Sheppard, David N., and Michael J. Welsh. Structure and Function of the CFTR Chloride Channel. Physiol. Rev. 79, Suppl.: S23–S45, 1999. — The cystic fibrosis transmembrane conductance regulator (CFTR) is a unique member of the ABC transporter family that forms a novel Cl− channel. It is located predominantly in the apical membrane of epithelia where it mediates transepithelial salt and liquid movement. Dysfunction of CFTR causes the genetic disease cystic fibrosis. The CFTR is composed of five domains: two membrane-spanning domains (MSDs), two nucleotide-binding domains (NBDs), and a regulatory (R) domain. Here we review the structure and function of this unique channel, with a focus on how the various domains contribute to channel function. The MSDs form the channel pore, phosphorylation of the R domain determines channel activity, and ATP hydrolysis by the NBDs controls channel gating. Current knowledge of CFTR structure and function may help us understand better its mechanism of action, its role in electrolyte transport, its dysfunction in cystic fibrosis, and its relationship to other ABC transporters.
Cystic fibrosis transmembrane conductance regulator (CFTR) is a phosphorylation-dependent epithelial Cl− channel. It is located primarily in the apical membrane, where it provides a pathway for Cl− movement across epithelia and regulates the rate of Cl− flow. Thus CFTR is central in determining transepithelial salt transport, fluid flow, and ion concentrations. In the intestine, pancreas, and sweat gland secretory coil, CFTR plays a key role in fluid and electrolyte secretion, and in sweat gland duct and airway epithelia, it participates in fluid and electrolyte absorption. Dysfunction of CFTR Cl− channels in the genetic disease cystic fibrosis (CF) disrupts transepithelial ion transport and hence the function of a variety of organs lined by epithelia (142). This leads to the wide-ranging manifestations of the disease, which can include airway disease, pancreatic failure, meconium ileus, male infertility, and elevated levels of salt in sweat.
Studies during the 1980s showed that the apical membrane of several epithelia had a Cl− conductance activated by cAMP agonists and that this Cl− conductance was defective in CF. When the gene encoding CFTR was identified in 1989 (105), it was uncertain whether CFTR was itself an apical membrane Cl− channel regulated by cAMP-dependent phosphorylation or whether it functioned only to regulate such a Cl− channel. The loss of apical Cl− permeability observed in CF epithelia was consistent with either hypothesis. Several observations suggested that CFTR might regulate epithelial Cl− channels, either through direct association or by pumping a Cl− channel regulatory factor into or out of epithelial cells. First, the primary structure of CFTR placed it in a family of transport proteins called ATP-binding cassette (ABC) transporters (2 ,65). Several members of this family utilize the energy of ATP hydrolysis to actively transport substrates across cell membranes. Second, numerous phenotypic abnormalities have been observed in CF epithelia, including an increase in Na+ absorption by CF airway epithelia (21). These multiple phenotypic abnormalities seemed difficult to reconcile with a Cl− channel defect. Third, the primary sequence of CFTR did not resemble that of any other known ion channel (105).
However, studies of recombinant CFTR soon provided compelling evidence that CFTR is an apical membrane Cl− channel. First, investigators expressed CFTR in cells that do not normally contain cAMP-regulated Cl− channels and express little or no endogenous CFTR, including Chinese hamster ovary (CHO) cells, HeLa cells, NIH-3T3 fibroblasts, Sf9 insect cells, and Xenopus oocytes (4 ,5 ,13 ,39 ,69 ,125). In each case, expression of CFTR generated a unique Cl− current that was activated by cAMP agonists. Second, the biophysical properties and regulation of Cl− currents in cells expressing recombinant CFTR, in epithelial cells expressing endogenous CFTR, and in the apical membrane of Cl− secretory epithelia were similar (57). Third, mutation of specific residues in CFTR altered the anion selectivity sequence of Cl− currents (4). Fourth, when recombinant CFTR was purified and reconstituted into planar lipid bilayers, it formed Cl− channels with properties identical to those in native epithelia (14).
Once CFTR was demonstrated to be a Cl− channel, it was possible to investigate the relationship between structure and function. That is the focus of this review. Of note, data from several laboratories indicate that CFTR also regulates the function of other ion channels, as discussed in other contributions to this supplement.
B. Domain Structure of CFTR
After the identification of the primary amino acid sequence of CFTR, Riordan et al. (105) proposed a structure that turned out to be remarkably prescient. Figure 1 shows a model of CFTR. In large part, the structure proposed was based on a comparison of CFTR with members of the ABC transporter family, including periplasmic permeases in prokaryotes, such as the histidine and maltose transport systems; STE6, responsible for the secretion of the a mating factor in yeast; and P-glycoprotein (Pgp), which confers resistance to chemotherapeutic drugs (60). These transporters and CFTR are composed of two motifs, each containing a membrane-spanning domain (MSD) that is usually but not always composed of six transmembrane segments and a nucleotide-binding domain (NBD) that contains sequences predicted to interact with ATP (2 ,65). In CFTR, the two MSD-NBD motifs are linked by a unique domain, called the R (regulatory) domain, that contains multiple consensus phosphorylation sites and many charged amino acids.
C. Topology of CFTR
Several approaches have been used to investigate the topology of CFTR. Denning et al. (37) used antibodies in permeabilized and unpermeabilized cells to assign the loop between the first and second transmembrane segment (M1 and M2) to the extracellular surface and to assign the R domain and carboxy terminus to the intracellular surface. The glycosylation site in wild-type CFTR placed the loop between M7 and M8 on the extracellular surface. Functional studies described below suggest that the R domain and NBDs are located on the intracellular side of the membrane. Chang et al. (27) incorporated glycosylation sites throughout the protein and showed that extra- and intracellular loops are as predicted in the original description of the primary structure (Fig. 1). Similar results were reported using a cell-free expression system (29).
D. Description of Function, Role of Domains, and Mechanism of Regulation
Figure 2 A shows a recording from a single CFTR Cl− channel. These channels have several distinguishing characteristics. 1) They have a small single-channel conductance (6–10 pS) (Fig. 2 B). 2) The current-voltage (I-V) relationship is linear. 3) They are selective for anions over cations. 4) The anion permeability sequence is Br− ≥ Cl− > I−. 5) They show time- and voltage-independent gating behavior. 6) Their activity is regulated by cAMP-dependent phosphorylation and by intracellular nucleotides. These features are conferred on CFTR Cl− channels by the function of the MSDs, the NBDs, and the R domain. Knowledge of the contribution that these domains make to the overall function of CFTR has emerged from studies of wild-type CFTR, variants containing site-directed mutations, and many agents that alter function. The MSDs contribute to the formation of the Cl−-selective pore (4 ,31 ,67 ,76 ,78 ,84 ,87 ,117 ,126), the NBDs hydrolyze ATP to regulate channel gating (3 ,12 ,23 ,54 ,64 ,71 ,73 ,140), and R domain phosphorylation controls channel activity (16 ,28 ,30 ,63 ,64 ,101 ,125 ,144 ,146).
The opening and closing of the CFTR Cl− channel is tightly controlled by the balance of kinase and phosphatase activity within the cell and by cellular ATP levels. Activation of the cAMP-dependent protein kinase (PKA) causes the phosphorylation of multiple serine residues within the R domain. Once the R domain is phosphorylated, channel gating is regulated by a cycle of ATP hydrolysis at the NBDs. Finally, protein phosphatases dephosphorylate the R domain and return the channel to its quiescent state.
II. THE CHANNEL PORE: THE MEMBRANE-SPANNING DOMAINS
A. Biophysical Properties
When bathed in symmetrical Cl− concentrations, CFTR Cl− channels have a linear I-V relationship (16 ,127) (Fig. 2 B). However, when bathed in asymmetric Cl− concentrations, the I-V relationship rectifies. Originally, the outward rectification observed during cell-attached recordings was attributed to Goldman-type rectification caused by the Cl− concentration gradient. However, recent studies have offered additional explanations including the permeability of intracellular anions (92), voltage-dependent block of CFTR Cl− channels by large impermeant intracellular anions (75), and phosphorylation of CFTR by tyrosine kinases (44).
CFTR Cl− channels have a low single-channel conductance of between 6 and 10 pS (16 ,127). Some of the variation in the values reported is a consequence of the recording conditions used, including Cl− concentration, choice of buffers, and temperature (67 ,126). However, single-channel conductance also varies between CFTR from different species, decreasing in the rank order Xenopus > human > mouse > shark (56 ,71a ,95 ,127). Subconductance states of the CFTR Cl− channel are observed especially when CFTR is reconstituted into planar lipid bilayers (54 ,86 ,130). Perhaps the heavy filtering required to visualize CFTR Cl− channels in planar lipid bilayers accentuates their detection. Alternatively, subconductance states may result from the absence of divalent cations from the extracellular solution (130).
B. Ion Selectivity
Reversal potential measurements with different NaCl concentration gradients indicate that CFTR Cl− channels are selective for anions over cations. The permeability for Na+ over Cl− (P Na/P Cl) is in the range of 0.1–0.03 (4 ,14 ,126), suggesting significantly greater selectivity for Cl− than for Na+. Furthermore, substituting Na+ by other cations including K+ and N-methyl-d-glucamine was without effect on the permeability ratio. Thus it is likely that the channel shows little cation permeability.
The anion permeability sequence of CFTR was first determined using the whole cell configuration of the patch-clamp technique; the anion permeability sequence of CFTR in a whole cell patch is Br− ≥ Cl− > I− > F− (4). This anion permeability sequence distinguishes CFTR Cl− channels from other epithelial Cl− channels that have a higher permeability to I− than Cl−, for example, the outwardly rectifying Cl− channel (see other contributions in this supplement). Subsequent studies of CFTR Cl− channels in excised inside-out membrane patches under bi-ionic conditions suggested that the apparent I− to Cl− permeability is influenced by the ability of I− to block the pore. The data of Tabcharani et al. (126) using single channels suggest that I− is more permeable than Cl−, but because I− blocks the pore, it appears that I− is less permeable under some conditions, including the whole cell patch. Because other halides do not block the CFTR pore, they propose that the anion permeability sequence of CFTR is I− > Br− > Cl− > F−. This would suggest that like other Cl− channels, including ligand-gated Cl− channels in neurons (20) and outwardly rectifying Cl− channels in epithelia (55), the CFTR pore has a “weak field strength” selectivity site (147). Thus, depending on the experimental conditions and whether block of the channel by I− is considered, different values for anion selectivity sequence may be observed.
The permeability of CFTR Cl− channels to a large number of polyatomic anions of known dimensions was assessed by Linsdell et al. (77). From bi-ionic reversal potential measurements, the permeability sequence NO− 3> Cl− > HCO− 3 > formate > acetate was obtained. In contrast, pyruvate, propanoate, methanesulfonate, ethanesulfonate, and gluconate did not permeate CFTR (77). On the basis of these data, the minimum diameter of the CFTR pore was estimated to be ∼5.3 Å (77), similar to that reported for other Cl− channels (10 ,20 ,55). Water, urea, and ATP have also been reported to permeate CFTR Cl− channels, although the evidence for ATP permeation is controversial (51 ,59 ,72 ,93 ,99 ,100 ,111 ,129). Like the halide anion permeability sequence proposed by Tabcharani et al. (126), that of polyatomic anions follows a lyotropic sequence (35). This suggests that in CFTR, anion permeation is determined by the hydration energy of anions (126). Studies of CFTR permeation using polyatomic pseudohalide ions support this conclusion (123).
Several lines of evidence suggest that CFTR is a multi-ion pore. First, wild-type CFTR exhibits anomalous mole fraction behavior when bathed in symmetrical mixtures of Cl− and SCN− (128). Second, the inhibition of the CFTR pore by intracellular gluconate is relieved when the extracellular Cl− concentration is increased, suggesting that Cl− can expel gluconate from the pore (75 ,76). Third, the inhibition of CFTR by I− only occurs when I− is present on one side of the membrane and Cl− is present on the other, suggesting that interactions between Cl− and I− within the pore are responsible for the low P I/P Cl values reported for CFTR (126). These data provide strong evidence for interactions between anions within the CFTR pore. They also suggest that the CFTR pore may contain at least two anions simultaneously.
C. Identification of Residues That Influence Conduction and That Line the Pore
CFTR contains six positively charged amino acids within the putative transmembrane sequences [K95 (M1), R134 (M2), R334 (M6), K335 (M6), R347 (M6), and R1030 (M10)] (105). These residues are conserved across species, and two are the site of mutations associated with CF: R334Q/W and R347C/H/L/P (142). This suggests that these residues may have an important functional role. Mutation of two of these basic residues individually to acidic residues (K95D and K335E) altered the whole cell anion permeability sequence by converting CFTR from a low I− permeability pore (Br− ≥ Cl− > I−) to a high I− permeability pore (I− > Br− > Cl−) (4). The other mutants, R347E and R1030E, did not alter the anion permeability sequence, although P I/P Cl values were increased. These data suggest that the residues K95 and K335 determine, in part, the anion selectivity of CFTR. Although it is tempting to speculate that these residues line the pore, it could be that mutation of these residues disrupts the structure of the pore, thereby altering selectivity. Thus it could be that the peptide backbone lines the pore and determines the permeation properties. However, consistent with the idea that these residues line the pore, the mutants K95C and K335C interact with methanethiosulfonate (MTS) reagents, and mutations that eliminate the positive charge at K335 reduce single-channel conductance, alter the shape of the I-V relationship, and affect open-channel block by diphenylamine-2-carboxylate (DPC) (1 ,31 ,87 ,128). However, the mutation K335E did not affect anomalous mole fraction behavior of CFTR (128).
The basic residues R334 and R347, located in M6, are the site of CF-associated mutations (142). When expressed in heterologous epithelial cells, the CF-associated mutants R334W and R347P were correctly processed and generated cAMP-stimulated Cl− currents, although the amount of Cl− current was ≤30% that of wild-type CFTR (117). Analysis of the single-channel properties of R334W and R347P demonstrated that both mutants decrease single-channel conductance by ≥70% (117). A molecular explanation for why mutants at R347 decrease single-channel conductance has been provided by Tabcharani et al. (128). When the basic arginine at this position was replaced with the acidic residue aspartic acid, single-channel conductance decreased by ∼50%, and the anomalous mole fraction behavior of CFTR in mixtures of Cl− and SCN− was abolished (128). Moreover, Tabcharani et al. (128) could control the pore properties of the CF-associated mutant R347H simply by manipulating pH. At pH 5.5, when histidine is positively charged, R347H had normal pore properties (128). However, at pH 8.7, when histidine is uncharged, single-channel conductance was reduced to a similar extent as that of R347P, and the anomalous mole fraction behavior of CFTR was lost (128). These results suggest that as anions pass through the CFTR pore they interact with the positive charge at R347. They also suggest that CF mutations at R347 disrupt CFTR function by converting CFTR from a multi-ion to a single-ion pore. However, pending additional studies of structure, the caveats noted above regarding the contribution of residues to the pore versus less specific alterations of structure must be considered.
To identify residues that line the CFTR pore and make inferences about secondary structure, Akabas and colleagues (1 ,31 ,32) used the substituted cysteine accessibility method. At the extracellular end of M1 between G91 and P99, they identified three residues (G91, K95, and Q98) that line the CFTR pore, suggesting that between G91 and Q98, M1 has an α-helical structure (1). In M6, Cheung and Akabas (31) identified 11 residues between K329 and Q353 that likely line the CFTR pore (I331, L333, R334, K335, F337, S341, I344, R347, T351, R352, and Q353). The voltage dependence of the reaction rates of MTS reagents with cysteine mutations at these residues and single-channel studies of wild-type CFTR and the mutant R352C suggest that the selectivity of CFTR for anions over cations is determined by a site located at the intracellular end of the pore that involves the residue R352 (32 ,52). The data also suggest that electrical potential is not distributed linearly along the length of the CFTR pore; there is a major barrier to current flow through the CFTR pore at the intracellular end of the pore and R352C is located closer to the extracellular end of the pore than either T351C or Q353C (32). Based on these results, Cheung and Akabas (32) speculated that most of the secondary structure of M6 is probably α-helical, but at the intracellular end a segment of M6, including R352, may form a pore loop that contributes to the anion selectivity of the CFTR pore.
Proline residues located in the MSDs of transport proteins play an important structural role by kinking α-helices (22). In CFTR, the predicted MSDs contain four prolines: P99, P205, P324, and P1021 (105). Mutations of these residues suggest that P99 contributes either directly or indirectly to the formation of the CFTR pore (119). When P99 was mutated to leucine, the channel lost its ability to discriminate between Cl− and I−, and substitution of alanine, glycine, and leucine at P99 decreased single-channel Cl− conductance in the rank order: wild type ≥ P99G > P99L ≥ P99A (119). Interestingly, the propensity of these amino acids to occur in α-helices follows the reverse order (A ≥ L > G ≥ P) (104). Based on these observations, we speculated that P99 kinks M1 to form part of the channel structure. However, because P99C did not react with MTS reagents (1), P99 probably does not directly line the CFTR pore.
Knowledge of the structure and function of the CFTR pore has also emerged from studies using truncated and chimeric proteins (95 ,116 ,134). The amino-terminal portion of CFTR (D836X, which contains MSD1, NBD1 and the R domain), by itself, formed a regulated Cl− channel with conductive properties remarkably similar to those of wild-type CFTR (116). Interestingly, a physiological role for such a truncated channel has been proposed in the kidney (88). Based on biochemical and functional data, we speculated that D836X generates a Cl− channel by forming a homomultimer (116). This suggests that if MSD2 sequences contribute to the CFTR pore (as we discuss below), then in a D836X multimer MSD1 sequences may substitute for MSD2 sequences to form the pore. Alternatively, MSD2 sequences may have a different function. They might stabilize the channel complex, perhaps assisting in the arrangement of MSD1 sequences into a functional complex or in shielding some MSD1 sequences from the lipid bilayer. Consistent with this idea, biochemical and functional studies indicate that MSD1 associates with MSD2 to form a regulated Cl− channel (90).
Comparison of the anion permeability of human CFTR (Br− = Cl− > I−), Xenopus CFTR (Br− = I− > Cl−), and human-Xenopus CFTR chimeras in which either MSD1 or MSD2 of human CFTR was replaced with the equivalent region of Xenopus CFTR (hX1–6, Br− = I− > Cl−, and hX7–12, Br− > Cl− > I−, respectively) also suggest that sequences in MSD1 likely determine the anion permeability of CFTR (95). Interestingly, all the missense mutations discussed above that alter the pore properties of CFTR are the same in human and Xenopus CFTR. Therefore, other sequences must account for the differences in anion permeability between human and Xenopus CFTR. Because the anion selectivity of a CFTR construct lacking M1-M4 (Δ259) was similar to wild-type CFTR (134), whereas the carboxy-terminal portion of CFTR (R domain, MSD2, and NBD2) did not discriminate between Cl− and I− (38), these sequences may be located within M5 and M6 (84).
D. Contribution of the Intracellular and Extracellular Loops
Sequences in the intracellular loops (ICL) appear to be critical for correct protein processing and delivery to the cell membrane. First, incorporation of glycosylation sequences into ICL1, ICL3, and ICL4 disrupted protein folding, and there was little Cl− channel function measured (27). Second, deletion of exon 5 (residues 163–193, located in ICL1), a common CFTR splice variant in murine epithelial tissues, disrupted the processing and function of human CFTR (36 ,149). However, in cardiac tissues, the exon 5− splice variant may play an important functional role: it is found in cardiac tissue from a number of different species and when expressed in Xenopus oocytes forms a cAMP-stimulated Cl− current (46 ,58 ,61). Third, deletion of 19 residues from ICL2 (residues 267–285) disrupts the biosynthesis of CFTR (148). Fourth, 3 CF-associated mutations in ICL1, 1 in ICL2, 2 in ICL3, and 12 in ICL4 produced similar effects (33 ,112–114).
Site-directed mutations in the ICL also alter channel function. Deletion of 19 residues from ICL2 promoted transitions to a subconductance state, whereas the I-V relationships of the mutations S945L and G970R in ICL3 showed weak outward rectification in contrast to that of wild-type CFTR (114 ,148). These results suggest that ICL2 and ICL3 may be located close to the intracellular mouth of the CFTR pore. They also suggest that residues in ICL2 may stabilize the open state of the channel. Other mutations in the intracellular loops, however, had no discernible effect on pore properties (33 ,112–114). Instead, they altered gating behavior and channel regulation. In general, mutations in ICL1 and ICL2 increased mean closed time, whereas mutations in ICL3 and ICL4 decreased mean open time (33 ,112–114). Furthermore, the mutations I148T and G178R could not be locked open by the nonhydrolyzable ATP analog 5′-adenylylimidodiphosphate (AMP-PNP) (112), the mutation A1067T altered the relationship between open probability (P o) and ATP concentration (33), and the response of R1066L and F1052V to pyrophosphate (PPi) was less than wild type (33). These effects of mutations in the intracellular loops on channel activity are reminiscent of those observed with mutations in the NBDs (see sect. iii). This suggests that the intracellular loops may couple the activity of the NBDs to channel gating.
Studies of CFTR variants lacking the consensus glycosylation sequences in extracellular loop (ECL) 4 (N894, 900Q) suggest that glycosylation is not necessary for correct protein processing or for the formation of a phosphorylation-regulated Cl− channel (27 ,50). However, the time course of I− efflux by the N894,900Q mutant was prolonged compared with wild-type CFTR, suggesting that channel activity may be altered (27). Insertion of N-glycosylation consensus sequences into ECL1–3, ECL5, and ECL6 had no effect on protein processing or Cl− channel function (27). Similarly, insertion of the FLAG epitope into ECL4 did not alter the function of CFTR (109).
However, other studies suggest that residues in ECL1 may contribute to the CFTR pore. Arginine-117 located at the external end of M2 is the site of four CF-associated mutations (R117C/H/L/P) (142). When expressed in heterologous epithelial cells, R117H was correctly processed and generated cAMP-stimulated Cl− currents, although the amount of Cl− current was ≤30% that of wild-type CFTR (117). In addition, R117H had an altered sensitivity to external pH, suggesting that R117 may lie at the external mouth of the channel where it senses external pH, thereby influencing the conduction pathway (117). R117H caused a small decrease in single-channel conductance but a dramatic change in gating behavior: mutant channels were characterized by a large decrease in the time that they were open, and hence, P o was reduced (117).
Comparison of the sequence of ECL1 in CFTR cloned from different species indicates that ECL1 is a region of sequence divergence. For example, the sequence 111PDNKE115 in human CFTR is replaced by 111RDNEH115 in Xenopus CFTR (95 ,137). Although, Xenopus CFTR had a pattern of gating similar to human CFTR, the human-Xenopus CFTR chimera hX1–6 showed a pattern of gating characterized by a shortened open time, similar to that of R117H (95). When we inserted the 111PDNKE115 sequence into hX1–6, the gating pattern reverted to one similar to that of human CFTR and Xenopus CFTR (95). Based on these studies, we speculated that sequences in ECL1 contribute to the external vestibule of the pore.
E. Pharmacology of Block
Information about the CFTR pore has also emerged from studies using Cl− channel inhibitors. Because DPC inhibition of CFTR was voltage dependent and enhanced when the external Cl− concentration was reduced, McCarty and colleagues (86 ,87) speculated that DPC occluded the CFTR pore. Consistent with this idea, S341 located in M6 was identified as the major determinant of DPC binding to CFTR (87). Interestingly, the DPC-binding site could be transferred from S341 to S1141 in M12 while another mutation in M12, T1134F, increased the affinity of DPC block (87). McDonough et al. (87) interpreted these results to suggest that both M6 and M12 line the CFTR pore. They also suggested that these transmembrane segments have α-helical structures.
The estimated diameter of DPC (∼3 Å) suggests that it is small enough to pass through the CFTR pore. Consistent with this idea, DPC blocks CFTR when added to either side of the membrane (86). In contrast, the disulfonic stilbene derivatives 4,4′-dinitrostilbene-2,2′-disulfonic acid (DNDS) and DIDS are large anions (estimated diameters ∼13 Å) that only block CFTR when added to the intracellular side of the membrane (74). The failure of disulfonic stilbenes to inhibit CFTR Cl− channels when added to the extracellular side of the membrane distinguishes CFTR from other epithelial Cl− channels that are blocked by extracellular DNDS and DIDS (6). Like DPC, DNDS and DIDS inhibition of CFTR was voltage dependent (74). Moreover, the mutant R347D significantly weakened the binding of DNDS and DIDS to CFTR, suggesting that R347 contributes to the binding site for disulfonic stilbenes (74).
The pharmacology of ATP-sensitive K+ channels has been extensively investigated and found to be controlled by the sulfonylureas glibenclamide and tolbutamide (11 ,124). Because ATP regulation is a property shared by ATP-sensitive K+ channels and CFTR, these agents were studied with CFTR. The sulfonylureas glibenclamide and tolbutamide inhibited whole cell CFTR Cl− currents with half-maximal concentrations of ∼20 and 150 μM, respectively (120). Site-directed mutations in the NBDs did not alter the effect of glibenclamide (120). Addition of sulfonylureas to the intracellular surface of excised patches caused a dose-dependent decrease in the open time, suggesting that sulfonylureas are open channel blockers (118 ,139). Moreover, the effect of changes in pH suggested that the anionic form of glibenclamide inhibits CFTR. The effect of glibenclamide was voltage dependent and enhanced when external Cl− concentration was decreased (118). These results suggest that glibenclamide blocks the CFTR Cl− channel within a large intracellular vestibule that is part of the CFTR pore (108 ,118).
Other large anions, including gluconate, glutamate, and MOPS, cause a voltage-dependent block of CFTR Cl− channels, and their binding sites are located 30–60% of the way through the transmembrane electric field from the intracellular surface (67 ,75 ,76). Although electrical distance does not necessarily indicate physical distance, two studies have demonstrated excellent correlation between electrical distance and physical distance along the length of M6. Tabcharani et al. (128) found that SCN− bound within the CFTR pore at a site ∼20% of the electrical distance through the membrane from the intracellular side, in good agreement with the predicted location of R347 (assuming an α-helix) with which it interacted. McDonough et al. (87) found that DPC bound at a site ∼40% of the electrical distance through the membrane from the intracellular side; this value is in good agreement with the predicted location of S341 to which it bound. Although it is tempting to use this information to predict the identity of the residues that interact with gluconate, glutamate, and MOPS, caution is warranted for such speculation. Nevertheless, the data suggest that the CFTR pore contains a large intracellular vestibule where large anions bind to block Cl− permeation.
F. A Model of the CFTR Pore
A model of the CFTR pore is beginning to emerge based on the data reviewed above. Because mutation of specific residues within the MSDs altered conduction and permeation, these domains contribute either directly or indirectly to the structure of the CFTR pore. In contrast, mutation of specific residues within the NBDs and R domain have had little discernible effect on conduction and permeation, although they have frequently had profound effects on gating behavior (see below). These findings suggest that the NBDs and R domain probably do not contribute to the CFTR pore. However, Arispe et al. (9) found that when NBD1 was incorporated into planar lipid bilayers, it generated anion-selective channels. This result raises the possibility that NBD1 may also contribute to the channel pore. These divergent views serve to emphasize the limitations of our current state of knowledge.
The GABAA and glycine receptor Cl− channels are composed of five subunits that each contain four transmembrane segments, one of which (M2) contributes to the channel pore (150). Five M2 segments assembled in a quasi-symmetrical arrangement line the GABAA Cl− channel. In contrast, the number, identity, and organization of the transmembrane segments that line the CFTR pore are unknown at the present time. Because mutation of specific residues in M1, M5, M6, and M12 altered conductance, permeation, and open-channel block, these transmembrane segments appear to line the CFTR pore (1 ,4 ,31 ,84 ,87 ,117 ,128). In addition, M2 may contribute to the pore; when reconstituted into planar lipid bilayers, peptides with sequences representing M2 and M6 formed CFTR-like pores (89). There has also been speculation that the CFTR pore may be assembled from two or more CFTR monomers. Biochemical studies do not, however, support this idea. First, full-length CFTR did not associate to form a multimer (85). Second, CFTR sedimented as a monomer in a sucrose density gradient (91). Thus current data suggest that CFTR can function as a monomer in the cell membrane.
As has been suggested for ligand-gated channels, aligned residues in different transmembrane segments of CFTR may contribute to the pore properties. First, the residues K95 (M1) and K335 (M6) that determine, in part, anion permeation in the CFTR pore (4), are both predicted to lie toward the extracellular surface of the channel (105). Second, the DPC-binding site could be transferred from S341 in M6 to the aligned residue in M12, S1141 by site-directed mutation (87). However, most data suggest that the CFTR pore lacks the symmetry of the GABAA receptor. The data reviewed above suggest that M6 is one of the primary determinants of the pore properties of CFTR. Within M6, the residues R334/K335, S341, and R347 may directly or indirectly contribute to the control of Cl− permeation (4 ,87 ,117 ,126 ,128). Consistent with this idea, the permeation properties of CFTR were best described by a model with three Cl−-binding sites (76).
The topology of the CFTR pore is also beginning to be defined. Permeability studies indicate that the narrowest part of the CFTR pore is ∼5.3–6 Å in diameter (31 ,77). The findings that large anions in the intracellular solution cause a voltage-dependent block of CFTR, which is sensitive to the external Cl− concentration, suggest that the CFTR pore contains a large intracellular vestibule where these anions bind and prevent Cl− permeation (67 ,74 ,75 ,118). Because the mutant R347D significantly weakened binding of DNDS and DIDS to CFTR, R347 likely contributes to this site (74). Similarly, at the extracellular end of the CFTR pore, there is probably a large extracellular vestibule. Consistent with this idea is the accessibility of MTS reagents when added to the extracellular solution (31 ,32) and the structure of purified Pgp which suggests that Pgp has a large central pore that is closed at the intracellular side of the membrane (106).
Between the intra- and extracellular vestibules the pore appears to narrow. This constriction may be formed by a pore loop composed of residues at the intracellular end of M6, including R352 (32). Alternatively, because simultaneous mutation of the residues T338 and T339 to alanine (TT338,339AA) significantly increased both the single-channel conductance and the permeability of large anions, the residue(s) T338 and/or T339 may contribute either directly or indirectly to this constriction (77). This constriction may contribute to the anion selectivity filter of the CFTR pore. However, because the mutations T338C and T339C did not react with MTS reagents, the side chains of these residues do not interact with permeating ions (31 ,77). Instead, the peptide backbone of M6 may play an important role in determining the anion selectivity of CFTR (78).
III. REGULATION BY ADENOSINE 5′-TRIPHOSPHATE HYDROLYSIS: THE NUCLEOTIDE-BINDING DOMAINS
A. Requirement of ATP for Gating
The NBDs of CFTR contain a number of highly conserved sequences that were predicted to bind and hydrolyze intracellular MgATP (Fig. 3). These include the Walker A, Walker B, and LSGGQ motifs. Structural and functional studies of ATPases and ABC transporters suggest that the Walker A lysine interacts with either the α- or γ-phosphate of ATP and is essential for ATP hydrolysis, whereas the Walker B aspartate coordinates Mg2+ in MgATP and is required for ATP binding. Similarly, studies of GTP-binding proteins suggest that the conserved glutamine in the LSGGQ motif plays a key role in GTP hydrolysis. In CFTR, the importance of the NBDs is highlighted by the location of many CF-associated mutations in these domains (142). Although many CF-associated mutations in the NBD disrupt the biosynthesis of CFTR, some mutations that are associated with severe disease are correctly processed and delivered to the plasma membrane (141). This suggests that these mutations disrupt the function of CFTR.
Because the NBD were predicted to interact with ATP and because CFTR formed a regulated Cl− channel, Anderson et al. (3) investigated whether ATP regulated the activity of CFTR Cl− channels by using excised inside-out membrane patches from cells expressing recombinant wild-type CFTR. Addition of MgATP by itself to the intracellular solution had no effect, but once the channel was phosphorylated by PKA, MgATP activated channels (Fig. 4). When PKA and MgATP were removed, the channels closed. Then, the readdition of MgATP reactivated CFTR Cl− channels. These results indicate that MgATP regulates the channel, but only when the channel has first been phosphorylated with PKA. Other evidence suggesting a direct interaction of ATP with CFTR, not due to reversible phosphorylation, include the following. 1) Once the channel was phosphorylated by PKA, reversible regulation by ATP was unaffected by a number of kinase and phosphatase inhibitors. 2) CFTR Cl− channels in which part of the R domain had been deleted still required ATP to open (see below). 3) Adenosine 5′-triphosphate-dependent regulation of CFTR showed broad nucleotide specificity, unlike many known kinases and phosphatases. 4) Adenosine 5′-O-(3-thiotriphosphate) (ATPγS) serves as a cofactor for PKA phosphorylation but did not substitute for ATP in opening phosphorylated CFTR Cl− channels.
To understand better how ATP regulated channel gating, the effect of ATP on single phosphorylated CFTR Cl− channels was investigated using excised inside-out membrane patches and lipid bilayers (53 ,73 ,133 ,140 ,145). Figure 5 shows an example of the effect. As the ATP concentration increased, the mean closed time decreased, but mean open time did not change. Analysis of the kinetics of channel gating suggested that the gating behavior of CFTR was described by a minimal model containing one open and two closed states (C1 ➭→ C2 ➭→ O). Of course more complex models can also readily explain gating behavior. The data also suggested that ATP regulated channel gating through an interaction that increases the rate of transition from a long-lived closed state (C1) to a bursting state in which the channel flickers back and forth between an open state (Co) and a short-lived closed state (C2). However, several different kinetic models have been proposed to describe the gating behavior of CFTR (23 ,43 ,47 ,54 ,140 ,145).
Adenosine 5′-triphosphate interacts with both NBDs to control channel gating. Evidence supporting this idea include the following. 1) Analysis of the relationship between ATP concentration and either P o or bursting rate suggested that ATP may interact with two different sites in CFTR (7). 2) The CFTR variants containing site-directed mutations at crucial residues in the NBDs had different effects in NBD1 and NBD2. Substitution of either the Walker A lysine or the Walker B aspartate in NBD1 decreased CFTR activity; however, the mutations had different effects on absolute activity, on specific gating steps, as well as on the response to ATP and ADP (7 ,23 ,54 ,122 ,143). 3) Adenosine 5′-diphosphate inhibits channel gating by competing with ATP (7 ,110). Because mutations in NBD2 relieved this inhibitory effect, whereas mutations in NBD1 had no effect, the competition between ATP and ADP probably occurs at NBD2 (7). Thus the data suggest that ATP interacts with both NBDs. However, the data also suggest that the two NBDs are not functionally equivalent.
B. Hydrolysis of ATP
In the ABC transporter family, the NBDs are the site of ATP hydrolysis. In some members of the family, the energy released during the hydrolysis of ATP is used to actively transport substrate across the cell membrane. However, it was unclear why CFTR should have two domains that might hydrolyze ATP because once an ion channel is open ions flow passively down a favorable electrochemical gradient. Nevertheless, initial studies of channel gating by ATP indicated that nonhydrolyzable ATP analogs and Mg2+-free ATP were unable to support channel activity (3). The data suggested that ATP hydrolysis may be required for channel activation. However, there are additional reports that a very low level of activity can be observed in the absence of Mg2+ (23 ,107) and after removal of ATP from an activated channel (19). Such results may have implications for models of regulation, as described below.
Several biochemical studies suggest that CFTR functions as an ATPase. First, Ko and Pedersen (71) used three different ATPase assays to demonstrate that a wild-type recombinant NBD1 protein catalyzed the hydrolysis of ATP but that mutant NBD1 did not. Second, Randak et al. (97) showed that a recombinant NBD2 protein functioned as an active ATPase, GTPase, and adenylate kinase. Third, Li et al. (73) demonstrated that purified, reconstituted CFTR protein catalyzed the hydrolysis of ATP, albeit at a slower rate than either Pgp or the Na+-K+-ATPase. Interestingly, the rate of ATP hydrolysis of purified, reconstituted CFTR protein (0.5–1.0 molecules ATP/s) is comparable to the rate of channel gating (1–2 transitions/s). Based on these data and the findings that either chemical modification or mutation eliminated both ATPase activity and channel gating, Li et al. (73) speculated that channel gating is directly coupled to ATP hydrolysis.
Site-directed mutation of conserved residues in the NBDs also suggest that ATP hydrolysis at the NBDs regulates channel gating. Mutation of the conserved Walker A lysine residues in the NBDs produced specific alterations in channel gating without altering the permeation properties of CFTR (23 ,54). Mutations in NBD1 decreased the frequency of bursts, whereas mutations in either NBD2 or both NBDs simultaneously dramatically prolonged the duration of bursts of activity and decreased the frequency of bursts. Because none of the Walker A lysine mutations affected the affinity of CFTR for ATP and because similar mutations in related proteins slow the rate of ATP hydrolysis, the data were interpreted to suggest that ATP hydrolysis at NBD1 initiates a burst of channel activity, whereas ATP hydrolysis at NBD2 terminates a burst of activity (23 ,47).
Based on these data and sequence similarity between CFTR and GTP-binding proteins in the LSGGQ motif (82), Carson and Welsh (25) predicted that the terminal residue in the LSGGQ motif of CFTR may correspond to a highly conserved glutamine residue in GTP-binding proteins that directly catalyzes the GTPase reaction. Consistent with this idea, mutations of this residue in either NBD1 or NBD2, which were predicted to increase or decrease the rate of hydrolysis, altered the duration of channel open and closed times in a specific manner without altering ion conduction or ADP-dependent inhibition. Based on these data, they speculated that the rates of ATP hydrolysis at NBD1 and NBD2 determine the duration of the closed and open states of the channel, much as the rate of GTP hydrolysis by GTP-binding proteins determines the duration of their active state.
C. Effect of Nucleotide and Phosphate Analogs
In contrast to ATP, the nonhydrolyzable ATP analog AMP-PNP does not support channel activity by itself (24 ,64 ,110). However, Quinton and Reddy (96 ,98) showed that AMP-PNP in the presence of ATP and cAMP agonists increased CFTR Cl− currents in the apical membrane of sweat duct and T84 intestinal epithelia. Subsequent single-channel studies provided further information about the mechanism. In guinea pig cardiac myocytes, AMP-PNP in the presence of ATP and PKA at 25°C greatly prolonged the duration of bursts of CFTR Cl− channels (64). Figure 6 shows an example of similar results using recombinant CFTR Cl− channels. Because in the presence of ATP, AMP-PNP “locked” open channels that had a high P o , but was without effect on channels that had a low P o , Hwang et al. (64) proposed that AMP-PNP only interacts with highly phosphorylated CFTR Cl− channels and that the interaction of ATP with one site on CFTR was required for the interaction of AMP-PNP with a second site. Hwang et al. (64) concluded that 1) ATP hydrolysis at one NBD (probably NBD1) was a prerequisite for channel opening, 2) ATP hydrolysis at the other NBD was a prerequisite for channel closure, and 3) there was cross talk between the two NBDs. Consistent with this, Carson and co-workers (23 ,24) found that AMP-PNP only affected the channel in the presence of ATP and in the presence of PKA, which maximizes channel phosphorylation. However, some uncertainty about the effect of nonhydrolyzable analogs persists. Schultz et al. (110) have noted that because some nonhydrolyzable ATP analogs do not alter gating, they may not interact with the channel.
The Pi analogs VO4 and BeF3 potently inhibit ATPases by tightly binding at the site where Pi is released following the hydrolysis of ATP (12). Both VO4 and BeF3 greatly prolonged the duration of bursts of activity of CFTR Cl− channels that had been opened by PKA and ATP, but by themselves they did not support channel activity. Because VO4 and BeF3 cannot interact with ATPases until after ATP has been hydrolyzed and Pi released, Baukrowitz et al. (12) speculated that VO4 and BeF3 “lock” CFTR Cl− channels in the open configuration by interrupting cycles of ATP hydrolysis.
Like VO4 and BeF3 , polyphosphates, such as PPi , greatly prolong the duration of bursts of channel activity (26 ,53). However, kinetic analysis of the effect of PPi on single CFTR Cl− channels indicate that PPi affects two steps in channel gating. First, PPi increased the rate at which channels opened. Second, once channels were open, PPi greatly delayed the rate of channel closure. Carson et al. (26) speculated that PPi interacts with NBD2 where it regulates channel opening by NBD1, and then, because it is not hydrolyzed, it also slows the rate of NBD2-mediated channel closure. Consistent with this idea, mutations of conserved residues in NBD2, but not NBD1, significantly reduced the stimulatory effect of PPi on CFTR (33).
D. Two Conformationally Different Open States
The input of energy from ATP hydrolysis predicts that CFTR would have discrete conformations that do not exist at thermodynamic equilibrium. Moreover, the sequence of ATP hydrolysis by the two NBDs predicts that the channel will progress through an ordered, relatively irreversible series of distinct conformational states. Many studies with the patch-clamp technique have identified two conformations of the protein: the open and the closed states. However, with only two recognizable gating conformations, open and closed, it is not possible to observe directly asymmetric transitions in the gating cycle. An indication that additional conformations exist first came from the report of Gunderson and Kopito (54) that the open state may have two discrete conductances. Subsequent studies showed that the two “apparent” conductance states were due to block of the open channel by the buffer MOPS (67). The buffer MOPS blocks the open channel by binding to a site ∼50% of the way through the electrical field and in doing so reveals two distinct states, O1 and O2. Figure 7 shows an example of recordings obtained in the presence of 10 mM MOPS at 25°C. Figure 7, top, shows tracings that were heavily filtered at 10 Hz, and Figure 7, bottom, shows traces filtered at 500 Hz. Within a burst of activity, two different gating patterns were often observed. For example, in trace a in Figure 7, the first two-thirds of the burst show a flickery pattern of gating in which the open state was frequently interrupted by very short closings. In contrast, in the last one-third of the burst, the open state was less frequently interrupted.
The two gating states showed a strong asymmetry during bursts of activity; the first opening into a burst was most frequently into the O1 state, whereas the last opening before exit from a burst tended to be in the O2 state. Not all transitions proceeded in the sequence C-O1-O2-C, suggesting the reversibility of some transitions. Nevertheless, the majority of the transitions were asymmetric, C-O1-O2-C. Studies using nonhydrolyzable nucleoside triphosphates and site-directed mutations showed that interventions that blocked or slowed hydrolysis prevented the transition to the O2 state and prolonged the O1 state (54 ,67). These data indicate that ATP hydrolysis by the NBDs drives a series of asymmetric transitions in the gating cycle. They also indicate that ATP hydrolysis changes the conformation of the pore, thereby altering MOPS binding. These data link the activity of the NBDs to alterations in the CFTR pore and indicate that input of external energy from ATP hydrolysis causes a conformational change in the pore.
E. ATP-Dependent Gating Cycle
Several different models have been proposed to account for the regulation of channel gating by ATP hydrolysis (23 ,47 ,54). At present, it is difficult to develop one model that fits all the data, and it is therefore difficult to choose between models proposed by different groups. Here we describe two models that incorporate some of the features of earlier models. The important point to remember is that the main function of models is to help with the interpretation of the data and to help in the development of hypotheses that can be tested in the future.
Figure 8 shows a model modified from Carson et al. (23). This model proposes that ATP binds to both NBDs in the closed state. The hydrolysis at NBD1 opens the channel. Closing of the channel is the result of hydrolysis at NBD2. This model is a minimum model in that additional steps are almost certainly present and could be added to account for increased complexity. In the closed state, C1, the channel is closed and is not bound to ATP. This is consistent with the finding that in the absence of ATP, the channel does not open. The channel then progresses from the C1 state to the C2 and C3 states. Although for simplicity the model shows an ordered sequence of binding, the order might be reversed, or each NBD might bind ATP independently. In the C3 state, both NBDs are bound to ATP, but the channel remains closed. This result is consistent with observations that nonhydrolyzable nucleoside triphosphates are not able to open the channel. These results are also consistent with the observations that ATP concentration is an important determinant of the closed time between bursts of activity. Moreover, interventions that alter the rate of ATP binding, such as competitive inhibition by ADP or site-directed mutations in the NBDs, could result in a prolonged closed state between bursts of activities. If this interpretation is correct, then it is possible that some of the prolonged interburst intervals obtained with mutations in the NBDs might be overcome by increasing the ATP concentration. In the model shown in Figure 8, we also show that phosphorylation of the R domain influences ATP binding. This result is consistent with the observation that phosphorylation changes the duration of the interburst interval by changing the rate of ATP binding to the channel (see below). This model, however, makes no prediction about which NBDs interact with the R domain.
The channel opens with the transition from the C3 state to the O1 state. This step is modeled as occurring with hydrolysis of ATP by NBD1. Thus the rate of hydrolysis at NBD1 would also determine the duration of the long closed state between bursts of activity. Mutations predicted to decrease the rate of hydrolysis at NBD1, for example, K464A and Q552A, would thus increase the interburst interval.
The transition from the O1 to the O2 state is due to hydrolysis at NBD2. This transition would explain the large increase in duration of the O1 state observed when hydrolysis is slowed by site-directed mutations in NBD2 or by the addition of nonhydrolyzable analogs such as AMP-PMP. In addition, the observation that the H1350Q mutation, which would be predicted to increase the rate of hydrolysis, shortened the burst duration supports the notion that NBD2 hydrolysis is responsible for the O1 to O2 transition.
Finally, either release of the products of hydrolysis, ADP and Pi , from either one or both of the NBDs closes the channel. It is possible that manipulations that alter the binding of ATP to the NBDs might also alter the release of ADP. However, at this time, we have no knowledge of whether ADP or Pi remains bound to either of the NBDs.
An alternative model is shown in Figure 9. This model is based on the idea that both NBDs have symmetrical functions and is similar to the regulation of heterotrimeric G proteins (48). In contrast to the model shown in Figure 8, in this model hydrolysis of ATP at one NBD results in the release of ADP from the other NBD. When ATP binds to the vacant NBD, the channel activates, resulting in the gating transition either from an open to a closed or from a closed to an open state. Following the G protein analogy, the effect of the R domain is similar to a receptor or a nucleotide exchange factor that stimulates nucleotide exchange and/or a high affinity for GTP.
In the model shown in Figure 9, there are two closed states, C1 and C2, and two open states, O1 and O2. The transition from the closed C2 state to the open O1 state occurs when ATP binds at NBD2. Thus a major difference between this model and that shown above is that ATP binding opens the channel into the O1 state much like the binding of GTP activates a G protein. A model in which ATP binding causes conformational and gating changes also offers the possibility that binding is reversible. This could give rise to transitions from O1 to C2. In addition, phosphorylation of the R domain stimulates this step (see below). The transition from the O1 to the O2 state depends on hydrolysis at NBD2, just as it did in the model described above. Finally, the transition from the O2 to the closed C1 state depends on ATP binding at NBD1. Thus, in this model, the steps following channel opening are symmetrical with the steps involved in channel closing, with the two NBDs controlling different steps. As with the first model, the concentration of ATP would influence the duration of the long closed state between bursts of activity. However, measurements of the duration of the O1 and O2 states and the effect of ATP concentration could allow one to distinguish between these models. Although there have been no measured changes in the burst duration as a function of ATP concentration, there are several potential explanations. First, the rate of ATP binding may not be the rate-limiting step during the channel open time. Second, changes in O2 might go undetected unless the duration of the O2 state was specifically measured.
Two other models have been proposed to describe the gating behavior of CFTR Cl− channels. Gadsby and colleagues (45–47) developed a model that describes the regulation of channel gating by both incremental phosphorylation and ATP hydrolysis. In this model, partially phosphorylated channels interact with ATP only at one NBD (probably NBD1), resulting in brief channel openings. However, once fully phosphorylated, the interaction of ATP with one NBD allows the interaction of ATP or AMP-PNP with the second NBD, and hence, the open state of the channel is stabilized. In contrast, the model developed by Gunderson and Kopito (54) suggests that ATP hydrolysis at NBD2 controls channel gating, whereas ATP hydrolysis at NBD1 converts CFTR from an inactive to an active closed conformation.
As should be obvious even to the casual reader, much additional work needs to be done. An important point from all the models proposed is that there is an irreversible cycle of gating due to the high free energy of ATP binding and hydrolysis. Future work will allow further insights into the mysteries of how ATP and phosphorylation control the gating of the CFTR Cl− channel.
F. Implications for the Function of Other ABC Transporters
The CFTR belongs to the ABC transporter family of proteins (65). Members of this family have very diverse functions but share some structural features. The family is defined by the presence of NBDs, which have a highly conserved primary sequence, and the presence of MSDs, which share topological features but have very different amino acid sequences.
Based on our findings with CFTR, we speculate that a common mechanistic feature of ABC transporters is that energy from ATP hydrolysis by the NBDs causes conformational changes in the MSDs. In this way, the conserved NBDs could serve as a common engine that is coupled to a diverse set of MSDs. Depending on the structure of the MSDs, the conformational changes could have very different functional consequences. Consider the following examples. First, an ABC transporter (CFTR) can form an ion channel. In CFTR, ATP hydrolysis produces conformational changes that open and close the pore. Second, some ABC transporters regulate associated membrane proteins; an example is the sulfonylurea receptor that regulates an inwardly rectifying K+ channel (66). In sulfonylurea receptors, we speculate that ATP hydrolysis produces conformational changes in the MSD that alter the interaction with and thus the activity of associated K+ channels. Third, many ABC transporters perform active transport; examples include the multidrug resistance transporter (49) and prokaryotic transporters such as the maltose and oligopeptide transporters (60 ,121). In these, ATP hydrolysis may change the conformation of the MSDs which alters a substrate binding site, in a manner analogous to conformational changes that alter MOPS binding in CFTR. Alterations in substrate binding affinity could form, in part, the basis of active transport across a cell membrane. Thus a model for active transport that involves changes in substrate affinity might include transitions similar to those in CFTR, with the exception that the MSDs would form an occluded pathway, rather than an open pore.
This general model is supported by two additional findings. First, studies of CFTR-STE6 chimeras indicate that the NBDs from two different family members have functional and mechanistic similarities. STE6 is an ABC protein that transports the mating pheromone a-factor across the plasma membrane of yeast. When Teem and co-workers (131 ,132) replaced part of NBD1 of STE6 with the analogous portion of CFTR NBD1, they found that the chimeric STE6 was still capable of transporting a-factor. Second, it is the MSDs that appear to confer the unique functions of the diverse family members. For example, in CFTR, sequences in M6 and M12 line the ion channel pore, determining its passive conductive properties (31 ,87 ,117 ,128). In the multidrug resistance transporter, sequences in M6 and M12 can mediate the interaction between substrate and the protein, determining the selectivity of active transport (79).
The model of a highly related set of NBDs driving a very divergent set of MSDs provides for functional flexibility in ABC transporters. It seems very likely that the NBDs have a common mechanism of extracting energy from ATP; thus they form a standard engine that has been conserved through evolution. That standard engine is then coupled to a highly varied group of MSDs that have diverged through evolution to perform many different functions. When the NBDs change the conformation of the MSDs, ABC transporters can accomplish active transport, regulation of associated proteins, and the gating of an ion channel.
IV. REGULATION BY PHOSPHORYLATION: THE R DOMAIN
A. Phosphorylation of Residues in the R Domain
The predicted amino acid sequence of CFTR includes the unique R domain that contains a number of potential phosphorylation sites for PKA and protein kinase C (PKC) (105). Protein kinase A favors the consensus phosphorylation sequence R-R/K-X-S*/T* > R-X-X-S*/T* = R-X-S*/T*, where X is any amino acid, and the phosphoacceptor is indicated by an asterisk (for a review, see Ref. 70). There are 10 dibasic (R-R/K-X-S*/T*) PKA consensus sequences within CFTR: 8 serines (S660, S686, S700, S712, S737, S768, S795, and S813) plus 1 threonine in the R domain (T788) and 1 serine just before NBD1 (S422). However, PKA does not phosphorylate all 10 dibasic phosphorylation sites. Both T788 and S422 are poorly conserved across species, and neither residue was phosphorylated in vivo by cAMP agonists (30 ,94). Moreover, PKA phosphorylates a purified recombinant R domain peptide to a stoichiometry of only ∼5 mol/mol in vitro (94). Consistent with these data, site-directed mutagenesis and tryptic phosphopeptide mapping demonstrated that five serines in the R domain are phosphorylated by PKA in vivo (30 ,94): S660, S700, S737, S795, and S813. An isolated NBD1-R domain protein revealed PKA-dependent phosphorylation of these residues plus S712 and S768 when evaluated by mass spectrometry (135). When the other consensus phosphorylation serines are mutated to alanine, PKA also phosphorylates S753 (115). Protein kinase C phosphorylates a purified recombinant R domain peptide to a stoichiometry of ∼2 mol/mol in vitro and phosphorylates two serine residues in vivo (94): S686 and S700. However, stimulation of cells expressing CFTR with PKC activators before stimulation with cAMP agonists increases the phosphorylation of CFTR, suggesting that PKC may facilitate the phosphorylation of CFTR by PKA (68 ,125).
B. Control by Kinases and Phosphatases
Regulation of CFTR Cl− channel activity is precisely controlled by the balance of kinase and phosphatase activity within cells. Protein kinase A is the most important kinase responsible for regulating CFTR Cl− channel activity. However, other kinases can phosphorylate CFTR and open the channel. Initial studies suggested that both Ca2+-independent and Ca2+-dependent isoforms of PKC phosphorylate CFTR and activate CFTR Cl− channels, but only to ∼15% of the magnitude of that observed with PKA (18). They also suggested that PKC greatly potentiates the onset and magnitude of channel activation when PKA is subsequently applied (125). However, recent data suggest that PKC may play a more important role in channel activation than previously recognized. Using experimental protocols that prevent the phosphorylation of CFTR by PKC, Jia et al. (68) showed that PKA, by itself, fails to activate CFTR Cl− channels expressed in BHK and CHO cells. They speculate that constitutive phosphorylation of CFTR by PKC is required for PKA-dependent phosphorylation to activate CFTR Cl− channels.
Regulation of CFTR by the cGMP-dependent protein kinase (cGK) is isotype specific. Although both type Ia cGK purified from bovine lung and type II cGK (cGKII) isolated from pig small intestine phosphorylate CFTR, only cGKII activates CFTR Cl− channels (18 ,138). When compared with PKA, the magnitude of Cl− current activated by cGKII is similar, although the onset of activation is slowed probably because cGKII must first become membrane-associated before it can regulate CFTR. Recent studies suggest that the tyrosine kinase p60c-src may also phosphorylate and activate CFTR Cl− channels by a mechanism independent of PKA (44). In contrast, the multifunctional Ca2+/calmodulin-dependent protein kinase II failed either to phosphorylate or activate CFTR Cl− channels, suggesting that it has no direct effect on CFTR (18). This result is consistent with evidence suggesting that intracellular Ca2+ does not regulate CFTR Cl− channels.
When PKA is removed, the activity of CFTR Cl− channels in excised inside-out membrane patches decreases with time “runs down” even in the continued presence of MgATP. This channel inactivation likely results from the dephosphorylation of CFTR by membrane-associated phosphatases. In airway and intestinal epithelia, okadaic acid and calyculin A, inhibitors of protein phosphatase (PP) 1 and 2A, did not prevent the inactivation of CFTR Cl− currents when cAMP agonists were removed (18 ,136). However, in cardiac myocytes and NIH 3T3 cells expressing endogenous and recombinant CFTR, respectively, PP2A probably plays an important role (18 ,47 ,63). Because the rundown of CFTR Cl− channels in excised membrane patches is unaffected by the removal of Ca2+ from the intracellular solution, the Ca2+- and calmodulin-dependent PP2B is unlikely to dephosphorylate and inactivate CFTR. Other data indicate that some preparations of alkaline phosphatase dephosphorylate and inactivate CFTR Cl− channels expressed in CHO and human airway cells, whereas phenylimidazothiazole drugs that inhibit alkaline phosphatases activate wild-type and mutant CFTR Cl− channels expressed in CHO cells (15). These results suggest that PP2C or an unidentified protein phosphatase with properties similar to those of alkaline phosphatase may dephosphorylate CFTR in airway and intestinal epithelia. Consistent with the former idea, PP2Cα is expressed in airway and intestinal epithelial and recombinant PP2C dephosphorylates CFTR and inactivates CFTR Cl− channels (136).
C. Effect of Phosphorylation on Gating
Studies of CFTR variants that contain site-directed mutations in which multiple serines in the dibasic consensus PKA phosphorylation sequence are simultaneously mutated to alanine indicate that PKA-dependent phosphorylation of multiple serines in the R domain can open the channel (17 ,28 ,30 ,115). Moreover, as the total number of phosphoserines was decreased by mutation, the P o of the channels decreased. A surprising observation was that PKA still opened the channel when all of the serines that are phosphorylated in vivo or all of the potential phosphoserines and phosphothreonines were mutated to alanine (28 ,101 ,115). Similarly, PKC still potentiated the magnitude of channel activation by PKA in the 10SA mutant (28). These results suggest that there may be additional weakly phosphorylated sites in the R domain that are substrates for PKA and/or PKC. One such site is S753, which occurs in a R-X-S*/T* PKA consensus phosphorylation sequence. The mutation 10SA-S753A eliminates much of the residual PKA-dependent phosphorylation of the 10SA mutant and decreases the P o of the 10SA mutant by 40% (115).
Several studies have investigated the contribution of individual phosphoserines to the activation of CFTR by PKA-dependent phosphorylation (17 ,28 ,30 ,115 ,144 ,146). The data indicate that individual phosphorylation sites do not contribute equally to the activation of CFTR: S660, S795, and S813 appear to be functionally most important for stimulation, since mutations of these sites resulted in large decreases in CFTR Cl− channel activity (Fig. 10). Interestingly, mutation of two phosphorylation sites, S737 and S768, to alanine increased CFTR Cl− current in Xenopus oocytes (144). Reasons for the difference in the effect of the S737 mutation on channel activity in mammalian cells and in Xenopus oocytes are not presently known. This suggests that phosphorylation of these sites by PKA may inhibit CFTR in that system (144). The decrease in channel activity observed in individual or multiple dibasic phosphorylation site mutations is caused by a decreased rate of channel opening without altering the rate of channel closure. These results suggest that PKA-dependent phosphorylation state of the channel determines the interaction of ATP with the NBDs (53 ,73 ,145 ,146); the same may also be true for PKC (68). The data also suggest that PKA-dependent phosphorylation may alter the regulation of CFTR by intracellular ATP. Consistent with this idea, mutation of individual or multiple phosphoserines altered the affinity of ATP for CFTR (Fig. 10) (146). Thus the phosphorylation state of the R domain appears to regulate the interaction of ATP with the NBDs.
Additional evidence for phosphorylation-dependent control of CFTR channel activity comes from observations that there appear to be both high and low phosphorylation states correlating with a high and low P o (43 ,64). The action of AMP-PNP appears to be restricted to the highly phosphorylated state. These studies suggesting that modal gating of CFTR depends on the phosphorylation state also suggest that ATP hydrolysis is controlled by the phosphorylation state of the channel.
Which phosphorylation sites are the most important for regulation of channel activity by PKA? For two reasons, it has been proposed that S660, S700, S737, S795, and S813 are the major regulatory sites for PKA-dependent phosphorylation in vivo. First, those sites were phosphorylated in vivo after stimulation with cAMP agonists. Second, most of the decrease in P o occurred with simultaneous mutation of S660, S737, S795, and S813. Mutation of additional serines produced only a small additional decrement in P o . However, additional studies are required to understand further how individual sites in the R domain control CFTR activity.
D. Effect of Structural Alterations on the R Domain
Additional evidence that the R domain serves to regulate the channel comes from studies in which the R domain was partially deleted. Deletion of part of the R domain (residues 708–835, CFTRΔR) produced Cl− channels with biophysical properties similar to those of wild-type CFTR with the exception that they were constitutively active in the presence of MgATP even without phosphorylation (17 ,81 ,102 ,103 ,146). However, the P o of CFTRΔR was much reduced compared with that of wild-type CFTR and AMP-PNP and PPi failed to stabilize the open state of CFTRΔR, suggesting that the R domain may regulate the interaction of nucleotides with the NBDs (81 ,101). More extensive deletions of the R domain failed to generate functional CFTR Cl− channels (103). Although CFTRΔR produced a channel that was active even without phosphorylation, PKA-dependent phosphorylation produced a further increase in P o (17 ,101). This effect was attributed to S660, which remains in the sequence of CFTRΔR and is a substrate for PKA-dependent phosphorylation. When S660 in CFTRΔR was mutated to alanine (CFTRΔR-S660A), PKA failed to stimulate channel activity. These results suggest that in the construct CFTRΔR-S660A any residual PKA regulatory phosphorylation sites were either removed by deletion of part of the R domain or their influence on channel activation was eliminated by the deletion. However, the possibility cannot be ruled out that PKA cannot interact with CFTRΔR-S660A, even though it does interact with CFTRΔR.
Like CFTRΔR, the amino-terminal portion of CFTR (D836X, which contains MSD1, NBD1, and the R domain) formed Cl− channels that displayed some channel activity even without PKA-dependent phosphorylation (116). The activation of D836X Cl− channels by PKA indicates that the MSD1-NBD1 motif must contain many of the sequences with which the R domain interacts to control CFTR. The PKA-independent activity of D836X Cl− channels suggests that the R domain may also interact with region(s) of the second half of CFTR that are missing in D836X. Because deletion of part of the R domain at least partially suppressed the effect of a mutation in NBD2, the R domain may interact with NBD2 (102). Alternatively, the PKA-independent activity of D836X could be due to interference between R domains in a D836X homomultimer.
Substitution of six or more negatively charged aspartates for serines in the R domain generated Cl− channels that were open, even without phosphorylation by PKA (101). Thus multiple aspartate substitutions mimicked the effect of serine residues. The data also suggest that net negative charge or the charge density of the R domain may play an important role in channel activation, perhaps by electrostatic interactions with the protein. However, several lines of evidence suggest that an alteration in the conformation of the R domain might be a key step in channel activation. First, covalent modification of the R domain with the neutral hydrophobic adduct N-ethylmaleimide at residue C832 stimulated channel activity and increased the potency of ATP-dependent stimulation of CFTR Cl− channel activity (34). Interestingly, this modification occurred at a site that is not phosphorylated. Second, simultaneous mutation of three proline residues in the R domain to alanine (P740,750,759A) generated CFTR Cl− channels that were dramatically more active than wild type, while cyclophilin A, a cis-trans peptidyl-prolyl isomerase, greatly stimulated the activity of wild-type CFTR Cl− channels, but not that of P740,750,759A Cl− channels (152). Third, studies using circular dichroism spectroscopy demonstrated that either phosphorylation by PKA or mutation of multiple phosphoserines changed the conformation of purified R domain peptide (40 ,41). Thus multiple different structural alterations of the R domain are able to activate the channel.
Based on comparison of the amino acid sequence of the R domain from different species, Dulhanty and Riordan (42) proposed a two-domain model of the R domain: RD1 (amino acids 587–672) and RD2 (amino acids 679–798). The sequence of RD1 is highly conserved across species, but that of RD2 is not, despite the fact that this region contains the majority of the phosphorylation sites. Cystic fibrosis-associated mutations in RD1, but not those in RD2, disrupt the biosynthesis of CFTR (8). In addition, deletion studies of portions of the R domain suggest that there are two halves to the R domain with the first conserved in other ABC transporters and the second unique to CFTR (102 ,103). These data suggest that the structure of RD1 is tightly constrained but that of RD2 is not.
E. A Model of R Domain Function
The R domain was originally proposed to regulate CFTR by keeping the channel closed at rest. It was proposed that the unphosphorylated R domain was an inhibitor that maintained the CFTR pore in a closed state. Thus the function was proposed to perhaps be similar to that of the inhibitory “ball” of Shaker K+ channels (62 ,151). As described above, phosphorylation by PKA, deletion of part of the R domain, or substitution of six or more negatively charged aspartates for serines in the R domain relieved this inhibitory effect. Also supporting an inhibitory effect of the R domain was the finding that addition of an unphosphorylated recombinant R domain to CFTR blocked the channel (80).
However, countering the notion that the R domain is just an inhibitor, the activity of phosphorylated CFTR Cl− channels greatly exceeded that of CFTR variants that did not require phosphorylation to open in the presence of ATP (81 ,101 ,146). This suggests that the phosphorylated R domain might stimulate the activity of CFTR. Consistent with this idea, when added to the intracellular solution, a phosphorylated recombinant R domain stimulated CFTR and increased the rate of transition of CFTRΔR-S660A Cl− channels into a burst of activity (Fig. 11) (81 ,146). Only phosphorylated R domain stimulated activity and the R domain failed to influence wild-type CFTR. Mutation of phosphorylation sites in wild-type CFTR had the opposite effect on the same kinetic step, the rate of channel opening. These mutations also altered the relationship between ATP concentration and P o at low ATP concentrations, although all mutants had the same P o as wild-type CFTR at high ATP concentrations (Fig. 10) (146). Phosphorylation also enhances the ATPase activity of CFTR (73).
Thus the R domain does not function solely as an inhibitor that keeps the channel closed; it is not simply an “on-off” switch. Moreover, as individual phosphorylation sites do not contribute equally to activity, phosphorylation may have graded effects in stimulating the channel. The function of the R domain therefore differs mechanistically from the amino-terminal “ball” in Shaker K+ channels that is believed to physically obstruct the channel pore (62 ,151). How could the R domain exert both inhibitory and stimulatory effects? When phosphorylation modifies the R domain, it might have two effects: the first might be permissive, releasing steric inhibition, and the second effect might be stimulatory, facilitating interactions of the NBDs with ATP (Figs. 8 and 9). The data suggest that the R domain has a novel regulatory role. Once phosphorylated, the R domain stimulates ATPase activity and channel gating by a mechanism that is consistent with increased binding of ATP.
In this paper we have described some of our knowledge of the structure and function of CFTR. However, even to the casual reader, it should be apparent that our knowledge remains superficial. Many questions remain for each of the specific domains of CFTR and for CFTR in general.
In the case of the MSDs, additional work is needed to identify sequences that line the CFTR pore. It will be very interesting to learn how selectivity between anions and cations and between different anions is determined. It will be particularly interesting to learn how this compares with the structure of voltage-gated cation channels and the structure of other anion channels such as the ClC family of channels.
The NBDs are intriguing because of their conservation in ABC transporters. The fact that CFTR forms a Cl− channel allows application of the powerful patch-clamp technique to study the function of single CFTR molecules; the results may have broad implications for the ABC transporter family. Future studies will focus on how binding of ATP and its subsequent hydrolysis gates the channel. It will also be interesting to see how the two NBDs interact, because current studies have suggested significant cross talk between the two domains.
Regulation of channel activity by phosphorylation of the R domain also appears complex. Future studies will have to identify the role of the individual phosphorylation sites and interactions between sites to shed light on how graded phosphorylation yields graded channel activity. In addition, regulation by PKC and cGMP-dependent protein kinase may provide additional insights into the involvement of CFTR in secretory diarrhea.
A particularly interesting problem will be to learn how the different domains interact. How is it that ATP hydrolysis by the NBDs causes the pore to open and close. Where is the gate? Are the intracellular loops key to coupling the activity of the NBDs to the channel pore? Does the R domain interact directly with the NBDs to control their activity? In this regard, does the analogy between the NBDs and G proteins extend to the way in which the R domain governs NBD activity?
We think it is clear that a better understanding of the structure and function of CFTR may have broad applications as we begin to better understand the pathogenesis of CF and the pathogenesis of secretory diarrhea and as we begin to develop novel approaches to treat these diseases.
We thank our laboratory colleagues for helpful discussions.
D. N. Sheppard is supported by the Biotechnology and Biological Sciences Research Council and the Cystic Fibrosis Trust. M. J. Welsh is supported by the Cystic Fibrosis Foundation; the National Heart, Lung, and Blood Institute; and the Howard Hughes Medical Institute. M. J. Welsh is an Investigator of the Howard Hughes Medical Institute.
- Copyright © 1999 the American Physiological Society