Pore dilatation increases the bicarbonate permeability of CFTR, ANO1 and glycine receptor anion channels

Abstract

Key points

• Cellular stimuli can modulate the ion selectivity of some anion channels, such as CFTR, ANO1 and the glycine receptor (GlyR), by changing pore size.

• Ion selectivity of CFTR, ANO1 and GlyR is critically affected by the electric permittivity and diameter of the channel pore.

• Pore size change affects the energy barriers of ion dehydration as well as that of size‐exclusion of anion permeation.

• Pore dilatation increases the bicarbonate permeability ($P_{\mathrm{HC}{\mathrm{O}}_3/\mathrm{Cl}}$) of CFTR, ANO1 and GlyR.

• Dynamic change in $P_{\mathrm{HC}{\mathrm{O}}_3/\mathrm{Cl}}$ may mediate many physiological and pathological processes.

Abstract

Chloride (Cl⁻) and bicarbonate (HCO₃ ⁻) are two major anions and their permeation through anion channels plays essential roles in our body. However, the mechanism of ion selection by the anion channels is largely unknown. Here, we provide evidence that pore dilatation increases the bicarbonate permeability ($P_{\mathrm{HC}{\mathrm{O}}_3/\mathrm{Cl}}$) of anion channels by reducing energy barriers of size‐exclusion and ion dehydration of HCO₃ ⁻ permeation. Molecular, physiological and computational analyses of major anion channels, such as cystic fibrosis transmembrane conductance regulator (CFTR), anoctamin‐1(ANO1/TMEM16A) and the glycine receptor (GlyR), revealed that the ion selectivity of anion channels is basically determined by the electric permittivity and diameter of the pore. Importantly, cellular stimuli dynamically modulate the anion selectivity of CFTR and ANO1 by changing the pore size. In addition, pore dilatation by a mutation in the pore‐lining region alters the anion selectivity of GlyR. Changes in pore size affected not only the energy barriers of size exclusion but that of ion dehydration by altering the electric permittivity of water‐filled cavity in the pore. The dynamic increase in $P_{\mathrm{HC}{\mathrm{O}}_3/\mathrm{Cl}}$ by pore dilatation may have many physiological and pathophysiological implications ranging from epithelial HCO₃ ⁻ secretion to neuronal excitation.

Key points

• Cellular stimuli can modulate the ion selectivity of some anion channels, such as CFTR, ANO1 and the glycine receptor (GlyR), by changing pore size.

• Ion selectivity of CFTR, ANO1 and GlyR is critically affected by the electric permittivity and diameter of the channel pore.

• Pore size change affects the energy barriers of ion dehydration as well as that of size‐exclusion of anion permeation.

• Pore dilatation increases the bicarbonate permeability ($P_{\mathrm{HC}{\mathrm{O}}_3/\mathrm{Cl}}$) of CFTR, ANO1 and GlyR.

• Dynamic change in $P_{\mathrm{HC}{\mathrm{O}}_3/\mathrm{Cl}}$ may mediate many physiological and pathological processes.

Abbreviations

ANO1

anoctamin‐1

a₀

threshold ion diameter

Best1

bestrophin‐1

CFTR

cystic fibrosis transmembrane conductance regulator

ε

dielectric constant

GHK

Goldman–Hodgkin–Katz

GLRA1

human α1 GlyR

GluCl

glutamate‐gated chloride channel

GlyR

glycine receptor

I–V

current–voltage

MD

molecular dynamics

PMF

potential of mean force

SPAK

Ste20‐related proline and alanine rich protein kinase

TMEM16A

transmembrane member 16A

TM2

pore‐lining second transmembrane segment

TRPV1

transient receptor potential vanilloid 1

WNK1

WNK lysine deficient protein kinase 1

Anion channels are a component of cells essential for keeping them alive and mediating diverse functions. Because Cl⁻ is the most abundant anion in the body, anion channels are frequently referred to as Cl⁻ channels, but many anion channels are also permeable to various anions other than Cl⁻. In general, large halide ions, such as I⁻ and Br⁻, permeate anion channels more readily than Cl⁻ (Fatima‐Shad & Barry, 1993; Smith et al. 1999; Qu & Hartzell, 2000). This phenomenon can be explained by the small hydration/dehydration energy of large symmetrically charged ions, because ions in general pass through the channel after dehydration. Accordingly, it has been suggested that the pore of anion channels is composed of a large polarizable tunnel, where ion selectivity is determined by the electric permittivity (Smith et al. 1999; Qu & Hartzell, 2000). Based on this concept, a fixed value of the dielectric constant (ε, relative permittivity) in the channel filter has been considered the specific characteristic of a given anion channel that determines the ion selectivity (Born, 1920; Smith et al. 1999; Qu & Hartzell, 2000). However, recent studies suggest that ion selectivity of anion channels is not fixed and can be altered under specific cellular conditions (Park et al. 2010; Jung et al. 2013). Furthermore, in contrast to the polarizable tunnel theory, some anion channels, such as cystic fibrosis transmembrane conductance regulator (CFTR), are relatively impermeable to the large halide ion I⁻ (Sheppard et al. 1993).

Although many anions can permeate through anion channels, Cl⁻ and HCO₃ ⁻ are the two most abundant anions that can be the charge carriers of anion channels in animal cells. Increasing evidence indicates that HCO₃ ⁻ permeation though the anion channel is involved in many biological processes ranging from epithelial fluid secretion to neuronal excitation (Kaila et al. 1989; Lee et al. 2012; LaRusch et al. 2014). As a major component of the CO₂/HCO₃ ⁻ buffer system, HCO₃ ⁻ controls the cytosolic pH and guards against toxic pH fluctuations in extracellular fluids (Roos & Boron, 1981). Epithelial cells in respiratory, gastrointestinal and genitourinary systems secrete HCO₃ ⁻‐containing fluids, such as saliva, pancreatic juice, intestinal fluids, airway surface fluid and fluids secreted by reproductive organs. HCO₃ ⁻ is an essential ingredient in these fluids and plays critical roles. For example, HCO₃ ⁻ in pancreatic juice and duodenal fluids neutralizes gastric acid and provides an optimal pH environment for digestive enzymes to function properly in the duodenum (Lee & Muallem, 2008). Furthermore, HCO₃ ⁻ is a moderate chaotropic ion that facilitates the solubilization of macromolecules such as mucins (Hatefi & Hanstein, 1969). Inadequate HCO₃ ⁻ secretion in the epithelium leads to altered mucin hydration and solubilization (Quinton, 2010), resulting in hyper‐viscous mucus that blocks ductal structures of the lung and pancreas (Quinton, 2001, 2008). Consequently, aberrant HCO₃ ⁻ secretion is associated with a wide spectrum of diseases in the respiratory, gastrointestinal and genitourinary systems, including cystic fibrosis, pancreatitis and infertility (Wang et al. 2003; Quinton, 2008; Gee et al. 2011; Lee et al. 2012). In addition, ion permeation through synaptic anion channels such as the glycine receptor (GlyR) and GABA_A receptor plays a pivotal role in inhibitory neurotransmission in the central nervous system. Evidence suggests that HCO₃ ⁻ permeation through these synaptic anion channels is involved in the regulation of neuronal excitability (Kaila et al. 1989; Staley et al. 1995).

Therefore, the HCO₃ ⁻/Cl⁻ permeability ratio ($P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl), which determines the amount of HCO₃ ⁻ permeation through anion channels, is an important parameter of anion channel function. Of interest, $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl of anion channels can be modulated dynamically by multiple cellular stimuli. For example, activation of the WNK lysine deficient protein kinase 1 (WNK1)/Ste20‐related proline alanine rich kinase (SPAK) and Ca²⁺–calmodulin, respectively, increases $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl of CFTR and anoctamin‐1 (ANO1) (Park et al. 2010; Jung et al. 2013). In the present study, we aimed to identify the basic mechanism of $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl modulation by cellular stimuli and to further explore the underlying physicochemical principles of ion permeation through major anion channels, such as CFTR, ANO1 and GlyR, using an integrated molecular, physiological and computational approach.

Methods

Plasmids, siRNA and cell culture

The mammalian expressible plasmids for rWNK1, mSPAK, hCFTR and hANO1/transmembrane member 16A (TMEM16A) were previously described (Park et al. 2010; Jung et al. 2013). For the expression of hGlyR, cDNA of hGLRA1 was purchased from GE Dharmacon (Lafayette, CO, USA; Clone ID: 30915298) and subcloned into a pCMV‐myc plasmid using PCR amplification. The hGlyR P‐2’Δ mutant plasmid was generated using a PCR‐based site‐directed mutagenesis. The siRNAs against WNK1 and SPAK were purchased from GE Dharmacon (WNK1, SMARTpool L‐005362‐02‐0005; SPAK, SMARTpool L‐050614‐00‐0005). HEK 293T cells were cultured in Dulbecco's modified Eagle's medium (DMEM)‐HG (Invitrogen) supplemented with 10% (v/v) fetal bovine serum and 100 U ml^–1 penicillin and 0.1 mg ml^–1 streptomycin. Plasmids were transiently transfected into cells using Lipofectamine Plus (Invitrogen). siRNAs were transiently transfected into HEK 293T cells using Lipofectamine 2000 (Invitrogen). An average transfection rate over 90% was confirmed by transfection with a plasmid expressing green fluorescent protein.

Immunoblotting

Immunoblotting was performed as described previously (Park et al. 2010). Transfected HEK 293T cells were washed three times with ice‐cold phosphate‐buffered saline (PBS) and harvested. The protein samples were recovered in a sodium dodecyl sulfate (SDS) buffer and separated by SDS‐polyacrylamide gel electrophoresis. The separated proteins were transferred to a nitrocellulose membrane and blotted with appropriate primary and secondary antibodies. Protein bands were detected by enhanced chemiluminescence (GE Healthcare, Little Chalfont, UK). Antibodies against SPAK (no. 2281; Cell Signaling Technology, Danvers, MA, USA), WNK1 (ab53151; Abcam, Cambridge, MA, USA) and actin (sc‐1616; Santa Cruz Biotechnology, Dallas, TX, USA) were obtained from commercial sources.

Electrophysiology in cultured cells

Anion channel activities were measured in HEK 293T cells using the whole‐cell and excised patch clamp techniques as reported previously (Park et al. 2010; Jung et al. 2013). Briefly, cells were transferred into a bath mounted on a stage with an inverted microscope (IX‐70, Olympus). The conventional whole‐cell clamp was achieved by rupturing the patch membrane after forming a gigaseal. In the outside‐out excised patch clamp experiments, the electrode was slowly withdrawn from the whole‐cell state, allowing a bulb of the membrane to bleb out from the cell. When the electrode is pulled away by a sufficient distance, this bleb will detach from the cell and reform as a convex membrane on the end of the electrode. The bath solution was perfused at 5 ml min^–1. The voltage and current recordings were performed at room temperature (22–25°C). Patch pipettes with a free‐tip resistance of approximately 2–5 MΩ were connected to the head stage of a patch‐clamp amplifier (Axopatch‐700B, Molecular Devices, Sunnyvale, CA, USA). pCLAMP software v. 10.2 and Digidata‐1440A (Molecular Devices) were used to acquire data and apply command pulses. AgCl reference electrodes were connected to the bath via a 1.5% agar bridge containing 3 m KCl solution. Voltage and current traces were stored and analysed using Clampfit v. 10.2 and Origin v. 8.0 (OriginLab Corp., Northampton, MA, USA). Currents were sampled at 5 kHz. All data were low‐pass filtered at 1 kHz.

The standard pipette solution for whole‐cell and outside‐out patch clamp contained (in mm): 148 N‐methyl‐d‐glucamine‐Cl (NMDG‐Cl), 1 MgCl₂, 3 MgATP, 10 Hepes and 10 ethylene glycol tetraacetic acid (EGTA) (pH 7.2). The low Cl⁻‐containing pipette solution contained (in mm) 140 NMDG‐gluconate, 8 HCl, 5 EGTA, 1 MgCl2, 3 Mg‐ATP and 10 Hepes (pH 7.2). The bath solution contained (in mm) 146 NMGD‐Cl, 1 CaCl₂, 1 MgCl₂, 5 glucose and 10 Hepes (pH 7.4). For permeability measurements, 150 Cl⁻ in the bath solution was replaced with 146 X⁻ + 4 Cl⁻, where X is the substitute anion (I⁻, Br⁻, F⁻, NO₃ ⁻, HCO₃ ⁻, or gluconate⁻ in a form of NaX). When F⁻ currents were too small to measure E _rev, bath solutions containing an increased concentration of Cl⁻ (30 Cl⁻ and 120 F⁻) were used. For the anion permeability test, individual data were corrected by measuring the offset potential shift induced by the replacement of anion solution after each experiment (Park et al. 2010). The 146 mm HCO₃ ⁻‐containing solutions were continuously gassed with 95% O₂ and 5% CO₂ (pH 8.2), which is comparable to ionic compositions of human pancreatic juice. In some control experiments, the 146 mm HCO₃ ⁻‐containing solutions were gassed with 30% CO₂ and pH adjusted to 7.4. For the current measurements of ANO1, the free Ca²⁺ concentrations of buffer solutions were achieved by adjusting the Ca²⁺ chelator EGTA (10 mm) and CaCl₂ concentrations using WEBMAX‐C software (http://www.stanford.edu/∼cpatton/maxc.html). Because the Ca²⁺ chelating power of EGTA is weakened in the 1 μm and greater free Ca²⁺ range, the 3 μm free Ca²⁺‐containing solutions were buffered with the low‐affinity Ca²⁺ chelator dibromo‐BAPTA (5 mm). The osmolarity of the bath solution was made 10 mosmol l⁻¹ higher than the pipette solution's osmolarity by adding sorbitol to suppress volume‐activated anion channels. CFTR currents were activated by cAMP (5 μm forskolin and 100 μm 3‐isobutyl‐1‐methylxanthine (IBMX)) and WNK1/SPAK was activated by low [Cl⁻]_i (10 mm). Experiments in cells with WNK1/SPAK only indicated that the activation of WNK1/SPAK does not elicit discernible CFTR‐independent Cl⁻ currents in HEK 293T cells. Cation permeability (P _Na/P _Cl and P _NMDG/P _Cl) of CFTR (P _Na/P _Cl = 0.08 and P _NMDG/P _Cl = 0.07) and ANO1 (Jung et al. 2013) was very low and thus the P _X/P _Cl values were not corrected for the cation permeability. However, P _Na/P _Cl of the pore‐dilated GlyR mutant was considerably high (0.24 ± 0.03, see Fig. 8 B). Therefore, all experiments with GlyR were performed with Na⁺ salts and the Na⁺ permeability was corrected using the Goldman–Hodgkin–Katz (GHK) equation for each P _X/P _Cl calculation.

Figure 8. Characterization of wild‐type and P‐2′Δ GlyR currents .

Whole‐cell recordings were performed in HEK 293T cells expressing homomeric human α1 GlyR (hGLRA1). A, dose–response relationships of the WT and P‐2′Δ GlyR currents were analysed. Cells were transfected with WT and P‐2′Δ pCMV‐myc‐hGLRA1, and currents were activated by the application of glycine in the bath solution. Note that the P‐2′Δ mutation induced a right shift in the glycine dose–response curve. B, the WT and P‐2′Δ GlyR currents were measured in bath solutions containing various concentrations of NaCl (12, 25, 50, 75, 150 mm). Ion permeability was determined by the E _rev shift using the Goldman–Hodgkin–Katz equation. The black line represents expected E _rev values when the cation permeability of GlyR is 0, and the red and blue lines are the fitted lines of the GlyR WT and P‐2′Δ data points, respectively. The P _Na/P _Cl values of WT and P‐2′Δ GlyRs are 0.07 and 0.24, respectively. Values are the mean ± SEM (n = 5 at each point). C and D, the bicarbonate permeability ($P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl) of WT (C) and P‐2′Δ GlyRs (D) was examined using whole‐cell patch clamp recordings. Currents were activated with 10 μm (WT) and 1 mm (P‐2′Δ) glycine. The initial bath solution containing 150 mm Cl⁻ was replaced with a solution containing 4 mm Cl⁻ and 146 mm HCO₃ ⁻.

Calculation of relative ion permeability (P _X/P _Cl), dielectric constant (ε) and pore size

Current reversal potential (E _rev) was measured in zero‐current clamp experiments. To determine the current–voltage (I–V) relationship during the zero‐current clamp recordings, clamp mode was shifted to the voltage clamp mode and the I–V curve was obtained by applying ramp pulses from −100 to 100 mV (0.8 mV ms⁻¹; holding potential, near the resting membrane potential (RMP)) or step pulses from −100 to 100 mV (voltage interval, 20 mV; duration, 0.5 s; holding potential, near the RMP). The relative anion permeability was determined by the reversal potential shift (ΔE _rev = E _rev(X) – E _rev(Cl)) induced by replacing extracellular Cl⁻ with X⁻ anion using the GHK equation as follows: P _X/P _Cl = (exp(ΔE _rev/(RT/zF)) − ([Cl⁻]_o/[Cl⁻]′_o)) × ([Cl⁻]′_o/[X⁻]_o), where [Cl⁻]′_o is the bath concentration of Cl⁻, [Cl⁻]_o is the residual Cl⁻ in the substituted solution, [X⁻]_o is the concentration of substitute ion and R, T, z and F have their conventional thermodynamic meanings.

The size of dehydrated ions was estimated as reported previously (Linsdell et al. 1997; Smith et al. 1999). Briefly, the geometric mean of the two smallest dimensions for each ion was used for the size calculation, because the longest dimension of an ion would have little influence on its permeability compared with the two smaller dimensions (Linsdell et al. 1997). The dielectric constant (ε) of anion channels was calculated according to the electrostatic model of Born (Born, 1920; Smith et al. 1999). Briefly, the hydration energy of each anion can be calculated by the following equation: ΔG _hyd = −(K/2) × (1/r) × (1 − 1/ε_w), where K is a constant equal to 138.6 kJ nm mol⁻¹, r is the radius of the anion and ε_w is the dielectric constant of water (ε_w = 80). Lattimer correction of the equivalent radius of each ion was also applied to optimize anion–water interaction energy in the dielectric constant calculation (Smith et al. 1999). Δ(ΔG)_barrier for anions can be calculated from the following equation: P _X/P _A = exp[−Δ(ΔG)_barrier/RT]. A⁻, the largest anion examined in the present study (NO₃ ⁻ for CFTR, I⁻ for all other anion channels), was used as the reference anion. Using the Δ(ΔG)_barrier of each anion, the solvation energy (ΔG _sol) of the ion channel can be calculated by the following equation (Qu & Hartzell, 2000): |ΔG _sol| = |ΔG _hyd| − |Δ(ΔG)_barrier|. From the ΔG _sol value, the permittivity ε of the anion channel was calculated using the following equation (Born, 1920): ΔG _sol = −(K/2) × (1/r) × (1 − 1/ε).

Pore size of anion channels was determined by the excluded‐area model (partition coefficient model) using the P _X/P _Cl values of large non‐symmetrically charged anions (Dwyer et al. 1980; Cohen et al. 1992; Linsdell et al. 1997). According to the excluded‐area model, ionic permeation is proportional to the area of the narrowest region of pore left unoccupied by the ion. P _X/P _Cl is then given by the following equation: P _X/P _Cl = k(D _P – D _X)², where D _P and D _X are the respective diameters of the pore and anion X⁻, and k is a proportionality constant. Because the longest dimension of the ion did not affect its effective permeation, the geometric mean of the two smallest dimensions for each ion was used for D _X (Linsdell et al. 1997; Smith et al. 1999).

Intracellular pH measurement during whole‐cell recording

Intracellular pH was measured with the fluorescent pH probe dextran‐bis‐carboxyethyl‐carboxyfluorescein (dextran‐BCECF, 100 μm) included in the pipette solution. A stable fluorescence signal of pH_i was measured within 5 min after establishing the whole‐cell configuration. The low Cl⁻‐containing pipette solution contained (in mm): 140 N‐methyl d‐glucamine gluconate (NMDG‐gluconate), 8 HCl, 5 EGTA, 1 MgCl₂, 3 Mg‐ATP and 10 Hepes (pH 7.2). The bath solution contained (in mm): 146 NMDG‐Cl, 1 CaCl₂, 1 MgCl₂, 5 glucose and 10 Hepes (pH 7.4). The high HCO₃ ⁻‐containing bath solution was prepared by replacing 146 mm NMDG‐Cl with an equimolar concentration of choline bicarbonate. For activation of CFTR, cells were perfused with solutions containing forskolin (5 μm) and IBMX (100 μm). BCECF fluorescence was recorded at the excitation wavelengths of 490 and 440 nm at a resolution of 2 s⁻¹ using a recording set‐up (DeltaRam; Photon Technology International Inc. (PTI), Birmingham, NJ, USA). The 490/440 ratios were calibrated by standard pipette solutions at pH 6.5, 7.2 and 7.8.

Molecular dynamics (MD) simulations of GlyR structure

Preparation of simulation systems

Homology models of human α1 GlyR (GLRA1) WT and P‐2’Δ mutant were constructed using MODELLER v9.3 (https://salilab.org/modeller/) based on the crystal structure of an open anion‐selective GluCl (PDB: 3RIF). The sequence of α1 GlyR was edited to remove the large intracellular loop connecting TM3 and TM4 since the GluCl template does not have this homologous loop. Sequence alignments were generated using ClustalW and are shown in Fig. 11 A. For wild‐type and P‐2’Δ mutant GlyR, sequence alignment with GluCl is the same except at the position P250 (P‐2’). For each receptor channel, 100 homology models were built and the model with the lowest ‘MODELLER objective function’ score was selected to further MD relaxations. The MD simulation system was set up using the Visual Molecular Dynamics (VMD) analysis and visualization software (Humphrey et al. 1996). The TM domain of the receptor channel was inserted into the centre of a cylinder of pre‐equilibrated 1‐palmitoyl‐2‐oleoyl‐sn‐glycero‐3‐phosphocholine (POPC) lipid mixture. The outer radii of the cylinder POPC lipid construct was approximately 52.3 Å. Fully equilibrated TIP3 waters were added to the system to form a hexagonal boundary condition of 104.6 × 104.6 × 150 Å. Na⁺ and Cl⁻ ions corresponding to a 0.15 m solution were added to neutralize the system. There were one receptor channel, 196 POPC, 88 Cl⁻ ions, 83 Na⁺ ions and about 29,100 water molecules for a total of over 141,000 atoms.

Figure 11. Molecular dynamics (MD) simulations of the GlyR pore .

A, sequence alignment of wild‐type glycine receptor channel (GlyR) and C. elegans glutamate‐gated chloride channel (GluCl). The predicted sequence identity is 45%. Sequence alignment for P‐2′Δ GlyR mutant is the same except that P at position 250 (shown in yellow star) is replaced by ‘–’. B, channel pores inside the TM domains are depicted in green (2 Å < r ≤ 3 Å, radius) and blue (r > 3 Å, radius). The pore‐lining residues of TM2 are shown with their position numbers. C, the P‐2′Δ mutation enlarged the GlyR functional pore. Black and red lines represent the pore radii averaged over the last 10 ns MD simulations of GlyR wild‐type and P‐2′Δ, respectively. Dotted lines show the standard deviations of the calculations. The TM pore radius of the open anion‐selective GluCl (PDB: 3RIF) is depicted by the grey line as a reference. D and E, GlyR P‐2′Δ mutation altered the orientation of the pore lining TM2 helices, contributing to the enlargement of selectivity filter region. D, left, top view of five TM2 helices. Radial and lateral directions for calculating the tilting angles are shown. D, right, side view of the TM2 subunit radial angle in GlyRs. (WT: black cylinder; P‐2′Δ: red transparent cylinder.) Perpendicular arrow points to the channel z direction. Decreasing TM2 radial tilting enlarges the selectivity filter in the GlyR P‐2′Δ mutant. E, histograms of TM2 radial tilting angles and TM2 lateral tilting angles in GlyR wild‐type (black) and P‐2′Δ mutant (red). A total of 5000 structures and bins of 0.1 deg were used in each histogram analysis, which contains independent MD runs from 10 to 50 ns.

Molecular dynamics simulations

The CHARMM36 force field with CMAP corrections was used for protein, water and lipid simulations (MacKerell et al. 1998, 2004; Klauda et al. 2010). MD simulations were performed using the NAMD (version 2.9) program (Phillips et al. 2005). Two independent MD runs were performed for the WT GlyR (WT_1 and WT_2) and its P‐2’Δ mutant (PΔ_1 and PΔ_2), following the same simulation procedure. The system was first energy minimized for 50,000 steps. It then underwent a 0.5 ns constant volume and temperature (T = 310 K) (NVT) simulation and subsequent 4 ns constant pressure and temperature (NPT) simulation, during which the protein was fixed and the constraint on the POPC head groups was gradually released to zero. Subsequently, the constraint on the protein backbone was gradually reduced from 10 kcal mol^–1 to zero within 4 ns. Finally, the unconstrained protein underwent NPT simulation for 46 ns. The simulation protocol included periodic boundary conditions, water wrapping, hydrogen atoms constrained via the SHAKE algorithm and long‐range electrostatic forces evaluated via the Particle Mesh Ewald (PME) algorithm (Darden et al. 1993). Bonded interactions and short‐range non‐bonded interactions were calculated every time step (2 fs) and every two time steps (4 fs), respectively. Electrostatic interactions were calculated at every four time steps. The cut‐off distance for non‐bonded interactions was 12 Å. A smoothing function was employed for the van der Waals interactions at a distance of 10 Å. The pair‐list of the non‐bonded interactions was calculated every 20 time steps with a pair list distance cut‐off of 13.5 Å.

VMD (Humphrey et al. 1996) was used to analyse structural and dynamical features of the system, such as the root‐mean‐square deviation (RMSD), average pore radii and TM2 helical tilting angles. For RMSD calculations, α‐carbons of the entire receptor channel from each MD snapshot were first aligned with those of the initial model channel, following a standard approach. For structural averaging calculations, each MD snapshot was aligned with the initial model before the determination of an average structure. Radii of the model channels were calculated using the HOLE program (Smart et al. 1996). The average channel pore radii were calculated based on 500 snapshots over the last 10 ns MD simulations, during which the systems were well‐equilibrated with almost flat backbone RMSDs. The number of water molecules inside the selectivity region was obtained by counting water molecules inside the pore between −2’ and 2’ (−50 Å ≤ z ≤ −40 Å). Histogram analysis was performed after the system was well equilibrated (40 ns to 50 ns).

Halide ion parameterization in the CHARMM format

Non‐bonded Lennard–Jones (LJ) parameters for Na⁺ and Cl⁻ are directly taken from Chemistry at HARvard Macromolecular Mechanics (CHARMM) General Force (CGenFF) field (Vanommeslaeghe et al. 2010). LJ parameters for halide ions F⁻ and I⁻ were adapted from CGenFF (Vanommeslaeghe et al. 2010) and further refined to satisfy experimentally observed ion solvation properties in water solution, such as radial distribution function (RDF) of ion and oxygen atoms from water (R _o) and number of coordinated water molecules. Calibrations of LJ parameters for halide ions were carried out using MD simulations. For each system, one Na⁺ and one halide ion were placed in a 40 Å³ cubic water box. For each calibration, 6 ns standard Nosé–Hoover constant pressure (Nosé, 1984; Hoover, 1985) (P = 1 bar) and temperature (T = 310 K) (NPT) simulation was carried out using the NAMD program (Phillips et al. 2005). Table 1 lists the force field for ions.

Table 1.

Lennard–Jones (LJ) and electrostatic parameters used in the simulations

				R o (Å)
Ions	Charge (e)	ε1 (kcal mol−1)	R min/2 (Å)	MD	Expa
Na+	+1	–0.047	1.36	2.35	2.3–2.4
Cl−	–1	–0.150	2.27	3.15	3.1–3.2
F−	–1	–0.135	1.85	2.65	2.6–2.7
I−	–1	–0.215	2.65	3.55	3.5–3.6

ε₁ and R _min are depth of the LJ potential well and lowest energy interaction distance, respectively. R _o is the distance between the ion (column 1) and the closest oxygen atom on the first hydration shell. ^aValues obtained in the present MD simulations are compared with experimental results shown in Dang (2002).

MD calculations of energy barriers for halide ion transport through the charge selectivity filter region of GlyR

The energy barriers for halide ion (F⁻, Cl⁻, or I⁻) transport through the charge selectivity filter region were evaluated by MD calculations of the single ion potential of mean force (PMF). The PMF calculation was carried out using the adaptive biasing force (ABF) method (Chipot & Hénin, 2005) implemented in the NAMD software (Phillips et al. 2005) and following the previous approach for similar calculations in other family members of pentameric ligand‑gated ion channels (pLGICs) (Cheng et al. 2010; Cheng & Coalson, 2012). Briefly, the PMF calculation started from the intracellular bulk solution and covered the entire region of the charge selectivity filter, which was subdivided into approximately four to six different windows along the channel z axis (perpendicular to the membrane lipids). The width of each ABF window was 5 Å and 5–10 consecutive 1 ns ABF calculations were performed for each window until the variation of the PMF at any point along the z‐axis was less than 1 kJ mol^–1 within two consecutive runs. Calculations for different halide ions followed the same procedure and were carried out in the same initial configuration (20 ns MD equilibrated) except that only the target ion (transport ion) was replaced by the specific halide ion (I⁻, Cl⁻ or F⁻). A total of over 100 ns and 140 ns ABF calculations (for all three halide ions) were performed for the GlyR WT and P‐2’Δ mutant, respectively. Overall, the calculated PMFs converged, as indicated by the < 2 kJ mol^–1 variation in PMF along the z‐axis between two consecutive runs for all tested halide ions. All halide ions were assumed to have the same free energy value (adopted as reference of 0 kJ mol^–1) near the intracellular entrance to the TM domain because this portion was already in the internal solution and the PMF does not change significantly if integrated further into the internal solution region (Cheng et al. 2010; Cheng & Coalson, 2012). Thus, estimated energy barrier for ion transport through the selectivity filter region provides an upper bound for the free energy change (ΔG _barrier). The results are shown in Figs 11 and 12 and Table 2.

Figure 12. Pore dilatation increases ε of GlyR selectivity filter .

A, histograms depicting water inside the selectivity filter region of the equilibrated GlyR wild‐type (grey) and P‐2′Δ mutant (red). The water content inside the selectivity filter region is increased by 25% due to the P‐2′Δ mutation. B and C, adaptive biasing force (ABF) calculations of the single ion potential of mean force (PMF) for ion transport through the selectivity filter region. Comparisons of the single ion PMF for transporting an I⁻ (blue), Cl⁻ (red), or F⁻ (green) ion through the selectivity filter region of GlyR WT (B) and GlyR P‐2′Δ mutant (C) are shown. For each halide ion (I⁻, Cl⁻ or F⁻), ABF calculations of the PMF started from the intracellular entrance and were carried out in four (WT GlyR) and six (GlyR P‐2′Δ mutant) windows along the ion transport direction (the channel z axis). Each window had a width of 5 Å, and 5–10 consecutive 1 ns ABF calculations were performed for each window. For all three halide ions, over 100 ns and 140 ns of ABF calculations were performed for WT GlyR and GlyR P‐2′Δ mutant, respectively. D and E, The P‐2′Δ mutation increased ε in the selectivity filter. Data sets of Δ(ΔG _barrier) between different halide ions presented in Table 2 (D, MD simulation) and Fig. 7 C (E, whole‐cell recording) were used to calculate ε, using the biggest ion, I⁻, as a reference.

Table 2.

Comparison of energy barriers for ion transport across the selectivity filter region deduced from MD calculations

	GlyR WT		GlyR P‐2′Δ mutant
	ΔG barrier (kJ mol−1)	Δ(ΔG barrier)a (kJ mol−1)	ΔG barrier (kJ mol−1)	Δ(ΔG barrier)a (kJ mol−1)
F−	14.2 ± 2.5	8.9 ± 2.5	5.8 ± 1.5	3.0 ± 1.5
Cl−	6.8 ± 1.0	1.5 ± 1.0	3.8 ± 1.5	1.0 ± 1.5
I−	5.3 ± 1.0	0	2.8 ± 1.5	0
Na+	21.2 ± 4.0		12.8 ± 2.0

^aI⁻ was used as the reference anion.

Modelling of P _X/P _Cl using free energy changes of anions and pore diameter of anion channels

Computation of anion free energy changes

We employed the polarizable continuum model (PCM) and the density functional theory (DFT) to compute the free energy change of solvation of anions in a dielectric medium. First, the structures of various anions in the gas phase and in aqueous environment were optimized using DFT at the level of B3LYP/6‐311++G(d,p). The integral equation formalism PCM (IEF‐PCM) for solvents was then used to calculate the solvation and hydration free energies (ΔG _solv and ΔG _hyd) of the anions (Mennucci et al. 2002; Tomasi et al. 2005). Due to computational complexity of ions with too many orbitals, I⁻ was excluded from the free energy computation. All free energy calculations were carried out using the GAUSSIAN03 package (Gaussian, Inc., Wallingford, CT, USA).

Computation of pore size effect

To account for the size‐dependent effect, we employed a partition coefficient model. Finite pore size reduces the channel permeability when large charged solutes travel through ion channels; however, models describing the pore size effect are limited. Here, we used a modified version of the partition coefficient model, which describes the ion channel as a soft‐walled cylinder and the permeating ion as a sphere, in order to represent the effect of pore size on permeability. According to the conventional partition coefficient model, ion permeability is reduced in proportion to the ratio of the diameters of permeating ion and the pore (i.e. $\lambda =a/d$, where a is the diameter of the ion and d is the diameter of the channel pore) by an excluded area effect. Therefore, as solute particles pass through semipermeable membranes, solute permeability can be reflected by the pore size according to $P_{\mathrm{size}}={\left[1-\lambda \right]}^2$ (Linsdell et al. 1997). However, because large pores are unlikely to inhibit the transfer of small spherical ions after a certain level, the conventional partition coefficient model overestimates the inverse size dependence of permeability for small anions. In addition, recent studies have shown that ion channels are rather flexible (Heads et al. 2008; Wieczorek & Zielenkiewicz, 2008). Therefore, in this work we used a shifted partition coefficient model to incorporate the flexibility of ion channel pore upon ion permeation as

$$P_{\mathrm{size}}={\left[1-\frac{\lambda -{\lambda }_0}{1-{\lambda }_0}\right]}^2={\left[1-\frac{a-a_0}{d-a_0}\right]}^2$$ (1)

where$\lambda =a/d$, ${\lambda }_0=a_0/d$ with a being the diameter of the ion, d being the diameter of channel pore and a ₀ being the threshold anion diameter. As a result, P _size of ions with a diameter larger than a ₀ is reduced, whereas P _size of ions smaller than a ₀ remains constant. The threshold ion diameter (a ₀) and the channel pore size are characteristic of each channel and were determined by fitting the experimental values.

Modelling of P _X/P _Cl

To evaluate the relative permeability (P _X/P _Cl) of anion X with respect to Cl⁻, thermodynamic and size effects were taken into account as

$$P_{\mathrm{X}}/P_{\mathrm{Cl}}={P_{\Delta G}}^{\alpha }\times P_{\mathrm{size}}$$ (2)

where P _{Δ G} is the P determined by the free energy of anions, α is a channel‐specific weight factor and P _size is the P determined by the pore diameter of anion channels.

The thermodynamic transfer free energy contribution was evaluated as the following

$$P_{\Delta G}=\exp \left(-\Delta \Delta G_{\mathrm{trans}}/RT\right)$$ (3)

where R is the gas constant and T is temperature. The free energy change of transfer is defined as $\Delta G_{\mathit{trans}}=\Delta G_{\mathit{solv}}-\Delta G_{\mathit{hyd}}$, and $\Delta \Delta G_{\mathit{trans}}$ is the difference between the free energy of transfers of Cl⁻ and X.

A channel‐specific weight factor α was incorporated into the model as a power to P _{Δ G} to describe the relative contribution between thermodynamic and size effects. The threshold diameter and channel factor were estimated via the non‐linear least square method using the nls function of stats package implemented in R software (v. 3.0.2, freeware).

Statistical analysis

The results of multiple experiments are presented as means ± SEM. Statistical analysis was performed with Student's t test or with analysis of variance followed by Tukey's multiple comparison test, as appropriate. P < 0.05 was considered statistically significant.

Results

$P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl increases in CFTR and ANO1 are associated with increased dielectric constant (ε) and pore size

Previous study on the Ca²⁺–calmodulin‐induced regulation of ANO1 shows that Ca²⁺–calmodulin affects the permeability ratio (P _X/P _Cl) of other anions as well as HCO₃ ⁻ and changes the ε of the hypothetical ANO1 selectivity filter (Jung et al. 2013). Therefore, we first examined whether a similar phenomenon occurs in the WNK1/SPAK‐induced regulation of CFTR. The relative anion permeability (P _X/P _Cl) was measured using a whole‐cell patch recording of the human CFTR expressed in HEK 293T cells with or without activated WNK1/SPAK (Figs 1, 2, 3, 4, 5, 6). P _X/P _Cl was determined by the current‐clamped reversal potential shift (ΔE _rev = E _rev(X) – E _rev(Cl)) induced by replacing extracellular Cl⁻ with X⁻ anion using the Goldman–Hodgkin–Katz (GHK) equation. As reported earlier (Park et al. 2010), the CFTR $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl was greatly increased by the low [Cl⁻]_i‐induced WNK1/SPAK activation (Fig. 1 E and 2 E). Expression of inactivated WNK1/SPAK in cells with a high [Cl⁻]_i (150 mm) did not increase $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl (Fig. 3). On the other hand, the low [Cl⁻]_i (10 mm) in CFTR only expressing HEK 293T cells, which may have endogenous WNK1/SPAK, evoked a partial increase in $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl (Fig. 4); however, this was abolished upon depletion of WNK1/SPAK by siRNA treatments (Fig. 5). Collectively, the above results indicate that activation of WNK1/SPAK is responsible for the low [Cl⁻]_i‐induced increase in CFTR $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl. In addition, control experiments of measuring pH_i and whole‐cell currents at various pH_i and pH_o confirmed that alterations in [HCO₃ ⁻]_i or pH were not responsible for the WNK1/SPAK‐induced increase in $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl (Fig. 6).

Figure 1. Anion selectivity of CFTR in control cells .

CFTR was expressed in HEK 293T cells, and whole‐cell currents were measured. CFTR currents were activated by cAMP (5 μm forskolin and 100 μm IBMX) after whole‐cell configuration was established. The initial bath solution containing 150 mm Cl⁻ was replaced with a solution containing 4 mm Cl⁻ and 146 mm X⁻, with X⁻ representing the substitute anion. To determine the current–voltage (I–V) relationship during zero‐current clamp recordings, clamp mode was shifted to the voltage clamp mode and the I–V curve was obtained by applying ramp pulses from −100 to 100 mV (0.8 mV ms⁻¹; holding potential, near the resting membrane potential). A–E, representative voltage and current measurements are shown. F, a summary of the P _X/P _Cl values from zero‐current clamp recordings.

Figure 2. WNK1/SPAK activation alters the halide ion permeability of CFTR .

CFTR was expressed in HEK 293T cells, and whole‐cell currents were measured. CFTR currents were activated by cAMP (5 μm forskolin and 100 μm IBMX) after whole‐cell configuration was established. The initial bath solution containing 150 mm Cl⁻ was replaced with a solution containing 4 mm Cl⁻ and 146 mm X⁻, with X⁻ representing the substitute anion. To determine the current–voltage (I–V) relationship during zero‐current clamp recordings, clamp mode was shifted to the voltage clamp mode and the I–V curve was obtained by applying ramp pulses from −100 to 100 mV (0.8 mV ms⁻¹; holding potential, near the resting membrane potential). CFTR anion permeability (P _X/P _Cl) was analysed in WNK1/SPAK‐activated cells. A–E, representative voltage and current measurements. F, a summary of the P _X/P _Cl values from zero‐current clamp recordings. Note that WNK1/SPAK activation increased P _I/P _Cl and narrowed the interval between the P _X/P _Cl values of large and small ions (e.g. compare $P_{\mathrm{N}{\mathrm{O}}_3}$/P _Cl and P _F/P _Cl in Figs 1 F and 2F).

Figure 3. Anion selectivity of CFTR in cells with inactivated WNK1/SPAK .

CFTR anion permeability (P _X/P _Cl) was analysed in WNK1/SPAK‐inactivated cells. HEK 293T cells were transfected with plasmids expressing CFTR, WNK1 and SPAK. A high concentration of Cl⁻ (150 mm) in the pipette did not activate WNK1/SPAK. To determine the I–V relationship during zero‐current clamp recordings, clamp mode was shifted to the voltage clamp mode and the I–V curve was obtained by applying ramp pulses from −100 to 100 mV (0.8 mV ms⁻¹; holding potential, near the RMP). Representative membrane potential measurements under zero‐current clamp and I–V curves of F⁻, Br⁻, NO₃ ⁻, I⁻ and HCO₃ ⁻ measurements are shown in A–E, and a summary of the P _X/P _Cl values from the zero‐current clamp recordings is shown in F.

Figure 4. Anion selectivity of CFTR at low [Cl⁻]_i in cells expressing CFTR only .

CFTR was expressed in HEK 293T cells, and whole‐cell currents were measured. Anion permeability (P _X/P _Cl) was analysed at low (10 mm) [Cl⁻]_i. To determine the I–V relationship during zero‐current clamp recordings, clamp mode was shifted to the voltage clamp mode and I–V curve was obtained by applying ramp pulses from −100 to 100 mV (0.8 mV ms⁻¹; holding potential, near the RMP). Representative membrane potential measurements under zero‐current clamp and I–V curves of F⁻, Br⁻, NO₃ ⁻, I⁻ and HCO₃ ⁻ measurements are shown in A–E, and a summary of the P _X/P _Cl values from the zero‐current clamp recordings is shown in F. The low [Cl⁻]_i (10 mm) in CFTR only expressing HEK 293T cells evoked partial alterations in P _X/P _Cl, possibly due to an activation of endogenous WNK1/SPAK.

Figure 5. Anion selectivity of CFTR in WNK1/SPAK‐depleted cells .

Whole‐cell currents were measured in HEK 293T cells after transfection with siRNAs against WNK1 and SPAK and plasmids expressing CFTR. Anion permeability (P _X/P _Cl) was analysed at low (10 mm) [Cl⁻]_i. To determine the I–V relationship during zero‐current clamp recordings, clamp mode was shifted to the voltage clamp mode and I–V curve was obtained by applying ramp pulses from −100 to 100 mV (0.8 mV ms⁻¹; holding potential, near the RMP). Representative membrane potential measurements under zero‐current clamp and I–V curves of F⁻, Br⁻, NO₃ ⁻, I⁻ and HCO₃ ⁻ measurements are shown in A–E, and a summary of the P _X/P _Cl values from the zero‐current clamp recordings is shown in F. Depletion of WNK1/SPAK abolished the low [Cl⁻]_i‐induced alterations in P _X/P _Cl. G, immunoblotting of WNK1 and SPAK. Treatment with siRNAs against WNK1 and SPAK inhibited endogeous expression of WNK1 and SPAK in HEK 293T cells by 86 ± 3% and 81 ± 9% (n = 3), respectively.

Figure 6. pH_i alterations do not affect the CFTR $P_{\mathrm{HC}{\mathrm{O}}_3}$ / P _Cl change by WNK1/SPAK activation .

A and B, measurements of pH_i during whole‐cell current recordings. HEK 293T cells were transfected with plasmids expressing CFTR, WNK1 and SPAK, and whole‐cell recordings were performed. WNK1 and SPAK was activated by low (10 mm) [Cl⁻]_i. During the whole‐cell recording, intracellular pH was monitored with the fluorescent pH probe dextran‐BCECF (100 μm) included in the pipette solution. A representative trace is shown in A, and a summary is presented in B (n = 4). During the zero‐current clamp recording, ramp pulses (−100 to +100 mV, 0.8 mV ms⁻¹) were applied (indicated by arrows 1 and 2) to obtain I–V relationships. BCECF fluorescence was recorded at excitation wavelengths of 490 and 440 nm at a resolution of 2 s⁻¹ using a recording set‐up (DeltaRam; PTI). The 490/440 ratios were calibrated using standard pipette solutions of pH 6.5, 7.2 and 7.8, respectively. Note that pH_i did not significantly change during the $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl measurements (P = 0.8399). C–F, measurements of $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl in acidic, neutral and alkaline pH_i conditions. HEK 293T cells were transfected with plasmids expressing CFTR, WNK1 and SPAK, and whole‐cell recordings were performed. The pH in the pipette solutions was adjusted to 6.9, 7.2, 7.5 and 7.8, respectively. C and D, $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl in cells with inactivated WNK1/SPAK was measured using a high Cl⁻‐containing pipette solution (150 mm). E and F, $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl in cells with activated WNK1/SPAK was measured using a low Cl⁻‐containing pipette solution (10 mm). Summaries of these experiments are presented in D and F, respectively (n ≥ 6). Note that $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl always increased in cells with activated WNK1/SPAK regardless of the pH_i changes, indicating that pH_i did not contribute to the WNK1/SPAK‐induced increase in $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl (P = 0.5004 and 0.9860 in D and F, respectively). G and H, measurements of $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl using a 146 mm HCO₃ ⁻‐containing solution gassed with 30% CO₂ (pH 7.4) in cells expressing CFTR and WNK1/SPAK. Representative voltage and current measurements are shown in G, and a summary of the $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl values from zero‐current clamp recordings is shown in H. Reduction in the pH of HCO₃ ⁻‐containing bath solution from 8.2 (gassed with 5% CO₂; in Figs 2 and 3) to pH 7.4 (gassed with 30% CO₂) did not affect the effects of WNK1/SPAK activation.

Notably, activated WNK1/SPAK also affected the P _X/P _Cl of halides and NO₃ ⁻ in addition to HCO₃ ⁻ (Fig. 2). With the exception of I⁻, the P _X/P _Cl change pattern of CFTR by activated WNK1/SPAK was similar to that observed in the Ca²⁺–calmodulin‐induced modulation of ANO1 (Jung et al. 2013) (see also Fig. 7 B). WNK1/SPAK activation narrowed the interval between the CFTR P _X/P _Cl values of large and small ions (e.g. $P_{\mathrm{N}{\mathrm{O}}_3}$/P _Cl and P _F/P _Cl in Fig. 1 F and 2 F), possibly due to an increase in the permittivity of the CFTR selectivity filter. The permittivity means how much electric field is generated in the medium and its value in the ion channel pore can be exhibited by ε calculated according to the electrostatic model of Born (Born, 1920; Smith et al. 1999). When the ε of the channel pore region is increased and closer to that of water (ε of a vacuum = 1, water = 80), the dehydration energy penalty between the large and small ions become smaller. Indeed, WNK1/SPAK activation increased the CFTR pore ε (estimated from the P _X/P _Cl values of NO₃ ⁻, Br⁻ and F⁻) from 16 to 43 (Fig. 7 A).

Figure 7. Increased $P_{\mathrm{HC}{\mathrm{O}}_3}$ / P _Cl of CFTR, ANO1 and GlyR are associated with increases in dielectric constant (ε) and pore size .

CFTR, ANO1 and GlyR were expressed in HEK 293T cells, and whole‐cell currents were measured. The x‐ and y‐axes represent the diameter and the relative permeability of each anion, respectively. The dashed lines are the fitted lines for dielectric constant estimation using the electrostatic model of Born and the continuous lines represent the fitted lines for pore size estimation using the excluded‐area model. A, the effects of WNK1/SPAK activation on the CFTR dielectric constant (ε) and pore size were analysed using the P _X/P _Cl values of symmetrically charged ions (open circle) and non‐symmetrically charged polyatomic ions (filled circle), respectively. B, the ε and pore size of ANO1 were analysed using submaximal (0.4 μm) or high (3 μm) [Ca²⁺]_i stimulation. C, the ε and pore size of wild‐type and P‐2′Δ GlyRs were analysed. The P _X/P _Cl values of halide ions and NO₃ ⁻ were from Figs 1 and 2 (A, CFTR), Jung et al. (2013) (B, ANO1) and Fig. 9 A–D (C, GlyR). A, acetate; As, aspartate; G, gluconate; Is, isethionate; M, methanesulfate; Pr, propionate; Py, pyruvate.

According to the simple dielectric tunnel theory, the P _I/P _Cl should be greater than $P_{\mathrm{N}{\mathrm{O}}_3}$/P _Cl or P_Br/P _Cl and decrease when ε increases because I⁻ is bigger than other anions (Born, 1920; Smith et al. 1999; Qu & Hartzell, 2000). Although the ANO1 P _I/P _Cl pattern followed this principle (Jung et al. 2013) (Fig. 7 B), the CFTR P _I/P _Cl values did not in our study. The P _I/P _Cl of CFTR was 0.41 ± 0.03 in the control state (Fig. 1 D and F) indicating that the I⁻ permeability of CFTR was even smaller than that of the Cl⁻ ion. Moreover, WNK1/SPAK activation increased P _I/P _Cl (Fig. 2 D and F), which was the opposite of the expected response based on the ε increase in the electrostatic model.

If the CFTR pore is not big enough to allow free passage of I⁻, the size exclusion of I⁻ in the CFTR channel pore may limit I⁻ movement. Therefore, we analysed the size of the functional CFTR pore by measuring P _X/P _Cl of large polyatomic anions. Estimation using an excluded‐area model (partition coefficient model) indicated that the size of the CFTR pore is 4.8 Å in the control state (Fig. 7 A), which is only 9% bigger than the diameter of I⁻ (4.4 Å). Importantly, WNK1/SPAK activation increased the CFTR pore size to 5.3 Å (Fig. 7 A), which is 20% bigger than the diameter of I⁻. This result may explain why the P _I/P _Cl of CFTR was increased by WNK1/SPAK activation.

Due to the small pore size of CFTR, we could only obtain measurable P _X/P _Cl values from the relatively small polyatomic ions. The high P _I/P _Cl of ANO1 (Jung et al. 2013) suggests that this channel may have a larger pore than CFTR. Therefore, we next performed the above experiments in ANO1 to better estimate changes in pore size. ANO1 is activated by elevations in [Ca²⁺]_i (Caputo et al. 2008; Schroeder et al. 2008; Yang et al. 2008). It has been shown that ANO1 is highly permeable to HCO₃ ⁻ at a high [Ca²⁺]_i (> 1 μm), but is poorly permeable to HCO₃ ⁻ at a submaximal [Ca²⁺]_i (400 nm) (Jung et al. 2013). The ANO1 pore size with 400 nm [Ca²⁺]_i was 8.0 Å (Fig. 7 B), which is approximately 80% bigger than the diameter of I⁻. Therefore, the size‐exclusion effects were likely to be minimal and the P _I/P _Cl of ANO1 was principally determined by the permittivity of the pore rather than size‐exclusion effects. Importantly, in addition to the increase in ε reported earlier (Jung et al. 2013), the 3 μm high‐[Ca²⁺]_i stimulation increased the ANO1 pore size to 8.6 Å (Fig. 7 B). Taken together, the above results indicate that the dynamic increase in $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl of CFTR and ANO1 was associated with increases in ε and the diameter of the channel pore.

Pore dilatation increases $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl and ε of GlyR

We further investigated whether other anion channels have similar characteristics to those described above and if so, how pore dilatation induces an increase in ε. We first examined the GlyR Cl⁻ channel, which mediates synaptic inhibition in the spinal cord, brain stem and several other regions of the central nervous system (Lynch, 2004). It has been shown that deletion of proline at the ‐2 position (P‐2’Δ) of the pore‐lining second transmembrane segment (TM2) increases pore size (Lee et al. 2003). Therefore, we analysed the effects of P‐2’Δ on anion selectivity using whole‐cell recordings of homomeric human α₁ GlyR (GLRA1) expressed in HEK 293T cells (Figs 7 C, 8 and 9). Because P‐2’Δ induces a right shift of the glycine dose–response curve and increases Na⁺ permeability (Fig. 8 A and B), a higher dose of glycine was used for the activation of P‐2’Δ GlyR and the Na⁺ permeability was corrected using the GHK equation for each P _X/P _Cl calculation. The pore size of wild‐type GlyR was estimated to be 5.3 Å and that of P‐2’Δ was enlarged to 7.1 Å (Fig. 7 C), which were comparable to the previously reported values of 5.4 and 6.9 Å, respectively (Lee et al. 2003). Similar to the responses observed in the WNK1/SPAK‐induced regulation of CFTR and Ca²⁺–calmodulin‐induced modulation of ANO1 (Fig. 7 A and B), pore dilatation induced by the P‐2’Δ mutation increased the $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl of GlyR from 0.13 to 0.51 (Fig. 8 C and D). Notably, the P‐2’Δ‐induced pore dilatation of GlyR evoked a narrowing of P _X/P _Cl in the whole‐cell recordings (Fig. 9 A–D) and consequently increased ε of the GlyR pore from 14 to 33 (Fig. 7 C). The ε of wild‐type GlyR has been previously calculated as 13 (Smith et al. 1999) and is comparable to the present value of 14. Furthermore, we repeated the experiments with excised cell‐free patches to avoid problems that might incur from the patch clamp recordings under bi‐ionic conditions (Fig. 9 E–H). Almost identical ionic selectivities were observed in the whole‐cell and excised patch recording configurations (Fig. 9 B vs. F, Fig. 9 D vs. H), which exclude the possibility of erroneous contributions from the bi‐ionic potential measurements (see also Fig. 10 and Discussion).

Figure 9. Pore dilatation increases ${\mathit{P}}_{\mathit{}\mathrm{HC}{\mathit{}\mathrm{O}}_{\mathit{}3}}$ / P _Cl and alters the halide ion permeability of GlyR .

P _X/P _Cl changes in response to the P‐2′Δ mutation‐induced GlyR pore dilatation were analysed. A–D, homomeric human α₁ GlyR (hGLRA1) was expressed in HEK 293T cells and whole‐cell patch recordings were performed. Wild‐type (WT, A and B) and P‐2′Δ (C and D) GlyRs were stimulated with 10 μm and 1 mm glycine, respectively. The representative reversal potential traces under zero‐current clamp are shown in A and C, and summaries of the P _X/P _Cl values from multiple experiments are shown in B and D. E–H, homomeric human α₁ GlyR (hGLRA1) was expressed in HEK 293T cells and outside‐out patch recordings were performed. WT (E and F) and P‐2′Δ (G and H) GlyRs were stimulated with 10 μm and 1 mm glycine, respectively. The representative reversal potential traces under zero‐current clamp are shown in E and G, and summaries of the P _X/P _Cl values from multiple experiments are shown in F and H.

Figure 10. Comparison of P _X / P _Cl values obtained from zero‐current clamping and from I–V curves of voltage clamping .

P _X/P _Cl values determined by the reversal potential shift obtained from voltage recording under zero‐current clamp (I = 0), in which ion depletion/accumulation will be minimal, were compared with those obtained from the I–V curves recorded under voltage clamp (I–V), in which ion depletion/accumulation would be maximized. CFTR and GlyRs were expressed in HEK 293T cells, and whole‐cell currents were measured. The initial bath solution containing 150 mm Cl⁻ was replaced with a solution containing 4 mm Cl⁻ and 146 mm X⁻, with X⁻ representing the substitute anion. To determine the current–voltage (I–V) relationship during zero‐current clamp recordings (I = 0), clamp mode was shifted to the voltage clamp mode and I–V curve was obtained by applying ramp pulses from −100 to 100 mV (0.8 mV ms⁻¹). A and B, CFTR anion permeability (P _X/P _Cl) was analysed in control (CFTR only) and WNK1/SPAK‐activated cells. Panels A and B are related to Fig. 1 F and 2 F, respectively. C and D, whole‐cell recordings were performed in cells expressing wild‐type and P‐2′Δ GlyRs, and P _X/P _Cl values were analysed. Panels C and D are related to Fig. 9 B and D, respectively. None of the comparison pairs (I = 0 vs. I–V) showed significant differences.

Pore dilatation increases ε of GlyR in the selectivity filter region

The high homology between hGlyR and the Caenorhabditis elegans glutamate‐gated channel (GluCl), whose open‐channel structure was crystallized recently (Hibbs & Gouaux, 2011), enabled us to explore the effect of pore size on ε using structural modelling (Fig. 11 A). The molecular dynamics (MD) simulation of homopentameric GLRA1 suggests that the P‐2’Δ mutation increased the GlyR functional diameter from 5.0 to 7.0 Å (radius from 2.5 to 3.5 Å, Fig. 11 B and C) by altering the orientation of the pore lining TM2 helices (Fig. 11 D and E). These pore diameter values obtained from MD simulations were comparable to those measured from whole‐cell recordings (5.3 and 7.1 Å, respectively) (Fig. 7 C). Five P‐2’ residues line the intracellular entrance and constitute the most constricted region of the channel pore in the WT GlyR. The pore radius near P‐2’ was stabilized at ∼2.5 ± 0.2 Å in independent runs (Fig. 11 C), which is similar to the open GluCl crystal structure (Hibbs & Gouaux, 2011) and the experimentally measured minimum pore radius of GlyR (2.65 Å) (Fig. 7 C, diameter 5.3 Å). P‐2’Δ mutation causes the exposure of A‐3’ and A‐1’ to the channel lumen, which enlarges the radius of selectivity filter region more than 1 Å. The equilibrated pore radii near A‐3’/A‐1’ fluctuated around ∼3.5 ± 0.4 Å over the simulation time course

(Fig. 11 C). In conclusion, the P‐2’Δ mutation overall enlarged the selectivity filter region by ∼1 Å (radius) and slightly constricted the extracellular entrance to the TM domain pore (hydrophobic gate, Fig. 11 C).

The TM2 helices line the ion permeation pore, and the charge selectivity (Keramidas et al. 2004) and channel gating properties of pLGICs (Hilf & Dutzler, 2009) have been attributed to them. The orientation of the TM2 helix is characterized by the lateral (δ) and radial (θ) tilting angles (Fig. 11 D) (Cheng et al. 2007) and variations in these angles have been implicated in conformational changes leading to the channel opening/closure (Nury et al. 2010; Cheng & Coalson, 2012; Mowrey et al. 2013). In particular, the comparison of locally closed and open conformations of pLGIC indicates that reductions in TM2 lateral and radial tilting angles by a few degrees are enough to close the channel (Mowrey et al. 2013). Following a previously introduced approach (Cheng et al. 2007; Mowrey et al. 2013), we calculated the TM2 helical tilting angles along the radial and lateral direction. Average radial tilting angle in the open WT GlyR was 8.6 ± 1.3 deg, compared with 5.4 ± 0.8 deg in the P‐2’Δ mutant (Fig. 11 E). Average TM2 lateral angles were 2.6 ± 1.0 deg and 5.2 ± 0.7 deg in the WT and P‐2’Δ GlyR, respectively (Fig. 11 E). Our independent MD runs consistently demonstrated that the P‐2’Δ mutation induced an ∼2–4 deg inward TM2 radial tilting. This inward tilting, along with the increase in the lateral tilting by a few degrees, may contribute to the enlargement of the selectivity filter region in the open conformation of P‐2’Δ GlyR mutant.

It has been shown that water molecules in confined geometries like ion channels exhibit a space‐dependent reduction in the pore water dielectric constant (ε_w, down to 20) due to the restriction of the translational and rotational mobility of water molecules (Aguilella‐Arzo et al. 2009). The pore dilatation will relieve this restriction of water molecule movement and increase ε_w. In fact, the P‐2’Δ‐induced pore enlargement increased water occupancy in the selectivity filter region by 25% (Fig. 12 A), indicating an increase in the free space for water molecule movement. In addition, water occupancy was highly ordered and in nearly a single layer near the P‐2’ of WT GlyR, whereas the water molecules were in multiple layers and comparatively less ordered in the same region of the P‐2’Δ mutant (i.e. A‐3’). This may increase the effective local dielectric constant inside the channel pore and thus reduce the dehydration energy barrier for permeant ions. Consequently, these changes reduced energy barriers for halide ion transport (Fig. 12 B and C and Tables 1 and 2), suggesting that pore dilatation increased the permittivity of the GlyR pore. The energy barrier data for halide ion transport from MD simulations revealed that the P‐2’Δ mutation increased the estimated ε from 16 to 35 (Fig. 12 D), and these values were similar to those obtained from whole‐cell recordings (14 and 33, respectively) (Fig. 7 C and 12 E).

Prediction of anion channel ion selectivity using ε and pore size

To outline a generalized principle of ion permeation via anion channels, P _X/P _Cl values of GlyR, ANO1 and CFTR were estimated by multiplying the thermodynamic hydration energy effect (P _{Δ G}) by the pore size effect (P _size). P _{Δ G} of each ion in a dielectric medium was calculated by the integral equation formalism–polarizable continuum model (Fig. 13 A) and P _size was calculated by the shifted partition coefficient model (Fig. 13 B). The polarizable continuum model (PCM) is a widely used computational method for determining accurate and reliable solvent effects (Mennucci, 2012). Modelling the solvent as a polarizable continuum is advantageous for making ab initio computation feasible on a practical scale because the computational cost of modelling chemical reactions in condensed phases grows prohibitively high. Of the variations in the PCM models, we used the integral equation formalism (IEF) PCM because it is less sensitive to diffused solute charge distribution and is known to agree well with experimental data for solvents (Mennucci et al. 2002).

Figure 13. Modelling of ion selectivity of anion channels using the pore size and thermodynamic energy effects .

A, relationship between the solvent dielectric constant and the thermodynamic hydration energy effect on P _X/P _Cl (P _{Δ G}). Free energies of transfer ($\Delta \Delta G_{\mathit{trans}}=\Delta G_{\mathit{solv}}-\Delta G_{\mathit{hyd}}$) in a dielectric medium were computed using the polarizable continuum model and the density functional theory. In each ion, the relative free energy of transfer with respect to Cl⁻ was calculated ($\Delta \Delta G_{\mathit{trans}}=\Delta G_{\mathit{trans}}\left[\mathit{ion}\mathit{X}\right]-\Delta G_{\mathit{trans}}\left[{\mathit{Cl}}^{-}\right]$), and the value was converted to the relative ion permeability ($P_{\Delta G}=\exp (-\Delta \Delta G_{\mathit{trans}}/\mathit{RT})$). B, comparison of the (conventional) partition coefficient model and a shifted partition coefficient model at the threshold diameter a ₀ = 0.5d. Because the conventional partition coefficient (excluded area) model overestimates inverse size dependency of P _size particularly for tiny ions, threshold diameter (a ₀) was employed to limit the maximum size‐exclusion effect. The x‐axis represents the ratio between the diameters of the ion and channel pore (λ = a/d) and the y‐axis represents the pore size effect on P _X/P _Cl (P _size). P _size of ions with a diameter smaller than a ₀ is progressively increased in the conventional partition coefficient model (up to 4‐fold at a ₀ = 0.5d), whereas P _size remains constant in the shifted partition coefficient model. C–E, the P _X/P _Cl values of GlyR (C), ANO1 (D) and CFTR (E) were estimated by multiplying the hydration energy effect (P _{Δ G}) by the pore size effect (P _size). P _size was calculated by the shifted partition coefficient model. The channel‐specific factor (α) and a ₀ were determined by fitting the experimental values. The predicted values were fitted well with experimental values. Dashed line represents P _size (α = 0). Parameters used for P _X/P _Cl prediction are presented in Table 3.

In calculating P _size using the shifted partition coefficient model, we employed the threshold ion diameter (a ₀) to limit the maximum size‐exclusion effect because the conventional partition coefficient model overestimates inverse size dependency of P _size particularly for tiny ions, such as F⁻ (Fig. 13 B). a ₀ and the channel pore sizes were determined by fitting the experimental values. In the final modelling of P _X/P _Cl, we introduced a channel‐specific weight factor (α) into the model as a power to P _{Δ G} to describe the relative contribution of P _{Δ G} and P _size. Specific parameters of the dielectric constant, channel pore diameter, channel‐specific weight factor and threshold ion diameter for the modelling of P _X/P _Cl are listed in Table 3. The computed values fit well within the P _X/P _Cl values obtained from the patch clamp recordings of GlyR, ANO1 and CFTR (Fig. 13 C–E). In these figures, anions are classified into two groups. P _X/P _Cl of symmetrically charged anions, such as halides and NO₃ ⁻, is principally under the control of the dielectric constant, whereas the P _X/P _Cl of non‐symmetrically charged polyatomic anions is mainly governed by the pore size because of their large dehydration energy penalty in the anion channel dielectric medium.

Table 3.

Dielectric constant (ε), channel pore diameter (d), threshold ion diameter (a ₀) and channel‐specific weight factor (α)

Channel	ε	d (Å)	a 0(Å)	α
CFTR only, 150 mm [Cl−]i	16	4.8
			4.02	1.11
CFTR+WNK1+SPAK, 10 mm [Cl−]i	43	5.3
ANO1, 0.4 μm [Ca2+]i	23	8.0
			3.96	2.03
ANO1, 3 μm [Ca2+]i	53	8.6
GlyR WT	14	5.3
			3.68	1.30
GlyR P‐2′Δ	33	7.1

Discussion

The combined molecular and biophysical analyses in the present study indicate that change in pore size can dynamically modulate ion selectivity of anion channels by not only influencing energy barriers of size exclusion but also affecting those of ion dehydration by altering the electric permittivity of the filter. Although the change in pore size may affect the permeability of many anions, increased permeation of HCO₃ ⁻ through the enlarged pore is especially noticeable. Electrophysiological data obtained from three different anion channels indicated a common finding that an increase in $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl is associated with increases in ε and the diameter of the channel pore. When ε is below or near 20, $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl principally follows the size‐exclusion rule. However, $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl increases substantially when the pore size and ε increase, due to the reduction in the size‐exclusion force and the decrease in the solvation energy gap between Cl⁻ and HCO₃ ⁻ in a high dielectric medium.

Dynamic increase of $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl in anion channels is relevant in multiple areas of human physiology and pathophysiology. For example, HCO₃ ⁻ permeation through the synaptic anion channels has been shown to induce membrane depolarization, which may play a role in the development of early neuronal networks and also in pathological epileptogenesis (Kaila et al. 1989; Staley et al. 1995; Ben‐Ari et al. 2007). In addition, we recently demonstrated that mutations in the CFTR gene that are associated with aberrant HCO₃ ⁻ secretion in humans confer a major risk for chronic pancreatitis, chronic rhinosinusitis and male infertility (LaRusch et al. 2014). Lastly, a recent report showed that HCO₃ ⁻ secretion via ANO1/TMEM16A helped pH recovery from an acid load caused by bulk proton release from exocytosis events in the pancreatic acinar lumen, where cytosolic calcium easily reaches micromolar concentrations during physiological stimulation (Han & Thorn, 2014). Interestingly, ANO1 is co‐expressed in the apical membrane of several epithelia in which CFTR mediates HCO₃ ⁻ secretion. Therefore, identification of the ANO1 $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl‐modulating mechanism and augmentation of ANO1‐mediated HCO₃ ⁻ secretion may provide a pharmacologically suitable target for treating diseases caused by defective CFTR.

Ion selectivity of a given channel is traditionally believed to be an invariant feature. However, increasing evidence suggests that ion selectivity is dynamically altered by cellular stimuli or strong agonist stimulations. In addition to the CFTR and ANO1 previously discussed (Park et al. 2010; Jung et al. 2013), ion selectivity of the transient receptor potential vanilloid 1 (TRPV1) cation channel is altered by prolonged exposure to capsaicin, which induced a [Ca²⁺]_o‐dependent increase in the P _Ca/P _Na of TRPV1 (Chung et al. 2008). Interestingly, the capsaicin‐induced change in TPPV1 P _Ca/P _Na was also associated with pore dilatation (Chung et al. 2008), although the underlying mechanism was not clarified in the study. Moreover, recent X‐ray structures of chicken bestrophin‐1 (Best1), a Ca²⁺‐activated Cl⁻ channel, indicate that the channel pore is dilated when Ca²⁺ is bound to the Ca²⁺ clasp of the channel, thereby activating the channel (Kane Dickson et al. 2014). Taken together, the above results suggest that pore dilatation of an ion channel is not a rare phenomenon.

Measurements of the bi‐ionic potentials using whole‐cell recordings might contain erroneous results because of the series resistance and ion depletion/accumulation problems (Yu et al. 2014). To rule out these possibilities, we performed several control experiments and analyses in this study. First, for the GlyR experiments in Fig. 9, we included measurements with excised cell‐free patches, in which the series resistance and ion depletion/accumulation problems are minimal and smaller than in whole‐cell current recordings. Yet similar results were obtained in the two recording conditions, which exclude the possibility of any contribution from these problems. Next, we provided direct controls for lack of HCO₃ ⁻ depletion/accumulation during the whole‐cell recordings by showing that there has been no changes in pH_i (thus no changes in HCO₃ ⁻ concentration) during the $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl measurements (Fig. 6 A and B). Lastly, to minimize any potential series resistance and ion depletion/accumulation issues during whole‐cell recordings, zero‐current clamping mode was used to measure the membrane potentials in the present study, and these values were used for all mechanistic analyses including statistical comparisons. During zero‐current clamping, we occasionally applied ramp pulse to obtain current–voltage (I–V) curves to confirm the characteristics of measured currents. Theoretically, in zero‐current clamping there will be no potential change problems caused by series resistance (Jung & Lee, 2015). Also, we performed additional analyses comparing P _X/P _Cl values for CFTR and GlyR obtained from voltage recording under zero‐current clamp, in which ion depletion/accumulation is minimal if any, and the P _X/P _Cl values obtained from the I–V curves recorded under voltage clamp, in which ion depletion/accumulation would be maximized. As shown in Fig. 10, the results are virtually identical, thus providing further evidence for excluding influence from ion depletion/accumulation in our recording.

Structural analyses of the GlyR WT and pore‐dilated mutant (P‐2’Δ) indicated that pore dilatation increases the electric permittivity of selectivity filter. The quality of these structural models was supported by high sequence identity with the template, dynamical structure stability and statistical reproducibility (Fig. 11). The minimum channel pore radii obtained in our MD simulations were consistent with experimental measurements (Fig. 11 C), further supporting the reliability of our model channels. The stability of the simulated structures was assessed from time evolution of the overall receptor channel RMSD (based on α‐carbons) from its initial homology model structure. For both WT and the P‐2’Δ mutant, the RMSDs were stabilized at ca 2.7 ± 0.3 Å within 10 ns of MD simulations and remained nearly constant during the subsequent MD simulations, indicating that systems were well‐equilibrated after 10 ns MD relaxations. Additionally, the pentameric GLRA1 structures were maintained well and continuous water occupancy was maintained during the course of MD simulations, indicative of the opening of the channel. Notably, our multiple MD simulations resulted in a consistent structural difference between WT GlyR and its P‐2’Δ mutant, which provides a molecular basis for understanding the effects of P‐2’Δ mutation on the charge selectivity of GlyR.

Three conclusions can be drawn from these MD simulations. First, the P‐2’Δ mutation altered the channel constriction and shifted the position of the energy barrier. In the WT GlyR, the proline residue at the ‐2’ position constricted the channel pore and contributed to charge selectivity. However, the peak of the free energy occurred at the threonine occupying the 6’ position in the P‐2’Δ mutant, indicating the mutation induced a shift in charge selectivity (Fig. 12 B and C). Second, the minimum channel pore size determined the relative permeability ratio among halide ions. This is generally true if no charged residues line the selectivity filter region, as was the case in the WT GlyR and its P‐2’Δ mutant, and the energy barrier for ion transport is mainly due to the ion dehydration penalty. In the WT GlyR (minimum pore radius ∼2.5 ± 0.2 Å), partial dehydration was observed for all permeant ions. Our MD calculations showed that when the minimum pore diameter of the channel was larger than 8 Å (radius 4 Å), the difference in permeability among halide ions was almost eliminated. Third, enlarging the minimum channel pore size increased ε mainly due to an increase ε_w in the filter region. Accordingly, the difference in the free energy barrier (Δ(ΔG _barrier)) between different halide ions generally agreed with the calculations using effective dielectric constants (Fig. 12 D and F and Table 2).

In this study, we effectively quantified the ion selectivity of a given anion channel using transfer free energy and pore‐size effects. We expect that the ion selection rules described here can be applied to other anion channels that have a reasonably large‐sized pore. However, channel‐specific factors may also play a role. For example, the observed $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl values of ANO1 and CFTR were somewhat higher than those predicted by the modelling (Fig. 13 D and E). A plausible assumption is that dilatation of the channel pore in a single dimension, which creates an elliptical increase in the cross‐sectional area of the pore, would preferentially increase the HCO₃ ⁻ permeability because HCO₃ ⁻ has a planar triangular shape whereas Cl⁻ is spherical. The P _I/P _Cl of CFTR is smaller than $P_{\mathrm{N}{\mathrm{O}}_3}$/P _Cl or P_Br/P _Cl even after WNK1/SPAK activation (Fig. 2). The slit‐like elliptical pore may also explain the low permeability of CFTR to the large spherical ion I⁻, in which the smallest dimension of the diameter still limit the passage of I⁻ although the geometric mean of the CFTR pore is 20% bigger than the diameter of I⁻ under WNK1/SPAK‐activated conditions. In addition, different values of the channel‐specific weight factor (α) in each channel (Table 3) indicate that P _{Δ G} and P _size differentially affect the ion selectivity of each channel. In the present study, to identify the general mechanism of $P_{\mathrm{HC}{\mathrm{O}}_3}$/P _Cl changes, we employed a simple model of anion permeation and assumed that all the energy effects were from a single hypothetical major barrier. However, an anion channel may have accessory barriers, and effects from electrostatic interactions with specific charged side chains of amino acids in the filter region may also contribute to the individual selectivity pattern of anion channels. Future studies using high‐resolution channel structures will elucidate the mechanisms associated with channel‐specific factors.

In conclusion, we show that ion selectivity of given anion channels can be predicted by combination of the anion transfer free energy effect and pore‐size effect, and pore dilatation increases $P_{\mathrm{HC}{\mathrm{O}}_3/\mathrm{Cl}}$ of anion channels by reducing energy barriers of size‐exclusion and ion dehydration of HCO₃ ⁻ permeation. These findings help elucidate the mechanisms behind the determination of ion permeation and anion channel selectivity, and offer potential therapeutic targets to treat human diseases associated with aberrant HCO₃ ⁻ permeation.

Additional information

Competing interests

The authors declare no competing financial interests.

Author contributions

I.J., M.C. and E.S. contributed equally to this study. I.J., J.J. and Y.K. performed electrophysiological and molecular experiments and analysed the data. M.C. and I.B. performed the MD analysis. E.J., B.L.S, H.K. and K.P. performed mathematical modelling. C.H.K., J.‐H.Y., D.C.W. and M.G.L. planned the study and contributed to analysis and interpretation of the data. All authors have approved the final version of the manuscript and agree to be accountable for all aspects of the work. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Funding

This work was supported by grants 2013R1A3A2042197 (M.G.L.), 2007‐0056092 (M.G.L.) and 2014R1A1A3049671 (E.S.) from the National Research Foundation, the Ministry of Science, ICT & Future Planning, and grant HI14C0070 (M.G.L.) from the Korea Health Technology R&D Project, KHIDI, the Ministry of Health & Welfare, Republic of Korea. M.C. and I.B. gratefully acknowledge financial support by NIH P30 DA035778, 5R01 GM099738‐04 and P41 GM103712.