Structural Basis of Multiple Drug-Binding Capacity of the AcrB Multidrug Efflux Pump

See allHide authors and affiliations

Science  09 May 2003:
Vol. 300, Issue 5621, pp. 976-980
DOI: 10.1126/science.1083137


Multidrug efflux pumps cause serious problems in cancer chemotherapy and treatment of bacterial infections. Yet high-resolution structures of ligandtransporter complexes have previously been unavailable. We obtained x-ray crystallographic structures of the trimeric AcrB pump from Escherichia coli with four structurally diverse ligands. The structures show that three molecules of ligands bind simultaneously to the extremely large central cavity of 5000 cubic angstroms, primarily by hydrophobic, aromatic stacking and van der Waals interactions. Each ligand uses a slightly different subset of AcrB residues for binding. The bound ligand molecules often interact with each other, stabilizing the binding.

Multidrug efflux pumps are now known to be present in most living cells. In human cells, pumps such as P-glycoprotein prevent the entry of toxic molecules at the mucosal surface of the intestinal tract or at the blood-brain barrier. When overexpressed, P-glycoprotein makes cancer cells resistant to a wide range of anticancer agents (1). In bacteria, resistance to drugs is often associated with multidrug transporters functioning to decrease cellular drug accumulation (2, 3). Understanding the broad chemical specificity of these transporters has challenged our scientific community for decades. The central issue concerns the nature of substrate recognition. Important suggestions came from the study of ligand binding to soluble regulatory proteins of bacterial multidrug efflux pumps. The regulatory domain of the Bacillus subtilis BmrR regulator was crystallized in the presence of an inducer tetraphenylphosphonium (4). The binding was shown to involve mostly hydrophobic interactions with some electrostatic effects in a relatively large and loose binding site. The subsequent study of the Staphylococcus aureus QacR regulatory protein (5) showed conclusively that different ligands bind to different parts of a large and flexible binding site, each time again relying mostly on hydrophobic interactions. However, the structural basis of drug binding to transporters themselves has never been elucidated.

Escherichia coli AcrB is a transporter that is energized by proton-motive force and that shows the widest substrate specificity among all known multidrug pumps, ranging from most of the currently used antibiotics, disinfectants, dyes, and detergents to simple solvents (2, 3). Since its discovery (6), its properties have been studied by reconstitution (7), and it has served as a prototype bacterial multidrug transporter for studies of drug-transport mechanisms. Its ligand-free, trimeric structure was determined recently by x-ray crystallography in the laboratory of Yamaguchi (8), and this work has opened the approach to inquiry on the structural details of drug capture and transport. We have now solved the AcrB structure at a resolution of 3.5 to 3.8 Å in a complex with four different ligand molecules (Table 1). Our study shows that each ligand binds to different locations in the central cavity of the transmembrane domain of the trimer, not only confirming and extending the prediction of Brennan and co-workers on the nature of multidrug binding sites (5) but also showing small conformational changes in the protein upon substrate binding.

Table 1.

Data collection and crystallographic analysis.

Parameter Unliganded AcrB-R6G AcrB-Et AcrB-Dq AcrB-Cip
Space group R32 R32 R32 R32 R32
Cell constants (Å) a = b = 143.5, c = 519.6, α = β = 90, γ = 120 a = b = 144.8, c = 518.6, α = β = 90, γ = 120 a = b = 144.7, c = 517.5, α = β = 90, γ = 120 a = b = 144.8, c = 517.9, α = β = 90, γ = 120 a = b = 145.1, c = 517.2, α = β = 90, γ = 120
Resolution (Å) 3.70 (3.83-3.70) 3.63 (3.78-3.63) 3.80 (3.94-3.80) 3.80 (3.94-3.80) 3.50 (3.63-3.50)
Completeness (%) 100 (100) 100 (99.9) 99.0 (100) 98.0 (99.3) 100 (100)
Total reflections 204,935 260,474 280,672 212,408 596,530
Unique reflections 34,832 36,610 36,493 35,008 49,289
Rsym (%) 8.6 (44.4) 10.1 (45.0) 10.1 (44.2) 8.4 (53.8) 8.7 (54.6)
Rwork (%) 27.2 24.5 28.3 28.4 25.7
Rfree (%) 33.0 32.2 34.4 33.8 32.3

The structure of ligand-free AcrB (8) shows that it is a homotrimer of ∼110 kD per subunit. Each subunit contains 12 transmembrane helices and two large periplasmic domains (each exceeding 300 residues) between helices α1 and α2, and helices α7 and α8 (6). X-ray analysis of the overexpressed AcrB protein demonstrated that the three periplasmic domains form, in the center, a funnel-like structure and a connected narrow (or closed) pore (Fig. 1). The pore is opened to the periplasm through three vestibules located at subunit interfaces. These vestibules were proposed to allow direct access of drugs from the periplasm as well as the outer leaflet of the cytoplasmic membrane (8). The three transmembrane domains of AcrB protomers form a large, 30Å-wide central cavity that spans the cytoplasmic membrane and extends to the cytoplasm (8).

Fig. 1.

Structures of the trimeric AcrB transporter with bound ligands viewed from the side parallel to the membrane. (A) AcrB with three bound R6G molecules. The figure shows the transmembrane domain (inner and outer leaflets), the periplasmic domain, and the location of cavity, vestibule, pore, and funnel (8). The drugs are bound approximately at the level of the outer surface of the membrane lipid bilayer. (B) through (D) show the center of the side view in (A), with bound Et, Dq, and Cip molecules. This figure and Fig. 2 were prepared with PyMOL (22).

We purified the wild-type AcrB (9) (without any “tags”) and crystallized it in the presence of four structurally dissimilar agents, rhodamine 6G (R6G), ethidium (Et), dequalinium (Dq), and ciprofloxacin (Cip). The agents were added at 50 μM. In vitro reconstitution studies (7) showed that for most drugs, concentrations of ≥100 μM are needed to achieve 50% inhibition of the AcrB-catalyzed phospholipid-extrusion reaction. Thus, the drug concentrations used were modest in relation to the expected affinity of AcrB to these ligands.

The crystal structures illustrate that these ligands bind to various positions of the central cavity (Fig. 1 and Fig. 2), each using a different subset of residues (Fig. 2), thus greatly increasing the range of potential drug-protein interactions. (The electron density maps of the ligands are shown in Fig. 3.) The interior surface of the upper part of the cavity is surrounded with many hydrophobic residues, including 12 well-conserved phenylalanine residues, each protomer contributing Phe386, Phe388, Phe458, and Phe459. These residues are involved in drug binding, as seen below, suggesting that the binding is mainly governed by hydrophobic and perhaps also aromatic π–π interactions. Below we describe the structure of ligand-AcrB complexes. Another notable feature is that all these ligands bind more or less to the area that should be near the outer surface of the lipid bilayer, if, as seems likely, this cavity is filled with the bilayer.

Fig. 2.

The binding sites for the four ligands. Amino acid residues within 6 Å of the bound ligand molecules are shown. With the exception of (C), the view is approximately from the top (periplasmic side) of the trimer. Unmarked, primed, and double-primed residues, respectively, belong to the three subunits of the AcrB trimer. (A) R6G-binding site. (B) Et-binding site, including Phe388 that is slightly farther away (see text). (C) Dq-binding site. The side view shows the binding of the two quinolinium moieties within each Dq molecule (as in Fig. 1). The phenylalanine residues interacting with the bottom quinolinium moieties are shown even though they are 6.3 Å away from the ligand. Ile102 is not shown to avoid cluttering the figure. (D) Cip-binding site.

Fig. 3.

The final 2Fobs - Fcalc electron density maps (green) of the bound ligands contoured at 1.5σ. (Top left) Rhodamine 6G. (Top right) Ethidium. (Bottom left) Dequalinium. (Bottom right) Ciprofloxacin.

In the R6G complex structure, the loop regions between helices α3 and α4, and helices α5 and α6 form a ligand-binding domain in the large cavity facing the cytoplasm (Figs. 1 and 2). The ligand-binding cavity is extensive and occupies most of the upper half of the transmembrane region in the central cavity. Its total volume is ∼5000 Å3. The binding site in one subunit contains, within 6 Å distance from the ligand, Phe386, Ala385, Leu25, Val382, Lys29, and Phe386 from the neighboring subunit (the residue number with the prime symbol indicates a residue from another subunit). AcrB is a homotrimer, and each R6G molecule binds to the area corresponding to the border between the two neighboring subunits. These three identical sites face each other in the central cavity. They appear to assist and interact with each other, forming a large, single binding pocket to bind three drug molecules. Lys29 located at the vestibule in each subunit appears to point its side chain toward the ester group of each substrate. It is noteworthy that all other residues within 6 Å of the bound ligand are hydrophobic amino acids, suggesting the importance of the hydrophobic and possibly van der Waals interactions in drug binding. The three bound ligand molecules appear to interact with each other to stabilize the final configuration.

In the AcrB-Et complex, Et binds at a site that is distinct from but partially overlaps with the R6G binding site. Compared with the R6G binding, Et is bound ∼6 Å above the R6G binding site (compare Fig. 1, A and B), and its binding site is closer to the vestibule than that of R6G. The amino acid residues within 6 Å of Et include Phe386 and Ala385 from one subunit and Phe386 from the neighboring subunit; Phe388 is located slightly farther away (7 to 8 Å) (Fig. 2B). Et is close to Phe386 (∼4 Å). The phenyl moiety of Et is bound above Phe386, interacting closely with this residue. In this way, the bound Et molecule is sandwiched between two subunits of the transporter. As with R6G, most of these interactions are hydrophobic. In the binding pocket, the carbonyl oxygen of Phe386 is close to one of the amino groups of Et (3.6 Å), and may contribute to the neutralization of the formal charge of the substrate. The three bound Et molecules are ∼3.2 Å apart from one another, indicating that these ligands interact strongly in the central cavity.

The three bound Dq molecules occupy a large portion of the upper part of the binding pocket (Fig. 1C). Within 6 Å of a Dq molecule, we find two acidic residues, Asp99 and Asp101, as well as Ile102, close to the quinolinium moiety at the top (i.e., closer to the outer membrane) (Fig. 2C). Interaction with the two acidic residues appears to neutralize the formal positive charge of the quinolinium at the top. Asp99 is highly conserved (10), and this observation shows the importance of this residue in recognizing cationic drugs. This quinolinium group at the top forms a tight cluster with two other quinolinium moieties from the other two drug molecules. In contrast, the bottom part of the ligand, containing the second quinolinium moiety, is not within 6 Å of any residues on the cavity wall, although it is ∼6.3 Å away from the two phenylalanine residues (Phe386 and Phe386) from the two neighboring subunits. Thus the Dq contacts again are dominated by hydrophobic interactions, but the top quinolinium moiety has a strong electrostatic interaction. Also the drug-to-drug interaction appears to be quite tight in this complex.

As with R6G, the Cip molecule is bound between two protein subunits. Residues within 6 Å of the bound ciprofloxacin include Phe458, Phe459, and Lys29, Leu25 and Ala385 from the neighboring subunit (Figs. 1D and 2D). In the complex, again most of the interaction with the ligand is hydrophobic. The cyclopropyl group of the drug molecule interacts with Phe458 and Phe459 (the distance to the latter is 3.3 Å). From the neighboring subunit, the side chains of Leu25 and Lys29 are close to the ring nitrogen of the quinolone moiety. The fairly close distance (5 Å) between the ϵ-amino group of Lys29 and the electron-poor ring nitrogen [an analog, 1-methyl-4-quinolone, has a pKa of 2.46 (11)] also suggests some dipolar contributions. The carbonyl oxygen of Ala385 is also 5 Å away from one of the carboxyl oxygens of Cip, and these two groups may interact with the intervening water molecule. The geometry of the other carboxyl oxygen and the 4-carbonyl oxygen of Cip suggests that the carboxyl oxygen is protonated and that these atoms are parts of the six-membered, hydrogen-bonded ring that also includes C3 and C4 of the quinoline ring (12).

The details of these binding interactions confirm and extend the observation on the binding of various ligands to the regulatory proteins BmrR and QacR (4, 5). Thus the binding cavity is large, and each ligand binds to a different part of the cavity by using a different set of amino acid residues. There are, however, features that we did not expect from the studies of the binding proteins: (i) The binding cavity is extremely large and binds several molecules of ligands at the same time. With the multidrug transporter MdfA of E. coli, kinetic studies suggested the simultaneous binding of chloramphenicol and tetraphenylphosphonium (13). The present result indeed supports such a mechanism of binding. (ii) With the regulatory protein QacR, ligand binding involved the enlargement of the binding site by the expulsion of two tyrosine side chains (5). No drastic enlargement was seen in the already very large cavity of trimeric AcrB. (iii) With QacR, electrostatic interactions played a very important role in the binding of cationic dyes (5). With AcrB, the binding of the dyes R6G and Et appeared to involve mostly hydrophobic interactions rather than strong electrostatic interactions. This result is probably due to the large size of the binding pocket, possible interaction between the drug molecules, and the delocalized nature of the charge in these dyes. In contrast, for the cationic disinfectant Dq, with its more localized charges, the electrostatic interaction with acidic amino acid residues was apparent, at least with the quinolinium moiety on top, as described above. Unexpectedly, there was no obvious electrostatic stabilization of the quinolinium moiety at the bottom. In view of its location close to the surface of the presumed bilayer within the cavity (Fig. 1), however, this moiety (as well as other cationic moieties such as the protonated piperazine 4-N in Cip) may be stabilized by its interaction with the head groups of acidic phospholipids.

Comparison of the R6G-bound structure with the drug-free structure reveals that binding of the drug triggers a 1° rigid-body rotation in each subunit. The axis of rotation, passing through the side of each subunit, is approximately parallel to the plane of the lipid-bilayer. This motion enlarges the diameter of the periplasmic domain by ∼2.5 Å. The rotation is probably triggered by interactions between the ligand and the transporter and was seen also in the structure of Et-, Dq-, and Cip-bound AcrB (see reference 1 in supporting online material).

Previous biochemical studies have shown that AcrB exists as a complex with the periplasmic protein AcrA (14). These two proteins, presumably, interact specifically in the periplasm. The role of AcrA, which is absolutely required for transport, is still not clear. It is possible that the 1° rigid-body rotation of AcrB brings the periplasmic domain of AcrB closer to AcrA and that this triggers the conformational changes of these proteins that further trigger the drug transport. Hydrodynamic and electron crystallography studies of AcrA (15, 16) indicate that AcrA is an elongated, asymmetric protein. Such a flexible protein is ideal for amplifying transmembrane signal, because a small change at one end of the protein may eventually cause a large conformational change at the other end.

With the binding of R6G and Dq, we see a downward movement of a loop between residues 300 and 310 in the periplasmic domain, which causes the Arg307 side chain to flip downward and shift toward the center of the vestibule.

The drug molecules that are bound to the central cavity are probably pushed out through the periplasmic domains of AcrB, into the α-helical tunnel of TolC, and finally into the medium (3). The most direct route for drug extrusion may be through the central pore, which is formed by the pore helix in the periplasmic domain, residues 102 to 115 (6). In the unliganded as well as liganded AcrB, these residues make direct coil-coil interaction right at the center among the three subunits, and the pore is thus closed. Possibly the proton flux that follows the ligand binding produces a large conformational change of AcrB that leads to the opening of the pore and to the elevator-type mechanism of drug extrusion.

Recent genetic studies identified domains and residues that appear to be important in the substrate efflux in AcrB and its homologs. Thus, domain swapping (17, 18) and point mutation (19) studies both showed that the periplasmic domain plays a major role in determining the substrate specificity of these pumps. Yet most of the residues identified as parts of drug-binding sites in this study come from the transmembrane domain, with the exception of Asp99, Asp101, and Ile102. This is partly because drugs presumably traverse the vestibule, which primarily belongs to the periplasmic domain, to reach the cavity, and much discrimination may be exerted in the entry into and traversal through this channel. Also, the binding seen here represents only the first step of the efflux process, and drugs may interact with other sites before their final extrusion into the tunnel of the TolC protein.

The liganded structures show that the diverse chemicals were able to bind complementary polar and hydrophobic groups in the large central cavity in the efflux transporter firmly (but transiently, because the chemicals must be transported out). The observation that the ligands induce the same 1° rotation of the protein indicates that all participate in a thermodynamic shift toward the same conformational state. In this regard, that four different ligands induce a similar conformational change in the transporter, a finding that is consistent with the hypothesis (20) that the induced conformational state reflects evolutionary changes that select for amino acid residue interactions that optimize the kinetic and thermodynamic values needed for the conformational change.

It has been proposed earlier by one of us (2) that amphiphilic drug molecules partition partly into the outer leaflet of the cytoplasmic membrane, diffuse laterally within the leaflet, and eventually are captured by AcrB. The location of the binding site fits perfectly with this hypothesis. The hydrophilic head group of the drug would travel through the vestibules (8) (Fig. 1). Below the vestibule, there is a wide opening between the neighboring subunits in the upper part of the transmembrane domain. The hydrophobic part of the drugs may diffuse through this opening, presumably filled with the continuation of the outer leaflet of the lipid bilayer, to reach the binding sites seen in this study, close to the outer surface of the putative lipid bilayer within the cavity. The presence of the large substrate-binding cavity in AcrB is also reminiscent of the similarly large cavity in the low-resolution structure of mammalian P-glycoprotein (21).

Supporting Online Material

Materials and Methods

Fig. S1


References and Notes

View Abstract

Navigate This Article