Inward rectifier potassium (Kir) channels are expressed in almost all mammalian tissues and play critical roles in the control of excitability. Pancreatic ATP-sensitive K (KATP) channels are key regulators of insulin secretion and comprise Kir6.2 subunits coupled to sulfonylurea receptors. Because these channels are reversibly inhibited by cytoplasmic ATP, they link cellular metabolism with membrane excitability. Loss-of-function mutations in the pore-forming Kir6.2 subunit cause congenital hyperinsulinism as a result of diminished channel activity. Here, we show that several disease mutations, which disrupt intersubunit salt bridges at the interface of the cytoplasmic domains (CD-I) of adjacent subunits, induce loss of channel activity via a novel channel behavior: after ATP removal, channels open but then rapidly inactivate. Re-exposure to inhibitory ATP causes recovery from this inactivation. Inactivation can be abolished by application of phosphatidylinositol-4,5-bisphosphate (PIP2) to the cytoplasmic face of the membrane, an effect that can be explained by a simple kinetic model in which PIP2 binding competes with the inactivation process. Kir2.1 channels contain homologous salt bridges, and we find that mutations that disrupt CD-I interactions in Kir2.1 also reduce channel activity and PIP2 sensitivity. Kir2.1 channels also contain an additional CD-I salt bridge that is not present in Kir6.2 channels. Introduction of this salt bridge into Kir6.2 partially rescues inactivating mutants from the phenotype. These results indicate that the stability of the intersubunit CD-I is a major determinant of the inactivation process in Kir6.2 and may control gating in other Kir channels.
Introduction
Inward rectifier potassium (Kir) channels are phospholipid-gated, tetrameric protein complexes that are crucial in membrane potential control in various tissues, including the brain, heart, pancreas, and skeletal and smooth muscle (Nichols and Lopatin, 1997). Seven distinct Kir channel subunit families (Kir1–7) differ in their biophysical properties, sensitivity to phospholipids, organ distribution, and cellular localization (Hibino et al., 2010). Each Kir subunit consists of intracellular N- and C-terminal domains, two transmembrane helixes (TM1 and TM2), and a potassium-selective pore loop (Bichet et al., 2003). The ATP-sensitive Kir (KATP) channel complex is unique in requiring four sulfonylurea receptor (SURX) subunits in addition to the four pore forming Kir6.x subunits (Inagaki et al., 1995, 1996; Shyng and Nichols, 1997). All eukaryotic Kir channels require binding of phosphatidylinositol-4,5-bisphosphate (PIP2) for activation (Huang et al., 1998; Shyng and Nichols, 1998; Zhang et al., 1999). In addition, ATP binding to the Kir6.x subunit inhibits KATP channels, whereas SURX subunits are responsible for MgADP activation and pharmacological modulation (Nichols et al., 1996; Gribble et al., 1997; Shyng et al., 1997b; Drain et al., 1998).
In WT Kir6.2 plus SUR1 KATP channels, removal of ATP from the cytoplasmic face results in a maintained channel opening. However, an early study attempting to identify positively charged Kir6.2 residues involved in PIP2 binding revealed three intracellular arginine residues that, when mutated to alanine (R192A, R301A, and R314A), all resulted in channel activation after removal of intracellular ATP but then rapid channel closure (Shyng et al., 2000). This phenomenon was termed “inactivation” as channel activity reappeared (and again decayed) after subsequent application and removal of ATP. In a follow-up study (Lin et al., 2003), it was shown that individual R314E/C or E229R/C mutations also induced an inactivation phenotype, whereas double charge reversal (R314E/E229R) or double cysteine substitutions (R314C/E229C) reverted to a noninactivating behavior. R314 and E229 map to the intersubunit cytoplasmic domain interface (CD-I), suggesting that salt bridge–induced stability of this interface might normally be important for maintenance of channel activity.
Pancreatic KATP channels composed of Kir6.2/SUR1 subunits are active during low metabolic conditions, keeping the β cell membrane hyperpolarized and inhibiting insulin secretion (Nichols and Remedi, 2012). Gain-of-function mutations in either Kir6.2 or SUR1 result in maintained β cell hyperpolarization, leading to inhibition of insulin secretion, and neonatal diabetes mellitus (Koster et al., 2000; Gloyn et al., 2004). Conversely, loss-of-function (LOF) mutations result in maintained β cell electrical activity and hyperinsulinism (HI; Nichols et al., 1996). LOF in such mutations can result from decreased surface expression through deficits in protein production and membrane trafficking but can also arise from decreased channel activity (Koster et al., 2002; Remedi et al., 2006; Loechner et al., 2011; Martin et al., 2013). Several such HI-linked Kir6.2 LOF mutations (Shyng et al., 2012) reduce channel activity by inducing an inactivating behavior. The majority of these mutations are located at the CD-I and are predicted to disrupt salt bridge contacts between subunits, suggesting that disruption of CD-I interactions might be the molecular mechanism responsible for the inactivation and ultimately the disease phenotype (Lin et al., 2008, 2013; Shimomura et al., 2009; Bushman et al., 2010, 2013; Loechner et al., 2011).
This inactivation phenomenon has only been identified in Kir6.2-containing KATP channels. In this study, we set out to further examine the underlying molecular mechanism and to consider the possibility that manipulation of the CD-I stability may affect gating in other Kir channels.
Materials and methods
Molecular biology and heterologous expression system
Cosm6 cells were grown in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin in 95% O2/5% CO2 at 37°C. Cells between passages 10 and 30 were plated on glass coverslips in 40-mm, 6-well plates. At 30–50% confluency, cells were transiently transfected with plasmids in pcDNA3.1 encoding either mouse Kir6.2 (mKir6.2) and hamster SUR1, or human Kir2.1 (hKir2.1), along with green fluorescent protein (GFP) using Fugene6 (Promega). Each well was transfected with 0.2 µg GFP and either 0.3 µg mKir6.2 and 0.5 µg SUR1 or 0.8 µg hKir2.1. Transfection mixtures were incubated for 1.5 h at room temperature and then added to each corresponding well. Mutations in both mKir6.2 and hKir2.1 were introduced using the QuikChange II method (Agilent Technologies), and the entire coding region was verified by sequencing.
Electrophysiology
Cells at 70–90% confluency were used for electrophysiological experiments, 24–48 h after transfection, with electrodes pulled from glass capillaries with a final resistance of 1.0–2.5 MΩ. All experiments were performed using a symmetrical high-potassium pipette and bath solution (Kint) consisting of (mM) 140 KCl, 1 EGTA, 1 K2-EDTA, and 4 K2HPO4 and adjusted to pH 7.4 (KOH). Glass coverslips plated with transfected cells were first broken into shards and placed in an oil-gate perfusion chamber (Lederer and Nichols, 1989). Fluorescent cells expressing GFP were selected for patch clamp experiments. Current responses were low-pass filtered at 1 kHz (Axopatch 1B), sampled at 5–10 kHz (Digidata 1200), converted to digital files in Clampex7 (Molecular Devices), and stored on an external hard drive for offline analyses.
Patches were excised into zero ATP solutions, and KATP channel currents were recorded with a gap-free continuous voltage-clamp protocol (−50 mV membrane potential). WT KATP channel activity is characterized by a stable steady-state current response; however, over time the current slowly runs down. This process is correlated with the loss of inner leaflet membrane PIP2 through diffusion or hydrolysis by various phosphatases and can be minimized by including divalent ion chelators in the recording solutions (Hilgemann and Ball, 1996; Ribalet et al., 2000). To minimize rundown, EDTA was therefore included in all phosphate-buffered Kint recording solutions used throughout these experiments.
ATP (Sigma-Aldrich) solutions were prepared daily from 200 mM stocks in Kint (pH 7.4, KOH) and stored at −20°C. Working stocks of 5 µg/ml (4.8 µM) porcine brain PIP2 (Avanti Polar Lipids, Inc.) in Kint were stored at −80°C and sonicated on ice for 20 min before daily use.
Kir2.1 PIP2 potentiation experiments were performed with a continually repeating voltage ramp protocol (holding potential 0 mV, ramp from −100 to 100 mV potential over 1 s). Current responses were elicited first in the cell-attached configuration, during patch excision into the inside-out configuration, and then, after loss of rectification, during subsequent application of PIP2. Leak conductance was assessed at the end of each recording by the degree of residual current at −100 mV after application of 10 µM spermine (Fluka). [PIP2] responses were assessed with a gap-free continuous voltage-clamp protocol at −100 mV. After excision, the patch was briefly exposed to a 10 mM CaCl2 Kint solution to deplete endogenous PIP2 before application of 1,2-dioctanoyl-sn-glycero-3-phospho-(1′-myo-inositol-4′,5′-bisphosphate) (DiC8-PIP2; Avanti Polar Lipids, Inc.; prepared daily from a 100 µM stock stored at −80°C) Kint solution of varying concentrations. A high concentration (5 µg/ml) of brain PIP2 was applied at the end of the recording to determine the maximum current response.
Homology modeling
A homology model of mKir6.2 was built using MODELLER9.8 (Eswar et al., 2007) using chicken Kir2.2 (cKir2.2; PDB ID 3SPI) as a template (Hansen et al., 2011). A pairwise sequence alignment was performed using ClustalW2 (Sievers et al., 2011), and 50.6% sequence identity was observed when compared for the region structurally resolved in the cKir2.2 crystal structure (cKir2.2: 43–372; mKir6.2:33-359). A tetrameric cKir2.2 structure was generated using PyMOL (PyMOL Molecular Graphics System, version 1.8; Schrödinger, LLC), and the tetramer was used to build a tetrameric mKir6.2 homology model.
Data analyses
Inactivation parameters
KATP channel inactivation parameters were evaluated as follows. The remaining current after inactivation was calculated as the ratio of steady-state (Iss) to peak (Ipk) current (Iss/Ipk) after ATP removal, and extent of inactivation was calculated as (1 − Iss/Ipk). The time course of inactivation (τinact) was determined by fitting a mono-exponential function, A(t) = A1e−t/τ, to the current decay phase, in Clampfit (Clampfit 10.3 software; Molecular Devices). The degree of PIP2 potentiation was calculated from the final current in zero ATP (Imax) after application of PIP2, and the initial current in zero ATP (Iinitial), before PIP2 application, for both Ipk (Ipk PIP2 potentiation) and Iss (Iss PIP2 potentiation) current amplitude after patch excision: PIP2 potentiation = [Imax − Iinitial]/Iinitial.
Single-channel recordings
Preprocessing and analyses were performed offline with QuB kinetic and statistical software (University at Buffalo, Buffalo, NY). Individual traces were preprocessed to correct for instances of baseline drift and infrequent noise spikes. Single-channel amplitudes were first estimated and fit with the “Amps” function, and the corresponding closed and open amplitudes were idealized with a 50% threshold method with no imposed dead time (Qin, 2004). Idealized data were segmented by the number of active channels, and only single-channel bursts were used for state and kinetic modeling. Simple 1 open (O) and 2–4 closed (C) state models (Cn+2 ↔ Cn+1 ↔ Cn ↔ O ↔ Cf) were constructed and used to fit the open interval distributions and to estimate mean open times (MOTs) and transitions rate constants (O → Cf, k+Cf; Cf → O, k-Cf; and O → Cn, k-1) using the maximum interval likelihood (MIL) algorithm in QuB with an imposed 0.4-ms dead time for each individual recording (Qin et al., 1996, 1997; Qin and Li, 2004).
Stationary noise analysis
1-s segments of KATP channel currents recorded at −50 mV in zero and 1 mM ATP were used to estimate open probability (Po) in zero ATP (Po,zero) from individual macroscopic current recordings containing an unknown number of channels (N) with Eq.1:
where I is the mean current in zero ATP − mean current in 1 mM ATP, σ2 is the difference in the mean variance measured at zero and 1 mM ATP, and i is the single-channel amplitude (3.75 pA; Shyng et al., 1997a; Enkvetchakul et al., 2000).
Kir2.1 PIP2 potentiation
The degree of potentiation of Kir2.1 currents by PIP2 was calculated as the fold increase of the excised current amplitude (at −100 mV) after application of PIP2. The DiC8-PIP2 dose–response was fitted in OriginPro by Eq. 2:
where
is the current (IDiC8) at defined concentration of DiC8-PIP2 [DiC8] normalized to the maximum PIP2 potentiated current (IPIP2), EC50 is the half-maximal effective DiC8-PIP2 concentration, and nH is the Hill coefficient.
Statistics
Values are reported as mean ± SEM, and differences were evaluated with a two-tail Student’s t test and were considered significant for P < 0.05.
Results
Neutralization of CD-I–positive charges induces KATP current inactivation
WT KATP channels display negligible activity in cell-attached patch-clamp configuration, being mostly inhibited by intracellular ATP. Patch excision relieves this inhibition, resulting in maintained channel activity that is reversibly inhibited by subsequent application and removal of ATP (Fig. 1 A). Neutralization of certain positively charged intracellular residues, including R192A located at the CD-I (Fig. 1 B), has previously been demonstrated to induce a rapid spontaneous current decay following channel activation after patch excision (Shyng et al., 2000). This inactivation phenomenon is repeatedly observed after subsequent application and removal of ATP (Fig. 1 C).
Various models of KATP channel activity have been proposed to describe both macroscopic and microscopic kinetic parameters and sensitivity to numerous ligands (Nichols and Lederer, 1991; Alekseev et al., 1997; Enkvetchakul et al., 2000; Proks et al., 2001). Although more complexity is required to account for detailed single-channel behavior, a simplified three-state model in which the unliganded closed state (C0) can either bind ATP to stabilize the closed channel (CA) or open (O) as a result of binding PIP2 (Fig. 1 D, Model I) can adequately describe macroscopic channel properties (Enkvetchakul and Nichols, 2003). As discussed further below, addition of a fourth state, the closed inactivated (CI) state, accessible from the C0 state or perhaps the O state (see below; Fig. 1 D, Model II or IIa), can then account for the novel inactivating behavior exhibited by the R192A mutation (Loechner et al., 2011).
Residues responsible for KATP channel inactivation are conserved throughout Kir channels
A similar inactivating phenotype has been shown for mutations at residues R314 and E229 (Shyng et al., 2000; Lin et al., 2003). These residues are conserved across the Kir channel family (Fig. 2 A) and, although there is no available Kir6.2 crystal structure, homology models based on crystal structures of chicken Kir2.2 (cKir2.2) indicate that these residues are all located close to one another and form a network of intersubunit contacts (Fig. 2 B; Tao et al., 2009; Hansen et al., 2011). R314 as well as R192 from the same subunit is predicted to interact with E229 in the adjacent subunit. Human Kir2.1 (hKir2.1) contains the equivalent electrostatic contacts, as well as an additional salt bridge, that is not present in Kir6.2. This is formed by residue D205, located one residue below R204 (equivalent to the conserved Kir6.2[R192]) and R235 on the adjacent subunit (Fig. 2 C). The equivalent residues in Kir6.2 are H193 and T223, but their relatively short sidechains are predicted to be too distant to interact (Fig. 2 B).
Strength of CD-I contacts controls the extent and time course of inactivation
Neutralization of any member of the intersubunit salt bridge network between R192, E229, and R314 induces inactivation in Kir6.2 (Lin et al., 2003). This suggests that inactivation may result from a relative dissociation of the subunits from one another and leads us to specifically hypothesize that the strength of CD-I interactions will control the degree of channel inactivation. Thus, we predict that weaker CD-I interactions will result in a greater degree of channel inactivation and that, vice versa, stronger CD-I interactions between subunits could result in less inactivation.
To test this hypothesis, we examined gating in a series of Kir6.2 mutant channels in which we manipulated the number and location of CD-I salt bridges by mutagenesis. Fig. 3 shows representative recordings from this series of mutant channels. There is essentially no inactivation in mutants for which the endogenous R192–E229 salt bridge is preserved (e.g., H193D) or in which the additional salt bridge equivalent to the Kir2.1 D204–R235 interaction is introduced (e.g., H193D–T223R). However, in the absence of the H193D mutation, introduction of the T223R mutation itself causes inactivation, potentially caused by introduction of repulsive interaction between positively charged R223 and H193 sidechains. Significant inactivation is present in the R192A mutant, as well as R192A/H193D and R192A/T223R double mutants, in all of which the endogenous R192A-E229 salt bridge is broken. Importantly, both the extent and rate of inactivation resulting from R192A are markedly reduced by introduction of the second Kir2.1 equivalent salt bridge (H193D–T223R) in R192A/H193D/T223R channels. Inactivation parameters immediately after patch excision, and then degree of potentiation by PIP2, were assessed for all CD-I mutants (Fig. 4). Channels containing the R192A mutation (R192A, R192A/H193D, and R192A/T223R) all exhibited >70% inactivation (Fig. 4, A and B). Inactivation of these channels was well fitted by a single exponential component with a time constant of inactivation (τinact) of 2–4 s. Importantly, introduction of the H193D–T223R salt bridge on the R192A background (in the triple R192A/H193D/T223R mutant) significantly reduced the extent of inactivation (to ∼50%) and slowed τinact (to >5 s; Fig. 4, A and B).
The CD-I controls inactivation specifically, without affecting open state stability or ATP sensitivity
All Kir channels are activated by PIP2, which shifts the open probability (Po) to favor the O state. In WT Kir6.2 + SUR1 KATP channels, the Po in zero [ATP] (Po,zero) is typically ∼0.4–0.6 (Inagaki et al., 1995; Trapp et al., 1998; Enkvetchakul et al., 2000, 2001; Ribalet et al., 2006). A simplified linear three-state model of KATP gating in which the channel exists in open (O), unliganded closed state (C0), or closed ATP bound state (CA) can account for reduced apparent sensitivity to ATP and a characteristic and very nonlinear relationship between apparent K1/2,ATP ([ATP] causing half-maximal inhibition) and Po,zero (Enkvetchakul et al., 2000, 2001; Enkvetchakul and Nichols, 2003), by PIP2 acting to shift the C0 ↔ O equilibrium toward the O state (Fig. 1 D). Mutations that intrinsically stabilize the O state can be introduced at multiple locations, prominently in the pore of the channel (Loussouarn et al., 2000, 2001; Enkvetchakul et al., 2001), and these result in shifts along this same relationship (Enkvetchakul and Nichols, 2003). Inactivation might proceed from either the unliganded closed (C0) or the open (O) state (Fig. 1 D). If the latter, then PIP2 would also have to shift the O ↔ CI equilibrium toward the O state, but either model would then predict that application of PIP2 will also decrease steady-state occupancy of the CI state (i.e., the extent of inactivation) as the O state is stabilized. As can be seen from the recordings in Fig. 3, all mutant channels responded to PIP2 as predicted, with not only profound loss of ATP sensitivity, but also loss of macroscopic inactivation. The potentiation of Iss was ∼0.5-fold for WT channels but was more marked, essentially in proportion to the initial extent of inactivation, in the CD-I mutant channels (Fig. 4 C).
An important prediction of either model is that the inactivation process itself is independent of the intrinsic ATP-independent gating. Therefore, if CD-I mutations affect specifically the inactivation process, the CA ↔ C0 ↔ O ↔ Cf equilibria should be unaffected. Given the kinetic separation between channel activation after ATP removal and the relatively slow inactivation, Ipk can provide an estimate of channel activation (i.e., CA → C0 ↔ O ↔ Cf transitions). As shown in Fig. 4 D, there was no significant difference in the degree of Ipk potentiation by PIP2 (Ipk was potentiated ∼0.5-fold by PIP2 for all mutations), implying that the initial peak Po,zero and hence channel activation transitions are the same for all channels. Consistent with this conclusion, Po,zero calculated from stationary noise analysis (Eq. 1) was not significantly different between WT (0.52 ± 0.05, n = 6) and H193D/T223R (0.45 ± 0.06, n = 5) channels, further indicating a similar C0 ↔ O ↔ Cf equilibrium.
Recordings from inside-out patches with resolvable single-channel openings revealed no significant difference in single-channel amplitudes (Fig. 5, A and B) nor MOTs between WT, R192A, or R192A/H193D/T223R channels (Fig. 5 D). There was no significant difference in the open state lifetime (Fig. 5 C) between WT and mutant channels, indicating that the strength of the CD-I interface does not contribute to the stability of the O state (Table 1). WT Kir6.2/SUR1 channels display prolonged bursts of channel openings and brief high-frequency closures (Alekseev et al., 1998; Trapp et al., 1998; Enkvetchakul et al., 2000). Apparent burst lengths were shorter and prolonged closures were longer for both R192A and R192A/H193D/T223R channels in patches with only single openings detectable after patch excision. However, all such patches actually contained many more channels that were trapped in prolonged closures, as was obvious after application of PIP2 (Fig. 5 E), and thus measurements of these prolonged closures would be underestimated and were not quantified.
Conserved and unique CD-I salt bridges also control Kir2.1 channel activity
Although mutations that disrupt the R192–E229 CD-I salt bridge in Kir6.2 cause an obvious inactivating phenotype in Kir6.2 channels, the consequences of similar mutations at equivalent residues in other Kir channels are not known. Kir2.1 channel gating is complex and not well characterized; channels exhibit one to four open conductance levels (Lu et al., 2001; Schwalbe et al., 2002; Amorós et al., 2013) and require binding of a secondary anionic phospholipid (PL) for activity (Lee et al., 2013). Lack of any rapid inhibitory ligands (i.e., equivalent to ATP for Kir6.2) makes it impossible to directly assess any macroscopic inactivation. However, if we were to assume similar minimalistic gating models to those proposed for Kir6.2 (Fig. 1 D), but lacking the ATP-bound CA state, occupancy of CI in Kir2.1 would still be reflected in the degree of PIP2 potentiation of steady-state current. WT, R204A, R235T, and R204A/R235T mutant Kir2.1 channel activity was assessed after the loss of polyamine-dependent rectification that occurs after patch excision (Fig. 6 A; Lopatin et al., 1994; Kurata et al., 2010). WT channel currents exhibited only minimal PIP2 potentiation, consistent with a high intrinsic Po, but the R204A mutation (which disrupts the R204–E241 salt bridge) resulted in significantly greater potentiation. Abolishing the D205–R235 salt bridge with the R235T mutation also resulted in greater PIP2 potentiation, and these two effects were essentially additive in the R204A/R235T double mutation (Fig. 6, A and B). The change in degree of potentiation by PIP2 indicates that the strength of the CD-I indeed also controls an underlying long closure in Kir2.1 channels.
Weakening the Kir2.1 channel CD-I lowers PIP2 sensitivity
The consequence of a weaker CD-I on PIP2 sensitivity was examined to further characterize the role of these contacts in Kir2.1 channel gating. Brief exposure to 10 mM CaCl2 abolishes Kir2.1 channel currents by sequestering or depleting endogenous PIP2 from excised patch membranes (Suh and Hille, 2008), such that channel activity can only be restored by subsequent application of exogenous PIP2 (Fig. 7 A). To obtain a direct measure of sensitivity to PIP2, the current recovery was measured by application of short chain DiC8-PIP2, after CaCl2-mediated PIP2 depletion (Fig. 7 B). The DiC8-PIP2 response relationship was fit with a modified Hill equation (Eq. 2). From such empirical fits, it is unclear whether the apparent change in slope is mechanistically meaningful, but the half-maximal DiC8-PIP2 concentration (EC50) calculated from the dose–response relationship (Eq. 2) significantly increased with the number of abolished CD-I contacts (Fig. 7 C).
Discussion
Kir6.2 inactivation results from destabilization of the unliganded closed channel
Mutations of multiple residues in Kir6.2 (G156, R192, S208, S225, E229, V290, R301, and R314; Shyng et al., 2000; Lin et al., 2003, 2008, 2013; Loechner et al., 2011; Bushman et al., 2013) have now been shown to result in the distinct phenomenon whereby channel opening after removal of ATP is rapidly followed by channel closure. We have termed the phenomenon inactivation, by analogy to the closure of voltage-gated channels that follows channel activation in response to depolarization. All of these mutations, with the exception of G156, are located in or close to the CD-I (Fig. 9 A), and in each case, inactivation is attenuated as membrane PIP2 levels are increased (Fig. 3; Shyng et al., 2000; Lin et al., 2003; Loechner et al., 2011; Bushman et al., 2013). WT Kir6.2 channel activity does not show any appreciable inactivation (Fig. 1 A) and can be modeled by assuming that each channel (or more accurately by assuming each subunit) exists in just three or four states (CA, C0, O, and Cf). Such models can account quantitatively for the very nonlinear change of apparent ATP sensitivity that occurs with change in Po,zero, whether these are induced by mutations within the channel pore or by changing membrane PIP2 levels (Trapp et al., 1998; Enkvetchakul et al., 2000, 2001). Inclusion of an additional CI state, accessed from the unliganded C0 state, can qualitatively explain inactivation (Enkvetchakul et al., 2001; Enkvetchakul and Nichols, 2003; Lin et al., 2003; Loechner et al., 2011). In the present paper, we have identified additional CD-I mutations that induce inactivation by breaking predicted salt bridges, as well as mutations that are predicted to introduce a novel salt bridge based on structural similarity to Kir2 channels (Fig. 2). Importantly, introduction of this salt bridge into inactivating Kir6.2 mutant channels slows and reduces the extent of inactivation.
Modeling Kir6.2 channel inactivation
The addition of an inactivated state is necessary to account for the macroscopic current decay after removal of ATP. Analysis of specific mutations indicates that they do not affect the apparent O state stability or the intraburst gating (i.e., the C0 ↔ O ↔ Cf equilibria; Fig. 5 and Table 1). These findings solidify the conclusion that the inactivation phenomenon is a reflection specifically of the transition to the CI state and that mutations in the CD-I which induce the phenomenon (see below) are acting exclusively on this process, without affecting the stability of the “bundle crossing” gate that controls the C0 ↔ O equilibrium (Jiang et al., 2002). Consistent with this notion, Bushman et al. (2013) noted that, when PIP2-dependent open state stability was substantially increased by addition of a second gain-of-function mutation, G156P inactivation was abolished. Conceivably, the inactivated (CI) state could be accessed from either the unliganded C0 state or the open (O) state. The experimental observations that (a) for different mutations, the greater the extent of inactivation, the faster the rate of inactivation (Fig. 8 B) and that (b) when PIP2 is added, the rate of inactivation slows as the extent of inactivation decreases are readily accounted for by CI being accessed from the unliganded closed state. If accessed from the open state, PIP2 would have to act to slow the O → CI transition. Moreover, if CI were accessed from the open state, the reduced O to CI rate would also prolong the open or the open burst time. Instead, we see reduction of the interburst intervals and, for this reason, favor the interpretation that CI is accessed primarily from the C0 state. However, realistic gating models of KATP activity are complex (Enkvetchakul et al., 2001; Enkvetchakul and Nichols, 2003; Proks and Ashcroft, 2009), and firm conclusions are difficult to draw.
Kir6.2 inactivation results from destabilization of the CD-I, and the same gating process may be common to other Kir channels
Lin et al. (2003) first recognized that inactivation could be associated with breaking of salt bridges or charge networks at the CD-I. Here, we have confirmed the generality of this mechanism: breaking of the R192–E229 salt bridge results in Kir6.2 inactivation, and inactivation is reversed by introduction of the equivalent of the D205–R235 salt bridge found in Kir2.1. Importantly, breaking of either the D205–R235 or the R204–E241 salt bridges in Kir2.1 also results in decreased channel activity. WT Kir2.1 typically has a high Po after patch excision (Lopes et al., 2002; Schwalbe et al., 2002), and there is minimal activation by subsequent addition of exogenous PIP2 (Fig. 6 B). However, in Kir2.1[R204A] or in Kir2.1[R235T], basal channel activity is lower than WT, and the current is markedly potentiated after application of PIP2 for both mutations (Fig. 6 B). Moreover, the extent of potentiation increases with the number of salt bridges broken; channel activity is even lower, and PIP2 stimulation greater, in the R204A/R235T double mutation (Fig. 6 B). Without rapid activation gating driven by the dissociation of ATP, it is impossible to directly observe inactivation as a distinct macroscopic kinetic process in Kir2.1 channels, but the same gating scheme that describes Kir6.2 gating (but lacking the CA state) could account for effects of salt bridge mutations on Kir2.1 gating. Thus, we suggest that the underlying molecular mechanism is likely common to both Kir6.2 and Kir2.1. Given the highly conserved nature of the Kir cytoplasmic domain structure (Fig. 2) and the presence of identical predicted or observed salt bridges in other members of the eukaryotic and prokaryotic Kir channel families (Kuo et al., 2003; Clarke et al., 2010; Hansen et al., 2011; Whorton and MacKinnon, 2011; Zubcevic et al., 2014; Linder et al., 2015), we predict that the mechanism may be common to all members.
What is the structural basis of inactivation?
The CD-I is not only bridged by charge networks in the protein itself, it is also the location of the binding site for PIP2 in Kir channels (Hansen et al., 2011; Whorton and MacKinnon, 2011), as well as the nonspecific anionic lipid site in Kir2.1 (Lee et al., 2013, 2016) and the ATP-binding site in Kir6.2 (Enkvetchakul and Nichols, 2003; Trapp et al., 2003; Antcliff et al., 2005). The simple state model that we use to account for Kir channel gating thus suggests an appealing physical mechanism for the inactivation phenomenon. Clearly, inactivation and recovery are relatively slow processes occurring over seconds, which might suggest that a relatively large protein motion is involved. Given that inactivation is induced and enhanced by breaking of salt bridges between subunits, we hypothesize that disruption of these salt bridges makes the unliganded C0 state physically unstable, such that inactivation (reflecting the C0 → CI transition) might actually result from the CD-I essentially disconnecting. Because the inhibitory ATP-binding site is also located in the CD-I, ATP binding may also act to stabilize the connected CD-I (but with the channel closed, i.e., in the ATP-bound [CA] state), such that after removal of ATP, the channel rapidly transitions from CA → C0 → O states and reaches a pre–steady state before equilibrating with the CI state. The lack of effect of inactivation-inducing mutations on either the degree of Ipk potentiation by PIP2 or the O state stability is consistent with the finding that these mutations have no effect on channel activation (i.e., the CA → C0 ↔ O ↔ Cf transitions; Fig. 8 A). In the presence of high [ATP], channels will be gradually trapped in the absorbing CA state as the CD-I undergoes spontaneous reconnection (i.e., as the subunit undergoes occasional transitions out of the CI state) such that, macroscopically, channels will open and then inactivate after ATP addition and removal. The strong positive correlation between the residual noninactivating current and τinact (Fig. 8 B) is consistent with the major consequence of disruption of the CD-I being to specifically increase the rate constant for entry into the CI state. Because PIP2 also binds at the subunit interface, it may also intrinsically stabilize the “connected” CD-I, shifting the C0 ↔ O equilibrium toward the O state, thereby reducing the extent and rate of inactivation.
Two distinct Kir2 channel structures have thus far been identified crystallographically: one in the absence of PIP2 (Tao et al., 2009), in which the PIP2-binding site is unstructured, and one bound to PIP2 (Hansen et al., 2011), in which the binding site is highly structured and the whole cytoplasmic domain is pulled closer toward the membrane surface (Fig. 9 B). Comparison of the CD-I in the apo (PDB ID 3JYC) and PIP2-bound (PDB ID 3SPI) Kir2.2 structures indicates major differences in the disposition of the cytoplasmic domain relative to the transmembrane domain (Fig. 9 B): the whole cytoplasmic domain is displaced from the membrane ∼6 Å. This is accommodated by unfolding of the linkers between the cytoplasmic N and C termini, leading to unstructuring of the PIP2-binding site. Close inspection also indicates subtle yet potentially significant differences in the CD-I itself: an alignment of both structures at R204 illustrates how the CD-I widens ∼3 Å in the apo conformation such that the R235–D205 and E241–R325 residue pairs no longer form contacts (Fig. 9 C). These structural changes are very consistent with the novel hypothesis that the apo (3JYC) structure, rather than representing channels in the unliganded C0 state, is actually in the inactivated (CI) conformation (Fig. 10). This could then explain both why PIP2 cannot bind to the CI state (because the PIP2-binding site is unformed in this structure) and the role of CD-I salt bridge destabilization in enhancing the transition rate into the CI state.
If the PIP2-bound 3SPI and 5KUM structures represent the PIP2-bound O and Cf states, and 3JYC represents the CI state, the question then arises: what is the structure of the C0 state? In addition to PIP2, Kir2 channel activity normally requires additional binding of anionic PL at a distinct nonspecific site (Lee et al., 2013). We have recently identified additional cKir2.2 structures in which introduction of a tryptophan at the distinct “anionic lipid”–binding site mimics the anionic lipid-bound state and allows activation by PIP2 in the absence of PL (Lee et al., 2016). The cytoplasmic domain is tightly tethered to the membrane surface, and the PIP2-binding site is fully structured in cKir2.2[K62W] crystal structures, in both the absence (PDB ID 5KUK) and the presence of PIP2 (PDB ID 5KUM; Lee et al., 2016). Moreover, the CD-I is closed, with the inter-charge distances at the CD-I being nearly identical to those in the 3SPI structure. We thus suggest that 5KUK, in which the cytoplasmic domain is tethered to the membrane and the PIP2 site is formed but not occupied by PIP2, may reflect the C0 state (Fig. 10).
Parallels to glutamate receptor desensitization
Ionotropic glutamate receptors (iGluRs) are ancestrally related to K channels, with high structural similarity in the TM1 and TM2 transmembrane domains (M1 and M3 in iGluRs; Sobolevsky et al., 2009), although they are inserted in opposite orientation in the membrane, such that the extracellular bilobal ligand-binding domains (LBDs) are analogous to the intracellular cytoplasmic domains in Kir channels. There are evident parallels between the physical mechanism we propose for inactivation and the extensively studied structural mechanism of iGluR desensitization (Sobolevsky, 2015). The LBDs are dimerized in a back to back orientation, which allows the transduction of agonist induced LBD closure to open the channel gate. However, inherent strain in this conformation results in rupture of the LBD dimer interface, closing the channel while still bound to agonist (Kumar and Mayer, 2013). Recent cryo-EM structures of iGluRs in the resting, preopen, and desensitized states reveal that the structural rearrangements associated with desensitization involve larger domain movements of multiple intersubunit interfaces than previously thought (Dürr et al., 2014; Zhu et al., 2016). Direct evidence for or against the structural basis that we propose for Kir channel inactivation (Fig. 10) awaits further studies, but parallels between Kir channels and iGluRs might be a clue to a common role of intersubunit rearrangement in desensitization and inactivation in multiple ion channel families.
Acknowledgments
This work was supported by the National Institutes of Health grants HL54171 and DK109407 (to C.G. Nichols). W.F. Borschel was supported by National Institutes of Health training grants DK007120-38, HL007275-35, and HL125241-01.
The authors declare no competing financial interests.
Author contributions: W.F. Borschel, S. Wang, and C.G. Nichols conceived the project. S. Lee performed structural homology modeling. W.F. Borschel carried out experiments. W.F. Borschel and C.G. Nichols analyzed the data and wrote the paper, which was edited by S. Wang and S. Lee.
Richard W. Aldrich served as editor.