Transport of drugs by the multidrug transporter AcrB involves an access and a deep binding pocket that are separated by a switch-loop

Binding of Minocycline and Doxorubicin by the AcrB Wild-Type Protein.

We describe here the crystallization and structural elucidation of AcrB/DARPin complexes in the presence of minocycline and doxorubicin based on 1.9-Å and 2.25-Å diffraction data (Table S1). AcrB/DARPin complexes were cocrystallized in the presence of minocycline (2.0 mM) or doxorubicin (0.4 mM), yielding orthorhombic crystals. The structures were refined to an R_free of 23.1% and 22.7%, respectively, and describe an asymmetric trimer comprising three different monomer conformations representing the consecutive states L, T, and O (15). Electron densities for minocycline or doxorubicin could be clearly identified in the hydrophobic deep binding pocket of the T monomer (Fig. 1 and Fig. S2). Whereas the position of minocycline is congruent with the structure of Murakami et al. (10), the orientation of the doxorubicin molecule inside the deep binding pocket differs substantially (Fig. S2 and Table S2). Notably, differences in crystallization conditions and resolution, and our use of crystallization chaperones (DARPins), are possible reasons for the observed distinctions. In the T monomer, another density located in the lateral cleft determined by the PC1 and PC2 subdomains of the periplasmic porter domain could be assigned to dodecyl-α-d-maltoside (Fig. S3). In the L monomer of the AcrB/doxorubicin cocrystal structure, but not in the AcrB/minocycline cocrystal structure, large densities were observed at the PC1/PC2 subdomain cleft (access pocket) and could be assigned to a twofold symmetric sandwich of stacked doxorubicin molecules (Fig. 2 and Fig. S4). The AcrB/doxorubicin cocrystal structure therefore reveals an asymmetric AcrB trimer containing three doxorubicin molecules: two doxorubicin molecules bound to the access pocket in the L monomer and one bound to the deep binding pocket in the T monomer (Fig. S5). This observation indicates a possible link between the initial binding of drugs in the L monomer and the subsequent binding of the drug in the deep binding pocket mediated by the L to T transition in the functional rotation cycle (Figs. S1 E–G and S5).

Coordination of doxorubicin by residues inside the AcrB access pocket includes hydrophobic interactions with F666 and L828 and charged/polar interactions with T676, R717, and N719 (Fig. S4). In a recent study, these coordination sites and further proximate residues (D566, F664, L668, and E673; Fig. S4) were Cys-substituted and accessible to the AcrB substrate Bodipy FL maleimide (28), matching the crystallographic data presented here (Fig. S4B). Furthermore, single alanine substitution of residues F664, F666, E673, and R717 markedly interfered with the ability of these AcrB variants to confer resistance toward various toxic compounds (27).

The L monomer is considered the access protomer (10, 11) involved in the initial stage of drug binding. The lateral access pocket at the periplasmic cleft confined by the PC1 and PC2 subdomains might be the initial entry point for drugs to the AcrB functional rotation cycle. However, many of the AcrB substrates are expected also to partition into the outer leaflet of the inner membrane because of their hydrophobic properties. A suggested entry pathway for drugs partitioned in the outer leaflet of the inner membrane is the hydrophobic groove in the transmembrane domain defined by TM8 and TM9 (6, 15), where in the high-resolution structures presented here, binding of the detergent and AcrB substrate dodecyl-β-d-maltoside is apparent (Fig. S6 A and B).

The TM8/TM9 groove defines the entrance to tunnel 1 (Fig. S1F) (12, 15) inside the periplasmic porter domain, which leads toward the hydrophobic deep binding pocket (10–12). This tunnel and the opening of the deep binding pocket with subsequent minocycline or doxorubicin binding are apparent after the L to T transition (Fig. 1 and ref. 10). The L monomer of the AcrB/doxorubicin structure comprises one dodecyl-β-d-maltoside molecule at the TM8/TM9 transmembrane groove and a doxorubicin dimer sandwich at the access pocket (Figs. S4 and S6). The access pocket and the TM8/TM9 groove might represent exclusive entry points for drugs with different properties, or consecutive gates in a multistep transport mechanism.

A Switch-Loop Separates Access and Deep Binding Pocket Areas.

The doxorubicin molecules in the access pocket of the L monomer are separated from the deep binding pocket (Fig. 1) by a loop containing F615, F617, and R620 (switch-loop; Fig. 2 A and B).

The switch-loop adopts alternative conformations in the L and T monomer. It seems to modulate the path from the access pocket in the L conformation to the deep binding pocket in the T conformation. Superimposition of the L and T conformations indicates that the side chain conformation of F615 (located on the switch-loop) and Q176 (located on the PN2 subdomain) seen in the L monomer would interfere with doxorubicin binding in the deep binding pocket, and the switch-loop main chain conformation in the T conformation would sterically interfere with the stacked doxorubicin location as observed in the L conformation. Therefore, steric hindrance presumably precludes binding of the drugs in the deep binding pocket in the L monomer, whereas the alternative loop conformation seems to prevent binding of doxorubicin to the access pocket in the T conformation (Fig. 2B). The switch-loop conformation in the structure of the symmetric AcrB (i.e., the LLL conformation) is less inward oriented and would theoretically interfere with doxorubicin binding if these would localize at the same position as observed in the asymmetric LTO AcrB trimer (Fig. S7). Substrates like ethidium (PDB ID code: 1T9X, 3.1 Å) (27) or taurocholate (PDB ID code: 2W1B, 3.85 Å) (24) seem to bind more at the periphery of the PC1/PC2 cleft compared with doxorubicin binding in the asymmetric cocrystal structure (Fig. S7C).

It was recently reported (29) that a single substitution in the switch loop (G616N) decreases AcrB's ability to extrude macrolides, resembling the poor efflux activity toward macrolides shown by the Pseudomonas aeruginosa homolog MexB (wild-type MexB contains N616). The switch-loop conformation in the L conformation of the AcrB G616N variant (solved at 2.9 Å in the presence of minocycline; Table S1) resembles the loop conformation of wild-type AcrB in the T monomer and of the switch-loop conformation found in the wild-type MexB L monomer structure (Fig. 2 B and C). The differences in the switch-loop conformations between the wild-type AcrB and G616N variant are clearly defined by the differences in electron density in this region of the protein (Fig. S6 C and D).

Superimposition of the L monomer structure of MexB or AcrB G616N variant on the AcrB wild-type L monomer structure predicts a minor steric clash with one of the doxorubicin molecules bound the access pocket of wild-type AcrB (Fig. 2C). We conducted a drug resistance assay on solid medium using E. coli BW25113ΔacrB comprising wild-type or G616N AcrB equally well expressed from plasmids (Fig. S8). Clearly, an effect of the G616N substitution on the resistance against erythromycin could be detected, and a subtle difference in growth in the presence of doxorubicin was observed. Growth on other substrates like novobiocin, ethidium, or chloramphenicol was, however, also slightly affected by the substitution in various degrees. In minimal inhibitory concentration (MIC) experiments reported recently (29) using liquid media and chromosomal substitution of the G616N variant in an E. coli AG100 background, larger macrolide molecules were substantially less well transported by this variant, whereas other substrates like novobiocin, ethidium, and chloramphenicol showed wild-type resistance. Specific sensitivity toward macrolides was also conferred when F615 (localized on the switch-loop) was substituted with Ala, or when residues 615–617 were deleted from the loop (30).

Access, Binding, and Extrusion, the Three Cyclic actions Mediated by the L, T, and O Monomer.

The structural information obtained in this study from crystallization and structural elucidation of the wild-type AcrB with bound minocycline and doxorubicin at unprecedented high resolution, as well as the structures of the AcrB variant G616N, can be combined in a model for access, binding, and extrusion of drugs catalyzed by AcrB. Homotrimeric AcrB can adopt three different monomer conformations representing the consecutive states L, T, and O. However, during transition of the conformations within the trimer, AcrB is anticipated to exist in intermediate states [e.g., TTO (15, 16)], a hypothesis that is supported by quantitative cysteine cross-link experiments and molecular dynamics studies (21, 31).

The LTO conformation most likely represents the lowest energy form of the AcrB trimer in the presence of substrate (15, 16, 31), given that in vivo cross-link studies (21) and analytical ultracentrifugation experiments in the presence of DARPins (12, 13) imply the presence of this conformation in the membrane and in detergent rather than the readily crystallizable LLL form. The latter state might preferably exist in the absence of ligands (i.e., “resting state”), as has been suggested previously (7, 15, 31), and apparently is preferred within the R32 crystal lattice. Once substrates are present in the periplasmic space and/or partitioned in the periplasmic space/inner-membrane boundary, AcrB-catalyzed efflux of the drugs over the outer membrane is counteracting the influx. The L monomer provides probable initial binding sites for drugs to enter the AcrB protein. One entry point might be the TM8/TM9 groove located at the height of the periplasm/membrane boundary (i.e., the outer leaflet of the inner membrane). In the high-resolution structures presented here (Fig. S6) and in the 2.5-Å structure by Sennhauser et al. (12), a dodecyl-β-d-maltoside molecule is deeply buried inside the TM8/TM9 transmembrane groove, unlike the other more loosely bound dodecyl-β-d-maltoside molecules that were assigned from the data of the electron density maps. Moreover, this groove is located exactly below tunnel 2 (Fig. S1F) in the T monomer leading to the deep binding pocket (12, 15). In addition, recent Bodipy FL N-(2-aminoethyl) maleimide labeling data indicated that residue S836, located at the entrance of tunnel 2 on the PC2 subdomain, was highly accessible for this AcrB substrate (8). A second entry point is postulated at a lateral access pocket at the periplasmic cleft confined by the PC1 and PC2 subdomains (Fig. 2 and Figs. S1 B and D, S4, and S7). Both putative entry sites might represent exclusive access points for drugs with different properties, or consecutive binding sites. Interestingly, both sites collapse during the proton motive force dependent T to O transition simultaneous with the closure of the deep binding pocket and the disappearance of the tunnels leading from both these putative drug entry points toward the deep binding pocket (11, 12, 15).

The postulated entry step at the periplasmic cleft is depicted by the doxorubicin sandwich inside the access pocket of the L monomer (Fig. 3A). Binding of substrates (ethidium, rhodamine 6G, ciprofloxacin, nafcillin, Phe-Arg-β-naphthylamide, and taurocholate) was also shown in the symmetric L-state trimer (24, 27) at a more lateral position of the PC1/PC2 cleft (Fig. S7). The more lateral binding might indicate an earlier step in substrate acquisition by the complete ligand-free “resting-state” LLL trimer or might indicate the promiscuity of binding different drugs to different sites within the cleft area. Although it is unclear why two doxorubicin molecules are bound to the L monomer of the asymmetric trimer, it seems that one of the drug molecules can be further transported deeper into the AcrB protein toward the T monomer deep binding pocket (Fig. 3B). Very recently, cocrystal structures from 3.3-Å X-ray data showed the binding of high-molecular-mass drugs erythromycin and rifampicin to the access pocket, whereas the low-molecular-mass substrates like minocyclin and doxorubicin were postulated to be transported directly to the deep binding pocket, without binding first to the access pocket (32). Because doxorubicin is mainly present as dimer in solution at concentrations at which the AcrAB-TolC efflux system confers resistance to E. coli against the drug (and at the concentrations used for cocrystallization) (33), the binding in a dimeric state in the access pocket might represent a preliminary stage to the binding of the doxorubicin monomer in the deep binding pocket. This entails prerequisite conformation flexibility between the L and T monomer, as well as a binding affinity difference of the access pocket (low affinity) and deep binding pocket (high affinity). The switch-loop and the β-sheets of the PN2/PC1 subdomains prohibit entry of the drug in the deep binding pocket in the L conformation (Fig. 2), and rearrangement of these structures in the L to T transition facilitates access to the deep binding pocket in the T monomer (Fig. 3B). The presence of dodecyl-α-d-maltoside in the PC1/PC2 cleft of the T monomer seems to contradict the predicted function of the switch-loop. The maltoside moiety of dodecyl-α-d-maltoside is localized at the same position as the daunosamine sugar ring of one of the doxorubicin molecules in the L monomer, whereas the aliphatic tail protrudes into the porter domain and seems not to interfere with the switch loop conformation in the T monomer (Fig. S3). The presence of dodecyl-α-d-maltoside at this position in the T monomer is puzzling, even more so given that the high resolution of the electron density maps did allow to clearly distinguish between the α- and β-anomers of this detergent. Because only low concentrations of α-anomer are present in the commercial batch of dodecyl-β-d-maltoside, the binding must occur with high affinity for this ligand.

Once binding of drug to the deep binding pocket occurs, it is most likely stabilizing the T conformation and is enabling further stabilization by the PN1 subdomain tilting of the neighboring monomer (Fig. 3B). Thus, movement of the subdomains PN2/PC1 in the L to T transition not only takes effect on further conformational changes in the same monomer (i.e., T to O transition) (Fig. 3C) but is also implied to have an effect on the conformational changes in the neighboring monomers.

Movement of the subdomains during the T to O transition is coupled to an extensive kinking of TM8, which seems to be coupled to conformational changes in the transmembrane domain. Most likely protonation events [i.e., uptake of protons from the periplasm to the proton relay residues (D407/D408) located in the transmembrane domain] lead to a conformational change of the TMs, and this movement is presumably transduced to TM8, which is connected to the PN1/PC2 subdomains. Once the drug has been released from the O conformation, a switch to the L conformation to initiate another round of drug binding in the access pocket seems necessary (Fig. 3D). The reorientation of the PN1/PC2 subdomains during the O to L transition leads to de-kinking of TM8, which possibly triggers the deprotonation events in the transmembrane domain, stabilizing the L monomer conformation, and a new round of antibiotic uptake via the TM8/TM9 groove and/or access pocket can be started (Fig. 3). Molecular insight into coupling of the protonation and deprotonation events in the transmembrane domain to the conformational changes in the periplasmic porter domain leading to drug efflux will eventually provide the complete chain of events of proton motive force-driven drug extrusion by AcrB.

Bacterial Strains, Plasmids, and Growth Conditions.

E. coli DH5α was used as host for cloning procedures. E. coli C43 (DE3) (34) harboring pET24acrB_His (35) and pET24acrB_His_G616N was used for protein overproduction. LB medium and LB agar (36) were used for routine bacterial growth at 37 °C. Kanamycin (Applichem) was used at 50 μg mL⁻¹ (Kan⁵⁰). Drug resistance was assayed using solid media, as described previously (37).

Site-Directed Mutagenesis.

pET24acrB_His (35) served as a template for site-directed mutagenesis. By using the Quikchange protocol (Stratagene), amino acid substitution was achieved with 5′ phosphorylated primers. Insertion of mutation was verified by sequencing (Microsynth).

Crystallization of AcrB Wild Type and G616N in Presence of DARPins.

Overexpression and purification of DARPin clone 1108_19 was accomplished as previously described (12). AcrB wild-type and AcrB_G616N were overproduced according to the method reported previously (11, 35), with the following variations for G616N. Expression was induced with IPTG (1 mM) at OD₆₀₀ = 0.8, and the culture was incubated overnight at 20 °C, producing up to 20 mg AcrB_G616N per liter of culture after affinity chromatography. Because of the tendency of AcrB_G616N to aggregate, the protein concentration for crystallization was experimentally determined to be optimal at 8 mg mL⁻¹. DARPins synthesis, purification, and AcrB/DARPin complex crystallization were done according to Sennhauser et al. (12) using n-dodecyl-β-d-maltoside (Glycon; with 0.01% α-isomer). Substrate was added before crystallization at concentrations of 2 mM and 0.4 mM for minocycline or doxorubicin, respectively. Crystals (AcrB/DARPin and G616N/DARPin complex) were grown by the hanging drop vapor diffusion method at 0.05 M N-(2-Acetamido)-iminodiacetic acid buffer (pH 6.5), 7–9% (wt/vol) polyethylene glycol (PEG) 4000, 6–10% (vol/vol) glycerol, and 0.2 M (NH₄)₂SO₄. AcrB_G616N/DARPin crystals were grown in the presence of minocycline (2 mM final concentration), but electron densities at 2.9 Å at the expected deep binding pocket location were too weak to assign minocycline binding at this position.

X-Ray Diffraction Dataset Analysis and Refinement Procedure.

Datasets of P2₁2₁2₁ crystals were collected at the beamline X06SA of the Swiss Light Source (Paul Scherrer Institut) (wavelength 0.8–1.0 Å). Data reduction was done with the XDS package (38). The structures were solved by molecular replacement using MOLREP (39) or PHASER (40). Refinement was performed with the program phenix.refine from the PHENIX package (41) using rigid body refinement followed by restrained refinement with translation libration screw (TLS) restraints. Model rebuilding was performed using the program COOT (42). For superimposition of the AcrB monomers and ligands, the program SUPERPOSE (43) was used matching the DN (residues 181–272) and DC (residues 724–812) subdomains (11). Figures were created using Pymol (www.pymol.org).