Nucleotide-binding Domains of Cystic Fibrosis Transmembrane Conductance Regulator, an ABC Transporter, Catalyze Adenylate Kinase Activity but Not ATP Hydrolysis*

The cystic fibrosis transmembrane conductance regulator (CFTR) is an anion channel in the ATP-binding cassette (ABC) transporter family. CFTR consists of two transmembrane domains, two nucleotide-binding domains (NBD1 and NBD2), and a regulatory domain. Previous biochemical reports suggest NBD1 is a site of stable nucleotide interaction with low ATPase activity, whereas NBD2 is the site of active ATP hydrolysis. It has also been reported that NBD2 additionally possessed adenylate kinase (AK) activity. Knowledge about the intrinsic biochemical activities of the NBDs is essential to understanding the Cl– ion gating mechanism. We find that purified mouse NBD1, human NBD1, and human NBD2 function as adenylate kinases but not as ATPases. AK activity is strictly dependent on the addition of the adenosine monophosphate (AMP) substrate. No liberation of [33P]phosphate is observed from the γ-33P-labeled ATP substrate in the presence or absence of AMP. AK activity is intrinsic to both human NBDs, as the Walker A box lysine mutations abolish this activity. At low protein concentration, the NBDs display an initial slower nonlinear phase in AK activity, suggesting that the activity results from homodimerization. Interestingly, the G551D gating mutation has an exaggerated nonlinear phase compared with the wild type and may indicate this mutation affects the ability of NBD1 to dimerize. hNBD1 and hNBD2 mixing experiments resulted in an 8–57-fold synergistic enhancement in AK activity suggesting heterodimer formation, which supports a common theme in ABC transporter models. A CFTR gating mechanism model based on adenylate kinase activity is proposed.

Cystic fibrosis transmembrane conductance regulator (CFTR) 2 is an ATP-binding cassette (ABC) transporter that functions as a chloride channel. The ABC transporter superfamily has a large number of functionally diverse transmembrane proteins that transport a variety of substrates (from ions to proteins) within, into, and out of the cell (1). The typical ABC transporter consists of two transmembrane domains and two cytoplasmic nucleotide-binding domains (NBDs). The NBDs are thought to utilize energy from the hydrolysis of ATP to transport substrates across the membrane. Some ABC transporters function actively against a concentration gradient, and others are passive transporters (e.g. CFTR). In the human genome sequencing efforts, 51 ABC transporter genes have been identified, 13 of which have been linked to genetic diseases (2). The most common and intensely studied of these diseases is cystic fibrosis.
Primary amino acid sequence analysis suggests CFTR is composed of homologous halves. Each half contains six membrane-spanning segments and an NBD. The two CFTR halves are linked by a cytoplasmic regulatory domain (R-domain) which is unique and not shared with other ABC transporters. The R-domain contains a number of protein kinase A (PKA) phosphorylation sites (3). PKA phosphorylation at multiple sites is a prerequisite for the activation of CFTR chloride channel by ATP (4). To date, over 1300 disease-causing mutations have been identified in the CFTR gene (www.genet.sickkids.on.ca/cftr/). These mutations are located throughout the coding sequence and include deletions and missense and nonsense mutations. Most mutations are rare in the cystic fibrosis (CF) population. The exception is the deletion of a phenylalanine residue at position 508 (⌬F508) that accounts for about 70% of CF mutations worldwide (5). The ⌬F508 mutation is located in NBD1 and results in a CFTR protein that has a dual defect. First, the protein is improperly processed from the endoplasmic reticulum to the cell plasma membrane. Second, the mutant protein that reaches the membrane is reported to have only ϳ30% of wild-type channel function (5). In contrast, the second most common CF mutation G551D is only defective in channel function (6) and has a prevalence of 2-7% in the CF population (7).
Schemes of ATP binding and hydrolysis by CFTR have been proposed to govern channel gating responses (8). In CFTR, the NBDs are thought not to be functionally equivalent because of the lack of sequence conservation in three of the ABC transporter signature motifs (9). Similarly, biochemical reports suggest that NBD1 is a site of stable nucleotide interaction (10), whereas NBD2 has been reported to be the site of ATP hydrolysis activity (11). In contrast, other biochemical studies where the NBD1 construct includes the regulatory (R) domain (NBD1-R) suggest that the protein had the ability to function as an ATPase (12,13). Most interesting were biochemical reports by Randak et al. (14,15) that indicated the NBD2 possessed both adenylate kinase as well as ATPase activity. Adenylate kinases catalyze the phosphotransfer reaction (ATP ϩ AMP u ADP ϩ ADP). In experiments using patchclamp analysis together with the Walker A lysine mutations in the NBDs, Randak and Welsh (16) concluded that NBD2, not NBD1, was responsible for adenylate kinase activity. The structural similarity of the CFTR NBDs to the adenylate kinases was reported (17) soon after the CFTR gene was cloned in 1989. However, this observation was never investigated experimentally until 1997 (14). The crystal structure of mouse NBD1 (mNBD1) was recently determined (18). Comparing the mNBD1 and the adenylate kinase structures, we saw resemblances that raised the possibility that both NBDs may have adenylate kinase activity.
To gain insight into the CFTR functional mechanism (Cl Ϫ ion gating) we have pursued a detailed biochemical examination of highly purified soluble and refolded cytoplasmic domains (i.e. NBD1, NBD1-R, and NBD2). Knowledge about the individual domains and their biochemical activities is essential for the proper understanding of the CFTR gating mechanism. Insight into the CFTR gating mechanism would be extremely useful in designing functional assays to both find and understand how small molecules function to restore CFTR channel activity.
In this report, we provide evidence that the mouse NBD1, human NBD1, and human NBD2 all have the ability to function as an adenylate kinase (AK). Careful separation of all nucleotide products by thin layer chromatography (TLC) reveals that, in contrast to extant reports in the CFTR literature, the isolated CFTR NBDs do not catalyze ATP hydrolysis because no liberation of [ 33 P]phosphate is observed from ␥-33 Plabeled ATP substrate. Although we have detected a nucleotide diphosphatase activity (EC 3.6.1.8), which affinity-purifies with the NBDs, this can be separated by a subsequent gel filtration chromatography step. Substitution of the Walker A box lysine for alanine in the hNBDs results in the loss of AK activity, which demonstrates that the activity is intrinsic to these proteins. We further demonstrate that mNBD1 protein can be specifically labeled with a related AMP substrate analog 8-azido-[ 32 P]AMP. Each NBD at low protein concentration displayed an initial slower nonlinear phase in AK activity, which suggests that homodimerization may be important for activity. The G551D gating mutation has an exaggerated nonlinear phase as compared with wild type, and may indicate this mutation affects NBD1 ability to form dimers. In hNBD2 immunoprecipitation experiments, hNBD1 co-precipitates suggesting a physical association between the two NBDs. Mixing experiments with hNBD1 and hNBD2 at low protein concentration resulted in a significant synergistic enhancement in AK activity suggesting heterodimer formation. From our biochemical findings we therefore propose a CFTR gating mechanism based on AK activity.

EXPERIMENTAL PROCEDURES
Cloning of the Mouse and Human CFTR NBDs-The NBDs were amplified by PCR from mouse and human cDNA libraries using Pfu/ Taq polymerases (Stratagene). Oligonucleotide primers complementary to the 5Ј-and 3Ј-ends of the open reading frames were designed to introduce unique restriction sites. The PCR products were digested with appropriate restriction endonuclease and ligated to an Escherichiacoli expression vector pET28b (Novagen). The resulting human NBD plasmids, pET28b-hNBD1 and pET28b-hNBD2, produced both human NBDs with an N-terminal hexahistidine (His 6 ) tag hNBD1 (Thr 390 -Leu 671 ) and hNBD2 (Lys 1200 -Leu 1480 ). The mouse NBD1 fragment (Thr 389 -Asp 673 ) was fused to the C terminus of the His 6 -tagged Smt3 fusion protein as described (19) and was identical to the construct that yielded the NBD1 crystal structure (18). The mouse and human CFTR gene fragments were sequenced to ensure their integrity and were identical to the GenBank TM sequence, accession numbers (NM_021050) and (NM_000492), respectively. All mutant clones were sequenced to confirm the presence of the mutations and the absence of unwanted mutations.
Expression, Purification, and Refolding of the CFTR NBD Proteins-Mouse and human CFTR NBD clones were expressed in E. coli using BL21 DE3 pLysS cells induced with 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside at 25°C. Soluble mouse NBD1 was initially purified by nickel ion affinity chromatography, followed by a Sepharose S200 (16/60 mm column) sizing chromatography step; the Smt3 tag then was removed using the Ulp1 protease (19). Subsequently, a second nickel affinity step was employed to separate the tag from the cleaved mNBD1 protein.
The human NBD proteins were found in the insoluble fraction. The cell pellets (30 g) were resuspended in 600 ml of buffer 1 (25 mM Tris-HCl, pH 7.5, 300 mM NaCl, 0.1% Tween-20, and 10 mM 2-mercaptoethanol) followed by the addition of protease inhibitors (E-64, pepstatin, and leupeptin at 2 g/ml, DFP at 50 M). The cell pellets were douncehomogenized and incubated for 1 h at 4°C. Samples were spun at 54,000 ϫ g for 1 h at 4°C. Supernatants were removed, and the pellets were resuspended in 600 ml of buffer 2 (25 mM Tris-HCl, pH 7.5, 6 M guanidine hydrochloride, and 10 mM 2-mercaptoethanol) overnight on a magnetic stirrer at 4°C. Samples were spun at 54,000 ϫ g for 1 h at 4°C. To the supernatants, 30 ml of Nickel Hi-Selects resin (Sigma) were added and incubated overnight at 4°C. Samples were centrifuged to pellet resin and resuspended in buffer 2 and poured into a column. The packed column was washed with buffer 3 (25 mM Tris-HCl, pH 7.5, 8 M urea, and 10 mM 2-mercaptoethanol) containing sequentially increasing amounts of imidazole. Fractions were analyzed using SDS-PAGE and Coomassie Blue staining. Fractions containing protein were concentrated by Centricon (Amicon) and loaded onto Superdex S-200 16/60 mm column. Proteins were eluted in buffer 4 (25 mM Tris-HCl, pH 7.5, 8 M urea, 150 mM NaCl, and 10 mM 2-mercaptoethanol). Protein containing fractions were concentrated to 1 mg/ml. Protein concentrations were determined from A 280 using calculated extinction coefficients (20). The initial "in house" NDSB-256 (dimethyl benzyl ammonium propane sulfonate) (Anatrace Inc.) refolding buffer was (25 mM Tris-HCl, pH 7.0, 1 mM EDTA, 10 mM ␣-cyclodextrin, 0.1 mM GSH/0.01 mM GSSG, and 0.5 M NDSB-256). The refolding screen for the denatured proteins has been described (21). Larger batches (10 mg) of purified proteins were refolded (hNBD1 100 g/ml, hNBD2 50 g/ml) in either condition 10 (21), condition 10 plus 0.5 M NDSB-256 or an NDSB-256 buffer from above. After 24 h incubation at 4°C, samples were concentrated, filtered and sized on a Superdex S-200 10/30 mm column. Sizing fractions that yielded monomeric/dimeric NBD proteins were pooled. Protein concentrations of the refolded samples were determined by quantitative Westerns using specific monoclonal antibodies (hNBD1 clone L12B4) and (hNBD2 clone M3A7) (Chemicon Inc.) and known quantities of denatured hNBD proteins as a standard curve. Western detection was accomplished using infrared fluorescence-labeled secondary antibody and the Odyssey system (LI-COR Bioscience) (22).
Determination of Kinetic Parameters and Data Analysis-The apparent K m and V max values for ATP, AMP, and ADP were determined by fitting the rectangular hyperbolic plot of linear rates of product (M/ min) versus increasing substrate concentration (M) to the Michaelis-Menten equation. IC 50 values for the ATP competitive inhibitor P1-(5Јadenosyl)P5-(5Ј-adenosyl) pentaphosphate (AP 5 A) (Sigma), AP 5 G, AP 4 A, AP 6 A (Jena Bioscience), were obtained by measuring the decrease in linear rates of adenylate kinase activity (M/min) at increasing concentrations of inhibitor and fitting the data using a two parameter hyperbolic fit: y ϭ IC 50 /(B ϫ (X ϩ IC 50 )), where B is proportional to 1/V max , y is the rate of inhibition (M/min), and X is the concentration of the inhibitor. For all IC 50 determinations, unless otherwise stated the substrate concentrations were at or within 2-fold of apparent K m , and the enzyme concentration was less than or equal to 2ϫ the reported IC 50 .

RESULTS
Expression and Purification of CFTR NBD Proteins-To understand the CFTR gating mechanism we have pursued a detailed biochemical examination of highly purified soluble and refolded cytoplasmic domains (i.e. NBD1 and NBD2) of the CFTR protein. The CFTR literature has demonstrated that the CFTR domains are extremely challenging to study because of their poor levels of expression and insolubility in most recombinant protein expression systems. The mNBD1, being the most tractable, was pursued as described by others (18) with modification to the protein purification procedures (see "Experimental Procedures"). The mNBD1 proteins (wild type and G551D) were purified to homogeneity as judged by a Coomassie-stained SDS-PAGE gel (Fig.  1A). A total of nine mNBD1 wild type and four mNBD1 G551D independent protein samples were purified for this study. Western blotting confirmed the protein identity as it cross-reacted with the human NBD1 monoclonal antibody (clone L12B4, Chemicon) (Fig. 1B). The identity of the mNBD1 proteins was also confirmed by matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) and N-terminal sequencing (not shown).
The six histidine-tagged versions of hNBD1 and hNBD2 were overexpressed as insoluble proteins and purified under denaturing conditions to homogeneity (Fig. 1B). Four independently denatured wild-type hNBD1 and four hNBD2 samples were purified for this study. To obtain active, soluble hNBD proteins, we worked out conditions to refold these proteins. We initially pursued three methods of protein refolding; the first two have been described (23,24) and the third was developed inhouse (see "Experimental Procedures"). Multiple attempts of refolding both hNBD1 and hNBD2 using the bovine serum albumin (BSA) method described by Wang et al. (23) were unsuccessful. BSA removal after the refolding resulted in protein that reformed soluble aggregates and lacked adenylate kinase or ATPase activity. We conclude that BSA is not an effective chemical chaperone and instead acts as a dispersant. The other two methods relied on 0.5 M arginine (24) or NDSB-256 as chemical chaperones. For hNBD1, the refolding method employing 0.5 M arginine or NDSB-256 yielded active protein. Because active hNBD2 protein was obtained with the NDSB-256 procedure, the arginine method was not pursued. For both these methods, NBD protein recovery yields were Ͻ1%.
In an effort to increase protein recovery yields we used a fractional factorial refolding screen (21). Each protein was refolded at a 400-l scale and after a 24-h, 4°C incubation period, each screening condition was directly assayed for activity. For the hNBD1 protein, the best refolding conditions were condition 10 (50 mM Tris, pH 8.2, 0.5 mM Tween-80, 5 mM TCEP, 550 mM GdnHCl, and 2 mM MgCl 2 ) and condition 30 (50 mM MES pH 6.5, 0.5 mM Tween-80, 5 mM TCEP, polyethylene glycol 3350, 440 mM sucrose) with the hNBD1 protein screened at 100 g/ml (not shown). Tween-80 and TCEP additives appeared important in hNBD1 refolding as these two components were shared between the two conditions. For hNBD2 protein (screened at 50 g/ml), the two best conditions were also conditions 10 and 30. Whereas condition 10 for hNBD2 appeared to be superior (in terms of activity) to the NDSB-256 condition, the overall protein recovery yields did not improve dramatically. Large batches (10 mg) of hNBD1 and 2 were refolded three ways: condition 10, condition 10 plus 0.5 M NDSB-256 and the "in-house" NDSB-256 buffer. After the incubation period, samples were concentrated and sized on a 10/30 Superdex S200 column to remove aggregated (unfolded) material. Fractions corresponding to monomeric material were assayed and pooled. Protein concentrations of the pooled fractions were determined by quantitative Westerns using A 280 determined amounts (20) of the denatured NBD proteins for a standard curve. Each pooled fraction that was used for further study contained a known amount of hNBD1 ( Fig. 1D) or hNBD2 (Fig. 1, E and F). During the purification and by Western analysis we saw that hNBD1 and 2 proteins had the propensity to form homodimers even under SDS-PAGE conditions. This was most apparent with hNBD2 ( Fig. 1F) as higher concentrations of refolded protein were more readily obtained. With developed procedures in hand to purify mNBD1 protein and, having refolded the hNBD1 and 2 proteins we proceeded to characterize their biochemical activities.

FIGURE 1. SDS-PAGE and Western analysis of mouse and human CFTR NBD proteins.
A, Coomassie-stained SDS-PAGE of native mouse NBD1 (mNBD1) wild type and G551D mutant proteins (2 g of protein). B, Coomassie-stained SDS-PAGE of denatured human NBD1 (hNBD1), hNBD2, and hNBD1-R proteins after gel filtration. For each sample, 2 g of protein was loaded. C, wild type and mutant native mNBD1 samples (5 ng) were Western blotted and probed using the hNBD1-specific antibody (clone L12B4) at 1 g/ml. D, identification of the refolded and resized hNBD1 proteins by Western analysis using the hNBD1 specific monoclonal antibody (clone L12B4) at 1 g/ml. For each refolding condition (condition 10 (left) or condition 10 plus 0.5 M NDSB-256 (right)) 1.9 ng of protein sample was loaded. E, identification of the refolded and resized hNBD2 proteins by Western analysis using the hNBD2-specific monoclonal antibody (clone M3A7) at 1 g/ml. For refolding condition NDSB-256 (left side) 10 ng of protein was loaded. For refolding condition 10 (right side), 20 ng of protein was loaded. F, Western blot of refolded and resized hNBD2 protein sample containing 240 ng of protein from condition 10 showing a prominent dimeric band at ϳ70 kDa.

The CFTR NBDs Function as Adenylate Kinases and Not as ATPases-
Previous biochemical reports on the CFTR NBDs have suggested low catalytic rates (1-10 turnovers per minute) and the possibility of multiple enzymatic activities residing in a single domain (14). These observations indicated a need for a biochemical assay system that is highly sensitive to and capable of discriminating between multiple enzymatic activities (e.g. adenylate kinase and ATPase activity). We chose a PEI-cellulose TLC sys-tem and a radiolabeled [␥-33 P]phosphate ATP substrate because of its sensitivity and ability to follow reactants and products directly. The TLC system also had a solvent (0.75 M LiCl and 1 M formic acid) that is able to separate multiple reaction products that can result from enzyme action on an ATP substrate (ADP, AMP, P i , and pyrophosphate (PP i )). The importance of a sensitive and discriminatory assay system became obvious as we evaluated CFTR NBD affinity-purified samples. In the presence of 33 P ␥-labeled ATP and cold AMP substrate, all mNBD1 samples purified through a His tag affinity column displayed two activities (not shown): the formation of labeled ADP (i.e. adenylate kinase activity) and the formation of labeled pyrophosphate (i.e. ATP diphosphatase activity (EC 3.6.1.8)). The other product of the latter reaction would be unlabeled AMP from the 33 P ␥-labeled ATP. In the presence of 33 P ␥-labeled ATP, the affinity-purified NBDs displayed only ATP diphosphatase activity (not shown) as evidenced by the formation of [ 33 P]pyrophosphate. The ADP formation was only detected in the presence of AMP, consistent with adenylate kinase activity. The observation that no labeled-free phosphate was detected above background (not shown) suggests that mNBD1 cannot hydrolyze ATP.
The mNBD1 samples were further purified through a subsequent Sepharose S200 sizing chromatography step, which removed ATP diphosphatase activity and demonstrated that the activity was not intrinsic to the mNBD1 protein (i.e. contaminant). The adenylate kinase activity; however, associated with the mNBD1 (Fig. 2A) and was dependent on added AMP (Fig. 2B) and the divalent cation Mg ϩ2 (not shown). All purified mNBD1 samples displayed adenylate kinase and not ATPase activity. GMP could not be substituted for the AMP substrate (not shown), indicating that the mNBD1 ADP-forming activity had the same specificity as other adenylate kinases. Bi-substrate inhibitors, diadenosine polyphosphates (Ap n A), have been shown to be potent inhibitors of adenylate kinases (25,26). We found that the mNBD1 adenylate kinase activity was inhibited by Ap 5 A with IC 50 ϭ 70 nMϮ 4 nM (ATP and AMP substrates were at K m ) (Fig. 2C). From this IC 50 data and linear regression, we estimate the percent active mNBD1 molecules to be between 20 and 80%. A more potent inhibitor is needed for a more accurate active site determination. At 500 M ATP/400 M AMP, the IC 50 increased to 500 nMϮ 30 nM (not shown) indicating a competitive inhibition mechanism. Shorter and longer Ap n A analogs were less effective in inhibiting the mNBD1 activity (Ap 4 A IC 50 ϭ 66 (not shown). Similarly, Ap 5 G was also a significantly poorer inhibitor (IC 50 ϭ 50 M) consistent with the observation that GMP could not substitute for AMP as a substrate (not shown).
A mNBD1 time course showed that AK activity was linear and stable at 37°C for up to 2.5 h (Fig. 2D). At 900 nM enzyme, the kinetic parameters of mNBD1 AK activity were determined (Table 1); at saturating ATP the apparent AMP K m was 30 M and V max was 0.9 M/min whereas, at saturating AMP (400 M) the apparent ATP K m was 30 M and V max was 1.2 M/min. At concentrations above 500 M AMP, the mNBD1 AK activity clearly demonstrated substrate inhibition (not shown). No substrate inhibition was detected up to 5 mM ATP.
For the refolded human NBDs, the AK activity associated with both the hNBD1 and hNBD2 ( Fig. 2A) and was dependent on added AMP ( Fig. 2A). All purified hNBD1 and hNBD2 samples displayed AK and not ATPase activity. For the hNBDs, GMP could not be substituted for the AMP substrate (not shown). In the presence of [␥-33 P]ATP substrate ATP hydrolysis has not been detected (Fig. 2B) in any of our hNBD1 and hNBD2 preparations, demonstrating that the purified CFTR NBDs catalyze only AK activity. An alternative possibility is that the refolding of hNBDs results in a loss of ATPase while retaining AK activity. This scenario, however, requires the retention of a more complex reaction such as phosphoryltransfer to AMP, whereas the simpler hydrolysis reaction is lost. The hNBD1 and hNBD2 time courses showed that AK activity was linear and stable at 37°C for 3 h (Fig. 2D). The kinetic parameters of the hNBDs adenylate kinase activity (Table 1) were determined using 4.7 nM hNBD1 and 30 nM hNBD2; both proteins were refolded using condition 10. For the hNBD1 protein the apparent AMP K m was 60 M and V max was 0.7 M/min, whereas, at 400 M AMP the apparent ATP K m was 80 M and V max was 0.7 M/min. For the hNBD2 protein the AMP K m ϭ 40 M and V max ϭ 0.7 M/min, whereas, ATP K m ϭ 70 M and V max ϭ 0.6 M/min. We also found the hNBDs AK activity was inhibited by Ap 5 A (hNBD1 IC 50 ϭ 300 nMϮ 30 nM and  hNBD2 IC 50 ϭ 160 nMϮ 6 nM (Fig. 2C)) when ATP and AMP substrates were fixed at 80 M. NBD Active Site Mutations and AK Substrate-labeling Studies Provide Further Evidence That the CFTR NBD Catalyzes AK Activity-In an effort to provide additional evidence that the CFTR NBD catalyzes AK activity, we generated a lysine to alanine substitution in the Walker A motifs of both hNBD constructs (hNBD1:K464A and hNBD2:K1250A). The conserved lysines were chosen because they hydrogen bond with the phosphates of the ATP substrate. We expressed and purified these mutant proteins in an identical manner to the wild-type samples. Denatured wild-type and mutant samples were refolded simultaneously at a 400-l scale using same condition for both. After the 24 h, 4°C incubation period samples were assayed directly for AK activity. The wild-type NBD proteins possessed AK activity, whereas the mutant proteins lacked AK activity (Fig. 3) demonstrating that the AK activity is intrinsic to the hNBD1 and -2 and not a contaminant activity that co-purifies with the NBD proteins. These proteins were each independently refolded three times, and each time the wild-type protein possessed activity whereas, the mutant proteins did not. The hNBD2 (K1250A) mutant result is in agreement with the results reported by Randak and Welsh (16).
In addition to the active site mutant studies, we were interested in providing additional evidence that the AK activity was intrinsic to the purified proteins. Another approach is to show that the AK-specific substrate AMP actually binds/labels the purified NBD proteins. To accomplish this, we exposed the purified mNBD1 protein (60 nM) to photoactivated 32 P-labeled 8-azido-AMP (7.5 M) (ALT Inc. Lexington, KY) in the absence of ATP (no substrate conversion to ADP occurs) and resolved the protein samples on a SDS-PAGE gel to show that the mNBD1 protein became radiolabeled (Fig. 4, A and B). When AMP competitive substrates (Ap 5 A, AMP, and ATP) were added, the 8-azido-AMP labeling efficiency decreased, showing that the labeling is specific for the nucleotide-binding site(s) of the mNBD1 (Fig. 4, A and B, lanes 2,  3, 5, 6). In contrast, when 400 M GMP were added, no decrease in labeling was noted (Fig. 4, A and B, lane 4). This 8-azido-AMP labeling experiment was repeated twice at these ligand concentrations with a similar outcome each time. We also exposed the purified mNBD1 protein (60 nM) to photoactivated 32 P-labeled 8-azido-ATP (7.5 M) (Fig. 4, C and D) in the absence of AMP to discern whether the competitive substrate profile was different from 8-azido-AMP (Fig. 4, A and B). The 8-azido-ATP labeling experiment was repeated twice at these ligand concentrations with a similar outcome both times. We found that AMP substrate did not compete effectively with the labeling by 8-azido-ATP (Fig. 4C, lane 3), suggesting possible separate binding sites for both AMP and ATP. We noted that ATP was competitive with the azido-AMP substrate but that this observation could be explained by noting that in other adenylate kinase enzymes the specificity determinant of the AMP site is the adenine ring whereas, in the ATP site the triphosphate component is the determinant of specificity (27). To our knowledge, all known adenylate kinases have separate ATP-and AMP-binding sites; even known multimeric AK enzymes (28) have two substratebinding sites per subunit. Taken together, the labeling of a protein corresponding to the molecular weight of mNBD1 with 8-azido-[ 32 P]AMP provides further evidence that mNBD1 of CFTR catalyzes the AK activity. An attempt was made to identify the 8-azido-[ 32 P]AMP labeling site by proteolysis, but because of very low labeling efficiency this was unsuccessful.
Reversibility of the CFTR NBDs Adenylate Kinase Activity-We have demonstrated that mNBD1, hNBD1, and hNBD2 proteins have the ability to catalyze AK activity. All known adenylate kinases also have ability to catalyze the reverse reaction in the presence of just the ADP substrate (i.e. the conversion of two ADP molecules into ATP and AMP). Using 14 C-labeled ADP, we examined the mouse and human NBD1 and hNBD2 preparations and demonstrated each protein is capable of catalyzing the reverse reaction (Fig. 5). Three independent purified mNBD1, hNBD1, and hNBD2 protein samples were tested; all displayed the ability to catalyze the reverse adenylate kinase reaction (not shown). For each NBD sample, equal amounts of ATP and AMP were generated (not shown). The reverse reaction kinetic parameters for each enzyme were determined (Table 1) (Table 1) suggesting the NBDs function within a narrow nucleotide concentration range.
Evidence for Homodimer Formation in the CFTR NBDs-Western blot analysis of the CFTR NBDs indicated to us early on that purified NBDs could dimerize under the appropriate conditions (Fig. 1F). If homodimerization was necessary for AK activity, an initial slower nonlinear rate of activity might be evident at low NBD concentrations. This appears to be the case for mNBD1 protein. When AK activity measured as a function of mNBD1 protein concentration, a slower nonlinear rate is observable at concentrations below 100 nM followed by a faster linear relationship from 100 nM to 800 nM (Fig. 6A) suggesting homodimerization may have occurred between the 50 -100 nM of protein. This initial slow nonlinear rate was seen with all four mNBD1 samples tested. It is unclear from this data whether the AK activity below 100 nM results from a monomer NBD or transient NBD dimer. To further support the hypothesis, we also examined the mNBD1 protein samples at 5 mg/ml (ϳ160 M) by static light scattering and found that the mNBD1 protein does indeed form entirely homodimers at this concentration (not shown).
Next, we wanted to examine three key CFTR residues in mNBD1 (K464A, G551D, and ⌬F508) to determine what effect these mutations had on their ability to form homodimers or alter AK activity. All three mutant proteins overexpressed, however, during their purifications two mutants (⌬F508 and K464A) were problematic. The majority of protein from these two mutants was found in the soluble aggregate fractions. The remaining monomeric mutant protein co-purified with GroEL, an E. coli chaperone, in a 1:1 ratio (not shown). Because of the chaperone contamination we were forced to abandon our analysis of the K464A and ⌬F508 proteins. In contrast, the mNBD1 G551D mutant protein purified with the same characteristics as the wild-type protein. The AK activity of G551D mutant was measured as a function of protein concentration (Fig. 6A). From this data, we find the G551D gating mutation lacks AK activity at low protein concentrations (Ͻ250 nM) when compared with the wild type protein, suggesting the mutation affects the ability of the NBD1 to dimerize. This was confirmed with two independently purified mNBD1 G551D samples. The kinetic parameters of mNBD1 G551D protein (1800 nM) were determined (Table 1) at saturating ATP (2 mM). The apparent AMP K m was 70 M and V max was 0.4 M/min. At high AMP (400 M) the apparent ATP K m was 80 M and V max was 0.4 M/min. At high protein concentrations, the G551D mutation results in only a modest decrease in catalysis and 2-fold decrease in substrate affinity when compared with wild type protein.
These data suggest the G551D mutation does not significantly disrupt catalytic function directly but rather the mutation may reduce the affinity between the two NBDs in the intact CFTR protein. Experimentally, this putative association defect can be overcome in vitro by increasing the protein concentration whereas, in vivo this is apparently not possible as this mutation causes a severe CFTR gating defect. We surmise that the G551D mutation renders NBD1 unable to associate or rearrange productively with NBD2 upon nucleotide substrate binding and that this subsequently leads to a defect in Cl Ϫ ion transport. Interestingly, in two other ABC transporter proteins the corresponding glycine has also been proposed to function in NBD dimerization (29,30). However, we cannot formally rule out the possibility that the CFTR G551D folds less efficiently than the wild-type protein (i.e. that there are a lower percentage of active molecules in G551D preparation as compared with wild type preparation) as a good active site titrating compound is not available.
Similar to the mNBD1, both hNBD1 and hNBD2 when assayed individually demonstrated a slower nonlinear AK rate at low protein concentrations suggesting homodimer formation might be important for stimulated AK activity (Fig. 6, B and C). These results were true for four independently refolded hNBD1 and hNBD2 samples (not shown). Consistently, there was a noticeable exaggerated nonlinear phase when hNBD1 and hNBD2 proteins were refolded with NSDB-256 procedure and less so when the proteins were refolded using condition 10 (Fig. 6C,  hNBD2 plots). If hNBD1 protein was refolded in condition 10 without NDSB-256 (Fig. 6B, squares) and in condition 10 buffer with 0.5 M NSDB-256 (Fig. 6B, circles) a slower nonlinear was also more apparent when NDSB-256 was included. This observation suggests that the addition of NSDB-256 (in otherwise identical buffer) alters the monomerdimer equilibrium of the refolded hNBD proteins such that a greater  percentage of monomer is present when the additive is included. Taken together, these results suggest that all three NBD proteins examined appear to have the ability to form homodimers, which leads to AK or stimulated AK activity.
Evidence for Heterodimer Formation in the Human CFTR NBDs-The functional unit of the full-length CFTR chloride channel is suggested to be a monomeric CFTR protein (31,32). Within each CFTR protein there are two NBDs (NBD1 and NBD2) that are believed to govern channel gating. Channel gating models in CFTR and in other ABC transporters invoke association of the two NBDs. Nucleotide binding may drive association or rearrangement of the NBDs. Subsequent association or conformational changes within the NBDs is hypothesized to enhance catalysis, which may in turn lead to opening of the transport channel. We were interested in determining (a) if the isolated CFTR NBDs could form heterodimers, (b) whether nucleotide binding drives association between NBDs, and (c) whether NBD association enhances AK activity. Light scattering and other biophysical experiments examining human NBD association were not easily achievable as refolded NBD protein supply was limiting and the separation of homo-and heterodimers would be a challenge as the calculated molecular masses of the hNBD1 and hNBD2 are very similar (ϳ34 kDa). In addition the monomer to dimer transitions most likely occur at the nanomolar concentration range, which is beyond the limits of many biophysical techniques. However, the hNBD proteins do not run at the same position on SDS-PAGE gel (Fig. 1B). Using commercially available specific monoclonal antibodies for hNBD1 and hNBD2 we were able to conduct coimmunoprecipitation studies. Using a hNBD2 specific antibody, we immunoprecipitated 25 nM of hNBD2 protein which was mixed with 4 nM hNBD1 for 30 min under heterodimer activity reaction conditions (see below) with or without nucleotide substrates/inhibitor. The precipitated material was then separated on SDS-PAGE gel, Western blotted, and probed with a specific hNBD1 antibody. The Odyssey secondary detection system revealed that the 4 nM hNBD1 protein in the reaction was quantitatively immunoprecipitated by the hNBD2 antibody ( Fig. 7) with very little remaining in post-immunoprecipitated supernatant (not shown). Having ATP/AMP or Ap 5 A present during the incubation and wash steps of the experiment did not block the immunoprecipitation of hNBD1 (Fig. 7) at these protein concentrations. This pull-down exper-iment demonstrates that hNBD2 and hNBD1 proteins physically associate and form a stable complex in vitro.
Finally, we wanted to determine if there is a synergistic increase in AK activity upon heterodimerization. To accomplish this, the concentration of one NBD was fixed whereas the other NBD was titrated into the reaction. Control reactions included each NBD at the same concentration as in the heterodimer reaction. The synergistic heterodimer AK activity was determined by subtraction of the fixed-NBD control reactions and the corresponding variable-NBD reaction from the heterodimer reaction value. Initially, we held hNBD2 fixed at 13 nM and varied hNBD1 from 0 nM to 2.3 nM (Fig. 8, A and B). In the heterodimer reaction with the lowest amount of hNBD1 (0.2 nM) added, a significant synergistic increase in AK activity was apparent in the heterodimer lane when this lane was compared with NBD alone control lanes (Fig. 8, A  and B). Quantitative analysis indicated the synergistic activity was 57-fold greater than hNBD1 alone and 8-fold greater than hNBD2 alone (Fig. 8B). We note the synergistic activity did not increase with increasing amounts of hNBD1 (Fig. 8B). One explanation may be that all free monomer hNBD2 is already bound up by the initial quantity of hNBD1 and the addition of more hNBD1 (up to 2.3 nM) is insufficient at altering the equilibrium of hNBD2 homodimer. We also set up the converse experiment where we fixed hNBD1 (2 nM) and varied hNBD2 from 0 to 25 nM (Fig. 8, C and D). In this case, the synergistic AK activity increases in a linear relationship with increasing amounts of hNBD2 (Fig. 8D). This linear increase in AK activity suggests that only a small fraction of the 2 nM hNBD1 is complexed with each addition of hNBD2 protein in this titration. This mixing experiments and the converse mixing experiment were repeated three times with a synergistic increase in adenylate kinase easily apparent in all experiments (not shown). Our data convincingly demonstrate that when hNBD1 and hNBD2 are combined a significant synergistic increase in AK activity occurs suggesting association of hNBD1 with hNBD2 enhances catalytic activity of one or both NBDs. These observations give credence to ABC transporter gating models that hypothesize that the proper association of two NBDs leads to enhanced catalytic activity which may be the molecular mechanism for opening and closing the transport channel.
NBD1-R C-terminal Domain Has an Inhibitory Role on the Adenylate Kinase Activity-We believe that our experiments have clearly demonstrated that the NBDs catalyze AK activity, but the possibility remains that larger CFTR protein fragments may gain extra functionality (e.g. ATPase) from additional contiguous residues. Two CFTR biochemical reports suggested this might be a possibility (12,13). To examine this, pure denatured hNBD1-R protein (Thr 390 -Ile 840 ) with an N-terminal His tag was put through the refolding screen at 100 g/ml multiple times. However, unlike the hNBDs, no detectable ATPase or adenylate kinase activity was detected when assayed four ways (ATP alone, ATP and AMP, and with and without the presence of PKA. Sufficient PKA was added to phosphorylate the NBD1-R fragment completely in the first 10 min of the 1 h activity assay as judged by SDS-PAGE and autoradiography. Based on observations that the R-domain phosphorylation is a prerequisite to nucleotide-stimulated Cl Ϫ ion transport (4), we believed assaying the phosphorylated NBD1-R protein for enzymatic activity was critical. The lack of activity in the refolded NBD1-R protein may be attributed to the possibility that this protein may fold less efficiently than hNBD1. Alternatively, the phosphorylated R-domain may still prevent activity in the NBD1-R construct if the preferred binding site for the phosphorylated R-domain is only present in the full-length CFTR protein (33) (i.e. in the absence of the phosphorylated R-domainbinding site, rearrangement does not occur and, as a result, the NBD inhibitory property is not relieved). We also repeated the NBD1-R refolding condition described in Annereau et al. (13) without success. In this experiment the only difference was the NBD1-R constructs (Annereau: Gly 404 -Lys 830 and ours Thr 390 -Ile 840 ). As our construct encompassed theirs, this reason for failure seemed less probable.
We chose to pursue the NBD1-R study further by expressing a truncated version of the construct (Thr 390 -Ile 708 ) to test the hypothesis that R-domain contains an inhibitory domain. Studies have been reported (4,34) that the removal of residues between 708 and 835 in the full-length CFTR protein results in chloride channel that no longer requires activation by PKA. We reasoned that similar truncation would restore NBD1-R activity. In a refolding screen (21) we observed that this refolded truncated NBD1-R construct possessed AK activity, which was comparable to hNBD1 but not ATPase activity (not shown) and supports the possibility that C-terminal R-domain possesses inhibitory properties regardless of its phosphorylation state. This also suggests that a more complete biochemical picture on the regulation of the CFTR protein activities requires analysis of the purified full-length protein.

CFTR Gating Model Based on NBD1 and NBD2 Adenylate Kinase
Activity-We have demonstrated that isolated NBD1 and NBD2 fragments of CFTR function as adenylate kinases both individual and together, but not as ATPases. CFTR patch-clamp studies by Randak and Welsh (16,35,36) suggest that CFTR gating is regulated by AK activity in vivo. Using the patch-clamp technique a number of observations are made, which support an AK gating hypothesis including: 1) They showed that at intermediate ATP concentrations AMP can reversibly increase the CFTR current 20 -30% and GMP cannot substitute (16). 2) Randak and Welsh demonstrate that in the presence of AMP the rate of channel opening increases whereas, the adenylate kinase inhibitor Ap 5 A has the opposite effect (16). 3) The transfer of a phosphate between ATP and AMP was required for normal current generation and positive cooperativity (16). 4) They noticed that the ADP ligand induces positive cooperativity, which supports a conclusion that AK activity gates the CFTR channel rather than ATPase (16). 5) It was found the Ap 5 A also attenuated the ADP inhibition of CFTR current (16,35) suggesting that ADP is functioning through an AK reaction. Together, their CFTR gating data and our own characterization of the CFTR NBDs biochemical properties are consistent with a CFTR gating model based on adenylate kinase activity. In such a model, after R-domain phosphorylation by PKA, and in the presence of ATP and/or AMP, the NBD1 and NBD2 domains associate or conformationally rearrange such that they become active in or stimulate AK activity. Nucleotide(s) binding to the NBDs causes the proper interaction of the two membrane spanning domains (MSD) that in turn permits Cl Ϫ ions to move through the channel. In fact, others have shown that purified dimeric MSD2 from CFTR can indeed mediate Cl Ϫ ion flux whereas, monomeric forms of MSD2 cannot (37). This model could also explain a well known observation that when ADP is added to patch-clamped CFTR channels the current is inhibited (16,35). In the presence of a threshold concentration of ADP it is possible that the heterodimer AK activity functions in reverse driving the disassociation or nonproductive rearrangement of the heterodimer or ADP simply obstructs the mechanism. This would suggest that Cl Ϫ ions are transported through the channel when the heterodimer is in the ATP-and/or AMP-bound state.
How and why might adenylate kinase activity be needed in a CFTR mechanism to transport Cl Ϫ ions? It is known that AK activity generates very little free energy (38) and may be an insufficient amount of energy to drive conventional energy-dependent transport. However, CFTR is known to passively transport Cl Ϫ ions (i.e. not against a Cl Ϫ gradient) and may not need a large input of energy. Indeed, ATP and/or AMP binding maybe all that is required by the CFTR protein to rearrange itself in a productive Cl Ϫ ion transport conformation after R-domain phosphorylation has occurred. The role of the adenylate kinase activity in CFTR may be to convert the inducing ligand(s) (ATP and/or AMP) to another form (ADP), which dissociates more quickly from the protein and thereby provides the CFTR containing cell with more rapid tempo- FIGURE 8. CFTR heterodimer NBD AK activity studies. A, image of a thin layer chromatography plates showing the AK activity when hNBD1 is titrated (from 0 to 2. 3 nM) into a fixed concentration of hNBD2 (13 nM). The left side shows the AK activity of hNBD1 titration alone. The far right lane shows the AK activity of hNBD2 alone. B, AK activity in A was quantified by the phosphorimager and graphed as a function hNBD1 protein concentration. The filled triangles represent the AK activity of hNBD1 alone. The filled squares represent the total AK activity of the heterodimer reaction. The filled circles represent the synergistic AK activity of the heterodimer reaction and are determined by the subtraction of the fixed NBD control reaction and corresponding variable NBD reaction from the heterodimer reaction value. C, image of a thin layer chromatography plates showing the AK activity when hNBD2 is titrated (from 0 to 25 nM) into a fixed concentration of hNBD1 (2 nM). D, AK activity in C was quantified by the phosphorimager and graphed as a function of hNBD2 protein concentration. The triangles represent the AK activity of hNBD2 alone. The squares represent the total AK activity of the heterodimer reaction. The circles represent the synergistic AK activity of the heterodimer reaction. ral control of Cl Ϫ ion transport. For a passive transporter this seems the simplest model. In this ligand-induced model, the stoichiometry of Cl Ϫ ion transport with respect to adenylate kinase activity is not required to be fixed and may be dependent on the electrochemical (Cl Ϫ ) gradient that is applied. Brownian movement of the Cl Ϫ ion provides the kinetic energy for the movement across the membrane and may be assisted by the chemical attributes of the channel.
Nucleotide Stoichiometry of the CFTR NBDs and Implications for Full-length CFTR Protein-Currently, it is unclear whether a monomer of either hNBD1 or hNBD2 is capable of AK activity on its own or whether dimerization is necessary for activity. Comparing the structures of mNBD1 with the adenylate kinases, it seems possible that an AMP substrate could bind adjacent to the ATP site in ␣-helical region (18), suggesting a monomer may be an active species. If NBD monomers are an active species then a minimum of four nucleotide-binding sites are present in the heterodimer. However, we favor that NBD dimers are the only AK active species since nucleotide substrate sites are partially solvent exposed in the monomer (18) and the AK catalytic machinery cannot easily be protected from hydrolysis by predicted monomeric conformational changes. In contrast, in the NBD dimer models, water molecules are more easily excluded from interfering with the AK catalysis. Based on our protein labeling studies and strict requirement for AMP (GMP will not substitute), we speculate that there must be at least two nucleotide-binding sites per NBD monomer. Future efforts should determine the nucleotide stoichiometry and whether the dimer is the only active species. These activities will further refine the proposed gating model. An interesting observation with AP 5 A in the patch-clamp studies by Randak and Welsh (16) shows that the maximum current inhibition is 50%, suggesting it may block only two of four suspected nucleotide substrates sites at a time.
Although we were unable to show any ATPase or AK activity with a longer hNBD1-R construct there remains the possibility that the fulllength CFTR protein may encode both AK and ATPase activities. Ramjeesingh et al. (39) have purified the full-length CFTR protein and examined it in vitro. The Bear and other laboratories ( (11,12,13,32,39,40) have presented both biochemical and functional studies that indicate that CFTR could function as an ATPase and that this activity may be responsible for gating the CFTR channel.
This study suggests four important considerations that are relevant to future biochemical investigation of full-length CFTR or other ABC transporters. 1) The possibility of both AK and ATPase activity in CFTR requires methods that can distinguish between two activities (note: ␣-phosphate-labeled ATP cannot distinguish between the two activities). 2) AK enzymes generally have low ATP and AMP K m values (Ͻ50 M); impure ATP stocks could unintentionally provide the AMP substrate. 3) Contaminant enzymatic activities can confound assay results (e.g. our initial ATP diphosphatase contaminant). Abundant membrane bound and extracellular enzymes exist that manipulate ATP (41) and may purify with the desired protein. Therefore, assay methodologies need to resolve competing or unwanted ATP utilizing activities and confirm the identity of the product(s) with known standards. 4) ABC transporters often possess improved biochemical activity with lipids present. Lipids from natural sources may contain trace quantities of enzymes or nucleotides that could provide the missing AMP substrate. Awareness of these considerations should help in revealing all the ATP utilizing activities intrinsic to the full-length CFTR or ABC transporter.
Implication for Other ABC Transporter Systems-In the literature, there exist experimental observations that suggests some ABC transporters may operate as adenylate kinase engines rather than ATP hydrolytic engines. In one example, the addition of AMP to an ATP-stimu-lated K ATP channel increased the open probability time and rather than invoke an accessory AK enzyme (42) the sulfonylurea receptor component of the K ATP channel could be this AK enzyme. In LmrA, a net synthesis of ATP from ADP and a downhill substrate transport was demonstrated (43). An alternate explanation is that ATP synthesis arose from AK activity rather than the reversal of ATP hydrolysis, which requires energy input. A recent MRP1 NBD report, showed very little ATPase activity when assayed individually or when combined (30). It would be interesting to determine if these NBDs have the potential to catalyze AK activity, and even more importantly, if AK activity is more widely distributed among the ABC transporter family. Based on our biochemical data, we stress the importance of characterization of ABC transporters through direct experimentation, and not merely via assumptions based on similarity to other protein family members.