Assembly and Channel Opening of Outer Membrane Protein in Tripartite Drug Efflux Pumps of Gram-negative Bacteria*

Background: Pseudomonas aeruginosa mainly achieves multidrug resistance by use of the MexAB-OprM pump. Results: We determined the crystal structure of MexA. Electron microscopy work using MexA and OprM reveals that MexA makes a tip-to-tip interaction with OprM. Conclusion: We suggest an assembly and channel opening model for the pump. Significance: This study provides a better understanding of multidrug resistance in Gram-negative bacteria. Gram-negative bacteria are capable of expelling diverse xenobiotic substances from within the cell by use of three-component efflux pumps in which the energy-activated inner membrane transporter is connected to the outer membrane channel protein via the membrane fusion protein. In this work, we describe the crystal structure of the membrane fusion protein MexA from the Pseudomonas aeruginosa MexAB-OprM pump in the hexameric ring arrangement. Electron microscopy study on the chimeric complex of MexA and the outer membrane protein OprM reveals that MexA makes a tip-to-tip interaction with OprM, which suggests a docking model for MexA and OprM. This docking model agrees well with genetic results and depicts detailed interactions. Opening of the OprM channel is accompanied by the simultaneous exposure of a protein structure resembling a six-bladed cogwheel, which intermeshes with the complementary cogwheel structure in the MexA hexamer. Taken together, we suggest an assembly and channel opening model for the MexAB-OprM pump. This study provides a better understanding of multidrug resistance in Gram-negative bacteria.

Gram-negative bacteria are capable of expelling diverse xenobiotic substances from within the cell by use of three-component efflux pumps in which the energy-activated inner membrane transporter is connected to the outer membrane channel protein via the membrane fusion protein. In this work, we describe the crystal structure of the membrane fusion protein MexA from the Pseudomonas aeruginosa MexAB-OprM pump in the hexameric ring arrangement. Electron microscopy study on the chimeric complex of MexA and the outer membrane protein OprM reveals that MexA makes a tip-to-tip interaction with OprM, which suggests a docking model for MexA and OprM. This docking model agrees well with genetic results and depicts detailed interactions. Opening of the OprM channel is accompanied by the simultaneous exposure of a protein structure resembling a six-bladed cogwheel, which intermeshes with the complementary cogwheel structure in the MexA hexamer. Taken together, we suggest an assembly and channel opening model for the MexAB-OprM pump. This study provides a better understanding of multidrug resistance in Gram-negative bacteria.
One mechanism of multidrug resistance in bacteria is enabled by the activity of efflux transporters that expel a diverse range of xenobiotics and antibiotics within the cell (1). In Gram-negative bacteria, the tripartite efflux pumps, which span the inner and outer membrane and the periplasmic space, translocate these diverse substances out of the cell (2). These efflux pumps are comprised of an inner membrane transporter, an outer membrane efflux protein (OEP), 6 and a membrane fusion protein (MFP) that links the two components in the periplasm (3).
In Escherichia coli, the AcrA-AcrB-TolC pump plays a central role in the multidrug resistance (3). The inner membrane transporter AcrB belongs to the resistance nodulation cell division (RND)-type transporter family, which utilizes the proton gradient across the inner membrane (4) to drive the process. AcrB forms a homotrimeric assembly and translocates diverse materials through the pore at the top of the periplasmic domain (5). The OEP TolC exhibits a homotrimeric assembly that includes the membrane-anchoring ␤-barrel domain and the cavity-forming ␣-barrel domain, which is composed of six long ␣-hairpin equivalents (6). In the absence of the other components, the TolC channel remains closed at the periplasmic end of the cylinder due to the interactions of adjacent ␣-hairpins with salt bridges and hydrogen bond networks (6,7).
Pseudomonas aeruginosa is characterized by an innate multidrug resistance, which is significantly conferred by the RNDtype efflux pump MexA-MexB-OprM (20). The MFP MexA is a lipoprotein that contains a palmitoyl chain linked on the N-terminal cysteine residue. The crystal structure of the MexA in the lipid moiety-free form exhibited the six-or seven-membered twisted horseshoe assembly (21,22), which does not directly explain the MexA assembly in the membrane bound state. However, cryo-microscopic results in the reconstructed membrane layer showed that the MexA forms a hexameric complex with a dimension of 13-nm height (15,23,24). This finding is consistent with the hexameric model of MexA proposed in the previous report (14).
The MacA-MacB-TolC pump exports macrolides and protein toxins in E. coli (25,26), and the inner membrane component MacB is a member of the dimeric ATP-binding cassette transporter family (27,28). The crystal structure of the MFP MacA revealed a funnel-like hexameric assembly, where the six ␣-hairpin domains from the hexamer pack to form a hollow cylinder (29). The functional importance of the hexameric funnel structure also has been demonstrated (29,30). A chimeric protein containing the TolC ␣-barrel periplasmic end region formed a tight complex with MacA in an intermeshing cogwheel-to-cogwheel manner (30,31). Results from the chimeric studies suggested that the assembly mechanism of AcrA and TolC share that of MacA and TolC (13,14,31). Recently, the crystal structure of CusB in complex with CusA, which are the components of an RND-type tripartite pump CusB-CusA-CusC, revealed that the MFP CusB assumes a funnel-like hexameric assembly upon interaction with the RND-type transporter (19). Nevertheless, the assembly mechanism for the RND-type efflux pump has been considered to remain unclear (16).
In this study, we determined the crystal structure of a chimeric protein containing the P. aeruginosa MexA ␣-hairpin domain. The electron microscopic study of the protein complex containing the MexA ␣-barrel and OprM ␣-barrel illustrates an intermeshing cogwheel-to-cogwheel interaction. Furthermore, the role of peptidoglycan in the channel opening of the outer membrane channel proteins is discussed. Overexpression and Purification of Recombinant Proteins-A. actinomycetemcomitans MacA-MexA␣ hybrid and A. actinomycetemcomitans MacA-OprM␣ hybrid dimer proteins were overexpressed in E. coli and purified using the same method used for A. actinomycetemcomitans MacA as described previously (29).

EXPERIMENTAL PROCEDURES
Electron Microscopy and Image Processing-The sample was applied to a freshly plasma cleaned carbon-coated copper grid and allowed to adhere for 60 s before being blotted. Immediately after blotting, 3 l of a 1% solution of uranyl formate was applied onto the grid and blotted off directly. This was repeated three times. Data were acquired using a Tecnai F20 Twin transmission electron microscope operating at 200 kV, using a dose of ϳ40 e Ϫ /Å 2 and nominal underfocus ranging from 1.5 to 2.5 m. At the specimen level, 1,440 high magnification images were automatically collected at a nominal magnification of 80,000ϫ and pixel size 0.107 nm. All images were recorded with a Tietz F415 4k ϫ 4k pixel CCD camera (15-m pixel) utilizing Leginon data collection software (32). Experimental data were processed by the Appion software package (33), which interfaces with the Leginon database infrastructure. Automatic selection of 44,711 particles was conducted (34) from the micrographs and extracted at a box size of 400 pixels. XMIPP reference-free maximum likelihood alignment was applied to identify the best particles (35) from the selection. The final stack contained 2,244 particles that represented side views of the complex. The initial model was created by applying C6-symmetry on a manually selected two-dimensional average, which was formed using the XMIPP reference-free maximum likelihood alignment procedure (36). The three-dimensional reconstruction was carried out using the XMIPP reconstruction package (36). Resolution was assessed by calculating the Fourier shell correlation at a cut-off of 0.5, which provided a resolution of 24 Å. Automatic Docking was performed using the Chimera software package (37).
Crystallization, Data Collection, and Structure Determination of A. actinomycetemcomitans MacA-MexA␣-A. actinomycetemcomitans MacA-MexA␣ hybrid (20 mg/ml) was crystallized in a precipitation solution containing 0.16 M calcium chloride dehydrate, 0.1 M HEPES sodium (pH 7.5), and 23.5% PEG 400 at 14°C. The data set was collected on beamline 44 at Spring-8 with the CCD detector MX-225 (Marresearch) at Ϫ100°C. The diffraction data were processed and scaled using the HKL2000 package (38). The resulting crystal belongs to the space group P321 with cell dimensions of a ϭ b ϭ 151.7 Å and c ϭ 216.9 Å. Initial phases were determined using the molecular replacement package MOLREP (39) and the coordinates of a truncated version of the A. actinomycetemcomitans MacA structure at the ␣-hairpin tip region as a search model. Model building was performed using the program COOT (40), and model refinement was conducted using the programs PHE-NIX.REFINE (41) and REFMAC5 (39). Crystallographic data statistics are summarized in Table 1. All figures were prepared using PyMOL (42).

Refinement of A. actinomycetemcomitans MacA-
The crystallization, data collection, and structural determination of A. actinomycetemcomitans MacA were described previously (29,43). To build the complete model for A. actinomycetemcomitans MacA, the membrane proximal domain from ZneB was fitted to the MAD-phasing electron density map of A. actinomycetemcomitans MacA. The residues located in the ZneB membrane proximal domain were substituted with those of A. actinomycetemcomitans MacA and subsequently refined using the program PHENIX.REFINE (41).
Size Exclusion Chromatography-To determine the molecular weight of the proteins in this study, size exclusion chromatography was performed at a flow rate of 0.5 ml/min on Superdex S-200 HR 10/30 (GE Healthcare) and equilibrated with 20 mM Tris buffer (pH 8.0) containing 150 mM NaCl.
Purification of Peptidoglycans-Peptidoglycan from E. coli was purified according to the methods of Leduc et al. (44) and Marcyjaniak et al. (45) with modifications. Briefly, E. coli strain K12 was grown at 37°C in LB medium and harvested at an OD 600 of ϳ0.9. The cell pellet was suspended in 10 ml of distilled water, mixed with an equal volume of 8% SDS, and incubated for 30 min at 100°C. After standing at room temperature overnight, the sample was centrifuged at 30°C for 30 min at 1,800 ϫ g (13,000 rpm) in a high speed centrifuge (Hanil). The gel-type pellet was suspended in 1 ml of water and then washed by four cycles of resuspension and recentrifugation in distilled water to remove traces of SDS. Finally, it was resuspended in 0.5 ml of 10 mM Tris buffer (pH 7.5). The resulting insoluble material was treated with DNase at 37°C for 30 min, after that with ␣-amylase at 37°C for 16 h, and then with pronase at 50°C for 16 h. The used enzymes were washed from the peptidoglycan solution by boiling and centrifugation. The purified peptidoglycans were finally resuspended in 0.5 ml of 10 mM Tris buffer (pH 7.5).

RESULTS
Crystal Structure of MexA ␣-Hairpin Domain-The MexA in the crystal tends to form the twisted horseshoe assembly during the crystallization (21,22). The hexameric MexA could be observed only when it anchors to the membrane and/or binds to OprM (15,23,24). To obtain the hexameric form of MexA, we initially attempted to determine the crystal structures of the complex of MexA and OprM. However, it was not straightforward due to the integral membrane protein OprM, and the low affinity between the two proteins in the presence of the detergent that stabilizes the membrane protein. To overcome the problem, we had to construct the forced hexamer of MexA using the chimeric approach reported previously (14,31). This approach has been proven to be effective in representing the functional structure of ␣-hairpin domain of AcrA, a MexA homolog, in vivo and in vitro (14).
A chimeric protein containing a MexA ␣-hairpin domain with A. actinomycetemcomitans MacA was generated as a hexamerization framework because A. actinomycetemcomitans MacA has the highest propensity to form a hexamer in solution as compared with other MFPs (29). The MacA ␣-hairpin domain was substituted by the corresponding region of MexA and resulted in the chimeric protein (named A. actinomycetemcomitans MacA-MexA␣ hybrid in this study) that is intended to mimic the hexamerized MexA ␣-hairpin domains that are capable of binding to OprM (Fig. 1A). Because only ␣-hairpin domains of MFPs are involved in the binding to the cognate OEPs (17,30), the MexA ␣-hairpin domain in the chimeric protein should be suitable for investigating the interaction with OprM.

TABLE 1 X-ray data collection and refinement statistics for A. actinomycetemcomitans MacA-MexA␣ hybrid
The numbers in parentheses are statistics for the highest resolution shell. R free was calculated with 5% of the data set. r.m.s.d., root mean square deviation.

Data
Source BL-NW3A at photon factory AR Wavelength (Å) 1 The chimeric protein A. actinomycetemcomitans MacA-MexA␣ hybrid was purified and crystallized to analyze the structure of the ␣-hairpin domain of MexA. The hexagonally shaped crystals were obtained and x-ray diffracted to a resolution of 3.5 Å. To build the complete model for the chimeric protein, we refined the crystal structure of A. actinomycetem-comitans MacA, whose experimental electron density and partial model were described previously (Fig. 1B, right) (29). Because the length of the ␣-hairpin of MexA is shorter than that of MacA, the end portion of the ␣-hairpin of A. actinomycetemcomitans MacA was cut to make a searching model for the molecular replacement approach (Fig. 1B, left). The overall crystal packing of A. actinomycetemcomitans MacA-MexA␣ hybrid was similar to that of A. actinomycetemcomitans MacA, albeit a different space group (Fig. 1C). The six ␣-hairpin domains from the six protomers were packed sideby-side to make an ␣-barrel with a central hollow, as observed in the A. actinomycetemcomitans MacA crystal (Fig. 1D). When compared with ␣-hairpin domains from the full-length MexA structure (21,22), any significant conformational change was not observed (supplemental Fig. S1). The end of the ␣-barrel was slightly flared because the first ␣-helices of the ␣-hairpins is twisted outward by ϳ60°at the tip region, exposing a six-bladed cogwheel (supplemental Fig. S2). The cogwheels of the chimeric protein intermeshed with the ␣-barrel from the neighboring molecules in the crystal (Fig. 1C, left). This crystal structure illustrates the propensity for hexameric cylinder formation in the ␣-hairpin of MexA, suggesting that MexA can form a funnel-like hexamer when the MexA is anchored to the membrane or binds to MexB or OprM.
MexA ␣-Barrel Domain Interacts with OprM ␣-Barrel Tip Region-To examine whether or not the MexA ␣-hairpin domain is capable of interacting with the ␣-barrel tip region of the cognate OEP OprM, as observed with AcrA and TolC, we constructed a chimeric protein containing the OprM ␣-barrel tip region based on the hexameric protein A. actinomycetemcomitans MacA, as used in construction of the TolC ␣-barrel tip region in the previous reports (14,31). The OprM monomer consists largely of two repeats, which contain one ␣-hairpin at the ␣-barrel tip region such as TolC, whereas A. actinomycetemcomitans MacA has only one ␣-hairpin per monomer. To mimic the ␣-barrel tip region of OprM, two A. actinomycetemcomitans MacA protomers were fused in tandem and each of the A. actinomycetemcomitans MacA residues 124 -147 were replaced with corresponding OprM residues, 188 -212 (in repeat 1) or 396 -419 (in repeat 2) ( Fig. 2A). We designated the chimeric protein as A. actinomycetemcomitans MacA-OprM␣ hybrid dimer and generated a model based on the crystal structure of A. actinomycetemcomitans MacA (29), whose structure was refined in this study. The overall structure of the model for the chimeric protein was expected to be similar to A. actinomycetemcomitans MacA, to include the feature that the central channel is wide open in the OprM residue tip region (Fig. 2B). This open conformation is distinguished from the closed conformation observed in the crystal structure of the full-length OprM (Fig. 2C) (46). This chimeric protein containing the OprM ␣-barrel tip region (A. actinomycetemcomitans MacA-OprM␣ hybrid dimer) exhibited a strong affinity to another chimeric protein containing MexA ␣-hairpin on a size exclusion chromatographic column (Fig. 3A), as observed in a study of E. coli AcrA and TolC (14). Interestingly, the E. coli TolC ␣-barrel tip-containing protein, referred to as A. actinomycetemcomitans MacA-TolC␣ hybrid dimer in the previous report (14), demonstrated a strong affinity to the MexA ␣-hairpin-containing protein and vice versa (Fig. 3B). This result indicates that E. coli AcrA and P. aeruginosa MexA share common structural motifs in TolC binding, which is consistent with a previous observation that a chimeric AcrA containing the MexA ␣-hairpin domain was as functional as the wild type AcrA in vivo (17).
It was reported previously that the stoichiometry of the MexA-OprM complex in a reconstituted membrane bilayer shows strong pH dependence (15). To examine the pH dependence of the MexA-OprM complex formation under detergentfree condition, we carried out the in vitro binding assay. The MacA-OprM␣ hybrid dimer protein was coupled to agarose resin, and then the MacA-MexA␣ hybrid protein was incubated at various pH values. After washing, the binding was monitored in Coomassie Blue-stained SDS-polyacrylamide gel (supplemental Fig. S4). The result implied that the binding of the MexA ␣-hairpin domain to the OprM ␣-barrel tip is not dependent on the external pH. We think that the protein-protein interaction may be interfered at a certain pH range by the detergent molecules used in a reconstituted membrane layer.

Structure of OprM and MexA Complex, Visualized by Electron Microscopy-To depict the binding interface between
OprM and MexA, we carried out electron microscopic studies on the protein complex composed of the two chimeric proteins containing the OprM ␣-barrel tips and MexA ␣-hairpins. The hexagonal shape from the top view and indications of bridging from the side view were easily identified in the raw images (Fig.  4A). A projection-matching approach using side views of the negatively stained complex resulted in a three-dimensional density map at 24 Å resolution (Fourier shell correlation ϭ 0.5). Two-dimensional class averages indicate greater flexibility at one of the two ends of the complex (Fig. 4A, arrows). In the resulting three-dimensional electron density maps, this flexibility is represented by additional masses at one of the two ends. In previous reports, similar electron microscopic analyses were done with the TolC ␣-barrel tip and MacA (31), and the TolC ␣-barrel tip and AcrA ␣-hairpin (14).
Compared with previous results (14,31), the three-dimensional electron density map in this study demonstrated improved detail. The map displayed a characteristic hexagonal dumbbell-like structure with six spikes at both ends. However, the spikes were interlinked at only one end (Fig. 4B). A hexagonal cylindrical shape with a slight bulge was at the middle region of the dumbbell-like structure. A central channel of varying diameter with funnel-like openings on both ends threads through the complex.
Docking of Atomic Models-To gain more structural details of the MexA-OprM complexes, we docked the atomic models to the electron density map. Given the crystal structure of the chimeric protein containing the MexA ␣-hairpin (A. actinomycetemcomitans MacA-MexA␣; Fig. 1C) and the atomic model of the chimeric A. actinomycetemcomitans MacA protein containing the OprM ␣-barrel tips (A. actinomycetemcomitans MacA-OprM␣ hybrid dimer; Fig. 2B), we first generated the complex model of the two structures in the intermeshing cogwheel-to-cogwheel manner. The atomic models were well fitted to the density, but only when the two six-bladed cogwheels intermeshed. The configurations of the atomic models were refined using a three-dimensional cross-correlation function (Fig. 5A). The intermeshing ␣-barrels from the two proteins were fitted to the central hex-cylindrical electron density map, and the slight bulge at the middle region (about one-third from the bottom; Fig. 4, B, panel i) of the HEX cylinder from the bottom in the electron density map perfectly matched the bind-ing interface between OprM and MexA. The bulge region may display the slightly flared ends of each ␣-barrel of the proteins, and the position of the bulge reflects the length difference of the ␣-hairpins because MacA ␣-hairpins are about two times longer than MexA ␣-hairpins.
The lipoyl and ␤-barrel domains were well suited to the spreading region in the electron density map, and the mem-brane proximal domains were strikingly fitted to the six spikes protruding from each end. We were able to unambiguously identify which side contained the OprM tip despite the structural similarity between the two chimeric proteins because only the A. actinomycetemcomitans MacA-OprM␣ hybrid dimer containing the OprM tip region had the linker at the membrane proximal domains between the two MacA units. As a final step in building the model, the orientation of the side chains in the OprM and MexA ␣-barrel tips were adjusted by molecular dynamics using CNS (Fig. 5B) (47).
In this complex model of OprM and MexA ␣-barrel tips, the conserved amino acids, Arg-119, Leu-123, and Ser-130, of MexA contributed to OprM ␣-barrel tip binding. This observation was consistent with previous reports that demonstrated the role of the OEP in binding and the functional importance of the three conserved residues (13, 14, 29 -31). Arg-119 and Leu-123 were located at the end of the first ␣-helical region of the ␣-hairpin facing the right side of the cogwheel, whereas Ser-130 was located in the loop between the two ␣-helices facing the opposite side of the cogwheel (Fig. 5B). The long aliphatic chain of Arg-119 occupied the empty space between the convex tip of  (14) or MacA-MexA␣ hybrid (Pa MexA␣) was covalently linked to agarose using CNBr-activated resin (50 g; GE Healthcare) according to the manufacturer's instructions. Then, MacA-TolC␣ hybrid dimer protein (Ec TolC␣ tip; 1 mg) or MacA-OprM␣ hybrid dimer protein (Pa OprM␣ tip; 1 mg) was incubated with the agarose resin for 2 h at 4°C, followed by thorough washing with PBS, and then application by SDS-PAGE. The asterisk indicates that the proteins were liberated partially from the resin by denaturation, even though the proteins were covalently attached to the resin, due to the multimeric structure of the proteins. Because a part of the MacA-OprM␣ hybrid dimer protein did not form the hexameric structure (i), the amount of the bound protein is relatively small compared with the MacA-TolC␣ hybrid dimer protein. However, the bindings of the protein to AcrA ␣-barrel and MexA ␣-barrel are strong and specific because the MacA-OprM␣ hybrid dimer protein is not bound to the resin in the absence of the bait proteins.
the cogwheel in the OprM ␣-barrel and the concave face of the cogwheel in the MexA ␣-barrels (Fig. 5C). The guanidinium group of the arginine residue formed a hydrogen bond with the backbone carbonyl group of the ␣-barrel tip region of OprM. The hydroxyl group of Ser-130 in MexA formed a hydrogen bond with OprM Ser-198 (or Asn-410 in repeat 2). The most striking interaction was mediated by Leu-123, which is most conserved among MFPs. All the OEPs, including TolC, OprM, and CusC, had a conserved motif at the ␣-barrel tip region, Val/Thr-Gly-Val, which made a characteristic sharp bend and rendered Val/Thr and Val residues facing in the same direction. The Leu-123 residue of MexA, together with Leu-122, created a hydrophobic interaction with Val-201 in repeat 1 (or Val-408 in repeat 2) of OprM, as observed in the hydrophobic interaction between the helices in the coiled-coil structure. These interactions, mediated by the three conserved residues, are all preserved in the complex models of AcrA-TolC and MacA-TolC (Fig. 5E). Additionally, 18 polar interactions were found between MexA and OprM (supplemental Table S2). For instance, Asp-126 of MexA formed a salt bridge with Arg-190 of OprM repeat 1 or Lys-402 of OprM repeat 2.
TolC and AcrA Affinity to Peptidoglycan from E. coli-In Gram-negative bacteria, a peptidoglycan layer transverses the periplasmic space where OEP and MFP interact in vivo. It has been reported that a periplasmic domain of TdeA, a TolC homolog, has an intrinsic affinity to peptidoglycan (48). For these reasons, we examined whether or not the purified proteins of the full-length TolC and AcrA interact with peptidoglycan from E. coli. We used peptidoglycan from the Gram-positive bacteria Staphylococcus aureus as a negative control. The results indicated that TolC and AcrA co-precipitated with peptidoglycan from E. coli but not with that from S. aureus (Fig. 6). These observations indicated that both TolC and AcrA proteins have a specific affinity toward E. coli peptidoglycan. We also observed that the amounts of TolC and AcrA that co-precipitated increased by co-incubation of the two proteins with the E. coli peptidoglycan (Fig. 6). Our findings further suggested that the binding affinity of TolC to AcrA was promoted by peptidoglycan.

DISCUSSION
In the crystal structures of TolC, permuted salt bridges or hydrogen bonds between the neighboring protomers kept the channel closed (6,7). When the interaction was disrupted, the TolC channel partially opened spontaneously (7). Similar interactions were also found in the crystal structure of OprM in the closed state, demonstrating 12 polar interactions (46). However, the number of intermolecular interactions associated with channel closing was less than the putative interactions for the association of OprM with MexA residues. Considering the conserved leucine residue-mediated hydrophobic interaction at the six binding interfaces, the intermolecular binding energy should have overcome the interhelical channel closing energy. This finding indicates OprM binding to MexA is more thermodynamically favorable than OprM channel closing.
To allow OprM to efficiently and rapidly bind MexA, hexamerization of MexA is required to expose the six-bladed cogwheel at the ␣-hairpin tip region. We observed that the monomeric form of MexA or AcrA did not interact strongly with the OprM or TolC ␣-barrel tip regions in the hexameric form (14), whereas the hexamerized form of MexA ␣-hairpin domains demonstrated a strong affinity to OprM. The homohexamerization of MexA could be achieved by interaction with MexB. As observed in the case of CusB binding to CusA (19), the RNDtype inner membrane transporter provided the platform for the hexamerization of MFP. This mechanism implies that only MexB-bound MexA can bind to OprM, which prevents channel opening of the OEP by MFPs alone. In addition to the synergistic interaction with MexB, the interaction with OprM can also contribute to the hexamerization of MexA. It was observed previously that hexamerization of E. coli MacA and AcrA proteins was induced by the TolC tip-containing protein (14,31). These observations were consistent with the change in proteolytic sensitivity of the membrane proximal domain of MFPs by the OEP (16,49,50), which may stem from the global conformational change induced by the hexamerization of MFPs, in particular the mobile membrane proximal domains. There remains an obstacle for MexA to bind rapidly to OprM even though the binding reaction may be thermodynamically favorable. In our binding model, the residues for the channel closing are mostly involved indirectly in the binding to MFP. Consistent to the model, the TolC mutant generated by disruption of the hydrogen bonding network in channel closing were functional in vivo, suggesting that the residues involved in channel closing do not play essential roles in MFP binding (7). Thus, it is unlikely that the MexA residues directly induce the channel opening of OprM. If this is the case, what interaction would result in the opening the OEP channel? We note that thermal motion associated with the ␣-hairpins of the OEP could break the interhelical interactions responsible for channel closing and thus induce occasional partial channel opening. At the moment of thermally induced opening, the MexA ␣-hairpin tip region could make the necessary interaction with the OprM ␣-barrel tip region to induce full opening of the OprM channel by cooperative binding. It is noteworthy that the OEPs and MFPs are found to be associated in the in vivo environment. In this study, we observed partial affinities of TolC and AcrA to peptidoglycan, but also a synergistic binding of peptidoglycan to AcrA with TolC. This observation tempts us to speculate that peptidoglycan may be a candidate for increasing the likelihood of collision between OEPs and MFPs, which in turn facilitates capture of the open state of OEP by MFP.
Aside from our observations, other cross-linking experiments have implied that the MFP tip region is associated with the outside of the ␣-barrel of the OEP (11,12,(51)(52)(53)(54). Based on these results, several assembly models have been proposed and are generally accepted in this field (11,16,55). Janganan et al. (52) reported recently that the channel opening of the outer membrane protein channel is induced by interaction with the membrane fusion protein MtrC in the Neisseria gonorrhoeae MtrCDE pump. However, these studies cannot exclude the possibility that the models based on interactions between MFP and OEP may represent an intermediate or transient state of the assembly because the cross-linking experiments could capture both transient and intermediate states. Besides, the results from the cross-linking experiments were not supported by genetic data as the mutant genes used for the experiments were not as functional as wild type genes (13,14). In contrast, the crystallographic and electron microscopic tools used in this study provide structural information for the stable complexes, and our results were supported coordinately with biochemical and genetic data (13, 14, 29 -31). Therefore, although the interaction between MFP and the outside of the ␣-barrel of OEP may contribute to partial opening of the channel at an intermediate stage during the assembly, it is more likely that MFP makes a stable interaction with the tip region of OEP as described in this study.
The assembly and channel-opening mechanisms in the multidrug efflux pump are considered an important issue. This study detailed the adapter bridging model by illustrating the cogwheel-to-cogwheel interaction between MFP and OEP by use of crystal structure and electron microscope images. Owing to the atomic model for the complex, we make a reasonable explanation as to the mechanism for binding interaction between MFP and OEP. Although the exact mechanism must await confirmation through elucidation of additional ternary complex crystal structures, this study provides a better understanding of the underlying tripartite drug efflux pumps mechanisms that play an important role in the incidence of multidrug resistance in Gram-negative bacteria.