation, in parallel with the 6′Y forming a stable hydrogen bond to the same phosphate (Fig. 4 b, green trace). Furthermore, the 9′E does not form any hydrogen bonds with 6′Y.

We repeated the 400-ns simulations (simulation 2) to determine the reproducibility, and similar results were observed in the second simulation (Fig. S5, simulation 2). To quantify these observations, we calculated the probability of hydrogen bond formation between the 1′, 4′, or 5′ phosphate of PIP₂ and GIRK2 for all four pockets in both 400-ns simulations (Fig. 4, c and d; and Figs. S4 and S5). Interestingly, we found that the highest probability of hydrogen bonds with PIP₂ occurs between the side chain and backbone amide of R92 with the 1′-PO₄ in both WT 6′K and 6′Y simulations (Figs. S4 a and S5). This finding raises the possibility that R92, which is at the bottom of the M1 helix, serves as an anchor point for interactions of 1′-PO₄ of PIP₂ with the channel, whereas the 4′-PO₄ and 5′-PO₄ interact more dynamically with the tether helix and N-terminal hairpin.

The probability of hydrogen bonds with the 5′-PO₄ of PIP₂ significantly increased for the 0′K and 5′K in the 6′Y channel simulation (Fig. 4 d). Concurrent with these changes, the probability of hydrogen bonds between K64 and 5′-PO₄ decreased, whereas K64 hydrogen bonds with the 5′K increased in the 6′Y channel simulation (Figs. 4 d and S4 b). The probability of hydrogen bonds between the 4′-PO₄ of PIP₂ and K64 was unchanged but increased with the 5′K in the 6′Y channel simulation (Fig. S4 b). Furthermore, the 6′Y does not hydrogen bond with 9′E in the 6′Y simulation, allowing only hydrogen bond interactions with 5′-PO₄ (Fig. 4, c and d). Collectively, the 6′Y simulations appear to show a strengthening of the interaction of the 5′-PO₄ of PIP₂ with the 0′K and 5′K, limiting the downward movement of PIP₂ away from the channel’s PIP₂-binding pocket (Fig. 4, a–d).

The MD simulations also revealed an important interaction of the 6′K with the 9′E, as the 6′K adopts a downward orientation (Fig. 4 a). The positively charged 6’K forms a significant number of hydrogen bonds with 9’E that are absent in the 6’Y channel (Fig. 4 c). The 9′E also forms hydrogen bonds with K64 in the MD simulations (Fig. S4 c). In the 6′Y mutant channels, however, the number of hydrogen bonds significantly decreases between K64 and the 5′-PO₄ of PIP₂ (Fig. 4 d). Thus, a trifecta of hydrogen bond interactions among 6′K, 9′E, and K64 may determine the extent of hydrogen bonding between PIP₂ and the 0′K and 5′K sites in the channel. Collectively, the MD simulations provide a tenable model for how 6′Y might promote the channel to interact more strongly with PIP₂; anchored by the 1′-PO₄ at R92, the tyrosine substitution (6′Y) supports movement of PIP₂ deeper into the pocket toward the 0′K and 5′K.

The MD simulations indicate a prominent role for 5′-PO₄ of PIP₂ in GIRK2 channel gating. To test the role of the 5′-PO₄ in channel activation, we examined effect of different soluble forms of phosphoinositides (e.g., diC₈-PI(4,5)P₂) on GIRK2 activation using purified GIRK2 channels reconstituted into proteoliposomes (Whorton and MacKinnon, 2011; Glaaser and Slesinger, 2017). Potassium flux through GIRK channels was monitored via a fluorescence-based K⁺ flux assay (Fig. 5 a; Whorton and MacKinnon, 2011). With GIRK2-containing liposomes loaded with the pH-sensitive ACMA dye, the addition of the proton ionophore CCCP results in quenching of the ACMA dye if GIRK2 channels are open (i.e., H⁺ enters via CCCP if K⁺ can exit the proteoliposome; Fig. 5 a; Whorton and MacKinnon, 2013). As expected, diC₈-PI(4,5)P₂, the soluble form of the predominant phosphoinositide found in the plasma membrane (Hille et al., 2015), produced maximal activation of purified GIRK2 channels (Fig. 5, b and c). PI(3,4,5)P₃ elicited a K⁺ flux comparable to that of PI(4,5)P₂, whereas PI(3,5)P₂ and PI(3,4)P₂ produced an intermediate K⁺ flux (Fig. 5 b). As shown previously (Ho and Murrell-Lagnado, 1999; Rohács et al., 1999), PI(4)P, and PI did not increase GIRK2 activity (Fig. 5, b and c). However, we found that application of PI(5)P alone activates GIRK2 channels (Fig. 5 c). To quantify these changes, we measured the steady-state flux and normalized to the maximal response with PI(4,5)P₂. Indeed, PI(5)P significantly enhanced GIRK2 activity (Fig. 5 c). These data demonstrate that the two phosphates in the 4′ and 5′ position of PIP₂ are necessary for complete channel activation but that the 5′-PO₄ of PIP₂ is sufficient to partially activate GIRK2 channels.

To determine whether PIP₂ channel interactions observed in WT (6′K) and 6′Y channel simulations correlate with gating, we examined the time-dependent changes of PIP₂ in all four PIP₂ pockets at the level of F192 (-2’F), which forms the HBC gate. To quantify the relative association of PIP₂ with all four pockets of the tetrameric channel during the simulation, we calculated the total number of the hydrogen bonds for 0′K–5′-PO₄, 5′K–5′-PO₄, and 6′K–5′-PO₄ and subtracted the number of hydrogen bonds for non-PIP₂ interactions (i.e., 6′K–9′E). This relative PIP₂ association number provides an estimate of the extent of PIP₂ interaction, where a high number indicates a stronger 5′-PO₄ interaction within its binding site. Over the 400 ns, the relative PIP₂ association number is ∼10 for 6′Y (Fig. 6 a). In contrast, this number decreases toward zero for the WT channel.

To correlate these effects with channel opening, we calculated the mean diameter across the pore at the HBC, F192 (−2′F). For the beginning of the WT 6′K simulation, the diameter at the HBC is ∼10 Å and decreases to <8 Å by the end of the simulation (Fig. 6 b). Coincident with the reduction in the PIP₂ association number with WT channels, water density decreases in the pore around the HBC gate (i.e., “dewetting” occurs; Aryal et al., 2014), consistent with a closed channel (Fig. 6 c, arrow). In contrast, the diameter of 6’Y channel increases from 10 Å to ∼13 Å (Fig. 6 b), indicative of an open channel or a channel moving toward an open state. Furthermore, the M2 helix appears to move around G180, a site previously implicated in channel gating (Meng et al., 2012), in the 6′Y simulations (Fig. 6 f). These differences in the diameter of the pore are readily observed in the cross-sectional view of the channel at the HBC at the end of the 400-ns stimulation (Fig. 6, d and e). Therefore, as the channel transitions from a low to high PIP₂ association number, the gate expands to ∼13 Å, and hydration of the pore occurs (Aryal et al., 2014), suggesting an open channel. The increase in water density in the pore around the HBC gate in the 6′Y channel, compared with WT, was also observed in the second 6′Y simulation (Fig. S6). Collectively, the MD simulations of WT-6′K and 6′Y channels reveal dynamic interactions of PIP₂ within its binding pocket that are important for unique PIP₂-dependent gating characteristics of GIRK2 channels.

Although it has been known for nearly 20 yr that PIP₂ is required for activation of GIRK channels (Huang et al., 1998; Sui et al., 1998), the structural details of how the association of PIP₂ with GIRK channels leads to channel activation remain poorly described. The initial crystal structures of Kir2.2 and GIRK2 channels provided snapshots of how PIP₂ binds to Kir channels, implicating positively charged, basic amino acids in the tether helix, the M2 TMD, and the N-terminal domain in the binding of one PIP₂ molecule (Hansen et al., 2011; Whorton and MacKinnon, 2011, 2013). Indeed, a comparison of the amino acids in the tether helix among different Kir channels reveals a high degree of conservation among these basic residues (Fig. S1 a). However, atomic resolution structures are static and lack the dynamic interactions of ligands associating with the channel and inducing gating conformations. In the current study, we combined functional studies with MD simulations to provide evidence for a dynamic balance of hydrogen bonds between the channel and PIP₂, with the 6′K in the tether helix serving a unique role in regulating gating by PIP₂.

Previous functional studies and mutagenesis experiments revealed some of the first clues on the mechanism of PIP₂-dependent gating of Kir channels. Comparing the rates of current rundown while depleting PIP₂ revealed two important features of PIP₂-Kir gating. First, Kir channels differ in their apparent affinity for PIP₂, whereby GIRK channels exhibit an apparent PIP₂ affinity that is significantly lower than that for Kir2 channels (Huang et al., 1998; Sui et al., 1998). Second, the apparent affinity of GIRK channels for PIP₂ increases in the presence of Gβγ subunits (Huang et al., 1998) or alcohol (Bodhinathan and Slesinger, 2013). Hypothesizing that negatively charged phosphates on PIP₂ associate with positively charged basic residues in the channel, Lopes et al. systematically mutated basic residues in Kir2.1 and studied the effect of PIP₂ antagonists (e.g., anti-PIP₂ antibody or polylysine) on current levels. In the tether helix of Kir2.1, mutation of K188 to Cys (6′C) significantly reduced the association with PIP₂ (Lopes et al., 2002). Modification of the 6′C with MTSEA, however, restored the inward current and PIP₂ sensitivity, suggesting modification with the partial charge was sufficient to support gating. The 6′Q substitution, on the other hand, did not produce a functionally conducting channel (Lopes et al., 2002). Similarly, Soom et al. (2001) demonstrated that a 6′Q in Kir2.1 reduced the currents and PIP₂ binding. Together, these results suggest that the positive charge at the 6′K position is important for coordinating PIP₂ in Kir2.1.

In contrast, we find that positive charge at the 6′ amino acid in the tether helix is not essential for PIP₂-dependent gating. Chemical modification with MTS-HE, MTS-F, and MTS-Y all increased the basal current. Furthermore, the conservative Arg substitution at the 6′K produced little change in the basal current in contrast to the Tyr substitution, which significantly enhanced the basal current. In addition, the 6′Y in GIRK2 appeared to increase the relative affinity for PIP₂, primarily affecting the recovery rate after PIP₂ depletion. Consistent with this finding, a glutamine substitution in Kir1.1 at K187 (6′Q) also appeared to strengthen the PIP₂ interaction, whereas the neighboring 7′Q substitution (R188Q) dramatically reduced PIP₂ association (Lee et al., 2016). Collectively, these findings raise the possibility that 6′K serves a unique role in GIRK2 as compared with that in Kir2.1 channels. Other differences exist between Kir2 and GIRK channels. Kir2 channels possess a secondary anionic phospholipid-binding site, which is absent in GIRK channels (Lee et al., 2013). Lastly, the cytoplasmic domain in the apo structure of Kir2.2 (i.e., no PIP₂) is not engaged with the TMDs, whereas it remains fully engaged in the apo GIRK2 structure (Tao et al., 2009; Hansen et al., 2011; Whorton and MacKinnon, 2011).

In the atomic resolution structures of GIRK2-PIP₂ and GIRK2-PIP₂/Gβγ, which are proposed to be in a closed and preopen conformation, respectively, the 6′K forms a hydrogen bond with the 5′-PO₄ of PIP₂ in both structures (Whorton and MacKinnon, 2011, 2013). Similarly, MD simulations with KirBac3.1/GIRK1 channels showed that 5′K and the 6′K strongly interact with PIP₂ in both the constricted and dilated forms (Meng et al., 2012). Together, these studies suggest the electrostatic association of 6′K with the negatively charged phosphate may have a more complex role than simply stabilizing PIP₂ in the pocket. Our MD simulations confirmed that the 6′K (WT) forms hydrogen bonds with the 5′-PO₄, but they also revealed that this interaction is dynamic. When the 6′K side chain points toward the TMDs (Fig. 6 g), the 5′-PO₄ of PIP₂ hydrogen bonds more tightly with 0′K and 5′K. Based on the functional studies of 6′Y mutation in GIRK2, which showed increased basal current and stronger interaction with PIP₂, we also performed MD simulations with the 6′Y channel. Here, 6′Y stabilizes this upward orientation, increasing the probability of hydrogen bond formation for the 0′K and 5′K with the 5′-PO₄ while decreasing the probability of hydrogen bond formation between K64 and the 5′-PO₄ of PIP₂. The movement of 6′K toward the cytoplasm and its subsequent interaction with E203 (9′E) may destabilize the interaction of 0′K and 5′K with the 5′-PO₄ of PIP₂, leading to closure of the HBC gate (Fig. 6 g) and perhaps explaining the low open channel probability for GIRK channels. K64 in the N-terminal domain may also contribute to PIP₂-dependent gating of GIRK channels by hydrogen bonding with 9′E. It is possible GIRK channel activators, like Gβγ and ethanol, promote the interaction of K64 with 9′E and allow the 5′-PO₄ to engage more strongly with the 0′K (Fig. 6 g). Interestingly, the homologous amino acid in Kir2.2 is a glutamine, supporting a possible unique role for K64 in GIRK channels.

The MD simulations and functional studies also highlight an important role for the 5′-PO₄ of PIP₂ to open GIRK2 channels. Consistent with this, we found that PI(5)P alone is sufficient to partially activate GIRK2 channels in reconstituted bilayers, which contrasts with its lack of effect on Kir2.1 channels (D’Avanzo et al., 2010). Furthermore, activation of Dr-Vsp, which dephosphorylates the 5′-PO₄ of PIP₂ to produce PI(4)P (Rjasanow et al., 2015), completely inhibits GIRK channel activity. Although PI(5)P is capable of activating, it is not present at high levels in the plasma membrane.

Importantly, the MD simulations suggest new PIP₂-bound states that appear to capture a transition state on the path toward opening, an event that appears difficult to observe in any of the Kir channel crystal structures (Hansen et al., 2011; Whorton and MacKinnon, 2011, 2013). Recently, Meng et al. (2016) successfully modeled an open state of a chimeric GIRK1 channel using an M2-helix mutation, M170P. In these MD simulations, the M1 and M2 helices rotate counterclockwise, with a significant bending at G169, leading to a widening of the pore at the HBC gate (−2′F) that is sufficient to support K⁺ permeation. Interestingly, the binding mode of PIP₂ did not change between the M170P-dilated and WT simulations. With the 6′Y MD simulations, however, we observe a similar movement of the M2 helix near G180, a widening of the HBC, and a tighter interaction of the 5′-PO₄ of PIP₂ with 0′K and 5′K of the channel. Additional mutagenesis experiments will be needed to fully validate this MD model of an open channel.

In conclusion, our experiments demonstrate that the highly conserved positive charge of the 6′K in GIRK2 supports a dynamic interaction with PIP₂ that reduces the likelihood of channel opening in the absence of other activators (e.g., Gβγ or ethanol). This model may explain the intrinsically low open probability of GIRK channels and perhaps how the direct binding of physiologically relevant activators, such as Gβγ or ethanol, can modify these interactions to produce robust channel activation. It will be interesting to determine whether any particular features of this mechanism are involved in the regulation of channel activity in other types of Kir channel or whether this remains unique to the GIRK family.

We thank Rod Mackinnon for providing GIRK2 cDNA and P. pastoris strain, Yasushi Okamura for Dr-Vsp cDNA, and Lindsey McCracken and Marian Fernando for creating some of the mutations.

This work was supported by the National Institute on Alcohol Abuse and Alcoholism (grants AA018734 to P.A. Slesinger and F32AA024671 to I.W. Glaaser), the National Institute on Drug Abuse (grant DA037170 to P.A. Slesinger), the Wellcome Trust (OXION training fellowship WT084655MA to P. Aryal and grant WT092970 to M.S.P. Sansom), the Biotechnology and Biological Sciences Research Council (grants BB/L002558/1 to M.S.P. Sansom and BB/N000145/1 to M.S.P. Sansom and S.J. Tucker), and the American Heart Association (postdoctoral fellowship 12POST9830002 to K. Bodhinathan).

The authors declare no competing financial interests.

José D. Faraldo-Gómez served as editor.