Identification of Residues in the Drug Translocation Pathway of the Human Multidrug Resistance P-glycoprotein by Arginine Mutagenesis*

P-glycoprotein (P-gp, ATP-binding cassette B1) is a drug pump that extracts toxic drug substrates from the plasma membrane and catalyzes their ATP-dependent efflux. To map the residues in the drug translocation pathway, we performed arginine-scanning mutagenesis on all transmembrane (TM) segments (total = 237 residues) of a P-gp processing mutant (G251V) defective in folding (15% maturation efficiency) (glycosylation state used to monitor folding). The rationale was that arginines introduced into the drug-binding sites would mimic drug rescue and enhance maturation of wild-type or processing mutants of P-gp. It was found that 38 of the 89 mutants that matured had enhanced maturation. Enhancer mutations were found in 11 of the 12 TM segments with the largest number found in TMs 6 and 12 (seven in each), TMs that are critical for P-gp-drug substrate interactions. Modeling of the TM segments showed that the enhancer arginines were found on the hydrophilic face, whereas inhibitory arginines were located on a hydrophobic face that may be in contact with the lipid bilayer. It was found that many of the enhancer arginines caused large alterations in P-gp-drug interactions in ATPase assays. For example, mutants A302R (TM5), L339R (TM6), G872R (TM10), F942R (TM11), Q946R (TM11), V982R (TM12), and S993R (TM12) reduced the apparent affinity for verapamil by ∼10-fold, whereas the F336R (TM6) and M986R (TM12) mutations caused at least a 10-fold increase in apparent affinity for rhodamine B. The results suggest that P-gp contains a large aqueous-filled drug translocation pathway with multiple drug-binding sites that can accommodate the bulky arginine side chains to promote folding of the protein.

The human multidrug resistance P-glycoprotein (P-gp, ATP-binding cassette B1) 2 is an ATP-dependent drug pump that mediates efflux of a broad range of hydrophobic compounds out of the cell (1). It is expressed in the epithelium of liver, kidney, and gastrointestinal tract and at the blood-brain or blood-testes barrier where it functions to protect us from cytotoxic compounds. It is clinically important because it contributes to multidrug resistance in diseases such as cancer and AIDS (1). P-gp is an ATP-binding cassette transporter of 1280 amino acids that consists of two homologous halves (2). Each half begins with a transmembrane domain (TMD) containing six TM segments followed by a nucleotide-binding domain (NBD).
A key goal to understanding the mechanism of P-gp drug transport is to identify the amino acids that line the drug translocation pathway. Because P-gp extracts drug substrates from the lipid bilayer, the drug-binding pocket/drug translocation pathway are predicted to reside in the transmembrane (TM) segments. We previously showed that the TMDs alone were sufficient for drug binding (3). Expression of the TMDs as separate polypeptides showed that both TMD1 and TMD2 were required for binding drug substrate (4). The results of studies utilizing cysteine-scanning mutagenesis and labeling with thiol-reactive drug substrates suggested that all of the TM segments contribute to the drug-binding pocket/drug translocation pathway (reviewed in Ref. 5). The next step is to identify the specific amino acids that line the drug translocation pathway. It is important to identify amino acids that line the drug translocation pathway and to compare whether the biochemical evidence supports a model of P-gp structure in the closed conformation (6) (NBDs close together that was based on the bacterial Sav1866 crystal structure (7)) or the recent crystal structure of mouse P-gp in the open conformation (NBDs far apart) (8).
There have been concerns that the mouse P-gp structure may be a non-native structure or in a conformation that exists very transiently (9).
Our approach to map the drug translocation pathway has been to use arginine-scanning mutagenesis of the TM segments of a P-gp processing mutant (G251V) that shows partial maturation (ϳ15% maturation efficiency) (10). Maturation efficiency can be used to detect folding of P-gp in whole cells by monitoring the conversion of P-gp from a core-glycosylated (150 kDa) protein to a mature protein (170 kDa) that contains complex carbohydrate. Because mutant G251V shows partial maturation, we can detect whether an introduced arginine promotes, inhibits, or has a neutral effect on folding. The rationale for using arginine-scanning mutagenesis was that arginine has a large free energy barrier (17 kcal/mol) for insertion into the lipid bilayer because it is highly charged (11). Therefore, introduction of an arginine into a lipid face of the G251V mutant would likely inhibit maturation, whereas an arginine intro-duced into the aqueous face of the drug translocation pathway would not inhibit maturation of the mutant P-gp.
In an initial study on TM1, we demonstrated the feasibility of the approach (10). All arginines introduced into the predicted lipid-facing positions inhibited maturation, whereas those introduced into positions predicted to face the drug translocation pathway did not. A particularly intriguing observation was that some arginines promoted maturation. The residues at these positions were coincidentally at positions identical to those that reacted with thiol-reactive drug substrates in cysteine-scanning mutagenesis studies and were found to be within the drug-binding pocket (10,12). This suggested that arginine-scanning mutagenesis could be a useful approach for identifying residues in the drug translocation pathway and for determining the orientation of the TM segments in the membrane.
Arginines that promote maturation appear to identify positions that are important for P-gp-drug interactions because they appear to mimic drug rescue of P-gp. It was also found that the ability of arginines (such as I306R in TM5) to promote maturation involved global enhancement of P-gp folding rather than simply compensating for a localized mutation (such as G251V) because other processing mutants could also be rescued (12). Because these arginine mutations enhance folding of P-gp in general, they will be described as enhancer rather than suppressor arginines. In this study we performed argininescanning mutagenesis on TMs 2-12 of P-gp processing mutant G251V to determine their orientations in the membrane and to identify residues that line the drug translocation pathway.
For disulfide cross-linking analysis, the cDNA of mutant L339C (TM6)/F728C (TM7) (15) was modified to also encode the I868R or T945R mutations. The integrity of all the mutant cDNAs was confirmed by sequencing the entire cDNA (16).
Expression and Detection of Mutants-The mutant P-gps were transiently expressed in human embryonic kidney (HEK) 293 cells in Dulbecco's modified Eagle's medium with 10% (v/v) calf serum as described previously (14). Whole cell SDS extracts were subjected to SDS-PAGE (6.5% (w/v) polyacrylamide gels) and immunoblot analysis with monoclonal antibody A52 (17). The gel lanes were scanned, and the amount of mature P-gp relative to total P-gp was quantitated using the NIH Image program and an Apple computer. The glycosylation status of P-gp was monitored by treatment with endoglycosidase H or peptide N-glycosidase F (New England Biolabs, Mississauga, Canada) as described previously (3,18).
Purification of P-gp and Measurement of ATPase Activity-Histidine-tagged P-gps were expressed in HEK 293 cells and then isolated by nickel-chelate chromatography as described previously (14). Recovery of P-gp was monitored by immunoblot analysis with rabbit anti-P-gp polyclonal antibody (17). A sample of the isolated histidine-tagged P-gp was mixed with an equal volume of 10 mg/ml sheep brain phosphatidylethanolamine (Type II-S; Sigma) that had been washed and suspended in Tris-buffered saline. The sample was sonicated, and ATPase activity was measured in the presence of various concentrations of verapamil, vinblastine, rhodamine B, or colchicine.
Disulfide Cross-linking Analysis-The double cysteine mutants L339C (TM6)/F728C (TM7), I868R (TM10)/L339C (TM6)/F728C (TM7), and T945R (TM11)/L339C (TM6)/ F728C (TM7) were transiently expressed in HEK 293 cells. The membranes were prepared as described previously (19) and suspended in Tris-buffered saline, pH 7.4. A sample of the membrane was then incubated in the presence or absence of various concentrations of vinblastine for 15 min at 20°C. The samples were then cooled on ice for 10 min and treated with 0.2 mM of the homobifunctional methanethiosulfonate crosslinker 3,6,9,12-tetraoxatetradecane-1,14-diyl bismethanethiosulfonate (M14M, 20.8 Å) (Toronto Research Chemicals, Toronto, Canada) for 4 min on ice (20,21). The reactions were stopped by the addition of 2ϫ SDS sample buffer (125 mM Tris-HCl, pH 6.8, 20% (v/v) glycerol, and 4% (w/v) SDS) containing 50 mM EDTA and no reducing agent. The reaction mixtures were then subjected to SDS-PAGE (7.5% (w/v) polyacrylamide gels) and immunoblot analysis with a rabbit polyclonal antibody against P-gp. Intramolecular disulfide cross-linking between TMD1 and TMD2 can be detected because the crosslinked product migrates with a slower mobility on SDS-PAGE gels (22). The gel lanes were scanned, and the amount of crosslinked product was quantitated using the NIH Image program and an Apple computer.

Effect of TMs 2-12 Arginines on Maturation of Processing
Mutant G251V-Each half of P-gp consists of a TMD with six TM segments and an NBD (Fig. 1A). The two halves of P-gp can be expressed as separate polypeptides that will associate when expressed in the same cell to form a functional transporter (23). Fig. 1 (B-D) shows the models for N-half (residues 36 -631) C-half (residues 696 -1276), and full-length human P-gps, respectively (6). The models were based on the crystal structure of the Sav1866 multidrug transporter that was in the "closed conformation" (NBDs close together with TMDs in an outward facing conformation) (7). The models are consistent with cysteine mutagenesis and disulfide cross-linking studies of human P-gp (24,25). The G251V processing mutation is located in the cytoplasmic segment that connects TM segments 4 and 5 ( Fig. 1A). The protein is defective in folding because it shows only ϳ15% maturation efficiency compared with the greater than 90% maturation efficiency of wild-type P-gp ( Fig. 2A). Although mutant G251V shows inefficient folding, the mature form of the protein is functional. We previously observed that introduction of arginines into TM segments such as T199R (TM3), I306R (TM5), and F343R (TM6) promoted maturation of the P-gp G251V processing mutant (26). Arginines were introduced into these positions because cysteine-scanning mutagenesis and modification with thiol-reactive drug substrates suggested that they lined the drug-binding pocket. Covalent modification of cysteines at positions 199 (TM3), 306 (TM5), or 343 (TM6) with thiol-reactive drug substrates caused activation of P-gp ATPase activity (26 -28). Fig. 1B shows that residues 199, 306, and 343 are predicted to lie near the middle of the TM segments at the interface between TMD1 and TMD2. The results of a previous arginine-scanning mutagenesis study on TM1 (10) also support the models shown in Fig. 1 (B and D).
To test the effects of arginines, we constructed 219 A52tagged G251V mutants containing a single arginine at each position of TMs 2-12. HEK 293 cells were then transfected with the mutant cDNAs. The medium was changed the next day. After another 24 h at 37°C, whole cell SDS extracts were then subjected to immunoblot analysis using monoclonal antibody A52. Examples of the most common effects of the arginines on maturation of mutant G251V P-gp are shown in Fig. 2A. The  Identification of arginine mutations that promote, inhibit, or have little effect on maturation of mutant G251V. A, 48 h after transfection, whole cell SDS extracts of HEK 293 cells expressing A52-tagged wildtype P-gp (Wild-type), mutant G251V, mutants A230R/G251V, V231R/G251V, or W232R/G251V in TM4 or mutants F770R/G251V or Q773R/G251V in TM8 were subjected to immunoblot analysis with monoclonal antibody A52. B, 48 h after transfection, whole cell SDS extracts of HEK 293 cells expressing mutant W232R/G251V were treated with (ϩ) endoglycosidase H f (H), peptide N-glycosidase F (F), or no (Ϫ) endoglycosidase. Equivalent amounts were subjected to immunoblot analysis. C, 24 h after transfection, whole cell SDS extracts of HEK 293 cells expressing A52-tagged wild-type P-gp (Wild-type), mutant W232R in TM4, or I306R in TM5 were subjected to immunoblot analysis. D-E, the amount of mature P-gp (170-kDa protein) relative to total (mature 170-kDa protein plus immature 150-kDa protein) (% mature) was quantitated (D). Forty-eight hours after transfection, whole cell SDS extracts of HEK 293 cells expressing A52-tagged mutants ⌬Y490 or ⌬Y490/W232R (E), mutants ⌬NBD2 (deleted residues 1024 -1280) or ⌬NBD2/W232R (F) were subjected to immunoblot analysis. The positions of the mature (170 kDa), immature (150 kDa), and unglycosylated (Unglycos) P-gps are indicated. major P-gp product found in cells transfected with the A52tagged wild-type P-gp cDNA was the mature 170-kDa protein. By contrast, the major product in mutant G251V was the 150-kDa protein. Arginines introduced in TMD1 (at positions 230 -232 (TM4)) or in TMD2 (positions 770 and 773 (TM8)) of the G251V mutant showed different effects on maturation of the protein. The A230R (TM4) mutation had little effect. The V231R (TM4) and Q773R (TM8) mutations inhibited maturation of the protein because only immature 150-kDa P-gp was detected. The W232R (TM4) and F770R (TM8) mutations, however, promoted folding of the G251V mutant because the mature 170-kDa protein was the major product.
To confirm that the W232R (TM4) mutation did indeed promote maturation of G251V P-gp, the samples were subjected to treatment with endoglycosidases. Fig. 2B shows that only the 150-kDa P-gp protein was sensitive to endoglycosidase H, whereas both the 150-and 170-kDa proteins were sensitive to peptide N-glycosidase F. These results show that the 150-kDa protein represents core-glycosylated P-gp, whereas the 170-kDa protein contains complex carbohydrates added in the Golgi.
To test whether arginines that lie close to the G251V processing mutation promoted global folding of P-gp or simply compensated for the local effects of the G251V mutation, the W232R (TM4) and I306R (TM5) mutations were introduced into wild-type P-gp. HEK 293 cells were then transfected with mutant W232R (TM4) or I306R (TM5), the cells were harvested after a relatively short incubation period (24 h), and SDS whole cell extracts were subjected to immunoblot analysis. The short incubation period results in approximately equivalent amounts of mature (170 kDa) and immature (150 kDa) protein in wild-type P-gp, and this can be used to detect whether an introduced arginine promotes maturation of P-gp. Fig. 2C shows that the amount of mature (170 kDa) protein in wildtype P-gp was ϳ60% after 24 h. Introduction of the enhancer arginines (W232R (TM4) or I306R (TM5)), however, resulted in ϳ90% mature (170 kDa) protein (Fig. 2D). We previously showed that the I306R (TM5) mutation promoted maturation of mutants that had processing mutations located throughout the molecule (12). The W232R (TM4) mutation was also introduced into the ⌬Y490 and ⌬NBD2 processing mutants and expressed in HEK 293 cells. Immunoblot analysis showed that the W232R (TM4) mutation promoted maturation of these mutants (Fig. 2, E and F).
The effects of arginines introduced at other positions in the TMs 2-6 (TMD1) of mutant G251V were determined. HEK 293 cells were transfected with the mutant cDNAs. After 48 h, whole cell SDS extracts were subjected to immunoblot analysis. The results from a previous study on TM1 (Fig. 3A) are summarized in a helical wheel and shown for comparison (10). Fig.  3 shows the ratio of mature 170-kDa P-gp to total P-gp (mature 170-kDa plus immature 150-kDa protein) for arginine mutants in TMs 2-6. Wild-type P-gp showed a maturation efficiency of ϳ90% compared with ϳ15% for the G251V mutant (Fig. 3B). In TM2 (Fig. 3B), arginines introduced into the extracellular end of TM2 (positions 118 -128) of mutant G251V inhibited maturation of the protein. In contrast, most of the arginines introduced into the cytoplasmic end of TM2 (positions 129 -137) did not inhibit maturation (Fig. 3B). When TM2 was modeled as an ␣-helical wheel, arginines at positions 129 -137 that inhibited maturation (such as I131R, S134R, and W136R) appeared to reside within a hydrophobic face (composed of residues Gly 120 , Ala 123 , Gly 124 , Val 127 , Ile 131 , and Ser 134 ) (Fig.  3B, inset). One enhancer arginine (C137R) at the intracellular end of TM2 showed an ϳ3-fold (ϳ50%) increase in maturation efficiency. The results of arginine-scanning mutagenesis of TM2 were different from those observed with TM1 (10). It was previously found that all arginines introduced into the intracellular end of TM1 (residues 53-60) inhibited maturation of the G251V mutant. Enhancer arginines were identified at the extracellular end of TM1 at positions 64, 65, and 68 (Fig. 3A).
Multiple enhancer mutations were identified when argininescanning mutagenesis was performed on TM segments 3-5 ( Fig. 3, C-E). In TM3, the enhancer mutations Q195R, S196R, and T199R promoted maturation of the G251V mutant by 2-3fold. The T199R mutation was previously found to also promote maturation of the ⌬Y490 P-gp processing mutant (26). Arginines introduced into TM3 that inhibited maturation of the mutant were located at every third or fourth position. When the residues of TM3 (Fig. 3C, inset) were organized as an ␣-helical wheel, residues Thr 202 , Gly 191 , Ala 198 , Ile 205 , Phe 194 , and Phe 201 appeared to be on a hydrophobic face because insertion of arginines at these positions inhibited maturation of mutant G251V. By contrast, the other residues of TM3 appeared to face a more hydrophilic environment because insertion of arginines at 9 of 12 positions (on the opposite side of the wheel) did not inhibit maturation.
The results of arginine-scanning mutagenesis of TM4 showed a pattern that was similar to TM3 when the results are presented in a ␣-helical wheel (Fig. 3D). Residues Leu 227 , Ala 220 , Val 231 , Val 224 , Val 217 , and Ser 228 appear to face the lipid bilayer because introduction of arginines at these positions inhibited maturation of P-gp. Other positions in the predicted helix appear to face a hydrophilic environment because arginines introduced at 11 of the 12 other positions did not block maturation. Four arginine enhancer mutations were identified at positions I218R, I221R, L225R, and W232R. The W232R mutation was the most effective because it increased maturation efficiency to ϳ70%. These enhancer mutations appeared to be clustered on the same face of TM4 (Fig. 3D, inset).
In TM 5, four enhancer mutations (I299R, A302R, F303R, and I306R) were identified (Fig. 3E). The A302R, F303R, and I306R enhancer mutations were very effective in promoting maturation of the G251V processing mutant because greater than 60% of P-gp was present as the mature protein. The helical wheel projection (Fig. 3E, inset) suggested that TM5 has distinct hydrophobic and hydrophilic faces.
In TM6, however, arginine-scanning mutagenesis did not reveal a hydrophobic face (Fig. 3F), because most of the introduced arginines did not inhibit maturation. Only arginines introduced at positions 337 and 331 inhibited maturation of the G251V mutant. The seven enhancer mutations observed in TM6 at positions F336R, V338R, L339R, I340R, F343R, S344R, and Q347R promoted maturation of mutant G251V by at least 2-fold. The largest number of enhancer mutations in TMD1 was found in TM6. The lack of a hydrophobic face in TM6 might indicate that it is not in contact with the lipid bilayer.
Arginine-scanning mutagenesis was also performed on the six TM segments in the COOH-terminal half of P-gp. The effects of the TMD2 arginine mutations on maturation of mutant G251V are shown in Fig. 4.
The results observed with the TM7 arginine mutants (Fig.  4A) were similar to those previously observed with TM1 (Fig.  3A). Like TM1, introduction of arginines at all positions at the predicted cytoplasmic end of TM7 (positions 712-720) inhibited maturation of the protein. There appeared to be a hydrophilic face (Fig. 4A, helical wheel) at the extracellular end of TM7 because mutants N721R and Q725R did not inhibit maturation, whereas mutant F728R showed an ϳ2.5-fold increase in maturation efficiency.
In TM8, two enhancer mutations (S766R and F770R) were identified (Fig. 4B). They were predicted to reside on the same face of the TM segment on adjacent turns of a ␣-helix. The F770R mutation was very effective in promoting maturation because it showed ϳ70% maturation efficiency. Introduction of arginines at all other positions in TM8 inhibited maturation of mutant G251V.
No enhancer mutations were found in TM9 (Fig. 4C). The TM segment still appeared to have distinct hydrophobic and hydrophilic faces because all four arginines (Q838R, A834R, N839R, and N842R) that did not inhibit maturation were predicted to lie on one face of TM9.
In TM10, all arginines introduced into the extracellular half of the TM segment (positions 856 -867) inhibited maturation of mutant G251V. In segment 868 -873, however, it was found that introduction of arginines at positions Ile 868 , Gly 872 , and Val 873 promoted maturation by at least 2-fold.
Seven arginine enhancer mutations were found in TM12 that enhanced maturation of mutant G251V by at least 2-fold (Fig.  4F). This is the largest number of enhancer mutations found in TMD2. There was one notable difference between TM6 and TM12, however, when their helical wheel plots were compared (insets of Figs. 3F and 4F). TM12, but not TM6, appeared to have a distinct hydrophobic face because the F978R, A985R, V974R, V981R, V988R, and V977R mutations inhibited maturation of mutant G251V.
Effect of Enhancer Arginines on Drug-stimulated ATPase Activity of Wild-type P-gp-The induced fit mechanism of drug substrate recognition by P-gp predicts that the protein contains multiple flexible drug-binding sites in a common drug-binding pocket (29). Therefore, introduction of a relatively large charged arginine side chain at some positions in the translocation pathway should result in altered drug-stimulated ATPase activity with at least some substrates. By contrast, introduction of an arginine into a high affinity (low flexibility) drug-binding site or translocation pathway would inhibit its ATPase activity. If P-gp has a rigid structure, then it would be expected that most of the introduced arginines would inhibit all drug-stimulated ATPase activity.
To test whether enhancer arginines inhibited activity or disrupted P-gp-drug interactions, we introduced them into a wildtype P-gp cDNA. The wild-type cDNA encoded a 10-histidine tag so that the protein could be isolated by metal-chelate chromatography (14). Drug interactions can be measured with isolated P-gp because most drug substrates will activate P-gp ATPase activity. Activation of P-gp ATPase activity in the presence of drug substrates appears to be a measure of P-gp-drug interactions because there was a good correlation with drug transport (30).
The drug substrates vinblastine, colchicine, verapamil, and rhodamine B were selected for analysis. Vinblastine and colchicine were selected because they are "classic" drug substrates of P-gp (31). Verapamil and rhodamine B were selected because thiol-reactive derivatives of these drug substrates have been used extensively in cysteine mutagenesis studies to characterize the drug-binding sites of P-gp (32)(33)(34). Verapamil (35) and rhodamines (36) are transported by P-gp. Examples of the drugstimulated ATPase activities of wild-type P-gp and the M986R (TM12) enhancer mutation (in wild-type background) are shown in Fig. 5. The histidine-tagged P-gps were transiently expressed in HEK 293 cells. P-gp was extracted from the cells using the detergent dodecyl-␤-D-maltoside and then isolated by nickel-chelate chromatography. The isolated P-gps were mixed with lipid, and ATPase activities were measured in the presence of various concentrations of vinblastine, verapamil, rhodamine B, or colchicine. P-gp drug-stimulated ATPase activity showed the characteristic (37) concentration-dependent increase in activity to a maximum and then a decrease at higher drug concentrations (  observed with vinblastine and verapamil because less than 2-fold changes were observed in S 50 or S max . It was also observed that the mutant showed an ϳ2-fold increase in basal ATPase activity (0.26 Ϯ 0.04 compares to 0.14 Ϯ 0.02 for wildtype P-gp). No detectable change in the basal ATPase activity was observed in the other mutants. Therefore, it appeared that the major effect of the M986R enhancer mutation was to increase the ability of P-gp to interact with colchicine or rhodamine B.
The effects of the enhancer arginines on maximum (S max ) drug-stimulated ATPase activities are shown in Fig. 6. The effects on S 50 are shown in Fig. 7. Many of the enhancer arginines had detrimental effects on verapamil interactions with P-gp. Little or no verapamil-stimulated ATPase activity was observed with mutants I868R (TM10) and T945R (TM11) (Fig.  6A). These two mutants were the only ones that exhibited no detectable stimulation of ATPase activity with any of the drug substrates (Fig. 6). Reduced verapamil-stimulated ATPase activity (less than 50%) was also observed with mutants A302R (TM5), L339R (TM6), F728R (TM7), G872R (TM10), V873R (TM10), F942R (TM11), Q946R (TM11) V982R (TM12), and S993R (TM12). It appeared that a reduction in apparent affinity for verapamil was responsible for the reductions in activity with these mutants because most showed an ϳ10 -20-fold reduction in S 50 for verapamil (Fig. 7A). One mutant (V873R), however, showed an ϳ5-fold increase in apparent affinity for verapamil with an S 50 of 9.2 Ϯ 2.8 M (Fig. 7A).
Two mutations in TM12, however, appeared to cause an increase in apparent affinity for colchicine. Mutant M986R showed half-maximal activation with 142 Ϯ 10 M colchicine, and A987R showed half-maximal activation with 330 Ϯ 40 M colchicine (compared with 880 M for wild-type P-gp).
The effects of enhancer mutations on P-gp interactions with rhodamine B are shown in Figs. 6D and 7D. Mutant F983R (TM12) showed an ϳ3-fold increase in S max with rhodamine B compared with wild-type P-gp (Fig. 6D). Mutants F770R (TM10) and Q990R (TM12) also showed ϳ1.5-2-fold increases in S max with rhodamine B (Fig. 6D). No detectable rhodamine B-stimulated ATPase activity was observed for mutants I868R (TM10), F942R (TM11), T945R (TM11), Q946R (TM11), V982R (TM12), and S993R (TM12) (Fig. 6D).  None of the active mutants showed large reductions (less than 2-fold) in apparent affinity for rhodamine B (Fig. 7D). Instead, the most pronounced effect of the enhancer arginines was to increase the apparent affinity of P-gp for rhodamine B. Mutants F336R (TM6), F983R (TM12), M986R (TM12), A987R (TM12), and Q990R (TM12) required 4 -20-fold less rhodamine B for halfmaximal activation of ATPase activity (Fig. 7D). The results show that most mutants exhibited at least partial activity with hydrophobic substrates despite the presence of a bulky positively charged side chain. It should also be noted that all of the enhancer arginine mutants yielded mature P-gp (170 kDa) as the major product in the wild-type background (see Fig. 2, C and D).
Effect of Inhibitory Enhancer Arginines on P-gp Interactions with Vinblastine-Mutants I868R (TM10) and T945R (TM11) showed no drug-stimulated ATPase activity with any of the four substrates that were tested (Fig. 6). To test whether mutations I868R (TM10) or T945R (TM11) affected P-gp-drug interactions, we performed disulfide protection assays using mutant L339C (TM6)/F728C (TM7) and vinblastine as described previously (26). Mutant L339C (TM6)/F728C (TM7) is useful because the cysteines at positions 339 and 728 will cross-link when treated with the M14M thiol-reactive crosslinker. Cross-linking can readily be detected because crosslinked P-gp migrates slower on SDS-PAGE gels. The drug substrate vinblastine will protect the mutant from cross-linking. For example, we previously found that introduction of the L65R enhancer mutation into mutant L339C (TM6)/F728C (TM7) reduced its apparent affinity for vinblastine by ϳ60-fold (26).

DISCUSSION
We hypothesized that the mechanism that allows P-gp to recognize and transport many different hydrophobic drugs is that it contains multiple flexible drug-binding sites in a common drug-binding pocket and that drug binding occurs by an induced fit mechanism (29). The results of arginine-scanning mutagenesis of the TM segments are consistent with this mechanism. The drug-binding pocket and the translocation pathway appear to be quite flexible because introduction of the large positively charged arginine side chain at 89 of the 237 positions in the TM segments did not inhibit maturation (Figs. 3 and 4). There were 38 arginines that enhanced folding of P-gp. The enhancer arginines appeared to promote folding through a global effect rather than simply compensating for the local effects of the G251V mutation because they could enhance folding of wild type as well as other processing mutants (Fig. 2, C-F).
The various locations of the 38 enhancer arginines suggest that it is unlikely that each enhancer arginine would be positioned to hydrogen bond with a residue in another TM segment without some adjustment of the helices. Therefore, an induced fit mechanism could also explain enhanced folding of P-gp by the enhancer arginines.
The ability to accommodate the charged side chain of arginine at many positions within the drug-binding pocket is consistent with the finding that the drug-binding pocket is accessible to the aqueous environment (38) and is not totally hydrophobic as determined by fluorescence energy transfer (39).
In a previous study, we also showed that it was possible for an enhancer mutation to promote folding of P-gp processing mutants with processing mutations located throughout the molecule (12). The I306R (TM5) enhancer mutation could promote maturation of P-gp processing mutants containing mutations in the first intracellular loop (G251V), TM5 (G300V), the linker region connecting the two halves of the molecule (P709A), TM7 (G722A), the third intracellular loop (F804A), and NBD2 (L1260A). In addition, we previously showed that the M68R (TM1) (10), L65R (TM1), T199R (TM3), and F343R (TM6) (26) enhancer arginines also promoted maturation of processing mutants such as ⌬Y490 and ⌬NBD2. These observations suggest that processing mutations and enhancer mutations are affecting global rather than local folding events in P-gp. The results supported our model that explained the effects of processing mutations, drug substrates, and arginine enhancer mutations on folding of P-gp (26). The model predicted that processing mutations act as thermodynamic hurdles in the folding process and trap P-gp as a folding intermediate with incomplete domain-domain contacts and incomplete packing of the TM segments. Drug substrates or arginine enhancer mutations promote contacts between the TM segments and induce P-gp to complete the folding steps to yield a functional transporter. In a previous study, it was found that enhancer arginines appeared to promote folding through hydrogen bond interactions with residues in other TM segments (10).
Only two of the enhancer arginine mutants (I868R (TM10) and T945R (TM11)) showed little drug-stimulated ATPase activity with any of the drug substrates. The other 22 enhancer arginines (Figs. 6 and 7) showed drug-stimulated ATPase activity despite the presence of a positive charge. Most (21 of 22 mutant) showed altered activity (at least 3-fold) with at least one of the of the drug substrates. Some of these mutants had altered activity with one drug substrate but not with others. These results support the hypothesis for multiple flexible drugbinding sites. For example, mutants A302R and L339R caused over 10-fold reduction in apparent affinity for verapamil but had little effect on the apparent affinity for colchicine or rhodamine B (Fig. 7). Mutant F770R exhibited an ϳ5-fold increase in activity with colchicine but had little change with vinblastine (Fig. 7). Mutant F336R had undetectable vinblastine-stimulated ATPase activity, but its apparent affinity for rhodamine was increased ϳ10-fold. By contrast, mutant M986R had little effect on vinblastine-stimulated ATPase activity, but its apparent affinity for rhodamine B was increased by ϳ10-fold. These results show that different drug substrates interact with different residues in the drug-binding pocket and that there could be common residues for some drug substrates.
When the results of arginine-scanning mutagenesis of the TM segments were modeled as helical wheels, it was found that most TM segments appeared to have distinct hydrophobic and hydrophilic faces (Figs. 3 and 4). The hydrophobic faces would be predicted to face the lipid bilayer, whereas the hydrophilic face would be predicted to lie within the drug translocation pathway/drug-binding pocket. P-gp appears to contain an aqueous drug-binding pocket because cysteines introduced at sites within the chamber could be labeled with charged thiolreactive compounds (38). The arginine enhancer mutations that were identified in the TM segments were all located on hydrophilic faces. TM6, however, did not appear to contain a hydrophobic face because arginines introduced at nearly all positions did not inhibit maturation (Fig. 3). Therefore, TM6 does not appear to be anchored in the lipid bilayer but resides in the predicted aqueous-filled pocket. The location of TM6 within the aqueous chamber may explain why this TM segment appears to be quite mobile. For example, a disulfide cross-linking study suggested that ATP hydrolysis induced rotation (40) and tilting (41) of TM6. This TM appears to be a key segment that couples ATP hydrolysis to conformational changes in the TMDs (41)(42)(43).
The predicted locations of the enhancer mutations in the model of the closed conformation (Fig. 1D) of human P-gp are shown in Fig. 9A. For clarity, only the region of each TM segment that contained the majority of the enhancer mutations is shown and is viewed from outside the cell. Because some TM segments contained multiple enhancer mutations, only some are shown for clarity. It was found that hydrophilic faces of all the TM segments pointed toward the predicted drug-binding pocket. Most of the TM segments also appeared to contain a hydrophobic face because insertion of arginines into these regions inhibited maturation of the processing mutant (Figs. 3  and 4). The locations of the residues predicted to lie on the hydrophobic faces as identified by arginine-scanning mutagenesis are shown in Fig. 9B. It was found that all of the predicted hydrophobic faces of the TM segments were located on surfaces that would reside within the lipid bilayer. Therefore, there was a good correlation between the arginine-scanning mutagenesis results and the model of the structure of the TM segments of human P-gp.
The crystal structure of mouse P-gp (ATP-binding cassette B1a) was recently determined in the open conformation where the NBDs are far apart and the TMDs are facing inwards (8). Residues in mouse P-gp that are equivalent to those in human P-gp (Fig. 9, A and B) that were found to promote or inhibit maturation in the arginine mutagenesis studies are shown in Fig. 9 (C and D). The predicted orientations of all of TM segments except TM3 and TM5 were similar in the mouse and human P-gps (Fig. 9). In mouse P-gp, however, the orientations of TM3 and TM5 appeared to be opposite to the results observed with arginine-scanning mutagenesis. In both TMs, the hydrophilic and hydrophobic faces in mouse P-gp (Fig. 9, C and D) were in opposite directions to those in human P-gp (Fig. 9, A and B).
Biochemical studies, however, support the human models of P-gp for the orientations of TM3 and TM5 (Fig. 9, A and B). For TM3, the results of arginine-scanning mutagenesis suggested that residue Thr 199 was located in the drug-binding chamber because insertion of an arginine at this position promoted maturation of either the G251V or ⌬Y490 processing mutants (26). In the same study, it was shown that a mutant containing a cysteine at this position showed permanent activation of P-gp ATPase activity after modification with methanethiosulfonaterhodamine. Both of these observations are consistent with the human P-gp models (Fig. 9, A and B).
For TM5, the results of arginine-scanning mutagenesis are also in agreement with the models for human P-gp shown in Fig. 9 (A and B). Arginines introduced at positions 299, 302, 303, and 306 promoted maturation of human P-gp. These residues are predicted to be within the drug-binding pocket in human P-gp. In mouse P-gp, however, all of these positions are predicted to face the lipid bilayer. Previous biochemical studies support the human P-gp model because it was found that modification of a cysteine introduced at position 306 caused permanent activation of ATPase activity (12). In addition, cysteines introduced into TM5 (I299C) and TM8 (F770C) could be directly cross-linked when treated with oxidant at 4°C when thermal motion is reduced (44). These results are more compatible with the structures of human P-gp (Fig. 9, A and B).
Cysteine mutagenesis and disulfide cross-linking studies would suggest that human P-gp in cells is normally found in a closed conformation (Fig. 1D) because cysteines introduced into the NBDs could be directly cross-linked when the mutants were treated with oxidant at 4°C (45). The cross-linking study is also consistent with the projection structure of purified P-gp reconstituted into a lipid bilayer that shows close proximity of the NBDs (46). FRET analysis of hamster P-gp also showed close association of NBDs (47). In addition, kinetic studies on the human P-gp reaction cycle suggested that the NBDs do not undergo complete disassociation (48).
An explanation for the difference in the orientation of TM3 and TM5 in human and mouse P-gps is that they represent different stages in the reaction cycle. It has been demonstrated that there are many reaction intermediates in the P-gp transport cycle (49). The mouse P-gp structure may represent a reaction step that contains no bound ATP or ADP. It should also be noted that despite the high degree of amino acid identity between the mouse and human P-gps, they show quite different drug resistance profiles and sensitivity to modulators (50).
We predicted that arginine enhancer mutations promoted maturation of the P-gp processing mutant because they were located within the drug translocation pathway and mimicked the effects of drug rescue. This prediction is supported by the observation that arginine introduced at many of the positions that promoted maturation of mutant G251V P-gp also appeared to be involved in P-gp-drug interactions. For example, all of the positions that showed permanent activation of ATPase activity after modification of introduced cysteines with thiol-reactive drug substrates (L65C (TM1), T199C (TM3), I306C (TM5), F343C (TM6) (26), and F728C (TM7)) (22) also showed enhanced maturation when arginines were introduced at these positions into the P-gp processing mutant (this study and Ref. 26). Enhanced maturation was also observed with the F983R (TM12) mutation, a key residue required for interactions with flupentixols (51). In addition, it was found that ϳ95% of the positions in P-gp that showed inhibition of ATPase activity upon modification of introduced cysteines with the thiolreactive substrates methanethiosulfonate-verapamil or methanethiosulfonate-rhodamine (33,34) were located in the predicted drug-binding pocket of human P-gp (Fig. 9, A and B).
Mouse P-gp was also crystallized in the presence of two inhibitors of its ATPase activity (8). Residues predicted to interact with the two inhibitors or verapamil were identified. Comparison of the enhancer arginine residues with residues in mouse P-gp that are involved in drug binding showed that 16 of 32 drug-binding residues were identical.
The finding that P-gp folding defects could be corrected by introducing arginines at many locations throughout the TMDs suggests that it might be possible to correct folding defects at multiple locations in other mutant proteins associated with diseases. For example, defective folding and trafficking to the cell surface of mutants of the ATP-binding cassette transporters cystic fibrosis transmembrane conductance regulator (52) and SUR1 (53) cause the diseases cystic fibrosis and persistent hyperinsulinemic hypoglycemia of infancy, respectively. These mutant proteins show substantial function if they can be induced to fold and migrate to the cell surface. A goal to treat protein-folding diseases has been to mimic drug rescue of P-gp by identifying compounds that would directly interact with cystic fibrosis transmembrane conductance regulator or SUR1 mutants to promote folding and delivery to the cell surface (reviewed in Ref. 54). The results of arginine mutagenesis suggest that there may be many locations in the mutant proteins that could interact with compounds to correct defects in folding.
In summary, arginine-scanning mutagenesis provides biochemical evidence that all the TM segments surround an aqueous drug translocation pathway that contains multiple drugbinding sites. The mechanism for the ability of P-gp to