Molecular models of the open and closed states of the whole human CFTR protein
Cystic fibrosis transmembrane conductance regulator (CFTR), involved in cystic fibrosis (CF), is a chloride channel belonging to the ATP-binding cassette (ABC) superfamily. Using the experimental structure of Sav1866 as template, we previously modeled the human CFTR structure, including membrane-spanning domains (MSD) and nucleotide-binding domains (NBD), in an outward-facing conformation (open channel state). Here, we constructed a model of the CFTR inward-facing conformation (closed channel) on the basis of the recent corrected structures of MsbA and compared the structural features of those two states of the channel. Interestingly, the MSD:NBD coupling interfaces including F508 (ΔF508 being the most common CF mutation) are mainly left unchanged. This prediction, completed by the modeling of the regulatory R domain, is supported by experimental data and provides a molecular basis for a better understanding of the functioning of CFTR, especially of the structural features that make CFTR the unique channel among the ABC transporters.
KeywordsCystic fibrosis Structure Model Channel P-gp
Mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene cause cystic fibrosis (CF), the most common severe inherited genetic disease among the Caucasian population [1, 2]. The CFTR protein belongs to the large superfamily of ATP-binding cassette (ABC) transporters, which are integral membrane proteins that use the energy generated from ATP-binding and hydrolysis to translocate a wide variety of molecules across cellular membranes via an alternating access mechanism . CFTR (ABCC7) belongs to the subfamily C of ABC transporters, which also includes other proteins involved in human diseases, such as the multidrug resistance-associated protein 1 (MRP1, ABCC1) and the sulfonyl urea receptor 1 (SUR1, ABCC8). CFTR is, however, the only one of the numerous members of the ABC superfamily known to function as an ion channel, as it is a phosphorylation-dependent chloride channel located at the apical membrane of epithelial cells [4, 5]. According to the general domain architecture of ABC transporters, CFTR has two membrane-spanning domains (MSDs) and two nucleotide-binding domains (NBDs). However, while prokaryotic ABC transporters have several subunits composed of identical MSDs and NBDs, the CFTR and other eukaryotic ABC proteins have distinct MSDs and NBDs, which belong to a single polypeptide encoded by a single gene. The two NBDs form tightly interacting dimers upon ATP-binding at their interface, which later on dissociate upon ATP hydrolysis . These NBDs movements are transmitted to the MSDs to alternatively open and close the gates to the substrate binding sites. In CFTR, ATP-binding to the NBDs is believed to open the ion pore, whereas ATP hydrolysis leads to the channel closure . In addition, CFTR also has a large regulatory (R) domain, which links the two transporter halves and contains multiple phosphorylation sites, the channel activity being increased upon phosphorylation [7, 8].
Several CF-causing mutations, including the deletion of a phenylalanine at position 508 (ΔF508) which is the most prevalent one, prevent the protein from attaining its native global conformation. Such mutant proteins are consequently degraded by the proteasome, in contrast to the wild-type which is transported from the endoplasmic reticulum to the Golgi [9, 10, 11]. An important role of the F508 residue in the formation of inter-domain contacts was hypothesized by our modeling studies of the NBD1:NBD2 heterodimer [12, 13] and by crystal structures of the wild-type and mutated NBD1s [14, 15], which indicated that F508 was exposed to the solvent and that the ΔF508 mutation did not lead to a significant alteration of the overall NBD1 structure. This was further supported by experimental studies showing that interactions between MSDs and NBDs are disrupted by the ΔF508 mutation [16, 17].
The elucidation of the complete 3D structure of the multidrug transporter Sav1866 from Staphylococcus aureus  constituted a major advance in the overall understanding of the functions of ABC exporters and triggered studies of homology modeling of CFTR in an outward (or extracellular)-facing configuration, representing the ATP-bound state [19, 20]. Indeed, the bacterial and eukaryotic ABC exporters appear to form a relatively homogeneous family, as they share similar MSDs and NBDs, as well as long intracellular loops (ICLs) allowing the communication between their MSDs and NBDs. Even though these homology models have to be taken with caution, especially as regards some CFTR regions such as parts of the MSDs where the sequence identities are very low, they nevertheless provide a precious framework for further investigations of the CFTR functioning and for the design of pharmacotherapeutic approaches for CF. Along these lines, homology models of the CFTR MSD1:NBD1:MSD2:NBD2 assembly [19, 20] have revealed important features of the channel formed by the MSDs, as well as of the interfaces between the NBDs and MSDs. A groove on the surface of the NBD1:NBD2 heterodimer, especially rich in aromatic amino acids (among which F508), makes critical contacts with the intracellular loops 2 (ICL2 from MSD1) and 4 (ICL4 from MSD2), whereas the two other intracellular loops (ICL1 from MSD1 and ICL3 from MSD2) interact with the NBD ATP-binding sites. These studies highlighted in particular the critical position of F508 at the interface between NBD1 and ICL4. This was further supported by the observation that chemical cross-linkings could be established between non-native cysteine residues in the vicinity of F508 and ICL4 of wild-type CFTR, but that they were disrupted in the ∆F508 protein [20, 21]. It should be stressed here that, due to differences in the modeling strategies, the two models mentioned above [19, 20] do, however, differ significantly as regards the ICL4 residues involved. Most importantly, an arginine residue (R1070), located near the ICL4/NBD1 pivot, was observed in our model to establish particularly tight links with the main chain atoms of F508 and proximal residues, with almost all those bonds being lost in the ΔF508 protein . These particular features of our atomic model are at present strongly supported by the recent crystal structure of mouse P-gp (MDR) at 3.8 Å resolution (, see “Discussion”). Of note, recent work has also shown that the R1070P and R1070W mutations lead, respectively, to the absence or to reduced levels of insertion of the protein into the apical membrane of polarized epithelial cells , suggesting that these mutations may, as ΔF508, affect the inter-domain interactions of CFTR. In contrast, the protein with the R1070Q mutation was found to be located at the apical membrane of the polarized epithelial cells . In our model, this last mutation leads only to a relatively limited modification of the ICL4/NBD1 interface characteristics, a finding consistent with a moderate CFTR dysfunction ; however, a more severe phenotype has been observed when this mutation was associated with the nonsense mutation S466X .
Recently, several corrected X-ray structures of the ABC lipid flippase MsbA from three closely related bacterial orthologs were reported  (also see, for reviews, [26, 27]). The 3.7 Å resolution structure of S. typhimurium MsbA (StMsbA) in complex with adenylyl imidodiphosphate (AMPPNP) displayed an outward-facing conformation, which was similar to that observed with Sav1866 (<2.2 Å RMSD between the monomer Cα positions), this latter structure being identical to the crystal structure observed for StMsbA with bound ADP.Vi. Most interestingly, in addition to these two nucleotide-bound, outward-facing conformations, two inward (intracellular)-facing MsbA structures were obtained in the absence of nucleotide; these apo structures may represent the resting state of the exporter. The 5.3 Å resolution inward-facing E. coli MsbA (EcMsbA) structure showed fully dissociated, largely separated NBDs and was thus termed “open apo”, whereas the inward-facing conformation of V. cholerae MsbA (VcMsbA) solved at 5.5 Å resolution was characterized by closer NBDs and thus termed “closed apo”. It is noteworthy that, in the “closed apo” conformation, the NBDs, although closer than in the “open apo” conformation, display a smaller interface than in the outward-facing conformation; indeed, in the “open apo” configuration, the P-loops of the opposing NBDs face each other, whereas the nucleotides are sandwiched between the P-loop of one NBD and the ABC signature sequence of the other NBD in the outward-facing configuration. These structures thus provide a framework to understand the main differences between the inward- and outward-facing conformations.
In the present work, we took into account the different MsbA structures and constructed a model of the human CFTR MSDs:NBDs assembly in an inward-facing conformation, which probably represents the CFTR closed channel state, and compared it to our previous model of the human CFTR MSDs:NBDs assembly in an outward-facing configuration, which is thought to correspond to the CFTR open channel state. The changes between these two physiological states of CFTR were analyzed and discussed. Moreover, the analysis of the CFTR outward-facing conformation allowed us to highlight structural features of the intracellular loops, which are lacking in prokaryotic ABC transporters and may thus account for the unique channel function of CFTR. Furthermore, we also tentatively proposed a two-domain model of the regulatory (R) domain and indicated how this model might accommodate the two states of CFTR. Finally, we also evaluated how our present model of the overall CFTR MSD1:NBD1:R:MSD2:NBD2 assembly fits with recent structural data obtained by electron microscopy  and cryo-electron microscopy , as well as with previous work with electron crystallography .
Materials and methods
Modeling of the MSD:NBD assembly
Modeling of the R domain and of the N- and C-termini of CFTR
Hydrophobic cluster analysis (HCA) [33, 34, 35] was used in combination with other classical bioinformatics tools, in order to detect remote relationships between protein sequences, at high levels of divergence. HCA, which is based on a bidimensional representation of the protein sequence, combines the analysis of the sequence and of secondary structures, the regular secondary structures mainly corresponding to the hydrophobic clusters delineated by the method. The efficiency of this methodology has widely been demonstrated (see, for example, some references at http://www.impmc.jussieu.fr/~callebau). Homology modeling was then performed on the basis of the remote templates detected, using Modeller (9v2 release) . As no template could be detected for some sequence segments within the N-terminus, the C-terminus, and the R domain, those segments were modeled ab initio, taking into account the secondary structure predictions as well as well-defined neighboring segments.
Availability of the models
The 3D coordinates of our models can be downloaded from the following URL (http://www.impmc.jussieu.fr/~callebau/CFTR.html).
Inward-facing conformation of the CFTR MSDs:NBDs assembly
For modeling the inward-facing conformation of the human CFTR MSDs:NBDs assembly on the basis of a MsbA 3D template, we used a global strategy very similar to that which we have previously followed for building its nucleotide-bound, outward-facing conformation, on the basis of the experimental 3D structure of Sav1866 . Figure 1 shows the alignment of the MSDs and NBDs of S. aureus Sav1866 with those of human CFTR, as previously reported , to which we added the sequences of the three closely related bacterial MsbA orthologs already mentioned above: E. coli Msba (EcMsbA), V. cholerae MsbA (VcMsbA), and S. typhimurium MsbA (StMsbA). The Sav1866 and MsbA sequences share 22 and 51% identity within their MSDs and NBDs, respectively, with few insertions and deletions. Sequence identities between CFTR and Sav1866 are comparable to those between CFTR and MsbA (CFTR(A)/Sav1866: 14% (MSD) and 26% (NBD); CFTR(A)/VcMsbA: 10% (MSD) and 22% (NBD); CFTR(B)/Sav1866: 12% (MSD) and 27% (NBD); CFTR(B)/VcMsbA: 12% (MSD) and 27% (NBD)).
As already indicated above, MsbA can undergo a large range of motions between the apo and nucleotide-bound states . We chose here to model a possible, inward (cytoplasmic)-facing conformation of CFTR, likely representing the closed state of the channel, on the basis of the closed-apo 3D structure of VcMsbA (pdb 3b5x, ), in which the NBDs are not completely separated. Indeed, we thought that this conformation may be more relevant than any other mode of NBD separation, for several reasons. First, the CFTR-specific NBD1 regulatory insertion (RI) can indeed interact, in this closed-apo conformer, with the modified LSHGH signature sequence of NBD2 (see below). Moreover, it is likely that the R domain (which is C-terminal to NBD1) might prevent the complete dissociation of the CFTR NBDs, similar to what was observed in the structure of MalK NBDs in the absence of nucleotides . Another factor which may hinder the complete dissociation of the NBDs is the very poorly hydrolysable ATP, which remains tightly bound to the NBD1 non-canonical binding site for several minutes ), a time during which several gating cycles can occur . Finally, one can also argue that a wide separation of the NBDs may not be required for the functioning of CFTR as a chloride channel, while a large opening resulting from a complete and wide separation of the NBDs may be important for MsbA and other ABC exporters to allow access of very large substrates to the accessibility chambers.
Such a “closed-apo”-based model of CFTR is further supported by electron crystallography data , which showed a good agreement of the low resolution electron density of the 2D, ATP-free crystal 1-form of CFTR with the closed-apo VcMsbA structure solved at that time. Although that MsbA structure was incorrect, the whole volume that it occupies is similar to that occupied by the corrected VcMsbA structure which has been published since ; this is clearly also in favor of a limited separation of the NBDs for the apo configuration of CFTR (see also "Compatibility of the whole CFTR models with low resolution experimental data", below).
General features of the model
Figure 2a shows the resulting whole model of the MsbA-based, inward-facing CFTR MSDs/NBDs assembly, compared to our previous outward-facing model constructed using the Sav1866 experimental structure as template. Of note, the difference between the large and thin ribbons of the inward-facing model is due to the low resolution of the MsbA template structure (5.5 Å), which often does not provide canonical parameters for α-helices in the MSDs and therefore does not allow their automatic recognition. Figure 2b compares the global NBD dimeric organization (viewed from the MSDs) of the outward- and inward-facing conformations of CFTR and highlights the dimer contact points with the intracellular loops (ICLs). As for the MsbA closed-apo structure , the model of the inward-facing configuration of CFTR that we propose here is characterized, when compared to the outward-facing configuration, both by a twisting of the transmembrane helices (Fig. 2a) and by a sliding of the NBDs (Fig. 2b). Interestingly, the contact points of the ICLs with the NBDs (especially ICL2 and ICL4 with NBD2 and NBD1, respectively) appeared to remain unchanged (Fig. 2b), these contacts constituting stable anchors or pivots around which the conformational reorganization of the MSDs and NBDs can occur.
The experimental 3D structure of the closed-apo, inward-facing configuration of MsbA was observed in the absence of nucleotides . The MsbA-based model of the closed-apo configuration of the human CFTR NBD1/NBD2 assembly is shown on the right in Fig. 2b. In CFTR, in contrast to ABC transporters bearing only canonical ATP-binding sites, an ATP molecule may actually remain bound to the non-conventional binding site (NBD1) during many gating cycles and may thus be located in our closed-apo CFTR model at a possible junction between NBD1 (blue in Fig. 2b) and NBD2 (orange). Such a binding, through a half nucleotide-binding pocket (including the Walker A and Walker B motifs, but not the ABC signature sequence) is supported by the fact that ATP was observed bound to the isolated NBD1 crystal structure . Of note, in such a model (Fig. 2b right), the half nucleotide-binding pockets, formed by the Walker A/Walker B sides, are now in close contact. In contrast, the ABC signature motifs, which in the outward-facing conformation form the other sides of the nucleotide-binding sites, are here rejected outside the NBD heterodimer interface. In our Sav1866-based outward-facing conformation of CFTR, the large regulatory insertion (RI), located between strands β1 and β2 of the ABC-specific β sub-domain of NBD1 (green ribbon in Fig. 2b), was modeled using as template the isolated CFTR NBD1 experimental structure, with several adaptations toward a “semi-open” conformation in order to avoid steric clashes with NBD2 in the NBD1:NBD2 heterodimer structure (supplementary data 2 in ). In the present MsbA-based inward-facing configuration, it is likely that the RI has yet to move upward to a “completely open” conformation, in order to avoid steric hindrance problems with the NBD2 subunit in its novel position (Fig. 2b). Interestingly, this “completely open” RI conformation is likely to make new contacts with the NBD2 non-canonical ABC signature sequence (LSHGH). This potential interaction likely involves the RI region harboring the phosphorylated serine 422 residue, but the detailed atomic characteristics cannot be deduced here because of uncertainties of the model in this region. Nonetheless, this interaction might hinder the complete separation of the NBDs.
The transmembrane helices and the CFTR channel
Importantly, amino acid residues, which were previously shown to line the channel pore in our Sav1866-based outward-facing CFTR model , kept a similar orientation in the present inward-facing model (Fig. 3, and ESM: Fig. S3). Figure 3 indicates most of the critical residues localized in TM1 (K95, Q98, P99), TM6 (R334, K335, F337, T338, S341) and TM12 (T1134, T1137, N1138, S1141), as well as those recently highlighted in the TM11–TM12 loop (S1118, T1121, T1122, G1127, V1129, I1131, and I1132) . As already predicted by the outward-facing model, residues F337, T338, and S341, which play a crucial role in the chloride channel conductance [38, 39, 40], are located in a narrow region of the pore, likely constituting a constriction in the channel. Moreover, this constriction seems to be associated with a bending of the linear ion-conducting pathway, suggesting that it may participate in gating of the channel. R117 and R104 of the extracellular loop 1 (ECL1) may be exposed at the extracellular surface of the channel, at the entry of the pore (Fig. 3, and ESM: Fig. S3). They thus probably participate in the outer mouth, as R334 and K335 or S1118, T1122, and G1127 (TM11–TM12 loop). Altogether, these residues may thus constitute important determinants of anion-binding sites, as experimentally shown for R334  as well as for R104 and R117 . Two other basic residues occupy critical positions in the middle and at the intracellular end of the channel, respectively. K95 (located in the middle of the channel) is involved in the anion selectivity and permeation of the channel, and plays, together with Q98 and P99, a role in unitary conductance and macroscopic current properties [4, 43, 44]. R352, which is also involved in Cl− channel conductance [45, 46], is located at the cytoplasmic mouth of the pore.
Comparison of the features of the outward-facing and inward-facing conformations (Fig. 3) demonstrated the clear closure of the CFTR channel in the inward-facing conformation (see above), thus supporting the assumption that this conformer may indeed actually lead, at the end of the gating cycle, to the closed state of the channel. However, intermediate structures between this “tight” inward-facing conformation and the initially considered closed-apo structure (ESM: Fig. S1) may transitionally participate in the anion transport.
Finally, it should also be stressed here that an in-depth examination of the CFTR channel in its outward-facing conformation allowed us to highlight novel important structural features within the ICLs which may account for the opening of the channel at the cytoplasmic-side (in contrast to the 3D structures of bacterial exporters). This further supports the assumption that the outward-facing conformation does indeed correspond to the completely open state of the CFTR channel. Because of its speculative nature, this is discussed in detail in the “Discussion”.
3D model of the CFTR R domain
In contrast to the MSD:NBD assembly, which clearly belongs to the ABC family of exporters (even though sequence identities are often very low), no structural template has yet been described on which the structure of the R domain could be tentatively modeled. Many studies have considered that the regulatory R domain of CFTR (consisting of ~194 amino acids, from residue 646 to 843), which contains multiple phosphorylation sites, is intrinsically disordered [8, 47]. Among many other functions, the CFTR R domain has, from a structural point of view, to link the end of NBD1 (approximately amino acid 645) to the beginning of MSD2 (~residue 841). The shortest distance between these two extremities is ~76 and ~63 Å in the Sav1866-based (outward-facing) and MsbA-based (inward-facing) models, respectively. The segment linking NBD1 to MSD2 has, however, to get around the NBD2 globular domain, increasing thereby the minimal distance involved to ~100 Å. As reported in Table 1 in the ESM, proteins such as human MDR1 (P-gp) and human SUR1, which are related to human CFTR but do not have a CFTR-like regulatory R domain, have linkers whose length ranges from 58 to 93 amino acids. These relatively short linkers contain actually only few strong hydrophobic amino acids (V, I, L, F, M, Y, W) but many residues known to favor loops (P, G, D, N, S) . As a consequence, they mainly lack stable regular secondary structures and thus may well correspond to intrinsically disordered regions. This was indeed reported to be the case in the very recently published experimental structure of mouse P-gp (MDR , see “Discussion”). It should be stressed here that the CFTR R domain immediately appears quite different from those minimal linkers, as its sequence is clearly longer and globally contains more strong hydrophobic amino acids, often participating in hydrophobic clusters typical of stable regular secondary structures. Thus, it is probable that, in the full structural context of CFTR, the R domain may be, at least in part, relatively structured. Such a hypothesis is in agreement with recent simulations showing that the R domain may be more compact than had been anticipated .
The C-terminal part of the R domain: modeling strategy
The R domain joins the first and second halves of the CFTR protein, each made of a membrane-spanning domain (MSD) and a nucleotide-binding domain (NBD). These two halves likely occurred by duplication of an ancestral MSD–NBD module, as suggested for MDR . According to this duplication hypothesis, we suggest here that the sequences immediately preceding the MSDs [i.e., the CFTR N-terminal sequence (preceding MSD1) and the C-terminal part of the R domain (immediately preceding MSD2 and herein named the “R2” domain, residues ~775–840)] may also share a similar 3D arrangement. This hypothesis is supported by sequence similarities highlighted by hydrophobic cluster analysis (HCA) and by the good agreement observed between the secondary structures predicted for the two segments (ESM: Fig. S4). Moreover, these predictions are in good agreement with the secondary structure propensities of the R2 domain calculated on the basis of experimental NMR data  (ESM: Fig. S4). The R2 domain is characterized by 29% strong hydrophobic amino acids (V, I, L, M, F, Y, W) and by a ratio between hydrophobic and loop-forming residues of 1.14; such values are close to those generally found for well-structured domains (e.g., 30–32% and 1.35–1.39, respectively, for the CFTR NBDs). Given the absence of experimental 3D structures that could serve as templates, the R2 domain (~65 amino acids) was actually modeled ab initio, by taking into account the secondary structure propensities reported for the phosphorylated protein  and by masking the hydrophobic patches displayed on the surface of the NBD heterodimer in the outward-facing CFTR conformation with the hydrophobic faces of the ab initio modeled segments. According to our hypothesis of a relationship between the N-terminal sequence of CFTR and its R2 domain, both segments should be similarly structured, with the very N-terminus located in the vicinity of the CFTR C-terminal end, which follows NBD2.
The N-terminal part of the R domain: modeling strategy
On the other hand, we suggest here that the N-terminal part of the R domain (called herein the “R1” domain) may be structurally variable, but that a possible structure it may adopt might be related to the structure of regulatory domains located immediately after the NBDs in some bacterial ABC transporters (e.g., the maltose importer MalK  and the molybdate/tungstate ABC transporter ModC ), with which we identified potential similarities using HCA. These regulatory domains, which are located at the cytoplasmic bottom of the NBD homodimers, are known to be made of two similar, consecutive “TOBE” sub-domains (transport-associated OB, Pfam domain database PF03459, , classified in the SCOP structural classification database within the MOP (molybdate/tungstate binding protein) superfamily b40.6]. Such five-stranded TOBE sub-domains consist of four-stranded “Greek key” motifs for which the fourth strand of each sub-domain is locked by a fifth strand of the other sub-domain (ESM: Fig. S5). For clarity purposes, we herein call “R1a” and “R1b” the first and second “TOBE” sub-domains, respectively. The β16–β17 and β20 strands (TOBE labeling) in the R1b sub-domain appeared as the best markers for a possible link between the CFTR sequence and those of the regulatory domains of other ABC transporters (ESM: Fig. S6); this R1b sub-domain also displayed globally a better general alignment. On another hand, this R1b sub-domain almost perfectly matches the secondary structure propensities deduced from experimental NMR data for the phosphorylated form of the R domain  (ESM: Fig. S6). Also, according to the TOBE hypothesis, strand β20 within the R1b sub-domain locks strand β14, which corresponds to the fourth strand of the Greek-key motif of the R1a sub-domain. This R1b sub-domain strand β20 can readily be suggested in the CFTR sequence (ESM: Fig. S6). However, discrepancies appear to exist for the first part of the R1a sub-domain, in which the two small β12 and β13 strands may be substituted, according to the published secondary structure propensities, by an α-helix (between L662 and F669) (ESM: Fig. S6). These particular sequence features suggest that the N-terminal part of the R1a sub-domain might also be associated with a different fold and that a different location of the whole R1 domain relative to the whole MSD:NBD assembly might also be considered, as discussed below and in ESM: Fig. S7.
The R1 “globular-like” domain was thus tentatively modeled using as templates the structures of Glcv (pdb 1oxs)  and PH0194 (pdb 2d62) (Lokanath N. K. et al., unpublished data), the sequences of which are the most similar to the CFTR sequence. The template structures were respected, except for two regions (ESM: Fig. S6): (1) strands β12 and β13 (β-strands numbering according to the labeling of the templates) were replaced by an α-helix, and (2) the insertion present in human CFTR between amino acids 680 and 690 was also modeled as an α-helix, according to the secondary structure propensities of phosphorylated R . It is also worth noting that the upstream part of the R1 domain contains the so-called regulatory extension (RE) observed in the experimental structure of mouse NBD1 and found to be structured as an alpha helix packing against the NBD1 core. This RE, which contains phosphorylable serine residues (S660, S670), may in fact fold differently in the full length CFTR protein, as it may directly participate in the R1 core domain.
Finally, the sequence linking the R1 and R2 sub-domains of the CFTR R domain (R1–R2 linker, 28 amino acids) was also modeled ab initio, according to the secondary structure propensities reported for the phosphorylated protein  and to the same restraints as for the R2 domain. The short C-terminal extension of CFTR was also modeled ab initio. It includes a segment, PSDRVKLFP (1443–1451), which likely adopts an extended configuration (β propensity of the corresponding hydrophobic cluster = 1.93 vs α propensity 0.85 ), while another segment, extending from residue 1,462 to 1,480, may correspond to an α-helix. The extended region of the C-terminal end of CFTR may interact with this helix, as well as with the NBD1 helix following the Walker B motif (αcore3–4; ).
3D model of the R domain: results
The R1–R2 sub-domains were modeled on the Sav1866-based model of CFTR and found to fit to this structure. Interestingly, after we had built the MsbA-based (inward-facing conformation) model of CFTR reported herein, we observed that the proposed R1–R2 path was fully compatible with this new model and that there was no steric clash, except for a small displacement of part of the R2 sub-domain, shown in pink in Fig. 4. Of note, R1 and the first part of R2 (shown in light blue in Fig. 4) contact one side of NBD2 in the outward-facing model (mainly the NBD2 ABC-specific α sub-domain) and the other side in the inward-facing configuration (from Y1219 to 1445, and principally the α/β core).
Compatibility of the whole CFTR models with low resolution experimental data
Even more recently, during the revision of this manuscript, an article by Zhang et al. , following a previous work , reported the architecture of dimeric CFTR particles using cryo-electron microscopy (see Fig. 4 in  which describes the main features of the dimers). According to the dimensions of the Sav1866 structure fitted into the CFTR experimental density map, it appears that this dimeric architecture is closely related to that observed by Mio et al. already discussed above . Indeed, the sizes of the observed particles are similar as regards two of the three dimensions: 123 Å wide versus 120 Å and 106 Å deep versus 106 Å, with a packing of the fitted atomic model very similar to that reported in Fig. 5, as well as similar locations of the orifices. However, there is an obvious difference for the largest dimension along the CFTR pseudo two-fold axis (132 Å in  vs 162 Å in ). This difference may likely be explained by a different packing of the R region, which appeared located between the NBDs and MSDs near the ICLs region in the work by Zhang et al., whereas it was mainly located at the bottom of the particles in the report by Mio et al. Thus, these two reports described, for the first time, two possible states for the R domain, a feature which is not surprising for a region known as highly mobile. It is, however, noteworthy here that the location of the R domain may be dependent on the experimental conditions, rather than be biologically relevant. On the other hand, the mobility of the R domain is also consistent with an alternative folding of the R1a sub-domain, as proposed herein. Indeed, this region, which corresponds to the first TOBE-like domain and appears to be the less related to cognate TOBE domains (ESM: Fig. S6), might actually be unfolded and thereby cover the distance between the NBD1 end and the ICLs region, where the folded part of the R domain would then be located. As regards this hypothesis, it is worth noting that the yeast CFTR-related Yor1p protein possesses a R region, whose size is limited to that of the first TOBE domain of CFTR, and whose sequence texture is very similar to the CFTR first TOBE-like domain (ESM: Fig. S7). Taken altogether, these considerations support the assumption that the structural features of this part of the CFTR sequence may be highly versatile, being able to switch from relatively well-structured states to extended ones.
The good fit highlighted above between the aforementioned microscopy data and our CFTR models is further supported by the low resolution electron microscopy results obtained with negatively stained, 2D crystals of CFTR  (see Fig. 8 in , showing the three orthogonal views of CFTR, crystallized in the absence of nucleotides, in which was fitted the structure of the closed-apo conformation of VcMsbA, as solved at that time). Although incorrect, this nowadays obsolete VcMsbA structure has overall dimensions very similar to those of the corrected structure of the same VcMsbA protein reported afterwards , which was used here to model the inward-facing configuration of CFTR. Therefore, our model of the inward-facing, closed configuration of CFTR appears consistent with these data. Moreover, the shapes of the MSDs of the two CFTR states (with and without nucleotides) shown in Fig. 9 in  are in good agreement with the two models compared herein. In particular, the MSDs of the nucleotide-free crystal form (Fig. 9a in ) shows a hexagonal shape, with a central link between two holes. These features match our MsbA-based closed model of CFTR, in which the central link is constituted by helices 6 and 12, and the holes are located between helices 6 and 12 and helices 1 and 11 at one side, and helices 5 and 7 at the other side. The MSDs of the nucleotide-bound form of CFTR are characterized by a triangular shape with a large central hole (Fig. 9b in ), a feature which is also found in our Sav1866-based model (open state) of CFTR. Similar features have also been observed by Zhang and colleagues .
Comparison of the outward- and inward-facing conformations of ABC transporters has indicated that the ATP-bound state is generally associated with the most extensive interface between the two NBDs (head-to-tail configuration), whereas the structures of nucleotide-free transporters exhibit conformations characterized by a greater separation between the NBDs . In contrast to other ABC exporters which have symmetrical NBDs, an important particular feature of our human CFTR model is that, upon ATP hydrolysis, the sliding of the two CFTR NBDs along the heterodimer interface (which is coupled to the twisting motion of the MSDs) might be limited, and a complete separation of the NBDs may therefore be hindered, because of the presence of the CFTR-specific RI (green in Fig. 2) and that of an ATP molecule quasi-permanently bound at the non-conventional site (NBD1) and which likely maintains some contact at the NBD1:NBD2 interface. Indeed, by interacting in the inward-facing, closed-apo configuration with the modified LSHGH signature of NBD2 (ESM: Fig. S10), the RI might act as a “safety catch” limiting the NBD sliding motion and the associated twisting of the MSDs. This might constitute a mechanism slowing down the kinetics of the channel closing, in addition to that provided by the non-conventional poorly active ATP-binding site . The presence of ATP at the non-conventional binding site (Fig. 2b) for several minutes, a time during which many open-closed gating cycles occur , may favor the association of NBD1 and NBD2 in a situation similar to that observed for the closed-apo conformer shown in Fig. 2b, and may thus also probably influence the channel kinetics. Altogether, the general asymmetry of the CFTR NBDs may allow a longer duration of the open state of the channel and consequently permit the passage of more anions during each gating cycle. It is tempting to speculate that the R domain, at least its globular-like R1 region predicted herein, might act in a similar way, behaving like a brake for the overall movement. Moreover, as shown in ESM: Fig. S10, one can also observe that, during the transition from the inward-facing (closed state) to the outward-facing, ATP-hydrolizable form (open state), the NBD2 domain may rub against the R1 and R2 sub-domains. In addition, the NBD2-specific patches of basic residues [in (1) the linker insertion (H1197, K1199, K1200), (2) the NBD2 main core (R1239, R1385, R1386, K1389, K1429, R1434, R1438) and (3) the C-terminal extension (K1457, K1459, K1461, K1468)] appear well positioned to interact with the R1 and R2 phosphorylated serine residues in the outward-facing conformer (clearly rather than in the inward-facing one, ESM: Fig. S11). So, it is tempting to speculate that this putative phosphorylation-dependent interaction might favor the outward-facing form and thus the opening of the channel towards the extracellular milieu. Such a molecular mechanism may provide an explanation of the role of phosphorylation in the activation of CFTR (discussed in ). This hypothesis is supported by the conclusions of the studies (1) of Rich and colleagues , who suggested interactions of the R domain with NBD2, and (2) of Chappe and colleagues , who showed that phosphorylation of the R domain regulates the CFTR function by promoting its association with other CFTR domains, rather than by causing its dissociation from an inhibitory site. These considerations also appear to be supported by a recent review , highlighting the potential role of MSDs and substrates (also named allocrites) in driving conformational changes, rather than most directly ATP hydrolysis. Here, in the case of the CFTR channel, the size of small anions may not be sufficient for driving large conformational changes, which might thus be favored by additional elements such as the regulatory insertion (RI) and the R domain.
The rubbing of the R1/R2 sub-domains with NBD2 during inter-conversion also appears consistent with their content in strongly hydrophobic amino acids (ESM: Table 1) which is intermediate between fully disordered and canonical, folded globular domains. Indeed, the flexibility allowed thereby appears important for the fine-tuning of the concerted movements of the NBDs with the MSDs.
Finally, the present model of the inward-facing conformation of CFTR may allow some novel insights into the functioning of the ABC transporter CFTR as an anion channel. Indeed, it appears that the region around F337 (Fig. 3) may constitute a constriction of the channel, which, with the eventual help of the extracellular loops and their associated glycans, might form the structural basis of a gate allowing the channel to close. Conversely, our whole CFTR model does not provide an obvious structural basis for a second gate, the existence of which remains thus elusive. Accordingly, as already suggested by others , the cytoplasmic-side gate of CFTR may have become atrophied, or uncoupled from the outer gate, and the CFTR channel may thus function as a “broken” ABC transporter.
In conclusion, our results suggest that CFTR, as every ABC transporter, uses for its gating cycles a mechanism alternating between outward- and inward-facing conformations. However, our model of the outward-facing conformation also reveals the presence, at the level of the MSDs intracellular loops, of an aperture which is sufficiently large to allow the passage of chloride ions; this clearly supports the channel function of CFTR and the assumption that the outward-facing conformation corresponds to the open state of the CFTR channel. The transition from that conformer towards an inward-facing conformation finally leads to the closure of the channel. Indeed, the structural features described above for inward-facing conformation of CFTR, modeled on the “closed-apo” conformer of MsbA, which is characterized by a very small separation of the NBDs, should allow the closure of the channel. However, for CFTR to function as a more classic ABC protein transporting much larger substrates than small ions, it may be expected that it will have to reach an inward-facing configuration similar to the “open-apo” structure described for MsbA, with the two NBDs now fully separated. Such a configuration is otherwise observed for the inward-facing form of P-gp . Remarkably, the concerted movements between the transmembrane helices constituting the channel and the NBDs apparently occur around constant pivots corresponding to the transmission interfaces linking the ICLs coupling helices and the NBDs. Our results strongly emphasize the critical role of the residue F508 at these interfaces, not only for CFTR domain assembly but also for the signaling between the NBDs and MSDs. Finally, the models presented here are in good agreement with various experimental data, including those recently observed by electron microscopy, which support the predicted overall assembly of CFTR dimeric arrangement and the possible location of the R domain, partly below the NBD heterodimer. Moreover, the location of the N-terminal and C-terminal ends of CFTR, close to the R domain, is in good agreement with experimental data suggesting the existence of a physical interaction between them [63, 64]. Although some of the structural features of CFTR predicted herein as well as the above speculations about the CFTR functioning await experimental validation, our models provide helpful support for understanding of how CFTR functions and how mutations lead to dysfunctions, as well as for the rational design of drugs for a CFTR-specific pharmacological treatment of CF.
This work was supported by grants from the association "Vaincre La Mucoviscidose" (Paris, France).
- 14.Lewis HA, Buchanan SG, Burley SK, Conners K, Dickey M, Dorwart M, Fowler R, Gao X, Guggino WB, Hendrickson WA, Hunt J, Kearins MC, Lorimer D, Maloney PC, Post KW, Rajashankar KR, Rutter ME, Sauder JM, Shriver S, Thibodeau PH, Thomas PJ, Zhang M, Zhao X, Emtage S (2004) Structure of nucleotide-binding domain 1 of the cystic fibrosis transmembrane conductance regulator. EMBO J 23:282–293CrossRefPubMedGoogle Scholar
- 15.Lewis HA, Zhao X, Wang C, Sauder JM, Rooney I, Noland BW, Lorimer D, Kearins MC, Conners K, Condon B, Maloney PC, Guggino WB, Hunt JF, Emtage S (2005) Impact of the deltaF508 mutation in first nucleotide-binding domain of human cystic fibrosis transmembrane conductance regulator on domain folding and structure. J Biol Chem 280:1346–1353CrossRefPubMedGoogle Scholar
- 24.Seibert FS, Linsdell P, Loo TW, Hanrahan JW, Clarke DM, Riordan JR (1996) Disease-associated mutations in the fourth cytoplasmic loop of cystic fibrosis transmembrane conductance regulator compromise biosynthetic processing and chloride channel activity. J Biol Chem 271:15139–15145CrossRefPubMedGoogle Scholar
- 28.Mio K, Ogura T, Mio M, Shimizu H, Hwang T, Sato C, Sohma Y (2008) Three-dimensional reconstruction of human cystic fibrosis transmembrane conductance regulator chloride channel revealed an ellipsoidal structure with orifices beneath the putative transmembrane domain. J Biol Chem 283:30300–30310CrossRefPubMedGoogle Scholar
- 29.Zhang L, Aleksandrov LA, Zhao Z, Birtley JR, Riordan JR, Ford RC (2009) Architecture of the cystic fibrosis transmembrane conductance regulator protein and structural changes associated with phosphorylation and nucleotide binding. J Struct Biol (in press)Google Scholar
- 59.Kos V, Ford RC (2009) The ATP-binding cassette family: a structural perspective. Cell Mol Life Sci (in press)Google Scholar