Identifying the molecular target sites for CFTR potentiators GLPG1837 and VX-770

Cystic fibrosis (CF), a lethal genetic disease affecting 1 in every 2,500 newborns of Caucasian heritage (Zielenski and Tsui, 1995; Rowe et al., 2005), is a channelopathy caused by malfunction of the chloride channel CFTR (Riordan et al., 1989; Gadsby et al., 2006), whose main physiological function is to transport salt and water across many epithelium-lining organs (Quinton and Reddy, 1991; Bear et al., 1992). Loss-of-function mutations in the cftr gene result in multiorgan dysfunction, including chronic lung infection and destruction, the leading cause of morbidity and mortality in CF (Rowe et al., 2005). As a member of the ATP-binding cassette superfamily, CFTR inherits two cytosolic nucleotide-binding domains (NBDs) that exploit the energy of ATP binding and hydrolysis to drive the opening and closing (gating) of an anion-selective pore constructed by its two transmembrane domains (TMDs). In addition, CFTR possesses a unique regulatory domain containing serine/threonine residues for PKA-dependent phosphorylation (Ostedgaard et al., 2001). Once phosphorylated, CFTR’s gating cycle is triggered by ATP-induced NBD dimerization and hydrolysis-elicited separation of NBD dimer (Vergani et al., 2003, 2005; Hwang and Sheppard, 2009; Hwang et al., 2018; Csanády et al., 2019).

Since the first cloning of cftr gene (Riordan et al., 1989), ∼2,000 different variants (http://www.genet.sickkids.on.ca) have been identified and the disease-associated mutations are classified into six groups based on their molecular mechanisms (Wang et al., 2014; Veit et al., 2016). Both the most prevalent mutation, deletion of phenylalanine at the 508th position (ΔF508), and the third-most common mutation, G551D, are known to impair channel gating (Dalemans et al., 1991; Hwang et al., 1997; Cai et al., 2006; Cui et al., 2006; Bompadre et al., 2007; Miki et al., 2010). Hence, tremendous efforts have been made to develop small molecules known as CFTR potentiators to ameliorate the fundamental defect in CFTR gating (Hwang et al., 1997; Hwang and Sheppard, 1999; Van Goor et al., 2009; Jih et al., 2017). In 2012, the approval of VX-770, or ivacaftor (N-(2,4-di-tert-butyl-5-hydroxyphenyl)-4-oxo-1,4-dihydroquinoline-3-carboxamide), by the U.S. Food and Drug Administration for the treatment of CF patients carrying the G551D mutation ushered in a new era of personalized CF therapeutics. In the following years, VX-770 was approved for the treatment of patients carrying a broader spectrum of CFTR mutants with gating abnormalities (Yu et al., 2012; Van Goor et al., 2014). Despite this revolutionary progress made in the past decade, developing new CFTR potentiators remains an urgent need, as VX-770 is not efficacious enough to completely eliminate the gating defect of G551D (Accurso et al., 2010; Jih and Hwang, 2013; Lin et al., 2016). In addition, VX-770 is shown to pose negative impacts in vitro on the action of VX-809 (Cholon et al., 2014; Veit et al., 2014), a CFTR corrector for mutations that cause trafficking/biogenesis defects such as ΔF508 (Van Goor et al., 2011). While the improved efficacy of second-generation correctors with VX-770 (Davies et al., 2018) and ongoing efforts in discovering new CFTR potentiators (e.g., GLPG1837 or N-(3-carbamoyl-5,5,7,7-tetramethyl-5,7-dihydro-4H-thieno[2,3-c]pyran-2-yl)-lH-pyrazole-5-carboxamide; Van der Plas et al., 2018) promise an encouraging future for CF therapy, the ability to rationally design effective compounds with improved pharmacological properties may hold the key to an ultimate cure of CF.

As VX-770 is a proven prototype of CFTR potentiators, its mechanism of action and potential binding sites may provide the essential insights for the development of next-generation CFTR potentiators. To date, however, the binding site for VX-770 remains unsettled, with several studies proposing different locations, including (1) the interface between the lipid bilayer and TMDs (Jih and Hwang, 2013; Yeh et al., 2015, 2017), (2) a region surrounded by amino acids in between the NBD1/NBD2 interface and the coupling helix 1 (or residue 167–172; Veit et al., 2014), and (3) a binding pocket close to intracellular loop 4 (Byrnes et al., 2018). On the other hand, its mechanism of action can be well described in the context of a gating scheme featuring an energetic coupling between dimerization of the NBDs and opening/closing of CFTR’s gate in TMDs (Eckford et al., 2012; Jih and Hwang, 2013). Specifically, by stabilizing the open channel conformations in the TMDs, VX-770 promotes channel opening nearly ubiquitously across a wide spectrum of CFTR mutants (Yu et al., 2012) and works synergistically with other CFTR potentiators with different mechanisms of action, including ATP analogues that target the ATP-binding sites (Aleksandrov et al., 2002; Zhou et al., 2005; Cai et al., 2006; Miki et al., 2010) and 5-nitro-2-(3-phenylpropylamino)benzoate, which appears to promote NBD dimerization (Lin et al., 2016; c.f. Csanády and Töröcsik, 2014).

Recently, we demonstrated that a newly developed CFTR potentiator, GLPG1837, shares the same mechanism of action and a common binding site with VX-770 (Yeh et al., 2017). Based on a classical allosteric modulation model, we further proposed that the two critical properties of a CFTR potentiator, affinity (or potency) and efficacy, can be mathematically defined because of energetic coupling between potentiator binding/unbinding and channel opening/closing. In short, the apparent affinity of a potentiator is determined by how tight it binds to both the open and closed channel conformations, whereas the efficacy is determined by the differences in free energy of binding between the open and closed states, regardless of the absolute value of the free energy of binding in each state. This idea of state-dependent binding demands that the structure of the binding site for a potentiator in an open state should differ from that in a closed state. Thus, in theory, one can manipulate the efficacy and affinity of the potentiator by altering its chemical interactions with the binding sites. However, to realize this ambition of structure-based drug design, one has to first identify the binding sites.

In the past three years, major breakthroughs in solving the atomic structures of zebrafish (zCFTR) and human CFTR (hCFTR) by cryo-EM reveal exquisite chemical features of this protein in different configurations (Zhang and Chen, 2016; Liu et al., 2017; Zhang et al., 2017, 2018b). Taking advantage of the detailed molecular pictures of CFTR and the extensively studied mechanism of action for CFTR potentiators, we envision the structure-based drug design a real possibility once the molecular targets for CFTR potentiators are identified. In this study, our aim is to identify the binding sites for GLPG1837 and VX-770. By combining in silico molecular docking and functional studies using the patch-clamp technique, we identified two potential binding sites, site I and site II_N, for GLPG1837 and VX770 at the interface of CFTR’s TMD1 and TMD2. The comparative contribution of these two sites to the binding of CFTR potentiators and the implication of our results for structure-based drug design will be discussed.

CFTR mutants were constructed with QuikChange XL kit (Agilent) and sequenced by the DNA Core Facility at the University of Missouri. The CFTR constructs and green fluorescence protein encoding pEGFP-C3 (Takara Bio) were carried by separate pcDNA plasmids and cotransfected with PolyFect transfection reagent (Qiagen) into Chinese hamster ovary cells for all patch-clamp experiments. After transfection, cells were incubated at 27°C for 2–6 d for patch-clamp recordings.

In all patch-clamp experiments, the patch pipettes were pulled with a two-stage micropipette puller (PP-81; Narishige) and polished with a homemade microforge to a resistance of 2–5 MΩ in the standard inside-out solution (see chemicals and solution compositions below). Transfected cells grown on a glass chip were transferred to a chamber filled with standard inside-out perfusate on the stage of an inverted microscope (IX51; Olympus) at room temperature. Membrane patches were then excised to an inside-out mode with a seal resistance >40 GΩ. The pipette tip was subsequently positioned at the outlet of a three-barrel perfusion system operated by a fast solution change device (SF-77B; Warner Instruments) with a dead time of ∼30 ms (Tsai et al., 2009). The signals were recorded with a patch-clamp amplifier (EPC9; HEKA), filtered at 100 Hz with an eight-pole Bessel filter (LPF-8; Warner Instruments) and digitized to a computer at a sampling rate of 500 Hz. The membrane potential was held at −30 mV unless otherwise indicated in the figure legends. Devices that contacted with VX-770 were washed with 50% DMSO after each recording to minimize contamination by residual VX-770 as described previously (Jih and Hwang, 2013).

In all experiments, pipette solution contains (in mM) 140 NMDG-Cl, 2 MgCl₂, 5 CaCl₂, and 10 HEPES, with pH adjusted to 7.4 by NMDG. The standard inside-out perfusate contains (in mM) 150 NMDG-Cl, 2 MgCl₂, 1 CaCl₂, 10 EGTA, and 8 Tris, pH 7.4 adjusted with NMDG.

PKA and magnesium ATP (MgATP) were purchased from Sigma-Aldrich and stocked at −20°C. The working concentrations for PKA and MgATP are 25 IU and 2 mM, respectively, unless indicated otherwise in the figures. VX-770 was provided by R. Bridges (Rosalind Franklin University, North Chicago, IL) and stored as a 100 µM stock in DMSO at −70°C. GLPG1837 was provided by Galapagos and stored as a 10 mM stock at −20°C. All chemicals were diluted with the standard inside-out perfusate, and the pH was adjusted to 7.4 with NMDG.

Considering the limited solubility of GLPG1837, we performed the dose–response experiments with a maximum concentration of 40 µM. While this limiting concentration worked well for most of our constructs, some mutations may have resulted in a more drastic shift of the dose–response relationship, and therefore, a complete dose–response curve was not obtained (e.g., Y304A- and Y304T-CFTR).

Igor Pro 7 (Wave-Metrics) was used to measure the steady-state mean current amplitude, estimate the relaxation time constant, and perform Hill equation fitting with built-in functions to determine the dose–response relationships. The current amplitudes (I) in response to GLPG1837 at different concentrations (µM) were normalized to the amplitudes at 3 µM of GLPG1837 in the same patch by the following equations:

The normalized response (y axis) was plotted against the corresponding concentrations (x axis), and Hill equation fitting was performed to estimate the half-effective concentration (EC₅₀). The base of the Hill equation was held at zero, and the maximal response predicted by the Hill equation fitting was normalized to one. Notice that this second normalization is performed for better presentation of the dose–response curves, and it does not affect the value of EC₅₀. The EC₅₀ is presented as fitted value ± one standard deviation.

For single-channel studies (Figs. S2 and S4), we decreased the amount of plasmids used for transfection, performed experiments within 2 d after transfection, and used smaller pipettes (4–5 MΩ) for patch clamping. The membrane potential was held at −30 mV throughout the experiment. CFTR channels were activated by PKA and ATP to a steady state before PKA was removed with the perfusate containing 2 mM ATP. Single-channel traces were inverted for a better visualization. Single-channel kinetic analysis was done with a program developed by Csanády (2000). To ensure the accuracy of the number of functional channels in the patch, we applied GLPG1837 in each experiment to determine the number of channels and analyzed data from patches yielding a maximum of four simultaneous opening steps for microscopic kinetics.

Student’s t test was used in Figs. 2 D and S2 B. ANOVA followed by the Dunnett test was used to compare every mean with a control mean in Fig. 6 E; Fig. 8, B and D; Fig. 9 C; and Fig. S3 B. P value < 0.05 was considered statistically significant, and the error bars represent SEM in all figures.

The structure of the phosphorylated, ATP-bound hCFTR was not available at the time when we launched our study. Two high-resolution cryo-EM structures of the whole CFTR protein were solved at the time: the structures of the dephosphorylated, ATP-free hCFTR protein (PDB accession no. 5UAK; Liu et al., 2017) and the phosphorylated, ATP-bound zCFTR protein (PDB accession no. 5W81; Zhang et al., 2017). Both structures were downloaded from the PDB (Berman et al., 2000). Thus, based on the sequence of hCFTR in 5UAK, a homology model of the phosphorylated, ATP-bound hCFTR conformation was built with Modeller (Šali and Blundell, 1993; Fiser et al., 2000; Martí-Renom et al., 2000; Webb and Sali, 2016) using the phosphorylated, ATP-bound zCFTR protein (5W81) as a template. The recommended parameters were adopted and the two bound ATP molecules in 5W81 were included in the modeling work. The output with the lowest value of the MODELLER objective function was selected as the modeled ATP-bound, phosphorylated hCFTR structure, which was used for molecular docking of the ligands, GLPG1837 and VX-770.

Docking of GLPG1837 and VX-770 onto the CFTR protein was performed through an in-house version of AutoDock Vina (Trott and Olson, 2010), which is capable of outputting a maximum of 500 docking modes (Yan et al., 2016). As compared with the original version, which only provides the top 20 docking modes, our modified version offers a broader view of the docking landscape of our targets. The modeled CFTR protein structure was prepared as described in the last section. The PDB file for the modeled hCFTR protein or the cryo-EM–solved, ATP-bound hCFTR structure was sanitized to retain only the protein structure. The 3-D atomic structures for GLPG1837 and VX-770 were built using the open-source software Avogadro Version 1.2.0 (Hanwell et al., 2012). The PDB files of the CFTR protein and the ligands were converted to the PDBQT format (Forli et al., 2016), which is specific for the AutoDock Vina. Through the conversion process, partial charges were calculated by the AutoDock Vina with the Gasteiger method and assigned to the atoms of the protein, and the nonpolar hydrogen atoms were united with the heavy atoms to which they were bonded. In docking, the protein structure was treated as a rigid body, whereas the ligands were allowed to be flexible. The docking results targeting the TMDs of the hCFTR were ranked from the best to the worst according to their docking energy scores. The more negative the energy score, the better the rank of the binding mode.

Fig. S1 provides the chemical structures and docking modes for GLPG1837 and VX-770 in the cryo-EM structure of phosphorylated, ATP-bound hCFTR. Fig. S2 presents microscopic recrodings of F312A-CFTR. Fig. S3 demonstrates the response of site IV mutants to the addition/removal of VX-770. Fig. S4 shows microscopic recordings of Y304A- and Y304F-CFTR. Figs. S5 and S6 show the effects of GLPG1837 on zCFTR and R347D/D924R hCFTR, respectively.

We recently demonstrated that the CFTR potentiators GLPG1837 and VX-770 share a common mechanism of action as well as a common binding site (Yeh et al., 2017). In this study, our goal was to identify their exact binding sites in the CFTR protein. As a first step to hunt for the potential binding modes, we performed a docking calculation of GLPG1837 by running the AutoDock Vina program on a hCFTR homology model (see details in Materials and methods). This homology model was built mainly using the template structure of the phosphorylated and ATP-bound zCFTR (Zhang et al., 2017), which represents a channel closest to an open configuration among the available atomic structures at the time when we launched this project.

Our docking was specifically focused on CFTR’s TMDs, where the binding sites for both GLPG1837 and VX-770 are presumably located (Jih and Hwang, 2013; Yeh et al., 2015, 2017). The output poses of GLPG1837 were clustered visually, and the pose exhibiting most interactions with CFTR was selected as the representative conformation for each site. These poses and the corresponding sites were numbered according to their energy scores (−8.8, −8.2, −7.8, and −7.7, respectively) and are shown in Fig. 1 A. The expanded views of each site, labeled as site I, II, III, and IV, are displayed in Fig. 1, B–E. We caution our readers that due to the limited accuracy of the existing docking scoring functions (Grinter and Zou, 2014; Gaieb et al., 2019), the small differences in these energy scores should not be used to determine the rankings of these poses; mutations and functional studies are required to verify these potential binding sites. Nonetheless, the docking methodology does provide a blunt yet valuable screening tool to preselect otherwise innumerable potential binding sites for functional studies.

A two-dimensional chemical structure of GLPG1837 can be found in Fig. S1 A. In the predicted binding sites, we first identified the amino acid residues that have any atoms within 5 Å of GLPG1837. Among these residues, we then defined the ones that have a side chain pointing toward the bound ligand and thus can potentially interact with GLPG1837 as the candidates for further examination with the patch-clamp functional assay. The predicted binding sites were then tested by first making mutations at the above-identified positions and then measuring CFTR currents at different potentiator concentrations with the patch-clamp technique to quantitatively assess the effects of mutations in each binding site on the dose–response relationships of GLPG1837.

This straightforward assay, however, is impractical for VX-770, since we have previously determined that 200 nM is already a saturating concentration for VX-770, and the actual EC₅₀ may be much lower (Jih and Hwang, 2013; but compare Hadida et al., 2014; Van Goor et al., 2014). As concentrations in the range of picomolars must be used to complete the dose–response curve for VX-770, it will take very long time to achieve a steady state. In addition, the stickiness of VX-770 to the recording device also adds another level of technical difficulty in assuring the correct concentrations used in each experiment. Therefore, we used GLPG1837, which assumes a micromolar affinity and is readily removed from the system, in all the following dose–response experiments and took a different strategy to estimate the affinity for VX-770, as described latter in this section.

The top-scoring site I consists of three amino acids (D924, N1138, and S1141) located at the interfaces between CFTR’s two TMDs (Fig. 1 B). We started with D924 to test whether site I is a potential binding site for GLPG1837. In this binding mode, the aspartate side chain appears to be hydrogen bonded with the two nitrogen atoms in the pyrazole ring of GLPG1837. We reasoned that the binding affinity of GLPG1837 may be changed by introducing mutation to disrupt the hydrogen bonding. Fig. 2 shows the effects of mutations at D924 on the apparent affinity for GLPG1837. As shown in Fig. 2 A, while the prephosphorylated WT-CFTR currents can be potentiated by GLPG1837 at a saturating concentration of 3 µM in the continuous presence of ATP, an application of 20 µM GLPG1837 further enhances the currents of D924N-CFTR, indicating a rightward shift of the dose–response relationship or an increase in EC₅₀. We next replaced the aspartate with alanine (D924A) in an attempt to observe a more drastic change in the affinity of GLPG1837. Although a steady-state dose–response relationship and a corresponding EC₅₀ cannot be attained because we could not obtain macroscopic currents for D924A-CFTR (but a representative microscopic recording is shown in Fig. 2 B), we can safely conclude that the apparent affinity of D924A-CFTR for GLPG1837 is decreased by ≥10-fold (EC₅₀ = 0.26 µM for WT-CFTR) based on two observations. First, we barely discerned any potentiating effects on D924A-CFTR at 3 µM GLPG1837, which was the saturating concentration for WT-CFTR. Second, upon changing the [GLPG1837] from 20 µM to 3 µM, the channel activity immediately decreased to the level comparable to the activity in the absence of GLPG1837, suggesting a rapid dissociation of GLPG1837 from D924A-CFTR. These two observations indicate that the EC₅₀ of D924A-CFTR should be >3 µM. Fig. 2 C summarized the dose–response curves for WT-, D924E-, and D924N-CFTR. Of note, the EC₅₀ of the charge-conserving mutant D924E-CFTR, which preserves the negative charge at this position, is slightly different from that of WT-CFTR. These side-chain–dependent incremental changes of EC₅₀ among D924E, D924N, and D924A mutations support the idea that D924 is involved in the binding of GLPG1837.

While the dose–response experiment is a straightforward assay to obtain the apparent affinity, one can also measure the apparent dissociation rate (or the off-rate) of the drug by monitoring the time course of current decay upon sudden removal of the compound. In Fig. 2 A, the current decay upon removal of 3 µM GLPG1837 can be fitted with a single exponential function, yielding a time constant of 24.4 s for WT-CFTR and 3.8 s for D924N-CFTR, respectively. The statistical comparison shown in Fig. 2 D confirms a faster dissociation of GLPG1837 (i.e., shorter relaxation time constant) from D924N-CFTR, which is consistent with a lower apparent affinity seen in the dose–response relationships (Fig. 2 C). It should be noted that the affinity of a drug is determined by both its on-rate and off-rate, and this relaxation analysis only assesses the off-rate. Despite this caveat, this measurement can be used as a complementary tool to evaluate any changes in the binding affinity of a molecule when performing dose–response experiment is not feasible, as in the case of VX-770 (see Figs. 8 and 9 below).

The next amino acid in site I is N1138 in transmembrane segment (TM) 12. Interestingly, the side chain of N1138 resides close to the hydrophobic region of GLPG1837. We thus speculated that substitutions of N1138 with hydrophobic amino acids may actually improve the binding of GLPG1837. To test this idea, we chose G551D-CFTR as the background for two reasons. First, due to its intrinsic low open probability (P_o; Bompadre et al., 2007; Lin et al., 2014), the maximal fold increase of G551D-CFTR currents in response to GLPG1837 is 20-fold greater than that of WT-CFTR (Yeh et al., 2017), which allows us to clearly observe the stepwise changes in current amplitudes at different concentrations of GLPG1837 (Fig. 3 A). Second, since the apparent affinity for GLPG1837 on G551D-CFTR is 10-fold lower than that on WT-CFTR mostly due to state-dependent binding (Yeh et al., 2017), it leaves a potentially larger room for the affinity to be increased. Fig. 3 B shows a representative recording of the dose-dependent change in N1138Y/G551D-CFTR currents in response to GLPG1837. A similar protocol was used to obtain the dose–response relationships for GLPG1837 on N1138F/ and N1138L/G551D-CFTR. In Fig. 3 C, the dose–response curves for these three mutants are leftward shifted compared with G551D-CFTR, supporting the hypothesis that hydrophobic amino acids substitutions at N1138 enhance binding of GLPG1837 (see Discussion for more details).

As seen in the docking mode shown in Fig. 1 B, the side chain of S1141 is ∼3 Å away from the thiophene ring of GLPG1837. Compared with N1138, S1141 is somewhat distant from the bound drug, suggesting a weaker interaction. One maneuver that may enhance its interaction is to convert S1141 to a positively charged amino acid to take advantage of a possible cation–π electron interaction. We therefore altered S1141 to lysine or arginine in the G551D background and determined the dose–response relationships for the double mutants S1141K/G551D-CFTR and S1141R/G551D-CFTR. Indeed, a leftward shift was observed, whereas the threonine substitution, which carries a hydroxyl group like serine, poses minimal effect (Fig. 3, D and E).

To this point, we found that mutations on the three residues in site I either decrease (D924) or increase (N1138 and S1141) the affinity for GLPG1837, supporting the notion that these amino acids may contribute to the binding of GLPG1837.

We next tested the effect of alanine substitution on the amino acids in the three lower-scoring sites (II, III, and IV) predicted by docking. Again, we reasoned that if any of the residues were involved in the binding of GLPG1837, substitution of the original side chain with alanine should disrupt the drug–protein interaction and hence decrease the affinity. Fig. 4 summarizes the dose–response relationships for mutations in site II, III, and IV. Site II consists of three hydrophilic amino acids in CFTR’s TM3 and TM6, namely K190 (TM3), T360 (TM6), and S364 (TM6), and a phenylalanine (F191) in TM3. Replacing the former three residues with alanine poses minimal effect on the affinity for GLPG1837 (Fig. 4, A and B), as the dose–response curves show little shift. We were not able to assess the effect of F191A mutation, as F191A-CFTR failed to generate detectable currents in our experimental setting. Similarly, we tested all three residues in site III, including Q237 in TM4, D993 in TM9, and S1149 in TM12. While the dose–response curves for D993A-CFTR and S1149A-CFTR are very close to that for WT-CFTR (Fig. 4 D), Q237A-CFTR appears to decrease the affinity and shift the curve rightward (Fig. 4, C and D). However, it should be noted that this slight change in apparent affinity could be attributed to a decreased P_o of this mutant based on the idea of state-dependent binding (Yeh et al., 2017). The clue for a lower P_o is that the maximal percent increase of Q237A-CFTR currents potentiated by GLPG1837 (615%; Table 1) is much larger than that of WT-CFTR (115%; Table 1). A simple calculation yields a maximum P_o of 0.14 (1/7.15) for Q237A-CFTR (vs. ∼0.4 for WT).

The site with the lowest score or least negative binding energy (IV) has two serine residues located within 5 Å of the bound GLPG1837, namely S182 in TM3 and S263 in TM4. Similar to the results for site II and site III, alanine substitution of either of these two serines has little effect on the dose–response relationships of GLPG1837 (Fig. 4, E and F). The dose–response curves for S182A-CFTR and S263A-CFTR almost overlap with that for WT-CFTR, suggesting that neither amino acid is involved in the binding of GLPG1837.

By combining in silico docking and the functional assay, we identified site I as a likely binding site for the CFTR potentiator GLPG1837. However, one caveat in the aforementioned study is that the docking simulation was based on a homology model, which could be different significantly from the actual structure of hCFTR. Fortunately, the atomic structure of the ATP-bound, phosphorylated hCFTR was solved with cryo-EM in late 2018 (Zhang et al., 2018b). We immediately performed molecular docking on this structure and compared the results with those shown in Fig. 1. While both GLPG1837 and VX-770 could be docked to the same four binding sites in this new cryo-EM structure (Fig. S1, A–D), an additional binding site located at the interface between TM4, TM5, and TM8 was identified (Fig. 5 A; and Fig. S1, E and F). This new site (named site II_N, because its score is between site I and II) consists of four phenylalanine (F229, F236, F312, and F931) residues and one tyrosine (Y304) residue, forming a hydrophobic cradle to accommodate the potentiators (Fig. 5 B). This binding pocket was not found in our homology model, because the side chain of residue F931 in the original zCFTR structure (PDB accession no. 5W81) protrudes in a direction that clashes with the potentiators. As both GLPG1837 and VX-770 are hydrophobic molecules carrying multiple rings (Van Goor et al., 2009; Van der Plas et al., 2018), we speculated that ring–ring stacking interactions could play an important role in stabilizing drug binding. To test this idea, we determined the dose–response relationships for GLPG1837 with each residue replaced with alanine one at a time. Among the five residues identified in site II_N, alanine substitution at Y304 and F931 decreases the apparent affinity for GLPG1837 (Fig. 6, A and B), whereas F229A and F236A have little effect (Fig. 6 B). The EC₅₀ for F312A appears to be similar to that for WT-CFTR, but the efficacy (i.e., maximal percent increase of currents) is reduced from 115% in WT-CFTR to 26% for F312A-CFTR (Table 1), suggesting that F312A may increase the P_o. According to the idea of state-dependent binding (Yeh et al., 2017), the apparent affinity should increase as the P_o increases. The fact that F312A-CFTR may bear a higher P_o but unchanged EC₅₀ implies its actual affinity for GLPG1837 is reduced. This idea will be tested in Fig. 7 below.

Y304 is in close proximity to the pyrazole group of GLPG1837, where both the hydroxyl group and aromatic ring of Y304 could potentially interact with GLPG1837. We substituted Y304 with phenylalanine, threonine, and alanine. Fig. 6 C shows a representative recording of Y304F-CFTR in response to GLPG1837, and the dose–response relationship yields an EC₅₀ of 2.28 ± 0.34 µM (Fig. 6 D). Y304T-CFTR, which preserves the hydroxyl group, also exhibits a decreased affinity. As the currents of Y304T-CFTR at 3 µM GLPG1837 are still <50% of that with 30 µM GLPG1837, the highest concentration applied in the experiments, the EC₅₀ for Y304T-CFTR is expected to be >3 µM. Similar results were observed with Y304A-CFTR. These data suggest a major role of the side chain of Y304 in the binding of GLPG1837.

Although for technical reasons (see Materials and methods) we did not obtain a complete dose–response relationship of GLPG1837 for Y304A- and Y304T-CFTR, the current relaxation analysis upon removal of GLPG1837, which reflects the apparent dissociation rate, did provide another line of supporting evidence for a significant decrease in the affinity (Fig. 6 E). Both Y304A- and Y304T-CFTR shorten the relaxation time constant to <5 s (c.f., ∼17 s for WT-CFTR).

As briefly mentioned above, although F312A in site II_N appears to have little impact on the EC₅₀ of GLPG1837 (Fig. 6 B), the true binding affinity for GLPG1837 in F312A-CFTR may be much lower when we take the effects of state-dependent binding into consideration. Fig. 7 A shows a real-time recording of F312A-CFTR currents in response to GLPG1837 in the presence of 2 mM ATP. The maximally potentiated currents at 20 µM GLPG1837 are only ∼1.3-fold of the initial currents in the presence of ATP, implying that the P_o of F312A-CFTR could be greater than WT-CFTR, which responds to GLPG1837 with twofold increase in macroscopic currents (Fig. 2). Indeed, single-channel kinetic analysis confirmed that the P_o of F312A-CFTR (0.63 ± 0.04; Fig. S2) is higher than the P_o of WT-CFTR mainly due to a prolonged open time constant (∼0.4, Zhou et al., 2006; Yeh et al., 2017).

According to the idea of state-dependent binding, the positive correlation between apparent affinity and P_o demands that the measured apparent affinity increases as the P_o increases (Yeh et al., 2017). Thus, the observation that the measured EC₅₀ for GLPG1837 in F312A-CFTR is slightly higher than that of WT-CFTR (0.4 vs. 0.26 µM) suggests a much lower affinity for GLPG1837 than WT-CFTR. In other words, the true binding affinity for GLPG1837 is significantly decreased by the F312A mutation, but the effect is masked by the opposing effect of its high P_o. This idea can be further examined simply by adjusting the P_o of F312A-CFTR to the level of WT-CFTR by lowering the concentration of ATP. In Fig. 7 B, the currents of F312A-CFTR are decreased by 45 ± 3% with 30 µM ATP compared with the initial currents at 2 mM ATP (n = 7), reducing the P_o to ∼0.35 (0.63 × 55% = 0.35). In this condition, the dose–response relationship is rightward shifted, and Hill equation fitting yields an EC₅₀ of 1.30 ± 0.26 µM (Fig. 7 C). Of note, for WT-CFTR, a 20-fold reduction of the P_o to 0.02 only increases the EC₅₀ from 0.26 to 1.7 µM (Yeh et al., 2017), suggesting that an EC₅₀ of 1.3 µM at a P_o of 0.35 in F312A-CFTR is unlikely accounted for simply by a state-dependent binding mechanism. We thus conclude that alanine substitution of F312 indeed reduces the apparent affinity for GLPG1837, supporting the notion that F312 contributes to binding of GLPG1837.

So far, we narrowed down the potential binding sites for GLPG1837 to two loci, site I and site II_N. As discussed earlier, it is impractical, at least in our experimental setting, to perform the same dose–response experiments for VX-770 to confirm whether it binds to the same sites. Here, we took the alternative strategy of relaxation analysis described in Fig. 2 D, in which the reduced affinity of GLPG1837 is reflected by a faster current decay upon its removal, implicating a faster dissociation rate. Fig. 8, A and B show the current relaxation upon removal of VX-770 in WT-CFTR and D924N-CFTR. While the WT-CFTR currents relaxed back to the initial level with a time constant of 361 ± 39 s, D924N shortened the relaxation time constant to 115 ± 7 s (P < 0.05). The accelerated dissociation of VX-770 in D924N-CFTR is consistent with our observation for GLPG1837 in Fig. 2 D and supports the hypothesis that D924 is involved in the binding of both VX-770 and GLPG1837. Next, we tested whether the other two residues in site I, N1138 and S1141 (in the G551D background), also contribute to the binding of VX-770 in a manner similar to their interaction with GLPG1837. In Fig. 8, C and D, single-exponential fitting yields a time constant of 68 ± 14 s for the currents decay upon removal of VX-770 in G551D-CFTR, whereas the relaxation time constant is prolonged to 221 ± 50 s for S1141K/G551D-CFTR (P < 0.05) and 268 ± 25 s for N1138L/G551D (P < 0.05). It should be noted that the effect of VX-770 cannot be washed out completely within the experimentally permissible time, which is likely due to its extremely high affinity and “stickiness” to the recording system (Jih and Hwang, 2013; Yeh et al., 2015, 2017). As the currents do not relax back to the initial level before the application of VX-770 (Fig. 8, A and C), the value of the relaxation time constant is likely to be somewhat underestimated. Nevertheless, some observed changes in the decay time constants are visually discernible (e.g., Fig. 8 and also see Fig. 9 below) and hence support the notion that mutations in site I change the affinity for VX-770.

While the decay time constant reflects the dissociation rate of the drug, the association rate of VX-770 to the channels can be assessed by measuring the time constant of the current rising phase (τ_on) upon the application of VX-770. Here, we report and compare τ_on and τ_off only, instead of using these numbers as inputs to calculate the theoretical K_d, because we have no evidence for a simple bimolecular reaction between CFTR and VX-770 (or GLPG1837) especially in light of the likelihood that binding occurs in the interface between lipid bilayer and the protein (see Discussion). Fig. 8 B shows that D924N-CFTR not only accelerates the current decay upon removal of VX-770 but also shows a slower rising phase upon the application of VX-770 than WT-CFTR, further strengthening the conclusion of a lower affinity for VX-770. On the other hand, the τ_on values for N1138L/G551D- and S1141K/G551D-CFTR do not differ from the τ_on for G551D-CFTR (Fig. 8 D). As a negative control, the same experiments were performed on S182A- and S263A-CFTR. Fig. S3 shows that neither τ_on nor τ_off of VX-770 were affected by these mutations in site IV.

Similar relaxation analysis was performed to test the changes in the affinity for VX-770 by mutations in site II_N. In Fig. 9 A, alanine substitution at Y304, which drastically decreases the affinity for GLPG1837 (Fig. 6), reduces τ_off to 13 s (vs. ∼360 s for WT-CFTR), suggesting a faster dissociation rate or off rate. In addition, the current rise upon application of VX-770 for Y304A-CFTR is visibly slower than that of WT channels (Fig. 9 A), with a τ_on prolonged to 41 s (vs. ∼6.7 s for WT-CFTR). Similarly, Fig. 9 B shows an accelerated relaxation upon removal of VX-770 with F312A-CFTR. Fig. 9 C and Table 1 summarize the τ_on of VX-770 for site II_N mutants and the significant shortening of the τ_off of VX-770 for F236A-, Y304A-, and F312A-CFTR. These results support the proposition that amino acids in both site I and site II_N that interact with GLPG1837 also play important roles in the binding of VX-770. These data shown in Figs. 2, 3, 6, and 7 for GLPG1837 and Figs. 8 and 9 for VX-770 also reaffirm our previous conclusion that GLPG1837 and VX-770 may share a common binding site (Yeh et al., 2017).

In conclusion, our data support the notion that two CFTR potentiators, GLPG1837 and VX-770, share a common binding site and that two specific loci, site I and site II_N, initially identified by molecular docking, may serve as their molecular target sites in the CFTR protein.

In this study, we combined in silico docking and the patch-clamp functional assay to identify the molecular target sites for CFTR potentiators GLPG1837 and VX-770, which were previously shown to share a common binding site (Yeh et al., 2017). We provided evidence that two loci at the interface of CFTR’s two TMDs, site I (D924, N1138, and S1141) and site II_N (F229, F236, Y304, F312, and F931), are likely to be the binding sites where GLPG1837 and VX-770 exert their function as potentiators. In this section, we will first discuss the intrinsic limitations of the techniques used in the current study, including homology model, molecular docking, and the functional assay using mutants as the main tool. Next, we will delve into more detailed discussion on the complexity in interpreting affinity and efficacy in the context of the state-dependent binding and seek a possible yet imperfect solution to untangle the convoluted relationship between drug binding and channel gating. We will then provide a more in-depth analysis of the pros and cons for each of the identified binding sites for GLGP1837 and VX-770. Finally, we will speculate on the implications of our results on structure-based drug design, and discuss translational significance based on our current understanding of drug effects on patients carrying pathogenic mutations with gating defects.

It is imperative to realize that the chemical interactions between the potentiators and their targets, which determine their affinity and efficacy, are dynamic as the channel itself usually undergoes multiple conformational changes (Colquhoun, 1998). To hunt for the binding sites, we first need to ask which conformation of the channel is a better starting point. As CFTR is a phosphorylation-activated and ATP-gated channel, the ideal structure would be a phosphorylated and ATP-bound open CFTR. Since this study started at a time when the corresponding hCFTR structure was unavailable, we employed homology modeling using the cryo-EM structure of zCFTR as a template. Although human and zCFTR share 55% sequence identity that ensures the current modeling techniques could achieve reliable prediction accuracy (Marti-Renom et al., 2003), the possible side-chain rotameric orientations of every amino acid inevitably result in significant uncertainties. This issue is particularly critical, as the drug–protein interaction is highly sensitive to the local amino acid positioning in the binding sites. The fact that our docking software failed to identify site II_N in the homology model because the side chain of residue F931 in the modeled structure obstructs this binding pocket bespeaks this very point.

Although the problem of homology modeling was solved by the timely atomic structure of hCFTR, docking simulation bears its own limitations. First, the computer program allows the drugs to assume different orientations but treats the protein as a rigid body with immobile amino acid side chains. However, it is unrealistic to assume that the amino acids are positioned at the exact location without dynamic changes upon drug binding. For example, according to the induced-fit model, binding of a ligand could cause considerable changes in the shape of its binding site (Koshland, 1958). Thus, the docking software–calculated free energy of interaction between the ligand and its target may not represent the actual condition where the target could adjust to better interact with the ligand. Furthermore, the lipid bilayer is not considered in current docking. The docking simulation assigned a better (i.e., more negative) score to site I than site II_N without taking into account the possible roles their distinct local environments may play. For example, while site I resides close to the aqueous permeation pathway, site II_N, located on the lateral surface of CFTR’s TMDs, likely makes direct contact with the lipid bilayer. The affinity for site II_N may in fact be greater than site I in a physiological condition due to a more favorable partition of hydrophobic compounds such as GLPG1837 and VX-770 into the lipid phase. In fact, this lipid–protein interface may also account for the observation that alanine substitution of F229 or F236 fails to affect the EC₅₀ of GLPG1837 (Fig. 6 B), as the membrane lipids may fill the void and compensate for the loss of binding energy in F229A and F236A. Although what have been discussed here are of a more speculative nature, they highlight the potential limitations of the docking approach to identify the binding sites.

Despite these caveats, we recognized that the docking results can serve as a guide in our hunt for the drug binding sites. Once potential binding partners are identified through docking, we can then alter the amino acids involved in binding and measure the effects of the mutations on the sensitivity and efficacy of the drugs of interest. One immediate problem in mutational study is that we cannot exclude the possibility that observed change in the apparent affinity is actually caused by allosteric effects of the mutation rather than by a direct alteration of the binding event. There are at least two scenarios where the word allosteric applies. First, by allosteric, we mean the amino acid being mutated alters the structure of the actual binding site through a long-distance effect. In an effort to minimize this possibility, we affirm the binding site only when mutations of a majority of the identified residues in one binding site produce large changes in EC₅₀ (also see below and Table 1). For example, all three residues in site I (D924, N1138, and S1141) affect affinity for GLPG1837. Furthermore, the observation of side-chain–specific changes in affinity further supports a specific interaction between the referenced amino acid and the ligand. However, as we do not have evidence to prove this scenario wrong, we caution our audience to be aware of this potential caveat.

The second scenario is much more complex and features a classical struggle while studying the structure–activity relationships for ligands and their target receptors. Here, the term allosteric refers to a mechanism in which the protein exists in two different conformations where their affinities for the ligands could differ substantially. As this issue to some extent is model dependent, it deserves a separate section of discussion. Some thermodynamic considerations will be deliberated first in the next section. We will then discuss the complexity in the relationship between ligand binding and channel gating in the context of an allosteric modulation model.

For now, let us take a more literal interpretation of the measured EC₅₀ for the K_d of a CFTR potentiator, whose physicochemical meaning can then be expressed in an energetic term: the free energy of binding (ΔG) = −RTln(K_d). In a system without other sources of energy input, the differences in free energy change (ΔΔG) between a mutant protein and WT protein undergoing this reaction will be −RTln(K_d(mutant)/K_d(WT)), which represents the loss or gain of binding energy introduced by the mutation (see Table 1 for details). When we calculated the corresponding ΔΔG for each mutation in the five binding sites, we found that a majority of mutations in site I and II_N introduces a ΔΔG greater than the other mutations in sites II–IV (Table 1). Specifically, the three mutations in site I with the greatest ΔΔG include D924A, N1138L/G551D, and S1141K/G551D, which result in a ΔΔG (absolute value) >5 kJ/mol. In site II_N, mutations on Y304 and F312 change the ΔG by ≥4 kJ/mol. In contrast, mutations in site II, III, and IV cause smaller ΔΔG. If we assume each amino acid contributes to the binding of GLPG1837 independently, and therefore the summation of ΔΔG represents the total loss/gain of binding energy, we can safely conclude that sites I and II_N are more likely to be the binding sites than the remaining three sites.

Although the calculation of ΔΔG provides a quantitative way to evaluate and compare the effects of each mutation on the binding of GLPG1837, it is based on a questionable assumption. As binding of a potentiator affects gating of the channel (i.e., increases the P_o), the reciprocity is also valid. According to the idea of a cyclic model featuring energetic coupling between ligand binding and channel gating (Yeh et al., 2017), the apparent affinity depends not only on the absolute values of the closed and open state affinities but also on the distribution of the channel in these states. Thus, the measured EC₅₀ of a potentiator lies between the actual affinity for an open channel and that for a closed channel. In other words, the ΔΔG introduced by a mutation could also be attributed to an alteration of the gating equilibrium rather than changing binding (Colquhoun, 1998). Of note, although ATP hydrolysis, by providing an input of free energy into CFTR gating scheme (Csanády et al., 2019), may further complicate the interpretation of mutational effects, the general principle of state-dependent binding for CFTR potentiators likely still holds, since a P_o-dependent EC₅₀ has also been demonstrated for G551D-CFTR (Yeh et al., 2017), a mutant with limited capability of ATP hydrolysis (Li et al., 1996; Ramjeesingh et al., 2008).

It is therefore more precautious to state that when a mutation not only shifts the dose–response relationship but also changes the P_o of the channel, it becomes difficult to affirm the role of the residue of interest, since the issue of state-dependent binding discussed above has to be taken into consideration. Ideally, one could find a mutation that changes the dose–response relationship without affecting the P_o, and hence could more confidently attribute the role of the pertinent amino acid to ligand binding. However, the idea of state-dependent binding itself implies that the binding sites must undergo some conformational changes during gating. Thus, mutations meant to affirm the interaction between the ligand and its target very likely also affect gating. Indeed, the mutation D924N in site I decreases the P_o, as the maximal percent increase of currents by GLPG1837 is ∼500% (Table 1), implying a P_o < 0.17 (1/6). A similar problem is seen with mutation at N1138. N1138Y/G551D-CFTR increases the apparent affinity as well as P_o. The latter is qualitatively attested by the observation that the basal current of N1138Y/G551D-CFTR can easily reach a true “macroscopic” level (i.e., tens of picoamperes ), while G551D-CFTR in the same condition rarely generates current greater than a couple of picoamperes. In addition, the efficacy of GLPG1837 is reduced in N1138Y/G551D-CFTR, implying that the basal P_o of this double mutant is likely higher than that of G551D-CFTR. These observations and the resulting complications in data interpretation are perhaps not surprising, because it seems hard to imagine a stationary site I, which is buried in the interface between CFTR’s TMD1 and TMD2 and resides close the location of the gate (Gao and Hwang, 2015), during gating transitions.

Site II_N may not be totally immune to this predicament, as the F312A mutation indeed increases the P_o. Nonetheless, as described above, this nonideal condition can at least be amended by assessing the potency of GLPG1837 at a condition with a P_o similar to that of WT-CFTR (Fig. 7). Furthermore, while the Y304A and Y304F mutations show a drastic increase in EC₅₀ (>3 µM and 2.28 µM, respectively), the P_o of Y304A-CFTR (0.31) and Y304F-CFTR (0.39) is not very different from that of WT channels (0.4; Fig. S4). Given that these mutations in site II_N barely change P_o but significantly reduce the apparent affinity, we argue that the observed increases of EC₅₀ are not caused secondarily to alterations in gating. In summary, although state-dependent binding inevitably complicates the interpretation of our functional data, the fact that our patch-clamp data match well with molecular docking results warrants a more serious consideration of our conclusion that sites I and II_N are the real binding sites for GLPG1837 and VX-770.

While both site I and site II_N seem to have equivalent supporting evidence for being the binding site for GLPG1837, the effect of GLPG1837 shows no evidence of a cooperativity (Hill coefficient ≈1). More troublesome is the fact that these two sites are both located at the interface between TMD1 and TMD2; they are also situated on the same horizontal plan of the CFTR protein (Figs. 1 A and 5 A). This geographic relationship raises a distinct possibility that any mutations at one site may potentially affect the other. Thus, it is entirely plausible that only one of these two loci is actually the correct binding site. It is known that CFTR orthologues may respond differently to potentiators (Van Goor et al., 2009; Cui et al., 2016). We reason that by comparing the amino acid compositions of sites I and II_N in different species and their response to GLPG1837, we could gain indirect evidence to help differentiate site I and site II_N.

As our homology model was built on the zCFTR template, we first compared the amino acid compositions of both sites in zCFTR with those in hCFTR. When the two critical residues in site II_N, F312 and Y304, are unaltered at the corresponding positions in zCFTR, the hydrophilic asparagine and serine in site I of hCFTR are changed to hydrophobic residues in zCFTR (L1146 and G1149, respectively). The EC₅₀ for GLPG1837 is 3.9 ± 1.7 µM for zCFTR (c.f., 0.26 µM for hCFTR; Fig. S5, A and B). However, neither the single mutation L1146N or G1149S nor the double mutation L1146N/G1149S could restore the affinity (Fig. S5 B). Moreover, it is known that the P_o of zCFTR is much lower than that of hCFTR (Zhang et al., 2018a). Therefore, it is possible that the reduced affinity for GLPG1837 in zCFTR is mostly attributed to the preferential distribution of zCFTR in the closed state.

As zCFTR does not lead to a conclusive answer, we wondered whether a CFTR orthologue that has a distinct site II_N but the same amino acid composition in site I, as opposed to the case with zCFTR, would provide some insights. Mouse CFTR (mCFTR), which has an identical composition of site I but a different residue in site II_N, serves the purpose (F304 in mCFTR vs. Y304 in hCFTR). Interestingly, neither VX-770 (Fig. 10 A) nor GLPG1837 (Fig. 10 C) can potentiate mCFTR (Van Goor et al., 2009; Bose et al., 2019; but compare Cui and McCarty, 2015). Remarkably, F304Y-mCFTR restores the potentiating effects of both GLPG1837 and VX-770 (Fig. 10, B and D), with an EC₅₀ of 0.68 ± 0.18 µM for GLPG1837. It appears that this position determines an all-or-none effect of the potentiators for mCFTR. Although this line of evidence does not exclude site I or support site II_N to be the correct site, it does suggest that the local structure of site II_N is important for GLPG1837 (or VX-770) to act as a potentiator.

While it is striking that a single mutation in site II_N, F304Y, could restore the action of GLPG1837 and VX-770 in mCFTR, this is not the only position where mutations cause an all-or-none effect of the potentiators during our studies. In fact, mutations at R347 (R347C, R347D, and R347S) in hCFTR, the residue that forms a salt bridge with D924 to stabilize the open channel conformation, abolish the response of hCFTR to GLPG1837 (Fig. 11, A and B). Interestingly, as a well-defined CFTR potentiator, GLPG1837 paradoxically inhibits R347C-CFTR currents by 16 ± 3%. Although it seems puzzling at first glance, the idea of state-dependent binding predicts that the affinity for a ligand to a closed channel relative to an open channel determines whether it is a potentiator or an inhibitor. If the binding of a ligand favors an open (closed) channel, its binding will stabilize the open (closed) state, and the compound will function as a potentiator (inhibitor). Therefore, the reversed action of GLPG1837 on R347C-CFTR could be explained by a reversal of its affinities for the open and closed state. Here, we wonder if this simple idea could also account for the recent report that VX-770 at a micromolar concentration decreases the activity of mCFTR (Bose et al., 2019).

Despite that in the docking simulation R347 is not involved in binding of GLPG1837, its close proximity to site I and interaction with D924 presumably would affect the architecture of this binding site. Cotten and Welsh (1999) have shown that the reduced single-channel conductance of R347C-CFTR could be partially restored by acidification (i.e., lowering pH), implicating a partial reconstruction of the local structure. We thus tested the idea that a similar strategy of lowering pH may also restore the potentiating effect of GLPG17837. As shown in Fig. 11 C, GLPG1837 decreases the currents at pH 7.4 but generates appreciable increase at pH 5.5, supporting our hypothesis. The dual (inhibitory vs. potentiating) actions of GLPG1837 likely result from its binding to a single site with different affinities for GLPG1837 as pH varies (see the Fig. 11, C and D legend for details). While other more complicated scenarios may also explain this observation, the effect of R347C on the action of GLPG1837 suggests that the interaction of R347-D924, which potentially maintains an intact structure of site I, is essential for GLPG1837 to potentiate hCFTR. This idea is tested with R347D/D924R-CFTR, a double mutant in which the side chains of the R347-D924 salt bridge are swapped. Fig. S6 shows that R347D/D924R-CFTR can be potentiated by GLPG1837, supporting the idea that maintaining the local structure of site I is important for the potentiating effect of GLPG1837. The reduced apparent affinity and increased efficacy for GLPG1837 in R347D/D924R-CFTR also corroborate the conclusion that D924 in site 1 plays an important role in gating and/or binding of CFTR potentiators.

Although these additional investigations on zCFTR, mCFTR, and mutations at R347 in hCFTR reveal some interesting results, these data seem to support both sites as the targets for GLPG1837 and VX-770 without providing a firm answer to which is the correct binding site. While a single binding site for these CFTR potentiators is a simpler scenario, multiple binding sites existing in the CFTR protein remain a possibility, and more studies are needed at this juncture.

Regardless of the pros and cons analyzed above for these two sites, a key feature of them is that both sites are located at the interface of CFTR’s two TMDs. This interface likely undergoes conformational changes during the gating motion so that the structures of the binding sites in the closed state would differ from that in the open state, leading to different affinities for CFTR potentiators. For example, the potentiator and the amino acids in the binding site may be closer in an open state, yielding a stronger interaction. It is also possible that after channel opening, some interactions present in the closed state will be lost and other amino acids will impinge onto the potentiator to provide tighter binding. Ultimately, understanding of the specific interactions between the potentiator and its target site in both open and closed channels holds the key for designing compounds with better affinity and efficacy. As the allosteric coupling model between binding and gating suggests, the improvement in affinity relies on stronger binding to both closed and open states, whereas a better efficacy will be achieved by widening the difference between the affinities for each state. In this aspect, knowledge in the structural changes upon channel opening in CFTR’s TMDs, particularly in the interface where the proposed binding sites are located, is urgently needed. We believe that the current study in identifying the molecular target sites for CFTR potentiators, together with additional atomic structures of CFTR that could emerge in the near future, should provide a practical guide along the path to realize structure-based drug design for CF therapy.

We thank Cindy Chu and Shenghui Hu for their technical assistance. We are also grateful that Prof. Jue Chen (Rockefeller University, New York, NY) sent us the PDB file for the hCFTR structure before it became publicly available. Discussion with Prof. Jue Chen, Prof. John Hunt (Columbia University, New York, NY), and Prof. Ina Urbatsch (Texas Tech University, Lubbock, TX) at the 32nd North American Cystic Fibrosis Conference over the structural biology of CFTR potentiators was also exceedingly valuable for our pursuit. Prof. Hunt’s sharing of his unpublished cryo-EM structure at the 16th European Cystic Fibrosis Basic Science Conference is very much appreciated.

This work is supported by the National Institutes of Health (grant R01DK55835 to T.-C. Hwang and grant R01GM109980 to X. Zou) and the Cystic Fibrosis Foundation (grant HWANG15G0 to T-C. Hwang).

The authors declare no competing financial interests.

Author contributions: Original conceptualization of the project was based on our 2017 Journal of General Physiology paper (H-I. Yeh, Y. Sohma, K. Conrath, and T-C. Hwang) that shows a common mechanism and binding site for GLPG1837, a newly developed CFTR potentiator, and VX-770, a drug used clinically. It is recognized by H-I. Yeh and T-C. Hwang that the low affinity but high potency of GLPG1837 makes it an invaluable tool to hunt for the binding site(s) of CFTR potentiators. Subsequently the goals of the current study, using a combination of computational and electrophysiological approaches to identify the binding sites for CFTR potentiators, were encouraged by data showing some mutations do affect the potency and efficacy of GLPG1837 (H-I. Yeh) and eventually formulated by H-I. Yeh, L. Qiu, X. Zou, and T-C. Hwang. The creation of a homology model of the CFTR structure and the docking studies were performed by L. Qiu and supervised by X. Zou. The patch-clamp experiments were conducted by H-I. Yeh. The results of docking studies were analyzed by L. Qiu and X. Zou and interpreted mostly by L. Qiu and X. Zou. H-I. Yeh and T-C. Hwang also provided inputs. The electrophysiological data were analyzed by H-I. Yeh and interpreted by H-I. Yeh and T-C. Hwang. The presentation of docking results was made mainly by H-I. Yeh and assisted by L. Qiu. The figures for electrophysiological data were made by H-I. Yeh. The original draft of the manuscript was written by H-I. Yeh and T-C. Hwang. The structural modeling and docking parts were written by L. Qiu and X. Zou. The original draft of the manuscript was reviewed and edited by H-I. Yeh, L. Qiu, Y. Sohma, K. Conrath, X. Zou, and T-C. Hwang. Oversight of the planning and execution of the current project is exercised by T-C. Hwang, who also manages and coordinates the collaboration among all other authors involved in this work. All authors approved the final version of the manuscript. All experiments were performed at Dalton Cardiovascular Research Center, University of Missouri (Columbia, MO).

Merritt C. Maduke served as editor.