Allosteric coupling between transmembrane segment 4 and the selectivity filter of TALK1 potassium channels regulates their gating by extracellular pH.

Opening of two-pore domain K+ channels (K2Ps) is regulated by various external cues, such as pH, membrane tension, or temperature, which allosterically modulate the selectivity filter (SF) gate. However, how these cues cause conformational changes in the SF of some K2P channels remains unclear. Herein, we investigate the mechanisms by which extracellular pH affects gating in an alkaline-activated K2P channel, TALK1, using electrophysiology and molecular dynamics (MD) simulations. We show that R233, located at the N-terminal end of transmembrane segment 4, is the primary pHo sensor. This residue distally regulates the orientation of the carbonyl group at the S1 potassium-binding site through an interacting network composed of residues on transmembrane segment 4, the pore helix domain 1, and the SF. Moreover, in the presence of divalent cations, we found the acidic pH-activated R233E mutant recapitulates the network interactions of protonated R233. Intriguingly, our data further suggested stochastic coupling between R233 and the SF gate, which can be described by an allosteric gating model. We propose that this allosteric model could predict the hybrid pH sensitivity in heterodimeric channels with alkaline-activated and acidic-activated K2P subunits.

In potassium (K+) channels, the K+ ions pass through pore regions assembled from tetrameric pore-forming sequences, including the pore helix (PH) and the selectivity filter (SF; amino acid sequence, T-V/I-G-F/Y-G). The PH provides a supporting force to maintain the conformation of the SF, while the five carbonyl groups in the SF form the K+ binding sites (S0-S4) that mimic K+ hydration (1, 2). Mild distortion of the SF conformation destabilizes K+ binding in ways that can tremendously impact the channel properties, including opening probability (Po), conductance, and gating kinetics (3, 4, 5, 6, 7). This phenomenon has been demonstrated as the molecular basis of SF gating, that is, C-type gating, in potassium channels (8).

Two-pore domain K+ (K2P) channels (Fig. 1A) are a group of dimeric proteins responsible for setting the resting membrane potentials of various cell types (9, 10, 11). These channels are responsive to diverse external cues such as pH (12), membrane tension (13, 14, 15), and temperature (16, 17), which are widely thought to regulate K2P activities via C-type gating mechanisms. However, the cue-sensing residues are clustered in a hot-spot region formed by the M1P1 loop, the M2P4 loop, and the N-terminal or C-terminal portions of transmembrane segment 4 (TM4), which is structurally far from the SF (18, 19, 20, 21). Therefore, the molecular linkages between the cue sensors and C-type gates are not immediately apparent and must be defined for each type of K2P channel.

Figure 1

R233 is the primary pHo sensor of TALK1. A, the diagram represents the domain-swapping architecture of the TALK1 channel. Blue and yellow colors represent individual TALK1 subunits; white spheres represent K+. The M1P1 loop and P2M4 loop are highlighted in green and blue, respectively. TM1-TM4 are labeled as M1-M4 in the diagram. B, the amino acid sequence alignment of the M1P1 loop, PH1, P2M4 loop, and TM4 regions among pHo-sensitive K2P channels, including TALK1 (NP_001128578), TASK2 (NP_003731), TREK1 (NP_001017424), TREK2 (XP_024305396), and TRAAK (NP_201567). The dashed line represents missing residues. Red squares represent the four pHo sensor candidates in TALK1, K84, K226, H227, and R233. The critical residues R233 and L114 are labeled by red and black asterisks in (A) diagram and (B) sequence alignment, respectively. C, the I-V plot shows the whole-cell currents from COS7 cells transfected with WT TALK1 channels at various pHo values (pHo 5.0 − 10.0); the external and internal solutions contain 2 mM K+ and 150 mM K+ (EK = −110 mV), respectively. D, the pHo-sensitive curves of four candidates were acquired by normalizing to the current densities at pHo 10.0, 0 mV, and by fitting with the Hill equation. Two parameters, pK1/2 and nH, reveal the pHo sensitivity of TALK1-WT (pK1/2 = 7.6 ± 0.1; nH = 0.52 ± 0.05, n = 16), K84A (pK1/2 = 7.9 ± 0.2; nH = 0.48 ± 0.10, n = 5), K226A (pK1/2 = 7.3 ± 0.1; nH = 0.57 ± 0.08, n = 6), H227A (pK1/2 = 7.8 ± 0.6; nH = 0.40 ± 0.18, n = 6), and R233A (unsuccessful fitting, n = 7). The pHo-sensitive curves represent pHo sensitivities of (E) neutral, (F) positively charged, and (G) negatively charged residues at position 233. The black dashed lines represent TALK1-WT from Figure 1D (n = 5 for R233V and 6 for R233C, respectively); both pHo-sensitive curves cannot be fitted with the Hill equation. The pK1/2 of R233H is 5.2 ± 1.1 and nH is 0.41 ± 0.08 (n = 5); the pK1/2 of R233K is >10 (n = 5). The negatively charged R233E exhibits acidic activation (n = 7). K2P, two-pore domain potassium channel; PH, pore helix; TASK, TWIK-related acid-sensitive K+ channel; TM, transmembrane region; TREK, TWIK-related K+ channel.

Extracellular pH (pHo) regulates the C-type gate in three subtypes of K2P channels, including the TWIK-related acid-sensitive K+ channel (TASK), TWIK-related K+ channel (TREK), and TWIK-related alkaline pH-activated K+ channel (TALK) families. A set of crystallographic studies for TREK1 showed that the K+-dependent C-type gating involves disruption of S1 and S2 ion binding via asymmetric rearrangements of the backbone of SF1 and SF2 (22). Of note, heterodimeric TREK1 and TREK2 show a hybrid pHo sensitivity that can be activated by both acidification and alkalinization (23, 24, 25). In TASK3, a protonated H98 at the mouth of the SF guards the narrow entrance of side portals toward the S0 binding site (26). Compared to other pHo-operated K2P channels, TALK channels are activated at relatively high pHo (12). Previous studies suggested that this property may be due to high-pKa basic residues such as Arg and Lys in the TM4 acting as the pHo sensor (20, 27). Furthermore, a recently released structure for TASK2 suggested that a protonated Arg causes dilation of the S1 and S0 sites of SF1 through interaction with E228 and rearrangement of the long extracellular M1P1 loop (28). Although the TALK1 channel has a conserved Arg (R233), it lacks a Glu at the position equivalent to TASK2-E228 (i.e., TALK1-A237). Additionally, the low similarity between the M1P1 loops of TASK2 and TALK1 (Fig. 1B) suggests that TALK1 may exhibit a different pHo gating mechanism than TASK2. Interestingly, a recent case report showed that a novel TALK1 mutant, L114P, causes maturity-onset diabetes of the young (29). TALK1 is abundantly expressed in the pancreatic β cells (9, 30), and this gain-of-function L114P mutation was shown to inhibit glucose-stimulated insulin secretion (29). Since L114 is located near the outer mouth of the SF (Fig. 1A), it is likely to be involved in C-type gating. While these studies provide key pieces of information about the TALK1 gating mechanism, a comprehensive model has not been developed.

In this study, we investigated the mechanisms of pHo sensitivity in TALK1 channels by combining mutagenesis, electrophysiology, and molecular dynamics (MD) simulations. Based on our findings, we generated a model to describe the allosteric mechanisms of pHo gating in TALK1. Briefly, we found that the peripheral sensor, R233, stochastically regulates the C-type gate by flipping the carbonyl group of the S1 K+-binding site in SF1. Our model further predicts the behavior of heterodimeric R233 mutants, especially for the hybrid activation of heterodimeric K2P channels comprised of channel subtypes with distinct gating properties.

Results

R233 is the pHo-sensing residue in TALK1

To investigate the pHo-sensitivity of TALK1, we transfected pCDNA3.1 hTALK1 variant 3 into COS-7 cells and recorded whole-cell currents under a range of pHo values (pHo 5–10). The current-voltage plot of TALK1 WT showed typical outward-rectification under 2 mM [K+]o/150 mM [K+]i, and the currents were augmented by alkaline conditions with a pK1/2 of 7.6 ± 0.1 and Hill coefficient (nH) of 0.52 ± 0.05 (Fig. 1, C and D) (9). For other pHo-sensitive K2P channels, a single basic amino acid located either at the M1P1 loop, P2M4 loop, or the N-terminal TM4 domain has been shown to be the pHo sensor (18, 19, 20, 28). Sequence alignment of these pHo-sensitive K2P channels suggested four candidate pHo sensors, K84, K226, H227, and R233 (Fig. 1B). To evaluate whether any of these basic residues is the TALK1 pHo sensor, we mutated each residue to Ala and examined the resultant pHo sensitivities.

All the mutant channels exhibited stereotypical outward-rectification under 2 mM [K+]o/150 mM [K+]i (Fig. S1, A–D). Moreover, the TALK1-K84A and K226A mutants displayed comparable alkaline activation characteristics. The K84A mutant exhibited a pK1/2 of 7.9 ± 0.2 and nH of 0.48 ± 0.10, while the K226A mutant had a pK1/2 of 7.3 ± 0.1 and nH of 0.57 ± 0.08 (Fig. 1D, red and purple curves). Although the TALK1-H227A mutant also displayed alkaline activation, with pK1/2 of 7.8 ± 0.6 and nH of 0.40 ± 0.20, its span of current intensity was significantly reduced (Fig. 1D, magenta curve). By contrast, although the TALK1-R233A exhibited comparable current densities as the WT, its pHo sensitivity was completely abolished (Fig. 1D, blue circle). This result strongly suggests that R233 is the principal pHo sensor. Substituting R233 with other neutral amino acids such as Cys or Val also abolished the pHo sensitivity (Fig. 1E). Moreover, conservative mutation preserved the pHo sensitivity, as the pK1/2 values were 5.16 ± 1.17 for R233H and > 10 for R233K, which are close to the pKa for His (6.0) and Lys (10.5), respectively (Fig. 1F). Interestingly, the TALK1 R233E mutant displayed acidic activation rather than alkaline activation (Fig. 1G).

R233 protonation triggers flipping of the K+-ligating carbonyl group at the S1 site

Since R233 is located at the N-terminal part of TM4, far from the K+ permeation pathway (Fig. 1A), it cannot directly block K+ flow via electrostatic repulsion, as demonstrated for the pHo-sensing mechanism of TASK3 (26). To investigate how R233 might control the pHo sensitivity of TALK1, we compared MD simulations of protonated R233 (R233+) and deprotonated R233 (R2330). The dihedral angles of the 108T-I-G-Y-G112 residues lining the SF displayed greater variability in the MD simulations for R223+ than those for R2330. In particular, the backbone dihedral of Y111 lining the SF (ΨY111) rotated by ∼180° in 20% (replicates 1, 8, 11, and 16) of the R233+ simulations (Fig. 2A, top left and S2A), but this phenomenon was absent in the R2230 simulations (Fig. 2A, top right and S2B). This flip coincided with the loss of coordination between the Y111 carbonyl oxygen and K+ at the S1 binding site (Fig. 2B). Of note, the Ψ angle of I109 at the S3 binding site was also highly dynamic, but there was no observable difference in this angle between the R2330 and R233+ simulations (Fig. 2A, bottom panel). Therefore, although there are two locations (S1 and S3) in TALK1 that may contribute to C-type gating, only S1 is likely to be dependent on the protonation status of R233.

Figure 2

The carbonyl group at S1 K+ binding site is often flipped in R233+ MD simulations. A, the violin plots show the CαΨ angle of Y111 (S1 K+-binding site) and I109 (S3 K+-binding site) during R2330 or R233+ MD simulations. For each condition, ten sets of 200 ns MD simulations were performed; each column represents a single run of MD simulation for individual subunits (1–10 for subunit 1 and 11–20 for subunit 2). The black bars represent the average angles during each simulation. B, representative snapshots of the S1 K+-binding site from an R233+ simulation. The Y111 backbone carbonyl oxygen was either pointing toward (left, ΨY111 ∼ −50°) or away (right, ΨY111 ∼ 100°) from the S1 K+-binding site. The orange spheres represent K+. Normalized TALK1 currents from (C) TALK1-WT and (D) R233A upon a voltage step from -150 mV to 110 mV at a different pHo. E, the average time constants (τ) of voltage-dependent activation of TALK1 or TASK2 mutants were acquired by fitting with a single exponential equation: f(t) = A(1-e^(-t/τ)) + C. The alkaline solution accelerated the voltage-dependent activation in WT TASK2 but not TALK1. Neutralizing the pHo sensor (TASK2 R224A) also abolished the voltage-dependent activation (∗∗p< 0.01, paired t test). TASK, TWIK-related acid-sensitive K+ channel.

The pHo-dependent C-type gating of TASK2 involves the dilation of the S0 and S1 K+ binding sites in SF1 (28). Additionally, the kinetics of voltage-dependent K2P channels involves a slow rearrangement of the SF (31, 32). Such a slow rearrangement differs from the rapid flipping of the carbonyl groups of the K+ binding sites (4, 5, 6). Moreover, an alkaline extracellular solution would be expected to expedite the voltage-dependent process, but it may not affect the rapid gating by flipping the carbonyl group. As expected, elevating pHo accelerated the voltage-dependent activation of the TASK2-WT (48.41 ± 17.14 ms at pHo 7.4, 21.80 ± 10.44 ms at pHo 9.0) but not the pHo-insensitive TASK2-R224A mutant (41.57 ± 5.81 ms at pHo 7.4, 34.90 ± 8.07 ms at pHo 9.0) (Figs. S3, A and B, 2E). By contrast, varying pHo had no effect on the voltage-dependent activation of either pHo-sensitive TALK1-WT (0.19 ± 0.11 ms at pHo 7.4, 0.44 ± 0.73 ms at pHo 9.0) or pHo-insensitive TALK1-R233A (2.97 ± 1.65 ms at pHo 7.4, 3.04 ± 2.1 ms at pHo 9.0) (Fig. 2, C–E). These results imply that TALK1 and TASK2 may utilize distinct gating mechanisms even though their pHo sensors and corresponding gates are located at the equivalent position.

The interaction network of R233, PH1, and SF regulates C-type gate

Comparison of the structures between S1 normal and flipped forms shows that the side chains of R233 (at the TM4 N-terminus) and L114 (positioned halfway between the R233 and the SF) were rotated (Fig. 3A), suggesting extensive interactions might connect R233 at TM4 to L114 near the SF. To test this hypothesis, we examined the interaction partners of R233 and L114 during the R2330 and R233+ simulations.

Figure 3

Interaction network among R233, PH1, and SF determines the pHo sensitivity of TALK1. A, the diagrams represent conformational change between normal and flipped R233+, and the white and orange spheres represent K+. B–E, summary of the interaction strengths between critical residues from R233+ and R2330 simulations. Ten independent sets of 200-ns MD simulations were performed: replicate 1 to 10 for subunit one and replicate 11 to 20 for subunit two. B, R233/S97 hydrogen-bond interaction (in %). C, R233/L114 distance (in Å) van der Waals interaction energy (in kcal/mol) between F100 and L114 (D) or Y111 and L114 (E). The cumulative frequency distributions for each interaction were analyzed by the Kolmogorov–Smirnov test. F, the Heat map illustrates the Pearson correlation coefficient between various parameters including R233/S97 H-bond, R233/L114 distance, L114/F100 and L114/Y111 van der Waals interaction, ΨY111 angle, and S1 flipped/non-flipped in the R233+ and R2330 simulations (∗∗∗p < 0.001, Pearson correlation analysis). The flipped and non-flipped S1 were defined as 1 and -1, respectively. The correlation coefficients were presented as the absolute value, and the negative correlation was labeled with (-). Since we did not observe the flipped S1 in R2330 simulations, the correlation coefficients of S1 flipped/non-flipped in R2330 with other factors cannot be determined. G, the violin plots of the CαΨ angle (Y111) when the R233/L114 distance was set at 5.3 Å. Ten sets of 100-ns MD simulations were conducted; each column represents one replicate of MD simulations for individual subunits (1–10 for subunit 1 and 11–20 for subunit2). The black bars represent the average angles during the whole simulation. H, the current densities at 0 mV of COS7 cells transfected with TALK1 S97G, F100I, or L114P mutants at different pHo (∗p< 0.05, 2-way ANOVA). I, the normalized pHo-sensitive curves for S97G, F100I, and L114P mutants. The results for TALK1-WT (dash line) are from Figure 1D. PH, pore helix; SF, selectivity filter.

The positively charged R233+ side chain did not have a stable partner during the R233+ simulations; its most stable hydrogen bond was with the S97 side chain, which occurred in only 20 to 30% of the total simulation time (Fig. 3B). In the presence of the R233+/S97 hydrogen bond, R233+ rotated toward the L114 side chain (R233+/L114 distance ∼ 5.5 − 6.5 Å in Figs. 3C and S2A, Pearson coefficient = −0.628, Fig. 3F). We noticed that sometimes, but not always, the shortening of the R233+/L114 distance preceded the ΨY111 rotation (Fig. S2A, flipped; replicate 5, 6, 7,10 and 14 in non-flipped), implying the R233+-induced ΨY111 rotation is a stochastic process. Moreover, we also observed stronger van der Waals interactions for L114/F100 (-1.5 to -1.0 kcal/mol, Figs. 3D and S2A) and L114/Y111 (−2.0 to −2.5, Figs. 3E and S2A) in the closed conformation (S1 flipped). These stable van der Waals interactions of L114/F100 and L114/Y111 only occurred after the ΨY111 shifted to ∼1000 (Fig. S2A, flipped), suggesting a strong coupling between Y111/L114/F100 and the ΨY111 rotation (Fig. 3F). In contrast, R2330 occasionally formed a hydrogen bond with the backbone of V220 (occupancy of 37%) but the interactions with S97 became unstable (occupancy of 5%) (Figs. 3B and S2B). Most importantly, in all R2330 simulation replicates, R233 never came in close contact with L114 (R233+/L114 distance > 6.5 Å, Figs. 3C and S2B), and the F100-L114-Y111 interactions that linked the SF to R233 became infrequent (Figs. 3, D and E, S2B). Overall, these observations suggest that multiple transient interactions connect R233+ to the conformational changes of the S1 binding site.

To further establish a dynamic link between R233+ and the SF, we carried out a series of simulations in which the R233-Cζ and L114-Cγ distance was constrained to the shortest observed in the unconstrained simulations (R233(Cζ)---L114(Cγ) = 5.3 Å). In agreement with our hypothesis, we observed more frequent rotation of ΨY111 in these simulations (13/20, 65%) as compared with unconstrained R233+ simulations (Fig. 2A, top left and 3G; 4/20, p < 0.01, Fisher’s exact test).

To confirm the predicted role of the R223–PH1-L114 interaction in channel activation, we generated S97G, F100I, and L114P mutants and examined their pHo sensitivities. As expected, TALK1-S97G had an attenuated activation at high pHo and an acidic-shifted pK1/2 (Fig. 3, H and I). TALK1-F100I exhibited weakened proton inhibition with elevated current densities (Fig. 3, H and I). The electrophysiological recording of TALK1-L114P mutants revealed a disruption of pHo sensitivity and a 15-fold increase in current density (Fig. 3, H and I), in agreement with the previous study (29).

R233E reverses pHo sensitivity by establishing PH1 and M1P1 loop interaction networks in the presence of divalent cations

In contrast to positively charged Arg, Glu is a negatively charged at physiological pH; therefore, we predicted that the R233E mutation would exhibit different sensitivity to pHo. Indeed, TALK1-R233E showed an acidic activation (Fig. 1G). Based on the modeled structure of R233E, we then hypothesized that residues R233E, D94, and S97 could form a binding pocket for divalent cations (dications) such as Ca2+ and Mg2+, which would inhibit TALK1-R233E currents in alkaline conditions (pHo > 5.0), mimicking R233+ (Fig. 4A). To test this hypothesis, we recorded TALK1-WT and TALK1-R233E currents in a dication-free extracellular solution (Ca2+ and Mg2+-free plus 1 mM EDTA, pH 7.4). This dication-free solution at pHo 7.4 could activate TALK1 R233E (fold increase =16.23 ± 9.15, n = 9) but not WT (fold increase =1.12 ± 0.84, n = 6) (Fig. 4, B and C). Furthermore, although both Ca2+ and Mg2+ could inhibit TALK1-R233E, Ca2+ was twice as potent as that of Mg2+ at the same concentration (1 mM) (Fig. S4, A–E).

Figure 4

Dications inhibit TALK-R233E channels by mimicking interaction networks between R233 and PH1.A, the diagram illustrates the dication binding pocket formed by E233, D94, and S97 and the interacting partners of R233E with or without the Ca2+. B, sample traces of TALK1 WT and R233E recorded with or without dications. C, the normalized current amplitudes for TALK1-WT (n = 6) or R233E (n = 9) (∗∗p < 0.01, Student’s test). D, the current ratios of I pH 5.0/I pH 7.4 for TALK1-R233E (n = 6), D94A (n = 8), or R233E D94A double mutants (n = 7). (∗∗p < 0.01, One-way-ANOVA). PH, pore helix.

To further test our hypothesis, we examined the acidic activation of a double mutation, R233E + D94A. If D94 is involved in the acidic activation of the R233E mutant, this double mutant could abolish this acidic activation. We first confirmed that D94A single mutant exhibited normal pHo sensitivity (Fig. S4F). As expected, acidic activation was lost in TALK1-R233E + D94A (Fig. 4D). This result further suggested that R233E and D94 could form a dication-binding pocket. We then tested the ability of these acidic residues to capture a Ca2+ by creating three different starting conformations for MD simulations of the TALK1-R233E mutant in which a Ca2+ ion was placed close to either E233 or D94. These simulations revealed that Ca2+ tightly bound the carboxylates of E233 and D94. S97 could also coordinate Ca2+, but the interaction was more sporadic (Fig. S4G). Therefore, inhibition of TALK1-R233E at high pHo might result from a mimicking of the R233+–PH1 interaction.

Allosteric pHo-gating model for TALK1

Together, the electrophysiological experiments and MD simulations suggested an allosteric pHo-gating model for TALK1 (Fig. 5A). We proposed that the protomeric TALK1 can adopt various conformations, including R2330 (P1), R233+ with normal SF (P2), and R233+ with flipped SF (C). Both P1 and P2 represent K+-permeable conformations, whereas the C represents a nonconductive conformation. At low pHo, the R2330 (P1) will be protonated (R233+). KDn represents the proton dissociation constant of R233, and n denotes proton sensitivity. Because R233+ does not always induce a conformational change in the SF, we included a stochastic component α, an equilibrium constant between normal and flipped SF. Based on our model (Fig. 5A), the K+ permeation probability (Pp) of protomer can be calculated bywhere [H+] denotes the concentration of protons. Since a functional TALK1 channel is a dimer, the pHo-dependent Po relative to pHo 10 can be calculated by

Figure 5

The pHo-dependent allosteric gating model for TALK1.A, the gating models for the promoter and the dimer. P1 and P2 represent the K+ permeable forms; C represents the nonconductive form (S1 K+-binding site flipped). The open and closed states of a dimer follow a binomial distribution. B–D, the current ratio plots show experimental (dot) and predicted (line) pHo-sensitivities of various heterodimers, including (B) WT-R233A, (C) R233E-R233A, and (D) WT-R233E. The results of (B) WT and (C) R233E were fitted with Equations 2 and 3, respectively. The red line represents predicted pHo activation of WT-R233A and R233E-R233A heterodimeric channels by fitting Equations 1 and 5, respectively, whereas the red dots are the experimental pHo activation data for WT-R233A and R233E-R233A heterodimeric channels. D, the blue line represents predicted pHo activation of WT-R233E heterodimeric channels by Equation 6, and the blue dots show the experimental pHo activation data. TALK, TWIK-related alkaline pH-activated K+ channel.

Here, we assume that the protonation of R233 in one subunit is independent of protonation in the other subunit due to the distance between residues.

We then refitted the pHo sensitivity curve of TALK1-WT with Equation 2 (Fig. 5B) and obtained values for the parameters α, pKD, and n (Table 1). We found a mild acidic shift of pKD compared to pK1/2 estimated by the Hill equation. Also, the value of n was slightly smaller than that of nH. Of note, the value of α would yield a residual current around 15% at the extreme low pHo. This value is considered to be the TALK1 residual current that has been reported previously (33).

Table 1

Parameters of pHo-dependent allosteric gating model for TALK1 WT and R233E

Parameters	WT	R233E
α	1.64	4.27
pKD	6.18	6.01
n	0.36	0.88

All parameter values were acquired by fitting the data from Figure 1, D and G with Equations 2 and 3, respectively. The fittings were conducted using R software.

Next, we established a similar model for TALK1-R233E (Fig. S5A). Its pHo-dependent open probability (Po(E)) relative to pHo 5 can be calculated bywhere KD(E)n represents the proton dissociation constant of R233E, denotes the equilibrium constant between normal and flipped SF when R233E is deprotonated (R233E-), and n(E) denotes the proton sensitivity. It is worth noting that the value of was dependent on the presence or absence of dications. Because glutamate has a negatively charged side chain, Equation 3 has minor differences from Equation 2. Refitting the pHo sensitivities of TALK1-R233E with Equation 3 (Fig. 5C), we found that the residual current of TALK1-R233E dropped to only 7.9%.

The pHo sensitivities of heterodimeric TALK1 R233 mutants can be predicted by the allosteric pHo-gating model

A key feature of the allosteric pHo-gating model is independent gating for each protomer. Thus, we hypothesized that the pHo sensitivities of heterodimeric TALK1 R233 mutants, for example, WT-R233A, R233E-R233A, and WT-R233, could be predicted by the product of the Pp for each subunit, that is, . The former term represents the pHo-dependent PP of subunit 1, and the latter term represents the pHo-dependent PP of subunit 2. The pHo-dependent PP of TALK1-WT is given by Equation 1. Since TALK1-R233A is pHo-insensitive, its pHo-dependent PP is set to 1. Furthermore, the pHo-dependent PP of TALK1-R233E is

As an example, the pHo sensitivity of heterodimeric TALK1-R233E-R233A relative to pHo 5 could be predicted by

To test this model, we designed several concatenated TALK1 mutant plasmids that express a single functional channel comprised of two different linked subunits. All the concatenated homodimeric channels, including WT-WT, R233A-R233A, and R233E-R233E showed pHo sensitivities corresponding to those of dimers formed from plasmid-expressed single sequences (Fig. S6, A–C), suggesting the short linker did not affect the pHo sensitivities. Since the order of coding sequences did not affect pHo sensitivities (Fig. S6, D–F), we pooled the data from channels with the same combination (regardless of order) and used WT-R233A, WT-R233E, and R233E-R233A to represent three types of heterodimeric TALK1 channels.

Because TALK1 R233A is pHo-insensitive, the dimeric channels containing one R233A subunit had the pHo sensitivities dominated by the other subunit. Therefore, the pHo sensitivities of WT-R233A and R233E-R233A could be predicted by Equations 1 and 5, respectively (Fig. 5, B and C). All parameter values used for predictions are listed in Table 1. Note that Equation 1 is normalized to pHo 10, and Equation 5 is normalized to pHo 5. These equations predict the augmentation of residual currents and shifting of pK1/2. Indeed, the experimental results for WT-R233A and R233E-R233A corresponded well with these predictions (Fig. 5, B and C).

According to our hypothesis, the pH sensitivities of heterodimeric WT-R233E can be predicted by

For display as a relative pHo sensitivity, Equation 6 was normalized to Po (max) using the parameters listed in Table 1. This equation predicts a U-shaped pHo sensitivity curve for heterodimeric TALK1 WT-R233E, which corresponded well with our experimental results (Fig. 5D).

Discussion

TALK1 is an alkaline-activated K2P channel, and its activation is known to result from augmentation of open probability rather than single-channel conductance (33). Our results showed that the primary pHo sensor of TALK1 is R233, and its alkaline activation can be modeled as an allosteric process. Moreover, we found that protonation of R233 impacts the C-type gate via an interaction network between R233, PH1, and SF. Our results imply that an interaction network between PH1 and SF supports a pore conformation that allows K+ permeation. In an alkaline environment, deprotonated R2330 does not perturb this network and an open channel conformation is maintained. In an acidic environment, however, protonated R233+ could disturb the network of interactions among TM4, PH1, and SF (Fig. 3A) that destabilize the SF. Often, this disruption can induce the rotation of ΨY111 at the S1 binding site, causing Y111 to interact with L114, rendering the channel nonconductive.

The MD simulations suggested that R233, S97, and F100 are the critical residues for sensing pHo. These residues form a pocket located between the extracellular space and the lipid bilayer, which is far away from SF. This pocket has been reported to serve as a detector for multiple physiological stimuli and regulates the C-type gates in other K2P channels; for example, it is involved in the activation of TREK1 and TREK2 by ML335, a K2P channel opener (22, 34), proton-induced inhibition of TASK2 and TALK2 (20, 28), and membrane-tension-sensing of TRAAK (14). Hence, this motif is considered as a common C-type gate regulation center for K2P channels. Our studies on TALK1-R233E also support the importance of this motif. A dication-mediated interaction between R233E and D94 inhibits channel activity, analogous to the interaction between R233+ and PH1. Upon acidification or chelation of the divalent cations, this interaction is eliminated, and the TALK1-R233E channel returns to an open conformation. Interestingly, this inhibition appears to depend on the size of the divalent cation, as Ca2+ (ionic radius, 100 PM) is twice as potent as Mg2+ (ionic radius, 72 PM) (Fig. S4, A–E).

The flipping model we proposed for the TALK1 channel has not yet been observed in other K2P channels. Nevertheless, this model has been suggested for other potassium channels. In KcsA channels, the C-type inactivation is induced by disrupting the S1 and S3 potassium binding sites via flipping the carbonyl groups lining the SF. The glutamate at position 71 (E71) at the PH of KcsA channels serves as a pillar to support the SF by forming an extensive network of interactions with the neighboring residues (5, 6, 35). Mutating E71 to other noncharged residues disrupts this network and locks the carbonyl groups in place to prevent C-type inactivation (5, 6, 35). Solving the structures at various pHo may provide high-resolution mechanistic insights into the C-type gating in TALK1 channels.

Heterodimeric K2P channels with different subtypes have been found in several tissues. These chimeric channels could exhibit the properties of the original subtype K2P channels in a hybrid activation pattern, such as those seen for heterodimeric TREK1-TREK2, TASK1-TASK3, THIK1-THIK2, TALK1-TASK2, and TALK1-TALK2 (23, 25, 36, 37, 38). In our model, the parameter α represents the ratio of conformational change in the C-type gate to the conformational change in the physiological sensor. Because the calculated value of α accurately predicts the behaviors of heterodimeric TALK1 channels, we suspect that this stochastic conformational change in sensor and C-type gate can explain the hybrid activation properties of heterodimeric K2P channels formed from different subtypes.

The Hill equation is widely used to describe how a macromolecule responds to ligands (39, 40). In this equation, the parameter K1/2 represents the ligand concentration at which half response occurs. In the pHo-gated channel, this parameter can be used to define the pHo sensitivity (12). The equations we derived to describe the TALK1 kinetics are similar to the Hill equation; the relationship is given byfor Equation 1 andfor Equation 2. According to these relationships, the K1/2 of Hill equation could be greatly affected by α, whether the channel has one or two pHo-sensitive subunits. This implies that despite the same amino acid serving as the pHo sensor in K2P channels of different subtypes, the gating range might vary according to α. Moreover, the parameter nH represents the cooperativity in ligand binding. When it is equal to 1, all receptors act independently, but when it is not equal to 1, the binding of a ligand to the first receptor subunit could positively or negatively affect the binding affinity of other receptor subunits (40). In our model, the parameter n is equivalent to nH. Theoretically, fitting the data with Equation 2 of the Hill equation suggested almost identical values for n and nH. The n value of TALK1-WT is lower than 1, indicating a negatively cooperative binding.

In conclusion, our study on the pHo-gating mechanism of TALK1 not only expands the understanding of how physiological stimuli regulate C-type gate in K2P channels but also provides a functional model to explain the complex behaviors of the chimeric K2P channels. This gating model could offer a crucial molecular basis for understanding the actions of WT and disease-causing mutant K2P channels (29, 41, 42).

Experimental procedures

Plasmids

The pCDNA 3.1-human WT TALK1 variant 3 (NM_001135106) expression plasmid was a gift from Dr David A. Jacobson (Vanderbilt University). The pEXO-human TASK2 (NM_003740) was a gift from Dr Delphine Bichet (CNRS). Site-directed mutagenesis of the target plasmids was performed using the Quikchange strategy; primers were designed using the tool provided by Agilent Technologies (http://www.genomics.agilent.com/primerDesignProgram.jsp). The primers used to create TALK1 and TASK2 mutants are listed in Table S1. The sequences of the mutated clones were analyzed by an in-house sequencing facility (Sequencing Core, Institute of Biomedical Sciences, Academia Sinica). For experiments requiring heterodimeric or homodimeric tandem TALK1 mutant channels, we modified a method from a previous study (24). Briefly, the coding sequences of the first subunit (without stop codon) and second subunit were amplified by PCR. An amino acid linker Gly-Thr-Ala (GGTACCGCT) containing a KpnI restriction enzyme cutting site was added between the first and second sequences, and the whole sequence was ligated into the pCDNA 3.0 vector between the KpnI and XhoI restriction sites. The heterodimeric or homodimeric tandem channels were named as follows: subunit 1 and subunit 2, for example, WT (subunit 1)- WT (subunit 2).

Cell culture

COS-7 cells were maintained in Dulbecco's modified Eagle's medium (Gibco, 11966-025) supplemented with 10% fetal bovine serum in a humidified atmosphere of 5% CO2 at 37 °C. COS-7 cells were cotransfected with pN1-EGFP plasmid and target plasmid in a 1:9 ratio using PolyJet (SignaGen Laboratories) transfection reagent. One day after transfection, GFP-expressing cells were selected for patch-clamp recording.

Electrophysiological recording

An Axon700B amplifier was used for patch-clamp experiments. The standard external solution contained 148 mM NaCl (Sigma, 31434), 2 mM KCl (Sigma, P3911), 1 mM MgCl2 (Sigma, M2670), and 2 mM CaCl2 (Sigma, C5080); pH was adjusted with NaOH or HCl. Various pH buffers were used to maintain the pH values: 10 mM 2-(N-morpholino)ethanesulfonic acid (Sigma, M3671) for pH 5.0, 2-[4-(2-hydroxyethyl)piperazin-1-yl]ethanesulfonic acid (Sigma, H4034) for pH 6.0 to 8.0, and N-(1,1-Dimethyl-2-hydroxyethyl)-3-amino-2-hydroxypropanesulfonic acid (Sigma, A6659) for pH larger than 9.0. The dication-free external solutions were made by omitting MgCl2 and CaCl2 and adding 1 mM EDTA to the standard external solution. The Ca2+ or Mg2+ external solutions were modifications of standard external solution with dications of only 1 mM CaCl2 or 1 mM MgCl2, respectively. Data were acquired at 10 kHz using the pCLAMP software (MolecularDevice). The electrodes were pulled from 1.5-mm borosilicate glass capillaries (Sutter Inc). Pipette resistances were 2 to 5 MΩ when filled with the intracellular solution containing 150 mM KCl, 10 mM 2-[4-(2-hydroxyethyl)piperazin-1-yl]ethanesulfonic acid, 5 mM EGTA, and 1 mM MgCl2; pH was adjusted to 7.4 with KOH. The access resistances of whole-cell recording ranged between 5 and 20 MΩ and were compensated by 40 to 80%. All experiments were performed at room temperature (∼25 °C).

Modeling and MD simulations

The template structures of TASK2 at low (PDB ID: 6wlv) and high pH (PDB ID: 6wm0) were used for de novo structure predictions of WT TALK1 and various TALK1 mutants with Modeller 9.17 (43). The OPM database (44) was used to position the models respective to the membrane. The protonation states of the ionizable side chains respectively at pH 6.5 and 8.5 were determined using PROPKA3 (45). Missing hydrogen atoms were added using the HBUILD module in the CHARMM program (46) with the CHARMM36 forcefield (47). Each complex was then inserted into a POPC bilayer built using CHARMM-GUI (48). Once solvated with TIP3P (49) water molecules, the resulting system had a dimension of 100 × 100 × 127 Å (∼120,000 atoms), with 117 and 128 lipids in each leaflet, respectively. MD simulations were performed at a temperature of 310 K and 1 atm pressure using NAMD2.12 (50). All bonds to hydrogen atoms were constrained by the SHAKE algorithm (51). Long-range electrostatic forces were treated using the particle mesh Ewald method (52), with a grid spacing of 1 Å and a nonbond cutoff of 12 Å. First, the membrane-inserted system complex was minimized using 10,000 steps of conjugated gradient. Next, the system was equilibrated for 6 runs of 1 ns each, during which harmonic restraints on backbone and side chains were progressively lifted. An integration step of 1 fs was used in the first 3 runs, whereas 2 fs was used in the remaining rounds. For each complex built (dimer form), 10 independent replicates (replicate 1–10 for subunit 1 and replicate 11–20 for subunit 2) of a duration of 200 ns (unless stated otherwise) were generated (Table S2). Hydrogen bonds were analyzed using the CHARMM program (46) with an acceptor to hydrogen distance of ≤ 2.4 Å and a donor–hydrogen–acceptor angle of ≥ 130°. The simulation structures were visualized using Pymol.

Statistics

Electrophysiology Data were analyzed with pClamp10 software (Molecular Devices Corp.). Results are reported as means ± sd. Statistical analysis was performed using Prism 9 (Graph Pad), with differences considered significant at p < 0.05 (∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001). To calculate the effects of pH0, the current measured at 0 mV in whole-cell recording was plotted against pHo and fitted with the Hill equation: I = Imin + (Imax - Imin)/(1 + ([H+]/K1/2)nH) or the TALK1 kinetic model (see Results and Fig. 5). To calculate the voltage activating time constant (τ), the current measured at a specific membrane potential was plotted against time and fitted with the equation, f(t) = A (1 - e(-t/τ)) + C. The fittings were made using Prism 9, pClamp10 or R software.

Data availability

All data are contained within the article and supporting information, or are available from the authors: Wen-Hao Tsai (howard1024@ibms.sinica.edu.tw) and Cédric Grauffel (cedric@ibms.sinica.edu.tw) upon request.

Supporting information

This article contains supporting information, Figs. S1–S6, Tables S1 and S2.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.