The structure of the Aquifex aeolicus MATE family multidrug resistance transporter and sequence comparisons suggest the existence of a new subfamily

Significance

Multidrug and toxic compound extrusion (MATE) transporters play a significant role in the active export of drugs and toxic compounds in pathogens. The transport mechanism is still not fully understood. A comprehensive phylogenetic analysis indicates that Aq_128 of Aquifex aeolicus is a member of a subfamily of MATE transporters together with transporters from ε-Proteobacteria and Archaea. The high-resolution crystal structure (2.0 Å) in combination with drug efflux assays, mutagenesis studies, docking experiments, and molecular dynamics simulations reveals structural features that differ from previously described MATE transporters, and sheds light on their transport mechanism.

Abstract

Multidrug and toxic compound extrusion (MATE) transporters are widespread in all domains of life. Bacterial MATE transporters confer multidrug resistance by utilizing an electrochemical gradient of H⁺ or Na⁺ to export xenobiotics across the membrane. Despite the availability of X-ray structures of several MATE transporters, a detailed understanding of the transport mechanism has remained elusive. Here we report the crystal structure of a MATE transporter from Aquifex aeolicus at 2.0-Å resolution. In light of its phylogenetic placement outside of the diversity of hitherto-described MATE transporters and the lack of conserved acidic residues, this protein may represent a subfamily of prokaryotic MATE transporters, which was proven by phylogenetic analysis. Furthermore, the crystal structure and substrate docking results indicate that the substrate binding site is located in the N bundle. The importance of residues surrounding this binding site was demonstrated by structure-based site-directed mutagenesis. We suggest that Aq_128 is functionally similar but structurally diverse from DinF subfamily transporters. Our results provide structural insights into the MATE transporter, which further advances our global understanding of this important transporter family.

Living organisms have evolved a wide range of mechanisms to protect themselves against external environmental stress factors such as toxins and xenobiotics or internal metabolic waste products (1, 2). Multidrug resistance (MDR) transporters confer resistance against antibiotics and anticancer drugs to infectious pathogens and cancer cells (2–5), making them a central factor in two serious global health threats. Currently known MDR transporters belong to seven families (4): the adenosine triphosphate–binding cassette (ABC) superfamily, multidrug and toxic compound extrusion (MATE) family (6), major facilitator superfamily, resistance–nodulation–division superfamily, small MDR family, p-aminobenzoyl-glutamate transporter family (7), and the recently identified proteobacterial antimicrobial compound efflux family (8). Apart from primary-active ABC transporters, the other six families contain secondary-active transporters utilizing transmembrane electrochemical ion gradients to drive transport.

MATE transporters can use either Na⁺ or H⁺ as coupling ions to drive the extrusion of various structurally diverse, usually polyaromatic and cationic compounds, which can diffuse into the cytosol across the cell membrane (9–15). Members of the MATE transporter family are widely distributed in all three domains of life. Human MATE transporters, which are highly expressed in kidney and liver, mediate the proton-dependent excretion of various cationic drugs into urine and bile (16–19), thereby decreasing the plasma concentrations of these drugs and thus lowering their therapeutic efficacies (20). On the other hand, they help to reduce drug-induced nephrotoxicity by affecting the pharmacokinetics of anticancer agents in platinum-based chemotherapy (5, 21). In plants, MATE transporters are involved in the secretion of a variety of secondary metabolites for detoxification or as a defense strategy against herbivores and pathogens (22, 23). Bacterial MATE transporters are primarily employed as xenobiotic efflux pumps, contributing to the MDR phenotype of important human pathogens (24, 25).

The MATE family belongs to the multidrug/oligosaccharidyl-lipid/polysaccharide exporter superfamily (26). On the basis of amino acid sequence similarity, the MATE transporters have been previously classified into three major subfamilies: the NorM, DNA damage–inducible protein F (DinF), and eukaryotic MATE (eMATE) subfamilies (6, 25, 27). Bacterial and archaeal MATE transporters, subdivided into the NorM and DinF subfamilies, harness energy from transmembrane H⁺ and/or Na⁺ gradients, while eMATE transporters appear to be exclusively H⁺-coupled. Since the first crystal structure of a MATE transporter, that from Vibrio cholerae (NorM-VC), was determined (28), the structures of several MATE homologs including representatives from all three subfamilies have become available (9, 23, 29–33). All structures show the MATE transporters to be present in a monomeric state consisting of 12 transmembrane helices (TMs) with two 6-helix bundles (N and C bundles) arranged pseudosymmetrically. These bundles are linked by a large intracellular loop between TM6 and TM7. The drug- and cation-bound structures of a NorM transporter from Neisseria gonorrhoeae (NorM-NG) suggested that several conserved acidic residues in both N and C bundles are responsible for the allosteric regulation of uptake and efflux of substrates (9, 34). In contrast, atomic structures of the DinF transporter from Bacillus halodurans demonstrated the presence of only one competitive binding site in the N bundle and highlighted the importance of one highly conserved aspartate residue (D40^DinF-BH) for transport activity (29). In addition, a pH-dependent protonation/deprotonation of this aspartate residue was proposed to drive a bent–straight conformational change of TM1 in DinF transporters (30, 31), although this proposal still remains a matter of controversy (34, 35). Furthermore, a corresponding aspartate residue is absent in the N bundle of eMATE transporters. The crystal structure of AtDTX14, an eMATE transporter from Arabidopsis thaliana, revealed a direct competition between H⁺ and substrate that occurs in the C bundle (23). The crystal structure of the DinF subfamily transporter from Pyrococcus furiosus (PfMATE) in an inward-facing state was solved with bound native lipids (36), indicating a function as a lipid transporter. The transition between the two states is enabled by conformational flexibility and loose binding of TM1 and TM7 to their respective bundles.

Here we purified and characterized the MATE transporter Aq_128 from the hyperthermophilic bacterium Aquifex aeolicus. Our phylogenetic analyses indicate that Aq_128 belongs to a previously unknown subfamily of the MATE transporter family, which also comprises transporters from ε-Proteobacteria and Archaea so far. The phylogenetic placement of Aq_128 outside the known diversity of MATE transporters is consistent with the fact that Aq_128 lacks the conserved acidic residues which are crucial for the transport activity of other MATE transporters. X-ray structural and functional analyses characterize Aq_128 as an Na⁺-dependent drug transporter, and our data show that the cation and substrate binding site are located in the N bundle. A detailed structural and functional comparison between Aq_128 and the MATE transporters from other subfamilies is presented.

Results

Phylogenetic Analysis of MATE Family Transporters.

Earlier phylogenetic analysis of MATE proteins, including Aq_128, defined three major subfamilies (NorM, DinF, and eMATE), and Aq_128 was classified in the DinF subfamily (6, 27). However, amino acid sequence comparisons revealed that several otherwise strictly conserved and functionally essential residues (e.g., D41, N180, and D184 of PfMATE) are not present in Aq_128, raising the question of whether Aq_128 functions in the same manner as other DinF transporters. To place our functional studies on a solid phylogenetic basis, we expanded on earlier studies (6, 32) and reconstructed the evolutionary history of the MATE transporter family using 386 representatives (including 262 bacterial, 25 archaeal, and 99 eukaryotic proteins). The resulting tree revealed that the MATE transporters can be grouped into six major clades (Fig. 1 and SI Appendix, Fig. S1). The eMATE proteins are confined to two clades, referred to as eMATE-1 and eMATE-2. The eMATE-1 clade comprises animal, fungal, plant, oomycete, and alveolate proteins. It covers the full eukaryotic taxonomic diversity, indicating that its evolutionary age dates back to the last eukaryotic ancestor. In contrast, the eMATE-2 clade comprises MATE orthologs from plants only. It is interesting to note that eMATE-2 transporters are placed in a monophyletic group together with bacterial sequences, indicating that eMATE-2 was introduced either horizontally into the plant kingdom or was secondarily lost in other eukaryotes. The NorM and DinF subfamilies both form highly supported clades comprising both bacterial and archaeal sequences. Notably, our analysis revealed two additional clades that clearly fall outside the diversity of the hitherto-described MATE subfamilies, and we refer to them as aMATE-1 and aMATE-2. Next to Aq_128, members belonging to aMATE-1 are found, so far, in two representatives of the ε-Proteobacteria and in several of the Archaea. The aMATE-2 clade comprises mainly Proteobacteria and Firmicutes. For the remainder of the analysis, we concentrate on the aMATE-1 clade. The aspartate residues D41^PfMATE and D184^PfMATE, which are essential for cation and substrate binding in DinF transporters (29–31), are semiconserved among aMATE-1 proteins. The glutamate residue (E255^NorM-VC or E265^AtDTX14), which is conserved in the NorM and eMATE-1 subfamilies, is absent from the aMATE-1 subfamily (SI Appendix, Fig. S2).

Fig. 1.

Phylogeny of the MATE transporter family. The phylogenetic analysis indicates that 386 MATE family proteins cluster into six major groups: the eMATE-1, eMATE-2, NorM, DinF, aMATE-1, and aMATE-2 subfamilies. Aq_128, the MATE transporter from A. aeolicus, is indicated by a purple star. MATE proteins with available structures are marked with red stars. Bootstrap support is indicated by the size of the red circles. Individual clades have been collapsed to improve readability, and the number of subsumed sequences is given in parentheses after the clade label. The UniProt or NCBI reference sequence ID code is listed in Dataset S1. The full tree including numerical bootstrap values is shown in SI Appendix, Fig. S1.

Structure Determination and the Crystal Structure of Aq_128.

Aq_128 was initially crystallized using the vapor diffusion method. Tetragonal crystals were obtained at pH 8.5 and the structure was determined by molecular replacement at 3.8-Å resolution (SI Appendix, Table S1). Subsequent lipidic cubic phase (LCP) crystallization led to monoclinic crystals grown at pH 8.0 and to orthorhombic crystals grown at pH 5.0, which diffracted to 3.0- and 2.0-Å resolution, respectively (SI Appendix, Table S1). It should be noted that the latter is the highest-resolution structure of a MATE transporter to date. Regardless of the different methods or crystallization conditions, the overall fold of the structures is highly similar, as indicated by the low rmsds between the structures of 0.87 to 1.20 Å. Therefore, we mainly describe the highest-resolution structure of Aq_128 throughout the following paragraph.

Architecturally, Aq_128 consists of 12 TMs (TM1 to TM12) that are divided into N-terminal (TM1 to TM6) and C-terminal (TM7 to TM12) bundles (Fig. 2) connected by a long cytoplasmic loop (residues 222 to 231). The two bundles can be related by a pseudosymmetric-twofold rotation axis perpendicular to the membrane plane resulting in an rmsd of 3.35 Å over 206 Cα atoms (SI Appendix, Fig. S3), despite low sequence identity (15%). The overall V-shaped architecture of Aq_128 with the central cavity opened toward the periplasmic side represents an outward-facing state which is similar to that observed for other structurally characterized MATE proteins (9, 23, 28–33). The central cavity is connected to the N and C bundle cavities mainly at a position approximately halfway through the membrane-spanning region (SI Appendix, Fig. S4). An electrostatic surface analysis showed that the N and C bundle cavities exhibit different charge states in the outward-facing conformation (Fig. 2C and SI Appendix, Fig. S5). The N bundle cavity is dominantly negatively charged, attributed to two glutamate residues, E36 and E207. On the contrary, several potentially positively charged residues (R250, H275, K380, R400, and R428) are clustered in the C bundle cavity, resulting in a totally positively charged environment. The two arginine residues R250 and R400 are strictly conserved only within the aMATE-1 transporter subfamily. The difference of the charge distribution between the N and C bundle cavities is highly striking. By comparison, the whole internal cavity is either mainly negatively charged as in NorM- and eMATE-1–type transporters (9, 23, 28, 32) or positively charged as in PfMATE (31) (SI Appendix, Fig. S5). These differences may be attributed to different substrate specificities of each MATE transporter subfamily.

Fig. 2.

Crystal structure of Aq_128. (A) Cartoon representations of Aq_128 (ribbon model), with the periplasmic and cytoplasmic sides indicated. The structure is colored using a rainbow gradient from the N (blue) to the C terminus (red). (B) In-plane 90° rotation from A viewed from the periplasm. The molecule is colored rainbow as in A and the TMs are numbered. (C) Cross-section of Aq_128 in molecular surface representation and the electrostatic surface potentials calculated by APBS are colored ranging from blue (+10 kT/e) to red (−10 kT/e).

Crystallographic Dimer and Lipid Binding Site.

In contrast to the 3.8-Å structure, where Aq_128 is present as a monomer, the high-resolution structures (2.0 and 3.0 Å) derived from crystals grown by the LCP crystallization method contained two closely attached molecules in the asymmetric unit. In the 2.0-Å structure, four monoolein lipids are clearly visible at the dimer interface, while in the 3.0-Å structure, electron density is present at the same position but cannot be interpreted. The monooleins are arranged in a bilayer-like manner with the hydrophobic hydrocarbon chains pointing to the membrane center and the hydrophilic headgroups pointing toward the bulk solvent (Fig. 3A). The two Aq_128 molecules are closely linked by interactions mainly between monooleins and TM3, TM7, and TM11 of both molecules, and no strong direct protein interactions were observed (Fig. 3B and SI Appendix, Fig. S6). The two molecules are related by a twofold axis perpendicular to the dimer interface. Different from the other reported structures of MATE transporters, whose pseudosymmetry axis is perpendicular to the membrane plane (9, 29, 31), the pseudosymmetry axis of Aq_128 forms a 15° angle with the membrane normal (SI Appendix, Fig. S7A), perhaps due to the action of the nonphysiological monoolein as the hosting lipid. To further investigate the oligomeric state of Aq_128, chemical cross-linking experiments were performed in detergent solution and within the membrane to investigate the oligomeric state of Aq_128. The results show that part of the protein shifted from monomeric to higher–molecular mass species (SI Appendix, Fig. S7 B–D), which suggests that Aq_128 might exist in different oligomeric states. However, considering the fact that Aq_128 was heterologously produced in Escherichia coli, it is an open question whether such an oligomerization exists in A. aeolicus and whether it is of physiological relevance. Altogether, Aq_128 can be considered a crystallographic homodimer and an occurrence as a biological dimer would be compatible with the structural data.

Fig. 3.

Firmly bound monoolein lipids in the Aq_128 structure. (A) Cartoon and surface representation of the postulated Aq_128 dimer. Molecules A and B are colored cornflower blue and light sea green, respectively. The monoolein molecules are numbered and shown as tan sticks. (B) Close-up view of lipids (nos. 4 and 6) at the dimer interface. Several hydrophobic residues from TM3, TM7, and TM11 form a groove to accommodate the lipid tails. The hydroxyl groups of the glyceryl moiety are surrounded by three lysine residues (K234 from molecule A, and K384 and K442 from molecule B) at the cytoplasmic side, while the glycerol headgroup is hydrogen-bonded to N258 from molecule B and in van der Waals contacts by F115 and F118 from molecule A at the periplasmic side. (C) Close-up view of monoolein (no. 1) found in the central cavity. (D) Close-up view of monoolein (no. 2) found in the N bundle cavity. Lipids and the interacting residues are shown as sticks. Hydrogen bonds are depicted as black dotted lines.

In the 2.0-Å crystal structure of Aq_128, two monoolein molecules occupy the internal cavity of the outward-facing binding cleft. One monoolein lipid is located only in the central cavity of molecule A of the asymmetric unit. Its headgroup is projected to the entrance of the central cavity and forms hydrogen bonds with the adjacent residues N37 and S60 (Fig. 3C), whereas the lipid tail is buried deeply in the central cavity contacting W63 and F257. The second monoolein molecule is accommodated inside the N bundle cavity of molecules A and B with its carbonyl oxygen forming a hydrogen bond with N151. Several amphipathic residues, mostly tyrosine residues (Y33, Y66, and Y154), allow the hydrocarbon chain of this lipid to be cemented into the N bundle cavity through multiple van der Waals interactions of hydrophobic residues (Fig. 3D). The second bound monoolein molecule may mimic the biologically relevant substrates. Analogous to Aq_128, the monoolein molecule identified in the N bundle of PfMATE was also located at its drug binding site (31).

Functional Characterization of Aq_128.

To assess the drug susceptibility of Aq_128, in vivo complementation assays were carried out using the E. coli drug-hypersensitive strain C43(ΔacrAB) (33). Our results indicate that Aq_128 could confer cellular resistance to fluoroquinolone antibiotics including norfloxacin (SI Appendix, Fig. S8) and moxifloxacin (Fig. 4 and SI Appendix, Fig. S9). Furthermore, using C43(ΔacrAB) cells, the transport activity of Aq_128 could be inhibited by verapamil, as observed for DinF-BH and NorM-NG (35).

Fig. 4.

In vivo drug susceptibility assays of Aq_128. The moxifloxacin susceptibility test was performed using the E. coli C43(ΔacrAB) strain. (Left) In the experiment shown, transformants harboring the wild-type (wt) Aq_128 and its variants were inoculated on an LB plate supplemented with 0.009 μg/mL moxifloxacin and 0.5% l-arabinose. (Right) In the experiment shown, 100 μg/mL verapamil was used to inhibit the MATE-mediated drug efflux. Cells harboring the empty pBAD-A2 vector were used as negative controls (control). The experiment was performed at least three times with similar results, and representative results from one experiment are shown.

On the other hand, cation-dependent drug efflux assays were performed using E. coli C43(ΔacrAB) to study the cation specificity of Aq_128. Deenergized cells were preloaded with norfloxacin and drug efflux was initiated by the addition of glucose (to energize the cell) or Na⁺. Our results show that norfloxacin is significantly extruded from the C43(ΔacrAB)/pBAD-Aq_128 vector when an inwardly directed Na⁺ gradient is imposed, as indicated by the decrease of the norfloxacin fluorescence signal inside the cell (Fig. 5A). However, the difference between C43(ΔacrAB)/pBAD and C43(ΔacrAB)/pBAD-Aq_128 was less significant when 50 mM NaCl was added, because an ∼20% decrease of the relative fluorescence signal was observed in C43(ΔacrAB)/pBAD. This is probably due to the presence of an endogenous multidrug transporter from E. coli—YdhE, which was shown to be an Na⁺/norfloxacin antiporter (37). A similar observation was also reported for drug efflux experiments with NorM-NG (14). The increase of the Na⁺ concentration enhanced the driving force of the substrate efflux mediated by Aq_128, thereby leading to a greater efflux of norfloxacin and a more significant difference between C43(ΔacrAB)/pBAD and C43(ΔacrAB)/pBAD-Aq_128. By contrast, no obvious drug efflux could be observed when an inwardly directed proton gradient was imposed on the cells (Fig. 5B). This finding suggests that Aq_128 is an Na⁺-dependent, rather than a proton-dependent, MATE transporter.

Fig. 5.

Cation-dependent norfloxacin efflux assay. Norfloxacin efflux from E. coli C43(ΔacrAB) cells harboring pBAD-A2 or pBAD-Aq_128 was measured in the presence of 0 to 200 mM NaCl (A) or 0 to 1% glucose (B). Each experiment was performed at least three times and the mean values ± SD are indicated.

Identification of the Substrate Binding Site.

Our attempts to identify the substrate binding site of Aq_128 by cocrystallization or soaking with norfloxacin or moxifloxacin were unsuccessful, due to the poor quality of the resulting crystals. Therefore, in silico substrate docking experiments were performed to investigate the substrate binding site. The results show that substrates are accommodated inside the N bundle cavity of Aq_128 (Fig. 6 C and D). For both compounds (norfloxacin and moxifloxacin), the most populated cluster shows a very similar conformation. In this binding mode, the carboxyl group of two compounds has an ionic interaction with R250, while W63 and Y33 interact with the quinoline ring of the drugs by π–π conjugation. On the other side, the residue E207 interacts with the piperazine group of the substrates. This finding supports our assumption that the lipid identified in the N bundle cavity occupies the substrate binding site.

Fig. 6.

Cation and substrate binding sites identified by MD simulation and substrate docking experiments. (A) The distance between the sodium ion and Oδ1 atom of residue E36 during a 1-μs-long simulation, which is used as an indication of sodium ion binding or not. In the first binding event, the sodium ion stayed bound for ∼200 ns and then left the binding site. Afterward, a second sodium ion bound at ∼560 ns of the simulation, and stayed bound for the rest of the simulation. (B) Transient binding of a sodium ion in the binding events. The sodium ion is shown in purple. (C and D) Norfloxacin and moxifloxacin binding sites identified in substrate docking experiments are shown in C and D, respectively. Residues involved in ion and substrate coordination are shown as sticks.

To further corroborate the functional relevance of the above structural observations, we mutated several amino acids in the vicinity of the putative substrate binding site (Fig. 4). A Western blot analysis showed that all variants were expressed to a similar level as wild-type Aq_128 (SI Appendix, Fig. S10). In vivo drug susceptibility data demonstrated that substitution of W63 and Y154 by alanine completely abolishes the moxifloxacin resistance. Both residues are involved in binding the substrate to be transported. In the E249A or R250A variants, the resistance to moxifloxacin was considerably reduced, although not completely absent. In addition, the mutation of Y33, Y66, and E207, all positioned in the N bundle cavity and interacting with the docked substrate, to alanine did not abolish the ability of the resulting Aq_128 variants to confer drug resistance. Altogether, our results suggest that residues W63, Y154, E249, and R250 play significant roles in the process of MDR.

Molecular Dynamics Simulation of the Cation Binding Site.

Molecular dynamics (MD) simulations were performed to identify possible cation binding sites. During the 1-μs-long simulation, two spontaneous Na⁺-binding events were observed (Fig. 6A). In both binding events, a sodium ion bound to the N bundle near E36, first transiently for ∼200 ns and then again after ∼560 ns for the rest of the simulation. Compared with PfMATE, where a sodium ion stayed bound during the entire simulation process (36), the transient binding of a sodium ion to Aq_128 during the simulation indicated the instability of the current binding mode (SI Appendix, Fig. S11). The transient binding site of both binding events comprises residues E36, N174, H177, and E207 (Fig. 6B). While bound, the sodium ions were sandwiched between E36 and E207. However, it should be noted that this simulation does not consider the protonation state of titratable groups. The pK_a calculation in the absence of any ion showed that the pK_a values of E36 and E207 are 5 and 8, respectively. Therefore, a second 500-ns-long simulation was performed with a sodium ion put near a charged E36 and protonated E207. Results obtained from the second simulation showed an immediate escape of the sodium ion and no sodium ion binding throughout the simulation. Taken together, the simulation results suggest that the transient sodium binding to residues E36 and E207 mainly shows the charge imbalance in this area when E207 is charged. In order to further elucidate these observations, drug susceptibility assays were also conducted with the single substitutions E36A and E207A and the doubly mutated variant E36A/E207A (Fig. 4). Our results showed that all variants are equally well expressed and display wild-type levels of transport activities. These seemingly contradictory results led us to further explore whether there might be other potential candidates for an Na⁺ binding site. The positively charged nature of the C bundle cavity of Aq_128 almost excludes the possibility that the cation binding site lies within this region. In the N bundle, two negatively charged residues (E36 and E207) were shown to participate in transient binding of sodium ions but not to be critical for the transport activity. Together with the mutagenesis data, which indicate a functional importance of residue Y154, we propose that the sodium ion might be bound stably at Y154 through a cation–π interaction in the N bundle cavity, where the substrate binding site is also located (Fig. 6 C and D). Two glutamates, E36 and E207, may facilitate the entrance of Na⁺ into the binding site and, in the absence of E36 and E207, sodium ions may also enter the N bundle cavity from the central cavity. In this sense, the transient sodium-binding event identified in the MD simulation might represent an intermediate step during sodium ion translocation.

Discussion

Structural Comparison with NorM, DinF, and eMATE.

Although the structure and function of MATE transporters have been studied for many years, it is still a matter of debate how the substrate and the cation transfer are coupled during the transport cycle.

Prior to this work, Aq_128 from A. aeolicus was classified as a DinF-type MATE transporter. However, our sequence comparisons and the establishment of a comprehensive MATE transporter phylogeny led to the definition of two additional MATE subfamilies, which we term aMATE-1 and aMATE-2. Aq_128 belongs to the aMATE-1 subfamily, which otherwise comprises proteins from ε-Proteobacteria and Archaea. Notably, despite the distant evolutionary relationships of these taxa, they have been shown to heavily exchange genes via lateral gene transfer (38). This fact may explain the close evolutionary relationship of the aMATE-1 proteins to the exclusion of other bacterial MATE transporters. Besides, the presented structural data reveal substantial differences between Aq_128 and other MATE transporters (Fig. 7). In particular, one aspartate residue that is functionally essential and strictly conserved in the prokaryotic DinF and NorM transporters (e.g., D41^PfMATE and D36^NorM-VC) is replaced by a serine (S40) in Aq_128 (Fig. 7A). For the DinF subfamily, it was proposed that the protonation of this aspartate provokes a straight-to-bent transition of TM1 and a subsequent allosteric release of the bound substrate from the N bundle cavity (30, 31). This transport mechanism was postulated based on the finding that TM1 in PfMATE or VcmN crystal structures adopts a bent conformation at low pH and a straight conformation at high pH (30, 31). This proposal, however, remains contentious. In a recent study (36), PfMATE was crystallized at low pH using both the vapor diffusion and LCP methods, and the resolved outward-facing structures showed that TM1 remains straight. In the crystal structures of Aq_128 presented here, the conformations of TM1 at different pH values (pH 5 to 8) are rather similar (SI Appendix, Fig. S12). In line with this finding, Aq_128 lacks several conserved acidic residues (N180^PfMATE and D184^PfMATE) involved in the crucial hydrogen-bonding network in the N bundle of DinF transporters. Aq_128 also contains a pronounced hydrogen-bonding network in the N bundle which, however, is formed by different amino acids (E36, Y136, and H177) and several bridging water molecules (Fig. 7A). Therefore, the proposed transport mechanism based on a protonation-dependent structural transition of TM1 is not compatible with our results obtained for Aq_128.

Fig. 7.

Comparison of the structures of different MATE transporters. Close-up views of the N bundle (A) and of the C bundle (B) of Aq_128 (blue; PDB ID code 6Z70), NorM-VC (red; PDB ID code 3MKU), AtDTX14 (purple; PDB ID code 5Y50), and PfMATE (green; PDB ID code 3VVN). The key residues are shown as sticks. Hydrogen bonds are shown as dashed lines. Water molecules are shown as red spheres.

MATE transporters can use either Na⁺ or H⁺ as counterions to drive drug efflux. The cation-dependent drug efflux assay of Aq_128 strongly indicated that Aq_128 is an Na⁺-dependent transporter, which raises interest for a structural comparison between Aq_128 and Na⁺-dependent MATE transporters like those from V. cholerae (NorM-VC) and N. gonorrhoeae (NorM-NG). The proposed Na⁺ binding site in the NorM subfamily is located in the C bundle, as deduced from their Rb⁺- or Cs⁺-bound structures. In the NorM transporters, several conserved aromatic or acidic residues are involved in the coordination of the alkaline metal ions (9, 28). Among them, E255^NorM-VC and D371^NorM-VC, which are invariant in the NorM and eMATE-1 subfamilies, are absent in Aq_128 (Fig. 7B). In contrast, the C bundle cavity of Aq_128 contains two arginine residues, R250^Aq_128 and R400^Aq_128, strictly conserved in the aMATE-1 subfamily, which provide a positively charged character and nearly exclude the presence of an Na⁺ binding site.

A comparison of the crystal structures of Aq_128 and eMATE-1 transporter AtDTX14 reveals similarities in their N bundle cavities. S40 and H177 of the N bundle of Aq_128 are absent in DinF and NorM transporters but present and highly conserved in the eMATE-1 subfamily (Fig. 7A). Although a hydrogen-bonding network was not found in AtDTX14, probably due to the lack of an acidic residue (E36 in Aq_128), the notable structural similarity might reflect a common evolutionary origin of aMATE-1 and eMATE-1. On the other hand, the C-terminal bundles of Aq_128 and AtDTX14 differ. In eMATE-1 transporters, the formation of a hydrogen-bonding network centered around E265^AtDTX14 or D383^AtDTX14 implicates a protonation of these acidic residues which induces a bending of TM7 and finally enables the transition from the outward- to the inward-facing state (23). This mechanism highlights the importance of three residues (E265^AtDTX14, W266^AtDTX14, and D383^AtDTX14) in the substrate–proton recognition. In Aq_128, a similar hydrogen-bonding network was also observed in the C bundle (SI Appendix, Fig. S13). However, all three essential residues participating in the formation of the hydrogen-bonding network in the C bundle of eukaryotic MATE transporters are missing.

Functionality of Aq_128.

Previous studies on MATE transporters revealed that members of different subfamilies do not share a common substrate binding site. The substrate-bound structures of NorM-NG showed that substrate is captured on the extracellular loop and the substrate binding site is located in the central cavity (9), although this model is still debated (39). For NorM-VC and the eMATE-1 transporter AtDTX14, the putative substrate binding site is formed mainly by negatively charged residues located in the C bundle (39). In contrast, the DinF subfamily transporters PfMATE and DinF-BH use the N bundle cavity to bind the substrate (29, 31). For Aq_128, the observation of lipid molecules bound in the high-resolution structure and the substrate docking experiments suggests that the substrate binding site is located in the N bundle cavity, which is spatially similar to that of the DinF subfamily transporters. However, significant differences are present regarding the composition and configuration of their substrate binding sites. Based on the results of our mutagenesis studies, several amino acid residues, which are not present in the DinF subfamily, are shown to be important for substrate binding. In particular, W63 may play a key role in the recognition and binding of quinolone antibiotics. This tryptophan residue is located at the junction between the substrate exit channel (central cavity) and the N bundle cavity, and therefore may act as a gating residue. Mutation of either E249 or R250 moderately impairs the cellular resistance to moxifloxacin and norfloxacin. These two residues are strictly conserved only in the aMATE-1 subfamily. Taken together, the results indicate that they might contribute to the transport activity of Aq_128.

In this work, we present evidence that Aq_128 functions as a sodium-dependent multidrug efflux pump. Previous MD simulation studies on proton-coupled PfMATE showed that the crystallographically observed electron density peak near D41^PfMATE and N180^PfMATE should be assigned to a sodium ion (40). This assignment was later supported by the Cs⁺-bound structure of PfMATE (36). These findings suggest that PfMATE might possess a dual specificity for sodium ions and protons. As already mentioned, Aq_128 does not possess an aspartate at the equivalent positions. One glutamate (E207) is located in the opposite direction of the above-mentioned equivalent position inside the N bundle, and another glutamate residue (E36), which is involved in the hydrogen-bonding network of the N bundle, was found to be located one helix turn toward its N bundle cavity. The results of MD simulations and functional assays show that both glutamates cannot stably coordinate the sodium ion. However, the negative charge of residues E36 and E207 might promote the capture of the cation, and the sodium ion might pass through the transient binding site before it is stably bound to the protein. This scenario would explain why a transient sodium-binding event was observed during simulation, although this step might not be required. On the other hand, because the C bundle of Aq_128 bears a positively charged character, the true sodium binding site must be located in the N bundle. The functional importance of residue Y154 highlights the possibility that the sodium ion might be coordinated by Y154 through a cation–π interaction. Similar cation–π interactions were observed for the coordination of sodium ions in the sodium-dependent sugar cotransporter hSGLT1 (41) and a betaine/Na⁺ symporter, BetP (42). Therefore, we propose that both sodium ions and drugs might compete for the same binding site. This perspective should be further examined by future structural and functional studies.

In conclusion, the presented study demonstrates that Aq_128, the MATE transporter from A. aeolicus, shows unique structural features in the functionally crucial N- and C-terminal bundle cavities. Our data suggest that both cation and substrate binding sites are located in the N bundle. In this aspect, Aq_128 may employ a similar transport mechanism as DinF transporters, although the architecture of the active site is different. Based on a comprehensive phylogenetic analysis, a previously unknown MATE subfamily with Aq_128 as its prototype, aMATE-1, was defined. Future structural analyses of the cation-bound and inward-facing structure are expected to shed further light on the mode of coupling between drug extrusion and the influx of cations. It will also be interesting to extend the functional studies to include representatives of the aMATE-2 clade to see if, and to what extent, this advances our understanding of the functional evolution of MATE transporters.

Methods

Evolutionary Analysis.

We compiled a dataset of orthologs representing the currently known diversity of pro- and eukaryotic MATE proteins based on information provided by the Orthologous Matrix (OMA) orthology database (43). To this end, we identified OMA orthologous groups (OGs) harboring any of the following proteins: A. thaliana DTX1 (SwissProt ID code DTX1_ARATG; OG ID code 596033), A. thaliana DTX14 (SwissProt ID code DTX14_ARATH; OG ID code 663514), A. thaliana DTX19 (SwissProt ID code DTX19_ARATH; OG ID code 572862), A. thaliana DTX35 (SwissProt ID code DTX35_ARATH; OG ID code 586642), A. thaliana DTX41 (SwissProt ID code DTX41_ARATH; OG ID code 575730), A. thaliana DTX42 (SwissProt ID code DTX42_ARATH; OG ID code 571734), A. thaliana DTX50 (SwissProt ID code DTX50_ARATH; OG ID code 588636), Oryza sativa PEZ1 (UniProt ID code Q10HY1; OG ID code 580299), V. cholerae putative MDR protein (UniProt ID code C3LML9; OG ID code 572181), V. cholerae DinF (UniProt ID code C3LSA0), V. cholerae Na⁺-driven multidrug efflux pump (National Center for Biotechnology Information [NCBI] ID code ACQ59865.1; OG ID code 572720), Helicobacter hepaticus conserved hypothetical membrane protein (UniProt ID code Q7VK61; OG ID code 572161), and B. halodurans BH2163 protein (UniProt ID code Q9KAX3; OG ID code 575168). Subsequently, we concatenated the OGs and concentrated the downstream phylogenetic analysis on a subset of 128 species such that each species is represented in at least three OGs, and the species collection spans the full diversity of organismal life. We extracted the corresponding 372 sequences and complemented them with 14 further MATE proteins representing 11 additional species from the database TransportDB (http://www.membranetransport.org) to have all MATE proteins with an experimentally determined three-dimensional structure represented and to increase the taxon sampling of Archaea. The final dataset comprised 139 species and 386 sequences (Dataset S1).

For the phylogenetic tree reconstruction, we aligned the sequences with MUSCLE v3.8.31 (44). We then used a custom PERL script to remove columns containing more than 50% gaps from the multiple sequence alignment. The resulting postprocessed alignment served then as input for a maximum-likelihood tree reconstruction with RAxML v8.2.11 using the LG (45) model of sequence evolution substitution, allowing for invariant sites and modeling substitution rate heterogeneity across sites with a gamma distribution (PROTGAMMAILG). To assess the statistical support of the reconstructed tree, we performed 100 nonparametric bootstrap replicates using the rapid bootstrapping algorithm implemented into RAxML (46). The resulting tree was visualized and annotated using iTOL (47).

Construction of the Expression Vector.

The Aq_128 gene (NCBI gene ID code 1192671) was amplified from genomic DNA of A. aeolicus strain VF5 by PCR using the following two primers: 5′-gcagatcttgatgcaaaggattattgtgaatcccaac-3′ (BglII site underlined) and 5′-cggaattccgcgttactttactccttacttcgtcaagag-3′ (EcoRI site underlined). The PCR product was digested with BglII and EcoRI and cloned into the equivalent sites of pBAD-A2 (48), resulting in the expression construct pBAD-Aq_128 which encodes Aq_128 with a tobacco etch virus protease–cleavable His₁₀ tag fused to the C terminus.

Protein Expression and Purification.

E. coli Top10 cells transformed with pBAD-Aq_128 vector were grown at 37 °C in lysogeny broth (LB) medium supplemented with 100 μg/mL carbenicillin to an OD₆₀₀ of 0.5. Production of Aq_128 was induced by addition of 0.05% (weight/volume; wt/vol) l-arabinose and incubation was continued for 3 h. Cells were harvested by centrifugation at 10,000 × g for 15 min, frozen in liquid nitrogen, and stored at −80 °C until use.

All steps of membrane preparation and protein purification were performed at 4 °C. Cells were resuspended in resuspension buffer (20 mM Hepes·NaOH, pH 7.5, 100 mM NaCl, 2 mM phenylmethanesulfonyl fluoride, 20 μg/mL DNase I) at a ratio of 7 mL buffer per 1 g cells and were disrupted three times by passing through a precooled microfluidizer (Microfluidics) at 8,000 pounds per square inch. After centrifugation at 27,000 × g for 60 min, membranes were collected by centrifugation at 144,000 × g for 2.5 h. Crude membranes were resuspended in solubilization buffer (20 mM Hepes·NaOH, pH 7.5, 100 mM NaCl) and membrane proteins were solubilized by moderate stirring with 1% (wt/vol) n-dodecyl-β-d-maltoside (DDM) for 1.5 h. The insoluble membrane fraction was removed by centrifugation at 222,000 × g for 1 h, and the supernatant containing solubilized membrane proteins was loaded onto an Ni²⁺-NTA column (IBA) equilibrated with 20 mM Hepes·NaOH (pH 7.5), 300 mM NaCl, 1% (wt/vol) DDM, and 30 mM imidazole. After extensive washing, the bound proteins were eluted using the equilibration buffer supplemented with 300 mM imidazole. The eluted fractions were pooled and concentrated using Amicon Ultra-15 concentrators (50K molecular weight cutoff; Merck Millipore). The concentrated protein was purified further by size-exclusion chromatography using a Superdex 200 10/300 column (GE Healthcare) equilibrated in 20 mM Hepes·NaOH (pH 7.5), 150 mM NaCl, 0.025% (wt/vol) DDM or 0.075% (vol/vol) pentaethylene glycol monodecyl ether. Peak fractions containing Aq_128 proteins were collected, concentrated, and stored at −80 °C. Protein concentration was determined by the bicinchoninic acid assay (Thermo Scientific Pierce).

Vapor Diffusion Crystallization.

Aq_128 protein was crystallized with the sitting-drop vapor diffusion method at 18 °C using the CrystalMation System (Rigaku) at the Max Planck Institute of Biophysics (49). Initial crystallization screening was performed by mixing 100 nL protein sample (∼3.5 mg/mL) with an equal volume of commercially available screening solutions (MbClass Suite; Qiagen) in 96-well plates. After optimization, high-quality crystals were obtained in 0.1 M Tris·HCl (pH 8.5), 1.2 M (NH₄)₂SO₄. Protein crystals usually appeared within 60 d and grew to a maximum size in 3 mo.

Lipidic Cubic Phase Crystallization.

Aq_128 protein was crystallized using the LCP crystallization method according to previously described procedures with slight modifications (50) using the 96-well MPI tray plate (51). The concentrated protein sample (8 to 15 mg/mL) was mixed with monoolein using two coupled syringes. Fifty- to 80-nL protein-laden LCP boluses were dispensed on 96-well MPI tray plates (51) and covered with 1 to 1.5 μL of reservoir solution. Crystallization plates were set up with a ProCrysMeso drop setter (Zinsser Analytic), sealed, and stored at 22 °C. Crystals, monoclinic (space group: P 1 2₁ 1) and orthorhombic (space group: P 2₁ 2₁ 2₁), were obtained in two different conditions: monoclinic in 0.1 M Tris·HCl (pH 8.0), 30% (vol/vol) polyethylene glycol monomethyl ether 500, 0.2 M (NH₄)₂SO₄, and orthorhombic in 0.1 M sodium citrate-citric acid (pH 5.0), 30% (vol/vol) polyethylene glycol dimethyl ether 500. For cocrystallization experiments, 50 mM norfloxacin or moxifloxacin was added to the protein solution prior to crystallizing. Crystals grew to full size in 4 wk.

Data Collection, Structure Determination, and Refinement.

X-ray diffraction data were collected at the X10SA (PXII) synchrotron beamline at the Swiss Light Source, Paul Scherrer Institute. Data processing and scaling were performed using XDS (52) and SCALE (52). They were further analyzed using PHENIX software (53) and the CCP4 package (54). All structures were determined by molecular replacement by PHASER (53) using the coordinates of PfVc1-MATE (Protein Data Bank [PDB] ID code 4MLB). The molecular replacement solution was refined using the program Refmac (55), and the structure model was manually rebuilt using Coot (56). The structure was refined with PHENIX.refine (53). The final refinement statistics are summarized in SI Appendix, Table S1.

Substrate Docking.

Protein was prepared for docking using the protein preparation wizard in the Schrödinger Maestro Program Suite using the X-ray structure (PDB ID code 6Z70). Subsequently, the protein model was minimized with the OPLS_2005 force field. The ligands, norfloxacin and moxifloxacin, were minimized using Maestro LigPrep with an OPLS_2005 force field while retaining the stereochemistry. The ligands were then docked to the minimized MATE applying ligand docking using the Glide program, which supports full flexibility of the ligand (57). The standard precision scoring function with a softened van der Waals potential (scaling of 0.8), was used during the docking process. The whole binding cavity was divided into 24 overlapping docking boxes and docking was performed for each box. A maximum of 60 poses were retained in this stage for each box. Around 1,000 binding poses were produced for each ligand. The docking binding poses were clustered using the Jarvis−Patrick method with a cutoff of 3 Å implemented in the g_cluster command of GROMACS v2019.6 (58).

Molecular Dynamics Simulation.

The Aq_128 structure was embedded in a lipid bilayer of 1-palmitoyl-2-oleyl-sn-glycero-3-phosphoethanolamine/1-palmitoyl-2-oleyl-sn-glycero-3-phosphoglycerol (50/50%) composition using CHARMM-GUI (59). The pKas were calculated using PROPKA3 (60) and the protonation states of the titratable Asp, Glu, and His residues were assigned. In the first simulation run, E207 was set charged while in the second run it was assigned as protonated and the other Asp and Glu residues were assigned as charged. All His residues were neutral, and were protonated at either their Nδ or Nε atom according to the network of hydrogen bonds.

All systems were hydrated with 150 mM NaCl electrolyte. The all-atom CHARMM36 force field was used for protein, lipids, and ions, with TIP3P water (60–64). The MD trajectories were analyzed with visual molecular dynamics software (65). All simulations were performed using GROMACS 2019.6 (66). The starting systems were energy-minimized for the 5,000 steepest descent steps and equilibrated initially for 500 ps of MD in a canonical ensemble and later for 8 ns in an isothermal-isobaric ensemble under periodic boundary conditions. During equilibration, the restraints on the positions of nonhydrogen protein atoms of initially 4,000 kJ·mol⁻¹·nm² were gradually released. Particle mesh Ewald summation (67) with cubic interpolation and a 0.12-nm grid spacing was used to treat long-range electrostatic interactions. The time step was initially 1 fs, and was then increased to 2 fs during the NPT equilibration. The LINCS algorithm (68) was used to fix all bond lengths. A constant temperature was set with a Berendsen thermostat (69), with a coupling constant of 1.0 ps. A semiisotropic Berendsen barostat was used to maintain a pressure of 1 bar. During the production run, the Berendsen thermostat and barostat were replaced by a Nosé–Hoover thermostat (70) and a Parrinello–Rahman barostat (71). Analysis was carried out on unconstrained simulations. The simulation with E207 charged was run for 1,000 ns, while the second run with E207 protonated was run for 500 ns.

Drug Resistance Assay.

The drug export activity of Aq_128 was determined based on its ability to confer drug resistance to E. coli cells. Briefly, the drug-sensitive strain E. coli C43 (DE3) ΔacrAB harboring the pBAD-Aq_128 vector (or mutant vectors) was grown to an OD₆₀₀ of 0.5, and 0.05% (wt/vol) l-arabinose was added to induce the expression of Aq_128 at 37 °C for 30 min. The OD₆₀₀ was adjusted to 0.5 using liquid LB media and further diluted 1:10, 1:50, 1:100, 1:500, and 1:1,000 with LB media. Three microliters of the diluted cell suspensions was placed onto the LB-agar plate supplemented with 50 μg/mL carbenicillin, 0.5% (wt/vol) l-arabinose, and 0.009 μg/mL moxifloxacin. The drug resistance assay was also performed in the presence of 100 μg/mL verapamil. The cell growth was evaluated by visual inspection after 24 h of incubation at 37 °C.

Efflux of Norfloxacin in the Presence of Cations.

The assay of norfloxacin efflux was performed as described previously (14, 72) with some modifications. Briefly, cells were grown in LB media at 37 °C to an OD₆₀₀ of 0.5, and 0.05% (wt/vol) l-arabinose was added to induce the expression of Aq_128 at 37 °C for 1 h. Cells were harvested and washed three times with ice-cold buffer containing 0.1 M Tris·HCl (pH 7.0) by centrifugation at 3,000 × g at 4 °C for 10 min. To load the cell with norfloxacin, cells were incubated in the same buffer containing 25 μg/mL norfloxacin and 50 μM carbonyl cyanide m-chlorophenylhydrazone at 37 °C for 30 min. The cells were then pelleted, washed twice, and resuspended with 0.1 M ice-cold Tris·HCl (pH 7.0) to an OD₆₀₀ of 2.0. After 10 min, NaCl or glucose (glucose was used to initiate the respiratory chain) was added accordingly. Samples of 1 mL were taken after 10 min and washed once with the same ice-cold buffer. Pellets were resuspended in 1 mL of 100 mM glycine·HCl (pH 3.0), shaken vigorously for 1 h at room temperature to release the fluorescent content, and then centrifuged at 18,000 × g for 10 min at room temperature. The fluorescence of the supernatants was measured at an excitation λ (λ_ex) and an emission λ (λ_em) of 277 and 456 nm, respectively. The maximum fluorescence was normalized to 100%.

Polyacrylamide Gel Electrophoresis and Western Blot.

Sodium dodecyl sulfate–polyacrylamide gel electrophoresis analysis was performed using 4 to 12% Bis-Tris NuPAGE gels (Invitrogen) followed by Coomassie blue staining. The His₁₀-tagged Aq_128 was immunodetected using a monoclonal anti-polyhistidine alkaline phosphatase–conjugated antibody (Sigma) following the manufacturer’s instructions. The nitroblue tetrazolium/5-bromo-4-chloro-3-indolylphosphate system was used to detect the alkaline phosphatase activity.

Data Availability

Atomic coordinates and structure factors reported in this article have been deposited in the Protein Data Bank under ID codes 6Z70 (native Aq_128, LCP, 2 Å), 6FV8 (native Aq_128, LCP, 3 Å), and 6FV6 (native Aq_128, vapor diffusion, 3.8 Å).