Distinct allosteric mechanisms of first-generation MsbA inhibitors

Abstract

ATP-binding cassette (ABC) transporters couple ATP hydrolysis to substrate transport across biological membranes. Although many are promising drug targets, their mechanisms of modulation by small molecule inhibitors remain largely unknown. Intriguingly, two first-generation inhibitors of the MsbA transporter, TBT1 and G247, induce opposite effects on ATP hydrolysis. Using single-particle cryo-electron microscopy and functional assays, we show that TBT1 and G247 bind adjacent yet separate pockets in the MsbA transmembrane domains. Two TBT1 molecules asymmetrically occupy the substrate binding site, leading to a collapsed inward-facing conformation with decreased distance between the nucleotide-binding domains (NBDs). In contrast, two G247 molecules symmetrically increases NBD distance in a wide inward-open state of MsbA. The divergent mechanisms of action of these MsbA inhibitors provide important insights into ABC transporter pharmacology.

Introduction

ATP-binding cassette (ABC) transporters perform diverse functions including importing and exporting small molecules, mediating ion channel opening, and translocating lipids across cell membranes (1–8). ABC transporters have two transmembrane domains (TMDs), which interact with and transport substrates, and two nucleotide-binding domains (NBDs), which bind and hydrolyze ATP. Decades of investigation have led to various models to describe ABC transporter mechanisms. In the alternating access model (9–13), substrate is first recognized by the inward-facing transporter, subsequent ATP binding in the NBDs promotes a transition to the outward-facing state and substrate release, and finally ATP hydrolysis resets the transporter in the inward-facing state. Interrupting such conformational transition cycle has direct applications in treating cancer, regulating cholesterol homeostasis, and developing new antibiotics (14–16).

Small molecule inhibitors have been developed against human multidrug ABC transporters, including the ABCB1 inhibitors zosuquidar, tariquidar, and elacridar, and the ABCG2 inhibitors MZ29 and MB136 (3, 17–19). Structural studies have revealed how these compounds bind to the TMDs, interrupting conformational transition and consequently decreasing ATPase activity (3, 19, 20). In contrast, there is little structural insight into small-molecule modulation of most other ABC transporters, which demonstrate narrower substrate specificity. We do not know if modes of small-molecule inhibition are shared across the ABC superfamily or if there are general druggable conformations or pockets. These knowledge gaps limit our ability to rationally design drugs targeting many ABC transporters.

MsbA is a model system to study ABC transporter mechanism (21, 22). Due to its essential role in lipopolysaccharide (LPS) biogenesis in Gram-negative bacteria, MsbA is also an attractive target for developing antibiotics (23). X-ray crystallography and cryo-EM studies have determined MsbA structures in several conformations including ligand-free, LPS-bound, adenylyl-imidodiphosphate (AMP-PNP)-bound and vanadate-trapped (6, 22, 24–26). Importantly, the structural basis of specific binding and transport of LPS by MsbA has been elucidated (6, 25, 27). Compared to the recent rapid progress in studying MsbA function, understanding MsbA inhibition has lagged behind (27).

Two types of specific MsbA inhibitors have been reported: one has a tetrahydrobenzothiophene (TBT) scaffold, named herein as TBT1, and stimulates MsbA ATPase activity despite abolishing LPS transport (28); the other group, here referred to as G compounds, blocks both ATP hydrolysis and LPS transport (25, 29). Although these first-generation MsbA inhibitors represent important breakthrough, their mechanisms of inhibition are not well understood. Particularly, TBT1-induced decoupling of ATPase activity from substrate transport is puzzling because most known inhibitors suppress ATP hydrolysis of ABC transporters in membrane environment (12). We thus sought to understand how TBT1 and G compounds exert opposite allosteric effects on ATP hydrolysis even though both block LPS transport.

TBT1-bound MsbA adopts an asymmetric, collapsed inward-facing conformation

TBT1 was identified as an LPS transport inhibitor and MsbA ATPase stimulator in strains from the Acinetobacter genus (28). We therefore sought to explore the mechanism of TBT1 inhibition in Acinetobacter baumannii, an ESKAPE pathogen responsible for antibiotic-resistant infections in patients (30). A. baumannii MsbA was expressed in Escherichia coli cells, purified in dodecyl maltoside (DDM), and reconstituted in palmitoyl-oleoyl-phosphatidylglycerol (POPG) nanodiscs (Fig. S1A–G) (31, 32). Basal ATPase activity of A. baumannii MsbA in nanodiscs was ~1 μmol ATP/min/mg MsbA (Fig. 1A), which is ~4-fold lower than E. coli MsbA (6). Exposure of MsbA to TBT1 resulted in 4-6 fold stimulation in ATPase activity (Fig. S1D, S1G, S7B), as well as dose-dependent increase in reaction rate in Michaelis-Menten kinetics (Fig. 1A). Notably, TBT1-induced ATPase stimulation was much more pronounced in nanodiscs than in DDM (Fig. S1G), suggesting that nanodiscs are a more native-like system than detergent for investigating TBT1 action on MsbA. We thus used single-particle cryo-EM to determine the structure of TBT1-bound MsbA in nanodiscs at an overall resolution of 4.3 Å, with the TMDs at 4.0-Å resolution showing sufficient side-chain densities for model building (Fig. 1B, S2, and S3).

Fig. 1. TBT1 binding induces a collapsed inward-facing conformation of MsbA.

(A) ATPase activity of A. baumannii MsbA in nanodiscs, measured at varying ATP and TBT1 concentrations. TBT1 increases the rate of ATP hydrolysis in a dose-dependent manner. Each point represents mean ± SD (n=3). (B) Left, 4.0-Å resolution cryo-EM map of A. baumannii MsbA in complex with TBT1, with domain-swapping TM4-TM5 colored fuchsia. The unsharpened map filtered at 10-Å resolution is displayed as outline to show the nanodisc and tightened NBDs positioning. Right, cartoon of TBT1-bound MsbA with structural elements indicated, illustrating the structural asymmetry of the two MsbA chains. EH, elbow helix; CH, coupling helix. A flipped cartoon can be found in Fig. S4A. (C) Cryo-EM reconstruction low-pass filtered at 6-Å resolution. The dashed box points to the N-terminal end of TM4.B, where TM4-TM5.B becomes disordered and absent from the reconstruction. (D) Representative 2D class averages of TBT1-bound MsbA, showing constricted TMDs yet separate NBDs. Box size is 203 Å.

The conformation of TBT1-bound MsbA is asymmetric and distinct from all previously determined structures of type IV ABC transporters that are characterized by their domain-swapped TMs (33). Notably, the domain-swapping TM4-TM5 bundle is well resolved in chain A yet disordered in chain B (Fig. 1B), which is evident also in unmasked and unsharpened 3D reconstructions (Fig. S4B). This chain mismatch results in asymmetrical TMDs, with TMD1 including 6 helices (TM4,5.A, TM1,2,3,6.B) and TMD2 including only 4 helices (TM1,2,3,6.A). While TM4.B is mostly unresolved, its N-terminal end is clearly discernable and forms a 61º angle with the C-terminal end of TM3.B, which is ~15º greater than the corresponding angle between the well resolved TM4.A and TM3.A (Fig. S4C). Accordingly, TM4.B kinks outwards, pushing the nanodisc density to bulge upwards from the membrane plane (Fig. 1C and Fig. S4C, left). Interestingly, while the consensus map features a bent nanodisc, a subset of particles was incorporated in comparatively flatter and larger nanodiscs yet still adopt the asymmetric TBT1-bound conformation, suggesting that TM4-TM5 bundle destabilization is not dependent on nanodisc size (Fig. S2B). The domain-swapping TM4-TM5 bundle is of critical importance because the coupling helix (CH) at the TM4-TM5 junction (CH2), together with CH1 at the TM2-TM3 junction, is responsible for transmitting movement from the catalytically active NBDs to the TMDs (Fig. 1B, right). Due to disordered TM4-TM5.B, CH2.B is absent for interaction with the NBD in chain A (NBD.A). Accordingly, NBD.A tilts up, moving towards the NBD dimerization interface (Fig. 1B). Notably, CH1.A also appears destabilized, likely because it cannot form proper interaction with the drastically shifted NBD.A (Fig. S4D). In summary, the raised NBD.A is disengaged from its two coupling helices and seems decoupled from the TMDs.

At first sight, 2D class averages and 3D reconstruction of the cryo-EM images of TBT1-bound MsbA (Fig. 1D and Fig. S4E) resemble those of nucleotide-bound MsbA (6), both exhibiting constricted TMDs and reduced inter-NBD distance compared to nucleotide-free MsbA. However, the TBT1-bound MsbA conformation is still inward-facing because no nucleotide is present in the cryo-EM sample and the NBDs remain fully separate. Accordingly, TMD1 is well aligned to the structure of nucleotide-free E. coli MsbA (Fig. S4F, root-mean-square-deviation (RMSD) of 5.5 Å over 314 Cα atoms), leaving significant structural mismatch from TMD2. The most noteworthy structural rearrangement in TMD2 is the 13-Å displacement of TM6.A towards the central cavity (Fig. S4G), which seems to pull NBD.A upwards and to close proximity of NBD.B. Thus, the TBT1-bound MsbA structure described here is an unusual “collapsed inward-facing” conformation characterized by two unprecedented features: 1) complete destabilization of one domain-swapping TM4-TM5 helix bundle; 2) highly asymmetric positioning of the NBDs.

TBT1 binding drives the closure of a wide open MsbA

We then sought to characterize the structure of drug-free A. baumannii MsbA to understand how TBT1 binding affects MsbA conformation. To this end, we imaged A. baumannii MsbA in identical conditions as described above, but without addition of TBT1. The resulting cryo-EM reconstruction at an overall resolution of 5.2 Å enabled unambiguous tracing of each TM helix, revealing a wide inward-facing conformation (Fig. 2A, Fig. S5). To our knowledge, this is the first structure of membrane embedded MsbA in a wide inward-facing conformation. Such conformation had only been characterized by X-ray crystallography using MsbA in detergents (21, 24), and its physiological relevance has been heavily debated (6, 26, 34). Our result demonstrates that the wide inward-facing conformation of MsbA can exist in a lipid bilayer and may not be an artifact caused by deprivation of membrane environment and usage of detergents. Furthermore, the wide opening of MsbA seems not due to lack of bound substrate, because a clear LPS-like density is present in the inner cavity (Fig. S5G).

Fig. 2. Drug-free A. baumannii MsbA in nanodiscs adopts a wide inward-facing conformation.

(A) 5-Å resolution cryo-EM reconstruction of A. baumannii MsbA in the absence of ligand. The unsharpened map filtered at 8-Å resolution is displayed as outline to show the nanodisc. (B) Section through surface representation and comparison for three structures: A. baumannii MsbA, TBT1-bound A. baumannii MsbA, and E. coli MsbA (PDB: 5TV4, (6)). (C) Top-down view of the substrate binding pockets. (D) Side views of the NBDs. Cα distances between ABC signature motif serine and opposing Walker A glycine residues are indicated as dashed lines, with values rounded to the nearest integer shown in each panel. (E) Top view of the NBDs, seen from the transmembrane domains. Cα distances between NBDs are indicated as in D.

Since drug-free A. baumannii MsbA has well defined domain-swapping TM4-TM5 bundle in both TMDs and distanced NBDs, TBT1 binding is solely responsible for converting the wide open MsbA to the collapsed inward-facing conformation. To gain insights of TBT1 stimulated ATP hydrolysis of MsbA, we compared three cryo-EM structures: TBT1-bound or drug-free A. baumannii MsbA, and E. coli MsbA (Fig. 2B–E). Notably, the central substrate-binding pocket of TBT1-bound MsbA is much more constricted than that in the other two structures (Fig. 2B,C), consistent with the notion that the TMDs in TBT1-bound state present similarities with an outward-facing transporter. Upon TBT1 binding, the NBD distance is drastically decreased from ~47 Å to ~20 Å and even slightly shorter than the NBD distance in LPS-bound E. coli MsbA (Fig. 2D,E). The two NBDs of TBT1-bound MsbA are positioned asymmetrically, such that one ATP binding site is tightened compared to the opposite ATP site (~19 Å vs. ~22 Å, Fig. 2E). The structural analysis of TBT1-bound MsbA provides an intuitive explanation for ATPase activity stimulation: removal of TM4-TM5.B bundle and sliding of TM6.A into the central cavity greatly reduce inter-NBD distance, thus increasing the speed of NBD dimerization and ATP hydrolysis.

TBT1 hijacks the LPS binding site to modulate MsbA

Two densities, each consistent with a TBT1 molecule, are present in the upper region of the TMDs (Fig. 3A). While the structure resolution is limited, the distinctive features of the TBT1 densities allow for reliable modeling of the compounds and some critical sidechains (Fig. 3B, Fig. S6A–C). Interestingly, the two TBT1 ligands are positioned asymmetrically in the central substrate pocket of MsbA and related by a ~60º rotation angle, with the carboxyl group pointing downwards to the cytoplasm. The general binding mode of TBT1 is reminiscent of how E. coli MsbA binds the amphiphilic LPS using a large hydrophobic pocket, which accommodates the lipid acyl chains in LPS, and a ring of basic residues (Arg78, Arg148 and Lys299), which stabilize the phosphate groups on glucosamines (Fig. 3C, top) (6). Similarly, the two TBT1 molecules are located right above the ring of basic residues, with the aromatic ring structures of TBT1 positioned in the hydrophobic pocket (Fig. 3D).

Fig. 3. TBT1 binding to MsbA.

(A) Cartoon representation of TBT1-bound A. baumannii MsbA colored as in Fig. 1B, with TBT1 colored blue. (B) Close-up view of the TBT1 binding site (top) and superimposition of model and cryo-EM density (bottom). Distances between TBT1 carboxyl group and neighboring lysine and arginine residues are shown as colored dashed lines. (C) Top-down view of E. coli MsbA bound to LPS (PDB: 5TV4, (6)) and A. baumannii MsbA bound to TBT1. TMD1 is conformationally similar in both structures and indicated with a gray background. The phosphate groups on LPS glucosamines and the carboxyl groups of TBT1 are marked with red circles. (D) Surface representation of the TBT1 binding pocket, with hydrophilic and hydrophobic surfaces colored blue and orange, respectively. (E) TBT1-induced ATPase stimulation of wild-type and mutant MsbA. The activity of each protein was normalized to its basal activity without TBT1. Error bars correspond to mean ± SD (n=3). (F) ATPase activity of A. baumannii MsbA with W13 at increasing concentrations. Each point represents mean ± SD (n=3). The molecular structure of W13 is shown in inset.

Despite being in asymmetric binding pockets (Fig. S6B), each TBT1 molecule is stabilized by an analogous set of hydrogen-bonding and electrostatic interactions between TBT1 carboxyl groups and neighboring basic sidechains. TBT1.A is oriented parallel to the membrane plane, with its carboxyl group within hydrogen-bonding distance of Lys290.A and forming long-range electrostatic interactions with Lys290.B and Arg72.A, and TBT1.B is rotated relative to TBT1.A, forming a salt bridge with Arg72.B (Fig. 3B, and Fig. S6A).

In our structure, the carboxyl group of TBT1 is analogous to the phosphate groups on LPS glucosamines, both forming electrostatic interaction with MsbA. We thus sought to mutate Arg72 and Lys290 to determine whether MsbA retains sensitivity to TBT1-induced ATPase stimulation. A. baumannii MsbA with R72A, K290A, or R72A/K290A mutation was purified as wild-type MsbA and reconstituted in nanodiscs (Fig. S7A). All MsbA mutants demonstrated largely diminished TBT1-induced stimulation compared to the wild-type protein, highlighting the importance of electrostatic interactions in TBT1 binding (Fig. 3E, Fig. S7B,C). Consistently, chemical modification of the carboxyl group of TBT1 renders the compound incompetent for MsbA inhibition (28). Although the electrostatic interactions described herein may be the main drivers of TBT1 binding, hydrophobic contacts through the ring structures of TBT1 also participate in stabilizing the inhibitor. For instance, L68F and L150V mutants, which were previously shown to confer resistance to TBT1 (28), are in the vicinity of the TBT1 binding sites (Fig. S7D). While not directly adjacent to TBT1, it is conceivable that these mutations allosterically alter the pocket and prevent efficient inhibitor binding.

Despite striking resemblance in LPS and TBT1 recognition by MsbA, differences in TM positioning lead to divergence in ligand binding (Fig. 3C). LPS and TBT1 stabilization both involve TM2 (Arg72) and TM6 (Lys290), although unlike LPS, TBT1 does not clearly interact with TM3. Notably, the two TBT1 molecules together form a distorted mimic of LPS, with decreased distance between TBT1 carboxyl groups compared to LPS glucosamine phosphates. Upon TBT1 binding, TMD2 (TM1,2,3,6.A) undergoes significant changes: TM6.A moves into the center of the inner cavity, presenting Lys290.A for interaction with the hydroxyl group of TBT1.A and dragging the connected NBD.A towards the central axis (Fig. 3C, bottom). TMD2 then collapses around TM6.A, with TM2.A moving closer to TM6.B. It is conceivable that the convergence between TM2.A and TM6.B would then push TM4-TM5.B laterally. The unsolvable contradiction between NBD.A moving towards the central axis and TM4-TM5.B being pinched outwards results in the dissociation of the CH2.B coupling helix from NBD.A.

TBT1 offers an example of a small molecule inhibitor targeting the central substrate binding site of a non-multidrug ABC transporter. Because most ABC transporters have relatively narrow substrate spectra, our findings suggest that mimicking substrate binding can be a generally applicable strategy for developing small molecule modulators for all ABC transporters. To test if the central pocket is useful for structure-based compound discovery, we performed an in silico chemical library screen against the TBT1 induced-fit pocket. Virtual screening of ~800,000 compounds followed by ATPase assay of select hits resulted in identification of W13, a compound that stimulated ATPase activity with an EC₅₀ lower than TBT1 (~5.5 μM vs. 13 μM, Fig. 3F and Fig. S8B). Notably, the same MsbA mutants that were insensitive to TBT1 stimulation of ATPase activity also demonstrated lower W13-induced stimulation (Fig. S7B,C). W13 is structurally dissimilar to TBT1 and considerably larger in size (499 Da vs. 336 Da, Fig. S8A), and a single W13 molecule is expected to occupy the central MsbA pocket (Fig. S8C,D). Molecular dynamics simulation of W13-bound MsbA in a POPC lipid bilayer suggests a reminiscent binding mode to TBT1, including hydrophobic groups of W13 extending upward in the hydrophobic pocket and carbonyl groups of W13 near TBT1-recognizing arginine and lysine residues (Fig. S8D).

G247 symmetrically increases inter-NBD spacing and prevents MsbA closure

As the only other class of known MsbA-specific inhibitors, G compounds are mechanistically distinct from TBT1, because they bind the inward-facing MsbA and prevent ATP hydrolysis (25). Crystal structures of G compound-bound E. coli MsbA in FA-3 detergent present asymmetrical NBDs positioning (25), although it remains unclear how the asymmetric conformation translates to impaired conformational cycling. Furthermore, the extraordinary conformational flexibility of ABC transporters can result in different states being captured depending on whether the protein is studied by crystallography (21, 22, 24) or cryo-EM (6, 26). We thus sought to further investigate the mechanism of G compounds and acquired cryo-EM datasets of G247-bound E. coli MsbA in nanodisc (Fig. 4A and Fig. S9) and DDM (Fig. S10).

Fig. 4. G247 symmetrically increases inter-NBD spacing.

(A) 3.9-Å resolution cryo-EM reconstruction of E. coli MsbA in complex with G247. The unsharpened map filtered at 6-Å resolution is displayed as outline to show the nanodisc. (B) Comparison of MsbA structures determined by cryo-EM or X-ray crystallography (PDB: 6BPL, (25)). The cryo-EM structure adopts C2 symmetry, while NBDs are asymmetrically positioned in the crystal structure. (C) Comparison of the cryo-EM structures of drug-free (PDB: 5TV4, gray, (6)) and G247-bound (colored) MsbA, both in nanodiscs. G247 pushes NBDs away from each other, compared to the uninhibited state. The NBD shift is apparent when viewing MsbA from the side (left panel), and from the cytoplasmic space (right panel). The dashed box in the MsbA side-view (left) represents the cross-section of the NBDs shown in the panel on the right.

In previous crystal structures of G compound bound MsbA (25), the NBDs are brought in close proximity and positioned asymmetrically with one NBD raised relative to the other (Fig. 4B, right). In contrast, our cryo-EM structures in nanodisc and detergent are essentially identical, both exhibiting clear C2 symmetry, even when refined without symmetry constraints (Fig. 4B, left, and Fig. S11B,C). Asymmetry in the crystal structures stems from Arg190 which forms a salt bridge with the acrylic acid tail of the G compounds only in one subunit of MsbA (25). In cryo-EM structures, Arg190 is best ordered in the higher-resolution DDM structure and appears too distant from G247 to form a salt bridge (Fig. S11E). Furthermore, TM4,5,6 bundle is positioned differently relative to TM1,2,3 in the crystal and cryo-EM structures (Fig. S11F), which ultimately impacts the spacing between coupling helices (CH1 and CH2) and NBD positioning. While CH1-CH2 spacing is reduced in crystal structures compared to the drug-free conformation, CH1-CH2 distance is symmetrically increased in the cryo-EM structures (Fig. S11G). Since the closed asymmetric conformation in crystal structures is not observed in any round of 3D classification (Fig. S9E,S10A) or cryoSPARC variability analysis (Fig. S12A) (35) of cryo-EM images, we posit that the open C2-symmetric state is the predominant conformation of G247-bound MsbA in solution.

Comparing the cryo-EM structures of G247-bound and drug-free MsbA (6) reveals a mechanism of G compound inhibition which is different from previously anticipated: instead of pushing one NBD towards the dimerization interface by 10-15 Å (25), G247 binding displaces the NBDs symmetrically and away from each other, increasing inter-NBD distance by ~13 Å (Fig. 4C). This model is consistent with 3D variability analysis (Fig. S12A), in which G247-bound MsbA NBDs move away from the dimerization axis relative to the NBDs of drug-free MsbA. Whereas drug-free MsbA has NBDs aligned in a head-to-tail manner and thus primed for ATP hydrolysis, the increased inter-NBD distance in G247-bound MsbA is expected to reduce NBD dimerization efficiency and ATPase activity. To formally test this hypothesis, we exposed MsbA to the non-hydrolyzable nucleotide AMP-PNP, in the presence or absence of G247 (Fig. S12B). When subjected to 1 mM AMP-PNP, nearly all MsbA particles showed a closed conformation with dimerized NBDs. In contrast, 5 μM of G247 was sufficient to shift more than 60% of particles to an open conformation with separate NBDs (Fig. S12B).

Discussion

The effect on MsbA conformation and the inhibition mechanism of TBT1 (Fig. S13) are different from those of small molecule modulators for other ABC transporters, which typically induce more modest conformational changes (e.g., ABCB1 and ABCG2 in Fig. S6D). The narrowed inter-NBD spacing of TBT1-bound MsbA is more akin to the NBD tightening observed when the human multidrug transporter ABCC1 is exposed to its endogenous leukotriene substrate (Fig. S6D) (1), further suggesting that TBT1 triggers substrate recognition-like behavior in MsbA by mimicking the much larger LPS substrate. Yet unlike the authentic ligand, TBT1 breaks the symmetry of homodimeric MsbA by destabilizing one domain-swapping helix bundle. Uncoupling of ATPase function and substrate transport by TBT1 occurs through the domain-swapped coupling helix, which is a landmark feature of the type IV exporter fold (33). Thus, small molecule decouplers acting like TBT1 may be identified for many other type IV exporters such as ABCB1 and ABCC1. In contrast with TBT1, we have shown that G247 suppresses ATPase activity by increasing inter-NBD distance (Fig. S13). This inhibition mechanism is similar to that of previously described ABC transporter inhibitors, including zosuquidar, which reduces the activity of the homologous ABCB1 transporter by pushing its NBDs away from each other (17). As the vast majority of ABC transporter inhibitors characterized so far are ATPase suppressors, it is likely that more compounds act similarly to G247 than to TBT1.

TBT1 and G247 currently face severe drawbacks preventing their application in clinical settings, including low binding affinity to MsbA (28) and interaction with animal serum (29). Yet compound-bound ABC transporter structures provide essential templates for structure-based drug development, because molecules such as TBT1 and G247 reveal unexpected binding pockets in previously unidentified MsbA conformations. Our identification of W13 as an ATPase stimulator would have been unlikely if the structure of TBT1-bound MsbA was not available. One promising direction for compound optimization may consist of tethering together two TBT1-like molecules to take advantage of cooperative binding. Although TBT1 only works on A. baumannii MsbA, future efforts may yield molecules able to extend this inhibition mechanism to MsbA in other Gram-negative bacteria, including additional antibiotic resistant pathogens.

Data and materials availability:

Atomic models are available through the Protein Data Bank (PDB) under accession codes 7MET (TBT1-bound A. baumannii MsbA), 7RIT (drug-free A. baumannii MsbA), 7MEW (G247-bound E. coli MsbA in nanodisc). Cryo-EM maps are available through the Electron Microscopy Data Bank (EMDB) under accession codes EMD-23803 (TBT1-bound A. baumannii MsbA), EMD-23804 (drug-free A. baumannii MsbA), EMD-23805 (G247-bound E. coli MsbA in nanodisc), EMD-24446 (G247-bound E. coli MsbA in DDM, C2) and EMD-24447 (G247-bound E. coli MsbA in DDM, C1). Unprocessed motion-corrected micrographs for the TBT1-bound MsbA in nanodiscs are accessible through EMPIAR under accession code 10778. All other data are present in main text or Supplementary Materials.