Purification and Crystallization of the Cystic Fibrosis Transmembrane Conductance Regulator (CFTR)*

The cystic fibrosis transmembrane conductance regulator (CFTR) is a membrane protein that is mutated in patients suffering from cystic fibrosis. Here we report the purification and first crystallization of wild-type human CFTR. Functional characterization of the material showed it to be highly active. Electron crystallography of negatively stained two-dimensional crystals of CFTR has revealed the overall architecture of this channel for two different conformational states. These show a strong structural homology to two conformational states of another eukaryotic ATP-binding cassette transporter, P-glycoprotein. In contrast to P-glycoprotein, however, both conformational states can be observed in the presence of a nucleotide, which may be related to the role of CFTR as an ion channel rather than a transporter. The hypothesis that the two conformations could represent the “open” and “closed” states of the channel is considered.

The cystic fibrosis transmembrane conductance regulator (CFTR) 1 (product of the gene mutated in patients with cystic fibrosis) is a novel member of the ATP-binding cassette (ABC) superfamily that functions as an ion channel rather than a membrane transporter as most other members do (1). This chloride channel activity plays an important role in both secretion and reabsorption of ions and fluid at epithelial surfaces (2). More than 1000 different mutations in the CFTR gene have been identified in cystic fibrosis patients. Many of these impair different properties of the ion channel and its regulation, but the most common mutation, ⌬F508, results in misfolding and the failure of maturation (3). Electrophysiological and biochemical studies of wild-type and different mutant CFTRs have shed considerable light on its mechanism of action (1). The overall control of channel activity is exerted by the phosphorylation by protein kinase A of the central regulatory domain, which enables the control of gating by ATP interactions with the two dissimilar nucleotide binding domains (NBDs). Stable binding of ATP to NBD1 where little or no hydrolysis occurs together with binding and formation of the transition state for hydrolysis at NBD2 opens the channel (4,5), and relaxation of the strained structure possibly following dissociation of bound nucleotide allows for the channel closing (6). Allosteric interactions between the regulatory domain, the NBDs, and the channelforming portions of the protein are likely to underlie this sequence of events. However, before further insight can be gained into this mechanism, knowledge of the threedimensional structure is required.
Among eukaryotic ABC proteins in which all domains are contained within a single polypeptide, low resolution structures have been determined for P-glycoprotein (7-9) and MRP1 (10). Electron crystallography of P-glycoprotein has enabled the visualization of the major shape changes of the transmembrane domains (TMDs) as a consequence of nucleotide binding to the NBDs. High resolution structures by x-ray diffraction have not yet been determined for these or other whole eukaryotic ABC proteins but they have in the cases of two bacterial multisubunit transporters (11)(12)(13). Although potential analogies between domain interactions in the BtuCD vitamin B 12 importer (12) and in CFTR are intriguing and useful in the design of further biochemical experiments, it is essential that the structure of CFTR itself be determined.
Although a high resolution structure of the first nucleotide binding domain of mouse CFTR has recently been determined (14) progress toward a structure determination of the whole protein has been hindered by the lack of abundant sources for its purification and its limited solubility (15). No adequate natural source of the protein is known, and the levels of overexpression that can be achieved in mammalian cells (16) and yeast (17) are much lower than those of other proteins, at least partly because it is recognized and degraded by the endoplasmic reticulum quality control system (18). The highest levels of expression have been achieved in baculovirusinfected insect cells (19,20), but soluble preparations suitable for crystallization trials have not been obtained from this source.
We have now generated sufficient amounts of protein by heterologous mammalian cell expression for purification by conventional techniques and the formation of two-dimensional crystalline arrays in the presence of the high affinity ligand, AMP-PNP. Electron crystallography, yielding three-dimensional structural information to a resolution of ϳ2 nm indicates that the crystallized CFTR is monomeric, has overall structural similarities to P-glycoprotein, and can assume two different conformational states possibly reflecting differential nucleotide binding to the two NBDs.

EXPERIMENTAL PROCEDURES
Protein Purification, Nucleotide Binding, and Hydrolysis-Baby hamster kidney-21 cells expressing C terminally deca-histidine-tagged wild-type human CFTR were grown to confluence in roller bottles (850-cm 2 surface area) in 100 bottle lots yielding ϳ10 10 cells. Cells were washed, disrupted, and fractionated by conventional differential centrifugation steps (21) in the presence of protease inhibitors to obtain a crude microsomal membrane preparation containing most of the CFTR protein. Peripherally associated proteins were removed from the membranes by a brief incubation at pH 10.8 on ice (22). Solubilization of integral proteins was achieved with 1% n-dodecyl-␤-D-maltoside (DDM) in 20 mM Tris-HCl, pH 7.4, 0.5 M NaCl, 2 mM MgCl 2 containing 0.2% lipids (Escherichia coli crude lipids/phosphatidylcholine, 3:1), and 20% glycerol. The supernatant, after centrifugation for 1 h at 100,000 ϫ g, was mixed gently with Ni-NTA beads in the presence of 5 mM imidazole at 4°C. After washing extensively with 20 mM Tris-HCl, pH 6.8, 0.5 M NaCl, 2 mM MgCl 2 , 20% glycerol, 1% DDM, and 60 mM imidazole, His-tagged CFTR was eluted by increasing the imidazole concentration to 400 mM. The eluate was mixed gently with wheat germ lectin-agarose beads to enable the binding of the mature CFTR protein with exposed N-acetylglucosamine-containing binding sites for the lectin in the complex oligosaccharide chains. These beads were washed extensively with the same buffer but with reduction of the DDM concentration to 0.1% and addition of 1 mM dithiothreitol and a lipid mixture (phosphatidylethanolamine/phosphatidylcholine/phosphatidylserine/cholesterol, 5:3:1:1) to provide a lipid:protein ratio of ϳ100:1 on elution, which was with the same solution but also containing 0.5 M N-acetylglucosamine and 50% ethylene glycol. Finally the purified protein was reconstituted by dialysis (10-kDa cut-off membrane) for 48 h at 4°C against 20 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 2 mM MgCl 2 and adjusted to 10% glycerol.
The binding of AMP-PNP to CFTR in the same membranes used for its purification was assessed using the photoactive analogue, [␣-32 P]8-N 3 AMP-PNP (5). Radioactivity associated with specifically immunoprecipitated tryptic fragments containing either NBD1 or NBD2 of CFTR was determined by electronic autoradiography as described previously (23). The response of single CFTR chloride channels to AMP-PNP in the presence and absence of ATP was measured in planar lipid bilayers employing exactly the same procedures we have described elsewhere (24).
Crystallization and Electron Crystallography-Crystallization of CFTR used a method similar to that described for P-glycoprotein (8) and adapted from the method initially described in Refs. 25-27. Briefly, purified CFTR (0.1 mg/ml, 5 g) was added to a crystallization buffer (11.5% (w/v) polyethylene glycol 6000, 110 mM ammonium sulfate, 50 mM Tris-HCl buffer, pH 8.0, 0.1 mg/ml dodecylmaltoside, and 5 mM AMP-PNP). The protein solution was centrifuged at 10,000 ϫ g for 30 s and incubated (2-5 l) on a silanized glass slide against a reservoir containing 1.0 M MgCl 2 sealed with vacuum grease at 4°C. After ϳ20 h, the two-dimensional crystals were transferred to a glow-discharged (30 s in the presence of air) copper grid (400 mesh) coated with a continuous carbon film. These were immediately negatively stained with 2% (w/v) uranyl-acetate and viewed under the electron microscope. Untilted and tilted images were recorded under low dose conditions using Tecnai 10 and Philips CM100 transmission electron microscopes. Negatives were developed using full-strength D19 developer. Multilayered crystals were selected against by adjusting the protein concentration (25)(26)(27), and only single layered crystals (1-unit cell thick) were processed. Images were processed as described previously using the Medical Research Council crystallographic program suite (9). A procedure for determining the correct orientation in which each crystal is merged with the core data set, is described in Ref. 9. The three-dimensional Coulomb density maps were generated using the CCP4 software (28) and viewed using XFIT within the XTALVIEW software suite (29). Modeling of the MsbA homodimer and fitting of the model to the CFTR density map was carried out as detailed in Ref. 9.

CFTR Expression, Purification, and Hydrolysis of ATP-The
wild-type human CFTR cDNA with a deca-histidine C terminus was stably integrated into the genome of baby hamster kidney cells at a high copy number employing a plasmid con-taining a mutant dihydrofolate reductase gene enabling selection with a high concentration of methotrexate as we described originally (30). Membranes from these cells contain a larger amount of functional CFTR than other stable mammalian cells, for example ϳ5ϫ more than in Chinese hamster ovary cells (16,31), but still ϳ10ϫ less than the amount of the related MRP1 transporter in the same baby hamster kidney expression system (21). The effectiveness of each step in the purification protocol is demonstrated in Fig. 1. Substantial amounts of peripheral membrane proteins were removed from the membranes by alkaline pH treatment without a significant loss of CFTR. The remaining integral proteins including CFTR were solubilized in 1% DDM. Nearly all of the CFTR was bound to and eluted from the Ni-NTA matrix resulting in a high degree of purification. A subsequent WGA-agarose affinity step removed most remaining contaminating bands.
To determine whether the detergent-soluble purified protein was active, its ability to hydrolyze ATP was assayed. There was a linear relationship between the rate of hydrolysis and the amount of purified protein ( Fig. 2A), and Michaelis-Menten behavior was exhibited with a K m in the millimolar range with either ATP or GTP as the substrate (Fig. 2B). This low specificity among purine nucleoside triphosphates is common with other ABC proteins (32). The V max with ATP as the substrate was ϳ60 nmol/mg/min, which is quite similar to the V max with the protein purified from insect cells (33).
AMP-PNP Interactions with CFTR-The non-hydrolyzable ATP analogue, AMP-PNP, has been employed extensively in studies of CFTR channel activity (34). Because of its high affinity binding and locking of the CFTR channel, we reasoned that the non-hydrolyzable nucleotide might be a good ligand to employ in attempts to crystallize the protein, perhaps restricting its flexibility. Moreover, AMP-PNP has been employed successfully to increase the yield of two-dimensional crystals of another ABC protein, P-glycoprotein (8,9). AMP-PNP was found to greatly prolong the open state of the channel when it was present with ATP, and it is capable of opening the channel itself when present at high concentrations (4). Both of these effects are demonstrated in Fig. 3 using the CFTR present in membranes used for its purification. It is not known if these effects at different concentrations correspond to binding of AMP-PNP to different NBDs. However, N 3 AMP-PNP binds with a much higher affinity to NBD1 than NBD2, whereas N 3 ATP interacts with a similar affinity to both domains (23). Fig. 4 illustrates that N 3 AMP-PNP binds to NBD1 with a K d of ϳ10 M. As described below, two-dimensional crystalline arrays were obtained in the presence of 5 mM AMP-PNP, i.e. at a concentration where we might expect the NBD1 binding site to be saturated and the NBD2 site to be at least partially occupied (Fig. 4). Assessment of the Electron Crystallographic Data-Two-dimensional crystals of CFTR were grown using the hanging drop methodology described previously (8,(25)(26)(27). An initial assessment suggested that the crystalline order was high with strong peaks in the computed Fourier transform of images of crystals extending to the third order and ϳ20-Å resolution. A computed transform is displayed in Fig. 5 alongside a micrograph of a negatively stained CFTR crystal. Analysis of Ͼ100 separate crystals showed that there were two discrete crystal forms with significantly different unit cell parameters (Tables  I and II). One crystal form was ϳ3ϫ more frequent than the other, although we have calculated a three-dimensional structure for both crystal forms, which are presented here. Tables I   and II summarizes the electron crystallographic data collection. The number of observations/structure factor in the two data sets was ϳ4 -5 (i.e. a mean redundancy of 4 -5). The resolution limit for the data in the xy plane of the crystals was ϳ20 Å, as judged by the interimage phase residuals calculated for the untilted crystals. Resolution along z (perpendicular to the crystal plane) was estimated to be ϳ30 Å as judged by examination of the phase variation along lattice lines (Fig. 5d) as well as by a visual assessment of the features distributed along z in the three-dimensional maps (Fig. 6).
Three-dimensional Structure of CFTR- Fig. 6 shows images of three-dimensional density maps obtained from crystal forms 1 and 2, respectively. With both crystal forms, the CFTR mol- ecule in the unit cell is monomeric, in agreement with some other studies (35) and the P 1 plane group assignments and the unit cell sizes. Viewed from above the crystal plane (Fig. 6, top panels), CFTR is 6 -7 nm in diameter and displays a roughly hexagonal profile (crystal form 1) or an opened-out triangular profile (crystal form 2).
Crystal Form 1-When viewed from the side (parallel to the crystal plane), the density map is divided into two regions, a high density region (Fig. 6, HD) and a lower density region (LD). The high density region is composed of a ϳ6 -7-nm diameter cylinder of a depth of ϳ6 nm with a double-barreled central cavity with the two shafts separated by small central domains (Fig. 7, CD) that protrude from the walls of the barrel. This high density region also dominates the projection map for this crystal form (Fig. 5). The outer wall of the barrel is 1-2 nm wide and ϳ6 nm in length, and the barrel has a roughly hexagonal shape when viewed from the top (Fig. 6). A better impression of the paths of the two shafts through the high density region and the shape of the central domain is given in Fig. 7, A and B, where the front wall of the barrel is stripped away to reveal the inner structure. One shaft emerges at the top of the high density region, whereas the other shaft emerges toward the side of the barrel. The paths of both shafts join at the bottom into a common chamber. Interestingly, it has been reported that CFTR may be a double-barreled ion channel (36). The CFTR molecule elongates downwards into a more open lower density region (Fig. 6, LD) that extends a further ϳ6 nm and is composed of several ϳ2-nm diameter domains linked together (Fig. 6A, bottom). Linkage between the higher and lower density regions is via downward extensions from the barrel. Contacts between molecules in the crystal lattice are made in this lower density region; the high density regions do not make contact. The low density region will be less deeply embedded in heavy atom stain (being more distant from the grid support film) hence accounting for the lower mean density in this region. Such a differential staining effect has been reported previously (37,38). Comparisons with negatively stained crystals of P-glycoprotein (8,9) and cryo-electron crystallography of unstained P-glycoprotein crystals 2 suggest that the NBDs are likely to be located in the low density region and may be less well defined than the TMD region.
Crystal Form 2-The structure of CFTR in this crystal form shows a major conformational rearrangement compared with the first crystal form (Figs. 5e and 6B). The higher density region has a roughly triangular profile when viewed along the long molecular axis (Fig. 6B, top panel) with three major domains resolved. The central chamber now appears to be singlebarreled, because of a movement of the central domain, and a large gap in the density is apparent to one side (Fig. 6B, white  arrow). These changes are not attributable to resolution differences (see Tables I and II). An impression of the central pore can be gained in Fig. 7C. In this conformation, the pore is a 1.5-nm-wide channel that extends through the high density region without any noticeable constriction or capping. The impression gained from this conformational state is that it is a much more open structure than in crystal form 1. Perhaps significantly, in this crystal form, there is a more even distribution of density between the higher and lower density regions as a result of better stain penetration through the crystals. Contacts between CFTR molecules in this crystal form are intimate in both the high density region as well as in the low density region in contrast to the situation in crystal form 1.
Comparison with High Resolution Structures-The CFTR three-dimensional maps were compared with the high resolution structures for the bacterial ABC transporters, MsbA and BtuCD (11)(12)(13). No significant correspondence was noted between BtuCD and CFTR, but this was not unexpected as BtuCD has 2 ϫ 10 transmembrane spans and does not have connecting domains (ICDs) between the TMDs and NBDs. CFTR is predicted to have six transmembrane ␣-helices/transmembrane domain and has ICDs connecting the TMDs with the NBDs. A good correspondence could be found between the CFTR structure in crystal form 1 and the Vibrio cholera MsbA structure (13) but not to the E. coli MsbA structure (11). Fig. 8 shows a comparison between the V. cholera MsbA structure and the density map obtained for CFTR in crystal form 1. The V. cholera MsbA homodimer has 2 ϫ 6 transmembrane ␣-helices and ICDs and readily fits inside the CFTR molecular envelope with a good match to the overall size and shape of CFTR. This modeling places the transmembrane domains of MsbA within the high density region in the CFTR map. The CFTR envelope contains some small features that are not coincident with MsbA especially in the low density region. This is expected as CFTR contains additional sequence (such as the regulatory domain) and posttranslational modification (glycosylation of extracellular loop 4) that are not present in MsbA. The similarity between MsbA and CFTR in crystal form 1 implies that the latter could be a starting point for modeling CFTR using a similar approach to that described for P-glycoprotein (39). In contrast, CFTR in crystal form 2 could not be readily fitted to the MsbA structure. Although the overall dimensions of CFTR in crystal form 2 are similar to MsbA, the arrangement in the high density region is clearly different. The MsbA structure was obtained in the absence of nucleotide (see "Discussion").
DISCUSSION Three x-ray crystallography-derived structures for two bacterial ABC transporters have now been published (11)(12)(13) and the TMD structures for the transporters were revealed for the first time. There was no significant structural homology between the transmembrane domains of BtuCD and MsbA. Low (2-3 nm) resolution studies by electron microscopy and single particle analysis of a bacterial ABC transporter YvcC (40) have also been reported. Structural homology between the NBD Lattice parameters a ϭ 7.0 Ϯ 0.08 nm (6.6 Ϯ 0.06 nm) b ϭ 7.1 Ϯ 0.07 nm (   folds in the high resolution structures of the bacterial ABC transporters was apparent, and both correlated well with the folds determined for soluble NBDs (reviewed in Ref. 41) (42)(43)(44)(45). Recently, the structure of NBD1 of CFTR has been published (14), and it too shares a close overall structural homology with other ABC transporters.
In contrast to progress with bacterial ABC transporters, eukaryotic ABC transporters have remained relatively resistant to high resolution structural analysis, although the CFTR structure herein represents the fourth eukaryotic ABC trans-porter for which low resolution three-dimensional structural data have been obtained (8,46,47). Projection data (two-dimensional) for a fifth eukaryotic ABC transporter have also been obtained (8), and a projection map for the P-glycoprotein in two-dimensional crystals in lipid membranes was also published (48). Electron microscopy coupled with image analysis has revealed low resolution three-dimensional data for single particles (7) and two-dimensional crystals (8,9) of P-glycoprotein. In addition recent studies have led to low resolution three-dimensional structures from single particles of the transporter associated with antigen presentation (TAP) (46), a transporter of short peptides across the endoplasmic reticulum membrane, and Pdr5p, a yeast multidrug transporter (47). The CFTR three-dimensional structure presented here compares closely to P-glycoprotein (8,9). TAP has similar overall shape and dimensions to CFTR crystal form 1 and displays a central cavity region, but the cavity in TAP is single-barreled. This may be because the TAP structure is at significantly lower resolution (ϳ35 Å). In contrast, Pdr5p displays no central cavity within the ϳ160-kDa monomer, although in this study, a large internal cavity was observed that was created by the formation of a homodimer (47).
The three-dimensional structures of P-glycoprotein and CFTR are remarkably conserved, as shown in Fig. 9. Both proteins display at least two structures that are related by a major conformational rearrangement. Crystal form 1 of CFTR (Fig. 9A) corresponds closely to a crystal form of P-glycoprotein grown in the absence of nucleotide (Fig. 9C and Ref. 9). Crystal form 2 of CFTR (Fig. 9B) corresponds closely to P-glycoprotein crystals grown in the presence of nucleotide AMP-PNP ( Fig. 9D and Ref. 9). A double-barreled central cavity with a central domain is a common characteristic in Fig. 9, A and C. The main difference between the CFTR and P-glycoprotein maps in this crystal form is that in the latter, the central cavity is more occluded at the bottom, and in the former, the barrel has a more rounded profile. In the alternative conformation (Fig. 9, B and  D), the three-lobed triangular configuration is evident, with a gap in the density on one side (red arrows). The main difference here is that the CFTR structure appears to have a significantly wider gap. In Fig. 9, the white arrows indicate density that extends upwards into the putative extracellular side of the molecule in both maps but at different locations with respect to the gap on one side. The significance of this difference between CFTR and P-glycoprotein is not yet clear, although CFTR and P-glycoprotein do differ on the extracellular face especially in terms of the major glycosylation site, which is in extracellular loop 1 in TMD1 in P-glycoprotein but in loop 4 in TMD2 in CFTR (49).
One might speculate that the two conformations of CFTR observed relate to the quiet and actively gating states of the channel, which we have postulated exist prior to and after nucleotide binding, respectively (24). So far it is not possible to confirm this nor to predict which, if any, form is active based on structural information alone. For P-glycoprotein, the cloverleaf form is only observed in the presence of nucleotide (AMP-PNP), whereas the barrel form is restricted to the nucleotide-free condition. Because nucleotide binding is associated with the prolongation of channel opening in CFTR then by crude extrapolation from P-glycoprotein, the cloverleaf form might be predicted as the more likely active state of the channel. Clearly, further investigations are required to test this prediction. Alternatively, the two conformational states may represent differential nucleotide binding to the two NBDs that have different affinities for the nucleotide (23). This could explain why crystal form 1 but not crystal form 2 is comparable with the MsbA structure that was obtained in the absence of nucleotide ( Fig. 8 and Ref. 13). A wider study of AMP-PNP concentrations during crystallization could be used to test this idea.
There has been some debate about the quaternary structure of CFTR with evidence both for and against a dimeric organization (35, 50 -52). The structural data provided in this article show a monomeric organization. This adds extra weight to the idea that CFTR may exist as a monomer in vivo, although the possibility that purification and crystallization affects CFTR oligomeric behavior cannot be excluded, and a very recent report provides evidence of CFTR dimers in the plasma membrane (53). The purified CFTR monomers we observed have been shown to be enzymatically active and sufficiently structurally homogeneous/rigid to allow crystallization into two well ordered crystal forms. These data do not preclude the possibility of oligomer formation in vivo particularly if the latter was mediated by intercalating proteins that do not co-purify.