Multidrug Efflux Pumps and the Two-Faced Janus of Substrates and Inhibitors

CONSPECTUS:

Antibiotics are miracle drugs that can cure infectious bacterial diseases. However, their utility is challenged by antibiotic-resistant bacteria emerging in clinics and straining modern medicine and our ways of life. Certain bacteria such as Gram-negative (Gram(−)) and Mycobacteriales species are intrinsically resistant to most clinical antibiotics and can further gain multidrug resistance through mutations and plasmid acquisition. These species stand out by the presence of an additional external lipidic membrane, the outer membrane (OM), that is composed of unique glycolipids. Although formidable, the OM is a passive permeability barrier that can reduce penetration of antibiotics but cannot affect intracellular steady-state concentrations of drugs. The two-membrane envelopes are further reinforced by active efflux transporters that expel antibiotics from cells against their concentration gradients. The major mechanism of antibiotic resistance in Gram(−) pathogens is the active efflux of drugs, which acts synergistically with the low permeability barrier of the OM and other mutational and plasmid-borne mechanisms of antibiotic resistance.

The synergy between active efflux and slow uptake offers Gram(−) bacteria an impressive degree of protection from potentially harmful chemicals, but it is also their Achilles heel. Kinetic studies have revealed that even small changes in the efficiency of either of the two factors can have dramatic effects on drug penetration into the cell. In line with these expectations, two major approaches to overcome this antibiotic resistance mechanism are currently being explored: (1) facilitation of antibiotic penetration across the outer membranes and (2) avoidance and inhibition of clinically relevant multidrug efflux pumps. Herein we summarize the progress in the latter approach with a focus on efflux pumps from the resistance–nodulation–division (RND) superfamily. The ability to export various substrates across the OM at the expense of the proton-motive force acting on the inner membrane and the engagement of accessory proteins for their functions are the major mechanistic advantages of these pumps. Both the RND transporters and their accessory proteins are being targeted in the discovery of efflux pump inhibitors, which in combination with antibiotics can potentiate antibacterial activities. We discuss intriguing relationships between substrates and inhibitors of efflux pumps, as these two types of ligands face similar barriers and binding sites in the transporters and accessory proteins and both types of activities often occur with the same chemical scaffold. Several distinct chemical classes of efflux inhibitors have been discovered that are as structurally diverse as the substrates of efflux pumps. Recent mechanistic insights, both empirical and computational, have led to the identification of features that distinguish OM permeators and efflux pump avoiders as well as efflux inhibitors from substrates. These findings suggest a path forward for optimizing the OM permeation and efflux-inhibitory activities in antibiotics and other chemically diverse compounds.

Graphical Abstract

MULTIDRUG EFFLUX IS SYNERGISTIC WITH THE MEMBRANE PERMEABILITY BARRIER

The synergistic relationship between multidrug efflux pumps and the low permeability barrier of the outer membrane (OM) has been evident since their discovery more than 25 years ago.⁵ Since then, many examples have emerged in which a modest overproduction of efflux pumps has contributed to clinical antibiotic resistance.⁶ Recent kinetic modeling studies combined with experiments described this relationship in mathematical terms and explained why this mechanism of antibiotic resistance is so effective (Figure 1).^2,3,7,8 The major features of the system are (1) a propensity for saturation of both drug efflux and uptake fluxes at high concentrations of the drug; (2) the resulting highly nonlinear behavior of the system that is controlled by two kinetic parameters, the efflux constant (K_E) and the barrier constant (B); (3) a large reduction in steady-state drug accumulation levels in the range of low drug concentrations, which is specified by the value of K_E; and (4) a bifurcation at high drug concentrations that can either amplify or negate the effect of active efflux, depending on which of the fluxes, inward or outward, is saturated first. This bifurcation is controlled by the value of the barrier constant B, which relates the maximal outward and inward fluxes across the OM; at B = 1, these two fluxes are equal to each other. Even relatively small changes in the value of B can result in dramatic changes in intracellular drug accumulation levels (Figure 1).

Figure 1.

(A) Schematic representation of a Gram(−) cell envelope. The OM of these bacteria is an asymmetric bilayer of lipopolysaccharides and phospholipids, whereas the inner membrane is formed by phospholipids. (B) Drug accumulation in cells that differ in efflux efficiency. Steady-state drug accumulation levels were modeled for bacteria with the barrier constant (B) values of 0, 0.5 (B < 1), or 2 (B > 1). The barrier constant is defined as the ratio of maximal attainable active efflux to the maximal attainable uptake.² Also labeled are the MICs of drugs with B < 1 (MIC₁) and B > 1 (MIC₂), assuming that the intracellular inhibition constant K_I and efflux constant K_E are the same in both cases.

In principle, this mechanism of resistance, which is achieved through an increase of B above 1, could result in a large increase in minimal inhibitory concentrations (MICs) of antibiotics, as can be seen for example for macrolides, which are effective only against Gram-positive (Gram(+)) bacteria. The only way to sensitize bacteria to such compounds would be to bring B down to below 1, either by increasing the maximal influx rate or by decreasing the maximal efflux rate. The good news here is that this transition from the effective efflux at B > 1 to ineffective efflux at B < 1 can be achieved by reasonably small changes in a flux, which are well within reach of small-molecule inhibitors. Therefore, the two general approaches to overcome antibiotic resistance, namely, increasing influx across the OM or inhibiting active efflux, suggest a feasible path forward for potentiation of activities of antibiotics against Gram-negative (Gram(−)) and Mycobacteriales species and are at the heart of efforts to develop new antibacterial therapeutics. Approaches to overcome the OM permeation barrier are summarized in recent excellent publications.^8,9 This Account is focused on the efflux inhibition and avoidance approaches.

MULTIDRUG TRANSPORTERS OF THE RESISTANCE–NODULATION–DIVISION SUPERFAMILY AND THEIR ACCESSORY PROTEINS

Multidrug efflux transporters are members of various protein families and vary dramatically in molecular structures and mechanisms.¹⁰ All of these transporters, however, are polyspecific and can expel multiple structurally unrelated compounds from cells. The choice of a target transporter for inhibition is a critical question and is defined by its contribution to efflux of a given antibiotic. Transporters of the resistance–nodulation–cell division (RND) superfamily are notorious for their contributions to clinical levels of antibiotic resistance and are validated targets for efflux pump inhibitors (EPIs).^11,12 In Gram(−) bacteria, RND pumps expel various compounds from cells, including antibacterial agents, organic solvents, dyes, and hormones as well as intrinsically produced signaling molecules and pigments.¹³ In Mycobacteriales species, these pumps are important for transporting glycolipids that are essential for the assembly of the OM, and some of these transporters are also implicated in drug efflux.¹⁴

The diverse functions of RND pumps are enabled by an electrochemical gradient of protons across the cytoplasmic or inner membrane (IM), which drives the influx of protons from the periplasm into the cytoplasm through the transmembrane domains of the transporters, causing large conformational changes.¹⁵ These conformational changes alter the substrate binding properties of the proteins and are needed to move a substrate across or away from the membrane.¹⁶ In many transporters driven by a proton-motive force, protons and substrates share a binding site located in the transmembrane domain of the protein. In contrast, the substrate-binding sites of RND pumps are located in the periplasmic domains and are physically separated from the protonation sites of their transmembrane domains.^17,18 As a result, RND pumps do not translocate their substrates across the cytoplasmic membrane, in which they are located, but instead capture them from the periplasm or at the water–lipid interface and move them away from the lipid bilayer (Figure 2). In addition, such separation of proton and substrate binding sites could enable unidirectional drug efflux by RND pumps that cannot be reversed by changing the directionality of the proton gradient. This mechanism was validated by early biochemical reconstitution studies, which showed that the AcrB transporter from the model bacterium Escherichia coli transfers fluorescently labeled lipids from AcrB-containing proteoliposomes into acceptor lipid vesicles.^19,20

Figure 2.

Structures of AcrB from E. coli (PDB entry 3W9H)²¹ and MmpL3 from M. smegmatis (PDB entry 6AJG).²² Drugs bind AcrB at the access site and reach the deep binding pocket (Deep pocket) through tunnels. An inhibitor, D13-9001, is shown in the Deep pocket of AcrB, interacting with aromatic residues of the hydrophobic trap. MmpL3 shares a transmembrane topology with AcrB, but its periplasmic and cytoplasmic domains are structurally different. A molecule of 6-n-dodecyl-α,α-trehalose, which is a mimic of the MmpL3 substrate trehalose monomycolate, is bound in the periplasmic binding site. The indicated proton-translocating networks of Asp residues are located at the center of the transmembrane domains of AcrB and MmpL3.

Glycolipid substrates of Mycobacteriales RND pumps are delivered onto the cell surface. How these substrates reach their destination remains unclear, but RND pumps of these species interact with various proteins that could participate in transport.^14,23 In Gram(−) bacteria, drugs and signaling molecules are expelled into the external medium. At least two different accessory proteins associate with RND pumps and create a protein conduit spanning the periplasm and the OM (Figure 3). The E. coli AcrB transporter functions in complex with AcrA from the membrane fusion protein (MFP) family and the OM channel TolC from the outer membrane factor (OMF) family.²⁴ The complex is assembled sequentially, with AcrA and AcrB forming a “drug-sweeping” IM intermediate, which was originally postulated on the basis of biochemical analyses^19,25–27 and recently visualized by cryogenic electron tomography studies of AcrAB–TolC²⁸ and cryogenic electron microscopy studies of TriABC, an AcrAB homologue from the human pathogen Pseudomonas aeruginosa.²⁹ In this intermediate state, three MFP dimers interact with a trimeric RND transporter, and each MFP dimer establishes two different protein–protein interfaces with an RND protomer (Figure 3). MFP1 associates tightly with the periplasmic domain of an RND protomer in close proximity to an access tunnel leading to the substrate binding pocket located deep in the periplasmic domain of the transporter (the Deep pocket). Extensive molecular dynamics simulations demonstrated that this association reduces the conformational flexibility of MFP1 and stabilizes the complex with the transporter.^29–31 In contrast, MFP2 binds weakly at the interface between the two neighboring RND protomers and remains conformationally flexible.

Figure 3.

(A) Homology model of the AcrA MFP with its four domains labeled.³² MP denotes the membrane proximal domain. (B) Structure of the TriABC pump (PDB entry 6VEJ)²⁹ in a partially assembled intermediate state with the unresolved domains of the MFPs TriA and TriB shown schematically in red and blue, respectively. (C) Structure of the tripartite AcrAB–TolC pump (PDB entry 5NG5)³³ with MFP1 and MFP2 shown in red and blue, respectively.

In the drug-sweeping state, the Deep pocket of the transporter is not accessible from the periplasm because the access tunnels leading to this pocket are closed.²⁹ Structures of the fully assembled AcrAB–TolC complex suggest that the interactions with an OMF channel drive oligomerization as well as conformational changes in the MFPs needed to activate the transport.³³ Reconstitution studies showed that MFPs stimulate energy consumption by the transporters and that the substrates appear to contribute to this stimulation.^19,26 However, neither the presence of a substrate nor the functionality of the transporter is needed to engage the OMF. For example, transporters inactivated by mutations in the proton-translocating pathway may still assemble into tripartite complexes with the MFP and OM channel, and apparently such complexes can be even more stable than the complex with a fully functional transporter.²⁴ Therefore, the complex between a transporter and an MFP is conformationally dynamic, whereas the engagement of an OMF channel rigidizes MFPs into a trans-envelope assembly.

CURRENT ADVANCES IN INHIBITION OF RND TRANSPORTERS AND THEIR ACCESSORY PROTEINS

The mechanism of multidrug efflux by tripartite efflux pumps described above suggests several approaches for their inhibition. In one approach, the target is the active component of the complex, the RND transporter, and its conformational changes needed to transport a substrate. Alternatively, the accessory proteins and/or the assembly of protein complexes can be targeted. Independent of the mechanism, all EPIs are expected to bind specifically to pump components and potentiate the activities of antibiotics in an efflux-pump-dependent manner, such that their presence does not affect the activities of antibiotics in efflux-deficient cells. In addition, EPIs are not expected to have their own antibacterial activity unless the transporter is essential for bacterial growth or during infection.

RND Pumps as Targets

The first EPIs discovered by screening of large chemical libraries were found to target MexB, the RND transporter of the MexAB–OprM efflux complex from P. aeruginosa.³⁴ The initial hit, MC-207,110 (PAβN), and its optimized peptidomimetic analogues potentiated the activities of fluoroquinolones and several other antibiotics against P. aeruginosa strains overproducing several RND-type efflux pumps (Figure 4). These peptidomimetics as well as subsequently discovered MexAB–OprM inhibitors from the pyridopyrimidine class and pyranopyridine-based AcrAB–TolC inhibitors all apparently bind in the so-called “hydrophobic trap” located in the Deep pocket of RND transporters.^13,21 Binding in this site, which is lined with aromatic amino acid residues, traps the transporter in the ligand-bound conformation and slows the rotation of the transporter out of this state, which is required for the substrate to be extruded.

Figure 4.

Chemical structures of the best-characterized inhibitors of AcrA, AcrB/MexB, and MmpL3.

More recently, several other classes of inhibitors have been identified that bind in the hydrophobic trap of Gram(−) RND pumps.^35,36 Because the site of inhibition is located within the substrate binding pocket, many of these inhibitors are also substrates of the transporter and, depending on their interactions in the Deep pocket, can either inhibit, have no effect, or even enhance efflux of certain compounds.³⁷ Hence, these EPIs must be carefully paired with specific antibiotics to match their activities and pharmacological properties in order to achieve the desired therapeutic effect. Interestingly, several discovered inhibitors of MmpL3, an essential RND transporter in many Mycobacteriales species, act on the proton-translocating activity of these proteins. These inhibitors belong to several chemical classes, but all bind to MmpL3 directly and specifically.³⁸ Furthermore, resistance to most of these inhibitors is caused by single amino acid substitutions in the transmembrane domains of MmpL3.³⁹ Structural analyses showed that binding of SQ109, an ethylenediamine containing N-geranyl and N-adamantyl groups, as well as other inhibitors (Figure 4) expand the narrow proton-translocating channel of MmpL3 from Mycobacterium smegmatis by pushing away all six transmembrane α-helices of the protein.²² As a result, the two Asp-Tyr hydrogen-bonded pairs that are critical for proton translocation are disrupted. Importantly, conformations of the side chains and positions of Asp-Tyr pairs in the transmembrane domains of M. smegmatis MmpL3 can be essentially superimposed with the proton-translocating channel of E. coli AcrB (Figure 2). In both transporters, mutational analyses confirmed the functional importance of these residues.³⁹ However, SQ109 and other MmpL3 inhibitors are quite specific and have no activities against Gram(−) RNDs, and no inhibitors targeting the proton-translocating channel of Gram(−) RNDs have been identified to date.

OMFs as Targets

The second potential target for inhibiting efflux in Gram(−) bacteria is the OMFs. Their exposure to the external medium makes these proteins potentially susceptible to inhibition by a variety of molecules because such inhibitors do not need to permeate the OM for their action. Structural analyses of OMFs have revealed that the periplasmic tip of these channels is closed by an irislike arrangement of hydrogen-bonded α-helices.⁴⁰ The interactions with MFPs lead to opening of the channel and stabilization of the tripartite efflux complex. Hexaamminecobalt fits tightly into the closed TolC iris and was reported to bind with high affinity.⁴⁰ However, despite the potent binding to the closed TolC, it does not reduce MIC values of antibiotics in E. coli. The major roadblock here is that the amount of TolC and perhaps other Enterobacteriales OMFs does not limit efflux of antibiotics, as can be seen from the lack of efflux dependence on the expression levels of TolC⁴¹ and the lack of competition between multiple E. coli transporters functioning with this OMF.⁴² On the other hand, P. aeruginosa OMFs such as OprM or OprN function specifically with MexAB and MexEF efflux pumps, respectively, and their genes are coexpressed in the operons with the corresponding transporters. The kinetic contribution of such OMFs is likely to vary depending on the substrate. For slowly permeating antibiotics, even a single active pump could be sufficient to provide high levels of resistance.

In an alternative approach, surface-exposed domains of abundant OM porins and efflux OMFs could be exploited as vaccine candidates.⁴³ Along with a protective immune response, coating the surface of bacteria with antibodies against OM proteins could render bacteria susceptible to antimicrobials (including host defense compounds) exported by RND pumps.

MFPs as Targets

Lastly, EPIs could potentially target the periplasmic MFPs. The dual role of these proteins, namely, stimulation of transporter activity and engagement of an OMF, and their intrinsic conformational flexibility suggest potential vulnerabilities. These proteins consist of four flexibly linked domains, and the interfaces between those domains have been identified as potential binding sites for small-molecule inhibitors.¹ Furthermore, as discussed above, MFPs establish two different interfaces with transporters and must undergo large conformational rearrangements when bound to OMFs.

This team developed a combined virtual and empirical screening approach to identify compounds that bind to E. coli AcrA, the MFP component of AcrAB–TolC (Figure 3). Ensemble docking was used to screen compound libraries for their binding to various conformations of AcrA, and the top candidates were next prioritized on the basis of their ability to potentiate the activity of antibiotics in E. coli and to bind to AcrA in vitro using surface plasmon resonance (SPR) techniques.¹ In the initial validation of the approach, screening of a compound collection from the National Cancer Institute identified two promising hits, NSC 60339 and NSC 33353 (Figure 4), that inhibit efflux in E. coli by different mechanisms. The dihydroimidazoline NSC 60339 was found to disrupt the assembly of AcrAB–TolC, as determined by in situ limited proteolysis of AcrA, which was previously found to report on whether the tripartite complex is assembled.⁴⁴ In contrast, NSC 33353 did not affect the complex assembly and differed in its kinetics of inhibition of efflux of fluorescent probes. Subsequent biophysical analyses showed that NSC 60339 binds between the lipoyl- and α–β-barrel domains of AcrA and likely interferes with TolC-dependent oligomerization of AcrA needed to stabilize the complex.³² In contrast, NSC 33353 and 4-(3-aminocyclobutyl)pyrimidin-2-amine compounds identified by screening of a prefiltered ZINC library apparently inhibit efflux by binding to both AcrA and the hydrophobic trap of AcrB.⁴⁵ These compounds are substrates of AcrAB–TolC and were found to interact directly with both AcrA and AcrB. Interestingly, many other substrates of this efflux pump, including the antibiotics novobiocin and clorobiocin, interact with both subunits of the complex, suggesting that MFP1 bound at the access site of the transporter might contribute to the selection of efflux substrates (Figure 2).

For all EPIs that are effective against Gram(−) bacteria, general nephrotoxicity, attributed to the primary amine groups required for permeation across the OM, is an important consideration.⁴⁶ The search for new chemical scaffolds and mechanisms of action discussed above, applications of machine learning approaches to optimization of compound permeation and efflux inhibition (also see below)^4,47 as well as generation of prodrugs and soft drugs to overcome toxicity are among currently explored approaches.⁴⁶

THE CONTRIBUTION OF PASSIVE PERMEATION BARRIERS TO THE ACTIVITIES OF EPIS AND ANTIBIOTICS

Inhibitors and substrates of Gram(−) RND transporters follow the same path by first penetrating the OM and then binding to MFP and RND components of the complex and reaching the Deep pocket. Both antibiotics and EPIs are affected by the synergy between active efflux and permeation across the OM, but this effect differs for specific compounds. As a result, optimization of antibiotics for permeation and efflux avoidance and optimization of the inhibitory action of EPIs require dissection of the contributions of efflux and the OM barrier to their activities. In our studies, we use hyperporinated strains that express a nonspecific OM pore that is large enough for large molecules (up to 5 kDa) to penetrate the hyperporinated OM.^3,48 The expression of such pores reduces and for some compounds eliminates the effect of synergy in the activities of antibiotics and EPIs, and the strains producing the pore can be used to determine the impacts of active efflux and the OM barrier separately.

By comparing the activities of antibiotics in strains with native and hyperporinated OMs and with and without efflux pumps, we were able to define and distinguish substrate specificities of efflux pumps in several Gram(−) pathogens.^1,49–51 For example, novobiocin and erythromycin both are thought to be excellent substrates of AcrAB–TolC efflux pump because inactivation of the pump leads to the same 64–128-fold decrease in MICs of these antibiotics. However, these two antibiotics are affected differently by the OM barrier. Novobiocin is able to permeate the E. coli OM through the general porins, whereas erythromycin cannot, and its permeation is very slow. Accordingly, hyperporination only mildly (2–4-fold) reduces the MIC of novobiocin but has a dramatic 16–32-fold effect on MICs of erythromycin.⁴⁸

These findings also raised the following question: which substrates should be paired with EPIs to achieve the best potentiation of their antibacterial activity? When not masked by their synergy with the OM barrier, homologous RND pumps, which were previously thought to have similar substrate specificities, display biochemically distinct properties.^49,51 For example, the three major efflux pumps of the important human pathogen Acinetobacter baumannii, AdeABC, AdeIJK, and AdeFGH, possess definable substrate specificities based on their abilities to protect cells from different antibiotics in the absence of the permeability barrier of the OM (Figure 5).⁵¹ AdeFGH is the most effective in protection against trimethoprim and chloramphenicol. However, this pump is rarely overproduced in clinical isolates of A. baumannii because target-based and enzymatic resistance against these antibiotics is widespread among these isolates. Overproduction of AdeABC and AdeIJK is the major contributor to clinical antibiotic resistance. These two pumps differ in their ability to protect against azithromycin and erythromycin, two structurally similar antibiotics from the macrolide class, yet both pumps can protect against fluoroquinolones. AdeABC and AdeIJK not only differ in their substrate specificities but also are integrated into different physiological programs. Inactivation of these pumps in multidrug-resistant A. baumannii isolates triggers specific non-overlapping changes in phenotypes and gene expression profiles.⁵² Homologues of AdeABC, AdeIJK, and AdeFGH with similar substrate preferences are conserved in other pathogenic bacteria such as E. coli, P. aeruginosa, and Burkholderia spp (Figure 5).^49,50 Hence, EPIs targeting these three classes of pumps have the potential to be broad-spectrum inhibitors when paired with substrates specific to different pumps. The choice of antibiotic is also determined by the clinical context, such as its use and resistance burden. It is especially important to establish whether active efflux of a given antibiotic is the major mechanism of resistance in clinical settings or other mechanisms of resistance such as target modifications or enzymes are prevalent.

Figure 5.

Substrate specificities of RND-driven efflux pumps from Gram(−) bacteria. AdeABC, AdeIJK, and AdeFGH are RND-driven pumps of A. baumannii. MexAB–OprM, MexXY–OprM, and MexEF–OprN are efflux pumps of P. aeruginosa. AcrAB–TolC and AcrAD–TolC are E. coli RND-driven pumps. The structures shown are cloxacillin of the β-lactam class, azithromycin of the macrolide class, trimethoprim, and the fluoroquinolone (FQ) norfloxacin.

Other remaining questions concern which substrates can be optimized for avoidance of efflux pumps and whether the development of substrate–inhibitor chimeras is a plausible solution to the permeation problem. All of the “good” efflux-pump substrates, for which efflux reduces the intracellular concentration dramatically, and the “bad” efflux-pump substrates, for which changes are modest, bind to the Deep pockets of RNDs. Surprisingly, docking of antibiotics in the Deep pockets of homologous RNDs with dramatically different substrate preferences suggested that there is no specificity in these pockets. In AcrB from E. coli, with its broad polyspecificity, and TriC from P. aeruginosa, which is specific to triclosan and sodium dodecyl sulfate, the Deep pockets are large enough to accommodate substrates of varying sizes and properties.²⁹ However, to reach the Deep pockets, different substrates must be attracted into the access sites and then moved through one of the tunnels during the conformational transition. In the TriC transporter, the tunnel connecting the access site and Deep pocket is constricted by a large loop such that only small compounds such as triclosan can pass through the constriction. The same loop in AcrB is much shorter and is apparently more flexible, so that even bulky molecules such as macrolides can reach the Deep pocket and be extruded. Therefore, the substrate specificities of RNDs are largely defined by the affinities for the access sites. In contrast, as discussed above, the effectiveness of EPIs is defined by their affinities for the hydrophobic trap in the Deep pocket.

Along the same lines, the major bottleneck for EPIs acting on RND pumps is their difficulty in reaching the Deep pocket (Figure 1). The lack of affinity for the access site or inability to penetrate the tunnels limits the potency of an EPI, even if its affinity for the hydrophobic trap is high. Hence, it is not surprising that many EPIs targeting the hydrophobic trap also have features of substrates. For many EPIs with antibacterial properties, their activities have been shown to increase in efflux-deficient cells.^1,4 This finding further suggests that certain combinations of antibiotics might be synergistic in bacterial killing because one or both substrates could act as EPIs and antibiotics can also be modified to inhibit their own efflux.

In light of this discussion, the desired properties of EPIs targeting MFPs should be distinct from the biochemical features of substrates because EPIs with affinity for the access site will be efficiently pumped out of cells. It remains unclear whether the interactions of MFPs with substrates and EPIs contribute to their affinities for the access site of the transporter or instead play an allosteric role in the assembly or functioning of the pump. The only currently available crystal structure of an MFP in a ligand-bound state is for ZneB, which is involved in the efflux of zinc ions.⁵³ The bound metal stabilizes the cis conformation of ZneB by binding at the interface between the α–β-barrel and MP domains of the MFP. We found that this interface in AcrA is critical for the transition into the trans conformation needed for assembly of the complex and active efflux.³¹ Cysteine residue pairs introduced at this interface trap AcrA in the cis conformation, disrupting the proper assembly of the complex and leading to the loss of its function.³¹ EPIs that bind at this interface could potentially also interfere with conformational transitions in AcrA and the assembly of the functional efflux pump.

On the basis of our current understanding of the efflux mechanism, EPIs acting on MFPs are expected to be the most effective if they prevent conformational transitions of MFPs needed to seal up the periplasmic funnel.

TOWARD PHYSICOCHEMICAL PREDICTORS OF EFFLUX INHIBITION AND AVOIDANCE

On the chemistry side of EPI discovery and optimization, the major challenge is the dissection of physicochemical determinants that are associated with permeation across the OM and efflux recognition in various compounds. To identify those properties, we analyzed MICs of a set of 70 antibiotics, including β-lactams and fluoroquinolones, in wild-type, efflux-deficient, hyperporinated, and “barrierless” (i.e., both hyperporinated and efflux-deficient) strains of E. coli and P. aeruginosa. More than 100 physicochemical properties were then calculated for each antibiotic, and a random forest binary classification model was trained to identify specific properties that best classify the relative potencies of each compound in each strain.⁵⁴

Important observations from that work are that (i) the molecular properties that promote uptake and efflux evasion are species- or strain-specific and (ii) these properties are complementary to each other, such that efflux pumps recognize substrates that penetrate the OM. Therefore, each pathogen presents its own unique challenges, and optimizing the properties of compounds to overcome one barrier tends to result in detrimental properties against the other barrier. Thus, both barriers (i.e., OM permeability and efflux avoidance) must be considered during antibiotic development efforts.

Antibiotics that are active against P. aeruginosa tend to have characteristic electrostatic and surface area properties, whereas those that are active against E. coli display distinct topological and physical properties. However, some properties, such as conformational rigidity, are somewhat general in this regard, with this property facilitating uptake in both E. coli and P. aeruginosa. Unfortunately, rigid antibiotics such as fluoroquinolones tend to undergo efficient efflux by RND pumps. Furthermore, during those studies it became clear that the traditional physicochemical descriptors of compound properties are insufficient for the dissection of interactions of compounds with efflux pumps and OM barriers.

Our current efforts are directed toward the identification of molecular-level properties specific to OM permeators, efflux avoiders, and inhibitors. Recently, using a series of peptidomimetic EPIs related to MC-207,110, we mapped the fate of compounds starting from structure–activity relationships through their dynamical behavior in solution, permeation across both the inner and outer membranes, and interactions with MexB of P. aeruginosa.⁴ Machine learning algorithms were then applied to identify properties that correlate with efflux avoidance and inhibition. Molecular shape (represented by acylindricity), amphiphilicity (anisotropic polarizability), aromaticity (number of aromatic rings), and partition coefficient (LogD) are among the top physicochemical predictors of efflux inhibitors. The propensity to be a substrate/inhibitor increases with increasing acylindricity and anisotropic polarizability but decreases with increasing partition coefficient and lipophilicity. In addition, interactions with Pro668 and Leu674 residues of MexB located in the access pocket and the affinity of compounds for the Deep pocket of MexB also distinguish between efflux inhibitors/substrates and avoiders. Compounds avoiding efflux have fewer contacts with L674 and lower preference for the Deep pocket but more contacts with P668. In contrast, substrates/inhibitors interact more with L674 and the Deep pocket and negatively trend with P668. These predictive models and efflux rules are applicable to compounds with unrelated chemical scaffolds and pave the way for the development of compounds with the desired efflux interface properties.

ABBREVIATIONS

OM

outer membrane

RND

resistance–nodulation–division

MFP

membrane fusion protein

OMF

outer membrane factor

EPI

efflux pump inhibitor

MIC

minimal inhibitory concentration

Gram(−)

Gram-negative

Gram(+)

Gram-positive

Biographies

Helen I. Zgurskaya is George Lynn Cross Research Professor of Chemistry and Biochemistry at the University of Oklahoma in Norman, OK. She earned an M.Sc. degree in Microbiology in 1989 from Dnepropetrovsk State University in Ukraine and a Ph.D. degree in Microbiology in 1992 from the Russian Academy of Sciences, Russian Federation. She subsequently held research appointments at the Max Planck Institute of Molecular Genetics in Berlin, Germany, Stanford University Medical School, and the University of California at Berkeley. She leads a research program on antibiotic discovery and resistance, drug permeation, multidrug efflux mechanisms, and efflux pump inhibitors.

John K. Walker is currently an Associate Professor of Pharmacology and Physiology at the Saint Louis University School of Medicine. He obtained both his B.S. and M.S. degrees in Organic Chemistry from Southern Illinois University at Edwardsville. He did his doctoral studies at Indiana University under the direction of Professor Paul A. Grieco. Prior to joining academia, he spent 10 years doing research as a medicinal chemist at Pfizer and legacy companies. His research interests include the design and synthesis of novel modulators of orphan nuclear receptors as well as efforts to develop new antibacterial agents, including efflux pump inhibitors.

Jerry M. Parks is a Senior R&D Staff Scientist and Group Leader for Molecular Biophysics in the Biosciences Division at Oak Ridge National Laboratory. He received his B.S. in Chemistry from Texas Christian University, M.S. in Chemistry from Southern Methodist University, and Ph.D. in Chemistry from Duke University. His research interests include the structure, function, and inhibition of microbial proteins and enzymes.

Valentin V. Rybenkov is a Professor of Chemistry and Biochemistry at the University of Oklahoma. He received his M.Sc. degree in 1989 and his Ph.D. degree in Biophysics in 1992 from the Moscow Institute of Physics and Technology. He then held appointments at the Institute of Molecular Genetics in Moscow, Russia, and the University of California at Berkeley in the USA. He is currently a faculty member at the University of Oklahoma, where he leads research programs on molecular biophysics, chromosome biology, and antibiotic discovery.