The Lipid Activation Mechanism of a Transmembrane Potassium Channel

Abstract

Membrane proteins and lipids coevolved to yield unique coregulatory mechanisms. Inward-rectifier K⁺ (Kir) channels are often activated by anionic lipids endemic to their native membranes and require accessible water along their K⁺ conductance pathway. To better understand Kir channel activation, we target multiple mutants of the Kir channel KirBac1.1 via solid-state nuclear magnetic resonance (SSNMR) spectroscopy, potassium efflux assays, and Förster resonance energy transfer (FRET) measurements. In the I131C stability mutant (SM), we observe an open-active channel in the presence of anionic lipids with greater activity upon addition of cardiolipin (CL). The introduction of three R to Q mutations (R49/151/153Q (triple Q mutant, TQ)) renders the protein inactive within the same activating lipid environment. Our SSNMR experiments reveal a stark reduction of lipid–protein interactions in the TQ mutant explaining the dramatic loss of channel activity. Water-edited SSNMR experiments further determined the TQ mutant possesses greater overall solvent exposure in comparison to wild-type but with reduced water accessibility along the ion conduction pathway, consistent with the closed state of the channel. These experiments also suggest water is proximal to the selectivity filter of KirBac1.1 in the open-activated state but that it may not directly enter the selectivity filter. Our findings suggest lipid binding initiates a concerted rotation of the cytoplasmic domain subunits, which is stabilized by multiple intersubunit salt bridges. This action buries ionic side chains away from the bulk water, while allowing water greater access to the K⁺ conduction pathway. This work highlights universal membrane protein motifs, including lipid–protein interactions, domain rearrangement, and water-mediated diffusion mechanisms.

Graphical Abstract

INTRODUCTION

The mutual regulation and interplay between membrane proteins (MPs) and their native bilayers is fundamental to life. MPs, encoded by a third of mammalian reading frames, are targets for over 60% of known pharmaceuticals.¹ Inward-rectifier K⁺ (Kir) channels are MPs found within all excitable cells and are responsible for establishing and maintaining resting membrane potential. They are implicated in multiple forms of heart disease, mental illness, and kidney/liver disorders.² Kir channels carry out their unique biological roles via their preference for inward K⁺ conductance^3,4 crucial for facilitating and maintaining hyperpolarization of cardiac muscle, neurons, and other excitable cells. KirBac1.1, a bacterial inward rectifier from native species B. pseudomallei,⁵ shares 52% sequence homology with mammalian Kir1 and Kir2 channels and a majority of the structural and topological features common to all Kir channels (Figure 1a,b).⁶ These homotetrameric proteins are composed of an eight-helix transmembrane bundle (two helices from each monomer), which forms an “inverted teepee” transmembrane pore. This transmembrane region is flanked by the K⁺ selectivity filter facing the extracellular space and an activation gate at the cytoplasmic interface. The activation gate is adjoined by an intracellular Kir gating domain formed by the union of the N- and C-termini. While a dimer of dimers structure was resolved in the original X-ray crystal structure,⁶ it is likely that many variations of closed and open conformations exist in vivo as revealed in more recent studies.^7–10 One universal structural feature is a large cationic pocket at the intracellular water–membrane interface formed by basic residues in the slide helix and a flexible linker (c-linker) that joins the inner transmembrane helix (TM2) with the C-terminal domain (CTD) (Figure 1a–d, blue highlighting). This locus of anionic lipid binding was identified in multiple X-ray crystallography studies of mammalian and bacterial Kir channels.^7,9,10 However, detailed structural changes brought about by anionic lipids have not been fully determined in biologically relevant contexts, though multiple X-ray crystallographic, Förster resonance energy transfer (FRET), and MD studies^11,12 have pointed to concerted motions of the C-terminal domain in conjunction with opening of the activation gate. However, all Kir channel structures solved in the presence of lipid head groups are closed.^7,9,10 Only with targeted mutations to force the channel into an open conformation has a purported activated state been captured via crystallography.⁸ However, in vivo Kir channels are constitutively open and active when tethered to anionic moieties.¹³ FRET studies have described the global gating motions of this channel in a native like environment.^14,15 However, these studies lack information at an atomistic level and describe the PIP₂ closed state rather than the native one. Thus, mechanistic and structural details of the native lipid-embedded open-activated and closed states of Kir channels are lacking. In addition, the crucial role of water in K⁺ conduction is still fiercely debated, with previous solid-state nuclear magnetic resonance (SSNMR) studies showing water is required for selectivity filter stability but may not be required for K⁺ conduction via the “knock-on” mechanism.^16,17

Figure 1.

Structural similarities of lipid-activated Kir channels to KirBac1.1. Surface charge maps of KirBac1.1 (a) (PDB ID: 2wll) and Kir2.2-PIP complex (b) (PDB ID: 3spi) are depicted with red denoting anionic residues and blue specifying cationic residues. The cationic sites thought to bind anionic lipids are boxed in magenta in each structure. These binding pockets are expanded in (c) and (d). Key cationic residues are labeled and shown in blue and gray. Cocrystallized lipid head groups are illustrated in each. The choline headgroup in 2wll is depicted in orange, gray, and blue in (c), and the PIP₂ analog is shown in red, green, and orange in (d). Distances are provided between R49/151/153 Cζ and the cocrystallized phosphocholine group. (e) Sequence alignment of KirBac1.1 and 3.1 along with various human Kir channels. Homologous regions are boxed in blue with residues in the same group shown in red. The highly conserved c-linker motif (R—P—K—K—R) is highlighted in purple.

Each membrane protein responds uniquely to changes in bilayer phase properties and composition.¹⁸ These include G protein-coupled receptors (GPCRs), ion channels, and transporters.¹⁹ Notably, a recent study showed the β2-adrenergic receptor (β2R) is allosterically modulated by lipid head groups, with effects on ligand binding affinity and receptor activity.²⁰ Additionally, a mechanosensitive ion channel from E. coli was shown to be modulated by membrane thickness and curvature.²¹ Multiple human Kir channels are activated solely by phosphoinositol-4,5-bisphosphate (PIP₂) or require PIP₂ in conjunction with other secondary activators.²² Other key activators include cholesterol (which can promote or suppress activity depending upon the channel), G proteins, ATP, ethanol, and bulk anionic lipids.^23–26 Conversely, KirBac1.1 is inactivated by PIP₂ and cholesterol but activated by bacterial anionic lipids like cardiolipin (CL) and 1-palmitoyl-2-oleoylglycero-3-phosphoglycerol (POPG). Both KirBac1.1 and KcsA are known to selectively recruit CL and POPG with evidence of direct binding in the case of KirBac1.1.^27,28 CL is a crucial membrane component found in mitochondrial and select bacterial membranes and is known to activate a variety of MPs such as cytochrome C.²⁹

In this work, we investigate the gating mechanism of KirBac1.1 through manipulation of three arginine residues that comprise the locus of lipid tethering required for channel activation (R49/151/153) (Figure 1c). These arginines were mutated to glutamine, rendering an inactive mutant for comparison to the stability mutant (SM; I131C)³⁰ of the wild-type (WT) protein. (The SM and WT were previously shown to have nearly identical activity and behavior, with the exception that the SM tetramer remains more stable throughout protein purification.³⁰) We utilize fluorescence-based K⁺-efflux assays of reconstituted SM–KirBac1.1 in various lipid mixtures. We demonstrate a functional dependence on CL and POPG in SM protein and confirm the loss of activity in our R to Q mutants. FRET measurements reveal changes in channel conformation as a function of anionic lipid presence consistent with previous studies of KirBac1.1 gating motions.¹⁴ Our FRET studies further determine changes between activated and inactivated states of the channel are absent in the (I131C, R49/R151/R153Q) protein, which we refer to as triple Q mutant or TQ. Clearly, multiple components and binding partners give rise to a complex system that is difficult to fully characterize through traditional methods. Thus, the finer mechanistic details of this functional relationship at an atomistic level are a subject of speculation.

Solid-state NMR (SSNMR) under magic-angle spinning (MAS) is a robust technique that can discern the dynamic structure underlying biological functions within a native-like membrane environment.³¹ Recent technical developments allow for solid and liquid state^32–36 NMR characterization of large MPs such as the homotetrameric 147 kDa KirBac1.1. Via SSNMR experiments on U-¹⁵N,¹³C-labeled SM–KirBac1.1 and the inactive TQ mutant, we observe conformers corresponding to the native active and inactive states of KirBac1.1. We provide evidence of phosphate binding to the aforementioned arginines by an ¹³C–³¹P rotational-echo double-resonance³⁷ (REDOR) dephasing experiment on a selectively labeled sample. Using 2D ¹³C–¹³C dipole-assisted rotational resonance³⁸ (DARR) and ¹⁵N–¹³C SPECIFIC CP³⁹ experiments, we observe a variety of chemical shift perturbations (CSPs) upon mutation of SM to TQ. We also employ water-edited experiments to determine changes in the water accessible surface correlated to activation/inactivation. These experiments highlighted residues in the conductance pathway and sites in the nonmembrane embedded regions, which become exposed or buried upon gating. In addition, our data reveals that water is proximal to the selectivity filter when the channel is in the open-activated conformation, but we cannot conclude water enters this region of the protein. Thus, our data inconclusively supports recent findings that water does not directly mediate the K⁺ conductance knock-on mechanism. In aggregate, our work definitively identifies and characterizes the lipid-activation site of KirBac1.1 and suggests that an inward twisting motion of the C-terminus upon activation is initiated by anionic lipid binding and stabilized by a network of intersubunit Coulombic forces.

MATERIALS AND METHODS

Molecular Biology.

Mutants were generated via site-directed mutagenesis PCR on the KirBac1.1(I131C)-6xHis gene with a thrombin cleavage site (LVPRGSRS) preceding the C-terminal Histag and a G2′ insertion. This gene construct was inserted into the pQE60 vector (a gift from Colin Nichols) as described previously.⁴⁰ This I131C stability mutant serves only to stabilize the tetramer and is not accessible to FRET labels.³⁰ We refer to this protein as SM–KirBac1.1 throughout the manuscript, though it is functionally identical with the WT protein. We also generated the following mutants: R49Q, R151/153Q, and R49/151/153Q. All mutants were verified via DNA sequencing.

Expression and Purification of KirBac1.1.

Natural abundance (NA) KirBac1.1(I131C) and all mutant constructs were expressed in M15pREP4 E. coli as described previously.^40,41 For expression of U-¹³C,¹⁵N KirBac1.1(I131C), cells were grown in U-¹³C,¹⁵N M9 minimal media and induced at an OD of 0.8 using 1 mM IPTG. For aromatic suppression, media was supplemented with 1 g/L N-phosphonomethyl glycine (glyphosate) and 50 mg/L NA [L-tryptophan, L-tyrosine, and L-phenylalanine. Cells were harvested by centrifugation and stored at −80 °C. The sample was kept at or near 4 °C throughout the entire purification. Frozen cells were lysed via homogenization at 10–15 kpsi, and membrane components were extracted with 30 mM decyl-β-D-maltopyranoside (DM) for 3–4 h on a rocker at 4 °C; the solution was centrifuged at 60 000 rpm in a type 70 Ti rotor (Beckman Coulter, Brea, CA) for 40 min at 4 °C to remove cell debris. The supernatant was filtered through a 0.22 μm PES bottle top filter and loaded onto a 5 mL HisTrap (GE Healthcare Life Sciences) column in wash buffer (50 mM Tris-Base, 150 mM KCl, 10 mM imidazole, 0.02% NaN₃, 2.5 mM DDM, pH 8.0). The column was then treated with 5 column volumes of wash buffer before eluting with 5 column volumes of elution buffer (50 mM Tris-Base, 150 mM KCl, 250 mM imidazole, 0.02% NN₃, 2.5 mM DDM, pH 8.0). Eluted protein was transferred into exchange buffer (50 mM Tris-Base, 150 mM KCl, 0.02% NN₃, 1 mM EDTA, 1–5 mM DDM, pH 8.0) via a HiPrep 26/10 desalting column (GE Healthcare Life Sciences). Desalted protein was concentrated to ~3 mg/mL using an Amicon Stirred Cell with an Ultracel 100 kDa Ultrafiltration Disc (Millipore) before loading on a HiLoad 16/600 Superdex 200 column (GE Healthcare Life Sciences) equilibrated in exchange buffer for the final purification step. Fractions containing KirBac1.1 tetramer were pooled, concentrated to ~0.5–1 mg/mL, and stored at 4 °C. The overall yield for all mutants was ~7 mg/L from minimal media, with the exception of the aromatic suppressed sample, which was ~5.6 mg/L.

Sample Reconstitution.

Lipids purchased from Avanti Polar Lipids (Alabaster, AL) dissolved in CHCl₃ were combined at desired ratios in 15 mL round-bottom flasks or vials and dried under a stream of N₂. Films were then resuspended once in n-pentane, dried again, and left to sit under vacuum for at least 4 h. Dried films were then resuspended at 5 mg/mL in K⁺ buffer (20 mM K-HEPES, 150 mM KCl, 1 mM EDTA, pH 7.4 w/ HCl) or NMR sample buffer (50 mM Tris-Base, 50 mM KCl, 1 mM EDTA, 0.02% NaN₃, pH 7.5 w/ HCl) plus 18.5 mM CHAPS by periodic water bath sonication to avoid overheating. Once suspensions appeared mostly clear and homogeneous, they were solubilized at 20 °C for at least 2 h. Purified KirBac1.1(I131C) in exchange buffer was combined at 0.5–1 mg/mL. For assay samples, a protein to lipid ratio (P/L) of 3:200 (w/w) was used and an equal amount of exchange buffer was added to protein free controls. For NMR samples, a protein/lipid ratio of 1:1 (w/w) was used. Samples were annealed for 20 min to 1 h at 20 °C before detergent was removed via sequential addition of BioBeads-SM2 (Bio-Rad, Hercules, CA) equilibrated in the corresponding resuspension buffer. BioBeads were added twice daily until a cloudy suspension indicative of liposomes was formed. For this set of experiments, a standard lipid mixture of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoe-thanolamine/1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPE/POPG; 3:1) (w/w) was used as it is a close mimic to native bacterial membranes.⁴² The POPC/POPG (3:2) (w/w) lipid mixture was used to reconstitute the samples used to generate our 3D assignment spectra of the activated conformation of the protein.

Efflux Assays.

Liposomes in K⁺ buffer were incubated in 100 μM 9-amino-6-chloro-2-methoxyacridine (ACMA) for approximately 15 min at 20 °C before each assay. Samples were diluted 1:20 with Na⁺ buffer (20 mM Na-HEPES, 150 mM NaCl, 1 mM EDTA, pH 7.4 w/ HCl) in a 96-well plate before transferring 20 μL to a 384-well plate with wells containing 20 μL of Na⁺ buffer. All experiments were conducted at 20 °C using a Biotek synergy NEO2 fluorescent plate reader (Biotek Instruments, Winooski, VT) with excitation and emission wavelengths of 410 and 480 nm, respectively. A baseline fluorescence (F_B) reading was taken every 5 s for 1 min before the addition of 16 μM (10 μL of 80 μM) carbonyl-cyanide m-chlorophenylhydrazone (CCCP) in Na⁺ buffer initiating flux (effective [ACMA] of 2 μM). Fluorescence was monitored for a 15 min reading every 5 s before the addition of 10–20 nM (0.5 μL of 1–2 ng/μL) valinomycin, giving a minimal fluorescence (F_V) reading every 5 s for 1 min. Normalized fluorescence (F_N) was generated using

$$F_{\text{N}}=(F-F_{\text{V}})/(F_{\text{B}}-F_{\text{V}})$$ (1)

FRET Experiments.

Mutants for FRET experiments were made via site-directed mutagenesis PCR using the KirBac1.1 G249C construct along with the stabilizing I131C mutation as a template. Protein was purified, labeled, and reconstituted as described previously⁴⁰ with minor alterations. We attached the Alexa-Fluor-546 C5 maleimide/DABCYL and C2 maleimide (A/D) fluorescence donor and acceptor to the G249C mutant of each protein (the C131 site is within the transmembrane region, facing away from the central pore, and is thus inaccessible to either dye). Purified protein was labeled after desalting at a ratio of 1:2.5:10 KirBac1.1/A/D to obtain equal labeling of donor and acceptor. Labeled protein was then loaded directly onto a HiLoad 16/600 Superdex 200 column (GE Healthcare Life Sciences). Additionally, BioBeads-SM2 were used for detergent removal during reconstitution. Initial fluorescence was monitored before the addition of proteinase K (0.16 units) and was monitored afterward until a signal plateau was reached. FRET efficiency (E_app) was calculated as

$$E_{\text{app}}=(F_{\text{max}}-F_0)/F_{\text{max}}$$ (2)

where F_max is maximal fluorescence and F₀ is the initial values before addition of proteinase K.

NMR Sample Preparation.

Proteoliposomes containing U-¹⁵N,¹³C KirBac1.1 were transferred into 26 mL ultracentrifuge tubes and spun at 70 000 rpm and 4 °C for 1.5 h in a type 70 Ti rotor discarding supernatant. Pellets were resuspended in 1–1.5 mL of NMR sample buffer each time spinning at 12 700 rpm and 4 °C for 1 min in a microcentrifuge swing bucket rotor discarding supernatant. Samples were repeatedly freeze–thawed 6–8 times with liquid N₂ and a 37 °C water bath and spun each time at 12 700 rpm and 4 °C for 5–10 min. Sufficiently packed samples were packed into an NMR rotor of desired size spinning at 3000–5000 rpm and 4 °C for 5–10 min. The aromatic suppression sample was packed into a 1.6 mm PENCIL rotor (Agilent Technologies). For all other samples, 3.2 mm PENCIL rotors (Revolution NMR, Ft. Collins, CO) were used. Fully prepared samples were stored at 4 °C when not in use.

SSNMR Spectra.

All chemical shift assignment and H₂O-edited SSNMR spectra were acquired on a 600 MHz Agilent DD2 spectrometer (Agilent Technologies, Santa Clara, CA and Loveland, CO) equipped with an HCN BALUN 3.2 mm probe. ¹³C was referenced indirectly to DSS using the downfield adamantane peak at 40.48 ppm.⁴³ All spectra of samples packed into 3.2 mm rotors were acquired with 13 333 Hz magic-angle spinning (MAS) and 80 kHz of SPINAL decoupling on the ¹H channel for a total (direct and indirect dimensions) of 25–40 ms. The recycle delay was set to 2 s for all experiments. 90° pulses for ¹H, ¹³C, and ¹⁵N were 2.1, 3.1, and 5 μs, respectively. rINEPT pulse delays were 1.25 ms. ¹H to ¹³C cross-polarization was facilitated through a 800 μs contact time with a tangent ramped CP spin lock power of ~60 kHz on ¹H and ~73 kHz on ¹³C. ¹H to ¹⁵N cross-polarization was facilitated through a 600 μs contact time with a tangent ramped CP spin lock power of 55 kHz on ¹H and 42 kHz on ¹⁵N. All 2D DARR spectra were acquired with 12 ms of DARR mixing. All H₂O-edited ¹H–¹³C and ¹³C–¹³C spectra were acquired using a T₂ filter of 3 ms, eliminating all protein polarization and nearly all flexible lipid signal. No flexible lipid polarization was transferred to the protein (Figure S1). The REDOR spectrum of the ¹⁵N,¹³C labeled sample with suppressed aromatic ¹³C signals was acquired using an Agilent FASTMAS 1.6 mm probe in an ¹H–³¹P–¹³C configuration with 25 kHz MAS and a sample temperature of ~–10 °C. ¹H, ³¹P, and ¹³C pulse widths were 1.6, 2.4, and 1.5 μs, respectively. 26 ms of REDOR dephasing was used with 110 kHz of SPINAL ¹H decoupling. Pulse sequences are depicted in Figure S2.

RESULTS AND DISCUSSION

Sequence Analysis.

As stated in the Introduction, several studies have identified key arginine residues required for channel activation. The Nichols lab found that tethering of R49 in the slide helix to the lipid bilayer is necessary for channel activation,¹³ and Clarke et al. observed choline phosphates bound to R151 and R153 but in a closed conformation.⁷ On the basis of this knowledge, we hypothesized that KirBac1.1 is activated via a similar mechanism to eukaryotic Kir channels but with a smaller binding pocket to accommodate for anionic lipid headgroups present in prokaryotic membranes such as phosphatidylglycerol (PG) and cardiolipin (CL). As depicted in Figure 1, we selected R49, R151, and R153 as likely anionic lipid binding partners. These residues align with purported PIP₂ binding sites found in Kir2.2 and Kir3.1 structures (Figure 1e). To confirm activation is a charge-dependent process, we mutated each arginine to glutamine as it is the nearest polar uncharged neighbor to arginine. We produced two mutant permutations, R151/R153Q or “double Q” (DQ), and the TQ mutant in the I131C SM construct of KirBac1.1. The expression and yield of this mutant protein was identical to the SM within quantification error (7 mg/L cell culture).

K⁺ Efflux Assays.

We quantified channel function via fluorescence-quenching efflux assays (Figure 2a,b) as described previously.^40,44 We specifically targeted differences in channel gating promoted by 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) and CL (E. coli extract) head groups in the SM, DQ, and TQ proteins. Our assay data show SM–KirBac1.1 is activated by both PG and CL head groups (Figure 2a,b) but with greater overall activity in the presence of CL. Activation by CL does not increase the rate of K⁺ flux based on a monoexponential decay equation:²⁴

$$(F_{\text{t}}=F_0\times e^{-t/\tau }+C)$$ (3)

but promotes greater overall channel conductance. Thus, cardiolipin stabilizes the active conformation and shifts the equilibrium between the active and inactive states of the channel. However, K⁺ flux is limited by the rate of diffusion via the knock-on mechanism. Flux assays of mutant proteins confirm all R to Q mutations reduce channel activity. DQ exhibited 60–70% reduced activity in the presence of PG and CL. The TQ mutant was completely inactive in the presence of PG, with less than 15% of basal SM activity upon the introduction of CL. This confirms all three arginines play a crucial role in activation by anionic lipids.

Figure 2.

K⁺ efflux assays and FRET analysis of SM and mutant channels. (a) K⁺ efflux assays of SM–KirBac1.1 and two inactivating mutants R151/153Q (DQ) and TQ reconstituted into POPE/POPG (75:25) (w/w). (b) K⁺ efflux assays of the same proteins, performed in POPE/POPG/CL (75:24:1) (w/w) bilayers. The protein free (PF) negative controls for each lipid condition are depicted in black. Decay rates (±RMS) (a, inset) are negligible for both mutants with the exception of DQ in 1% CL. (c) FRET buildup curves after the addition of proteinase K. Curves are for both DQ and TQ mutants of the SM–G249C (GC) FRET construct in POPE/POPG bilayers with or without 5% (w/w) PIP₂. (d) FRET efficiency values with significant reduction for GC in the presence of PIP₂, and negligible differences for each mutant in the presence of PIP₂.

FRET.

As in our previous study, we used FRET distances to confirm gating motions in the SM protein and the absence of these motions in the inactivated mutant. As shown in Figure 2c,d, the FRET efficiency between activated and PIP₂-inactivated SM protein is consistent with our earlier study⁴⁰ but in PE/PG bilayers compared to PC/PG. The mutant proteins, however, showed little to no domain motions as a function of the lipid environment, including in the presence of PIP₂. This confirms that the primary means of channel response to anionic lipids is removed in the mutant proteins and further suggests that the PIP₂ induced inactive state is different from the native closed state, which comes about in the absence of anionic lipid binding. Indeed, FRET distances for all mutants and SM protein in pure POPC (used here as a control instead of POPE due to difficulties in reconstitution into pure POPE bilayers) indicate that the intersubunit distances measured at G249 in the “native” closed state are slightly closer compared to the PIP₂-inactivated state. This differs from our previous FRET study along with that of others,^14,15 where we looked only at the PIP₂-induced closed state rather than the native closed state present in absence of anionic lipids. FRET further establishes that the TQ mutant adopts a similar confirmation to the native closed state. Thus, we have identified a distinct closed-inactivated state that is not identical with the PIP₂ closed state investigated in previous FRET studies.

Lipid Binding Arginines.

¹⁵N–¹³C correlation spectra of U-¹⁵N,¹³C-KirBac1.1 in an activating 75:20:5 POPE/POPG/CL (PE/PG/CL) lipid environment identify distinct arginine Cζ–Nη and Cζ–Nε outliers (Figure 3a, black). These Cζ chemical shift outliers are upfield from the bulk population, indicating their local environment is distinct from other arginine residues. These peaks disappear in spectra of U-¹⁵N,¹³C-TQ–KirBac1.1 (Figure 3a, red) in the same lipid environment. This both confirms the TQ mutant abolishes key arginine interactions and indicates these outliers are correlated with an open-activated state of the channel. These peaks also disappear in the spectra of SM–KirBac1.1 in a purely zwitterionic lipid composition (Figure S5a,b).

Figure 3.

Spectral evidence for arginine–lipid interactions. (a) 2D NcoCX spectra of SM KirBac1.1 in black and TQ overlaid in red. Both samples are reconstituted into a 75:20:5 POPE/POPG/CL (PE/PG/CL) lipid mixture. Boxed in blue are arginine Cζ outliers at ~156.9 ppm, which are observed in SM but absent in TQ. (b) ¹³C–³¹P REDOR dephasing experiment for SM targeting aforementioned Cζ outliers (gray box) with an inset of the proposed interaction with lipid phosphate with the control (S₀) in black and dephased spectra (S) in red. Percentage (S/S₀) of dephased indicated in red for the “bulk” and outlier arginine peaks. Residuals between S₀ and S in blue. (c–e) NA-¹³C-lipid rINEPT temperature series of SM (c), TQ (d), and lipids only (e) with first derivative (ξ) insets of the respective melting curves. Error was determined via RMS of the spectra (~3%) and is within the size of the symbols denoting each point.

We confirmed these arginine outliers are bound to lipid phosphate head groups by performing ¹³C–³¹P rotational echo double resonance (REDOR) dephasing.⁴⁵ To suppress the tyrosine Cζ signals, which are often degenerate with arginine Cζ chemical shifts, we grew a sample in U-¹⁵N,¹³C minimal media, supplemented with glyphosate and natural abundance tyrosine, phenylalanine, and tryptophan.⁴⁶ This produced an ¹⁵N and ¹³C enriched sample with minimal aromatic ¹³C signals. REDOR data was acquired at ~−10 °C. At this temperature, the Cζ outliers were better resolved from the bulk arginine Cζ peak. Unlike previous studies, lower sample temperatures were not required to obtain lipid–protein dephasing, indicating the tight binding events associated with channel gating.⁴⁷ As shown in Figure 3b, 26 ms of REDOR dephasing reduces the outlying arginine peak intensity by ~40%, indicative of an average arginine–phosphate contact of between 4.6 and 8 Å for all three arginines. This agrees with the bound phosphate to arginine Cζ distances of 7.4 to 4.9 Å from the crystal structure (Figure 1c); however, the structure we observe is open-activated rather than closed. It is difficult to further quantify these ¹³C–³¹P distances as it is unknown how many phosphates are close to each Cζ site (likely, 1–3 lipid head groups may be within 8 Å of each Cζ). Thus, the full REDOR curve is an admixture of dephasing curves for three different Cζ resonances from an unknown number of ³¹P resonances. However, the recorded dephasing is consistent with REDOR curves simulated in SIMPSON using these distances within experimental error (Figure S4). The observed T₂ relaxation time is 4.2 ms. This short transverse relaxation time severely diminished the signal-to-noise of the resultant spectra (where the S₀ signal is approximately 0.2% of the initial signal). However, this bracketed distance range is consistent with previously reported arginine–phosphate salt bridges.^48–50

KirBac1.1 TQ Mutant Alters Lipid Phase Properties.

We previously showed that KirBac1.1 can alter the phase properties of bacterial lipids.²⁷ In that work, we hypothesized that this noncolligative property of KirBac1.1 was due, in part, to the binding of anionic lipids to the channel. Thus, we assessed differences in the phase properties of the bilayer surrounding SM–KirBac1.1 and the TQ mutant versus the protein-free membrane (Figure 3c–e). We performed a temperature series of refocused insensitive nuclei enhanced by polarization transfer (rINEPT) experiments on natural abundance ¹³C lipid resonances in both proteoliposome samples and a protein-free sample (PE/PG/CL). These spectra report upon the phase properties of the bilayer in two ways. First, rINEPT acts as a dynamic filter as polarization is transferred from ¹H to ¹³C through J-couplings in the absence of high-powered ¹H decoupling. In SSNMR, the strongest ¹³C T₂ mechanism is ¹H–¹³C dipole–dipole couplings, which can be attenuated by ¹H radio frequency (rf) decoupling, very fast magic-angle spinning (MAS), or intrinsic molecular motions. Under our experimental conditions, rINEPT selects for fast-moving molecules, which possess naturally long T₂ relaxation. Thus, rINEPT effectively monitors changes in thermotropic lipid phase transitions as motion is directly correlated to lipid phase. Second, the ¹³C chemical shifts of lipid acyl chains depend upon rotameric conformation.⁵¹ For –¹³CH₂– resonances, the dominant populations are the all-trans (AT) and the trans–gauche (TG) conformations, which correspond to peaks at 35 and 33 ppm, respectively (Figure S5). In samples containing only one type of lipid, the AT conformer is observed in the gel phase (which is invisible to rINEPT) and transitions to a TG peak after melting.²⁷ However, in lipid mixtures, which may also include sterols, hopanoids, and proteins, significant AT conformers are present in rINEPT spectra above the thermotropic phase transition. These mobile AT conformers are indicative of liquid-ordered (L_o) lipid phases.^52,53

Thus, the evolution of assigned peak positions and intensities in rINEPT spectra as a function of temperature maps the lipid phase. In Figure 3c–e, the integrated intensities of peaks corresponding to unsaturated, AT, TG, and methyl resonances are plotted from 253 to 313 K. These rINEPT temperature series for each sample have distinct features. First, in the absence of protein (Figure 3e), this lipid mixture exhibits a narrow thermotropic phase transition consistent with a heterogeneous but purely lipid bilayer. This titration is confirmed by taking the first derivative of this rINEPT curve (Figure 3c, inset) where a single 5°-wide point of inflection is observed. This thermotropic phase transition begins at ~300–305 K, and the rINEPT signal plateaus at ~284 K. However, the introduction of SM–KirBac1.1 into this lipid mixture lowers the initial melting temperature to 258 K, and the rINEPT signal slowly increases over a range of 50° (Figure 3c). This rINEPT curve corresponds to a very broad thermotropic transition similar to the one we observed in ¹³C-labeled bacterial lipids in the presence of KirBac1.1. This is likely because there is phase separation between lipids within this bilayer defined by their proximity to the protein. These phases possibly range from gel to L_o and L_d phases. The first derivative of this curve, ε (Figure 3e, inset), reveals two broad points of inflection. The lipid phase properties in the presence of TQ–KirBac1.1 are intermediate to the protein-free and SM–KirBac1.1 extremes (Figure 3c,e). Here, the lipids begin to melt at 268 K, and the rINEPT curve exhibits two transitions that are discontinuous, unlike the SM–KirBac1.1 curve. The first derivative of this warming curve reveals two inflection points, like the SM–KirBac1.1 sample, but the second main transition is relatively sharp, similar to the sample without protein. In addition, the relative populations of TG and AT are different in each sample. In the protein-free liposomes, the ratio of TG/AT is 6.7:1 at 315 K. This ratio in SM–KirBac1.1 containing liposomes is ~3:1; ~25% of the –¹³CH₂– resonances assume an AT chemical shift. In the TQ containing sample, the TG content is 5.5 times AT at 315 K. As stated above, AT conformers visible to rINEPT are indicative of an L_o phase, and this content is increased in the presence of SM–KirBac1.1. In addition, lipid headgroup resonances corresponding to anionic PG and CL lipids are less visible in the SM–KirBac1.1 sample, one of the few lipid signals for which this is the case (blue region of Figure S1). This indicates a significant portion of these head groups are held in place and thus invisible to rINEPT. These observations indicate the three Arg residues in the lipid binding site are substantially responsible for many of the properties imparted onto the lipid bilayer by KirBac1.1. The difference in lipid ordering is further represented in ¹H T₂ edited ¹H–¹³C correlation spectra presented in Figure S1 and discussed below.

SSNMR.

In the 2D ¹³C–¹³C correlation spectra of each sample (Figure S6), we observed a number of resolved peaks that either shifted or dramatically changed in signal intensity (Figure 4). These included T120 in the selectivity filter loop (SFL), A154 in the c-linker, and I138 along TM2. I138 makes a critical steric contact in the open-activated state of the channel, locking the side chain conformation. In 2D ¹³C–¹³C correlation spectra of SM–KirBac1.1, polarization builds up from the Cα down to the Cδ2 in 12 ms of mixing, while there is little polarization transfer in the TQ–KirBac1.1 sample. T120 has been tied to differing states of the selectivity filter, and A154 changes conformation upon lipid activation. It is located directly beneath both the activation gate and the lipid binding pocket.

Figure 4.

Notable differences in standard ¹³C–¹³C DARR spectra for the U-¹³C,¹⁵N TQ(red) and SM (black) KirBac1.1 samples in the standard PE/PG/CL mixture. Residues A18/154 (left) undergo CSPs in TQ; T120 (middle) experiences a minor CSP with a stark reduction in polarization, and in I138 (right), the Cα-Cδ is only present in the SM. Full spectra are provided in Figure S6.

H₂O-Edited SSNMR.

While some conformational changes could be observed in 2D spectra, the dramatic global transitions related to gating were not apparent. This is not surprising, as we previously determined that the secondary structures of key regions of the protein are preserved in both activated and inactive states of the protein.⁴⁰ To structurally characterize the domain motions required for channel gating, we acquired water-edited SSNMR spectra. Water-edited SSNMR spectroscopy exploits short protein ¹H transverse relaxation times (T₂) present in solids. Through implementation of a T₂ filter (3 ms), we are able to select for only water polarization before mixing (Figure S2a–c). This allowed us to observe only the water exposed or water proximal regions of the channel and identify changes in solvent exposure between the open and closed states. Polarization is transferred between water and protein via three key mechanisms. Exchange, spin diffusion, and the nuclear Overhauser effect (NOE). Near room temperature and under moderate spinning, the dominate transfer pathways should be exchange and spin diffusion. We acquired ³H_water–¹H_protein polarization transfer buildup curves, H₂O-edited 2D ¹³C–¹³C DARR, and ¹H₂O–¹³C heteronuclear correlation (HETCOR) spectra on SM and TQ mutant channels in an activating PE/PG/CL lipid mixture. Because only small portions of the protein are close enough to water to be observed with 5 ms of ¹H₂O–¹H_protein mixing, though observed peaks are a mixture of ¹H₂O–¹H_protein and ¹H₂O–¹H_protein–¹H_protein relayed transfer as discussed below, there is far less spectral crowding in these spectra and many peaks are cleanly resolved. Thus, many changes were observed between SM and TQ samples. Most strikingly, in both the H₂O-edited ¹³C–¹³C DARR and ¹H₂O–¹³C HETCOR experiments, the site-resolved polarization (and thus water proximity) of the cytoplasmic domain, most notably for charged side chains (Arg, Lys, Asp, Glu), was dramatically greater in TQ compared to SM (Figures 5, S7, and S8). In addition, the full HETCOR spectra (Figure S1) reveal no discernible lipid ¹H polarization survives the 3 ms T₂ filter in the SM sample, while in the TQ sample a small portion of ¹H to NA-lipid ¹³C polarization transfer is observed. No lipid to protein signals are observed at any ¹H to protein spin diffusion mixing time in either sample. This further illustrates how KirBac1.1 orders its surrounding lipids. The lipids adjacent to the protein are too ordered to survive the applied T₂ filter and effectively shield the protein from ¹H polarization originating from the lipid bilayer. In the SM sample, this ordering is near complete, while in the TQ sample a small fraction of the lipid ¹H signals survive the T₂ filter but with no net polarization transferring to the protein.

Figure 5.

Spectral evidence of the increased water accessibility of the TQ mutant. (a–c) Water-edited spectra of SM and TQ samples in black and red, respectively. Charged residue populations Glu/Asp Cδ/Cγ that we have assigned (a) (Figure S1) and Arg Cζ populations that appear much stronger in TQ over SM in both the ¹H–¹³C HETCOR spectra (a, b) and Glu Cδ. (d) Close-up of CTD charged residues in a SWISS-MODEL of the 1p7b structure with missing residues 196–205 and 295–300 modeled in. (e, f) Buildup curves for both SM and TQ in black and red, respectively. The greater overall exposed backbone and charged residue surface were observed for TQ; the carboxyl groups in particular (f) show a dramatic increase. Lines are best-fits based upon eq 11. Error bars are determined via RMS of spectral noise.

The H₂O to protein ¹H–¹H spin diffusion rate and differential polarization saturation points show the water accessible surface area of KirBac1.1 decreases upon activation. First, we analyzed two-dimensional ¹H_water–¹H_protein spin diffusion buildup experiments, where the indirectly acquired dimension was ¹H–¹H spin diffusion time (¹H_mix). These ¹H_water–¹H_protein spin diffusion buildup experiments monitor the water accessible surface area of transmembrane proteins. Several groups have shown that these experiments effectively monitor the water or lipid accessibility of complex biomolecular polymers with particular emphasis on the water accessibility of MPs and the active sites of microcrystalline enzymes.⁵⁴ As discussed by Schmidt-Rohr and Spiess,^{55 1}H–¹H spin diffusion experiments can monitor domain changes of biomolecules. The Baldus lab first showed the viability of this method to quantify large-scale changes in activated K⁺ channels.⁵⁶ Further, there is a derived generalized equation that can account for magnetization transfer behavior using four variables:

$$\sqrt{t_{\text{m}}^{\text{s}}}=\sqrt{\frac{D_{\text{eff}}}{\pi }}\frac{V_{\text{P}}}{S_{\text{WP}}}$$ (4)

where $t_{\text{m}}^{\text{s}}$ is the time of mixing until saturation, D_eff is the effective diffusion parameter, V_P is the volume of the protein, and S_WP is the surface area of the water–protein interface. We made several parametric assumptions to analyze the changes in H₂O accessible surface area using eq 1. (1) We assume the water and protein constitute a two-phase system where there is no chemical relay between the two phases. This assumption is ensured by optimization of the ¹H T₂ filter such that all meaningful ¹H polarization begins in the H₂O bath for most sites on the protein. (2) We assume ¹H–¹H spin diffusion is uniform across all water chemical populations as t_m → ∞. This is justified as, beyond the three mutated sites, the samples are chemically identical. (3) We assume the fold (and thus volume and overall surface area) of the individual protein domains is the same between each sample. As stated above, we previously showed that channel gating is brought about by multiple small conformational changes in the extracellular loop and along TM2 that leave the overall secondary structure of the channel intact.⁴⁰ As stated further below, it also appears that motions in the cytoplasmic regions of the protein are dependent upon changes at a few key points, while the structures of each C-terminal paddle comprising the Kir domain largely retain their fold. The increase in the surface area of the inactive TQ mutant relative to the SM can be determined by

$$\frac{S_{\text{WP}}^{\text{TQ}}}{S_{\text{WP}}^{\text{SM}}}=\frac{{\sum }_i^n{I}_n^{\text{TQ}}}{{\sum }_i^n{I}_n^{\text{SM}}}$$ (5)

where I_n is the integrated intensity of the selected peaks, and this ratio is directly proportional to changes in the water adjacent surface area between the SM and the TQ mutant. Summations between all selected spin systems converge upon an increase of the water adjacent surface area of the TQ sample by roughly 46% compared to the SM protein. We hypothesize this is due in large part to a rearrangement of the orientation and interdomain contacts of the C-terminal “paddles”, which decrease the solvent-exposed surface in the gated conformation of the C-terminus (Figure 5d). This further explains why the TQ mutant exhibits faster mixing saturation time. As defined in eq 1, if surface area increases and all else remains equal, polarization transfer saturation should occur at a faster rate.

Clear trends are observed in the arrayed ¹H–¹H spin diffusion data shown in Figure 5. First, there is a gradual decrease in the intensity as ¹H–¹H spin diffusion time increases past the maximum buildup. This is due to the T₁ relaxation effects at such long mixing times. Using this previously elucidated information, we observed unexpectedly that the T₁ values for these bracketed spin systems are larger in the TQ mutant compared to the SM (Table 1). We presume the R₁ relaxation rates are larger in the SM protein because of fast nanoscale time scale CTD motions present in the gated conformation of the sample and the differing H₂O accessibility of each sample. Thus, to further quantify the increased water accessibility of the TQ mutant, we fit the rates of ¹H_water–¹H_protein spin diffusion buildup to a full analytical solution adapted from Najbauer and co-workers⁵⁷ but with additional parameters to describe intraprotein cross relaxation. We begin with a full differential rate law describing proposed ¹H–¹H cross relaxational rates as

$$\frac{\partial M_{\text{p}}(t_{\text{m}})}{\partial t}=-(R_{1,\text{p}}+R_{\text{wp}}+R_{\text{p}{\text{p}}^{\prime }})M_{\text{p}}(0)+(R_{\text{wp}})M_{\text{w}}(0)+(R_{\text{p}{\text{p}}^{\prime }})M_{{\text{p}}^{\prime }}(0)$$ (6)

$$\frac{\partial M_{\text{w}}(t_{\text{m}})}{\partial t}=-(R_{1,\text{w}}-\frac{1}{N}{R}_{\text{wp}})M_{\text{w}}(0)+\frac{1}{N}(R_{\text{wp}})M_{\text{p}}(0)$$ (7)

where subscript p specifies protein, subscript w specifies water, R defines a rate of longitudinal cross relaxation for specified species, and M(t) defines the magnetization on the specified chemical species at mixing time t. M_w(0) and M_p(0) represent the initial magnetization on water or protein after the T₂ filter. Equation 7 accounts for the loss of polarization in water species a result of contact with the protein. N denotes the number of water molecules in contact with the protein. Due to rapid ¹H–¹H mixing between contact water ¹Hs to the bulk, noninteracting water ¹Hs, the differential rate law of eq 7 simplifies to

$$\frac{\partial M_{\text{w}}(t_{\text{m}})}{\partial t}=-(R_{1,\text{weff}})M_{\text{w}}(0)$$ (8)

Table 1.

Fit Relative Populations and Relaxation Rates Based upon Fitting ¹H_water–¹H_protein Polarization Transfer Buildup Curves to Eq 11^a

	stability mutant (I131C)							triple Q
	Mw(0)	Rwp	R1w	R1p	Rpp	Mp(0)	RMS	Mw(0)	Rwp	R1w	R1p	Rpp	Mp(0)	RMS
amide	1.64	2.74	1.33	47.7	2.72	0.00	0.2	1.84	2.85	1.63	41.6	2.85	0.00	0.2
Arg-15Nε	1.52	2.73	0.86	53.7	2.73	0.00	0.7	1.84	3.08	1.37	50.1	3.18	0.00	0.6
Arg-15Nη	1.75	2.81	1.32	46.9	2.81	0.00	5.8	2.02	3.30	1.72	54.0	3.29	0.00	0.3
Lys-15Nζ	1.18	2.61	1.90	68.8	2.73	0.08	0.2	1.83	2.37	1.25	40.1	2.20	0.18	0.1
-13CO	2.04	2.61	1.20	49.8	2.61	0.00	0.2	2.45	2.73	1.63	43.1	2.80	0.00	0.2
-13COO−	1.15	3.36	1.43	27.7	3.40	0.00	7.6	4.07	2.62	2.06	32.6	2.65	0.00	0.9
CHa	1.70	2.84	1.22	47.9	2.84	0.00	0.2	1.93	2.62	1.60	41.1	2.88	0.00	0.2
CHx	1.97	2.45	1.31	44.4	2.45	0.00	0.2	2.71	2.62	1.57	44.4	2.26	0.00	0.3

^aAll rates are in units of s⁻¹.

Indeed, R₁ measurements for water ¹Hs show that the water population follows a single effective exponential decay rate. An analytical solution for the resulting cross relaxation with eqs 6 and 7 may be derived, integrating results in the following form with respect to the relaxation evolution on protein and water, respectively:

$$M_{\text{p}}(t_{\text{m}})=\left(\frac{-R_{\text{wp}}{R}_{\text{p}{\text{p}}^{\prime }}}{R_{1\text{p}}+R_{\text{wp}}+R_{\text{p}{\text{p}}^{\prime }}}\right)(e^{-(R_{1\text{p}}+R_{\text{wp}}+R_{\text{p}{\text{p}}^{\prime }})t_{\text{m}}})$$ (9)

M_p(0) disappears from this expression as it is removed by the T₂ filter. Thus, for the water population, the following rate equation applies:

$$M_{\text{w}}(t_{\text{m}})=M_{\text{w}}\left(\frac{-1}{R_{1\text{w}}}\right)(e^{-R_{1\text{w}}{t}_{\text{m}}})$$ (10)

When the relaxational rate is combined, eqs 9 and 10 produce the following equation:

$$M_{\text{p}}(t_{\text{m}})=M_{\text{w}}\left(\frac{R_{\text{wp}}{R}_{\text{p}{\text{p}}^{\prime }}}{R_{1\text{p}}+R_{\text{wp}}+R_{\text{p}{\text{p}}^{\prime }}-R_{1\text{w}}}\right)(e^{-R_{1\text{w}}{t}_{\text{m}}}-e^{-(R_{1\text{p}}+R_{\text{wp}}+R_{\text{p}{\text{p}}^{\prime }})t_{\text{m}}})$$ (11)

We empirically found that R_wp ≈ R_pp′ = R_p, allowing for the approximation:

$$M_{\text{p}}(t_{\text{m}})=M_{\text{w}}\left(\frac{2R_{\text{p}}}{R_{1\text{p}}+2R_{\text{p}}-R_{1\text{w}}}\right)(e^{-R_{1\text{w}}{t}_{\text{m}}}-e^{-(R_{1\text{p}}+2R_{\text{p}})t_{\text{m}}})$$ (12)

A full rate matrix would fully quantify all of the relaxation terms defining the ¹H_water–¹H_protein polarization transfer; we find eq 11 provided excellent fits to the experimental data, confirming approximate cross relaxation terms can be effectively derived from this equation. All characteristics of the experimental data effectively model the spin diffusion curves (Figures 5e,f, S7, and S8), reflecting its robustness. The close relation between R_wp and R_pp′ suggests that ¹H_water to ¹H_protein transfer may be the limiting rate in this large, fully protonated system. However, it also suggests these transfer mechanisms are highly coupled and would share a similar D_eff (eq 4). This suggests a steady state exists where water penetration is determined by fast exchange proximal to water sites, i.e., fast exchanging OH, NH₃⁺, and backbone amides.⁵⁸ Cross relaxation within the protein remains but dies off quickly with the high R_1p, resulting in a site-specific readout on what is proximal to water interacting directly with protein at shorter spin diffusion times of approximately less than 10 ms.^59,60 Thus, polarization transfer into the interior of the protein does not occur detectably unless mixing times greater than 5–10 ms are used, indicating, depending upon the chemical group in question and the surface area of the protein, 5–10 ms of mixing is required to fully polarize the exterior of KirBac1.1. However, this could be unique to this system, as KirBac1.1’s V_p (eq 4) is relatively large compared to many proteins previously studied using H₂O-edited SSNMR.^{56–58,61–64} The importance of exchange as a function of water accessibility is clearly observed in the profound differences in the Lys, Arg, and amide backbone buildup curves (Figures S8 and 5e) between the SM and the TQ mutant. The dramatic overall increase in both the rate of polarization transfer and maximal signal intensity indicates these sites are more water adjacent in the TQ sample. Likewise, Glu and Asp side chains also exhibit increased rates of polarization transfer from water and greater overall maximal signal, likely because of transient interactions with polarized H₂O. The greater water accessibility is further highlighted by the nonzero M_p(0) observed for lysine side chains in the TQ mutant. Thus, the rate of exchange for these cationic side chains is on a much faster time scale in the TQ mutant.

Most dramatic accessibility changes were observed in C-terminal side chain ¹³C and ¹⁵N resonances of charged residues, including glutamate Cδ5, aspartate Cγ, and arginine Cζ, Nη, and Nε (Figure 5a–c). The TQ mutant displayed greater H₂O accessibility compared to SM protein for these sites in both ¹H_water^–1H_protein–¹³C and ¹H_water–¹H_protein–¹⁵N polarization buildup experiments (Figure 6e,f). In the TQ mutant buildup curves, we also observed an overall increase in backbone solvent accessibility. As stated above, we believe this is attributable to waters’ increased access to exchangeable sites in the backbone and more visibly on side chains (Figure S7). More importantly, as confirmed below, the dramatic increase in water accessible ionic residues was found almost exclusively in the CTD (Figure 5). This is interesting as several studies suggest similar conformational states may exist.^65,66 Our data suggest that each subunit of the CTD rotates synchronously with the opening of the activation gate of KirBac1.1. This open-activated conformation is stabilized by multiple intersubunit salt bridges between these ionized side chains on neighboring subunits. These new salt bridges reduce both proximity to water and the rate of chemical exchange between Lys and Arg residues and polarized water. These new contacts are likely an intersubunit based upon the prevalence of charged side chains on the external regions of each subunit. When the channel is closed, these ionic locks do not exist and the CTD interfaces are subsequently splayed out, exposing these ionic residues. This splayed CTD conformation explains the dramatic increase in overall solvent exposed surface in the closed state of the channel compared to the activated state, which is somewhat puckered with a more open activation gate. This may be consistent with previous FRET and crystallography studies.^7,15 We compare our results with calculated solvent accessible surfaces of different crystal structures in Figure S9.

Figure 6.

Site specific solvent accessibility of KirBac1.1. (a) Water accessibility map of KirBac1.1 generated from assignments of water-edited 2D DARR experiments with 5 ms of ¹H–¹H mixing (b–e) of both SM (black) and TQ (red) samples in our standard PE/PG/CL mixture (Figures S16 and S17). Sites and assignments are color coded, corresponding to those that appear more strongly in SM (blue) or more strongly in TQ (red) or are CSPs from SM to TQ (orange). (b, c) Distinct differences observed for select proline and isoleucine side chain correlations. (d) Numerous charged residues found primarily in the C-terminus were more water exposed in TQ. (e) Glycine C′-Cα correlations for selectivity filter and TM2 residues. Green arrows indicate CSPs of G137C′-Cα, G114C′-Cα, and I138Cγ1-Cδ. (f) Residues not represented in the 1p7b crystal structure with those more exposed in TQ or SM in red and blue, respectively. Residues 310 to 333 do not appear in the crystal structure and remain uncharacterized.

Site-Specific Solvent Accessibility.

To better ascertain the site-specific solvent accessibility of the protein, we completed the chemical shift assignments for KirBac1.1 from residues 1 to 301 in both the open and closed states of the channel, corresponding to 3:2 PC/PG and PC bilayers, respectively. Thus, in addition to the assignments previously reported on the transmembrane region of KirBac1.1,⁴⁰ we have assigned the slide helix, N-terminus, c-linker, C-terminus, and adjacent regions (Figures S10–S15). Residues 4–35 (Figures S14 and S15), 199–205 (Figures S12 and S13), and 290–295 are not present in the crystal structures of KirBac1.1 (PDB ID: 2wll and 1p7b) as such; this study represents their first atomistic structural characterization. The extensive assignments of KirBac1.1 in both environments allows us to quantify differences in the water accessible surface between the activated and inactivated states of KirBac1.1. Thus, we site-specifically assigned most resolved peaks in H₂O-edited ¹³C–¹³C DARR spectra of both SM and TQ (Figures 6, S16, and S17, and Tables S1–S4). There is a large degree of asymmetry in these experiments, which reflects the degree of solvent exposure at the site of initial polarization and to a lesser extent CP efficiency. For example, cross peaks for the entire T120 spin system are visible in the SM, while for TQ, Cα-Cβ and Cβ-Cα are absent and Cβ-Cγ is diminished. This can be interpreted as the entire residue including backbone being exposed in the SM, while for TQ, only the Cγ and to a much lesser extent Cβ are water exposed. The increase of ¹H_mix (spin diffusion time from H₂O to protein) from 5 to 12 ms revealed sites more distal to the exposed surface but with significantly worsened spectral crowding (Figure S18). Logically, residues observed in water-edited spectra compose cytoplasmic and extracellular domains along with transmembrane residues lining the conductance pathway with the exception of central sites of the selectivity filter, S2 and S3, as observed in MD studies.¹⁷ Residues 302–333 may also be impacted, but chemical shift assignments are not complete for this region.

Significant differences in water accessibility or CSPs were observed along the ion conduction pathway between spectra of the SM and TQ mutants, which correspond to the open-activated and closed-inactivated states, respectively (Figure 6). First, chemical shift assignments for residues in the transmembrane region of the TQ spectrum match the inactive state assignments in the POPC bilayers, in particular, residues in the extracellular loops (V98), selectivity filter (T110, Y113), and TM2 (G137 and I138). G137 and I138 compose a critical hinge site in TM2 involved in activation. Both residues exhibit the same ~2 ppm C′ CSPs observed between the active and inactive states of the channel. The G137 Cα resonance also shifts downfield by 1.4 ppm relative to the activated state, further indicating there is no kink at this site (Figure 6e). Thus, the inner helix and selectivity filter of TQ adopts an inactivated conformation.⁴⁰ Interestingly, the outer selectivity filter residues Y113 and G114 corresponding to S1 were observed in spectra of both samples, while inner residue V111 was water proximal only in the SM and G112 was not observed in either sample. This indicates that S1 and S4 are water proximal in the SM, while for TQ only S1 is accessible to water. This is interesting in light of a recent study on the NaK2K channel that concluded water is absent from the selectivity filter under physiological conditions using ²H exchange SSNMR spectroscopy. This study proposed a direct “knock-on” diffusion model that does not require water to enter the selectivity filter for K⁺ conductance.¹⁶ However, it is possible the T₂ relaxation rate of water inside the selectivity filter decreases and this signal is filtered out along with the protein ¹H signals. Regardless, it is abundantly clear that H₂O is present at either end of the selectivity filter in the open-activated conformation to facilitate efficient hydration/dehydration of K⁺ as it passes through the selectivity filter. Unlike KcsA and Kv channels, the KirBac1.1 selectivity filter does not enter the pinched or “deeply inactivated” conformation.⁶⁷ It was previously shown that water behind the selectivity filter helps to stabilize this filter conformation. Thus, the lack of water behind the Kirbac1.1 selectivity filter may partially explain why this state is never observed in Kir channels.⁶⁸ Two residues in the turret, directly downstream of the selectivity filter (D115 and M116) are more pronounced in the TQ spectrum while T120, I89, and P88 were more pronounced in the SM spectrum. This observation is consistent with previously observed allosteric changes of this region upon opening of the activation gate.⁶⁹ Large helical populations for valine and isoleucine Cα-Cβ shifts were observed in both water-edited spectra. These resonances are almost exclusively embedded within the membrane, indicating these populations correlate to residues lining the transmembrane pore. As these populations appear in both mutants, we conclude trapped water is present in the cavity between the SF and activation gate in both samples. However, interestingly, this water does not readily access the S4 site at the base of the selectivity filter as stated above. Indeed, residues in the upper cavity and inner gate were more pronounced in the SM sample, while those in the central cavity were relatively similar between the two samples. A few central sites, including I138, were more pronounced in the TQ mutant, indicating that trapped water is centralized to this region. Interestingly, this is the site “rectification control” in strongly rectifying Kir channels.⁷⁰ In addition, slide helix residues D50, Y52, and K57 along with other residues become visible in the SM spectrum as lipid tethering shifts the slide helix outward into contact with water. We presume lipid binding allows water to access c-linker residues near the lipid binding site (P152, A154) only in the SM as the activation gate opens to allow water and ions to pass. While those further down (K155-M157) were more pronounced in the TQ as the subunits move apart in the closed state. Our 3D assignments agree well with these findings, indicating once again that the TQ closed state mimics the native closed state.

The most dramatic water accessibility changes occurred in the CTD, which was significantly more solvent exposed in TQ, as quantified above (Figure 5). We site-specifically assigned most Arg and Glu residues in this region, along with multiple isoleucine and proline residues (Figure 6c,d). We have identified 9 glutamate residues (37, 187, 198, 202, 218, 234, 244, 248, 258) and 12 basic residues (R22, R26, R30, R36, K155, R181, R189, K191, R196, K208, R216, R271) as prime candidates for key intersubunit salt bridges on the basis of our assignments. These resonances are more pronounced in the TQ spectrum or are completely absent in the spectrum of the SM protein. Charged residues appear to be clustered at the interfaces between subunits but are found all throughout the cytoplasmic domain. In particular, the small helical region at the base of the CTD (244–248) and nearby loops, which are absent in X-ray crystal structures, seem of particular importance. In the activated state, these residues likely form strong interactions, forcing the subunits closer together and tilting the top of the CTD outward, opening the gate. Four basic residues in the N-terminus listed above are also involved in the process of channel opening. R22 is absent from all X-ray crystal structures, but our data indicates it forms important functional contacts. Interestingly, the weak rectification sites E187 and E258⁶ are strongly observed in the TQ spectrum but are absent from the SM spectrum. This suggests that an underexplored means of rectification may be stabilization of the closed state of the C-terminus rather than simply obstructing the conductance pathway as these residues must be accessible to polyamine binding during rectification. Many of our observations correspond to previously uncharacterized regions of the cytoplasmic domain. Indeed, many of the residues we identify as key to channel function are not well characterized in any KirBac1.1 X-ray crystal structure, including 4–35, 199–205, 290–295, and 310–333. These uncharacterized regions appear to be key players in this mechanism as they contain many charged residues within the list of likely candidates listed above (Figure 6f). This gives insight into the functional importance of these regions as they form important ionic contacts, stabilizing the active state.

Lipid Binding Promotes “Paddle-Like” Mechanism.

The minimal changes in the standard DARR correlation experiment (Figure 4) suggest an orientational rearrangement of subunits relative to each other with minimal changes in overall protein fold. This along with the rearrangement of connective regions constitute a paddle-like mechanism. This mechanism is coupled with the formation of multiple intersubunit salt bridges between the CTDs upon activation, bringing subunits in close contact with one another. This agrees with recent studies that describe an inward twisting and puckering of CTDs more perpendicular with the membrane coupled with a widening and opening of the activation gate.^7,15,65 We observe a host of new ionic contacts that stabilize the active state in conjunction with those previously described. Many glutamate residues in the closed-inactive crystal structure face toward the intracellular vestibule along with aforementioned rectification residues. In addition to these are residues in flexible loops and a small helical region at the base of the CTD, which seem to play a critical role. These residues likely interact with basic residues in an N-terminal loop not visible in the crystal structure. Thus, a clockwise twist (bottom-up) upon activation depicted in the bottom panel of Figure 7 is likely responsible for burying many of the charged residues lining the vestibule and base of the CTD between subunit interfaces. These described processes are initiated by the ionic association of both c-linker and slide helix residues to lipid phosphates (Figure 7).

Figure 7.

Cartoon of the proposed structural changes to SM KirBac1.1 (left), upon removal of the anionic lipid binding site in the TQ mutant (right). Membrane and aqueous regions are highlighted in yellow and blue, respectively. Solvent exposed protein surface for each respective state is highlighted in light blue, and potassium ions are depicted as yellow circles. The bottom-up perspective (bottom) of CTD motions is shown for both SM and TQ.

CONCLUSION

The determination of the direct functional role played by bilayer lipids is crucial to the advancement of membrane protein structural biology and therapeutic development. Above, we site-specifically identify the activation locus, global motions, and stabilization mechanisms underlying functional states of a Kir channel. We explicitly characterized the KirBac1.1 anionic lipid binding and tethering site (R49/151/153), which shares close homology to eukaryotic channels. This site has been previously identified in part via further refinement of crystallographic data showing c-linker residues R151 and R153 with a bound phosphate.⁷ Additionally, the slide helix residue R49 was shown to be required for channel activation.¹³ However, neither of these studies elucidated the full structural and functional implications actuated by phospholipid binding. We find that the lipid binding site couples the slide helix-TM1 region with the c-linker and activation gate, leading to a global rearrangement of the channel. Our R to Q mutations show that this is a charge dependent process with near complete inactivation of the TQ mutant even in the presence of activating anionic lipids (PG,CL). Our FRET experiments indicate that the TQ inactivated state mirrors the native inactive state in POPC bilayers and that the PIP₂ induced inactive state differs from the native one. We also identified in our 2D spectra distinct arginine Cδ chemical shift outliers, which disappear in the absence of anionic lipids and in spectra of the TQ mutant. We confirmed these outliers correspond to lipid-binding arginine side chains by performing ¹³C–³¹P REDOR dephasing experiments, which indicated a ~5–8 Å Cδ-³¹P distance characteristic of a salt bridge with lipid phosphate groups. Our rINEPT experiments of SM and TQ variants in an activating lipid environment reveal further effects on lipid phase properties after removal of the lipid binding site. The TQ melting curve exhibited monophasic properties characteristic of lipids in absence of KirBac1.1.

We observed the effects of this mutation not only on activity and protein–lipid interactions but also on the solvent accessible surface and overall channel conformation via water-edited SSNMR. Our ¹H–¹H spin diffusion experiments showed a drastic 46% increase in water exposed surface in the TQ mutant. This was even more drastic for charged carboxyl and amino groups, suggesting a disruption of salt bridges in the inactive TQ conformer. This observation is also consistent with the water-edited 2D ¹H–¹³C and ¹³C–¹³C spectra (Figure 5). We believe this to be due to changes in tertiary structure of CTDs and the N-terminus in which the majority of these charged groups are located. We observed many sites exposed only in the TQ, which correspond to residues in the CTD. These residues are primarily located at the interfaces between the CTDs (Figure 6), and very few residues in this region have increased exposure in the SM by comparison. We also see significant differences in functional domains of the transmembrane region and extracellular loops.

The transmembrane cavity experiences large changes in solvent accessibility between the SM and TQ. It is seen in the water-edited spectra that many more assignable resonances along the conductance pathway are present in the SM than in the TQ. These manifest in unique α-helical resonances that closely agree with the assignments of KirBac1.1 in an activating lipid environment.⁴⁰ This increase in water accessibility of the transmembrane cavity shows that KirBac1.1 in its lipid activated state is able to allow water to enter its central pore. Although we cannot definitively say if water passes through the selectivity filter via the “knock-on” mechanism, we observe water interacting with the top (G114) and bottom (T110) of the selectivity filter. This suggests both the S1 and S4 sites are water proximal for the SM and to a much lesser extent the TQ.

All of these observations suggest that, even in an activating lipid environment, the TQ exhibits an inactive selectivity filter and closed activation gate or a closed-inactive conformation. This occurs in conjunction with an outward twisting motion of the CTDs relative to each other, and we believe closing/tightening of the activation gate upon disruption of multiple intersubunit salt bridges stabilizes the active conformer. This is in agreement with previously proposed closed-inactive conformations.^7,15,65 With the exception of the selectivity filter and TM2 along with flexible cytoplasmic domains, the overall protein fold does not change significantly in the “paddle-like” mechanism discussed earlier. We believe these motions occur on a time scale similar to those previously described in solution state NMR experiments.³⁴ All of these changes occur due to the abolishment of anionic lipid binding and tethering, which we show to be key in channel activation.^{3,4,6,9,14,18–29,31,44,56,65,67,69–77}