CFTR channel opening by ATP-driven tight dimerization of its nucleotide-binding domains

Crystal structures of most ABC-protein NBDs determined so far share the same fold^7,8 with a core subdomain (‘head’) that binds the ATP, and an α-helical subdomain (‘tail’) that includes the ABC-specific signature sequence (LSGGQ). Dimeric structures revealed nucleotide-bound NBD homodimers in rotationally symmetric ‘head-to-tail’ arrangement, enclosing two ATP molecules within interfacial composite sites, each comprising conserved ATP-binding motifs from the head of one monomer and signature sequence residues from the tail of the other^3,5,9,10. On the basis of this structural evidence and biochemical studies of reversible dimerization of isolated NBDs^4,5,11–13, opening and closing of CFTR channels can be interpreted¹⁴ in terms of cycles of NBD1–NBD2 dimerization and dissociation, induced by ATP binding and hydrolysis, respectively (Fig. 1a). Opening of a phosphorylated CFTR Cl⁻ channel seems to require ATP binding to both composite sites because, at low [ATP], mutations expected to weaken ATP binding can make nucleotide occupancy at either site rate-limiting for channel opening¹⁴. In addition, interfering with hydrolysis prevents the normal rapid closing of CFTR channels^1,2,14. Because photolabelling studies show that ATP can remain at the NBD1-head site for several minutes without being hydrolysed^15,16, whereas a CFTR-channel gating cycle lasts only seconds, channel opening and closing seem to be timed by nucleotide binding and hydrolysis at the composite site incorporating the NBD2 head, which we call the NBD2 catalytic site^14,15.

Evidence supporting our speculation (Fig. 1a) that the CFTR-channel open state corresponds to the dimerized NBD conformation is provided by the approximately 1,000-fold stabilization of the open-burst state that results from mutation of the possible catalytic base^4,17, Glu 1371 (a glutamate in the NBD2-head ‘Walker B’ motif), to Gln (E1371Q; Fig. 1b). Because CFTR channels do not open without ATP (see below), current decay upon the removal of ATP reflects channel closing, and its time course measures the open burst duration; closing was complete within about 1 s of ATP withdrawal for wild-type (WT) CFTR channels (mean open burst duration was less than 0.5 s), but had a time constant of 411 ± 64 s (mean ± s.e.m., n = 16) for mutant E1371Q channels (Fig. 1b). Correspondingly, in isolated NBD subunits impairment of ATP hydrolysis by the homologous Glu-to-Gln mutation induces the formation of very stable ATP-bound homodimers^4,5,11,12.

To test the dynamic domain rearrangements underlying the model (Fig. 1a), we investigated the interaction between residues on opposite sides of the predicted NBD1–NBD2 heterodimer interface. In crystals of ATP-bound prokaryotic NBD homodimers (MJ0796, the hyperthermophylic archaeon homologue of LolD, part of a putative lipoprotein transporter⁵, and MalK, the NBD of the maltose transporter of Escherichia coli¹⁰) a hydrogen bond connects the residues corresponding to CFTR’s Arg 555 (three positions after the signature sequence: LSGGQRAR) in the NBD1 tail, and Thr 1246 (within the ‘Walker A’ phosphate-binding loop, GRTGSGKS) in the NBD2 head. We chose to target this pair of residues on the basis of a statistical analysis (Fig. 2b) of more than 10,000 NBD sequences, which suggests that positions corresponding to CFTR’s Arg 555 and Thr 1246 are functionally coupled¹⁸. Partitioning the total multiple sequence alignment into subsets, on the basis of the side chain present at the site corresponding to Arg 555, yields one major subset with arginine, and another with lysine, at that position, both of which are potential hydrogen bond donors. The distribution of potential acceptor side chains at the position corresponding to Thr 1246 differs in these two subsets, the shorter lysine donor being more frequently paired with the longer asparagine acceptor, and the longer arginine donor being more frequently paired with the shorter serine acceptor (or the equivalent threonine, as found in CFTR) (Fig. 2b). This striking covariance suggests that the two sites have been subject to evolutionary pressure as a pair, rather than individually, so as to retain the ability to form a hydrogen bond that precisely distances the two α-carbons (Fig. 2a) in the dimer structure.

We first mutated Arg 555, and found that the charge-removing mutation R555Q slowed channel closing (mean open burst duration was 3.20 ± 0.35 s (n = 18) for R555Q, compared with 0.43 ± 0.02 s (n = 32) for WT; Fig. 2c), consistent with Arg 555 being part of the composite NBD2 catalytic site, where the ATP hydrolysis that times closing of WT channels occurs. Possibly, the positive charge of Arg 555 normally helps to stabilize the partial negative charges developed on the β- and/or γ-phosphate in the transition state for that ATP hydrolysis¹⁹; an equivalent mutation impairs ATP hydrolysis in the ABC multidrug transporter P-glycoprotein²⁰. Accordingly, the charge-conserving mutation R555K did not affect open burst duration (mean 0.39 ± 0.04 s (n = 26)). However, it substantially prolonged closed interburst duration (inversely related to opening rate) from 2.29 ± 0.46 s (n = 16) for WT to 8.53 ± 1.23 s (n = 15) for R555K (Fig. 2c, all measured at saturating [ATP]). On the basis of the evolutionary evidence suggesting the involvement of the side chain at this position in a conserved interaction (Fig. 2b), this slowing of channel opening could be explained if the R555K mutation were to weaken or remove a hydrogen bond between NBD1 and NBD2 that is absent in the closed, ground, state but present in the transition state for the channel opening reaction; the resulting destabilization of the transition state would increase the activation free energy (ΔG^‡) for channel opening and hence decrease the opening rate.

To quantify this suspected interaction between Arg 555 and Thr 1246 side chains (Fig. 2a), we applied double mutant-cycle analysis^6,21 (see Supplementary Information), after mutating Arg 555 to Lys and Thr 1246 to Asn, both individually and jointly. The WT protein, the two single mutants and the double mutant form the corners of a thermodynamic cycle (Fig 3a and Fig 4a, d). If the two residues do not interact, the effects of mutating Arg 555 to Lys should be the same in a Thr 1246 background as in a T1246N background (and vice versa); that is, the effects of the single mutations should be independent and hence additive, and mutation-linked changes on parallel sides of the cycles should thus be equal. Any difference signifies, and quantifies, energetic coupling (ΔΔG_int) between the two residues. Because changes in any path-independent variable can be used to evaluate the effects of the mutations, we used different kinetic measurements to assess energetic coupling between residues at positions 555 and 1246 at different stages of the channel gating cycle.

We first examined coupling in closed channels. The opening rate of WT CFTR shows simple Michaelis–Menten dependence on [ATP] (Fig. 3c), most probably reflecting ATP interaction with the NBD2-head binding site¹⁵. The measured maximal opening rate is relatively slow (0.27 s⁻¹, for pre-phosphorylated channels¹⁴; step C₂ to O in Fig. 1a), allowing the preceding ATP binding step (C₁ to C₂) to reach a steady state not far from equilibrium. Consequently, the apparent affinity for ATP (K_0.5) provides a good estimate of its dissociation constant at the NBD2 site on the closed channel (C₂). Changes in K_0.5 could therefore be used to quantify the effects of mutations and so assess coupling between Arg 555 and Thr 1246 in closed channels (Fig. 3). The apparent affinity for ATP was little influenced by the mutation R555K (R555K K_0.5 = 71 ± 14 µM versus WT K_0.5 = 55 ± 5 µM), but was reduced by the mutation T1246N (T1246N K_0.5 = 261 ± 49 µM) by the same extent in the WT background as in the R555K background (R555K T1246N K_0.5 = 257 ± 51 µM). The closely similar effects of mutations on parallel sides of the double mutant cycle signify a coupling energy not significantly different from zero (ΔΔG_{int(unbound–bound)} = 0.31±0.55kT). So, either Arg 555 and Thr 1246 do not interact when CFTR channels are closed, or they do interact but with identical coupling energies before, and after, binding of the second ATP (that is in state C₁ and in state C₂). But, given that the residue corresponding to Thr 1246 is observed to directly contact the γ-phosphate oxygens in all ATP-bound crystal structures, both monomeric and dimeric^3,5,10,22,23, it is highly improbable that there is significant energetic coupling between Arg 555 and Thr 1246 that is unaffected by ATP binding. We therefore conclude that the two target side chains do not interact in the closed-channel conformations (either with or without bound ATP), and hence that ATP binding occurs before the formation of a closely apposed NBD1–NBD2 dimer. Indeed, in dimeric crystals the bound ATP molecules are buried within the interface^5,10, implying that access to the binding sites must occur in a different conformation.

We next tested for energetic coupling as CFTR channels approach the open burst state, by determining changes in activation free energies for channel opening (in the presence of saturating [ATP]). The slowing of opening caused by the R555K mutation (Fig. 2c, above) corresponded to a 1.4 ± 0.4kT increase in the activation energy barrier. The T1246N mutation also greatly slowed channel opening, increasing the energy barrier by 2.5 ± 0.4kT. However, fast opening was partly restored when the two mutations were introduced simultaneously (Fig. 4b, c). Comparing changes in activation-energy barrier height caused by each mutation in the WT, and in the mutant background, we obtain an energetic coupling between residues at positions 555 and 1246 of $\mathrm{\Delta }\mathrm{\Delta }{G}_{\text{int}(\text{opening})}^{‡}$ of −2.7 ± 0.5kT. The negative sign is consistent with Arg 555 and Thr 1246 in WT CFTR forming a stabilizing interaction, present in the transition state but not in the closed, ground, state (the above conclusion that these two residues do not interact in closed channels allows us to rule out the alternative interpretation, that the negative coupling energy could arise from a destabilizing interaction in the closed state that is lost in the transition state for channel opening; see also Supplementary Information). The presence of the postulated hydrogen bond between Arg 555 and Thr 1246 in the transition state could represent that stabilizing interaction: either mutation alone would remove the hydrogen bond and destabilize the transition state, whereas both mutations together would restore favourable geometry for hydrogen bond formation (Fig. 2a) and so speed up channel opening.

Interaction between Arg 555 and Thr 1246 in the open burst state, rather than approaching it, was monitored by measuring channel open probability (P_o) to estimate the relative stability of the closed and open states, after reducing the channel gating scheme to a simple closed–open equilibrium. For the scheme in Fig. 1a, the latter condition holds, at saturating [ATP], if channel closing from open bursts by means of the hydrolytic pathway (O to C₁) is precluded. In other ABC-ATPases, mutating the key lysine in the phosphate-binding loop to arginine drastically reduces or abolishes hydrolysis (see, for example, ref. 24) and, as would be predicted if hydrolysis at CFTR’s NBD2 catalytic site were markedly slowed, CFTR channels carrying the corresponding mutation (K1250R) have prolonged open burst durations (Fig. 4e; mean time constant of current decay upon ATP removal τ = 9.3 ± 0.5 s; n = 49). Introducing the T1246N mutation into the K1250R background decreased P_o, corresponding to destabilization of the open burst state by 2.5 ± 1.0kT with respect to the closed state. However, adding the R555K mutation to T1246N–K1250R channels restored high stability of the open state (Fig. 4e, f). The coupling energy obtained from the mutant cycle (Fig. 4d) was again negative in sign (ΔΔG_{int(open–closed)} = −2.4 ± 1.0kT), which is consistent with the presence of a stabilizing interaction (for example a hydrogen bond) between Arg 555 and Thr 1246 in the open burst state that is absent from the closed state.

Thus, as conformational changes gate CFTR’s transmembrane Cl⁻-ion permeation pathway from closed to open, two residues, on opposite sides of the anticipated NBD1–NBD2 heterodimer interface, go from being independent to being energetically coupled, most probably by forming a hydrogen bond. The changes in the relative positions of CFTR’s NBD1 and NBD2 implied by our results need not be large. They need be no larger than the relatively small differences observed between members of NBD pairs, juxtaposed in a ‘head-to-tail’ fashion, in crystal structures of nucleotide-bound versus nucleotide-free conformations. The residues corresponding to Arg 555 and Thr 1246 are close enough (2.7Å) to form a hydrogen bond in the ATP-bound crystals of LolD (MJ0796; ref. 5) and MalK (PDB ID no. 1Q12; ref. 10) and are found only a few ångströms further apart in nucleotide-free structures (for example, 5.7Å in BtuCD (ref. ⁹); 8.5Å in MalK, PDB ID no. 1Q1B (ref. ¹⁰)), in which the NBDs share a limited contact surface but are kept in a dimeric arrangement by interactions occurring through contiguous domains (see also ref. 13). Nevertheless, however small the local change at CFTR’s NBD2 catalytic site, it is large enough to cause transmission of the long-range signal that results in opening of the channel pore.

By clearly linking NBD dimerization state to the disposition of the transmembrane domains in an intact, functioning, human ABC protein, our results establish dynamic NBD dimerization as the molecular mechanism that couples ATP binding and hydrolysis cycles to cyclic changes in the transmembrane domains. Conservation of the structural underpinnings of the hydrogen bond highlighted here implies that NBD dimerization occurs during the duty cycle of most, if not all, ABC proteins, although with subfamily-specific consequences. Thus, the tightly dimerized NBD conformation corresponds to the open-burst state in CFTR channels (Fig. 1), but in exporters or importers to conformations of the membrane domains that release drugs to the exterior²⁵, or receive substrate from the substrate-binding protein²⁶, respectively, and in DNA-repair ABC ATPases to relevant conformations in the DNA-binding regions²⁷.