Ligand binding to a remote site thermodynamically corrects the F508del mutation in the human cystic fibrosis transmembrane conductance regulator

Many disease-causing mutations impair protein stability. Here, we explore a thermodynamic strategy to correct the disease-causing F508del mutation in the human cystic fibrosis transmembrane conductance regulator (hCFTR). F508del destabilizes nucleotide-binding domain 1 (hNBD1) in hCFTR relative to an aggregation-prone intermediate. We developed a fluorescence self-quenching assay for compounds that prevent aggregation of hNBD1 by stabilizing its native conformation. Unexpectedly, we found that dTTP and nucleotide analogs with exocyclic methyl groups bind to hNBD1 more strongly than ATP and preserve electrophysiological function of full-length F508del-hCFTR channels at temperatures up to 37 °C. Furthermore, nucleotides that increase open-channel probability, which reflects stabilization of an interdomain interface to hNBD1, thermally protect full-length F508del-hCFTR even when they do not stabilize isolated hNBD1. Therefore, stabilization of hNBD1 itself or of one of its interdomain interfaces by a small molecule indirectly offsets the destabilizing effect of the F508del mutation on full-length hCFTR. These results indicate that high-affinity binding of a small molecule to a remote site can correct a disease-causing mutation. We propose that the strategies described here should be applicable to identifying small molecules to help manage other human diseases caused by mutations that destabilize native protein conformation.

domains (TMDs) of hCFTR, as schematized in Fig. 1. The F508del mutation causes degradation of hCFTR by cellular quality-control systems before being transported or "trafficked" to the site of function in the plasma membrane on the cell surface (24 -26).
Previously, we demonstrated that the F508del mutation destabilizes the native conformation of isolated human NBD1 (hNBD1) (10, 11) relative to a partially unfolded molten-globule conformation that aggregates aggressively in vitro (20,(27)(28)(29)(30)(31). This aggregation reaction is irreversible, and computational modeling of scan-rate-dependent differential scanning calorimetry (DSC) data indicated that it is controlled by a unimolecular conformational transition within the molten globule (10). Mg-ATP binds to native hNBD1 with sub-micromolar affinity, stabilizing it relative to this aggregation-prone molten globule (Fig. 1). Second-site mutations in hNBD1 that suppress defective trafficking of F508del-hCFTR in vivo also stabilize the native conformation of hNBD1 relative to the molten globule in vitro (32)(33)(34)(35), indicating that thermodynamic destabilization of hNBD1 by the F508del mutation is an important contributor to its molecular pathology. Similarly, the cosmotropic (structure-stabilizing) compound glycerol, which suppresses the trafficking defect in F508del-hCFTR in vivo (29,36), also stabilizes the native conformation of hNBD1 relative to the molten globule in vitro (10, 11,17,37). These analyses and others indicate that increasing the stability of the native conformation hNBD1 should counteract the pathological effect of the F508del mutation in hCFTR (Fig. 1).Thermodynamic principles suggest that the binding energy of any molecular species that interacts with hNBD1 should increase its stability ( Fig. 1) (38). Therefore, tight binding of a small molecule to hNBD1 should also be able to correct the "trafficking" defect of F508del-hCFTR, even if it binds at a remote location from the mutation site. Promoting more stable interaction of hNBD1 with other domains within hCFTR should also, in theory, correct the trafficking defect (27, 29, 30, 39 -41).
In this paper, we demonstrate that some nucleotide analogs, which bind ϳ20 Å from the site of the F508del mutation, bind to hNBD1 more tightly than the physiological ligand ATP and significantly enhance its thermal stability as well as that of fulllength F508del-hCFTR. These analogs were identified using a new high-throughput screen for NBD1-stabilizing agents that Its two nucleotide-binding domains, NBD1 (16 -18, 44) (blue, PDB code 2BBO) and NBD2 (green, PDB code 3GD7) bind two molecules of Mg-ATP (yellow and orange) at their mutual interface (4 -11). Its two transmembrane domains, TMD1 (blue) and TMD2 (green), structurally interdigitate to form two composite binding sites (20,21), one for NBD1 and the other for NBD2. Residue Phe-508 (red), which is deleted in F508del-hCFTR, is located on the surface of NBD1 that binds to the cognate binding site formed by TMD1/TMD2 (1,17,18,20,21). The regulatory or R region (magenta) is believed to be predominantly disordered but to have segments that reversibly bind to the other domains in hCFTR to modulate their behavior (16,(91)(92)(93). The folding pathway of human NBD1 (hNBD1) is schematized in the center and at the left, above a free-energy diagram. The native conformation of hNBD1, which is stabilized by binding to ATP and the TMDs of hCFTR, is in equilibrium with a low-energy "molten-globule" conformation (94 -101), which retains native-like secondary structure. Formation of this molten globule species is inhibited by a wide range of second-site mutations (10, 11,16) that also suppress the trafficking defect (32,34,41) in F508del-hCFTR that is responsible for causing cystic fibrosis (24). These observations support the hypothesis that the F508del mutation causes the disease by promoting formation of the highly aggregation-prone molten-globule intermediate. Thermodynamic theory suggests that chemical compounds that bind to F508del-hNBD1, either alone or at its interface with the TMDs of hCFTR, should pull the domain away from the molten-globule conformation (20,(27)(28)(29)(30) and thereby offset the defect caused by the mutation (orange and green dotted lines in the free-energy diagram). This thermodynamic effect can be thought of as a form of "mass action."

Thermodynamic correction of F508del-CFTR
we developed based on our previous biophysical studies showing that aggressive aggregation of NBD1 follows unfolding in vitro. This highly efficient screen directly detects aggregation and thereby indirectly monitors unfolding via self-quenching of a visible fluorophore that binds at a single site in an engineered variant of hNBD1. The most strongly stabilizing nucleotides identified in this screen represent the first small molecules demonstrated to significantly thermally stabilize full-length F508del-hCFTR. The thermodynamic approach and biophysical methods presented in this paper should be applicable to identifying small molecules representing lead compounds for the treatment of other human genetic diseases caused by mutations that destabilize native protein conformation.

Development of a fluorescence self-quenching (FSQ) assay for hNBD1 stability
The molecular pathology caused by the F508del mutation (42) in hCFTR is attributable at least in part to destabilization of the native conformation of hNBD1 relative to a molten globule folding intermediate (10,11). We were therefore motivated to develop a sensitive and efficient assay for use in high-throughput screening for hCFTR-stabilizing compounds. Because current assays for the stability of aggregation-prone proteins are technically complex, making them artifact-prone and/or difficult to adapt to a high-throughput format, we sought to develop an assay that monitors aggregation of the molten-globule intermediate of hNBD1, which we previously demonstrated to be tightly coupled to unfolding of native hNBD1 (10, 11). Fluorescence self-quenching, which strongly attenuates emission upon oligomerization of many fluorophores (43), offered a promising approach for development of an hNBD1 unfolding/aggregation assay. The fluorescence self-quenching is caused by resonanceenergy transfer between overlapping electronic absorption modes that have different fluorescent quantum yields, and it can monitor aggregation most efficiently when each molecule bears a single fluorophore before aggregation. Therefore, we sought to develop an hNBD1 construct with native-like stability harboring a single cysteine residue with high reactivity toward fluorescent maleimide reagents.
A previously characterized hNBD1 construct (10, 11,44) that spans residues 488 -646 in hCFTR was selected because it is well-behaved biochemically and biophysically ( Fig. 2A). This construct harbors an internal deletion of the regulatory insertion (RI) motif (residues 403-436), but it otherwise has the native sequence of hCFTR with Met at the polymorphic site at residue 470. This construct was named hNBD1⌬(RI,RE) in previous studies (10, 11, 44), but we have simplified its name to hNBD1⌬RI in this paper. It includes four endogenous cysteine residues present in hNBD1 in WT-hCFTR: Cys-491, Cys-524, Cys-590, and Cys-592 ( Fig. 2A). This construct was thermodynamically destabilized and aggregated upon reaction at roughly 1:1 stoichiometry with Alexa Fluor 546 (AF546) maleimide (Fig. S1B). Therefore, we engineered hNBD1⌬RI to remove endogenous reactive Cys residues and add a highly reactive non-native cysteine after the last residue at its C terminus ( Fig. 2 and Fig. S1), which creates the equivalent of a G646C mutation in the native human sequence. We performed free energy calculations using Eris (45) to predict the change in the thermodynamic stability upon mutation of each or all of the endogenous Cys residues. Calculated stability correlated well with experimentally measured stability by DSC for 11 different A, ribbon diagram of hNBD1⌬RI with Mg-ATP (yellow), residue Phe-508 (red), and all cysteine residues in the native domain shown in space-filling representation. The Cys residues are colored like the protein backbone in the subdomain in which they are located (orange for the F1-like ATP binding core, green for the ABC␤ subdomain, and blue for the ABC␣ subdomain. PDB code 2PZE). B, melting temperature (T m ) from DSC assays plotted against the change in the free energy of folding predicted by the program Eris (45,80,102,103) for 11 Cys-reduced hNBD1⌬RI variants. Assays were conducted at 0.5 mg/ml protein using 2 mM ATP in SSB, which contains 3 mM MgCl 2 , 150 mM NaCl, 10% (v/v) glycerol, 10% (v/v) ethylene glycol, 1 mM TCEP, and 20 mM Na-HEPES, pH 7.5. C, table showing the average value and sample variance of the T m from at least two replicate DSC assays conducted in the same ATP-containing buffer on protein constructs used to develop or implement the fluorescence self-quenching assay for hNBD1⌬RI stability (Fig. 3). The magenta prefix AF546-indicates protein covalently labeled with the fluorescent dye Alexa Fluor 546.

Thermodynamic correction of F508del-CFTR
single or multiple point mutants (Fig. 2B). The only construct retaining native-like thermodynamic stability in this study was the C592L single mutation (Fig. 2, B and C). Because Cys-592 is the only solvent-exposed Cys residue in hNBD1, we tried to optimize labeling conditions for the C592L/G646C-hNBD1⌬RI construct to obtain covalent attachment of AF546 exclusively to the engineered C-terminal Cys residue (Fig. S1, C and D). We identified conditions that consistently yield incorporation of 0.80 -0.90 molecules of AF546 per C592L/G646C-hNBD1⌬RI protomer. The resulting AF546-labeled protein preparation remains monodisperse as assayed by gel-filtration chromatography (Fig. S1D) and has a similar thermodynamic stability to the corresponding unlabeled parental construct, both when retaining the native Phe-508 residue and when harboring the disease-causing F508del mutation (Fig. 2C).
Triple spectroscopic detection experiments demonstrated that AF546 -F508 -C592L/G646C-hNBD1⌬RI has very similar biophysical properties to the unlabeled parental domain and that self-quenching of AF546 fluorescence in this construct provides an accurate and sensitive monitor of the tightly coupled unfolding and aggregation of the domain (Fig. 3A). These experiments were conducted as described previously (11) Figure 3. Efficient tFSQ assay for hNBD1 stability. A, triple detection experiment monitoring far-UV CD at 230 nm (top), SLS at 230 nm (middle), and fluorescence emission intensity at 573 nm (bottom) during thermal denaturation at 3°C/min of 0.05 mg/ml (2 M) AF546-labeled F508 -C592L/G646C-hNBD1⌬RI in the presence of 30 or 430 M Mg-ATP. The temperature corresponding to the steepest slope of decline in emission intensity, as determined using nonlinear curve fitting to the first derivative of the van't Hoff equation (104), is defined as the fluorescence self-quenching temperature (T SQ ).These experiments were conducted as described previously (10, 11) in a 1.6-ml cuvette using a fluorescence excitation wavelength of 556 nm. B, fluorescence emission intensity at 586 nm (top) and the first derivative of that intensity (bottom) during thermal denaturation of 0.005 mg/ml (ϳ0.2 M) AF546labeled F508del-C592L/G646C-hNBD1⌬RI in a 10-l volume in a 96-well microtiter plate. Fluorescence in all wells was monitored in parallel using a real-time PCR machine with 545 nm excitation. Results are shown from three replicate assays conducted at each of four different Mg-ATP concentrations indicated on the graph. The T SQ values here differ from A because the experiments employed different protein constructs (F508 versus F508del) and ATP concentrations. C, T SQ values measured in eight replicate assays on AF546-labeled F508del-C592L/G646C-hNBD1⌬RI conducted using the same methods in 3 or 2003 M Mg-ATP in the absence (left) or presence (right) of a 0.1% (w/v) concentration of the nonionic detergent C12E8 (ϳ10ϫ cmc). All assays in A-C were conducted in SSB.

Thermodynamic correction of F508del-CFTR
except for monitoring the visible fluorescence of the covalently attached AF546 dye instead of intrinsic tryptophan fluorescence. Circular dichroism (CD) spectroscopy was used to monitor protein secondary structure (top panel in Fig. 3A), and static light-scattering (SLS) was used to measure mass-averaged particle size during isothermal chemical denaturation (Fig. S2A) or thermal denaturation of the AF546-labeled protein (middle panel in Fig. 3A). The fluorescence emission of AF546, which was monitored at 573 nm using 556 nm light for excitation, shows complete quenching (bottom panel in Fig. 3A) during thermal denaturation closely coincident with the aggregation of the domain as monitored by SLS (middle panel in Fig.  3A). The midpoint of the fluorescence self-quenching transition (T SQ ) coincides with that of the SLS transition when thermal denaturation is conducted at either 30 or 430 M Mg-ATP.
The ϳ3.5°C increase in T SQ in the presence of the higher concentration of Mg-ATP demonstrates the efficacy of the assay in detecting a thermodynamic stabilization of the domain by a small-molecule ligand, in this case a natural physiological ligand. Mg-ATP stabilizes the domain because it binds to the native conformation with 200 nM affinity but has negligible affinity for the rapidly aggregating molten-globule intermediate state produced by denaturation (10, 11). This kind of protein folding intermediate loses the native tertiary structure but retains most of its secondary structure. Previous studies demonstrated that the rapid aggregation of the molten-globule intermediate of hNBD1 increases secondary structure slightly (10, 11), resulting in minimal change in the observed CD when this intermediate is formed during isothermal chemical denaturation (top panels in Fig. S2A) or thermal denaturation (top panel in Fig. 3A). However, a strong reduction in CD magnitude is observed at higher temperatures in thermal titrations (top panel of Fig. 3A) due to the unfolding of the molten-globule intermediate into a significantly less structured conformational state.
We evaluated the reproducibility and robustness of the thermal fluorescence self-quenching (tFSQ) assay by conducting it at different thermal scanning rates (Fig. S3A) in the presence of increasing concentrations of Mg-ATP in 96-or 384-well microtiter plates in real-time PCR machines (top panel in Fig. 3B and Fig. S3). Complete quenching of the fluorescence of AF546 -F508del-C592L/G646C-hNBD1⌬RI occurs in these thermal denaturation experiments in microtiter plates (Fig. 3B) over the same temperature range observed in the cuvette-based CD-SLS-FL assays (Fig. 3A). The maximum of the first derivative of the fluorescence emission intensity as a function of temperature gives a readily automated quantitative estimate of T SQ (bottom panel in Fig. 3B).
We used eight replicate assays conducted at 3 M or 2.003 mM Mg-ATP, which give mean T SQ values of 43.3 Ϯ 0.3 and 52.2 Ϯ 0.3°C, respectively, to calculate a Z-factor of 0.8 for the tFSQ assay. The Z-factor is widely used to assess the suitability of an assay for use in high-throughput screening (46), and assays characterized by Z-factors greater than 0.5, which corresponds to 12 standard deviations of separation between negative and positive controls, are considered to have excellent characteristics. The tFSQ assay performs similarly with a wide variety of AF546-labeled hNBD1 constructs harboring the C592L/G646C mutations, including the hNBD1⌬RI construct harboring the F508del mutation (Fig. S3, A and B) and several constructs including both the RI and RE segments (Fig. S3, C-E). The performance of the tFSQ assay is unchanged in the presence of the nonionic detergent C12E8 (Fig. 3C), suggesting that it could be applied directly to full-length hCFTR labeled by a single visible fluorophore. Therefore, the sensitive and efficient tFSQ assay that we have developed is applicable to highthroughput screening for compounds stabilizing a wide variety of protein constructs containing hNBD1, including in principle full-length hCFTR.

Mg-dTTP stabilizes hNBD1 more strongly than the physiological ligand Mg-ATP
To explore more thoroughly the characteristics of the tFSQ assay and its efficacy in detecting stabilizing ligands, we used it to evaluate the influence of the eight major physiological nucleotides on the thermal unfolding/aggregation of AF546 -F508 -C592L/G646C-hNBD1⌬RI (Fig. 4A). Whereas nucleotides bearing adenine, guanine, or uracil bases produce similar thermal stabilization and T SQ values, consistent with the promiscuous nucleotide-binding properties of many ABC family ATPases (8,47), thymine-containing nucleotides produce greater thermal stabilization, as manifested by higher T SQ . At a 2 mM nucleotide concentration, T SQ is consistently ϳ2.5°C higher for the Mg 2ϩ complex of thymine-containing nucleotides compared with adenine-containing nucleotides for all hNBD1 constructs tested, including AF546 -F508 -C592L/ G646C-hNBD1⌬RI (Fig. 4A), the equivalent construct harboring the F508del mutation (Fig. S3B), and a variety of constructs including the RI and RE segments (Fig. S3, C-E). A similar stabilizing effect for thymine compared with adenine is observed in isothermal fluorescence self-quenching (iFSQ) assays conducted in a high-throughput parallel format (Fig.  S2B).
The convenient properties of the high-throughput tFSQ assay enabled rapid completion of an extensive structureactivity study on nucleotide interaction with hNBD1 ( Fig. 4, A and B, and Figs. S4 and S5). Nucleotides with 14 different base structures all produce substantial stabilization of the domain compared with the control, indicating they all bind with high affinity (Fig. S5). A variety of nucleotides with exocyclic base modifications stabilize the domain more strongly than ATP, with 7-methyl-GTP (7Me-GTP) producing the greatest stabilization among the nucleotides analyzed (⌬T SQ ϭ ϩ3.3°C compared with ATP and ⌬T SQ ϭ ϩ1.0°C compared with dTTP). Addition of an exocyclic methyl group to a given base structure consistently produces an increase in T SQ of ϳ3°C (i.e. dTTP versus dUTP, 7Me-GTP versus GTP, and 5-methyl-dCTP versus dCTP in Fig. S5).
Equivalent T SQ values are observed for ribose and deoxyribose nucleotides bearing the same base, as well as for nucleotides harboring other modifications at the 2Ј and 3Ј positions on the ribose (Fig. 4A and Figs. S3C and S5). However, diphosphate nucleotides consistently give T SQ values ϳ4°C lower than the corresponding triphosphate nucleotides, whereas monophosphate nucleotides produce no significant shift compared with the control (Fig. S4A). Therefore, the monophosphates have Thermodynamic correction of F508del-CFTR negligible affinity for hNBD1, and the triphosphates have substantially higher binder energy than the diphosphates, at least in the temperature range of the unfolding reaction. Complete depletion of Mg 2ϩ reduces T SQ by more than 10°C (Fig. S4, B and C), indicating that the Mg 2ϩ cofactor contributes significantly to the energy of nucleotide binding to hNBD1. However, values from five replicate tFSQ assays conducted on AF546-labeled F508 -C592L/ G646C-hNBD1⌬RI in the presence of a 2 mM concentration of the Mg 2ϩ complex of the indicated nucleotide. The control assays contained 3 M Mg-ATP, which comes from the protein storage buffer and is present in all samples in addition to any added nucleotide. Assays were conducted in SSB using a heating rate of 3°C/min in a microtiter plate as in Fig. 3, B and C. Equivalent differences in T SQ are observed in assays conducted with AF546-labeled F508del-C592L/G646C-hNBD1⌬RI (Fig. S3B) or full-length C592L-hNBD1 (with native Cys-647) retaining the RI and RE sequences (Fig. S3, C and D). Sample variance for these assays ranged from 0.1 to 0.3°C, as reported in Fig. S5B. Data from four nucleotides are duplicated in panels in A and B to illustrate the consistent effect of adding an exocyclic methyl group to the bases. C, DSC assays conducted on 0.5 mg/ml protein in the presence of a 0.5 mM concentration of the Mg 2ϩ complex of the indicated nucleotide in SSB. D, T m from DSC assays plotted against the T SQ from the fluorescence selfquenching assays in Fig. 4, which were conducted in the same buffer containing a higher 2 mM concentration of the Mg 2ϩ complex of the indicated nucleotide. Linear regression gives a slope of 1.03 Ϯ 0.02 (red line). E, results from ITC measurements of ATP or dTTP binding to hNBD1⌬RI in the same buffer at 10°C. The fitted binding stoichiometry likely reflects some aggregated protein that does not bind nucleotide, which should not perturb inferred thermodynamic parameters, as explained under "Experimental procedures." Thermodynamic correction of F508del-CFTR even when Mg 2ϩ is fully depleted, 2 mM ATP increases T SQ by ϩ1.0°C, and 2 mM dTTP increases T SQ by ϩ3.0°C (Fig. S4C), indicating that the nucleotides retain significant binding affinity in the absence of the Mg 2ϩ cofactor, possibly reflecting its replacement by Na ϩ in the protein-bound state (8). The greater stabilization by dTTP compared with ATP even in the absence of Mg 2ϩ provides further evidence that the thymine base has higher binding energy than adenine for the native conformation of hNBD1.

Calorimetric studies confirm the thermodynamic inferences from the tFSQ assays
We used DSC, which directly monitors unfolding of the native conformation of hNBD1⌬RI into the molten-globule state (10), to confirm that the tFSQ assay accurately reports on thermodynamic stabilization of the domain. The shift in the thermal melting temperature (T m ) measured in DSC assays conducted in the presence of a 0.5 mM concentration of four representative nucleotides (Fig. 4C) closely matches the shift in T SQ in fluorescence self-quenching assays conducted in the presence of a 2.0 mM concentration of the same species (Fig.  4D). These results confirm our hypothesis that tight coupling of the unfolding of hNBD1 to aggregation enables the stability of the domain to be monitored reliably by an assay detecting aggregation.
Isothermal titration calorimetry (ITC) was used to quantify the binding affinities of 11 nucleotides characterized using tFSQ assays ( Fig. 4E and Table S1). These measurements were conducted at 10°C to minimize aggregation during data collection, which starts in a nucleotide-free buffer containing the domain at ϳ1 mg/ml protein concentration. Consistent with the greater stabilization of hNBD1⌬RI by dTTP compared with ATP in our tFSQ assays, ITC demonstrates that dTTP binds to hNBD1 with ϳ2.3-fold higher affinity than ATP (98 Ϯ 26 nM versus 228 Ϯ 32 nM) (Table S1), corresponding to ⌬⌬G dTTP-ATP of Ϫ0.47 kcal/mol.
The substantial increase in thermal stability produced by a small improvement in binding energy reflects the physical and thermodynamic complexities of the thermal unfolding reaction, which involves a very steep change in protein stability over a narrow temperature range. For most proteins, the stability of their native conformation relative to their unfolded state is determined by a small difference between large and offsetting enthalpy and entropy differences that vary substantially with temperature due primarily to changes in solvent dynamics. The complex thermodynamic processes controlling thermal unfolding enable relatively small energetic differences between the native and unfolded states to produce large changes in thermal stability. Because nucleotides bind tightly to the native conformation of hNBD1 but not significantly to the molten-globule intermediate produced by thermal denaturation (10, 11), their binding energy directly adds to the stability of the native conformation. The thermodynamic results showing that a Ϫ0.47 kcal/mol change in binding free energy produces a 2.5°C increase in thermal stability (Fig. 4) suggests that modest improvements in ligand-binding affinity should be able to offset the full 7°C change in the thermal stability of hNBD1 produced by the F508del mutation.
The ITC data demonstrate that all 11 nucleotides bind to hNBD1⌬RI with favorable enthalpy, reflecting a net gain in electrostatic and van der Waals interactions, but unfavorable entropy, reflecting a loss in motional freedom in the ligand and possibly the protein. Comparing the ITC data on dTTP and dUTP shows that the exocyclic methyl group at the 5-position on the pyrimidine base in dTTP, which is the only difference compared with dUTP, increases the entropy of binding to hNBD1⌬RI by ϩ5.0 cal/mol/°C (top of lower section in Table  S1). This effect increases the magnitude of Gibbs free energy of binding by 1.4 kcal/mol, but an offsetting 0.3 kcal/mol reduction in the magnitude of the enthalpy of binding decreases the net gain in the Gibbs free energy of binding to ϩ1.1 kcal/mol. These observations suggest that the improvement in binding affinity produced by the exocyclic methyl group on dTTP is an entropic effect derived from release of dynamically restricted water during the binding reaction. Similarly, comparing the ITC data on 7Me-GTP and GTP shows that the addition of an exocyclic methyl group to this purine base increases the entropy of binding to hNBD1⌬RI by ϩ1.2 cal/mol/°C (lower section in Table S1), consistent with the ϩ2.9°increase in T SQ in the presence of 7Me-GTP compared with GTP ( Fig. 4B and- Fig. S5B) also being attributable to an entropy gain due to greater water release upon binding a nucleotide base with an exocyclic modification.
Another noteworthy feature in the ITC data is the consistently strong entropy-enthalpy compensation when ribonucleotides versus 2Ј-deoxyribonucleotides bind to hNBD1⌬RI (bottom of lower section in Table S1). The ribonucleotides consistently show 2-4 kcal/mol more favorable enthalpy of binding (greater heat release) and a roughly equal increase in the entropy loss upon binding, resulting in approximately equivalent binding energies for both species. The previously published hNBD1 crystal structures, as well as those reported below, show the 2Ј-position on the ribose ring to be fully solvent-exposed, making it unlikely that the observed entropy-enthalpy compensation derives from differences in structural interactions in the bound state. Molecular dynamics calculations suggest that it derives instead from a large difference in the conformational entropy of the free nucleotides caused by the systematic differences in ribose ring pucker distribution caused by the presence or absence of the 2Ј-hydroxyl group. 13

X-ray crystal structures show significant variations in nucleotide-binding stereochemistry
We determined crystal structures at ϳ1.9 Å resolution for F508-hNBD1⌬RI bound to 11 different nucleotide triphosphates that were characterized in tFSQ assays (Table S2). Previous crystallographic studies demonstrated that the stereochemistry near the ATP-binding site in the domain is not altered by the F508del mutation (16,17). Our crystal structures show that all nucleotides bind in the canonical geometry observed for NBDs from ABC superfamily proteins ( Fig. 5 and Fig. S6) (8,17,18,47). Their triphosphate groups bind in a nearly identical geometry (Fig. S6, A and B (H-bonding) to the lysine, threonine, serine, and final two glycine residues in the Walker A motif (GXXGXGKTS) spanning residues 458 -466 and also to Gln-493 in the ␥-phosphate switch (8, 17) (alternatively called the Q-loop (48)).Their bases all adopt an anti-configuration and make aromatic stacking interactions with the indole group of Trp-401 in the antiparallel ␤-subdomain, but they adopt a wide range of positions that result in substantial differences in their closest contact to the methyl group on Thr-465 near the C terminus of the Walker A motif (Fig. 5, B-D, and Fig. S6B). This distance is 6.3 Å for the C8 atom ATP but only 4.1 Å for the exocyclic 5-methyl group in dTTP (Fig. 5B). In structures with different nucleotides, the base effectively swings parallel to the plane of the indole group of Trp-401 like a pendulum attached to a tether formed by the ribose group. The movement of the tether is produced by relatively small angular variations in the ribose group that keeps all dihedrals in the same potential energy well but collectively produce up to a 2.1-Å shift in the position of the atom on the base that is covalently bonded to the ribose group.

Thermodynamic correction of F508del-CFTR
To gain insight into the structural basis of the thermodynamic stabilization of the domain by dTTP, we compared structures of F508-hNBD1⌬RI with dTTP versus dUTP (Fig.  5C), which gives a T SQ 3.3°C lower than dTTP but differs exclusively by the absence of an exocyclic methyl group at the 5-position in the pyrimidine ring ( Fig. 4B and Fig. S5B). This comparison, combined with the ITC data for the binding of these two nucleotides (Table S1), suggests that the enhanced binding affinity of dTTP compared with dUTP is due to displacement of hydrating water from interaction with the methyl group on res-idue Thr-465 in the Walker A motif. The crystal structures show a relatively subtle ϳ1.1 Å shift in the position of the pyrimidine rings (Fig. 5C) in the two nucleotides, which enables the 5-methyl group on the thymine base in dTTP to make a weak 4.1 Å van der Waals contact to the methyl group of Thr-465. In contrast, the closest C5 atom in dUTP is 6.1 Å away, which leaves a gap large enough to accommodate hydrating water molecules in contact with the methyl group of Thr-465. Therefore, the crystal structures suggest that the 5-methyl group in dTTP will displace water molecules interacting with the methyl group of Thr-465 when dUTP is bound, which is likely to account for the ϩ5.0 cal/mol/°C increase in the entropy of binding of dTTP compared with dUTP (Table S1). The 4.1-Å distance of the 5-methyl group on dTTP responsible for this displacement effect corresponds to an enthalpically weak van der Waals interaction. The observation that such a weak interaction can produce such significant thermal stabilization of hNBD1 suggests that a nucleotide containing a base designed to make optimal packing interactions at this site could stabilize the domain more strongly and fully offset the thermal defect caused by the F508del mutation.
Analysis of the differences in the thermodynamics of binding of GTP versus 7Me-GTP is complicated by the fact that the corresponding crystal structures show a 2.1-Å shift in the positions of their purine bases (Fig. 5D), which is significantly larger than the shift observed for the pyrimidine bases in dTTP versus dUTP (Fig. 5C). Nonetheless, our structural and thermodynamic analyses suggest that the exocyclic methyl group in 7Me-GTP contributes to its increased stabilization of hNBD1 The protein backbone and carbon atoms in two residues from the dTTP structure are shown in gray. The Mg 2ϩ cofactor bound to dTTP is shown as a yellow sphere. Four nucleotides from different structures are shown in ball-and-stick representation with carbon atoms colored according to nucleotide identity: magenta for dTTP (PDB code 5TF8, R free ϭ 18.7% at 1.86 Å); green for ATP (PDB code 5TF7, R free ϭ 20.0% at 1.93 Å); cyan for 7Me-GTP (PDB code 5TFB, R free ϭ 19.9% at 1.87 Å); pink for dUTP (PDB code 5TFA, R free ϭ 19.6% at 1 .87 Å), and ruby for GTP (PDB code 5TFC, R free ϭ 20.2% at 1.92 Å). Oxygen, nitrogen, and phosphorous atoms are colored red, blue, and orange, respectively. Dotted lines connect atoms for which internuclear distances are given.

Thermodynamic correction of F508del-CFTR
compared with GTP by dehydrating the methyl group of Thr-465 in an equivalent manner to the exocyclic methyl group on dTTP. The crystal structures show that the exocyclic methyl group on 7Me-GTP is 4.0 Å away from the methyl group of Thr-465, whereas the closest C8 atom on GTP is 6.7 Å away, again leaving a large enough gap to accommodate hydrating water molecules when GTP is bound but not when 7Me-GTP is bound. Our ITC data demonstrate that the latter compound has ϩ1.2 cal/mol/°C greater entropy of binding (Table S1) and yields a ϩ2.9°C higher T SQ (Fig. S5B), consistent with the hypothesis that that dehydration of the methyl group of Thr-465 tends to increase the entropy of binding, the affinity, and the thermal stabilization of hNBD1⌬RI by nucleotides.
Comparing structures of F508-hNBD1⌬RI with ATP versus dATP show minimal differences in the positions or conformations of the ribose groups or bases (Fig. S6C). These observations support the inference presented above that the strong enthalpy-entropy compensation revealed by ITC binding assays on these compounds (Table S1) derives from differences in the conformational properties of the nucleotides in the unbound state.

Thermal rescue of F508del-hCFTR via stabilization of hNBD1 or the NBD1-NBD2 interface
We used temperature-controlled single-channel electrophysiology assays in black lipid membranes (6, 41, 49 -51) to evaluate the ability of representative nucleotides to stabilize full-length F508del-hCFTR against thermal inactivation ( Fig. 6 and Fig. S7). These assays quantify not only the conductance but also the nucleotide-dependent gating kinetics (i.e. opening and closing rates) of the chloride ion channel in individual functional hCFTR molecules. Channel opening is driven by formation of an hNBD1-hNBD2 interface that encapsulates two nucleotide triphosphates (13), one bound to each of these domains before interface formation, so the open probability (P o ) of the channel at saturating nucleotide concentration provides a measure of the thermodynamic stability of the nucleotide-containing hNBD1-hNBD2 interface.
The electrophysiology studies in Fig. 6 and Fig. S7 demonstrate that nucleotides that provide enhanced stabilization of either hNBD1 itself (orange dotted lines in Fig. 1) or the hNBD1-hNBD2 interface (green dotted lines in Fig. 1) thermally stabilize F508del-hCFTR ion channels. These results are consistent with the thermodynamic scheme in Fig. 1, which illustrates how compounds that bind to the native state of hCFTR (right side in Fig. 1) stabilize it relative to the unfolded molten-globule intermediate of hNBD1 (center left in Fig. 1) that causes permanent thermal inactivation when it aggregates (top left in Fig. 1) (10). Given this inactivation pathway, compounds with higher binding affinity that stabilize the native state more strongly should produce a greater reduction in the inactivation rate because they reduce the relative concentration of the irreversibly aggregating molten-globule intermediate. The temperature-controlled electrophysiology studies in Fig. 6 and Fig. S7 verify that this mass-action mechanism produces the predicted indirect thermodynamic correction of the destabilizing effect of the F508del mutation. Rescued channels harboring the F508del mutation (rF508del-hCFTR), which are expressed in BHK cells growing at reduced temperature in the presence of the pharmacological corrector VX809, exhibit a stable open state equivalent to that adopted by WT channels but also an unstable open state with significantly lower conductance called the fast-flickering mode (FFM) (right side of Fig. 6A) (41). This pathological conductance mode has been previously observed in electrophysiological experiments on rF508del-hCFTR in black lipid membranes (41), but it has not been reported in experiments conducted in excised membrane patches, where equivalent protein constructs show reduced channel lifetime (52-55) as well as evidence of a state with reduced conductance at physiological temperature (53). Possible reasons for the differences between experiments in black lipid membranes and excised membrane patches are outlined in the section entitled "Temperature-dependent electrophysiology assays" under "Experimental procedures." The FFM, which represents an early report of functional failure of F508del-hCFTR in black lipid membranes, presumably reflects a state with hNBD1 dissociated from its docking site on the TMDs due to the disruption of that interface by the F508del mutation, which leads to rapid fluctuation of the TMDs between the open and closed channel conformations via a transition pathway decoupled from hNBD1-hNBD2 interface formation.
At 25°C, rF508del-hCFTR alternates between the FFM and the normal gating mode in the presence of 2 mM ATP but exhibits exclusively the normal gating mode in the presence of the same concentration of dTTP (right side of Fig. 6A), indicating that dTTP stabilizes the physiological functional state of hCFTR more effectively than ATP. A more dramatic difference in gating properties is exhibited in experiments conducted at 30°C in which rF508del-hCFTR exclusively adopts the FFM in the presence of ATP but still exclusively adopts the normal gating mode in the presence of dTTP (Fig. 6B). Most strikingly, dTTP supported normal gating of rF508del-hCFTR at 37°C for ϳ1 min before channel inactivation (Fig. 6C), whereas none of the other nucleotides tested supported any channel function at this temperature. The thermodynamic studies presented above in conjunction with additional electrophysiological studies described below suggest that the qualitatively different behavior of rF508del-hCFTR in the presence of dTTP results from enhanced stabilization of the native conformation of F508del-hNBD1, which promotes its docking to the TMDs and formation of the proper physiological open-state conformation driven by nucleotide-mediated hNBD1-hNBD2 interaction (Fig. 1).
We evaluated a wider range of the nucleotides characterized in our T SQ assays ( Fig. 4 and Fig. S5) for their ability to support Thermodynamic correction of F508del-CFTR Thermodynamic correction of F508del-CFTR normal gating of rF508del-hCFTR at 30°C. These electrophysiological assays demonstrated that a variety of deoxyribonucleotides (dATP, dGTP, dTTP, and dUTP) and the ribonucleotides TTP (5-methyl-UTP) and 7Me-GTP support at least some normal gating at this temperature, whereas the other ribonucleotides tested (ATP, GTP, and UTP) do not (Fig. 6B and Fig. S7). To gain further insight into the thermodynamics of nucleotide interaction with hCFTR during gating, we performed full Eadie-Hofstee analyses of the gating behavior as a function of the concentration of eight representative nucleotides, four ribonucleotides (left in Fig. 6D), and four deoxyribonucleotides (right in Fig. 6D). We performed these analyses on WT-hCFTR because rF508del-hCFTR is unstable and does not gate normally with some of the nucleotides at 30°C, the temperature used to conduct our electrophysiological stabilization assays ( Fig. 6B and Fig. S7).
Eadie-Hofstee analysis of a ligand-gated channel, which examines P o as a function of ligand concentration, assumes channel opening depends on two factors, the effective dissociation constant or affinity of the ligand for the active conformational state of the channel undergoing gating (K eff ) and the probability of adopting the open state when that ligand fully saturates its binding sites on the channel (P o-max ). Given these assumptions, the observed P o for hCFTR should show a hyperbolic dependence on nucleotide triphosphate (NTP) concentration, reflecting the dependence of gating on an equilibrium binding process, as shown in Equation 1.
Rearranging Equation 1 yields a linearized form that is applied to analyzing the observed P o as a function of varying NTP concentration (Equation 2).
Therefore, the slope of the line in the resulting Eadie-Hofstee plot is the effective dissociation constant K eff for a given NTP, and the intercept on the vertical axis is P o-max when hCFTR is saturated by that NTP.
As indicated above, the opening of hCFTR channels is coupled to a conformational change in which spatially separated hNBD1 and hNBD2 each with a bound NTP (hCFTR CLOSED ⅐NTP 2 ) move together to tightly encapsulate the two bound NTPs in their mutual interface (hCFTR OPEN ⅐ NTP 2 ) (4, 5). This gating mechanism creates a simple relationship between P o-max in the Eadie-Hofstee plot and the equilibrium constant (K hNBD1-hNBD2 ) and Gibbs free energy (⌬G 0 hNBD1-hNBD2 ) for the functional conformational change mediating hNBD1-hNBD2 interface formation and channel opening in the presence of a given NTP (Reaction 1 and Equation 3).
This formula is justified more completely and is derived under "Relationship between the energy of hNBD1-hNBD2 interface formation and P o-max " under "Experimental procedures." The value of ⌬G 0 hNBD1-hNBD2 should contribute to net channel stability according the thermodynamic scheme in Fig. 1.
The Eadie-Hofstee analyses demonstrate that the deoxyribonucleotides dATP, dGTP, and dUTP give significantly higher P o-max and therefore thermodynamically stabilize the hNBD1-hNBD2 interface substantially more strongly than dTTP or the ribonucleotides ATP, GTP, UTP, and 7Me-GTP (Fig. 6D). The 2Ј-hydroxyl on the ribose is completely solvent-exposed in the crystal structures of hNBD1 (Figs. 5 and Fig. S6), making it very likely to contact hNBD2 directly in the hNBD1-hNBD2 interface in open hCFTR channels. The consistently higher P o-max values exhibited by the deoxyribonucleotides compared with the ribonucleotides with the same bases suggest that the 2Ј-hydroxyl on the ribose either loses more hydration energy or produces a mild steric clash upon packing in this interface.
Notably, the P o-max values measured for the different nucleotide analogs show a different rank-order than their effective affinities K eff determined in the Eadie-Hofstee analyses, which is attributable to the fact that K eff depends on the binding At least five different experiments of 2 min duration were used to calculate the mean P o value of for each NTP type and concentration. The values of the standard error of the mean are less than the size of the symbols and therefore not shown. E, Gibbs free energy change for opening the WT-hCFTR channel at saturating nucleotide concentration (⌬G 0 hNBD1-hNBD2 ) plotted against the fluorescence T SQ of AF546-labeled F508 -C592L/G646C-hNBD1⌬RI at a 2 mM concentration of the same nucleotide (from Fig. 4, A and B). The value of ⌬G 0 hNBD1-hNBD2 is calculated as ϪRT⅐ln(P o-max /(1 Ϫ P o-max )). Decreasing ⌬G 0 hNBD1-hNBD2 reflects greater stabilization of the functional hNBD1-hNBD2 interface, which is required for stable channel opening, whereas increasing T SQ reflects greater stabilization of hNBD1 itself. Closed symbols are used for nucleotides that maintain normal activity of rF508del-hCFTR channels at 30°C, and open symbols are used for those that do not. The ellipses highlight the correlation predicted by the thermodynamic scheme in Fig. 1 between the ability of nucleotides to rescue the thermal defect caused by the F508del mutation (B and C) and their efficacy in stabilizing either hNBD1 itself (orange dotted lines here and in Fig. 1) or the hNBD1-hNBD2 interface (green dotted lines here and in Fig. 1).

Thermodynamic correction of F508del-CFTR
energy of the nucleotide for both hNBD1 and hNBD2 in addition to its contribution to the stabilization energy of the functional hNBD1-hNBD2 interface, whereas P o-max depends exclusively on the latter factor. The highest effective affinities are exhibited by dTTP, 7Me-GTP, and dATP. The first two nucleotides have a relatively low P o-max of ϳ0.4 -0.5 (Fig. 6D) but provide the strongest thermal stabilization of hNBD1 (Fig.  4), indicating that they bind most tightly to hNBD1 at its unfolding temperature. In contrast, dATP does not stabilize hNBD1 as strongly (Fig. 4) but gives the highest observed P o-max of ϳ0.95 (Fig. 6D). Different nucleotides thus achieve high effective affinity via different molecular mechanisms.
To gain insight into the molecular mechanisms that enable compounds to thermally stabilize F508del-hCFTR, we plotted the value of ⌬G 0 hNBD1-hNBD2 determined for each nucleotide against the T SQ of hNBD1⌬RI measured in the presence of a 2 mM concentration of the same nucleotide, using closed symbols for the nucleotides that support normal channel gating and open symbols for those that do not (Fig. 6E). This graph shows that all of the nucleotides that restore normal channel gating and thermally rescue active F508del-hCFTR enhance the stability of either the hNBD1-hNBD2 interface (lower ⌬G 0 hNBD1-hNBD2 ) or the hNBD1 domain (higher T SQ ). These observations are consistent with the prediction from the thermodynamic theory that compounds that stabilize either hNBD1 directly or the hNBD1-hNBD2 interface can correct the thermodynamic defect caused by the F508del mutation in hCFTR (Fig. 1).

Discussion
In this paper, we use a novel thermal fluorescence selfquenching (tFSQ) assay to identify nucleotide analogs that stabilize the first nucleotide-binding domain (hNBD1) from hCFTR more strongly than the physiological nucleotide ligand ATP, and we use temperature-controlled single-channel electrophysiology assays to demonstrate that these analogs also correct the thermodynamic defect in full-length hCFTR caused by the predominant disease-causing F508del mutation in hNBD1. These results represent the first conclusive demonstration of correction of this defect by a small molecule binding directly to hCFTR (31,37,53,56). Thermodynamic theory predicts that compounds with sufficient binding affinity for either the hNBD1 domain destabilized by the mutation or for the interdomain interfaces it forms in full-length hCFTR should be able to offset the thermal defect caused by the F508del mutation (Fig. 1). Neither mode of correction has been documented convincingly in the literature to date, but our thermodynamic studies of hNBD1 ( Fig. 4 and Figs. S3-S5) combined with our electrophysiological assays on full-length hCFTR variants (Fig. 6, A-D, and Fig. S7) provide examples of both modes of correction (Fig. 6E).
Previous attempts to demonstrate direct correction of the thermodynamic defect caused by the F508del mutation focused on identification of small molecules that bind to hNBD1. Several compounds with relatively weak micromolar-level affinity for hNBD1 have been identified (39,57), but none of these compounds have been demonstrated to thermally rescue fulllength F508del-hCFTR. One such compound called RDR1 was identified using differential scanning fluorimetry (DSF) assays that monitor the unfolding of hNBD1 indirectly based on interaction of the unfolded conformation of the protein with a fluorescent reporter dye (40,58). Our tFSQ assays failed to reproduce the stabilizing effect on hNBD1 inferred from DSF assays (Fig. S8), and DSC assays also fail to show any improvement in the thermal stability of hNBD1 in the presence of this compound (59). In contrast, our electrophysiological assays (Fig. 6) demonstrate that nucleotides that stabilize hNBD1 (dTTP and 7Me-GTP in Fig. 4) and different nucleotides that stabilize the functional hNBD1-hNBD2 interface (dATP, dGTP, and dUTP in Fig. 6, D and E) restore normal channel gating and rescue the thermal defect in F508del-hCFTR.
A wide variety of "corrector" compounds improving F508del-hCFTR biogenesis in vivo have been identified using cell-based assays (60 -62), but their mechanisms-of-action (31,37,53,56) and binding sites remain unclear, which has prevented use of structure-based methods to improve their efficacy. The uncertainty concerning mechanism-of-action has left lingering doubt whether they bind directly to hCFTR or have a different molecular target. Importantly, none of these compounds have been demonstrated to thermally stabilize hCFTR molecules on a physiologically relevant time scale (31,37,53,56).
The corrector compound, VX809 or Lumacaftor, which has been approved for clinical use in humans, produces functional expression of F508del-hCFTR at ϳ14% of the level of WT-hCFTR (60). There are conflicting reports in the literature concerning whether Lumacaftor provides effective shortterm stabilization of hCFTR channel function in vivo on the time scale of minutes (31,37,53,63). Furthermore, one recent paper demonstrated significant thermodynamic destabilization of isolated hNBD1 in DSC experiments conducted in the presence of 1 mM Lumacaftor (63). Although additional research will be required to resolve the controversies related to the interaction of Lumacaftor with hCFTR, it is clear that this drug does not fully correct the thermal stability defect caused by the F508del mutation in full-length hCFTR (31), and it does not support any proper channel gating at 37°C in temperature-dependent electrophysiology assays like those employed in this paper (31), while dTTP does (Fig. 6C). The thermally destabilizing influence of the F508del mutation promotes more rapid degradation of the protein at 37°C, which reduces the steadystate level of F508del-hCFTR in the plasma membrane. The results reported in this paper prove that thermal stabilization of hCFTR channel function ( Fig. 6 and Fig. S7) can be achieved by a compound binding directly to the protein (Fig. 5 and Fig. S6).
The previously identified corrector compounds improve functional expression of hCFTR, but their inability to correct the thermal stability defect suggests that they may modulate its biogenesis pathway rather than binding directly to the protein.
These compounds were all identified using cell-based screens, so their molecular target is not necessarily hCFTR. Most corrector compounds were identified using assays measuring F508del-hCFTR ion conductance in the plasma membrane of living cells (60 -62), and their molecular targets have not been conclusively established. In contrast, the molecular target is known for a corrector recently identified using an alternative Thermodynamic correction of F508del-CFTR cell-based screen for compounds that increase the level of immature F508del-hCFTR (band B) in the endoplasmic reticulum without changing its mRNA level. One compound identified using this screen, PYR-41, is a known inhibitor of an E1 ubiquitin-activating enzyme, and it increases the level of functional mature protein (band C) in the plasma membrane by reducing the rate of proteolytic degradation of F508del-hCFTR (62). This compound provides a clear example of a corrector that modulates hCFTR biogenesis indirectly rather than by binding to the protein directly. Correctors that do not bind to the protein directly are unlikely to thermally stabilize F508del-hCFTR, in which case they cannot correct one of the two molecular defects caused by the F508del mutation. In contrast, the nucleotide analogs analyzed in this paper bind directly to hCFTR, and as predicted by thermodynamic theory, they correct the fundamental thermodynamic defect caused by the F508del mutation and thermally stabilize F508del-hCFTR molecules.
Our calorimetric (Fig. 4, C-E, and Table S1) and crystallographic ( Fig. 5 and Fig. S6) studies suggest that a nucleotide with a custom base designed using computational chemistry methods could fully rescue the thermal defect caused by the F508del mutation. This observation is noteworthy because a variety of nucleotide analogs are effective clinically-approved drugs (64 -66). We observe that two nucleotides that make suboptimal van der Waals interactions (Fig. 5), dTTP and 7Me-GTP, stabilize hNBD1⌬RI more strongly than the major physiological ligand ATP and offset approximately half of the 6.8°C reduction in its thermal unfolding temperature produced by the F508del mutation (Fig. 2C). The exocyclic methyl groups on the bases of these nucleotides are likely to displace entropically restricted water molecules hydrating the hydrophobic methyl group on the side chain of residue Thr-465 in the Walker A motif (GXXGXGKTS) spanning residues 458 -466 in hNBD1 in hCFTR, thereby producing an entropy increase upon binding due to the reduction in solvent-exposed hydrophobic surface area (Fig. 4, C-E, and Table S1). However, the internuclear distances between the methyl group of Thr-465 and the exocyclic methyl groups on dTTP and 7Me-GTP are 4.1 and 4.0 Å, respectively, which represents a sub-optimal and energetically weak van der Waals contact. A nucleotide with a base making optimal packing interactions in this binding site would have the same entropic advantage while also having larger interaction enthalpy, which would increase binding affinity for hNBD1 and enhance its thermal stabilization.
The observation that a 0.5 kcal/mol difference in the binding affinity of dTTP compared to ATP (Fig. 4E and Table S1) offsets half of the thermal defect caused by the F508del mutation (Figs.  2C and 4, A-D, and Fig. S3B) suggests that a relatively small gain in nucleotide-binding affinity should completely offset the thermal defect in F508del-hCFTR at the therapeutic concentration for most drugs, which is typically in the low micromolar range. Physiologically, dTTP is present at a similar concentration of ϳ30 M (67), which is ϳ1/100 of the cellular concentration of ATP. Given the comparative binding affinities and cellular concentrations of dTTP versus ATP, the former only provides ϳ1/30 of the stabilization of the latter under physiological conditions. We estimate that an increase in binding energy compared to dTTP on the order of ϳ3 kcal/mol, which corresponds to an ϳ160-fold increase in equilibrium constant, would be needed for an appropriately designed nucleotide analog to fully offset the thermodynamic defect caused by the F508del mutation at the low micromolar concentrations typically achieved for the related drugs currently in clinical use (64). Nucleotide analogs are widely used in human pharmacology thanks to the presence of promiscuous kinases in human cells that phosphorylate the corresponding nucleosides, which are generally cell-permeable (68,69). Notably, the dTTP analogs azido-thymidine triphosphate and d4TTP (65), the triphosphate derivatives of Zidovudine (66) and Stavudine (66), remain among the most widely used antiviral drugs worldwide (70).
The technical characteristics of our FSQ assay (Fig. 3) demonstrate that equivalent assays should be a powerful tool for high-throughput screening to identify small molecule lead compounds to treat the many human protein-folding diseases that involve aggregation. Our results demonstrate that FSQ is equivalently effective in monitoring protein aggregation when implemented in thermal-ramping assays (tFSQ, Fig. 4 and Figs. S3-S5) or isothermal kinetic assays (iFSQ, Fig. S2). The excellent sensitivity of the assay derives from the high signal-tonoise ratio provided by fluorescence detection combined with the nonlinear dependence of self-quenching on aggregate size. Monitoring protein stability in this indirect manner based on subsequent aggregation provides a very large gain in sensitivity compared with the available biophysical methods that monitor unfolding directly. Furthermore, our data demonstrate that using FSQ to monitor unfolding/aggregation provides exceptional precision and reproducibility (Figs. 3, B and C, and 4, A and B, and Figs. S3-S5) while closely tracking the results obtained from rigorous thermodynamic assays (Fig. 4, C-E) that require large amounts of protein and that are too slow and cumbersome to be used for high-throughput screening. Many protein unfolding reactions induce aggregation (71,72), and a wide variety of serious and prevalent human diseases are caused by toxic protein aggregation processes (73)(74)(75), some of which are triggered by destabilizing mutations, as is the case for F508del mutation in hCFTR (1). The results reported in this paper on hNBD1 and hCFTR demonstrate that the FSQ assay provides excellent performance screening for compounds that prevent protein aggregation. Therefore, tFSQ and iFSQ assays equivalent to those implemented in this paper should be applicable to screening for correctors to treat diseases caused by protein aggregation (73)(74)(75) as well as diseases caused by protein destabilization when unfolding of the protein triggers aggregation (71,72). Moreover, the thermodynamic stabilization strategy (Fig. 1) proven to work in our studies of nucleotide interaction with F508del-hCFTR ( Fig. 6 and Fig. S7) should be applicable to treating diseases caused by protein instability.

Computational prediction of protein stability changes induced by Cys-reducing mutations
Starting from the crystal structure of F508-hNBD1⌬RI (PDB code 2PZE), we reconstructed missing side chains using the Eris protein design suite (76). We repaired breaks in the protein Thermodynamic correction of F508del-CFTR backbone using discrete molecular dynamics (77-79) with peptide bond distance and angular constraints connecting the broken fragments. We allowed three residues on each side of the break to move freely to fulfill the constraints, whereas the rest of the protein remained static. We performed mutations to the structure and calculated the ⌬⌬G using the Eris suite (45,80). We ensured that the residues binding ATP did not move significantly in any step of the procedure. The prediction of protein stability changes by mutations can be performed online using the Eris server (45,80).

Protein engineering and purification of thermally stable hNBD1 constructs
Using the QuikChange site-directed mutagenesis protocol (Agilent Technologies), the Cys-reduced constructs were engineered from truncated F508-and F508del-hNBD1 domains that were used in previously published studies. These protein constructs comprise residues 387-646 of human CFTR with residues 405-436 deleted. All the Cys-reduced hNBD1 constructs were expressed in Escherichia coli at 18°C and purified using previously published methods. These constructs, which have an N-terminal His 6 -Smt 3 fusion, were purified by Ni-NTA chromatography, cleaved by Ulp1 protease, purified by Sephacryl S200 gel-filtration chromatography, recovered from the flow-through of a second Ni-NTA column (to remove residual His 6 -Smt3 tag), and concentrated to 2-5 mg in Standard Stabilizing Buffer (SSB). The composition of this buffer is given in the legend for Fig. 2.

Fluorescent labeling of Cys-reduced hNBD1 domains
1 mg of AlexaFluor 546 (AF546) C 5 maleimide powder was dissolved in 200 l of fresh anhydrous DMSO (D8418, Sigma) immediately prior to hNBD1 labeling. A 10-fold higher concentration of the fluorescent dye was mixed with the protein at ϳ20 M concentration (0.5 mg/ml) in Standard Stabilizing Buffer. The mixture was incubated at room temperature for 5 min and then chilled on ice for another 5 min. Finally, a 1000-fold higher concentration of 2-mercaptoethanol compared with the AF546 dye was added to the mixture to terminate the labeling reaction. Free fluorescent dye in the mixture was removed using a PD-10 desalting column followed by gel-filtration chromatography in Standard Stabilizing Buffer. Purified AF546-labeled proteins with a labeling stoichiometry of ϳ0.9 were concentrated to approximately 120 M (3 mg/ml) for storage at Ϫ80°C after flash-freezing in liquid nitrogen.

Thermal fluorescence self-quenching assays
Thermal fluorescence self-quenching assays were conducted in a J-815 CD spectropolarimeter equipped with a PFD-425 Peltier temperature-controlled cell and an FMO-427 fluorescence detector (Jasco, Easton, MD) using a 1.6-ml cuvette. Temperature was monitored directly in the CD cuvette via a thermocouple connected to the control computer, and the heating rate was 3°C/min. Thermal self-quenching assays were conducted in plate format (96-or 384-well, 10 l solution volume) using the Mx3500P qPCR system (Agilent Technologies) for 96-well PCR plates and the ViiA 7 real-time PCR system (Life Technologies, Inc.) for 384-well PCR plates. Both systems used a heating rate of 3°C/min and fluorescence excitation and emission wavelengths of 545 and 568 nm, respectively. Data processing and T SQ determination were conducted using GraphPad Prism 5 software (GraphPad Software, Inc.). AF546labeled protein was diluted from the protein stock down to a working concentration of 0.005 mg/ml (ϳ0.2 M) for tFSQ assays. 100 mM stock solutions of eight native nucleotides or nucleotide analogs were diluted 1:49 into each well to yield the final desired nucleotide concentrations. The RDR1 compound was first dissolved in DMSO at 80 mM and diluted from this stock into the tFSQ assay solution to yield the final working concentration of the compound.

DSC
Calorimetry was carried out with a VP-Capillary DSC System (MicroCal Inc., GE Healthcare) in 0.130 ml of cells at 2°C/min. An external pressure of 2.0 atm was maintained during all DSC runs to prevent possible degassing of the solutions upon heating. The DSC measurements were conducted in SSB (defined in the legend for Fig. 2). ATP-free protein stock solution was prepared as follows: the ATP was removed by repeatedly diluting the sample 10-fold into Mg-ATP-free buffer and then concentrating back to the original volume using an Amicon ultrafiltration device with a 10-kDa molecular weight cutoff. A total of 5 dilution/concentration cycles were performed. 2 mM EDTA was included in the buffer for the first 3 cycles to ensure residual magnesium removal. Various amounts of the nucleotides or analogs were added to the protein sample and incubated at 5°C for at least 1 h prior to the initiation of the DSC measurement.
Protein concentration was determined using the Pierce BCA kit in a microtiter plate. Bacillus subtilis NAD synthetase was used to establish the protein-concentration calibration curve each time this assay was performed. The standard deviation in the measured concentration among triplicate samples was usually less than 5%. The protein concentration assay was also validated to ensure accuracy using an hNBD1 sample with known concentration that was determined by quantitative amino acid analysis. DSC was conducted at a protein concentration of 20 M. DSC data were analyzed with the Origin 7.0 software (OriginLab, Northampton, MA), from which the unfolding temperature (T m ) and the calorimetric unfolding enthalpy (⌬H c ) were obtained.

ITC
F508-hNBD1⌬RI was purified as described previously (10, 11) and stored in Standard Stabilizing Buffer. Immediately before each ITC experiment, ATP was removed as described above for the DSC experiments. The residual ATP concentration in the resulting protein sample was determined by an HPLC assay, as described previously (81). Briefly, 100 l of the protein sample was mixed with 100 l of 6 M guanidine-HCl to completely denature the protein and release any bound ATP, and 30 l of this denatured protein sample was analyzed on a 4.6 ϫ 100-mm Synergi Polar-RP column (Phenomenex, Torrance, CA) connected to a Shimadzu (Durham, NC) HPLC system. The ATP concentration calibration curve was established using pure ATP in the same buffer. The detection limit of this method was 0.025 M ATP. Measured ATP concentrations Thermodynamic correction of F508del-CFTR after ATP removal did not exceed 1 M and in most cases were less than 0.5 M. The nucleotide solutions were prepared by diluting a 100 mM stock into the matched control buffer, which was the filtrate from the ultrafiltration device during the last step of protein concentration. The nucleotide concentration was verified by UV absorbance using the extinction coefficients provided by the manufacturer (ThermoFisher Scientific, Waltham, MA). 3 mM MgCl 2 was added to both the protein and the nucleotide solutions before the titration experiments.
ITC experiments were performed using a MicroCal Auto-iTC200 System (Malvern Instruments, Malvern, UK). In each experiment, 0.2 mM nucleotide was transferred to the injection syringe, and 20 -25 M hNBD1 was transferred into the cell by the autosampler. Before being transferred, the samples were stored in a 96-well sample plate inside the sample holder, which was maintained at 5°C. The titrations were carried out at 10°C. Each titration experiment consisted of 10 injections of 2 l each, followed by six injections of 3 l each. The interval between injections was 300 s. The instrument feedback gain was set to "none" to reduce noise. Data analysis was performed using the MicroCal Origin 7.0 (OriginLab Corp., Northampton, MA) curve-fitting routines supplied with the instrument. The binding parameters, which included the stoichiometry (N), the binding constant (K), and the binding enthalpy change (⌬H), were determined by fitting the integrated heats using the Origin curve-fitting method for one set of identical binding sites. The software calculated the entropy change (⌬S) for the binding reaction, and the Gibbs free energy change (⌬G) was calculated externally using the relationship ⌬G ϭ ϪRT⅐ln(K), where R is the ideal gas constant and T is the absolute temperature during the ITC experiment (283.15 K).
We investigated possible causes for the low (Ͻ1:1) values of binding stoichiometry consistently obtained from our ITC experiments on hNBD1. We concluded this phenomenon is attributable to a portion of the protein population forming aggregates during these experiments that are incapable of binding nucleotides, and we performed analyses demonstrating that the thermodynamic binding parameters determined under such circumstances should remain accurate. Protein aggregation during our ITC experiments is consistent with the observation of some precipitate in the samples recovered from the cells at the end of the titrations and furthermore with the known biophysical properties of hNBD1. As summarized in the main text and described in detail in previous papers (10, 11,17), ATP binding strongly stabilizes this domain compared with a rapidly aggregating molten-globule conformation with minimal affinity for ATP. Because of this effect, the domain is purified in the presence of ATP, which has to be removed to perform an ITC experiment, and removal of ATP from the protein solution is expected to accelerate formation of protein aggregates that are incapable of binding ATP.
We ruled out alternative explanations related to errors in concentration determination or carryover of residual ATP from the buffer used to purify the protein. Nucleotide concentrations were determined based on UV absorbance and closely matched expected values based on calculations. Protein concentration before the ITC experiments was verified using the Pierce 660-nm protein assay reagent (ThermoFisher Scientific), and the purity of the protein as assessed by SDS-PAGE was Ͼ95%. Finally, the HPLC method described above demonstrated that the amount of ATP carryover was always less than 5% of the total protein concentration.
Importantly, loss of active protein due to aggregation during an ITC experiment is not expected to change the measured value of the binding enthalpy as long as the experiment is conducted in the proper regime in which the concentration of active protein molecules greatly exceeds the dissociation constant for the binding reaction, which was the case in all of our ITC experiments on hNBD1 (i.e. active protein concentrations greater than 8 M compared with dissociation constants less than 0.9 M). In this regime, nearly all incoming ligand molecules bind to protein molecules at the early points in the titration, meaning the enthalpy of binding is given directly by the measured value of kilocalories of heat/mol of injectant in this region. Thermodynamic parameters can only be inferred reliably from curve fitting of ITC experiments conducted in this regime in which the measured value of enthalpy at the outset of the titration is essentially independent of active protein concentration (82). To confirm the insensitivity of the thermodynamic parameters inferred from our ITC experiments to variations in active protein concentration, we performed curve fitting of our data assuming that 50% of the protein population was inaccessible for nucleotide binding, and these analyses yielded ϳ1:1 binding stoichiometry without significant changes in the values of the binding enthalpy or entropy or the dissociation constant compared with our baseline curve-fitting analyses assuming that all the protein in the sample was active. 13 Crystallization and X-ray structure determination F508-hNBD1⌬RI was purified using a modified version of the previously published protocol (10, 11). In brief, following Ni-NTA chromatography, dialysis was performed using an ATP-free buffer containing 150 mM NaCl, 30% (v/v) glycerol, 1 mM TCEP, and 20 mM Na-HEPES, pH 7.5, in order to remove ATP from the protein solution along with imidazole. Subsequent size-exclusion chromatography was conducted in SSB without any nucleotide. Either 2 mM dTTP or 2 mM ATP was added to the pooled protein from the size-exclusion column prior to concentration to ϳ10 mg/ml using Amicon Ultra-15 centrifugal filters. Co-crystallization of the concentrated protein with each of these nucleotides was conducted at 6°C in 72-well microbatch plates under 100% paraffin oil using a 3:1 volume ratio of one of these protein/nucleotide stock solutions to a precipitant solution containing 40% (v/v) PEG 400, 100 mM NH 4 Cl, and 100 mM MES, pH 6. After 5-7 days of equilibration at 4°C, each drop was streak-seeded from a previously formed crystal using a human hair. Protein crystals over 100 m long were mounted onto CrystalCap TM SPINE HT mounting pins using 15% (v/v) glycerol as cryoprotectant. The mounted crystals were snap-frozen in liquid nitrogen pending X-ray data collection.

Thermodynamic correction of F508del-CFTR
dTTP-bound or ATP-bound co-crystals were first transferred into a cryoprotectant solution in a clean 72-well microbatch plate covered with 100% paraffin oil. The cryoprotectant contained a 4 mM concentration of the magnesium complex of the target nucleotide dissolved in the precipitant solution with the addition of 15% (v/v) glycerol. After 24 h of incubation at 4°C in this initial soaking solution, the crystals were transferred into the same nucleotide-containing cryroprotectant solution in a clean well for an additional 7 days at 4°C. One additional weeklong incubation at 4°C was conducted with fresh nucleotidecontaining cryoprotectant in another clean well before the nucleotide-bound hNBD1⌬RI crystals were mounted and flash-frozen in liquid nitrogen.
X-ray diffraction data were collected using the X4C beamline at the National Synchrotron Light Source at Brookhaven National Laboratory or the BL14-1 beamline at the Stanford Synchrotron Radiation Lightsource. Diffraction data were indexed and integrated using HKL (84). Structure solution by molecular replacement, model-building, refinement, and structure evaluation were conducted using PHENIX (85). The published hNBD1⌬RI structure (PDB code 2PZE) was used as the molecular replacement search model.

Membrane isolation for electrophysiology
Membrane vesicles containing WT-hCFTR were prepared as described previously from BHK cells stably expressing this protein (41). Some modifications were introduced to allow BHK cells to stably express F508del-hCFTR (31). The cells were grown under standard conditions in a 150-mm dish up to 80% confluence before shifting to reduced temperature (28°C) for 48 h during which the small molecule corrector VX809 was added during the last 12 h. After treatment, cells were harvested by scraping, pelleted by brief centrifugation, and resuspended in ice-cold hypotonic lysis buffer containing 5 mM ATP. Following a 15-min incubation on ice, cells were lysed by 10 strokes in a tight-fitting Dounce homogenizer, followed by an additional 15 strokes after the addition of an equal volume of sucrose buffer. The resulting suspension was centrifuged at 100,000 ϫ g for 45 min to sediment microsomes that were then resuspended in phosphorylation buffer. The expression of the mature form of the rescued F508del-hCFTR protein (rF508del-hCFTR) was confirmed by immunoblotting. Membrane vesicles were phosphorylated by incubation with 50 nM PKA catalytic subunit in phosphorylation buffer containing 5 mM ATP for 15 min at room temperature. The membranes containing rF508del-hCFTR were always stored on ice and used for the functional assay the same day.

Planar-bilayer-based single-channel electrophysiology
Planar lipid bilayers were prepared by painting a 0.2-mm hole drilled in a Teflon cup with a phospholipid solution in n-decane containing a 3:1 mixture of 1-palmitoyl-2-oleoyl-sn-glycero-3phosphoethanolamine and 1-palmitoyl-2-oleoyl-sn-glycero-3phosphoserine. The lipid bilayer separated 1.0 ml of solution in the Teflon cup (cis side) from 5.0 ml of a solution in an outer glass chamber (trans side). Both chambers were magnetically stirred and thermally insulated. CFTR ion channels were transferred into the preformed lipid bilayer by spontaneous fusion of membrane vesicles containing the WT-hCFTR or VX809/temperature-rescued F508del-hCFTR (rF508del-hCFTR) in symmetrical salt solution. Single-channel currents were recorded at Ϫ75 mV under voltage-clamp conditions using an Axopatch 200B amplifier (Molecular Devices, LLC, Sunnyvale, CA). For analysis, the single-channel current was digitized by Digidata 1322 (Molecular Devices) with a sampling rate of 500 Hz and analyzed using pCLAMP 9.2 software (Molecular Devices). Origin 7.5 (OriginLab Corp., Northampton, MA) was used to fit all-point histograms by multiple peak Gaussians. Single-channel current was defined as the distance between peaks on the fitting curve and was used for the calculation of the single-channel conductance. The singlechannel P o was calculated as the ratio of the area under the peak for the open state to the total area under both peaks on the fitting curve.

Temperature-dependent electrophysiology assays
Heating and temperature control were established by the temperature control system TC2BIP (Cell Micro Controls, Norfolk, VA). The values of the temperatures in the bulk solution and in the vicinity of the single channel were confirmed when the open-state conductance was identical to that of the WT-hCFTR at the same bulk solution temperature (41) at thermal equilibrium. Possible differences in the gating kinetics of the WT versus F508del-hCFTR constructs were not taken into account.
Thermal stability of an hCFTR construct was defined based on its ability to demonstrate single-channel function with open-state conductance identical to the WT-hCFTR under the same experimental conditions for a 10-min period following an initial 10 min of incubation at the indicated temperature. Consistent with this definition, each recording in Fig. 6B shows the last 2 min of a 20-min recording at 30°C, demonstrating a dramatic difference in the thermal stability of rF508del-hCFTR in the presence of a 2 mM concentration of different nucleotide analogs.
As reported previously, rF508del-hCFTR produced in BHK cells using the protocol employed here exhibits several gating modes in black lipid membranes, including a fast-flickering mode (FFM) with an unstable and less conductive open state (41). The FFM is a hallmark of thermal instability that becomes apparent at the elevated temperature of 30°C, and it was interpreted to represent an initial step in the functional inactivation of rF508del-hCFTR in black lipid membranes (41). The FFM has not been reported in experiments conducted on similar F508del-hCFTR protein constructs in excised membrane patches, although these experiments confirm that the mutation reduces functional protein lifetime (52)(53)(54)(55), and some of them show evidence of an open-channel state with reduced conductance (53). The failure to observe the FFM in studies on excised membrane patches could be attributable to differences in protein preparation procedures, differences in the lipid composition in the membranes used for the electrophysiological measurements, or differences in the observation protocol. Our rF508del-hCFTR preparations typically show normal openstate conductance at the beginning of experiments conducted at room temperature prior to the appearance of the FFM later Thermodynamic correction of F508del-CFTR during the 20-min time period over which measurements are conducted.

Relationship between the energy of hNBD1-hNBD2 interface formation and P o-max
Previous research has demonstrated that hNBD1 is catalytically inactive (17,18,86,87). Therefore, assuming diphosphate/ triphosphate exchange occurs rapidly on hNBD2 compared to nucleotide hydrolysis, both hNBD1 and hNBD2 will have NTPs bound the vast majority of the time in the presence of a saturating concentration of NTP, which is the condition that yields P o-max in the Eadie-Hofstee analysis. Because channel opening and closing rates have both been shown to be independent of NTP concentration under saturating conditions (88 -90), the dynamics of channel gating at saturation are well approximated as a simple two-state equilibrium in which individual hCFTR channels cycle between an open conformation with two triphosphates encapsulated in the hNBD1-hNBD2 interface (hCFTR OPEN ⅐NTP 2 ) and a closed conformation in which hNBD1 and hNBD2 are spatially separated but still both have bound NTPs (hCFTR CLOSED ⅐NTP 2 ), as shown in Reaction 1 above. This gating model assumes that, at a saturating concentration of a given NTP, all hCFTR molecules have that NTP species bound to both hNBD1 and hNBD2 the vast majority of the time and that the observed differences in the open probability (P o-max ) reflect different proportions of channels occupying the open ([hCFTR OPEN ⅐NTP 2 ]) versus closed ([hCFTR CLOSED ⅐NTP 2 ]) conformational states when that NTP is bound. The different proportions in the two states reflect variations in the equilibrium constants (K hNBD1-hNBD2 ) and corresponding Gibbs free energy changes (⌬G 0 hNBD1-hNBD2 ) for the conformational transition when the different NTPs are bound to the protein, as quantified in Equations 4-6.
(1/ c )[hCFTR OPEN  The parameters o and c in in Equations 4-6 above are the time constants for channel opening and closing, respectively. The population ratio (P o-max /(1 Ϫ P o-max )) from single-channel recordings corresponds to K hNBD1-hNBD2 because, according to basic statistical mechanical principles, the fraction of time an individual channel spends in the open versus closed conformation will match the steady-state distribution of open versus closed channels in a population of hCFTR molecules at equilibrium. Although the individual time constants o and c depend on the activation energy for channel opening, which should not influence net channel stability, ⌬G 0 hNBD1-hNBD2 depends on their ratio, which is independent of activation energy and contributes directly to net channel stability according to the thermodynamic scheme in Fig. 1.
The assumption underlying this model concerning rapid diphosphate/triphosphate exchange on hNBD2 following hydrolysis must be correct for dATP and dGTP, the two NTPs giving P o-max values close to 1.0 (Fig. 6D), because opening of the hCFTR channel requires an NTP to be bound to both hNBD1 and hNBD2 (4,5). Therefore, an open probability close to 1 can only be achieved if hNBD2 has an NTP bound in its active site close to 100% of the time, which requires rapid replacement of the diphosphate product by the triphosphate substrate following each round of hydrolysis. Our assumption that the other six NTPs characterized in Fig. 6 undergo comparably rapid diphosphate/triphosphate exchange on hNBD2 even though they give lower P o-max values seems reasonable based on several considerations. These NTPs have Gibbs free energies of binding to isolated hNBD1 that differ by only Ϫ1.0 to ϩ0.5 Ϯ 0.2 kcal/mol at 10°C compared with dATP (Table  S1), making it unlikely that they have dramatic differences in affinity for hNBD2 given the strong homology of its nucleotidebinding site to that in hNBD1. More importantly, the NTPs showing lower values of P o-max have effective binding affinities for the open state of the hCFTR channel, as given by the slopes in the Eadie-Hofstee plots in Fig. 6D, that vary from 2.2-fold higher to 16-fold lower than that of dATP, the species giving the highest P o-max . The observation that none of the NTPs has more than a slightly higher effective affinity for hCFTR than the rapidly exchanging nucleotide dATP makes it unlikely that they undergo much slower diphosphate/triphosphate exchange on hNBD2 than dATP. These observations provide support for our assumption that all of the NTPs characterized in Fig.  6 undergo rapid diphosphate/triphosphate exchange on hNBD2 following hydrolysis.