Conformational changes in the catalytically inactive nucleotide-binding site of CFTR

CFTR (ABCC7; Riordan et al., 1989) is a member of the asymmetric C subfamily of ATP-binding cassette (ABC) proteins. Most ABC proteins are active transporters that move a variety of substrates across biological membranes. The substrate translocation pathway, formed by two transmembrane domains (TMDs), cycles between inward- and outward-facing conformations driven by an ATP hydrolysis cycle catalyzed by the two cytosolic nucleotide-binding domains (NBDs). ABC NBDs contain three highly conserved sequence motifs: the Walker A (GXXXXGKS/T) and B (ΦΦΦΦDE, with Φ representing hydrophobic residues), and the ABC signature motif (LSGGQR/K). In each catalytic cycle, ATP binding to the Walker motifs of both NBDs promotes formation of a tight head-to-tail NBD dimer that occludes two ATP molecules in composite interfacial catalytic sites formed by the Walker motifs of one NBD and the signature motif of the other. ATP hydrolysis is required to disrupt this dimer to allow nucleotide exchange and initiation of a new cycle (Hollenstein et al., 2007). Members of the C subfamily share this overall architecture but show a marked asymmetry in the primary sequences of their two NBDs. Because of noncanonical substitutions in the Walker B and in other conserved motifs of the N-terminal NBD (NBD1) and in the signature sequence of the C-terminal NBD (NBD2), the catalytic site containing these motifs (site 1) is inactive (Aleksandrov et al., 2002; Basso et al., 2003), leaving ABC-C proteins with a single catalytically active site (site 2, formed by Walker motifs of NBD2 and signature sequence of NBD1). Noncanonical substitutions in the same key positions, resulting in reduced or absent hydrolysis at one composite site, are also seen in the clinically important transporter associated with antigen processing (formed by TAP1 and TAP2 of the human ABC-B subfamily; Procko et al., 2009), as well as in numerous yeast and prokaryotic transporters (Szollosi et al., 2011). The latter group includes TM287/288, a heterodimeric transporter from Thermotoga maritima, whose high resolution structure has been recently solved (Hohl et al., 2012).

CFTR, the protein mutated in cystic fibrosis patients (Riordan et al., 1989), is the only member of the ABC-C family whose TMDs are known to form the pore of a Cl⁻ ion channel. In addition to the canonical ABC domains, CFTR contains a unique regulatory (R) domain, phosphorylation of which by cAMP-dependent protein kinase (PKA) is a prerequisite for channel gating (Anderson et al., 1991). In phosphorylated CFTR, the pore opens to a burst upon ATP-induced dimerization of its NBDs and closes from a burst upon dimer dissociation (Vergani et al., 2005) after ATP hydrolysis at site 2; thus, in WT CFTR, each gating cycle (or burst of openings) is coupled to hydrolysis of one ATP molecule (Csanády et al., 2010). Deviations from this strict 1:1 coupling between ATP hydrolysis events and bursts of openings have been suggested to happen in mutants or in the presence of drugs. Catalytic site mutations can lower this coupling ratio leading to bursts that involve ATP occlusion/deocclusion but no hydrolysis (Csanády et al., 2010). In contrast, [ATP]-dependent prolongation of open burst durations for the W401F CFTR mutant (Jih et al., 2012b), or for WT CFTR gating in the presence of the potentiator compound Vx-770 (Jih and Hwang, 2013), has led to the suggestion that post-hydrolytic ADP/ATP exchange in site 2 might occur without intervening pore closure, thereby increasing the coupling ratio by allowing more than one ATP to be hydrolyzed in site 2 within a single gating cycle (Jih et al., 2012a).

Biochemical work using γ-³²P-labeled ATP analogues revealed that ATP remains bound at CFTR’s NBD1 for periods much longer (several minutes) than a channel gating cycle (∼1 s), without being hydrolyzed (Aleksandrov et al., 2002; Basso et al., 2003). In a recent study (Tsai et al., 2010), nucleotide exchange time constants at the two sites were measured using a functional approach. The high affinity ATP analogue N⁶-(2-phenylethyl)-ATP (P-ATP) was found to speed the opening and slow the closing of CFTR channels, but the onset of these two effects upon sudden replacement of ATP with P-ATP in the bath (cytosolic) solution followed distinct time courses. Whereas the effect on opening rate appeared instantaneously, suggesting that it was caused by P-ATP binding to the rapid-turnover site 2, the onset of slowed closure followed with a delay of 30–50 s, consistent with ATP/P-ATP exchange occurring more slowly at degenerate site 1. Furthermore, because the nucleotide exchange rate at site 1 was affected by mutations in the NBD2 signature sequence, which completes site 1 only in a formed dimer, the authors suggested only partial separation of the NBD dimer in closed channels, with the interface remaining closed around site 1 for several gating cycles. This notion obtained further support from thermodynamic studies (Szollosi et al., 2011) that examined state-dependent changes in energetic coupling between pairs of residues on opposing faces of the site-1 interface. A lack of change in energetic coupling for three such pairs was interpreted to be consistent with the model suggested by Tsai et al. (2010), i.e., with an “immobile” site 1 remaining closed throughout the gating cycle.

Nevertheless, site 1 clearly does play a role in gating energetics, as perturbations at site 1 can profoundly alter channel gating kinetics. For instance, the K464A mutation, which—by removing the side chain of the conserved NBD1 Walker A lysine—perturbs the NBD1 side of site 1, dramatically alters the mechanism of gating (Csanády et al., 2010). In addition, the H1348A mutation in the NBD2 signature sequence, which perturbs the NBD2 side of site 1, was found to slow channel closure (Szollosi et al., 2011). Finally, gating is affected by perturbations of the nucleotide structure at site 1, as P-ATP bound at site 1 also slows channel closure (Tsai et al., 2010). However, the exact mechanisms by which the H1348A mutation, or P-ATP bound at site 1, affects gating have not yet been elucidated. To gain further insight into the energetic consequences of perturbations at site 1, we examined in detail the effects of these two perturbations on CFTR channel gating. Our observations suggest that, despite maintained contact between residues on opposite sides of site 1, this site does not remain “frozen”; rather, significant conformational changes are likely to take place there both during channel opening and concomitant with ATP hydrolysis at site 2.

Human WT CFTR and CFTR segment 433–1480 in the pGEMHE plasmid (Chan et al., 2000) served as templates for mutants H1348A, K1250A, E1371S, K1250A/H1348A, E1371S/H1348A, E1371S/K464A, and 433–1480(K1250A), which were created using the QuikChange kit (Agilent Technologies). The entire coding sequence of each construct was verified by automated sequencing (LGC Genomics). T7 polymerase was used for in vitro transcription (Mmessage kit; Ambion), and purified cRNA was stored at −80°C.

Xenopus oocytes were extracted and treated with collagenase as described previously (Chan et al., 2000). Isolated oocytes were stored at 18°C in a solution containing (mM) 82 NaCl, 2 KCl, 1 MgCl₂, 5 HEPES, pH 7.5 with NaOH, 1.8 CaCl₂, and 50 µg/ml gentamycin. To obtain expression levels appropriate for single-channel or macroscopic recordings, oocytes were microinjected with 0.1–10 ng of CFTR cRNA, and further incubated at 18°C for 2–3 d.

For inside-out patch recordings, the pipette solution contained (mM) 136 NMDG-Cl, 2 MgCl₂, and 5 HEPES, pH 7.4 with NMDG. The bath solution was continuously flowing and contained (mM) 134 NMDG-Cl, 2 MgCl₂, 5 HEPES, and 0.5 EGTA, pH 7.1 with NMDG. MgATP (Sigma-Aldrich) was added from a 400-mM aqueous stock solution (pH 7.1 with NMDG) to achieve a final concentration of 2 mM (or 10 mM for channels bearing the K1250A mutation). A 10-mM aqueous stock solution of P-ATP Na⁺ salt (BIOLOG Life Science Institute) was stored at −80°C and diluted into the bath solution immediately before recording to achieve a final concentration of 10 µM (or 50 µM for K1250A mutants). CFTR channels were fully activated by a 1–2-min cytosolic exposure to 300 nM of catalytic subunit of PKA (Sigma-Aldrich); all experiments shown were done in the partially dephosphorylated state after PKA removal, which remains stable over the time course of several minutes (Csanády et al., 2000). Switching between various bath solutions was achieved using computer-controlled electronic valves (HEKA); solution exchange time constant was 20–50 ms. Experiments were performed at 25°C. Unitary CFTR currents were recorded (Axopatch 200B; Molecular Devices) at a pipette holding potential of +80 mV (V_m = −80 mV), filtered at 2 kHz, and digitized at 10 kHz (Digidata 1322A, Pclamp9; Molecular Devices). As CFTR gating is largely voltage independent (Cai et al., 2003), macroscopic currents were recorded at membrane potentials between −20 and −80 mV.

Mean burst and interburst durations were extracted from steady segments of record with one to six active CFTR channels as described previously (Csanády et al., 2000). Currents were digitally filtered at 100 Hz and idealized by half-amplitude threshold crossing. Events lists were fitted with a simple model in which ATP-dependent slow gating is pooled into a closed-open scheme and ATP-independent brief closures are modeled as pore-blockage events (Ishihara and Welsh, 1997). Rate constants (r_CO, r_OC, r_OB, and r_BO) of the resulting closed–open–blocked (C↔O↔B) scheme were extracted by a simultaneous maximum likelihood fit to the dwell-time histograms of all conductance levels while accounting for the filter dead time (Csanády, 2000). Mean burst duration was then calculated as (1/r_OC)(1 + r_OB/r_BO) (Figs. 1 F and 2 C), and mean interburst duration was calculated as 1/r_CO (Figs. 1 G and 2 D). The estimates of interburst duration, and consequently of channel open probability (P_o), are sensitive to correct estimation of the number of channels (N) in the patch. These parameters were therefore used only if N could be determined with reasonable confidence; for this subset of patches, N was ≤5.

To extract the gating pattern of channels loaded with either ATP or P-ATP at both sites, segments immediately following nucleotide exchange were not included in our analysis. The starting point for idealization was assigned individually for each patch, based on visual inspection, with analysis restricted to segments in which the gating pattern had already stabilized. The time required for stabilization after nucleotide exchange was typically between 10 and 60 s. To facilitate nucleotide loss from site 1, believed to happen faster in closed channels (Tsai et al., 2010), in some cases patches were rinsed with nucleotide-free solution for 20–30 s before applying a different nucleotide (e.g., Figs. 4 and 6).

Burst analysis was done as described previously (Csanády et al., 2010). In brief, open bursts from records with a single active channel were isolated by ignoring closures shorter than a cutoff, chosen using the method of Jackson et al. (1983). The distributions of the durations of bursts obtained in this way were fitted using maximum likelihood (Colquhoun and Sigworth, 1995) to either a single-exponential distribution (Fig. 7, A and B, blue dotted lines) or to the scheme in Fig. 7 C, with the slow rate k_-1 fixed to 0 (Fig. 7, A and B, red solid lines); events shorter than t_low = 12 ms were excluded from the analysis. The probability density functions of the two fitted models are given by f(k, t) = k · exp(−kt) and f(k₁, k₂, t) = (k₁k₂/(k₂ − k₁)) · (exp(−k₁t) − exp(−k₂t)), respectively; the improvement of the fit caused by introduction of the second free parameter was evaluated using the log-likelihood ratio test and found significant for both distributions in Fig. 7 (A and B).

Macroscopic current decay time courses were fitted with single-exponential functions using nonlinear least squares (pClamp9), and the closing rate was defined as the inverse of the fitted time constant. For the nonhydrolytic mutants (K1250A, E1371S, and double mutants), the decay time courses after nucleotide removal often required a double-exponential function—of the form I(t) = I₀(A₁exp(−t/τ₁) + (1 − A₁)exp(−t/τ₂))—with two slow time constants for a satisfactory fit (e.g., Fig. 3, B and E), suggesting the presence of two populations of open-channel bursts. In such cases, the average steady-state burst duration (τ*) was estimated as τ* = τ₁τ₂/(A₁τ₂ + A₂τ₁), and the average closing rate was defined as 1/τ* (Szollosi et al., 2011).

Mutant cycle analysis (Fersht, 2002) was used to give a measure of the energetic coupling (ΔΔG_int) between pairs of target structural elements, as described previously (Vergani et al., 2005; Szollosi et al., 2010, 2011). In brief, perturbation-induced changes (between corners i and j of a mutant cycle) in activation free energy barriers (ΔΔG^‡) for closing were calculated from the changes in (hydrolytic and nonhydrolytic) closing rates (r) as ΔΔG^‡_j−i = −RTln(r_j/r_i), and ΔΔG_int was defined as the difference between ΔΔG^‡ values along parallel sides of the mutant cycle (e.g., ΔΔG^‡₄₋₃ − ΔΔG^‡₂₋₁). All ΔGs are given as mean ± SEM. Because the numbers of observations, n, for each corner of the cycle are similar, SEMs were calculated using the mean value for n.

Data are given as mean ± SEM of at least five measurements. Statistical significance was evaluated using Student’s t test (*, P < 0.05; **, P < 0.01).

Fig. S1 compares single-channel conductances in ATP versus P-ATP. Fig. S2 shows the effect of the K464A mutation on nonhydrolytic closing rate measured in the E1371S background. Figs. S1 and S2 are available.

We first determined rough kinetic parameters for CFTR channels gating in P-ATP in our experimental system. Apparent affinities for ATP and P-ATP were compared in macroscopic patches by applying test concentrations of the nucleotide bracketed by applications of a saturating (2 mM and 32 µM, respectively) concentration (Fig. 1, A and B). Fractional currents plotted against nucleotide concentration (Fig. 1 C) were well fit by the Michaelis–Menten equation and yielded K_m values of 40 ± 2 µM for ATP but 2.4 ± 0.2 µM for P-ATP (Fig. 1 C, blue and red). We next compared steady-state gating of CFTR channels exposed either to a close-to-saturating concentration of 10 µM P-ATP or to 2 mM ATP, in patches containing a few (≤2) channels in which individual gating transitions remained clearly resolved (Fig. 1 D). Kinetic analysis (see Materials and methods) revealed significantly higher P_o in P-ATP as compared with that seen in ATP (Fig. 1 E), which was largely caused by approximately twofold longer mean burst durations (Fig. 1 F) together with a small reduction in interburst durations (Fig. 1 G). These results confirm earlier studies that have established P-ATP as a high affinity, potent CFTR stimulator (Zhou et al., 2005; Tsai et al., 2010).

In an earlier study, we found that the H1348A mutation prolongs burst durations of CFTR channels gating in 2 mM ATP by approximately threefold (Szollosi et al., 2011). Because for another site-1 mutant (W401F) mean burst durations were shown to increase at high millimolar ATP concentrations (Jih et al., 2012b), we tested whether that was the case for H1348A CFTR. However, comparing the patterns of gating in 2 and 10 mM ATP (Fig. 2 A) did not reveal any significant effect of raising [ATP] above 2 mM; kinetic analysis revealed identical P_o (Fig. 2 B), mean burst (Fig. 2 C), and interburst (Fig. 2 D) durations under these two conditions.

To quantitatively compare the effects of our site-1 perturbations on the slow rate of nonhydrolytic closure, we studied macroscopic closing rates after nucleotide removal for channels in which nucleotide hydrolysis at site 2 is abrogated by mutagenesis. Because different NBD2 mutations, in addition to disrupting ATP hydrolysis, likely differentially alter the stability of the open state, we do not know which nonhydrolytic mutant is the best model for the O₁ state of WT CFTR. We therefore compared the effects of our site-1 perturbations on the closing rates of two nonhydrolytic mutants, NBD2 Walker A mutant K1250A (Fig. 3, A–C) and NBD2 Walker B mutant E1371S (Fig. 3, D–F). Both P-ATP (Fig. 3, A and D, red fit lines and time constants) and the H1348A mutation (Fig. 3, B and E, green fit lines and time constants) caused markedly slower current relaxations than those observed for the two background constructs upon removal of ATP (Fig. 3, A and D, blue fit lines and time constants; cf. Zhou et al., 2005; Szollosi et al., 2011). Thus, the nonhydrolytic closing rate (Fig. 3, C and F, bars; calculated from the fitted relaxation time constants as described in Materials and methods) was similarly affected by both site-1 perturbations, and this was true regardless of whether the K1250A (Fig. 3 C) or the E1371S (Fig. 3 F) mutant was chosen as the nonhydrolytic model; i.e., both site-1 perturbations decreased this rate by two- to threefold.

The slowing of both hydrolytic and nonhydrolytic closure by P-ATP was suggested to result from occupation of site 1 by this nucleotide, based on the delayed onset of both effects upon application of the analogue (Tsai et al., 2010). We used thermodynamic mutant cycles (Fersht, 2002) as an independent approach to test this hypothesis. Nonadditive effects on gating of two structural perturbations suggest that those two structural elements are energetically coupled. Channels opened by P-ATP contain two structural “perturbations”: a phenylethyl group (P group) linked to N⁶ of the adenine of the nucleotide bound in site 1, and another P group on the nucleotide bound in site 2. If the presence of the P group in site 2 affects closing rate, then this effect is expected to be additive with effects on closing rate of mutations in the spatially distant site 1. In contrast, if the P group in site 1 is responsible for the effects of P-ATP on closing, then P-ATP and mutations of site-1 residues (which are energetically coupled to this P group) will affect closing rate in a nonadditive manner. We chose two site-1 perturbations to test for additivity of effects on closing rate with those of P-ATP. On the NBD2 side of site 1 we chose mutation H1348A, because this perturbation causes a large effect on closing rate in ATP, which we had already characterized (Fig. 2). To perturb the NBD1 side of site 1, we chose to delete most of a nonconserved insertion in CFTR’s NBD1 (called the “regulatory insertion” [RI]; Lewis et al., 2004) by coexpression of CFTR segments 1–414 and 433–1480. This “ΔRI” perturbation leaves ATP-dependent gating largely intact (Csanády et al., 2005) but is nevertheless expected to cause local distortion in areas of NBD1 that contact the adenine base of the site-1–bound nucleotide.

We first characterized the effect of P-ATP on hydrolytic closing rate in these two mutant backgrounds (Fig. 4) by recording single-channel activity of H1348A (Fig. 4 A), or of ΔRI channels (Fig. 4 D), alternately exposed to either 2 mM ATP (Fig. 4, A and D, green segments) or 10 µM P-ATP (Fig. 4, A and D, brown segments). Consistent with previous work, the H1348A and ΔRI perturbations differentially affected hydrolytic closure: the closing rate in ATP was unchanged for ΔRI (Fig. 4 E; compare green bar with blue bar; cf. Csanády et al., 2005), but two- to threefold slowed for H1348A (Fig. 4 B; compare green bar with blue bar; cf. Fig. 2 C). However, switching from ATP to P-ATP (Fig. 4, A and D, brown segments) did not elicit any noticeable change in the length of channel open bursts for either construct (Fig. 4, A and D; compare brown with green segments in expanded insets). Thus, P-ATP does not slow hydrolytic closure when applied in the H1348A or ΔRI mutant background (Fig. 4, B and E; compare brown with green bars). Although the twofold effect of P-ATP on hydrolytic closing rate of WT CFTR is small in energetic terms, the difference in the effects of the mutations on closing rates in P-ATP versus those in ATP yielded interaction free energies (ΔΔG_int(closing))—both between the P group and residue 1348 (Fig. 4 C) and between the P group and the RI region (Fig. 4 F)—which were significantly different from zero (P < 0.05). This is consistent with the effect on hydrolytic closing rate being caused by the P group in site 1.

Additivity of effects on nonhydrolytic closure of the same site-1 perturbations with those of P-ATP was tested in the K1250A nonhydrolytic background (Fig. 5) by measuring the macroscopic closing rate of K1250A/H1348A channels (Fig. 5 A), and of channels obtained by coexpression of segments 1–414 and 433–1480(K1250A) (K1250A/ΔRI; Fig. 5 D) upon nucleotide removal. The H1348A and ΔRI perturbations affected nonhydrolytic closing rate in opposite ways; whereas mutation H1348A slowed it by approximately threefold (Fig. 5 B; compare green bar with blue bar; cf. Fig. 3 C), the ΔRI perturbation accelerated it by approximately threefold (Fig. 5 E; compare green bar with blue bar; cf. Csanády et al., 2005). Interestingly, neither perturbation abolished the effect of P-ATP on nonhydrolytic closure (Fig. 5, A and D; compare brown with green single-exponential fit lines and time constants); rather, both perturbations seemed to potentiate it. Thus, whereas P-ATP slowed nonhydrolytic closure by approximately fourfold for K1250A channels with an intact site 1 (Fig. 3 C; compare red with blue bar; replotted in Fig. 5, B and E), this effect increased to greater than sixfold and to approximately ninefold, respectively, in the presence of the site-1 perturbations H1348A and ΔRI (Fig. 5, B and E; compare brown with green bars), again suggesting nonadditivity. Indeed, mutant cycles built on nonhydrolytic closing rates (Fig. 5, C and F) yielded interaction free energies between the P group and residue 1348 (Fig. 5 C), as well as between the P group and the RI region (Fig. 5 F), significantly different from zero (P < 0.05 and P < 0.01, respectively), consistent with the effect on nonhydrolytic closing rate also being caused by the P group in site 1.

Both P-ATP and the H1348A mutation prolong steady-state mean burst durations (Figs. 1 F and 2 C). Recent studies suggested a novel mechanism for a prolongation of steady-state CFTR burst durations for NBD1 mutant W401F (Jih et al., 2012a,b), by proposing the presence of a short time window at the end of each burst during which the hydrolysis products ADP and phosphate at site 2 can be replaced by a new ATP molecule without intervening channel gate closure. Such a “reentry” mechanism should allow for more than one ATP hydrolysis cycle to happen within a single burst, thereby prolonging the mean duration of bursts at steady state. A simply testable prediction of such a model is a discrepancy between the burst durations measured at steady state and the time constant of the macroscopic current relaxation upon sudden removal of ATP. This is because the latter closing rate is measured under conditions that preclude a reentry cycle, i.e., in the absence of nucleotides. Because the W401F mutation also resides in site 1, we asked whether a reentry mechanism might explain the longer burst durations in P-ATP or in the presence of the H1348A mutation. To this end, we measured macroscopic closing rates after nucleotide removal for WT and ΔRI channels opened by either 2 mM ATP or by 10 µM P-ATP (Fig. 6, A and C), and for H1348A channels opened by 2 or 10 mM ATP (Fig. 6 B, top) or by 10 µM P-ATP (Fig. 6 B, bottom). Current relaxation time courses were fitted by single exponentials (Fig. 6, A–C, colored lines). The time constants of these exponentials reflect the mean duration of bursts, which do not contain reentry events. When we compared these time constants (Fig. 6 D, colored bars) with the steady-state mean burst durations measured in the presence of the respective nucleotide (Fig. 6 D, striped bars; data replotted from Figs. 1 F, 2 C, and 4, B and E), we found in each case a very close agreement; importantly, in no case was the steady-state burst duration longer than the macroscopic estimate. This suggests that both for WT (and ΔRI) channels gating in P-ATP, and for H1348A channels gating in 2 or 10 mM ATP or 10 µM P-ATP, each steady-state burst involves occlusion of a single nucleotide at site 2, just as suggested for WT channels gating in ATP (Csanády et al., 2010).

Of note, enhancement of macroscopic WT CFTR current by P-ATP, relative to ATP, was slightly smaller (<1.5-fold; see Fig. 6 A) than expected from the fractional increase in P_o (∼1.55-fold; Fig. 1 E). This small discrepancy is accounted for by the slightly smaller average conductance of channels opened by P-ATP, which likely reflects rapid flickery closures not fully resolved at our recording bandwidth (Fig. S1). Thus, at −80 mV, average unitary current of open channels was −0.59 ± 0.01 pA (n = 6) in ATP but significantly (P < 0.01) smaller, −0.55 ± 0.01 pA (n = 6), in P-ATP.

Thus, under the conditions studied here, the duration of each open burst includes some time spent in a prehydrolytic open state (O₁; Fig. 7 C) in which a nucleoside triphosphate is occluded at site 2, followed by a shorter time interval in a less stable post-hydrolytic open state (O₂; Fig. 7 C). Because the rate (k_-1) of nonhydrolytic closure (step O₁→C₁; Fig. 7 C) is very slow (see Fig. 3), the overall mean burst duration is mostly determined by the rates of the O₁→O₂ and O₂→C₂ transitions (k₁ and k₂; Fig. 7 C). The latter two rates can be estimated by fitting the scheme in Fig. 7 C to the distributions of open burst durations using maximum likelihood (Colquhoun and Sigworth, 1995). To determine which of these two rates is affected by P-ATP and the H1348A mutation, respectively, we recorded currents from patches containing a single active channel (Fig. 7, A and B, insets). Open burst events were isolated by burst analysis (see Materials and methods), pooled from several patches, and fitted by maximum likelihood. The histograms of burst durations were distinctly peaked for both WT channels gating in 10 µM P-ATP (Fig. 7 A) and H1348A channels gating in 2 mM ATP (Fig. 7 B), consistent with a non-equilibrium gating cycle that involves nucleotide hydrolysis. The most striking difference between these distributions and that found earlier for WT channels gating in ATP (Csanády et al., 2010) was in the tail, which was distinctly longer in the presence of the site-1 perturbations, whereas the positions of the peaks were little moved. Accordingly, the maximum likelihood fit yielded two- to threefold slower k₁ values (Fig. 7 D, left) in the presence of P-ATP (red bar) or the H1348A mutation (green bar) compared with the earlier estimate for WT channels gating in ATP (blue bar; replotted from Csanády et al., 2010). In contrast, neither P-ATP nor the H1348A mutation seemed to dramatically affect the faster rate k₂ (Fig. 7 D, right).

In this paper, we have investigated in detail the effects on channel gating of two structural perturbations at CFTR’s degenerate interfacial ATP-binding site (site 1). The H1348A mutation removes a histidine side chain from the signature sequence of NBD2, thereby perturbing the NBD2 side of site 1, whereas the use of P-ATP introduces a phenylethyl group into this composite binding site. Although in our experiments, which involved prolonged exposure to P-ATP, both composite sites were presumably loaded with this ATP analogue, an elegant study by Tsai et al. (2010) convincingly demonstrated that the effect of P-ATP on burst length (compare Fig. 1 F) is caused by the P-ATP molecule bound at site 1, whereas P-ATP bound at site 2 is responsible for the high affinity, potent stimulation of opening rate (compare Fig. 1, B and C). In addition, experiments using the nonhydrolyzable ATP analogue AMPPNP together with P-ATP to lock WT channels in the open state suggested that the slowing of nonhydrolytic closure by P-ATP is also attributable to P-ATP bound at site 1 (Zhou et al., 2005). Here, we provide further independent support for this assignment by demonstrating nonadditive effects on both the hydrolytic and nonhydrolytic closing rate of P-ATP and of mutations on either face of composite site 1 (Figs. 4 and 5). Furthermore, combining maximum likelihood fitting of steady-state single-channel records with the analysis of macroscopic current relaxations, we have determined which microscopic rates are affected by these site-1 perturbations resulting in the increased stability of channel open states.

In two recent studies, Jih et al. (2012a,b) proposed the existence of a short time window at the end of each burst, which would follow the release of hydrolysis products from site 2 but precede gate closure, and allow a new ATP molecule to bind at site 2, returning the channel into a new prehydrolytic open state without intervening gate closure. They further proposed that mutations or pharmacological modulation might prolong CFTR channel bursts by increasing the frequency of such “reentry” events (Jih et al., 2012a,b; Jih and Hwang, 2013). Because one such mutation (W401F) resides in site 1, we have evaluated the possibility that the H1348A mutation, or P-ATP bound in site 1, might also act by such a mechanism. However, the lack of effect of increasing [ATP] on steady-state burst durations of H1348A CFTR (Fig. 2 C) and the close agreement of macroscopic closing time constants upon nucleotide removal with steady-state mean burst durations (Fig. 6) ruled out this possibility and confirmed that the two site-1 perturbations studied here do not disrupt the near 1:1 stoichiometry between ATP occlusion events and pore opening events, characteristic to WT CFTR gating in ATP (Csanády et al., 2010). Although comparison of macroscopic and steady-state closing rates (Fig. 6) might not be sensitive enough to rule out slight deviations from 1:1 stoichiometry, it is clear that small effects would not be sufficient to explain the two- to threefold prolongation of burst durations by the H1348A mutation and P-ATP. We have therefore used a simplified cyclic scheme (Fig. 7 C) as a framework for interpreting our kinetic results and have dissected the effects of site-1 perturbations on the three rates that determine the stability of the open states: those of steps O₁→O₂, O₁→C₁, and O₂→C₂ (Fig. 7 C).

Interestingly, the detailed kinetic consequences of these two perturbations turned out to be very similar, amounting to a simultaneous two- to threefold decrease in the rates of steps O₁→O₂ and O₁→C₁. This phenotype affords a simple explanation in terms of the free energy profile of the CFTR gating cycle (Fig. 8, blue energy profile). For a structural perturbation that selectively stabilizes the O₁ state, relative to both states C₁ and O₂ (by δΔG_O1; Fig. 8, red energy profile), a simultaneous increase in the heights of the transition-state barriers for both step O₁→O₂ (ΔG^‡₁; Fig. 8) and step O₁→C₁ (ΔG^‡₋₁) is expected (compare red and blue vertical double arrows in Fig. 8), resulting in slowed rates k_-1 and k₁, as observed. In addition, the distortion of the energy profile, as drawn, also suggests a lower energy barrier for step C₁→O₁, predicting an increase in opening rate. Interestingly, this is exactly what we observed for the H1348A mutant, the opening rate of which is approximately twofold increased relative to WT (compare Fig. 2 D, green bar, with Fig. 1 G, blue bar). We also observed a more modest increase in WT opening rate in the presence of 10 µM (subsaturating) P-ATP (compare red with blue bar in Fig. 1 G), but interpretation of this small effect is limited by the fact that stimulation of opening rate by P-ATP has been mostly attributed to P-ATP bound at site 2 (Tsai et al., 2010).

From an energetic point of view, the effects of the site-1 perturbations we have studied are modest, on the order of ∼1 kT. However, two facts have to be considered when interpreting these effects. First, the measured ΔΔG values by no means represent the overall energetic change of the site-1 structure between states C₁ and O₁, or O₁ and O₂, but the change in environment of a single chemical moiety (the histidine side chain at position 1348 or the phenylethyl group of P-ATP in site 1) associated with those gating transitions. Second, our ΔΔG values reflect changes in barrier heights, which suggest larger changes between ground states (see Fig. 8); e.g., an ∼1-kT change in the barrier height for nonhydrolytic closure might be associated with a ΔΔG of ∼2 kT between states C₁ and O₁. (As a comparison, ΔΔG_int between C₁ and O₁ is ∼2 kT for residues 555 and 1246 on opposing faces of the active site 2, which completely separates during the O₁→C₁ transition [Vergani et al., 2005].)

What mechanistic insight can be gained from these observations? The recent suggestion that composite site 1 of CFTR does not separate upon channel closure (Tsai et al., 2010; Szollosi et al., 2011) has fostered a view of the degenerate site as a simple rigid scaffold, which does not participate in the functional movements of asymmetric ABC-C proteins (although this proposal has remained a matter of contention; Artur et al., 2011). Our present data are hard to reconcile with such an assignment. The fact that removal of the histidine side chain at position 1348 selectively stabilizes state O₁ relative to C₁ (Fig. 8) is most simply explained by postulating that in the WT channel, the environment of that histidine side chain is not identical in those two states. The similar effect of P-ATP bound at site 1 suggests that the phenylethyl group also experiences a change in environment during the C₁↔O₁ transition. It follows then that significant movement is likely to take place at site 1 during the channel opening step, even if the composite site does not actually separate in a closed channel. Furthermore, the selective energetic stabilization of state O₁ relative to O₂ (Fig. 8) by both perturbations implies that in a WT channel, the physico-chemical environment of the H1348A side chain, or of the phenylethyl group of a P-ATP molecule bound at site 1, experiences a significant change also during the O₁→O₂ transition, i.e., upon ATP hydrolysis at the active composite site 2. The implication is a significant movement in site 1 upon ATP hydrolysis at site 2: a classical allosteric effect. Interestingly, the K464A mutation, which perturbs site 1 by removing the conserved Walker A lysine, was also shown to affect the energetics of both of the C₁↔O₁ and O₁→O₂ gating steps (Csanády et al., 2010), although in a different way: in this mutant, rate k₁ decreased approximately fourfold, whereas the rate of nonhydrolytic closure, in a K1250A mutant background, increased by ∼10-fold (this is also replicated in the E1371S background; Fig. S2). Thus, although the distortion of the energetic profile by the K464A mutation differs from that seen for the two perturbations studied here, the K464A result is again consistent with movements in site 1 occurring in both of the above gating steps.

In principle, we cannot exclude the possibility of a “static allosteric effect” as an alternative scenario to explain our results. That is, perturbations in site 1 might statically impose a slightly different overall conformation of the NBDs, which leads to altered transition rates associated with reactions and conformational changes occurring at the active site 2, without movements in site 1 during those transitions. However, such an interpretation is at odds with the generally accepted view of protein allostery. Thus, manipulations at one site in a protein (the “allosteric site”) that affect conformational changes at a distant site (the “active site”) generally involve movement at the allosteric site itself: allosteric modulators affect the conformational equilibrium of proteins because the affinity of their binding sites changes during the conformational transition. Examples range from 2,3-bisphosphoglycerate or proton modulation of oxygen binding in hemoglobin (Monod et al., 1965) to activation of ligand-gated channels (Changeux and Edelstein, 2005); in contrast, we are not aware of any example of a “frozen” allosteric site. We therefore feel that a conventional “dynamic allosteric effect,” which involves movements in site 1 of CFTR, is a more likely explanation of our findings.

Does site 1 separate then, or not, during a normal gating cycle? The answer might lie somewhere in between. The recent determination of the TM287/288 structure (Hohl et al., 2012) at least suggests such an intermediate scenario. In this structure, the TMDs are seen in an inward-facing orientation, generally believed to correspond to the closed-channel state in CFTR. At the same time, the NBD dimer interface is only partially separated (Fig. 9): the active site 2 is open and devoid of nucleotide (Fig. 9, top composite site), whereas the degenerate site (Fig. 9, bottom composite site) contains a bound AMPPNP molecule, and in it, the opposing faces remain in contact. In particular, the α-carbons of an asparagine in the TM288 D-loop (Asn521; Fig. 9, violet space fill) and that of a conserved threonine in the TM287 Walker A loop (T368; Fig. 9, red space fill) are separated by only 5.7 Å, and the asparagine is seen to make contact with the γ-phosphate of the bound AMPPNP. Interestingly, the corresponding positions in site 1 of CFTR formed one of three interface pairs (T460-H1375) for which a lack of state-dependent changes in energetic coupling was interpreted to suggest a maintained interaction throughout the gating cycle (Szollosi et al., 2011). However, consistent with the data we present here, a comparison between TM287/288 and crystals with tightly dimerized NBDs highlights clear differences in composite site-1 structure. The distance between the Walker A motif and the opposing signature sequence of the degenerate site is larger by ∼3 Å in the TM287/288 structure (Hohl et al., 2012) than that seen in tight dimers of ABC NBDs (e.g., Smith et al., 2002; Dawson and Locher, 2007). The TM287/288 structure therefore also suggests some opening, although not complete separation, of the degenerate site.

In conclusion, the data presented here strongly suggest that significant movements occur in CFTR’s composite site 1 concomitant with gating conformational changes. The precise extent and direction of these movements await further structural and functional studies.

We thank Dorottya Mayer for oocyte isolation and injection.

This work is supported by National Institutes of Health grant R01-DK051767, MTA Lendület grant LP2012-39/2012, and an International Early Career Scientist grant from the Howard Hughes Medical Institute (to L. Csanády).

Christopher Miller served as editor.