A Periplasmic Drug-Binding Site of the AcrB Multidrug Efflux Pump: a Crystallographic and Site-Directed Mutagenesis Study

Bacterial strains and plasmids used in this work are listed in Table 1. E. coli DH5α was used for construction and propagation of various plasmid constructs. Cells were grown at 37°C with agitation in Luria-Bertani broth (LB) (Difco) supplemented with ampicillin (100 μg/ml) (Sigma) when necessary.

Plasmid DNA was routinely prepared by using a QIAGEN Plasmid Miniprep kit (QIAGEN Inc.). Transformation of E. coli strain DH5α or HNCE1b with plasmid DNA was performed by heat shock treatment of competent cells. DNA sequencing was performed on double-stranded DNA templates by the use of universal primers or custom primers (Invitrogen). Nucleotide sequences were determined by Elim Biopharmaceuticals Inc. DNA alignments were obtained by using the BLASTN 2.02 program (3) provided by the National Center for Biotechnology Information.

The Asn109 residue of E. coli AcrB was changed into Ala by site-directed mutagenesis using a QuikChange kit (Stratagene) with the plasmid pUC151A (17). The mutated gene was sequenced completely to make certain that no other mutations were introduced during the PCR.

The native AcrB and N109A mutant AcrB proteins (both without “tags”) were purified following procedures similar to those described earlier (47, 49).

The N109A mutant AcrB was overproduced in E. coli C43(DE3) cells (21) by the use of the plasmid derived from pUC151A. Cells were grown in 10 liters of LB medium with 100 μg/ml ampicillin. Cells were disrupted with a French pressure cell. The membrane was collected and washed twice with high-salt buffer containing 20 mM sodium phosphate (pH 7.2), 2 M KCl, 10% glycerol, 1 mM EDTA, and 1 mM phenylmethanesulfonyl fluoride and once with 20 mM HEPES-NaOH buffer (pH 7.5) containing 1 mM phenylmethanesulfonyl fluoride. The purified membrane protein was solubilized in 1% (wt/vol) N-dodecyl-β-d-maltoside. Insoluble materials were removed by ultracentrifugation at 370,000 × g. The extracted protein was purified with hydroxyapatite, Cu²⁺ affinity (49), and G-200 sizing columns.

Liganded AcrB crystals were grown by cocrystallization with ligands as described earlier (48). Thus, crystals of the N109A mutant were grown by sitting-drop vapor diffusion at 25°C. A protein solution containing 24 mg/ml N109A protein in 20 mM Tris (pH 8.0)-0.1% N-dodecyl-β-d-maltoside-20 mM dithiothreitol was mixed with an equal volume of a reservoir solution containing 8 to 10% polyethyleneglycol 3000, 40 mM potassium citrate (pH 6.5), and 10% glycerol. To ensure that more than one ligand molecule would bind to the transporter, the ligand-to-protein (trimer) molar ratio in all complex-crystal drops was set to 2:1. This corresponds to a total ligand concentration of about 120 μM. Crystals appeared in the drops within 4 days. Cryoprotection was achieved by raising the glycerol concentration stepwise up to 35% with a 5% increment in each step.

All X-ray intensity data sets were collected at the Advanced Light Source (Beamline 8.2.2) at a cryogenic temperature (100°K). The diffraction data were processed with DENZO and scaled with SCALEPACK (30). The crystals of the N109A mutant took the R32 space group with unit cell dimensions listed in Table 1. Initially, the overall structures of the N109A mutant were determined by molecular replacement using a MolRep program (43) in the CCP4 package. The wild-type AcrB structure (1OY6) with residues 7 to 498, 513 to 710, 712 to 859, and 869 to 1,036 was used as a search model. Before refinement, 5% of all data were set aside for cross-validation (4). The model refinements were performed using the program Refmac (26) in the CCP4 package, and model rebuilding was conducted using the program O (12).

Oligonucleotide primers engineered with SalI and SmaI restriction sites were used to amplify, by PCR, the acrA wild-type gene under standard conditions using Pfu Ultra (Stratagene). The amplified DNA was purified, digested sequentially by SalI and SmaI restriction enzymes (New England Biolabs Inc.), and ligated into a similarly digested pAcrB plasmid (8) so that the acrA gene would be located upstream of acrB, with an intergenic distance of 30 nucleotides, and so that both genes would be under the control of the lac promoter of vector pSport1.

Point mutations were introduced into the acrB gene carried by plasmid pAcrB by the use of sense and antisense mutagenic primers in a one-step PCR procedure. Approximately 10 to 20 ng of plasmid DNA from E. coli served as the template for PCRs. The PCR mixture was composed of 1 μM of each primer, 6% dimethyl sulfoxide, each deoxynucleoside triphosphate at a concentration of 250 μM in 1× PCR buffer, and 1.25 U of Pfu DNA polymerase (Stratagene) in a final volume of 50 μl. The PCR program was 1 min at 95°C followed by 20 cycles of 30 s at 94°C, 30 s at 45°C, and 8 min at 68°C and an 8-min final extension at 68°C. Then, the PCR product was treated with DpnI restriction enzyme (New England Biolabs Inc.) to digest template DNA and to allow the enrichment of newly synthesized DNA containing the desired mutation. The digested DNA was used for transformation of E. coli strains DH5α by the cold CaCl₂ procedure (33). Plasmid DNA was extracted, sequenced to ensure the presence of the desired mutation. The intact acrA gene was then inserted ahead of the mutated acrB gene to produce pAcrAB-like plasmids, which were used for trans complementation of strain HNCE1b.

The susceptibilities to antimicrobial agents of E. coli strains (inoculum, 500 cells/ml) harboring pAcrAB-derived plasmids were determined by the twofold dilution method with LB agar medium containing 0.1 mM IPTG (isopropyl-β-d-thiogalactopyranoside). Bacterial growth was examined after 18 to 24 h at 37°C. Each assay was repeated at least five times to ensure the reproducibility of the results.

Cells were grown overnight at 37°C, diluted 100-fold in LB, and grown 1 h before the addition of 0.1 mM IPTG. When the cell suspension optical density at 600 nm reached 0.7 to 0.9, cells were harvested at 5,000 × g for 10 min at room temperature, washed once with 50 mM sodium phosphate buffer (pH 7.2), and resuspended in the same buffer at a cell optical density at 600 nm of 0.4. The accumulation of ethidium bromide was assayed as described by Li et al. (16).

Exponential-phase cells grown in LB were harvested, resuspended in 50 mM sodium phosphate buffer (pH 7.2), and broken by sonication. Unbroken cells were removed by centrifugation at 10,000 × g for 10 min. Total membrane extract was collected by ultracentrifugation at 100,000 × g for 30 min at 4°C. Proteins were resolved in sodium dodecyl sulfate (SDS)-7.5% polyacrylamide gel electrophoresis. Alternatively, proteins were transferred electrophoretically to nitrocellulose membrane (Bio-Rad) in Tris base (20 mM)-glycine (150 mM)-methanol (20%) for Western blot analysis. Binding of primary polyclonal antibody anti-AcrA or anti-AcrB (49) was detected with an alkaline phosphate-conjugated anti-rabbit secondary antibody (Sigma). Protein visualization was performed with nitroblue tetrazolium and 5-bromo-4-chloro-3-indolyl phosphate (5).

The coordinates for the protein structures have been deposited in the Protein Data Bank (PDB) as 1T9T (N109A mutant apoprotein), 1T9U (N109A-ciprofloxacin complex), 1T9V (N109A-rhodamine 6G complex), 1T9W (N109A-nafcillin complex), 1T9X (N109A-ethidium complex), and 1T9Y (N109A-Phe-Arg-β-naphthylamide complex).

After the binding of the drug molecules to the central cavity (47, 48), the most direct route for drug extrusion seemed to be through the central pore, which is formed by the pore helices in the periplasmic domain of AcrB, residues 102 to 115. However, in AcrB these helices make a direct coiled-coil interaction at the center of the three subunits (23, 48), and no channel is apparent within the “pore.” Side chains of a few amino acid residues protrude prominently into the center. Among these residues, we thought that the three Asn109 side chains might interact with each other. We therefore mutated this residue into alanine by site-directed mutagenesis. The mutant AcrB was functional in drug efflux, when expressed, together with AcrA, from a pUC19-derived plasmid (17) or from a pSPORT1-derived plasmid. The latter results are shown in Table 2 (top two rows). As can be seen, the levels of resistance generated by the mutant protein were nearly identical with those caused by the unaltered AcrB, except for tetracycline, acriflavine, and SDS. When expressed from pUC19-derived plasmid, both the mutant and the unaltered AcrB raised the MIC of nafcillin (Naf) from <0.2 μg/ml in the AcrB-deficient host strain to >500 μg/ml.

We solved the three-dimensional structure of the mutant AcrB, either alone or in a complex with five different ligand molecules, by X-ray crystallography through molecular replacement. Refinement resulted in R_work values between 25.6 and 28.0%, with a maximal resolution of 3.2 to 3.8 Å (Table 3). The overall structure of the mutant protein was very similar to that of the wild-type AcrB. In the mutant AcrB, an opening of 3 Å was created in the central pore according to surface plots, whereas the pore was totally closed in the wild-type AcrB (not shown). The loop (residues 860 to 868) that connects the C-terminal end of the C-terminal loop domain to TM helix 8 could not be modeled in the wild-type AcrB (23, 48). In the crystals of N109A mutant AcrB, however, these residues could be introduced into the model with the help of simulated annealing omit maps. The TM helices following this loop (TM8 through TM12) were displaced in the mutant protein, and changes in tilt increased the displacement to as much as 3.5 Å close to the C terminus (Fig. 2).

The mutant protein was crystallized in the presence of five different agents, ciprofloxacin (Cip), rhodamine 6G (R6G), ethidium bromide (Et), Naf, and Phe-Arg-β-naphthylamide (MC), an inhibitor of AcrB (32). The crystal structures showed that these ligands, in addition to the central cavity, bound to a second, novel binding site in the periplasmic domain, within the deep external depression formed by the C-terminal periplasmic loop domain. The structure of Cip-N109A AcrB complex is shown as an example in Fig. 1B. The simulated annealing omit maps of Cip and R6G, bound to the central cavity and periplasmic site, are shown as examples in Fig. 3.

The binding of the five different ligands to the periplasmic pocket is shown in Fig. 4. All drugs bind to the same region of the pocket, although Cip and R6G molecules extend more toward the membrane and Naf further away from it. The “right” wall (looking into the center of the trimer with its periplasmic domain on top, as in Fig. 1) and a part of the bottom of the binding pocket are formed by the loop (residues 663 to 680) that connects the two subdomains (PC1 and PC2) (23) of the C-terminal periplasmic loop. A short section of another loop (residues 564 to 568), connecting TM7 helix with the periplasmic domain, also functions as the bottom of the pocket. The wall on the opposite side is made of a short β-strand (Cβ6 of reference 23) (residues 714 to 722) and another β-strand (Cβ14 of reference 23) (residues 826 to 833). Four proline residues, Pro718, Pro669, Pro565, and Pro833, surround the pocket.

Figure 5 shows, as an example, the details of the Cip binding in the predominant conformation we have modeled. The limited resolution makes detailed interpretation difficult. However, the ligand binding seems to be rather loose. Thus, there are no clearly identifiable H-bonding interactions, and the bulk of amino acid side chains appear to be several angstroms away from the ligand. Nevertheless, in the model there seem to be some electrostatic interactions, which are facilitated by the apparent rotations of some side chains upon drug binding. Thus, in our model, the side chain amide nitrogen of Asn719 is close (3.7 Å) to the carboxylic acid oxygen of Cip, and the side chain carbonyl oxygen of Asn667 is close (3.3 Å) to the presumably protonated piperazine nitrogen of Cip.

The drug binding also produces significant displacement of several segments of the backbone. There is a large movement of the segment between Val107 and Leu113 (including the mutated residue Ala109), apparently because the segment between Gln104 and Gln108 becomes helical as a result of drug binding. Some of the side chain atoms appear to become displaced by nearly 3 Å upon drug binding. In addition, the segment between Gly861 and Gly870, just preceding TM helix 8, shows a slight backbone displacement and side chain movement up to 3 Å.

Similar interactions often involving the same set of residues appear to occur with the binding of R6G and Et in our model (not shown). Surprisingly, no acidic residue that would neutralize the positive charges of these dyes was found within 6 Å of the ligand (except Asp566, which is close to the benzoic acid moiety of R6G but more than 11 Å away from its amine nitrogens), although the partially negative π-electron cloud of the phenyl ring of Phe664 is about 4 Ε from these dyes. The charged atoms may also interact with the backbone oxygens (possibly of Leu828, Gly829, and Ser715), as well as the side chain oxygens of some residues (possibly Ser715 and Gln830). Interestingly, positive charges of pentamidine were recently found to interact exclusively with partial negative charges of oxygens in the backbone and the hydroxyl groups in the side chains in the regulatory protein QacR (25).

Compared with the binding of Cip, R6G, and Et, the binding of a negatively charged antibiotic, Naf, seemed to involve more electrostatic interactions in our model (not shown). Phe664 is the only hydrophobic residue close to the ligand. One of the partially positively charged hydrogen atoms of the phenyl ring of this residue is very close to the sulfur atom in the penicillin nucleus. The other drug-protein interactions also appear largely electrostatic or dipolar. The carboxylate group of the penicillin nucleus of Naf appears to interact with the backbone nitrogen of Gly720 as well as the side chain amide nitrogen of Gln577. These are reminiscent of the neutralization of ligand charges by formally uncharged atoms mentioned above. The carbonyl oxygen of the β-lactam ring is not far from the backbone nitrogens of Pro718 and Asn719.

In the binding of the inhibitor, MC, Phe664, and Arg717 appear to provide hydrophobic contacts to the naphthylamide and phenylalanine moieties, respectively, of MC. Again it is surprising that there are no acidic residues near this ligand with two positive charges, and participation of backbone oxygens (for example, those of Pro718 and Leu828) may be involved. The amino nitrogen of the Phe residue of MC is also not far from the side chain oxygen of Ser715. Possibly the guanidium group of the Arg residue in MC interacts with the π-electron cloud of the naphthylamide moiety within MC. Coupled with the inhibitor binding in the periplasmic domain is an outward movement of residues 650 to 659 located in the α-helix (Cα3 of reference 23) preceding the substrate-binding loop of residues 663 to 676. This shift is apparently caused by the 10-degree rotation (or partial melting) of a large portion of this helix. The movement is significant, with the Gln657 α-carbon moving 3.6 Å and some of the side chain atoms moving nearly 8 Å. We suspect that the movement of these residues, which is not seen in the other ligand complexes, may contribute to the inhibition of the transporter activity.

Drug molecules are also bound to the central cavity. However, for those ligands previously examined for their interaction with the wild-type AcrB protein (48), including R6G, Et, and Cip, the positions of binding appeared to be somewhat different from those seen earlier, although the significance of this difference is made somewhat unclear by the limited resolution of our crystals. Thus, these ligands, as well as Naf and MC, seem to bind close to the center of the cavity-lining wall of each protomer, in contrast to the position between protomers in the wild-type protein (Fig. 6; also compare Fig. 1B with Fig. 1A). In other words, most ligands are located close to the top of TM helices 5 and 6, although R6G and MC appear to bind to the location somewhat to the left of the others (as viewed in Fig. 6, from the center of the cavity), involving more strongly the participation of residues at the end of TM helix 3 (such as F386 and F388).

Residues within about 6 Å of the bound Cip are shown, as an example, in Fig. 7. The binding is very loose, yet in comparison with the wild-type structure, the binding with the mutant protein appears to involve larger number of residues, including Phe386, Phe388, Phe459, and Arg468.

The binding of Et and Naf (not shown) was similar to the Cip binding. R6G was found at a position similar to Et bound to the wild-type AcrB (48). Thus, the three R6G molecules are only ∼3 Å apart from one another, indicating that these ligands interact strongly with each other. For the inhibitor MC (not shown), Phe386 appears to provide hydrophobic contact. The three MC molecules also appear to interact strongly with each other here, as each of them is separated by only about 5 Å from the others. In our model, the MC molecules appear to be stretched vertically, in contrast to the highly bent conformation found in the periplasmic binding site.

To confirm that the ligand binding observed had functional significance, we mutated the residues near the ligand-binding sites and measured MICs in strains expressing these site-directed mutants of AcrB. Initial studies were carried out with recombinant plasmids containing only the mutated acrB genes. These experiments, however, often gave nonreproducible results. Overexpression of AcrB alone, without the parallel overexpression of AcrA, apparently often leads to variable drug susceptibility patterns, possibly because of the misfolding of the membrane protein, AcrB (H. I. Zgurskaya, personal communication; O. Lomovskaya, personal communication). To avoid these complications, in the following experiments we always used recombinant plasmids expressing both AcrA and AcrB from the same promoter, a strategy pioneered by other laboratories (for an example, see reference 40).

As described earlier, the mutant N109A protein produced nearly wild-type levels of resistance to most inhibitors, except tetracycline, acriflavine, and SDS (Table 2), when expressed together with AcrB in the ΔacrA::cat ΔacrB::kan ΔacrD host strain HNCE1b. Thus, the mutant transporter was largely functional. These results can be compared with the recent study (24) of the function of N109C mutant AcrB, which reported that the resistance to tetracycline and acriflavine was decreased. With the N109C mutant, however, decreased resistance to erythromycin and unaltered resistance to SDS are reported; it is unclear whether this difference between N109A and N109C is related to the spontaneous disulfide cross-linking between protomers seen in the latter mutant (24).

We converted into alanine those residues that appear to be close (usually within 6 Å) to the ligands in the structural models. These include D566, which forms the “bottom” (i.e., the end closest to the membrane surface) of the binding pocket, as well as F664 and F666, which are parts of a periplasmic loop that connects the AcrB PC1 and PC2 domains and form the “right” wall of the pocket (viewed from outside the trimer as in Fig. 4), and S715 and R717, which form the “left” wall. We also included E673, which is close to the bottom of the pocket, although it seems somewhat more distant from the ligands in our model. Finally, as a control, we mutated W859, which is located in the periplasmic domain in the cβ15 segment preceding TM8 but outside the binding pocket. Most of the mutations were generated in the background of both the wild-type acrB gene and the N109A mutant acrB. We compared the resistance levels caused by the various double-mutant proteins with those generated by the parent, single mutant N109A. As seen in Table 2, conversion of these residues to alanine strongly decreased the resistance to various agents compared with the results seen with the N109A parent protein except D566 and S715. With many mutations (see F664, F666, and E673) resistance to practically all of the agents was compromised, whereas with R717 a strong effect was seen only with novobiocin. These results suggest strongly that the residues surrounding this periplasmic binding site are required for the transport, and presumably the binding, of substrates, at least in the N109A mutant AcrB transporter. We note that all of these mutant proteins were produced at levels close to that of the parent proteins (Fig. 8A). Although there were small variations in the level of expression, these are not expected to produce large (fourfold or larger) differences in MIC as noted in Table 2, because calculations using the parameters derived in reference 38 indicate that decreasing tetracycline MIC from 8 to 2 μg/ml requires the loss of more than 80% of the transporter if its specific activity is unaltered.

We further examined whether the same residues are involved in the transport catalyzed by the wild-type transporter. As seen in Table 2, conversion to alanine of those residues found to be important in the N109A background, including F664, F666, E673, and R717, resulted in significant decreases of resistance to at least a few agents. Interestingly, the R717A mutation, which decreased only the novobiocin MIC in the N109A background, decreased the MIC of both novobiocin and acriflavine in the wild-type background. We point out also that the W859A mutation, used as a negative control, did not alter the MIC of any agents. Again, all transporters were expressed to a level similar to that of the wild-type protein (Fig. 8B).

The wild-type AcrB crystal structure in the presence of four different substrates (R6G, Et, dequalinium [Dq], and Cip) (48) suggested that residues F386, F388, F458 and F459 may be involved in drug binding. We examined the role of these residues, which are located at the wall of the upper part of the central cavity, by converting them into alanine.

All of the single substitutions resulted in transporters that retained most of the activity towards the tested molecules (Table 2, footnote b), although twofold decreases in MICs were frequently seen for tetracycline, erythromycin, Dq, and acriflavine. These decreases are likely to be significant, as the MIC determination was repeated at least five times and reproducible values were obtained. Similar results were obtained with double substitutions F386A/F388A or F458A/F459A (Table 2, footnote b). However, with the quadruple-substitution mutant F386A/F388A/F458A/F459A, there was a fourfold decrease in resistance to taurocholate and acriflavine (Table 2). As shown in Fig. 8A, Western blot analysis showed that all the mutant proteins were expressed at levels comparable to that of the wild-type proteins expressed by pAcrAB (Table 1).

Charged residues are known to play an important role among multidrug transporters. Residue K29, located close to the internal end of the vestibule, was found within 6 Å of the ligands R6G and Cip (48). Substitution of residue K29 with the neutral residue alanine resulted in a significant decrease in resistance to acriflavine and a small, but probably significant, twofold decrease in MICs of novobiocin, tetracycline, erythromycin, Et, and Dq (Table 2).

We (48) pointed out the possible electrostatic interaction for the dicationic disinfectant Dq with the acidic amino acid residue D101, located in the ceiling of the AcrB central cavity. As shown in Table 2, when D101 was replaced with the neutral amino acid alanine it caused only a modest, twofold reduction in Dq MIC. There was also a small decrease in the MIC of novobiocin and erythromycin. However, the strain expressing this mutant AcrB showed increased susceptibility to acriflavine and tetracycline. The latter results are consistent with a recent report (24) of a study in which the authors replaced residue D101 with the uncharged amino acid cysteine and found decreased MICs of acriflavine, tetracycline, chloramphenicol, and erythromycin. Western blot analysis showed that D101A and K29A mutant proteins were expressed at levels comparable to that of the wild-type AcrB (Fig. 8). Although charged residues within the TM helices are known to be crucial in drug efflux (1, 10), these results suggest that such residues outside the TM helices may also play significant roles in the efflux process, presumably in ligand-binding step(s).

To confirm the MIC results, we examined the intracellular accumulation of Et into intact cells expressing various constructs of AcrB. While the ΔacrA::cat ΔacrB::kan ΔacrD E. coli strain HNCE1b (pAcrB) rapidly accumulated Et owing to the virtual absence of its active efflux, HNCE1b strain trans-complemented with the wild-type acrAB genes (plasmid pAcrAB) showed a much slower accumulation, presumably as a result of efflux (Fig. 9). As expected, strains expressing mutant AcrB proteins with substitutions F664A/N109A (Fig. 9B) and F666A/N109A (Fig. 9A) showed higher rates of Et accumulation compared to that with the wild-type AcrB, while strains with AcrB containing single substitution F664A or F666A accumulated the dye somewhat less rapidly. These results show that alterations of F664 or F666 in the periplasmic binding site decrease the Et pumping activity of AcrB and that this defect becomes even more serious when an N109A mutation exists in the background.

We mutated the Asn109 residue, whose side chain protrudes into the narrow central pore of the periplasmic domain of AcrB (Fig. 1A), into alanine, with a one-carbon side chain. Although this did not produce a wide pore that would accommodate the ligands easily, the drugs now bound to a new site, corresponding to the external depression of the periplasmic domain, in addition to the wall of the central cavity (Fig. 1B). This may be important, because the large periplasmic domains of the RND transporters MexB, MexD, AcrB, and AcrD have been implicated in substrate recognition and specificity by using chemical mutagenesis (19, 20) or by engineering pump chimeras, first by Elkins and Nikaido (8) and Tikhonova et al. (39) and later confirmed by Eda et al. (7), and yet periplasmic substrate-binding sites have not so far been observed by crystallography.

Drug binding to the periplasmic site seems to involve a larger number of amino acid side chains (Fig. 5), in comparison with the binding to the central cavity of the wild-type AcrB (48). Furthermore, there are likely electrostatic interactions between the atoms of ligands and the protein atoms. Yet the binding pocket is quite loose. Similar expansive binding pockets and loose binding have been reported earlier for regulators of multidrug transport protein expression, based on X-ray structures with good resolution (34, 35, 51). For example, in the Et complex of the QacR protein (PDB file 1JTY) (34), only two atoms of the protein are within 3 Å of any atom of the ligand, and substantial portions of the aromatic rings of tyrosine and phenylalanine residues that sandwich the ligand are 5 to 6 Å away from the ligand, a situation similar to that found in this work. Neyfakh (27) argued persuasively that the tight binding of hydrophilic ligands is required only because of the energy needed to remove such ligands from the extensively H-bonded environment of water and that such tight binding is not needed for lipophilic ligands, because no such energy is needed to remove them from water.

In the central cavity, the positions of binding often appeared to be somewhat different from those seen in the wild-type AcrB (48). The extensive nature of binding sites for lipophilic ligands is often known to lead to the situation where a single ligand can bind in multiple positions or orientations. This was found for example with the pig odorant binding protein (45) and PXR xenobiotic resistance regulator (46). There are also data suggesting the presence of multiple ligand-binding sites in various multidrug resistance transporters (6, 14, 22, 31, 36, 44). Our finding of drug binding to different positions in the central cavity in the wild-type and the mutant AcrB and the inability to abolish activity by single (or even double) mutations also seem to emphasize the flexibility of substrate binding sites within this huge, possibly composite (47) space.

We showed in this study that the conversion, into alanine, of some of the residues surrounding the bound ligands in the periplasmic sites, such as F664, F666, and E673, made the AcrB pump less efficient with most substrates. Importantly, the same mutations also decreased the activity of the AcrB protein without the background N109A mutation, although the range of substrates affected became somewhat more limited (Table 2). These results suggest strongly that the periplasmic binding site newly discovered by the crystallographic study of N109A AcrB transporter is physiologically relevant, not only in the N109A mutant protein but also in the wild-type AcrB.

As with any site-directed mutagenesis study, we cannot exclude formally the possibility that the conversion of these residues into alanine affected the activity because of alteration of protein conformation rather than its direct effect on ligand binding. This seems unlikely, because transporter structures are amazingly flexible; for example, the complete cysteine mutagenesis of the lactose transporter LacY uncovered only 6 residues that were essential out of more than 400 residues (13). Furthermore, side chains of most of the residues altered in this study, both in the central cavity and in the periplasmic pocket, are extending out into hollow spaces in the apoprotein structure (23, 48) and are therefore unlikely to affect the global conformation of the protein. Nevertheless, we are currently trying to examine ligand binding kinetics to settle this issue in an unambiguous manner.

Since the binding of the drugs to the periplasmic site was seen only with the N109A mutant AcrB, its physiological relevance may be questioned. However, this second site is likely to participate in the normal pathway of substrate transport for the following reasons. (i) As discussed above, site-directed mutagenesis of the residues involved in the periplasmic binding decreased the MIC of various ligands. (ii) Mao et al. (19) isolated spontaneous mutants of P. aeruginosa MexD (a close homolog of AcrB) that acquired the ability to extrude dianionic β-lactams. Some of the mutants have amino acid substitutions right in the area of the periplasmic binding site, including N673T, which corresponds to Thr676 in AcrB, at the lower end of the substrate binding pocket (Fig. 5), as well as E89K, corresponding to Gln89 in AcrB, at the bottom of the periplasmic depression. Furthermore, random mutagenesis of P. aeruginosa MexB, an AcrB homolog, showed that R716H mutation alters the substrate specificity (20); this residue corresponds to Arg717 of AcrB, that seems to play a major role in the periplasmic binding (Fig. 5). The contribution of the periplasmic binding site to the normal export pathway is also consistent with the earlier genetic data mentioned above, showing that the substrate specificity of the RND pumps is determined largely by the periplasmic domain.

Our current data, however, do not allow us to conclude whether the two binding sites function sequentially or concomitantly. In the former case, we favor the hypothesis that the periplasmic binding occurs after the initial binding of ligands to the central cavity, because the periplasmic binding seems to be somewhat tighter than the binding to the central cavity. Possibly the slightly altered conformation of the N109A mutant transporter mimics one of the transient conformations of AcrB during drug transport, and the periplasmic binding in the N109A AcrB thus may mimic the ligand-binding step that occurs after the binding to the central cavity (48). It is not known how the ligands would travel between the central cavity and the periplasmic depression area. However, the periplasmic domain contains, in addition to the various cavities already mentioned, a possible passageway that we call a “fissure” (Fig. 1A) between the C-terminal loop domain on the outside and the N-terminal loop domain on the inside (23); possibly the ligands may migrate through these fissures. This is suggested by the mutant studies of Mao et al. (19), which also identified several residues at the entrance of the fissure as those affecting the substrate specificity of MexD, an AcrB homolog. We suspect that the fissure area may open up more when a drug molecule moves from the central cavity to periplasmic pocket. Alternatively, the substrates may not move between the two sites, and the simultaneous occupancy of the two sites may be necessary to activate the pump. In the mammalian P-glycoprotein, evidence suggests that there is an allosteric modulator-binding site that is separate from (but possibly overlaps) the two substrate-binding sites (6). In any case, these models are entirely speculative at present and there are no data that exclude other models (for example, ligand export through the central pore, as advocated by another laboratory) (24).