Role of protein dynamics in ion selectivity and allosteric coupling in the NaK channel

NaK is small and highly stable, facilitating high-resolution NMR studies. We first performed radioactive rubidium flux assays to confirm that NaK produced in our laboratory and reconstituted into proteoliposomes was functional (26) (Fig. S1). For NMR, NaK proteoliposomes were solubilized with DHPC to create lipid bicelles, as described in our previously published protocol (27).

The 2D-TROSY HSQC spectra of NaK in bicelles provide insight into both the structure and the dynamics of the protein. These spectra are of the NaK∆19 construct with the M0 helix truncated, which is referred to as NaK throughout. Information is contained in both the observed chemical shifts and the peak dispersion, as well as the number of peaks and their relative intensities. The spectrum of NaK solubilized in q = 0.33 DMPC:DHPC bicelles at 45 °C displays excellent peak dispersion (Fig. 2), indicating that NaK is well folded under these conditions. Helical membrane proteins often have significant peak overlap due to the modest dispersion of amide chemical shifts in helical regions and the uniformity of the membrane environment. The good dispersion observed for NaK suggests that the helices are tightly packed with a well-defined tertiary structure resulting in unique chemical shifts. Eighty-three percent of NaK peaks have been assigned at 65 °C (Figs. S2 and S3), and the secondary structure of NaK determined from the chemical shift index matches the crystal structure (PDB 3E8H) quite well (Fig. 2, Inset) (28, 29). These spectra confirm the structural stability of NaK in isotropic bicelles up to 65 °C, as expected from its thermal stability, and establish the suitability of these conditions for detailed investigation of channel structure and dynamics by solution NMR.

Our initial goal was to compare the backbone dynamics of NaK and NaK2K to experimentally test the role of dynamics in ion selectivity. However, ¹H-¹⁵N TROSY-HSQC spectra of NaK2K revealed widespread chemical shift differences (Fig. 3) that were completely unexpected on the basis of the available crystal structures. In the high-resolution crystal structures of NaK and NaK2K, differences are highly localized to the selectivity filter region only (Fig. 1B) (21). We therefore expected relatively few peak shifts when comparing spectra of NaK and NaK2K bound to K⁺, localized to the region in and around the selectivity filter. Instead, we observed widespread chemical shift changes (Fig. 3). The major difference between crystallography and NMR is that a single structural state of a protein is trapped within a crystal, whereas the NMR data represent the population-weighted average of all states present in the bicelle. Thus, these chemical shift differences may be explained by a shift in protein structure between NaK and NaK2K that is not captured in the crystal, or they could arise from changes in the equilibrium distribution of states sampled by NaK and NaK2K. Due to these significant chemical shift differences, NaK2K was independently assigned. Each of the individual point mutations (D66Y or N68D) has a spectrum with only minor chemical shift differences compared with NaK (Fig. S4).

Mapping the chemical shift changes between NaK and NaK2K onto the NaK structure shows the largest differences are in the selectivity filter region as expected (Fig. 3B). Significant chemical shift differences are also seen in the M2 helix in regions that change conformation upon channel opening/closing. This includes the hinge region, which bends upon opening/closing of the inner gate (residues G87 and G89), a hydrophobic patch that rearranges upon channel opening (F94 and I95) and the constriction point formed by the M2 helix (N101 and Q103) (Fig. 3 A and B, arrow) (20). Some of these residues are over 25 Å away from the site of mutation, well beyond the range of simple effects of mutation on neighboring residues or structural differences predicted by the crystal structures. This demonstrates that the sequence and structural state of the selectivity filter affects the entire NaK protein and is coupled to the inner gate.

Such coupling between the selectivity filter and the inner gate is well established in KcsA and several other K⁺ channels but has not previously been demonstrated for NaK, which has a much more stable structural scaffold. This leads to a number of questions. (i) Are the chemical shift changes an artifact of experimental conditions? (ii) Is coupling between the gates inherent to NaK or introduced upon creation of a selective filter? (iii) Can coupling be observed due to changes at the inner gate?

One striking difference between KcsA and NaK is the stability of the NaK selectivity filter in various ionic conditions. We therefore tested whether this was true in our NMR samples. Spectra of NaK in bicelles with K⁺ only, Na⁺ only, or low salt show very minor shifts that are localized to the selectivity filter and adjacent regions (Fig. S5). Residues D66, G67, and S70 are sensitive to ion identity, consistent with differences in ion binding at the top of the selectivity filter in the crystal structures (22). The NMR data are thus consistent with the lack of selectivity filter collapse observed crystallographically. This prominent feature of NaK makes it an interesting model system to compare with KcsA, where structural changes in the selectivity filter are integral to coupling between the gates.

The secondary structure determined from the chemical shift index is very similar between NaK and NaK2K and closely matches the crystal structures (Fig. S6) (29). Together, the secondary structure and ion binding data support the high stability and proper folding of NaK in bicelles. Then what can explain the observed chemical shift differences between NaK and NaK2K? Although the crystal structures represent a high-resolution snapshot of a single state, the single set of peaks observed in the NMR spectrum reflects the population-weighted average of all of the states sampled by NaK. A shift in the equilibrium distribution of structural states between NaK and NaK2K could lead to the observed chemical shift differences. To test this hypothesis, we further assessed the dynamics of NaK and NaK2K.

To determine if coupling between gates is an inherent property of the nonselective NaK channel, we measured NaK dynamics on the microsecond–millisecond timescale using ¹⁵N-TROSY-CPMG experiments (Fig. 4). Due to the inherent insensitivity of these experiments, full deuteration of the protein is required. This reduces the number of residues that can be monitored because of limited back exchange of amide protons in this highly stable channel. The figure shows ∆R₂ (Fig. 4), determined as described in Methods. CPMG experiments monitor the transverse relaxation rate (R₂), which includes intrinsic relaxation plus additional line broadening due to loss of coherence upon conformational exchange between states with different chemical shift values. The exchange contribution to the apparent R₂ can be suppressed by rapid pulsing during the CPMG experiment. By comparing CPMG experiments performed with slow (low ν_CPMG) and fast (high ν_CPMG) pulsing, intrinsic relaxation is eliminated and ∆R₂ reflects only the exchange contribution to the transverse relaxation rate. Thus, ∆R₂ will be 0 if there are no microsecond–millisecond timescale dynamics, and ∆R₂ will be positive if there is conformational exchange on this timescale. The results show conformational exchange throughout the protein, particularly at the bottom of the selectivity filter and the hinge point in M2, as well as at the constriction point on the intracellular side of the channel. Because CPMG experiments detect dynamics only when the exchanging states have different chemical shifts, not every residue in a region that is exchanging will have a measurable ∆R₂. The regions with the most significant ∆R₂ values are exactly where local structural changes occur between the open and closed channel (20) (Figs. 1C and 4). These results support the idea that transmembrane allostery between the selectivity filter and intracellular gate is an inherent property in NaK. Because there is no reason a priori to assume two-state exchange and the available electrophysiology data on NaK are not sufficient for quantitative comparison of rates and populations, we did not attempt further quantitative analysis of the exchange process.

Does the coupling observed upon mutation of NaK to NaK2K arise from changes in equilibrium dynamic state? To assess dynamic differences between NaK and NaK2K, peak intensities were compared between the two spectra. Increased protein motion leads to line-broadening and a decrease in peak intensity. For comparison, peak intensities were normalized to the C-terminal residue, N110, which is far from the site of mutation and has identical chemical shifts in NaK and NaK2K. This approach, rather than comparison of ∆R₂ values, was used because deuteration of NaK and NaK2K for CPMG experiments severely limits the number of residues that can be analyzed. Comparison of peak intensity provides data for more residues throughout the protein and is sensitive to dynamics on a broader range of timescales, including faster fluctuations of the amide bond vector. NaK2K has stronger peak intensities in the region around the selectivity filter, particularly for residues in the pore helix (I51 and L54), but also in the surrounding loops (Fig. 5). Outside this scaffold region, differences in peak intensity between NaK and NaK2K are less significant, with an increase in peak intensity for NaK at the constriction point. This suggests that the scaffold behind the K⁺-selective NaK2K selectivity filter is less dynamic or more structurally homogenous than in nonselective NaK. These results are consistent with a role for structural stability in maintaining a four-site ion-selective filter and K⁺-selectivity, a phenomenon that has been extensively studied previously (30). They also suggest that the observed coupling between the gates is tuned by the dynamic state of the protein.

One convincing way to demonstrate coupling between the inner gate and selectivity filter is to show that this coupling works both ways. To test this, we recorded spectra of full length NaK including the M0 helix (identified throughout as FLNaK). FLNaK provides a good model of the closed state based on both the crystal structure and ion flux assays (Fig. 1C) (19). Initial assessment of this spectrum reveals peak doubling, which reveals that FLNaK exists in two states in isotropic bicelles at 45 °C (Fig. 6 and Fig. S7). Upon raising the temperature to 65 °C, each pair of peaks merges to a single peak, demonstrating that these two states are able to interconvert (Fig. S8). Examining only well-resolved and assigned residues reveals this peak doubling occurs throughout the protein with similar populations of the two states. This suggests a global two-state exchange process, although the states cannot be assigned from these data alone (Fig. 6). Further, chemical shift differences between NaK (open-state model) and FLNaK (closed state model) extend above the hinge on the M2 helix (Fig. S9), demonstrating reverse coupling from the inner gate to the scaffold directly behind the selectivity filter.

The structural stability of NaK has enabled engineering of the selectivity filter, tuning the number of ion binding sites and ion selectivity of the channel (21, 23). Beautiful high-resolution crystal structures of these different mutants have contributed to our current understanding of the structure–function relationships underlying ion selectivity and support the model that four ion binding sites are necessary for potassium selectivity through a knock-on mechanism (21–23, 25, 31–33). Here we present NMR data on the NaK channel solubilized in isotropic bicelles, adding experimental insight into the dynamics of the protein.

The scaffold holding the selectivity filter has long been appreciated for its important role in tuning selectivity in K⁺ channels (7, 13). Our results extend this idea in the NaK channel by revealing increased rigidity of the scaffold when the selectivity filter is K⁺-selective and less rigidity of the scaffold when the selectivity filter is nonselective (Fig. 5B). This complements NMR studies of KcsA that show line broadening in the selectivity filter and scaffold for a mutation that decreases selectivity (12). Our data suggest that a more rigid scaffold may be necessary to stabilize a selectivity filter structure with four ion binding sites, and K⁺-selectivity is thought to require four ion binding sites (24, 34).

Coupling between gates has been well established for a number of ion channels; however, the mechanistic conservation of this coupling has not been studied. Our solution NMR provides initial experimental insight into the motion of the NaK channel. FLNaK, which crystallized in the closed form, displays clear peak doubling throughout the spectra (19). This is indicative of two states that interconvert slowly (millisecond–second timescale) (Fig. 6). Removal of the M0 helix creates the NaK construct, which crystallized in the open state. The NMR experiments reveal that NaK is still undergoing conformational exchange, just on a faster microsecond–millisecond timescale. These dynamics occur throughout the M2 helix, including the hinge and channel constriction point, providing a dynamic pathway connecting the selectivity filter to the inner channel gate (Figs. 3B and 4B). Motion on this timescale is often correlated with exchange between functional states in the case of enzymes (35). Although we cannot currently assign the conformational exchange process to functional states of the channel, we note that the residues sensing motion are localized to the regions of NaK that would change environment upon channel opening/closing.

Coupling between gates occurs in both directions, as changes in the selectivity filter lead to significant changes at the inner gate as shown by the NaK2K mutation (Fig. 3B). The regions with microsecond–millisecond dynamics in NaK are the same regions with the largest chemical shift changes between NaK and NaK2K (Figs. 3B and 4B). This suggests that formation of a rigid K⁺-selective filter leads to changes in the equilibrium state of the M2 hinge and the inner gate, a distant but functionally important site.

How does our observed coupling compare with KcsA? In KcsA, major structural rearrangements at the selectivity filter lead to its collapse and loss of ion conduction (3). This collapsed state is communicated to the inner gate by interactions with the scaffold (36), similar to our proposed path of communication between the gates in NaK. However, in NaK there is no evidence for collapse or even minor ion-dependent structural changes in the NaK selectivity filter by crystallography or NMR (Fig. S5). Thus, the dynamic coupling between the gates in NaK must work through a different mechanism. NaK is a flickery channel, and flicker gating has been suggested to arise from changes in conduction at the selectivity filter (21). However, NaK electrophysiology is limited, and the current structural data do not provide any insight into how the rigid NaK selectivity filter could cause changes in ion conduction. Thorough single channel recordings of this system combined with more quantitative NMR studies will be needed to resolve the role of the selectivity filter in channel gating and the mechanism linking the selectivity filter to the inner gate.

Our results emphasize the intimate connections between the selectivity filter and the overall channel structure. These interactions are important for both structural stability of the selectivity filter and dynamic coupling between the filter and inner channel gate, which will determine overall flux through the ion channel.

NaK constructs were missing the first 19 amino acids, corresponding to the NaK∆19 construct used to determine most crystal structure of NaK and NaK mutants (21, 23). NaK∆19 is referred to as NaK throughout to emphasize the different mutations within this construct that are the focus of this work. NaK and the mutants studied here were purified largely as published (20, 27), with a few changes to optimize expression in minimal media. NaK was transferred from NaK-pQE60 (generously provided by Youxing Jiang) to a pET15B vector with an N-terminal 6xHis tag. This led to 2–3× higher yield of NaK when expressed in minimal media for isotopic labeling. Expression of NaK is described in detail in Supporting Information. In brief, NaK was overexpressed in BL21 (DE3) cells in M9 minimal media and induced with IPTG (isopropyl-beta-D-thiogalactopyranoside) at 20 °C for 16–20 h. The membrane fraction was isolated and solubilized in 20 mM DM (n-decyl-beta-D-maltopyranoside). NaK was purified by IMAC (immobilized-metal affinity chromatography) with Talon cobalt affinity resin (Clontech) and size exclusion chromatography on a Superdex 200 column.

NaK concentration was determined by A280 using the calculated extinction coefficient of 3,840 L ⋅ mol^-1 ⋅ cm⁻¹. NaK was reconstituted into bicelles using our previously published protocol with slight alterations (27). DMPC (1,2-dimyristoyl-sn-glycero-3-phosphocholine) was hydrated at 20 mg/mL in NMR buffer, bath sonicated ∼1 min and solubilized with 10 mM DM for 20 min. NaK with the His-tag removed was added to the solubilized DMPC at a molar ratio of 1:100 NaK monomer/DMPC and rotated at room temperature for 3 h. Two aliquots of 45 mg Amberlite XAD-2 (BioRad) per milligram of total detergent were added to remove the detergent and incubated overnight at room temperature. Amberlite was removed, and the NaK proteoliposomes were ultracentrifuged at 150,000 × g for 2 h at 6 °C. The proteoliposome pellet was solubilized with DHPC (1,2-dihexanoyl-sn-glycero-3-phosphocholine) dissolved in NMR buffer to create q = 0.33 DMPC/DHPC bicelles. Four freeze–thaw cycles produced homogeneous bicelles containing NaK, which were flash-frozen and stored at −80 °C until use.

NMR samples contained 0.5–1.5 mM NaK in q = 0.33 DMPC/DHPC bicelles in 100 mM MOPS, 40 mM KCl, pH 7 with 10% D₂O. Samples with different salt conditions, as described in the text, were prepared using the desired final buffer/salt for the S200 column and all subsequent steps. The ¹H-¹⁵N TROSY spectra were collected on a 700-MHz Varian spectrometer with a room temperature probe. Three-dimensional backbone walk experiments were acquired on a 600-MHz Bruker spectrometer with cryoprobe (Washington University) or 750-MHz Bruker spectrometer with cryoprobe [National Magnetic Resonance Facility at Madison (NMRFAM)]. TROSY-HNCA, BEST-TROSY HNCA, BEST-TROSY HN(co)CA, and BEST-TROSY HNCACB (37, 38) were collected using a nonuniform 10% poisson-gap sampling schedule (39), reconstructed using istHMS (40), processed using nmrpipe (41) and analyzed with CCPNMR Analysis (42). Backbone walk experiments were supplemented with 1–¹³C′-amino acid specific labeling (43). Assignments were submitted to the Biological Magnetic Resonance Data Bank (BMRB).

Chemical shift differences (Δδ_tot) were calculated using Eq. 1,

TROSY-CPMG (44, 45) experiments were performed on an 800-MHz Varian spectrometer with a cryoprobe (NMRFAM), with a relaxation delay (T_relax) of 40 ms. Two CPMG field strengths were acquired at ν_CPMG = 100 Hz and ν_CPMG = 1,000 Hz. ∆R_2,app was determined with Eq. 2, with I_νCPMG and I₀ being the peak intensities with and without CPMG refocusing periods. ∆R₂ was calculated for all well-resolved residues using Eqs. 2 and 3 (46):

All NaK constructs were missing the first 19 amino acids, corresponding to the NaK∆19 construct used to determine many crystal structures of NaK and NaK mutants (21, 23). NaK∆19 is referred to as NaK throughout to emphasize the different mutations within this construct that are the focus of this work. Full-length NaK is always named explicitly as FLNaK. All constructs were purified largely as published (20, 27), with a few changes to optimize expression in minimal media. NaK was transferred from NaK-pQE60 (generously provided by Youxing Jiang) to a pET15B vector with an N-terminal 6xHis tag. This led to 2–3× higher yield of NaK, when expressed in minimal media for isotopic labeling. BL21(DE3) cells containing the NaK-pET15b were grown in LB media with ampicillin. After 2 h the culture was transferred to 1 L of M9 minimal media. For isotopic labeling, the standard media components were substituted with ¹⁵NH₄Cl, ¹³C-glucose, or D₂O plus 0.5 g/L isotopically labeled isogro (Sigma-Aldrich). Cells were induced with 0.4 mM IPTG at OD 0.8–0.9 and harvested after overnight expression at 25 °C. Cells were stored at −80 °C if purification was not performed immediately. To purify NaK, cells were resuspended in lysis buffer (100 mM NaCl, 200 mM KCl, 2.5 mM MgSO₄, 20 mM Tris pH 7.5, DNase, 1 µg/mL pepstatin, 10 µM leupeptin, 100 µM PMSF) with 250 mM sucrose and 1 mg/mL lysozyme and lysed by sonication. The membrane fraction was isolated by spinning for 1 h at 30,000 × g, resuspended in lysis buffer with 20 mM DM (Anatrace), and solubilized for 3 h with rotation at room temperature. Insoluble protein was removed by centrifugation at 30,000 × g for 10 min. Solubilized NaK was purified by IMAC with 0.5 mL Talon cobalt affinity resin (Clontech) per liter of growth and was prewashed with Buffer A (5 mM DM, 20 mM Tris⋅HCl, 200 mM KCl, 90 mM NaCl, pH 7.8). NaK was allowed to bind for 20 min at room temperature, and then the beads were washed with 10 bed volumes of Buffer A followed by 10 bed volumes of Buffer A containing 15 mM imidazole. NaK was eluted in 8 bed volumes of elution buffer (Buffer A plus 300 mM imidazole), concentrated to 0.5 mL and loaded on a Superdex 200 column equilibrated in NMR buffer (100 mM MOPS, 40 mM KCl, pH 7) containing 5 mM DM. For low-K⁺ and low-salt samples, gel filtration chromatography and all subsequent steps of the protocol were performed using 100 mM MOPS/40 mM NaCl for low potassium or 100 mM ultrapure MOPS for low salt. To cleave the 6xHis-tag, fractions containing NaK were incubated overnight with 10 U of thrombin (Sigma).

Without deuteration, 87 out of 95 expected nonproline resonances are observed (Fig. 2). As was observed for KcsA (47), full deuteration of NaK leads to a loss of ∼40% of peaks due to incomplete back-exchange within the core of the transmembrane helices, consistent with a stably folded structure. Assignment of 83% of the nonproline residues was achieved using a combined strategy of traditional backbone walk experiments (Figs. S2 and S3), amino acid-specific labeling to provide anchor points (43), and amide NOESY. The technical challenges of assigning such a large (8 transmembrane helix) membrane protein solubilized in isotropic bicelles were overcome by taking advantage of the thermal stability of NaK and recent developments in nonuniform sampling and BEST-TROSY NMR (38, 48). The inability to fully deuterate prevented complete assignment, but the assigned residues provide good coverage throughout NaK as shown in Fig. 2.

Liposome reconstitution procedures and flux assays were carried out as described before with minor modifications (26). Purified NaK was reconstituted into 10 mg/mL lipid with 3:1 1-palmitoyl-2-oleoyl-phosphatidylethanolamine (POPE)/1-palmitoyl-2-oleoyl-phosphatidylglycerol (POPG) hydrated in reconstitution buffer B (450 mM KCl, 10 mM Hepes, 4 mM NMDG, 1 mM EDTA, and 1 mM EGTA, pH 7.4). Lipids were solubilized with 32.5 mM CHAPS, and then protein was added in a ratio of 10 μg per mg lipid and incubated for 20 mins at room temperature. Proteoliposomes containing NaK were formed by centrifuging the protein/lipid mixture through a 1.5-mL Sephadex G-50 gel filtration column to remove the detergent. For flux to occur, an electrochemical gradient was formed by exchanging the outside buffer to low-salt buffer C (400 mM Sorbitol, 10 mM Hepes, 4 mM NMDG, 1 mM EDTA, and 1 mM EGTA, pH 7.4). For uptake assays, 5 uM of ⁸⁶Rb was added to buffer C. Flux assays were run in 400 µL total volume. At each time point, a 60–80 μL aliquot was removed and passed over a 1.5-mL Dowex cation exchange column to remove extraliposomal ⁸⁶Rb. Samples were eluted from the Dowex columns with two 1-mL aliquots of 400 mM Sorbitol and collected in scintillation vials. For normalization, valinomycin at a final concentration of 1 μg⋅mL^–1 was added to an extra aliquot at the final time point, incubated for 10 min, and then loaded on and eluted from a Dowex column as described above. All eluted samples were mixed with 16 mL scintillation fluid, and radioactivity was measured in a scintillation counter.