AcrB: a mean, keen, drug efflux machine

Gram‐negative bacteria are intrinsically resistant against cytotoxic substances by means of their outer membrane and a network of multidrug efflux systems, acting in synergy. Efflux pumps from various superfamilies with broad substrate preferences sequester and pump drugs across the inner membrane to supply the highly polyspecific and powerful tripartite resistance–nodulation–cell division (RND) efflux pumps with compounds to be extruded across the outer membrane barrier. In Escherichia coli, the tripartite efflux system AcrAB–TolC is the archetype RND multiple drug efflux pump complex. The homotrimeric inner membrane component acriflavine resistance B (AcrB) is the drug specificity and energy transduction center for the drug/proton antiport process. Drugs are bound and expelled via a cycle of mainly three consecutive states in every protomer, constituting a flexible alternating access channel system. This review recapitulates the molecular basis of drug and inhibitor binding, including mechanistic insights into drug efflux by AcrB. It also summarizes 17 years of mutational analysis of the gene acrB, reporting the effect of every substitution on the ability of E. coli to confer resistance toward antibiotics (http://goethe.link/AcrBsubstitutions). We emphasize the functional robustness of AcrB toward single‐site substitutions and highlight regions that are more sensitive to perturbation.


The origins of antibiotic resistance
To develop new strategies for the treatment of multidrug-resistant pathogens, it is important to understand the fundamentals of microbial resistance, including its origin. Antimicrobial resistance is not a modern phenomenon, but rather an ancient one. 1,2 Genes encoding resistance mechanisms toward β-lactams, tetracyclines, and glycopeptide antibiotics have been isolated from 30,000-year-old permafrost sediments, and a genotype-to-phenotype study identified 18 chromosomally encoded resistance markers. These included three mechanisms previously unknown to be involved in antibiotic resistance that were identified in a bacterium (Paenibacillus sp. LC231) originating from a cave that had been isolated from the outside world for 4 million years. However, the a These authors contributed equally. origin of these genes is much older. On the basis of comparison with orthologous genes, class A β-lactamases originated up to 2.4 billion years ago. 3 For comparison, life on earth, based on findings of fossilized microorganisms, is estimated to be at least 3.77 billion years old. 4 The age of modern humans (Homo sapiens) most likely started around a relatively modest 300,000 years ago. 5 It does not come as a surprise that natural antimicrobial production is evolutionary related to counteracting resistance mechanisms. 6 Most antimicrobials are secondary metabolites produced by certain fungi and bacteria. These microbes are supposed to produce and release these low molecular mass compounds to inhibit essential cellular processes of other microorganisms with whom they compete for resources. In this way, they create(d) an (evolutionary) growth advantage. Inhibition can either be bactericidal (killing of bacteria) or bacteriostatic (inhibiting bacterial growth). 7 The production of antibiotics requires complex modular biosynthetic pathways, including many enzymes and specialized carrier proteins that are encoded in large gene clusters. 8 Most known antibiotics, including β-lactams, tetracyclines, macrolides, aminoglycosides, rifamycins, and glycopeptides, have been isolated from actinomycetes, Gram-positive facultatively aerobic soil bacteria that form fungus-like branched networks. 9 Antimicrobial producers, such as actinomycetes, must be resistant to the antibiotics they produce in order not to be killed. This correlation between production and resistance can be used when searching for new antibiotics. 10 The resistance genes of microbial producers are often encoded on the same contiguous clusters that contain the genes for antibiotic biosynthesis; and the obvious necessity of these genes makes actinomycetes a likely origin of many resistance genes in other bacteria as well. 11,12 In natural environments, it appears relatively common for bacteria to encode resistance genes (often on plasmids), even if these organisms do not produce antibiotics themselves. 13 Further, there are "consumer" bacteria that can not only resist high concentrations of antibiotics but, in addition, can metabolize these as their sole carbon source. 14 The different resistance genes of various organisms form a genetic reservoir, the so-called "resistome," 15 which can be mobilized via horizontal gene transfer 16 into the microbial community. Such transfer can (by detours) also include into human microbial pathogens. Comparative studies, resulting in perfect nucleotide identities between the resistance cassettes of multidrug-resistant soil bacteria and those of clinical isolates, support the hypothesis of horizontal gene transfer. 17 There are many indications suggesting that the massive use of antibiotics has led to a significant shift in balance toward the wider spread and more pronounced forms of (multiple) antibiotic resistance. 18 In 2010, for example, an estimated 63,000 tons of these chemical products were used worldwide in livestock, a market that is responsible for approximately 80% of antibiotic consumption in the United States. 19 Most mechanisms of antibiotic resistance, especially when combined and leading to multidrug-resistant phenotypes, are associated with significant fitness costs, leading to reduced bacterial growth rates. 20 Accordingly, in the absence of antibiotic exposition, multidrug-resistant pathogens are likely to be outcompeted by microbes lacking resistance markers. 21 However, since resistant bacteria are known to acquire additional fitness cost compensatory mutations, some of the resistance markers remain within the population, even in the absence of antibiotic pressure. 20

Targets and functional principles of antimicrobial agents
Effective antimicrobial agents target essential cellular processes, including the evolutionary conserved and complex machineries of cell wall and protein biosynthesis. The mode of action of prominent antimicrobials and counteracting cellular processes are illustrated in Figure 1. The bacterial cell wall (murein) is a multilayer mesh-like peptidoglycan structure formed by long polysaccharides that are cross-linked via short peptides. The cell wall mechanically counteracts the osmotic pressure from the cell interior and influences the shape of bacteria.
Individual steps of murein synthesis are targets of different antibiotics. β-Lactams, such as oxacillin (OXA) and piperacillin (Fig. S1, online only), covalently bind to the catalytic center of the DD-transpeptidase, preventing the enzyme to catalyze the peptidoglycan cross-linking reaction. 22 Supramolecular ribosomes are the protein factories of the cell that translate mRNA-encoded genetic information into polypeptide sequences. Evolution has resulted in many active compound classes directed against this ribosome-dependent machinery; compounds that bind either to the large 50S or the small 30S subunits of bacterial 70S ribosomes, resulting in the inhibition of protein synthesis at different stages (initiation, elongation, or termination). Macrolide antibiotics-including erythromycin (ERY) and spiramycin (Fig. S1, online only), oxazolidones, like linezolid (LIN), and amphenicol antibiotics, such as chloramphenicol (CAT)-bind to the large 50S subunit, whereas aminoglycosides, such as kanamycin (KAN), and tetracyclines, such as minocycline (MIN), bind to the small 30S subunit. Fusidic acid (FUS) interacts with a protein required for a different step of protein synthesis, elongation factor G. 23 Other bacterial protein targets of antibiotics include the replication and transcription machinery. The most prominent being the bacterial DNA gyrase, which can be inhibited by fluoroquinolones,   (1). Molecules that are able to penetrate the lipopolysaccharide layer are confronted with the hydrophobic outer membrane bilayer. More hydrophilic molecules, like nutrients but also β-lactam antibiotics, enter the periplasm via porins localized in the outer membrane (2). The bacterial inner membrane (3), which separates the cytoplasm from the periplasm, poses an additional hydrophobic permeation barrier for hydrophilic compounds. Hydrophobic compounds, such as many of the known antibiotics, readily penetrate the outer leaflet of the inner membrane. The AcrAB(Z)-TolC complex, in particular the resistance nodulation and cell division (RND) component AcrB, is suggested to sequester these hydrophobic compounds from the periplasm and the outer leaflet of the inner membrane, and exports them out of the cell. This process is driven by the proton motive force present across the inner membrane. Targets for most antibiotics are located in the cytoplasm. Single-component multidrug-resistance efflux pumps of the small multidrug resistance (SMR), major facilitator (MF), multidrug and toxin extrusion (MATE), or ATP binding cassette (ABC) transporter (super)families transport antibiotics from the cytoplasm to the periplasm from where these molecules are sequestered and transported across the outer membrane by the AcrAB(Z)-TolC efflux complex. The intrinsic resistance provided by the outer membrane and the network of efflux pumps can be further increased by the downregulation of porins (2) and/or the upregulation of multidrug-resistance efflux pump genes (4). The AcrAB(Z)-TolC multidrug efflux pump recognizes many different classes of antimicrobials (Fig. S1, online only). Among those substrates are antibiotics targeting the bacterial murein (cell wall) cross-linking reaction (e.g., oxacillin (5)), DNA unwinding/replication (e.g., novobiocin (6)), DNA transcription (e.g., rifampicin (7)), the translation machinery (e.g., erythromycin, (8)), or the folic acid biosynthesis pathway (e.g., trimethoprim (9)). Besides efflux, bacterial cells can acquire further resistance mechanisms, including the enzymatic degradation of compounds (e.g., β-lactamases (10)) or by modification of target sites. Inhibitors of β-lactamases, such as clavulanic acid (11) and also polymyxins, which disrupt bacterial membranes (12), were reported to be substrates of RND homologs in Gram-negative bacteria. Under overexpressing conditions, the MacAB-TolC tripartite ABC efflux machinery was shown to confer resistance against diverse macrolide antibiotics, although its cognate function might be different. 40 Ann. N.Y. Acad such as ciprofloxacin (CIP), and aminocoumarins, such as novobiocin (NOV), 24 and the bacterial DNA-dependent RNA-polymerase, which can be inhibited by rifampicin (RIF). 22 Metabolic pathways, such as the folic acid biosynthesis pathway that leads to the production of tetrahydrofolate, the starting compound for the synthesis of purines and pyrimidines (the base groups of RNA and DNA), can be inhibited by sulfonamides (dihydropteroate synthetase) and trimethoprim (dihydrofolate reductase). 7 Interestingly, for some antimicrobial agents, additional lethal effects have been postulated, including the alteration of a given target protein to produce lethal products. 25 For instance, protein synthesis disruption in the presence of aminoglycosides leads to the formation of toxic peptides that destabilize/permeabilize the inner membrane. 26 Similarly, the effect of macrolide antibiotics appears to be based on context-specific inhibition of peptide bond synthesis. 27 β-Lactams and teixobactin have been found to induce cell wall autolysis, 28,29 and fluoroquinolones appear to convert gyrases into endonucleases by forming a covalent gyrase-DNA cleavage complex. 30

Acquired and intrinsic mechanisms of antibiotic resistance
Depending on the initial state of a given bacterial target, the inhibitory effect of antimicrobial agents can be counteracted by different intrinsic or acquired mechanisms of resistance that prevent antimicrobial compounds from binding to their targets. Enzymatic inactivation of antibiotics, for example, is a resistance mechanism typically acquired by horizontal gene transfer. Examples include aminoglycosides that can be inactivated by (substrate-specific) modifying enzymes, including acetyltransferases, phosphotransferases, and nucleotidyltransferases; 31,32 and β-lactam antibiotics that can be hydrolyzed by β-lactamases; the latter enzymes, in turn, can be inhibited by clavulanic acid. This is why a combination of both β-lactams and clavulanic acid is often used for treatment ( Fig. 1). 33 Alternatively, antibiotic drug targets can be protected by additional factors or altered in a way that prevents inhibitors from binding. 32 Target site alterations can be achieved by covalent modifications, by the expression of alternative subunits, or by mutagenesis. 34 For example, bacterial DNA gyrases or DNA-dependent RNA polymerases become insensitive to antibiotics by one of a number of different point mutations (single-site alterations). 35,36 Other fluoroquinolone resistance mechanisms result from the production of protecting Qnr proteins or acyltransferases that are effective against CIP and norfloxacin, as well as efflux pumps. 37 Under laboratory conditions, chloramphenicoland doxorubicin (DOX)-sensitive E. coli have been shown to acquire multiple mutations when cultured continuously under increasing concentrations of drugs. These mutations are often located in either the gene acrB, which encodes the multidrug efflux pump (MDEP) AcrB, or the promotor region of the gene encoding the MdfA (cmr) efflux pump, as well as in genes encoding the regulatory proteins for their expression (marR, acrR, and pcmr). 38 Nongenetic causes, such as alternative physiological states within a bacterial population, give rise to intrinsic antibiotic persistence. Persisters manage to evade the toxic consequences of antimicrobials possibly because of their reduced biosynthetic activity. Dormant pathogens are hard to eliminate and are considered to be associated with diverse latent microbial infections. 25 Furthermore, dormancy in E. coli has been also shown to be associated with a higher expression of various MDEP components (tolC, acrA, acrB, acrD, emrA, emrB, macA, and macB), which contribute substantially to the intrinsic resistance of Gram-negative bacteria. 39 The difference between resistance, persistence, tolerance, and other phenomena, such as heteroresistance, and how to measure these properties within a bacterial population have been summarized and discussed recently. 40

The intrinsic resistance of Gram-negative bacteria
MDEPs transport antimicrobial agents from the bacterial interior back to the environment. By reducing intracellular antimicrobial concentrations, they contribute to the intrinsic resistance of Gram-negative bacteria and support acquisition of other modes of resistance. 41 The efflux pumps of the Gram-negative bacterial subdivision are organized into a complex network with a mode of operation that is defined by Gram-negative morphology comprising two membranes, the inner and outer membranes.

Gram negatives-a two-membrane system reduces toxin permeation
In contrast to Gram-positive bacteria, which contain only an inner membrane and a much thicker cell wall, Gram-negative bacteria are surrounded by an additional outer membrane. The inner and outer membranes enclose the periplasm, an additional compartment next to the cytoplasm. Importantly, unlike the inner membrane, the outer membrane is highly asymmetric. While the inner leaflet consists of phospholipids, the outer membrane outer leaflet comprises lipopolysaccharide (LPS) (Fig. 1). The LPS molecules form, on the one hand, a hydrophilic permeability barrier at the bacterial surface, 7 while the acyl chains of the LPS in the outer leaflet-in conjunction with the inner leaflet phospholipidsconstitute a hydrophobic permeability barrier. The outer membrane is interspersed with porins to allow an exchange of substances with the environment (e.g., in order to absorb nutrients or facilitate exchange of ions). Porins contain a certain degree of selectivity (with reference to size and polarity) and facilitate transport of molecules between the environment and periplasm. 42 Altogether, the properties of the outer membrane favor the uptake of small, hydrophilic molecules over larger and more hydrophobic compounds. Most antimicrobials (see Fig. S1, online only), however, rather fall into the latter category. Diverse studies indicate that porins are involved in the uptake of various antibiotics. Antimicrobial drug resistance can be increased by porin gene expression downregulation or changes in the porin population composition, further reducing the permeability of the outer membrane. 42 Those antibiotics not directed against periplasmic targets must overcome the inner membrane as well to reach their targets in the cytoplasm (Fig. 1).

MDEPs from different transporter superfamilies counteract the influx of antimicrobials
The Gram-negative bacterial inner membrane, which separates the cytoplasm from the periplasm, is a distinct hydrophobic barrier with strongly regulated mechanisms of exchange. The presence of ion gradients (including the membrane potential) and the cytoplasm as source of chem-ical/redox energy enables active transport across this physical barrier. 43 Accordingly, the inner membrane contains the active components of MDEPs that either are secondary active antiporters driven by the proton motive force (PMF) or the electrochemical gradient of Na + ions, or are primary active exporters driven by ATP binding and hydrolysis. 44 Efflux pumps counteract the passive influx of their substrates along the concentration gradient by transporting them, in case of the singlecomponent transporters, from the cytosol to the periplasm, or as part of a two-membrane-spanning tripartite complex from the periplasm across the outer membrane. [45][46][47] Of note, it is important to emphasize that efflux pumps that transport toxic compounds across the outer membrane are heavily dependent on the permeability of the latter.
The efficiency of tripartite efflux pumps to confer resistance relies on the physicochemical properties of the transported compound not only related to its affinity toward the efflux pump, but also by its ability to cross the outer membrane. The composition of the latter (including different types of porins), which is different among Gramnegative bacteria, determines the rate of influx and hence the concentration in the periplasm or inner membrane.
This important concept has been addressed recently and summarized in a review by Zgurskaya and Rybenkov. 48 In this sense, resistance against the accumulation of the drugs in the cell is counteracted by the synergism of efflux pump(s) and the outer membrane. By establishing the equilibrium between influx and efflux, the concentration of antimicrobial substrates in the cytoplasm and periplasm is kept low, allowing MDEPs to contribute essentially to the intrinsic drug resistance of Gram-negative bacteria. 49 Known bacterial MDEPs belong to a total of six different transporter families, which are the ATP-binding cassette (ABC) superfamily, the small multidrug resistance (SMR) family, the major facilitator superfamily (MFS), the multidrug and toxin extrusion (MATE) family, the resistance-nodulation-cell division (RND) superfamily, and the proteobacterial antimicrobial compound efflux (PACE) family. 50 These will be briefly introduced below. Structures of representative transporters of each (super)family have been included in Figure 1, as indicated.

The ABC family
Members of the ABC superfamily are either homodimeric or (pseudo)heterodimeric ATPdriven transporters that are involved in the import and/or export of solutes. Each ABC transporter can structurally be subdivided into a transmembrane domain (TMD), which contains the substrate binding pocket, and two nucleotide-binding domains (which can be encoded by separate genes), responsible for ATP binding and hydrolysis. 51 In Gram-negative bacteria, ABC transporter efflux pumps can be either single-component transporters or part of a tripartite complex. The Thermus thermophilus ABC transporter efflux TmrAB, whose X-ray structure was solved at 2.7 Å resolution (PDB: 5MKK) 52 and predicted to function as a single (heterodimeric) component transporter, was shown to transport the antimicrobial dye Hoechst 33342 (HOE). 53 Recently, in the case of MacAB-TolC, a structural model of an assembled tripartite ABC transporter was, for the first time, obtained via single-particle cryo-EM at 3.3 Å resolution (PDB: 5NIK) 54 (see Fig. 1).

The families of SMR and PACE transporters
Transporters of the SMR family were shown to form parallel or antiparallel homo-or heterodimers with identical inward-and outward-facing states consisting of only four transmembrane (TM) helices in each subunit. 55,56 An optimized structure of the SMR family member EmrE (i.e., the apostructure and in complex with various substrates, including inhibitory peptide compounds) from E. coli was obtained by microsecond-long molecular dynamic (MD) simulations 57 on the basis of a low-resolution X-ray structure (at 3.8 Å resolution) in complex with tetraphenylphosphonium (TPP) (PDB: 4B5D). 58 The transporter was shown to confer resistance against acriflavine, ethidium (ETH), methylviologen, and benzalkonium (BEN), 46,59 as well as aminoglycosides. 60 Conformational dynamics of SMR transporters appear to be modulated in an allosteric fashion by protonation. 61 Homologs of the Acinetobacter baumannii AceI MDEP, which confer resistance against synthetic bactericidal agents, such as chlorhexidine, ACR, proflavine, and BEN, have been suggested to form the new PACE family of transporters. 50 Size and predicted secondary structure of these transporters, which are not found in E. coli, appear to be similar to SMR family-type transporters. 44

The major facilitator superfamily
The members of the extremely diverse MFS that are involved in multidrug efflux belong to the drug/proton (H + ) antiporter families 1 and 2 (DHA1/2), which differ in their substrate/proton stoichiometry. E. coli MdfA, a member of the DHA1 family, exports neutral or monovalent cationic compounds in exchange for a single proton. 62 MdfA substrates include CAT, tetracycline, norfloxacin, DOX, trimethoprim, ACR, ETH, and TPP. 62,63 An X-ray structure of MdfA with CAT bound to the central cavity of the transporter was solved at 2.5 Å (PDB: 4ZOW) in an inward-facing conformation. 64 Each MdfA single-component transporter is formed by two interconnected bundles of six TM helices, which were suggested to cooperate in an alternating access mechanism with ordered binding and release of proton and substrate. 62

The MATE family
Bacterial MATE transporters can be divided into the NorM and DinF subfamilies of 12 TM drug/Na + or drug/H + antiporters (xenobiotic-transporting eukaryotic MATE transporters are described in Miyauchi et al. 65 ). MDEPs of this family transport polyaromatic and cationic compounds, such as rhodamine 6G (R6G), ETH, and TPP. 44,66 Structures of Na + -dependent NorM subfamily transporters have been solved by X-ray crystallography at 3.6 Å resolution in cation-and substrate-bound states. 67,68 The X-ray structure of the single-component Bacillus halodurans DinF (DinF-BH) transporter of the DinF subfamily, in which R6G was bound to the central cavity via mostly hydrophobic interactions, was solved at 3.7 Å resolution (PDB: 4LZ9). 68 For the drug/proton antiporter, an alternating access mechanism with ordered binding and release of proton and substrate was proposed, similar to MFS transporters described above. However, MATE and MFS transporters have different topologies in their pseudo-symmetrical 6 TM helix bundles. 44 The RND superfamily

BOX 1. Diversity of RND-type transporters
As suggested by the terms resistance, nodulation, and cell division, RND transporters are involved in versatile cellular processes. On the basis of phylogenetic analysis, RND members are divided into a number of families 76 (www.tcdb.org). Members of the hydrophobe/amphiphile efflux 1 (HAE-1) and the heavy metal efflux (HME) families of Gram-negative bacteria transport and thereby confer resistance against diverse antibiotics and heavy metals, respectively. RND transporters of the Gram-positive HAE-2 and the Gramnegative HAE-3 subfamilies participate indirectly in resistance by transporting factors that contribute to the robustness of the cell wall (HAE-2) 69 or the outer membrane (HAE-3). 77 Members of the nodulation factor exporter (NFE) family have been suggested to export nodulation factors in order to stimulate the growth of root nodules as a (symbiotic) bacterial habitat. 69 With regard to cell division, Ptch1, a member of the dispatched family, was shown to bind the protein hedgehog, an important signaling factor in embryonic development and carcinogenesis. [78][79][80][81] In addition, members of the SecDF subfamily are involved in bacterial (and archaeal) protein translocation, where they act as membrane-integrated chaperons; 82 members of the eukaryotic sterol transporter (EST) family, such as the Niemann-Pick type C protein 1 (NPC1), play important roles in eukaryotic cholesterol homeostasis. 83 The diversity of RND transporters is also reflected in their oligomeric states. While members of the HAE-1 subfamily seem to exclusively form (homo)trimers, 69 the hopanoid transporter HpnN from Burkholderia multivorans, a member of the HAE-3 family, was found to form homodimers. 84 The SecDF protein of Thermus thermophilus 82 acts as a 12 TM heterodimer; and human NPC1, 85 as well as Patched, appears to be active as single RND subunits. Recently, the monomeric Mycobacterium tuberculosis RND mycolic acid/phosphatidylethanolamine transporter MmpL3 structure was solved. 86,87 The E. coli genome contains a total of 10 RND transporters (www.membranetransport.org). Characterized members are AcrB, AcrD, AcrF, MdtF, and the heterotrimeric MdtBC, all of which belong to the HAE-1 family, and the copper(I)/silver(I) transporter CusA, representing a member of the HME family. The heterodimeric protein-transporting SecDF belongs to its own family of transporters 82 (www.tcdb.org). The trimeric RND proteins from E. coli assemble into elongated tripartite complexes with six copies of a membrane fusion protein (MFP) and the homotrimeric outer membrane factors (OMFs) TolC or CusC. 88 in many different cellular processes as summarized in Box 1. All RND-type transporters show high structural similarity in their TMDs, which in general contain 12 TM helices (sometimes, by lateral extensions, 13 or 14 TMs). The TMDs are arranged in two pseudo-symmetric 6 TM bundles, each containing a large external loop between TM1 and TM2, as well as between TM7 and TM8. Those loops, which usually form a large soluble domain, can be very different in size and shape. 69 Also, the oligomeric state is different between members of the subfamilies (see Box 1). The TMDs of well-studied representatives contain a proton relay network that directs protons along the electrochemical gradient across the membrane, driving conformational changes in both TM and soluble domains. [69][70][71] Most RNDs appear to be PMF-driven antiporters; however, VexF from Vibrio cholerae has been shown to be dependent on the presence of Na + . 72 In Gram-negative bacteria, RND transporters of the hydrophobe/amphiphile efflux 1 (HAE-1) family contribute (essentially) to the intrinsic resistance toward antibiotics. In contrast to all other families of efflux pumps, the characterized HAE-1 (and heavy metal efflux) family members are only active (i.e., confer resistance) in conjunction with membrane fusion proteins (MFPs) and an outer membrane factor (OMF), with which they form elongated tripartite complexes. 44,69 For the AcrAB(Z)-TolC tripartite complex from E. coli, structural models of up to 5.9 Å resolution (PDB: 5O66) were elucidated on the basis of single-particle cryo-EM data 73 (see Figs. 1 and 2). It has been shown that the inner membrane components, for example, AcrB and AcrD, from E. coli interact with the same MFP (i.e., AcrA) and OMP (TolC). The latter outer membrane channel shows even a higher promiscuity, since it can partner with several MFPs and is integrated in various tripartite systems, such as RND-type AcrAB-TolC, AcrAD-TolC, AcrEF-TolC, but also with ABC-type transporters, like MacAB and HlyBD, as well as with EmrAB, a member of the MFS. It was demonstrated that the substrate specificity of the entire tripartite RND complexes is (predominantly) determined by the periplasmic loops of the RND inner membrane protein (IMP). 74,75 44 Ann

The AcrAB-TolC tripartite efflux machinery as the central element of a transporter network
Tripartite efflux pumps traverse the Gram-negative inner and the outer bacterial membrane with their IMPs and OMFs connected via their MFPs. In E. coli, the OMF TolC is shared by all tripartite RND efflux pumps of the HAE-1 family. 88 The constitutively expressed acrAB-tolC MDEP genes encode a machinery that contains three copies of the RND IMP AcrB, six copies of the MFP AcrA, and three copies of the OMF TolC 89 (see Figs the genes encoding components of the AcrAB-TolC MDEP (e.g., tolC or acrB) were heavily impaired in conferring resistance toward the drugs. 90,91 The importance of the intrinsic resistance toward multiple compounds conferred by this transporter system is well documented by the systematic analysis of the E. coli KEIO collection 92 grown under various stress conditions. 91 Transporters from the MFS and ABC superfamilies involved in tripartite systems appear to play only a minor role in antimicrobial resistance, since their genes are not expressed owing to stronger regulation, as is the case, for example, for the acrEF RND-type genes. 59 In addition, many tripartite systems confer only resistance toward a small number of drugs, as shown for the tripartite MFS pump EmrAB-TolC or the tripartite ABC transporter MacAB-TolC. 59,93 This might also be due to the cognate function of the latter systems; for example, EmrAB contributes to the efflux of free fatty acids, 94 whereas MacAB is involved in the secretion of protoporphyrin 95 and enterotoxin II, 96 and is suggested to transport LPS. 97 As demonstrated by many comprehensive studies, 59,90,91,98-100 the AcrAB-TolC MDEP confers (high level) resistance against a broad range of antimicrobial compounds, which are diverse in structure and size (structural formulas for some of these compounds have been compiled in Fig. S1, online only). These include different classes of antibiotics, such as macrolides, β-lactams, aminocoumarins, rifamycins, quinolones, oxazolidones, and tetracyclines, but also coloring agents, anticancer drugs, detergents, bile salts, and solvents. Apart from their differences, compounds recognized by AcrB have in common that they contain certain hydrophobic moieties. 70 More hydrophilic compounds, such as bi-anionic β-lactams (e.g., carbenicillin) and aminoglycosides (e.g., KAN) are not, or are poorly, transported by the AcrAB-TolC pump, but much better by the closely related AcrAD-TolC MDEP. [101][102][103][104] Whenever acrD expression is induced by environmental stimuli, 105 it can close this functional gap of the AcrAB-TolC system.
The broad substrate spectrum of AcrB corresponds to the sum of substrates transported by the diverse single-component MDEPs of the ABC, MFS, MATE, and SMR (super)families (in addition to those substrates with periplasmic targets, e.g., β-lactams). The individual single-component transporters each appears to recognize only a limited number of substrates. 59 However, a certain degree of redundancy is apparent, as some substrates are transported by more than one single-component transporter across the inner membrane. 46,59 This might be one reason why cells with individual chromosomal deletions of single-component transporter genes in most cases do not show higher susceptibility. 90 Single-component efflux pumps transport compounds across the inner membrane, while tripartite RND complexes, such as AcrAB-TolC or the homologous MexAB-OprM complex from P. aeruginosa, appear to mediate transport across the outer membrane as has been implied from in vivo 106 and in vitro 107 experiments.
Despite the fact that RND MDEPs span the inner bacterial membrane, they at least "prefer" to capture their substrates from the periplasm. 108 However, if bacterial cells lack all single-component transporter genes responsible for the efflux for a certain toxin across the inner membrane, sensitivity similar to bacteria lacking acrB is observed. [45][46][47] Clearly, if the single-component transporters are not present, the AcrAB-TolC system appears incapable of conferring resistance. These results might be interpreted as that the single-component transporters accumulate the drugs at the periplasmic side, in the outer leaflet of the inner membrane. If true, the increased drug concentration in the outer leaflet will be beneficial for the rate of drug transport by the AcrB component. Drugs with a high flip-flop rate equilibrate fast over both leaflets of the inner membrane, thereby decreasing the efficiency of binding and transport by AcrB. In addition, rapid equilibration with the cytoplasmic concentration (e.g., in the case of ETH) ultimately causes intoxication by the incoming drug. Single-component transporters shift this equilibrium toward the outer leaflet of the inner membrane and hence increase the drug concentration according to thermodynamics, depending on the number of protons coupled per drug. This hypothesis assumes that AcrB can only capture drugs from the outer leaflet of the inner membrane or from the periplasm. Structural features of the pump that support this assumption are described below.
Accordingly, drug efflux from the cytoplasm across the outer membrane, as illustrated in Figure 1, appears to be a two-stage process in which substrates are first transferred by various single-component transporters to the periplasm, 46 Ann. N.Y. Acad from where they are sequestered and transported across the outer membrane by AcrAB-TolC. 46 The fact that the latter step is only catalyzed by AcrAB-TolC for these substrates (see Fig. S1, online only) illustrates the importance of this complex pump machinery as the central element of the efflux network.

Tripartite RND efflux pumps from a clinical perspective
RND MDEPs, such as AcrAB-TolC from E. coli and Salmonella enterica, MexAB-OprM from P. aeruginosa, CmeABC from Campylobacter jejuni, and AdeABC from A. baumannii, are often upregulated in clinical isolates. 109,110 In addition, RNDs have been found encoded on mobile genetic elements, 108 conferring for instance nitrofurantoin resistance (by OqxAB) to uropathogenic E. coli. 111,112 The activity of MDEPs is also expected to support the acquisition of additional mechanisms of resistance 41,113,114 by reducing intracellular drug concentrations to sublethal levels that are supposed to enhance the genome-wide mutation rate. 115 The outer membrane, as the front defense line and essential part of the integral framework of tripartite efflux in Gram-negative bacteria, is a target of diverse peptide antibiotics, such as the cyclic polycationic polymyxins. Currently, polymyxins, which disrupt the outer membranes and kill bacteria by permeabilizing their inner membranes 116 (Fig. 1), are the strongest weapon in the fight against otherwise resistant Enterobacteriaceae. However, their efficiency appears to be impeded more and more by acquired resistance. 117,118 Not surprisingly, carbapenem-resistant A. baumannii, P. aeruginosa strains, and carbapenem and third-generation cephalosporin-resistant Enterobacteriaceae, including Klebsiella pneumoniae, Escherichia coli, Enterobacter spp., Serratia spp., Proteus spp., and Providencia spp., Morganella spp. (all Gram-negative organisms), were given top priority by the World Health Organization in the search for new antimicrobial agents. 117 The primary defense mechanisms of Gramnegative bacteria, which are based on a pronounced diffusion barrier with opposing selective properties (hydrophilic/hydrophobic), additional compartmentalization (periplasm/cytoplasm), and a multifunctional efflux machinery, complicate the development of new active substances that would ideally be well penetrating. 108 Alternatively, there are different approaches, such as making use of existing antimicrobials in combination with efflux pump inhibitors. [119][120][121] This challenging approach requires a more detailed understanding of the molecular binding and transport mechanisms of tripartite RND efflux pumps.

Structural properties and dynamics of the AcrAB(Z)-TolC complex
A first imaging method-derived structural model for the AcrAB(Z)-TolC tripartite complex was determined by Du et al. using single-particle cryo-EM. 89 The low-resolution structure (16 Å) gave a first impression of the structural organization of the elongated tripartite complex comprising homotrimeric OMF TolC, six copies of the MFP AcrA, the homotrimeric IMP AcrB, and three copies of the small single TM helix protein AcrZ. Further optimization of the experimental conditions led to the symmetrical (LLL) and asymmetrical (LTO) complex models at 6.5 Å (PDB: 5V5S) and 5.9 Å resolution (PDB: 5O66), 73 which provided more detailed insights into the interactions of the individual complex components, as described below (see Fig. 2). As these complexes were obtained by genetically engineered AcrA/AcrB fusions, it was important that Daury et al. (by negative-stain EM) showed that native AcrAB-TolC and MexAB-OprM complexes in nanodiscs appeared to have the same architecture. 122 Just recently, Tsutsumi et al. solved the MexAB-OprM structure at 3.6-3.8 Å resolution by single-particle cryo-EM. 123 Of note, the structure and, in particular, the interactions between MexA and OprM had been correctly predicted earlier by López et al. using sequence covariation analysis and microsecond-long MD simulations. 124 This and other successful predictive MD simulations used in the field of efflux pumps [125][126][127][128][129][130][131][132][133] emphasize that the predictive quality has been improving over the years, and will be an additional important asset in deciphering the structure/function relationship of efflux pumps.

Acriflavine resistance protein B
The first crystallographic structure of the 1049 AA (113.6 kDa) acriflavine resistance protein B (AcrB) from E. coli was solved at 3.5 Å resolution by Murakami  to represent the unoccupied resting state of the transporter. 135 Asymmetric (LTO) structures of AcrB were obtained under different crystallization conditions, 136,137 which also included the use of designed ankyrin repeat proteins (DARPins) as crystallization chaperones. 138 With the help of the latter, resolutions up to 1.9 Å (PDB: 4DX5) could be achieved. 139 In the asymmetric AcrB trimer, which appears to resemble the active state of AcrB, each protomer has a different conformation. The individual conformations were suggested to represent defined states of a transport cycle, which will be further discussed below.
According to their anticipated roles in the functional rotation-which are the (1) access of substrates, (2) their binding, and (3) their extrusion-the individual conformational states were named access, binding, and extrusion, 136  . Under crystallization conditions resulting in asymmetric AcrB trimers, several AcrB/substrate costructures were obtained in which substrate molecules were bound within one trimer to binding pockets of the L, T, or even to both L and T protomers. 136,139,142,143 AcrB can be structurally divided into a funnel domain (FD, or docking domain) important for AcrB trimerization, 144,145 a TMD responsible for the energy transduction to facilitate drug transport, and the porter domain (PD) that contains the binding pockets (proximal and distal, or access and deep binding pockets (DBPs)) and mediates substrate uptake, recognition, and translocation.

Acriflavine resistance protein Z
AcrB was found to stably interact with the small 49 amino acid (AA)-long (5.3 kDa) inner membrane acriflavine resistance protein Z (AcrZ), whose expression is coregulated with acrAB. 146 AcrZ essentially consists of a single tilted helix inside the membrane and thereby fits into a wide groove in the AcrB TMD. 73 The deletion of this small accessory protein led to a moderate reduction in activity with respect to some AcrB substrates. 91,146 Single TM proteins, such as AcrZ and YajC, shown to interact with AcrB, 73,146,147 appear to be present in some Enterobacteriaceae, and might function as allosteric modulators of RND MDEPs. 73

Acriflavine resistance protein A
The 42.2 kDa (397 AAs) periplasmic MFP acriflavine resistance protein A (AcrA) is produced as a precursor protein containing an N-terminal signal helix (residues 1-24). The mature protein, which lacks this signal peptide, is anchored to the inner membrane via palmitoylation (lipid anchoring) at Cys25. 148 MexA, an AcrA homolog from P. aeruginosa, has been experimentally shown to be palmitoylated 149 and was the first MFP for which the X-ray structure was solved at 2.4 Å. The structure revealed an elongated architecture formed by a membrane proximal (MP) domain, a β-barrel domain, a lipoyl domain, and an αhelical hairpin motif. 150,151 Shortly thereafter, the AcrA structure was published showing an identical architecture, 152 and, on the basis of a refined MexA structure, the thus-far elusive MP domain of AcrA was revealed. 153 In the latter study, comprehensive cross-linking studies with AcrA/AcrB (based on the refined MexA structure) led to the first insights into the AcrA/B interactions.
In the tripartite complex, as was shown from the first single-particle cryo-EM structures, AcrA forms a hexameric assembly 73,89 similar to the quaternary structure observed in crystals of MacA from Actinobacillus actinomycetemcomitans. 154 In the hexameric assembly, the AcrA hairpins, lipoyl, and β-barrel domains are densely packed, enclosing a central channel that merges at its lower end into the funnel on the upper side of the trimeric AcrB periplasmic domain. From the four AcrA subdomains, only the β-barrel and the MP domain are in direct contact to AcrB, whereby adjacent AcrA protomers differ in their interaction. Protomer I is in contact to the AcrB subdomains FC (or DC, Cterminal funnel or docking domain), PC1, and PC2 (C-terminal porter subdomains 1 and 2), while protomer II interacts with the FN (or DN, N-terminal funnel or docking domain) and PN2 (N-terminal porter subdomain 2) subdomains 73,89 (Figs. 2  and 3). The interaction of protomer II significantly differs from the interaction that was observed in the CusAB (Cu + /Ag + RND/MFP antiporter) complex structure, 155 which might be a preassembled state in the absence of the OMF (CusC). As has been shown for AcrAB(Z)-TolC 73  structures, 54 the MFPs (AcrA and MacA) bridge the periplasm and connect the IMP components to TolC via α-helical hairpin tip-to-tip interactions.
In doing so, the substantially longer hairpin of MacA appears to compensate the smaller (shorter) periplasmic domain of MacB (Fig. 1).

The outer membrane factor TolC
In contrast to other Gram-negative bacteria, such as P. aeruginosa, which contain many different OMFs, enterobacteria, such as E. coli, essentially have only a single tolC gene. TolC is shared between many different transporters covering different physiological processes, including the expulsion of metabolites, acid tolerance, cell membrane integrity, virulence, and antibiotic resistance. 157 TolC is synthesized as a 53.7 kDa (493 AAs) protein and cotranslationally exported into the periplasm, from where it is incorporated into the outer membrane, depending on the BamA/BamD complex. 158 As a homotrimer, TolC forms a 12-stranded β-barrel outer membrane pore domain whose opening merges into a 100 Å long α-barrel periplasmic domain that, in turn, consists of a total of 12 alternating continuous and pseudo-continuous α-helices (Fig. 2). TolC was crystallized at 2.1 Å resolution in a closed state in which the helices at the lower end of the α-barrel are twisted into each other like a closed iris diaphragm. 159 The closed state was shown to be stabilized by a network of intra-and intermonomer hydrogen bonds and salt bridges, whose disruption results in significantly higher ion conductivity, 160 and to structural characterization of (partially) opened states. 161,162 The presence of free (noncomplex associated) OMF in a closed state might prevent an uncontrolled exchange of (toxic) substances between the periplasm and the extracellular environment.

Structural and functional dynamics of the AcrAB(Z)-TolC complex
The structure of the apo (LLL) AcrAB-TolC complex (PDB: 5V5S) 73 and the low-resolution structures reported previously 89,163 were obtained from Cys-crosslinked or directly engineered protein fusions, which left the question open whether the complexes fully resemble the situation in the native tripartite efflux pump setup. Apo-complex formation was achieved with three of the six AcrA molecules Cys-crosslinked to AcrB, resulting in a structure at 6.5 Å. This tripartite setup was very similar, however, to structures of native AcrAB-TolC or MexAB-OprM complexes in nanodiscs. 122 In the apo-state, TolC was still in a closed conformation akin to that reported for the free protein, 159 and the AcrA protomers were only loosely packed without sealing the enclosed cavity toward the periplasm (Fig. 2). In the presence of substrate (e.g., puromycin (PUR)), however, the trimeric AcrB core complex adopted an asymmetric (LTO) conformation (PDB: 5O66). 73 Furthermore, the asymmetry in the AcrB trimer is associated with the repacking of the neighboring AcrA β-barrel and MP domains and in particular the reorientation of the α-helical coiled-coil hairpins, the flexibility of which was suggested important earlier on. 152 These reorientations, which involve all AcrA subdomains, lead to the iris-like opening of TolC and the sealing of the efflux channel toward the periplasm. Accordingly, the funnel of AcrB, from where the pumped substrates emerge from the AcrB O protomer, is in direct contact with the extracellular environment. In addition, the formation of an active pump system causes an axial compression of the entire complex by about 10 Å and the transition from a triple symmetry to a quasi-sixfold symmetrical interaction between AcrA and TolC. 73 In both AcrAB(Z)-TolC and the MacAB-TolC complexes, the opening of the OMF is associated with a tightly interlocked tip-to-tip interaction between the MFP hairpins and those of the TolC α-barrel domain. 54,73,124 However, these interactions involve only a few residues and are likely rather transient in order to rapidly close TolC and, presumably, prevent substrate reflux when AcrB is not actively pumping. In summary, the entire tripartite efflux pump appears to be a highly allosteric machinery in which conformational changes in the IMP are coupled to the OMF via the intermediate MFPs. 44,164 Concerning tripartite complex dynamics, the homologous MexAB-OprM complex was reconstituted into a two-liposome system, and substrate transport was associated with a rapid PMF consumption 107 followed by a subsequent dissociation of the complex. 165 Considering these findings and the observation from Kralj et al. that in E. coli a rapidly spiking (∼1 Hz) PMF coincided with the efflux of the potential AcrB substrate tetramethyl rhodamine methylester, 166 we and others have suggested that the AcrAB-TolC MDEP operates in a discontinuous fashion. 167 Some uncertainty still remains concerning the tripartite setup, since clear cross-linking between AcrB and TolC has been observed. 168 Moreover, the expression of genetic fusions between acrA and acrB in an E. coli acrAB strain resulted in drug resistance. Since trimerization (and no higher oligomeric state) of AcrB has been established, the latter results imply that stoichiometric AcrA/AcrB (e.g., 1:1) is sufficient for pump activity, and immunoblot analysis indicated that no free AcrA (e.g., as result from proteolytic cleavage of the AcrA-AcrB fusion) is present. Nevertheless, some degradation or transcriptional/translational separation of the AcrA-AcrB fusion must have occurred, since "free" AcrB was observed in the same samples. 200

Architecture of the RND core component-the AcrB inner membrane protein
The model of the asymmetric AcrB X-ray structure solved at 1.9 Å (PDB: 4DX5) 139 displays the homotrimeric core component of the AcrAB(Z)-TolC RND MDEP (Fig. 3). It can be structurally subdivided into an FD, a PD, and a TMD (Fig. S2, online only). The N-terminal (FN) and the Cterminal (FC) halves of the FD that constitute the funnel-like depression in the upper part of the periplasmic domain are involved in AcrB trimerization via a long intermonomeric loop, which penetrates the adjacent protomer. 144,169 Since the FDs of each protomer appear structurally identical between the different conformational states, 137 these are likely to form a scaffold for the conformational transitions occurring in the PD and TMD during the transport cycle. The 36 TM helices containing TMD are more loosely packed and are enclosing a central cavity. This cavity is occupied with lipids forming a hexagonal arrangement 170 and suggested to be involved in the interprotomer conformational transduction. The TMD of each protomer is arranged in two pseudo-symmetric bundles (TM1-TM6 and TM7-TM12). Each of the TM bundles comprises a central TM helix (TM4 and TM10), which contains titratable residues of a proton relay network (AAs D407, D408, and K940). These residues and R971 on TM11 are involved in sequential protonation and deprotonation and are central to the energy conversion required for substrate transport. Accordingly, the TMD and PD, the latter formed by the two periplasmic loops between TM1 and TM2 as well as between TM7 and TM8, are conformationally coupled. 128 Each PD can be subdivided into two N-terminal and two C-terminal subdomains (PN1, PN2, PC1, and PC2). These subdomains act as rigid bodies adapting different relative orientations during the conformational cycling and constitute the structural differences between the individual conformers (L, T, and O) of the asymmetric structure. In the L and T conformation, the PC1 and PC2 subdomains form a vertical cleft toward the periplasm. The interspace between these subdomains also comprises the access pocket (AP). A further substrate pocket, the DBP, is composed of PN2 and PC1 (and PN1) subdomains, and is present in the T state only. Both pockets are separated from each other by an eleven amino acid long switch loop motif, a feature important for the transport of substrates (Fig. 3).
Computational structure analysis gave rise to diverse substrate entry channels in the L and T conformation of the asymmetric trimer. Channel 1 (CH1), which was observed in the TMD/PC2 interface of the T protomer above the TM8/TM9 groove, 138 appears to guide substrates from the outer leaflet of the inner membrane to the AP-DBP transition zone. The cleft pathway via CH2, observed about 15 Å above the putative membrane plane in both the L and T conformation, provides alternative access to the AP 136,137,139,142 (Fig. 3). Some compounds, as has been suggested from symmetrical (LLL) AcrB/substrate costructures, 171,172 might also enter the central cavity via the vestibules between the protomer interfaces. For planar aromatic cations, an alternative input channel (CH3) was predicted for the T protomer, which would establish a direct connection between the central cavity and the DBP bypassing the AP and the switch loop. 173 In the O conformation, the cleft, AP, DBP, and all entry channels are closed, but an exit tunnel/gate between PN1 and PN2 is present. The exit gate and the exit channel (CHe), exclusively open in the O state, connect the transporter's interior to the AcrB funnel and the AcrA/TolC exit duct of the tripartite complex that is in contact with the extracellular environment.

Remote alternative access drug/proton antiport in AcrB
The individual protomer conformations (L, T, and O) in the asymmetric structures of AcrB were interpreted as intermediate states of a transport cycle in which each protomer undergoes consecutive conformational changes. 136,137 In order to visualize the molecular processes within the PD (PN1, PN2, PC1, and PC2 subdomains) and the conformationally coupled TMD (TM1-12), individual protomers of the high-resolution 1.9 Å resolution structure (PDB: 4DX5) 139 were superimposed via their rigid FDs (PD) and TM4-6 (TMD), respectively, as  previously described 128,137 (Fig. 4). The coupling between the proton-translocating TMD and the substrate-translocating PD appears to be carried out by vertical shifts of TM2, which is directly con-nected to the PN2 subdomain, and the lateral movement of TM8 as consequence of the disengaged state of the TMD bundles. It has been recently suggested that the coil-to-helix and helix-to-coil transitions in TM8, which is connected to the PC2 subdomain, are a consequence, rather than the cause, of interdomain coupling. 173 Nevertheless, the TM8 helix itself reacts to the protonation of the D407 and D408 residues, causing a lateral shift and, consequently, movement of the PC2 and PN1 subdomains (Fig. 4). Rearrangements in the hoisting loop (the region of TM8 where the coil-to-helix transition happens), also called the convertible TM-section, appear to be involved in the opening and closing of CH1.
Symmetrical LLL structures 73,134 and MD simulations 125 suggest that the L conformation is the preferred state in absence of substrates. Assuming that the L conformation is the starting point of a transport cycle, the AP between PC1 and PC2 subdomains will be open, while the DBP between PC1 and PN2 subdomains is closed. As a consequence of the L-T transition, the subdomains PN1 and PC1 swing apart, causing the DBP to open and the AP to become smaller. The binding of substrate molecules to the DBP stabilizes the T state, which is coupled to a downshift in TM2 and further, smaller TM displacements that make the TMD accessible for water molecules from the periplasmic side (Fig. 4). The formation of a water molecule network enables the protons to access the central titratable residues (D407 and D408) whose protonation drives the T-O transition. 128 Strong conformational changes within the TMD, such as a TM2 upshift, cause the reorientation of the PN1 and PC2 subdomains that did not move significantly during the L-T transition. The tilting of the PN1 subdomain in the direction of the adjacent protomer opens the exit gate between PN1 and PN2 subdomains through which substrates can be extruded toward the exit funnel as a consequence of the closure of the DBP and AP. Proton release to the cytoplasm leads to a widereaching revision of the shifts within the TMD. However, the position of TM2 is hardly affected, which is presumably why conformational changes in the PD during the O to L transition are restricted to PN1 and PC2 subdomains only. The opening of the cleft (AP) and the closing of the exit gate finally bring the system back to its initial L state (Fig. 4). Accordingly, the processes within the PD and TMD of each protomer can be considered as intermediates of two remote alternating access transport mechanisms. 128 As reversible Cys-crosslinking studies 174 and single-protomer exchange experiments 71 have shown, there is a mutual depen-dence not only within the subdomains of a protomer, but also between the individual protomers.
A single AcrB subunit with a critical mutation in the proton relay network (D407A) was sufficient to render the entire tripartite complex inactive. 71 Accordingly, the locking of a single subunit seems to prevent the others from completing their conformation cycle as well. These findings and the positive cooperativity observed in kinetic experiments with different penicillin antibiotics 175 support the bisite activation hypothesis in analogy to the F 0 F 1 ATPase, according to which the release of a substrate requires the binding of another substrate at a different binding site. 70 Different subprocesses were also suggested to take the advantage of each other when conformational transitions in the TMD were speculated to move FUS in an elevator-like mechanism from a membrane-associated state in the direction of a further potential entry site. 176 The heterotrimeric RND efflux pump MdtB 2 C might be considered as a special example of mutual dependence between individual protomers. While MdtC seems to be predominantly involved in substrate binding and transport, the MdtB components appear to translocate protons in order to provide the required energy. 177,178 Interestingly, for CmeB (an AcrB homolog from C. jejuni), it was proposed that individual protomers operate independently of each other. 179 However, the examination of the isolated IMP component might not necessarily resemble the conformational dynamics within the assembled tripartite complex.

Substrate recognition and inhibition of the AcrB RND core component
As elucidated by Nakashima et al. 142 and Eicher et al., 139 the PDs of the asymmetric AcrB trimer harbor two large internal substrate (binding) pockets that are separated from each other by an 11amino-acid-containing switch loop motif (residues 613-623) (Fig. 5). The AP and DBP are created by rigid body movements of the porter subdomains as elaborated above. In the L and T states, the PC1 and PC2 subdomains comprise the AP, while the DBP is formed by the PN2, PC1, (and PN1), exclusively in the T state. The boundaries of the pockets thus essentially consist of the rigid β-sheets of the subdomains, but also the more flexible loop sequences of the PC1-PC2 "connecting loop" (residues 666-679) at the bottom of the AP, the serine-rich "serine loop"   F178 and F628 (shown in orange) (J, M, N). Further MBX derivatives, like MBX2319 (blue, PDB: 5ENO), MBX2931 (marine, PDB: 5ENP), and MBX3135 (slate, PDB: 5ENR), are found in an orientation highly similar to MBX3132 (K). All residues within a radius of 3.5 Å from the bound ligand are indicated in Figure 6 and assigned to their potential interaction partners in Table S1 (online only).

The individual costructures can be considered as sequential intermediates along the substrate pathway from the outer AP to the inner DBP each representing a local energetic minimum for the interaction between AcrB and its ligand. The switch loop, indicated in dark red, contributes to the recognition in both pockets. The DBP can be subdivided into an upper groove, the narrow-elongated area of the DBP between the β-sheets of the PN2 and PC1 subdomains, and a lower more open cave, which is limited toward the lower side by the transmembane domain (TMD) (H). While MIN is bound to the groove and PUR is mainly bound in the cave, the inhibitors ABIPP (J) and MBX3132 (M) interact tightly with both DBP regions, including extensive π-π stacking interactions with
(residues [131][132][133][134][135][136], and the switch loop in both the AP-DBP interface as well as the short sequence between L177 and Y182, which contributes to the DBP. The flexibility provided by the loop sequences and the sum of the side chain degrees of freedom likely contribute to the enormous plasticity and multifunctionality of the internal pockets. 126 The plasticity of the internal pockets is also reflected through the multitude of AcrB/substrate costructures. Figure 73,121,139,142 These local energetic minima in the AP and DBP appear to be associated with critical steps in the transport cycles of the individual substrates as suggested by competition studies with fluorescent dyes, where compounds that are likely to dwell at the same subsites interfere with each other during transport. 142,[180][181][182] The binding of dimeric DOX to the outer AP was suggested to represent a preliminary stage to the binding of the monomeric ligand to the DBP in the sequential T conformation, 139 and might be seen as an example how the decision to finally extrude a compound is likely made in several successive steps (multistage recognition). While RIF appears to be stalled at the switch loop, ERY and PUR, which were both crystallized in a location below the switch loop (the former in the L and the latter in the T protomer), might represent states close to the conformational transition.
A closer look at the various AcrB/substrate costructures reveals how many residues are located within a radius of 3.5 Å around the individual ligands, which makes them likely to be involved in substrate binding. These residues are shown for the AP and DBP in Figure 6. A more precise subclassification of the residues into the outer and the inner AP, the AP-DBP interface at the switch loop, and the DBP groove and cave as well as an assignment of the (potentially) interacting ligands are given in  S1 (online only). In a direct comparison of the two internal substrate pockets, the DBP is more hydrophobic, and therefore likely has a higher affinity to the predominantly hydrophobic substrates and probably is the endpoint of their uptake. 126 However, especially for larger substrates, such as RIF, ERY, or the DOX dimer, the accessibility of the DBP within the crystallographic resolvable conformational states is limited because of constriction caused by the switch loop. 139,142 As summarized above, the stabilization of the T state is expected to be mechanistically decisive for the energization of transport.
From the comparison of the different T state costructures (Fig. 5H), the impression arises that individual substrates (MIN, DOX, R6G, and PUR) stabilize the T state by binding to different areas of the DBP. On the basis of docking 183 and MD simulations, 184 substrates are classified to preferentially bind to the upper groove, the lower cave, or both regions of the DBP (mixed binders). The groove region is formed by converging β-sheets of the PN2 and PC1 subdomains in the upper part of the DBP, while the cave (limited downward by the TMD) provides a much larger cavity. On closer examination of the computer simulations, several energetically similar binding poses were predicted. Moreover, comparison of the two X-ray AcrB/DOX costructures (PDB: 2DR6 and 4DX7) displays significantly different binding modes with approximately orthogonal ligand orientations. 136,139 Barring the possibility that the lower resolution structure prohibited the correct orientation of the bound ligand, the interaction of DOX with residues in a large pocket in several energetically de facto equivalent ways assumes ligands to oscillate between individual binding sites. 182 The availability of multiple binding modes would also increase the probability for a binding pocket to accommodate more than one ligand at the same time, as suggested earlier. 185 Substrate recognition by AcrB and other proteins with multidrug binding and transport properties appears less based on a precise network of hydrogen bonds and other specific interactions than predominantly entropically driven by the hydrophobic effect (and electrostatic attraction), which is less dependent on a defined geometry. 186 However, the DBP of AcrB is not just simply hydrophobic. The ligands in the costructures mentioned above are in contact (closer than 3.5 Å) to a few very hydrophobic residues (V139, I277, and V612), but also to some charged (K292 and R620) and weakly polar residues, such as S134, Q151, Q176, and N274 (see Fig. 6). The main interaction occurs, however, by the particularly prominent aromatic residues of the phenylalanine cluster (F136, F178, F610, F615, F617, and F628) and Y327. In all, the DBP is predominantly lined by weakly hydrophobic and weakly polar residues that are predicted to create a versatile and adaptable environment with numerous so-called multifunctional sites. These dynamic local microenvironments are suggested each to provide (inducible) complementarity toward parts of the substrate structure. Accordingly, their combination is thought to result in flexible and overlapping substrate binding sites. This flexibility in interaction might also explain the high substrate promiscuity and likely facilitates efflux. 126,187 In contrast to the binding of substrates, the potent AcrB inhibitors of the pyridopyrimidine (ABI-PP) ( Fig. 5I-K) and the pyranopyridine classes (e.g., MBX3132) (Fig. 5L-N) tightly bind to the entire DBP 121,143 and were shown to significantly potentiate the activity of known antimicrobials without exhibiting membrane disruption or antibacterial activity. [188][189][190] Extensive interactions to both the groove and the cave region (see Table S1, online only) were suggested to prevent substrates from entering the DBP and/or conformationally arrest the protomer in the T state. 121,143 The aromatic parts of the inhibitory ligands are thereby engaged in intensive π-π stacking interactions with residues F178 and F628 (Fig. 5J-M), whereby the pyranopyridine core structures of the MBX derivatives (MBX2319, MBX2931, MBX3132, and MBX3135) show essentially similar binding properties (Fig. 5K). 121 In addition to the apparently more complex and defined binding modes of these inhibitors, including water-mediated protein-ligand interactions, computational structural analysis suggests that the binding affinities of these inhibitors significantly exceed those of AcrB substrates. 121,189 Of note, substrates also appear to bind via hydrogen bonding interaction mediated by water molecules, as is apparent for the binding of MIN, where the 1.9 Å high-resolution AcrB/MIN costructure reveals that structural water molecules support the binding of this substrate. 139 Nevertheless, in contrast to inhibitor binding, the recognition of substrates appears to be based less 56 Ann. N.Y. Acad on specific interactions and more on their suitable physicochemical properties. Accordingly, for example, the residues of the phenylalanine cluster were classified as (at least partially) redundant, since individual alanine substitutions were not associated with major functional effects. 99 As an exception, the F610A substitution has been associated with more general functional impairments that, however, seem to have mechanistical causes. 129 More significant changes in substrate specificity appear to require more global changes in the properties of the binding pockets, as reflected, for example, in the comparison of the binding pockets of AcrB and AcrD. 126

An overview of molecular determinants for substrate binding and transport
Substantial mutagenesis studies have been conducted on acrB, leading to amino acid residue substitutions in the AcrB protein with concomitant effects on its ability to either confer resistance to E. coli or its ability to transport fluorescent dyes. A dataset of 430 publications mentioning AcrB was obtained by querying the NCBI PubMed database (April 2019) for "acrb and coli" using Biopython, 191 followed by manual download of the individual papers. In the last 17 years, 50 papers have been published describing acrB mutagenesis or acrB mutants, leading to AcrB variants with wild-type-like or other than wild-type properties. Using text mining, we constructed an initial list of residues mentioned in these publications in a diverse variety of formats, such as D407 and Asp407. Manual curation of this list, combined with extraction of the observed phenotypes, resulted in a comprehensive table comprising all substitutions (from N-to C-terminus, including multiple-site exchanges) mentioned in the respective publications, and their effects on AcrB (AcrAB-TolC) activity (in percent of wildtype activity) on the tested substrates (e.g., drugs, dyes, and solvents). Activities were determined using various methods mentioned accordingly (see Table S2, online only); whenever described, the expression levels of acrB (or the extent of synthesis of the AcrB protein) are indicated as well. The location of the substitution is given (indication of the AcrB subdomains), as well as whether the characterized mutants were expressed from plasmid or the chromosome. Moreover, the PMID number links the substitution with the respective publication (see http://goethe.link/AcrBsubstitutions).   Table S3, online only). 58 Ann. N.Y. Acad We found that 235 of 1049 amino acids in AcrB were subjected to exchange, with the majority of the substitutions (65%) involving replacement for alanine and cysteine residues. Reports on deletions inside acrB (removal of one or more codons) have not been included in this study. To discuss the various substitutions and their effects on AcrB activity (i.e., minimal inhibitory concentration (MIC) value changes or change in dye efflux activity), we structurally divided the AcrB protein in its TMD, PD (including the AP, DBP, and switch loop), and FD (a.k.a. docking domain) (Fig. S2, online only). The effects of single-site substitutions in the TMD (Fig. S3, online only), PD (Fig. 7A), and FD (Fig. S4, online only) are displayed in color code, resulting in a heat map. AcrB is highly insensitive toward substitutions; that is, single-site exchange of amino acid residues normally does not result in a severe phenotype (compared with wild-type activity). Hence, only a few killer mutations (i.e., substitutions leading to an E. coli acrB phenotype for all tested substrates) have been identified thus far. These substitutions are mainly located in the TMD (Table S2, online only) and include the characterized side chains involved in the binding and release of protons (transport) across the TMD (D407, D408, K940, and R971). 128,192,193 In the AcrB PD, substitutions are less detrimental, but the most prominent null phenotypes are caused by substitutions within the switch loop, which will be discussed below. One sensitive residue located in the FD, R780, was found to be important for stabilization of the AcrB protein. A suppressor mutation found by random mutagenesis within the gene encoding inactive R780A AcrB (M774K) restored the activity to almost wild-type levels. 194 The most interesting substitutions are within the AP and DBP, as these residues are involved in the binding and/or transport of drugs (and inhibitors) (Figs. 6 and 7A). As depicted in Figure 8, we selected six residues from the AP and five residues within the DBP involved in multiple drug binding (Table S1, online only). We subsequently derived the MIC values against various antibiotics of cells harboring the indicated single-site substitution variants from Figure 7A. As shown in Figure 8, the antibiotics were sorted according to three different properties, that is, molecular weight (MW), partition coefficient (logP), and minimal projection area (MPA, the minimal rectilinear parallel projection of a surface of any shape onto a plane, see www.chemicalize.com). Two interesting observations can be made.
First, mutagenesis/substitution experiments have been made to address specific questions within the respective studies involving substitution at the indicated residues, but analysis of substrate-specific effects is lacking (or incomplete). This is emphasized in Figure 8 by the amino acid side chains of the AP colored in gray (F664C, F666C, L668C, and T676C) and the substitution effects were tested against macrolides and aminocoumarins only. Consequently, a general tendency on substrate-specific effects cannot be deduced for the AP.
Second, for the DBP, there is a tendency for larger substrates or substrates with higher MPA values to be more affected by substitutions (except macrolides). There does not appear to be a correlation between the partition coefficient (logP) of the drug and the effect of the substitution. However, to support the above claims and to identify the molecular determinants for substrate (or inhibitor) specificity, a more systematic approach is needed, including alternative descriptors of hydrophobicity. Current hypotheses on substrate selection are mostly based on assumptions and are in need of systematic and consistent experimental procedures followed by stringent analysis. Furthermore, phenotypical data obtained by characterization of AcrB variants in whole cells should be interpreted carefully for each substrate with respect to its physicochemical parameters and the outer membrane permeability. 48 The switch loop motif separates the AP and the DBP The 11-amino-acid-containing switch loop (residues N613-N623) is a part of the PC1 subdomain and is a drug transport modulation loop. It emerges from a β-sheet (Cβ2) with a sharp downward curvature ( Fig. S5A and D, online only). According to the relative orientation, the switch loop can be structurally divided into two halves, that is, into a PC1 (N613-F617) and PN1 proximal parts (A618-N623) (Fig. S5A-C, online only). 195 The downward orientation of the PC1 proximal part of the switch loop, in the direction of the membrane plane, continues with residues G614, F615, and G616 until the peptide chain reaches F617 at the tip of the loop.  the PN1 proximal residues A618, G619, R620, and G621. Residues Q622 and N623 form a further wider bend to connect the end of the switch loop to the antiparallel strand Cβ3. The loop motif participates in the binding of substrates to both pockets as shown by the X-ray costructures. 139,142 During the functional rotation of the individual protomers, the switch loop likewise switches between the three distinct conformational states L, T, and O. As can be seen from the comparison of the wild-type AcrB L (blue) and T (yellow) states (PDB: 4DX5), the switch loop is shifted toward the AP during the L-T transition, contributing to a larger DBP in the T state ( Fig. S5D and E, online only). The comparison of different loops in the L protomer from apo, RIF, ERY, and DOX dimer (co)structures reveals a high conformational adaptability and flexibility at the tip of the loop close to F617 ( Fig. S5F and G, online only). By contrast, the different T states in complex with MIN, DOX, PUR, and R6G were far more invariant when superimposed ( Fig. S5H and J, online only). The switch loop contains four glycine residues, where each PC1 proximal residue (G614 and G616) faces a PN1 proximal counterpart residue (G621 and G619) in a highly symmetrical arrangement. These residues were considered of special importance for the flexibility of the loop to allow substrates to overcome the transport bottleneck between the AP and DBP. 139,142 The observed lower inherent macrolide transport capacity of MexB 75 could be attributed to position 616 within the switch loop (Gly in AcrB, and Asn in MexB). Indeed, E. coli AcrB readily lost most of its macrolide transport capacity, mimicking that of MexB, by a single G616N switch loop substitution. 196 The structural characterization of this variant indicated a loop conformation in the L state that was significantly shifted toward the AP, essentially resembling the situation in the T state of the wild-type protein. 139 On the basis of the assumption that the G616N L loop (PDB: 4DX6) would collide with the DOX dimer bound to the AP of the wild-type AcrB (PDB: 4DX7), it was postulated that the G616N variant switch loop was more rigid compared with the wild-type switch loop, and that it might interfere with the uptake of larger substrates by the L protomer. 139 This finding was further corroborated, as it was shown that the G616N variant could only effectively transport substrates with a low MPA, not necessarily a property related to molecular mass (Fig. S6, online only). It supported the notion that the side chain substitution did result in a lower flexibility of the loop, rather than from involvement of the Asn side chain in preventing macrolide binding. 197 A much stronger intervention in the loop characteristics was made when each of the two switch loop glycine residues were exchanged to proline, resulting in the symmetrical exchange variants G614P_G621P and G616P_G619P. Those variations resulted in a de facto inactivation of the entire efflux pump 142,195 (Fig. 7A; Figs. S5 and S7, online only). Moreover, even the G614P or G616P single-substitution variants resulted in null phenotypes (Fig. S7, online only). Surprisingly, however, substitution of the switch loop phenylalanine side chains F615 and F617 to alanine, in the Gly -> Pro background, recovered some of the activity depending on the substrate and location of the Gly -> Pro substitution. Whereas the F615A_F617A double substitution in the wild-type AcrB background only showed a major effect on sodium dodecyl sulfate (SDS) and NOV resistance (Fig. S7, online only), it rescued the G614P phenotype for ERY and TPP; for the G616P variant, however, the removal of the phenylalanine side chains (Fig. S5, online only) resulted in a rescue for almost all tested substrates (R6G, SDS, TPP, ERY, OXA, ETH, LIN, and NOV; see Fig. S7, online only). The larger rescue effect for the G616P variant might be explained by the immediate vicinity of the F615 and F617 residues flanking the proline at position 616. For the G616P_G619P double-substitution variant, the resistance phenotype was rescued as well by Ala substitutions of F615 and F617, albeit the effect (compared with the rescue phenotype of the G616P variant) was rather low (Fig. S7, online only).
Deletion of the entire switch loop also does not impair the binding of ERY, and even residual resistance activity against ERY and full activity in the presence of DOX was shown for a switch loop-less variant. 198,199 This appears somewhat in contrast compared with the nearly null-phenotype (against eight different AcrB substrates) shown for another switch loop variant, 195 but the variants are not identical and the conditions tested were different. The results shown for the switch loop thus far imply that especially its flexibility is important for optimal drug recognition and transport. Nonflexible loops impair transport; however, this is most likely due to the obstruction of the bulky switch loop side chains (i.e., F615 and F617), as a rigidified switch loop without these bulky side chains shows the recovery of activity up to wild-type levels, depending on the tested substrate. 195

Closing remarks
After 17 years of addressing individual questions on drug and proton recognition, we are in need of a much more systematic analysis for understanding the molecular basis of drug and proton recognition and transport. This will be, in the end, important for understanding the overall phenotype of "antibiotic resistance," where the contribution of the efflux pumps is considered not so marginal. After all, a very important basic rule applies for future directed research: understanding the fundamental principles of drug binding and transport by major efflux pumps is necessary for understanding the principles of the phenotype of multidrug resistance.

Acknowledgments
Work in the Pos lab is supported by the German Research Foundation (DFG-SFB 807, Transport and Communication across Biological Membranes and DFG-FOR2251, Adaptation and Persistence of the Emerging Pathogen Acinetobacter baumannii), the DFG-EXC115 (Cluster of Excellence Macromolecular Complexes at the Goethe University Frankfurt), and the Innovative Medicines Initiative Joint Undertaking Project Translocation (IMI-Translocation GA_115525). We also thank the European Union Seventh Framework Programme (FP7/2007-2013) and the European Federation of Pharmaceutical Industries and Associations (EFPIA) for in-kind contributions. M.S.K. is a recipient of a fellowship from the Friedrich-Naumann-Stiftung für die Freiheit (https://fnst.org). We also would like to thank Attilio Vargiu, Paolo Ruggerone, and Giuliano Malloci for valuable suggestions, corrections, and comments on the manuscript.

Supporting information
Additional supporting information may be found in the online version of this article.        Table S1. Assignment between bound substrates and potentially interacting residues. Table S2. http://goethe.link/AcrBsubstitutions. Table S3. Comments on the marked positions (with a plus (+)) in the heat maps of the funnel, porter, and TM domain ( Fig. 7 and Figs. S3 and S4).