Broadly conserved Na⁺-binding site in the N-lobe of prokaryotic multidrug MATE transporters

Significance

MATE transporters deplete the bacterial cytosol of natural and human-made antibiotics and contribute to confer multidrug resistance to a wide range of human pathogens. Despite being a compelling pharmacological target, little is known about their molecular mechanism. Based upon crystallographic and computational data, we gain insights into the mechanism by which drug efflux is energized by transmembrane electrochemical gradients of ions by identifying a Na⁺-binding site in the N-terminal domain of the transporter. This site can be clearly discerned in two different high-resolution structures and is broadly conserved within the MATE superfamily. Our structural analysis also provides a plausible rationale for the observation of H⁺-coupled drug transport and other pH-dependent effects.

Abstract

Multidrug and toxic-compound extrusion (MATE) proteins comprise an important but largely uncharacterized family of secondary-active transporters. In both eukaryotes and prokaryotes, these transporters protect the cell by catalyzing the efflux of a broad range of cytotoxic compounds, including human-made antibiotics and anticancer drugs. MATEs are thus potential pharmacological targets against drug-resistant pathogenic bacteria and tumor cells. The activity of MATEs is powered by transmembrane electrochemical ion gradients, but their molecular mechanism and ion specificity are not understood, in part because high-quality structural information is limited. Here, we use computational methods to study PfMATE, from Pyrococcus furiosus, whose structure is the best resolved to date. Analysis of available crystallographic data and additional molecular dynamics simulations unequivocally reveal an occupied Na⁺-binding site in the N-lobe of this transporter, which had not been previously recognized. We find this site to be selective against K⁺ and broadly conserved among prokaryotic MATEs, including homologs known to be Na⁺-dependent such as NorM-VC, VmrA, and ClbM, for which the location of the Na⁺ site had been debated. We note, however, that the chemical makeup of the proposed Na⁺ site indicates it is weakly specific against H⁺, explaining why MATEs featuring this Na⁺-binding motif may be solely driven by H⁺ in laboratory conditions. We further posit that the concurrent coupling to H⁺ and Na⁺ gradients observed for some Na⁺-driven MATEs owes to a second H⁺-binding site, within the C-lobe. In summary, our study provides insights into the structural basis for the complex ion dependency of MATE transporters.

The emergence of bacterial multidrug resistance (MDR) is a growing threat to public health worldwide, as it renders existing antibiotics ineffective against important infectious diseases (1, 2). In part, MDR results from innate mechanisms that enable bacteria to expel a broad range of cytotoxic compounds out of the cell. These mechanisms deplete the bacterial cytosol of man-made and natural antibiotics, thus protecting their intracellular targets (3). So-called MDR transporters are the class of membrane proteins involved in this type of defense mechanism. These transporters are also associated with the efflux of compounds that help pathogenic cells to colonize their host, increasing their chance of survival (4). Novel pharmacological strategies against MDR would therefore benefit from a comprehensive, detailed understanding of the structure and mechanism of these multidrug transporters.

Among the known five families of MDR transporters, so-called multidrug and toxic-compound extrusion (MATE) proteins are the most recently recognized and least characterized (5–7). It is well established, however, that MATE transporters catalyze the efflux of a variety of toxic compounds, most of which are hydrophobic and weakly cationic and able to diffuse into the cell across its envelope. To counter this flow and deplete the bacterial cytosol maximally, MATEs energize the drug-efflux process by coupling it to the influx of monovalent cations, down a preexisting electrochemical gradient. That is, MATEs function as ion-driven secondary-active antiporters.

Structurally, MATE antiporters consist of 12 transmembrane (TM) helices, organized in two distinct domains of six TM helices each (TM1–TM6 and TM7–TM12), referred to as N- and C-lobe (5, 6, 8). Like other secondary-active transporters, it is probable that MATE antiporters operate according to the alternating-access model, that is, that they possess the ability to interconvert between two conformational states that sequentially expose the bound substrate(s) to one side of the membrane, but not the other (9). To date, however, only structures of outward-facing conformations have been reported. In six of the seven crystal structures available, namely those of NorM-VC (10), NorM-NG (11, 12), PfMATE (13), ClbM (14), CasMATE (15), and AtDTX14 (16), the protein is captured in a V-like conformation, with the N- and C-lobes in contact on the intracellular side and projecting away from each other on the opposite side of the membrane. In between these two domains, a large cavity is open to the extracellular space and, sideways, to the lipid bilayer. The structure of DinF-BH (17) shows an intriguing variation of this conformation, possibly reflecting an intermediate, in which the TM7-TM8 hairpin is dissociated from the C-lobe and bundled with the N-lobe instead. Despite these insights, however, it is completely unclear how the inward-facing conformation is achieved. Unlike many other secondary-active transporters, MATE proteins do not contain inverted topological repeats; although the N- and C-lobes are pseudosymmetrical, their TM topologies are parallel, as is also observed for MDR transporters in the RND family (18). Thus, it is not straightforward to infer the features of the inward-facing state from the existing outward-facing structures.

Like most secondary-active transport systems, MATEs are driven by either Na⁺ or H⁺. However, it has been challenging to rationalize the factors that determine the cation specificity of any given homolog, even at the level of the amino acid sequence. NorM of Vibrio parahaemolyticus was the first MDR transporter shown to be driven by Na⁺ (19), and other MATEs studied since then appear to share this property [e.g., NorM-NG (20) and ClbM (14)]. However, other prokaryotic MATEs have been reported to be coupled to H⁺ instead [e.g., PfMATE (13), DinF-BH (12), and NorM-PS (21)] or, intriguingly, to both Na⁺ and H⁺ acting additively [e.g., NorM-VC (22)]. Other monovalent cations such as K⁺, Rb⁺, and Li⁺ have also been reported to influence substrate efflux by some MATEs (23–26). A caveat, however, is that the specificity of these transporters is often probed through cell-based resistance assays, and therefore it is not always completely clear that the effects observed reflect directly or exclusively on MATE activity.

Available structures of prokaryotic MATEs have not conclusively clarified this question either, as in most cases the resolution of the crystallographic data is insufficient to ascertain the presence of an ion-binding site or its detailed geometry. An exception is the structure of PfMATE (13), which was resolved at about 2.5-Å resolution. As mentioned, PfMATE has been reported to be coupled to H⁺, which would be recognized by a binding site in the N-lobe (13). Confusingly, though, the side-chain makeup of this putative site is highly similar to that of known Na⁺-binding sites (27–29). Oftentimes these sites may also bind H⁺, and indeed may be H⁺-selective, but at physiological pH and Na⁺ concentrations Na⁺ is expected to be utilized as a coupling ion (30).

We thus reasoned that the putatively ion-bound structure of PfMATE would be a suitable starting point to begin a theoretical investigation of the structural factors controlling the cation specificity of MATE transporters, and we report the outcome of this investigation below. We begin by reexamining the published crystallographic data and discuss several plausible interpretations thereof. We then present the results of molecular dynamics (MD) simulations designed to evaluate each of these interpretations in terms of their consistency with the experimental protein structure. Finally, we assess whether our conclusions can be generalized to other prokaryotic MATEs.

Results and Discussion

Reinterpretation of Crystallographic Data for Outward-Facing PfMATE.

Based on their crystallographic data for outward-facing PfMATE [Protein Data Bank (PDB) ID code 3VVO], Tanaka et al. (13) concluded that a functional H⁺ binding site exists in the N-lobe of the transporter, approximately halfway between the membrane midplane and the extracellular surface of the protein (Fig. 1). The proposed binding site is formed by four side chains, namely, Asp41 on TM1, Asn180 and Asp184 on TM5, and Thr202 on TM6. The backbone carbonyl of Ala198, also in TM6, is the fifth component. A strong spherical electron-density signal not attributable to the protein was also detected at the center of this site. Tanaka et al. (13) interpreted this signal as a bound water molecule (Fig. 1A) and further suggested that it mediates a network of hydrogen-bonding interactions whereby both Asp41 and Asp184 are protonated.

Fig. 1.

Reinterpretation of crystallographic data for PfMATE. (A) Outward-open conformation of PfMATE (PDB ID code 3VVO), proposed to be H⁺-bound. The N- and C-lobes are shown in yellow and blue, respectively. Crystallographic water molecules are shown as red spheres. The green sphere in the N-lobe, initially interpreted as a water molecule, indicates the proposed H⁺ binding site. (B) Temperature factors for the crystallographic water molecules illustrated in A against their number of neighboring polar contacts in the protein. (C) 2F_o − F_c map of the proposed binding site in the N-lobe, calculated after replacing the putative crystallographic water molecule with an Na⁺ ion (green sphere). Coordinating residues are highlighted. The data are shown at 2.9σ. (D and E) F_o − F_c omit maps for the cation-binding site in the N-lobe, and for one of the other sites occupied by a water molecule, respectively (indicated with gray arrow in B). In both cases the data are shown at 7.0σ.

The central electron-density signal is, however, unlike that from any of the other crystallographic water molecules identified within the protein surface. As shown in Fig. 1B, the corresponding atomic B-factor is much smaller; it is also an outlier in that the number of neighboring polar contacts in the protein is five, rather than one or two (Fig. 1B).

A coordination number of five raises the possibility that the electron-density signal in the center of the site corresponds to a Na⁺ ion rather than a water molecule. Na⁺ electrochemical gradients drive the mechanism of a wide range of secondary-active transporters, including some MATEs, and Na⁺ was reportedly present in the crystallization conditions (13). The specific chemical makeup of this putative binding site is also highly similar to that of functional Na⁺ sites in other membrane proteins of known structure (28, 29, 31). Indeed, a simple valence calculation (32) shows that the coordination number and geometry of this site in PfMATE match exactly what is empirically observed for Na⁺ (valence of 1.0) but not K⁺ or Ca²⁺ (valences of 3.7 and 1.6, respectively).

Water and Na⁺ have the same number of electrons and are therefore not distinguishable through X-ray crystallography at typical resolutions (in this case, 2.5 Å). Accordingly, an alternative refinement of the data in which we replaced the putative water molecule with a Na⁺ ion (Fig. 1C) resulted in R-factors that are nearly identical to those obtained originally (R_work = 19.9% and R_free = 25.1% vs. R_work = 19.7% and R_free = 25.2%), and the same geometry (Fig. 1C). Comparison of omit electron-density maps for the putative Na⁺ ion (Fig. 1D) and for the most ordered crystallographic water elsewhere in the structure (Fig. 1E) demonstrate the former produces a much stronger signal, which would be consistent with a narrowly defined geometry determined by ion–protein interactions.

In summary, we hypothesize that the published crystallographic data for PfMATE reveal a Na⁺-binding site in the N-lobe; however, this analysis is not conclusive, and therefore the original interpretation also appears plausible based on the data. It is also unclear whether the presence of Na⁺ would imply that the binding site is partially or fully deprotonated. This is an important consideration because, as mentioned, published evidence indicates the activity of PfMATE can be driven by H⁺, at least in nonnative experimental conditions (13). To clarify these questions, we proceed to further examine these alternative interpretations of the structural data using atomically detailed MD simulations.

MD Simulations of Na⁺- and H⁺-Bound PfMATE.

We begin by evaluating the two configurations shown in Fig. 2 A and B. In both cases, we assume a H⁺ is bound to Asp184, mediating a canonical carboxyl–carboxylate interaction with Asp41 (H-bonding distance of 2.7 Å). In one case, a second H⁺ is bound to Asp41 and a water molecule occupies the central electron density, as initially proposed (13). In the second case, Asp41 is deprotonated and an Na⁺ ion occupies the center of the site instead.

Fig. 2.

Initial configuration and conformational dynamics of outward-facing PfMATE in the Na⁺- and H⁺-bound states. (A and B) Close-up of the cation-binding site in the N-lobe, in either the H⁺- or Na⁺-bound state, at the start of the simulations. Note Asp184 is protonated in both cases, while the protonation state of Asp41 differs. (C) Subdomain definition employed in the analysis of the conformational dynamics of outward-facing PfMATE. (D) Analysis of the conformational dynamics of outward-facing PFMATE in the Na⁺- and H⁺-bound states, in the submicrosecond time scale. This analysis pertains to the backbone of the protein; error bars primarily reflect the variability of the data among independent simulations. To evaluate the internal dynamics of each of the subdomains color-coded in C, we evaluate the time-averaged rmsd of that subdomain relative to the starting X-ray structure (PDB ID code 3VVO), after an optimal superposition of all simulation snapshots and reference structure using that specific domain only. We refer to these cases as “self-fit”; in the figure, the subdomain for which the rmsd is quantified has the same color as the “fitting subdomain,” indicated underneath. To evaluate the relative dynamics of two subdomains, one of them is used as the fitting subdomain and the other is used for the rmsd calculations, or vice versa. These cases are referred to as nonself-fit. For clarity, rmsd values of the extracellular subdomains relative to each other (red vs. orange, and vice versa) are omitted. These values are 3.7 ± 1.3 and 4.2 ± 1.5 Å for the Na⁺-bound state and 4.1 ± 1.1 and 4.6 ± 3.0 Å for the simulations with only H⁺ bound. ext, extracellular; int, intracellular.

That Asp184 is protonated in both cases is not only plausible based on the binding-site geometry; it is also indicated by a probabilistic analysis of protonation states throughout the protein based on continuum-electrostatic calculations (Methods). This analysis indicates the protonation probability of Asp184 is virtually 100% for both configurations of the binding site, at physiologically relevant pH values. Asp41 is also highly likely to be protonated when a water molecule occupies the site (>95%), consistent with the two-proton assignment of Tanaka et al. (13), but its protonation probability is negligible in the Na⁺-bound state. The proton in Asp184 may therefore be seen as a “structural” proton—a notion that will be discussed further below.

Four independent simulations were carried out for each of the two binding-site configurations described above. These simulations were deliberately prepared using different protocols, to minimize any potential bias imposed by the starting condition on the conclusions of this analysis (Methods). For conciseness, all simulation data obtained for each case are combined, and the variability among and within trajectories is reported as error bars, when appropriate.

First, we assessed the overall conformational dynamics of the transporter, using the X-ray structure as the reference conformation, and the time-averaged root-mean-square deviation (rmsd) from that structure as a metric. We specifically aimed to identify subdomains that are internally quasi-rigid and to quantify and compare their relative motions. The results from this analysis did not reveal clear differences between the Na⁺- and H⁺-bound states but produced potentially significant insights otherwise (Fig. 2 C and D). In both cases, we observe that within both the N- and C-lobes the extracellular half of the lobe undergoes significant motions relative to the intracellular half (see nonself-fits of red vs. blue, orange vs. purple). Each of these halves, however, appears to be fairly compact and shows little internal dynamics on the time scale of the simulation (self-fits of red, blue, orange, and purple). Accordingly, the relative dynamics of the two lobes is significantly restricted on the intracellular side, where the two intracellular halves are in close contact (nonself-fits of blue vs. purple). In other words, in this outward-open conformation, and for the specific ion-occupancy states considered and simulated time scale, most of the relative dynamics between lobes originates from flexing of their respective internal structures, approximately at the level of the membrane midplane, rather than from rigid-body motions with a hinge at the point of contact between lobes, as often assumed (10, 11). Consistent with this observation, most of the TM helices in both lobes feature a significant number of glycine and proline residues in this central region. In addition, a notable observation is the high degree of flexibility of TM1, which flanks the cation-binding site in the N-lobe (Fig. 2 C and D). This helix is significantly more flexible than its symmetry equivalent in the C-lobe (i.e., TM7), whose dynamics are comparable to that of other helices (e.g., TM12). Because TM1 flanks the cation-binding site in the N-lobe, this observation might be mechanistically significant; indeed, a structure of PfMATE bound to norfloxacin differs from the structure considered here only in the conformation of TM1 (13). Nevertheless, the Na⁺- and H⁺-bound simulation systems are comparable also in this regard.

A more focused analysis of the dynamics of this cation site, however, clearly supports the notion that the signal detected in the center of the N-lobe binding site is that of an Na⁺ ion. In Fig. 3, we evaluate the degree to which the binding-site geometry sampled during the simulations of the Na⁺-bound state is consistent with that observed experimentally, both in terms of protein–protein and ligand–protein interatomic distances. From this data, it is apparent that there is a high degree of consistency between the simulated and experimental geometries, with no indication of Na⁺ dissociation (see also Fig. 6B). By contrast, in the simulation with an H⁺ bound to Asp41 (Fig. 4), we observe that the hypothesized water molecule occupying the central density signal dissociates readily (and is not replaced by other water molecules), irrespective of the simulation protocol (Fig. 4B). Notably, this observation appears to be consistent with the data presented in a simulation study of this doubly protonated state (33) that followed Tanaka et al. (13); yet, no alternative explanation was provided for the discrepancy between simulation and electron-density map. It is worth noting, and possibly a source of confusion, that the preferred geometry of the binding site is not significantly perturbed after water dissociation; aside from the expected reconfiguration of the hydrogen-bonding network, it appears to be comparable to that of the Na⁺-bound state, although with greater structural flexibility (Fig. 4C).

Fig. 3.

Structure and dynamics of the cation-binding site in simulations of the Na⁺-bound state. (A) Protein–protein (1–6; Left) and ion–protein (7–9; Right) distances used to represent the geometry of the binding site. (B) Probability distributions for each of the distances in A, calculated from the simulated data. The corresponding values from the experimental X-ray structure are indicated with red vertical lines.

Fig. 6.

Structure and dynamics of hypothetical variants of the Na⁺-bound state with no H⁺ or two H⁺ bound. (A) Hypothetical configuration with Na⁺ bound and no H⁺ bound to either Asp41 or Asp184. (B) Occupancy of the binding site by any Na⁺ ion in the simulation system, in the case where Asp184 is deprotonated (as in A, black), or when Asp184 is protonated (as in Fig. 3A, red), or when both Asp41 and Asp184 are protonated (blue). The occupancy is evaluated by quantifying the distance between the center of the binding site and the closest Na⁺ ion in the system, as a function of the simulation time. In both cases, restraints are employed during the system equilibration phase to keep the Na⁺ bound to the site and are then released. When Asp184 is deprotonated, Na⁺ dissociates readily and irreversibly; likewise, when both Asp41 or Asp184 are protonated. (C) Probability distributions for each of the protein–protein distances indicated in A, calculated from the simulated data. The corresponding values from the experimental X-ray structure are indicated with red vertical lines.

Fig. 4.

Structure and dynamics of the cation-binding site in simulations of the H⁺-bound state. (A) Protein–protein distances used to represent the geometry of the binding site. (B) Occupancy of the site by any water molecule in the simulation system, in two independent trajectories based on different equilibration protocols (gray). The occupancy is evaluated by quantifying the distance between the center of the site and the closest water molecule in the system, as a function of the simulation time. In one simulation, restraints are employed to keep the water molecule proposed to occupy the site bound therein, and are then released (black). In the other case, the restraints are released early in the equilibration (red). (C) Probability distributions for each of the distances in A, calculated from the simulated data, postequilibration. Note these distributions reflect a state with no water bound to the center of the site. The corresponding values from the experimental X-ray structure are indicated with red vertical lines.

The structural similarity between the two states is also apparent in Fig. 5, which compares the crystal structure of PfMATE with electron-density maps calculated from the MD simulations in both cases. These data make clear that the most plausible interpretation of the crystallographic data is that it captures a Na⁺-bound state, stabilized by an H⁺ shared between Asp184 and Asp41. The central density signal is therefore an Na⁺ ion, and not a water molecule. However, this analysis also suggests that a state with an additional H⁺ bound to Asp41, and the center of the site unoccupied, might also contribute to the measured electron density; that is, both structures might coexist in the crystal lattice, as has been noted for other ion-coupled membrane antiporters (34). Indeed, the chemical makeup and geometry of this Na⁺ site is highly similar to that observed in Na⁺-dependent rotary ATP synthases known to be promiscuous in regard to their specificity for Na⁺ or H⁺, including that of Pyrococcus furiosus (27). It stands to reason, therefore, that the observation that the activity of PfMATE can be driven by H⁺ owes to this promiscuity.

Fig. 5.

Comparison of the experimental geometry of the cation-binding site with the simulated ensemble. (A) The crystal structure (yellow/red/blue sticks, green sphere) is compared with a density map derived from the simulations of the Na⁺-bound state. The density map for the protein is shown in gray, and the signals for the Na⁺ ion and the H⁺ bound to Asp184 are shown in green and magenta, respectively. (B) Same as A, for the simulation of the H⁺-bound state (i.e., with a H⁺ bound to Asp41 in addition to that in Asp184). In both cases, the calculated density maps derive from the ensemble of simulation snapshots reflected in the peaks of the probability distributions shown in Figs. 3 and 4.

Finally, we evaluate in more detail the significance of the H⁺-mediated interaction between Asp184 and Asp41, which we propose is concurrent with Na⁺ binding. The simulation data described above suggest that this proton is an integral element of the interaction network defining the Na⁺-bound state, but we have not demonstrated that it is indeed necessary to explain the experimental geometry. We therefore considered a third case in which the N-lobe site is occupied by Na⁺ and both Asp41 and Asp184 are deprotonated. As shown in Fig. 6, this simulation resulted in the irreversible dissociation of Na⁺ and a pronounced distortion of the geometry of the cation-binding site. That is, the simulated geometry is in no way akin to that measured experimentally. Finally, we also considered a case where H⁺ are bound to both Asp41 and Asp184 in addition to Na⁺; in this case also, Na⁺ dissociated readily and irreversibly. These simulation data are therefore consistent with our initial evaluation of protonation states and reaffirm the conclusion that the experimental structure of outward-facing, substrate-free PfMATE represents a Na⁺-bound state where Asp184 and Asp41 form a carboxyl–carboxylate pair through a shared proton.

The Proposed Cation Site Is Strongly Selective for Na⁺ over K⁺.

The simulation data discussed up to this point very clearly support the hypothesis that the N-lobe of PfMATE features a Na⁺-binding site. Whether or not the Na⁺ occupancy of this site in the available substrate-free structure is mechanistically significant cannot be fully ascertained computationally. A conclusive answer will ultimately require careful experimental assays of transport and binding under well-defined conditions (of the kind that, to our knowledge, have not been reported for PfMATE). Nevertheless, computations can be used to evaluate whether the proposed site has some of the features expected from any functional binding site in a membrane transporter. Specifically, we can evaluate whether the site is selective for Na⁺ against other physiologically relevant cations, and specifically K⁺, and we can also evaluate the degree of conservation of the site within the MATE superfamily.

The results of the selectivity analysis are summarized in Fig. 7. Here, we quantify the free-energy cost associated with replacing Na⁺ by K⁺ in the protein binding site, relative to the bulk solution. That is, we quantify the difference in the binding free energy of these two cations (K⁺ is stably bound to the site in the time scale of all our simulations). We consider two cases: one in which the geometry of the binding site is restricted to that captured in the crystal structure, but to varying degrees, and another in which the geometry can adapt freely to the presence of the larger K⁺ ion. Consistent with the notion that the structure captures a Na⁺-bound state, we find that the selectivity of the site against K⁺ is maximal when the experimental geometry is imposed strongly, and that this selectivity decreases as the protein is allowed greater flexibility. Crucially, when the protein dynamics is entirely unrestricted there remains a significant free-energy penalty for K⁺ displacement of Na⁺, namely over 5 kcal/mol. This value suggests a preference for Na⁺ of over 1,000-fold, a value that is consistent with that deduced for known functional sites in Na⁺-dependent systems (35), if analyzed similarly. Insofar as selectivity indicates functionality, this analysis therefore suggests the proposed N-lobe site is a functional Na⁺ site.

Fig. 7.

Selectivity of the proposed Na⁺-binding site against K⁺. The y axis specifies the difference in the binding free energy of K⁺ relative to Na⁺, from four independent calculations. From top to bottom, the first three data points correspond to calculations in which the binding site is forced to preserve the geometry of the Na⁺-bound state (e.g., Fig. 5A) throughout the calculation, using a harmonic restraining potential of force constant k_F (Methods). A smaller value of k_F implies greater variability is permitted. The last data point is from a calculation in which the geometry of the site can adapt freely to either Na⁺ or K⁺. The x axis reports a measure of the change of the binding-site geometry when K⁺ replaces Na⁺; specifically, ΔD quantifies the mean increase in the distances indicated in Fig. 3.

The Proposed Na⁺ Site Is Broadly Conserved in Prokaryotic MATEs.

As mentioned, an expected feature of functional ion-binding sites in membrane transporters is a significant degree of conservation among different homologs. Therefore, we now ask whether the proposed Na⁺-binding site is indeed conserved among other members of the MATE superfamily. The Conserved Domain Database of the National Center for Biotechnology Information (NCBI) (36) lists 27 subclasses of MATE transporters, for a total of 1,294 sequences. In 16 of these 27 subfamilies, or about 60% of sequences, we find that the proposed Na⁺-binding motif is well conserved. Fig. 8 shows a statistical analysis of the amino acid make up of this subset, focused on the three TM helices that flank the proposed Na⁺-binding site (i.e., TM1, TM5, and TM6). It is apparent from this analysis that Asp41, Asn180, and Thr202 are highly conserved. At the position of Asp184, aspartate and asparagine are almost equally frequent, which also implies a high degree of conservation since protonation of Asp184 is concurrent with Na⁺ binding. As one might expect, there is some diversity in the position equivalent to Ala198, as this provides the backbone contact for the ion. However, the more frequent substitutions here are valine and serine, which also have small side chains. Moreover, the preceding residue, Gly197, is 100% conserved, suggesting that a glycine at this position provides the necessary backbone flexibility to facilitate the contact between the Na⁺ ion and the carbonyl of Ala198 or its equivalent.

Fig. 8.

Conservation of the proposed Na⁺-binding site among prokaryotic MATEs. (A) Primary-sequence logo for helices TM1, TM5, and TM6, calculated for 16 classes of the MATE superfamily that appear to conserve the proposed Na⁺-binding motif in the N-lobe (out of a total of 27 classes). This subset comprises ∼60% of the ∼1,300 sequences available in the NCBI Protein Conserved Domain Database. The five residues involved in the proposed motif are indicated with arrows. (B) Crystal structure of outward-facing ClbM (PDB ID code 4Z3N), which we propose also features a functional Na⁺-binding site in the N-lobe (green sphere). Like for PfMATE, the ion was originally interpreted as a water molecule. The figure shows the 2F_o − F_c map calculated after replacing this water with Na⁺, at 2.9σ (gray), and a F_o − F_c omit map for Na⁺ ion (green), at 7.0σ. (C) Like PfMATE (Fig. 9), ClbM appears to feature a H⁺-binding site in the C-lobe, in addition to the Na⁺-binding site in the N-lobe, created by Asp299 and His351.

Among the prokaryotic MATEs that feature this sequence motif in the N-lobe are transporters known to be coupled to Na⁺, such as NorM-VP and VmrA from V. parahaemolyticus (19, 37), NorM-VC from Vibrio cholerae (10), NorM-NG from Neisseria gonorrheae (11), and ClbM from Escherichia coli (14). In particular, the motif in VmrA is exactly identical to that observed in PfMATE, while NorM-VC and NorM-NG are examples of the asparagine substitution of Asp184, and ClbM exemplifies the valine substitution of Ala198. Although crystal structures of NorM-VC and NorM-NG have been resolved, their limited resolution seems to have obscured the functional Na⁺-binding site in these transporters. Nevertheless, that the residue equivalent to Asp41 is functionally crucial has been established by multiple studies of different MATEs (11–13, 22, 38). The case of ClbM is particularly notable, as an outward-facing structure of the protein is available at 2.7-Å resolution (14). ClbM is highly similar to PfMATE in the region of the proposed Na⁺-binding site. Indeed, the electron-density map for this structure also shows a clear signal at the center of the site. Mousa et al. (14) interpreted this signal as an ordered water molecule, presumably following the assignment made for PfMATE by Tanaka et al. (13). Our analysis, however, suggests that this signal is due to a bound Na⁺ ion (Fig. 8B) and suggests that this site underlies the observed functional Na⁺ dependence of ClbM.

In summary, multiple lines of evidence indicate that the proposed Na⁺ binding site in the N-lobe of PfMATE is very likely a characteristic functional motif in a large number of transporters of the MATE superfamily. Indeed, we believe that our sequence analysis probably underestimates the degree to which this motif is conserved, because the N- and C-lobes are exchanged in some MATEs. For example, the N-lobe motif does not seem to be present in MurJ or its closest homologs; however, the recently determined structure of this protein (39) shows the Na⁺ motif in the C-lobe (i.e., involving residues in TM7, TM11, and TM12).

A Hypothetical Functional H⁺-Binding Site in the C-Lobe.

Intriguingly, transport assays for NorM-VC have indicated that electrochemical gradients of Na⁺ and H⁺ can drive substrate efflux additively, with an apparent stoichiometry of 1Na⁺ and 1H⁺ per substrate (22). Our analysis supports the view that the functional Na⁺-binding site in NorM-VC is in the N-lobe (22), by extension of what we observe for PfMATE and ClbM. A site explaining the additional H⁺ coupling must therefore be elsewhere.

Inspection of available structures of MATEs suggest that this H⁺-binding site might be in the C-lobe. In PfMATE, the abovementioned probabilistic analysis of protonation states based on electrostatic calculations clearly points to Glu331, in TM9 (Methods). This side chain is relatively buried within the protein and within hydrogen-bonding distance of the carbonyl oxygen of Gly289, in TM8 (Fig. 9A). In our MD simulations, the hydrogen bond donated by protonated Glu331 to Gly289 is highly stable (Fig. 9B), and the geometry of this site is highly consistent with that observed in the input experimental structure (Fig. 9C), irrespective of whether the binding site in the N-lobe is assumed to be Na⁺- or a H⁺-bound. These results strongly indicate that Glu331 is protonated in the PfMATE structure.

Fig. 9.

Putative functional H⁺-binding site in the C-lobe of PfMATE. (A) In the outward-facing structure of PfMATE, E331 in TM9 is predicted to be protonated and forms a specific hydrogen bond with the backbone of TM8, at the point where this helix bends. The site is largely hydrophobic, consistent with the proposed protonation of this site. (B) Probability distribution for the interaction distance between protonated Glu331 and the backbone carbonyl of Gly289, from simulations in which N-lobe ion-binding site is bound to by Na⁺ or H⁺. The value in the crystal structure is indicated with a vertical gray line. (C) Comparison of the experimental structure with a calculated density map derived from the simulation of the Na⁺-bound state (as in Fig. 5A); the signal for the bound H⁺ is shown in magenta.

Whether this H⁺-binding site in the C-lobe is mechanistically important, or only a structural feature, will need to be clarified experimentally. For PfMATE, it is possible that the site that mediates H⁺ coupling is Asp184 itself, that is, that both Na⁺ and H⁺ bind to and dissociate from the same site in the N-lobe. However, it is worth noting that a protonatable site analogous to Glu331 in PfMATE also exists within the C-lobe of NorM-VC, as well as other MATEs known to be Na⁺-dependent. The precise location varies, but in all cases the site is approximately halfway across the TM span, partially buried within the protein though reachable from the cavity between the N- and C-lobes. Specifically, in ClbM this site is on Asp299/His351, in TM8 (Fig. 8C), and in NorM-VC and NorM-NG on Asp371 and Asp377, respectively, in TM10.

Confusingly, in the original crystallographic studies of these NorM homologs, the C-lobe was hypothesized to harbor the Na⁺ recognition site (10, 11). This notion stems from the detection of electron-density signals for Rb⁺ or Cs⁺ in the vicinity of the abovementioned acidic residues, from crystals soaked in buffers containing high concentrations of these ions [note neither ion would be able to displace Na⁺ from the site we propose, as its selectivity against Rb⁺ or Cs⁺ would be even higher than that against K⁺ (Fig. 7)]. The fact is, however, that neither Rb⁺ nor Cs⁺ actually mimics Na⁺ in any respect other than its charge. Indeed, a recent simulation study of outward-facing NorM-VC where the bound Rb⁺ was replaced by Na⁺ showed the ion to be stable only when an inward 400-mV electrostatic potential was applied across the protein (40). Evidently, that a membrane potential be present (chemical or electrostatic) is not a requirement for ion or substrate binding to secondary-active transporters (consider, for example, binding assays for purified, detergent-solubilized protein, or crystal structures). Contrary to the conclusions of this study, therefore, we would argue that these simulations reaffirm that the C-lobe of NorM-VC is unlikely to harbor a functional, selective Na⁺ site, consistent with the poor conservation and suboptimal amino acid makeup of this region. In our view, the experimental signals for Rb⁺ and Cs⁺ in the crystallographic conditions simply underscore that this region is distinctly electronegative. It would be therefore naturally attractive to positively charged ligands, or alternatively, protons, unless other cations are present at nonphysiological concentrations. Tentatively, therefore, we propose that the C-lobe is instead the primary site of H⁺ coupling in Na⁺-dependent MATEs like NorM-VC. As mentioned, however, the binding site in the N-lobe might also contribute to H⁺ coupling in this and other Na⁺-driven MATEs, as it appears to be naturally promiscuous in regard to its specificity for Na⁺ or H⁺.

Conclusions

An objective analysis of high-resolution crystallographic data for the drug-efflux transporter PfMATE reveals a Na⁺-binding site in the N-lobe. That Na⁺ occupies this site had not been previously recognized in structural studies of Na⁺-driven MATE transporters, which instead had hypothesized Na⁺ recognition occurs in a site in the C-lobe, based on the detection of nonspecific interactions with cations that do not mimic Na⁺. By contrast, the features of the site we propose are consistent with what might be expected for a functional binding site in a Na⁺-driven secondary-active transporter. The site is structurally well-defined, it is strongly selective for Na⁺ against K⁺, and it is broadly conserved across the MATEs superfamily, including isoforms of NorM known to be driven by Na⁺.

We further propose that an H⁺-binding site exists in the C-lobe of PfMATE and its homologs, and that this is the primary explanation for the observation that Na⁺-driven MATEs such as NorM-VC appear to be simultaneously coupled to a Na⁺ and H⁺ gradients. We also note, however, that the chemical makeup and structure of the N-lobe Na⁺-binding site indicates that it is weakly Na⁺-specific against H⁺; indeed, Na⁺ coupling in MATEs likely results from the fact that Na⁺ is in large excess over H⁺ in most settings. This phenomenon is not uncommon; inherently H⁺-selective sites that mediate Na⁺-driven activity in physiologically realistic conditions have been reported for other membrane enzymes and transporters, such as rotary ATP synthases (27, 41) and amino acid and sugar symporters (42, 43). This promiscuity between Na⁺ and H⁺, modulated by the specific protein environment, would explain the observation of H⁺-driven function or other pH-dependent effects in MATEs featuring the Na⁺-binding motif described here.

Not all prokaryotic MATE transporters are expected to be driven by or even dependent on Na⁺. Na⁺ is believed to have been the primary coupling ion in the evolution of transport systems (44), but as has been demonstrated for rotary ATP synthases, one or two hydrophobic mutations in a site that is specific for Na⁺ under physiological concentrations will cause a strong shift in the selectivity of the site for H⁺ (30, 45). Thus, strictly H⁺-dependent MATEs are entirely expected, particularly among eukaryotes. Nonetheless, our analysis does suggest that Na⁺ coupling in prokaryotic MATEs might be much more ubiquitous than previously recognized and pinpoints the associated structural and sequence motif.

That ion coupling in some MATEs involves binding sites in both the N- and C-lobes (e.g., for Na⁺ and H⁺, respectively) might contribute to the remarkable diversity of substrates transported by this family of proteins. Prototypical mechanisms of coupled ion–substrate antiport demand that binding of the transported substrate and the driving ions be mutually exclusive. We speculate that this competition might take place in different lobes for different substrates; PfMATE, for example, is known to recognize norfloxacin deep inside the N-lobe (13), causing conformational changes in TM1 (PDB ID code 3VVP) that would impair Na⁺ (or H⁺) binding to the site we have identified in this domain. However, biochemical evidence for, for example, NorM-PS, clearly indicates other substrates are directed to the C-lobe (21), where they would compete with H⁺. That distinct ion-recognition sites exist within each lobe might also suggest that the conformational mechanism of MATEs is unlike that of, for example, MFS transporters, whereby an N- and C-terminal domain rotate around a central site occupied by ions and/or substrates, for the most part preserving their internal structure (46). It seems entirely possible that the N- and C-lobes in MATEs undergo individual ion-dependent conformational cycles that alter their internal structure. Indeed, the abovementioned structure of DinF-BH (17) appears to capture a state in which the TM7–TM8 hairpin is dissociated from the C-lobe and is bundled with the N-lobe instead.

Irrespective of the precise nature of the molecular mechanism of MATEs, it is highly probable that it will fulfill the alternating-access model. If so, this mechanism must be tightly controlled by the occupancy state of the transporter, to enable the cell to deplete the cytosol from toxic compounds without dissipating the ionic electrochemical gradients that sustain this and other vital process. To make progress in our understanding of this intricate mechanism, it is important to conclusively ascertain the sites of ion and substrate recognition in these transporters. We posit that this computational study is a stepping stone in that direction.

Methods

Crystallographic Data Analysis.

Calculated 2F_o − F_c and F_o − F_c omits maps, atomic models, and R_work and R_free values are based on published structure factors as deposited in the PDB for the ID codes 3VVO (PfMATE) and 4Z3N (ClbM). The crystallographic data were processed with PHENIX version 1.8.4 (47, 48) and Coot (49). The valence of the site was calculated as ${\displaystyle {\sum }_i{(R_i/R_o)}^{-N}}$ (32), where i denotes the number of ligand–protein contacts within 3 Å, R_i is the contact distance in each case, and R_o and N are empirical parameters with the following values: 1.6 and 4.29 for Na⁺, 2.276 and 9.1 for K⁺, and 1.909 and 5.4 for Ca²⁺.

Protonation States of Ionizable Residues.

A Monte Carlo algorithm was used to identify the most likely set of protonation states reflected in the experimental structure of PfMATE (PDB ID code 3VVO), based on a continuum-electrostatics framework. Electrostatic-energy evaluations used the Poisson equation solver PBEQ in CHARMM 39b2 (50). The membrane was represented with a rectangular slab with dielectric constant of 2, while that of the protein interior was set to 4. The surrounding solution and solvent-accessible cavities within the protein were assigned a dielectric constant of 80. Atomic charges were those in the CHARMM36 force field (51) and atomic radii were those of Nina et al. (52). Only Asp, Glu, and His side chains were considered in this analysis (i.e., all Lys and Arg were assumed to be protonated). The Monte Carlo algorithm sampled a diverse set of all possible combinations of protonation states for these residues, favoring those with the best electrostatic energy at a given pH. From this “trajectory” the protonation probability P for a given side chain can be estimated from the frequency with which the protonated state is observed. Two cases were studied, either with or without an Na⁺ ion occupying the binding site in the N-lobe. (The bound Na⁺ ion was represented with a charge of +1e and a radius of 1.66 Å.) In both cases, most side chains favored their standard protonation state for the pH values considered, namely 6 and 7.5 (representing the periplasm and cytoplasm, respectively). For example, the calculated P for His416 is 46% at pH 6 and 3% at pH 7.5, consistent with its location in an exposed loop on the extracellular side. Likewise, for Glu51, which is also exposed, P is 2% at pH 6 but 0.08% at pH 7.5. Only a few residues showed atypical protonation probabilities at both pH 6 and 7.5. With Na⁺ bound these are Glu138 (16% and 0.7%, respectively), Asp184 (>99.9% in both cases), Glu331 (>99.9%), and His99 (<10⁻⁵%), and without Na⁺ also Asp41 (99.8% and 95.1%, respectively). In summary, this analysis predicts that H⁺ are bound to the outward-facing structure of PfMATE on Asp184 and Glu331 in the Na⁺-bound state, and additionally to Asp41, in the absence of Na⁺.

MD Simulations.

All MD simulations were carried out with NAMD version 2.7 (53), using the CHARMM36 force field (51, 54), with an NBFIX correction of sodium–carboxylate interactions (55). Temperature and pressure were kept constant at 298 K and 1 atm, respectively. A time step of 2 fs was used. Electrostatic interactions were calculated using PME with a real-space cutoff of 12 Å. The same cutoff distance was used for van der Waals interactions, with a switching function turned on at 10 Å. Periodic boundary conditions were used.

All simulations are based on PDB ID code 3VVO. A missing five-residue loop between TM9 and TM10 (residues 352–357) was added with MODELLER (56); 1,500 models were produced and the model with the lowest score was selected. Crystallographic water molecules within the protein solvent-excluded surface were retained, and additional water molecules with the protein were added with DOWSER (57). The resulting structure was briefly energy-minimized with CHARMM (500 steps) and embedded in a preequilibrated lipid bilayer (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) using GRIFFIN (58). The surrounding solution contained 100 mM NaCl and counterions to neutralize the total charge of the system. Altogether, the simulation system comprises ∼86,000 atoms in an orthorhombic box of ∼90 by 90 by 102 Å.

Four alternative configurations of the binding site in the N-lobe of PfMATE were considered (Results and Discussion). Several simulations were carried out in each case, each comprising an equilibration phase and a production phase. The equilibration phase consisted of a series of stages in which the dynamics of the protein–ligand complex is restricted, primarily through internal-conformation restraints, to a degree that is gradually reduced over ∼325 ns. To minimize the influence of the starting condition, each of the simulations carried out for the same binding-site configuration used a different equilibration protocol, varying in the exact definition of the restraints. In the production phase, the structural dynamics of the protein–ligand complex was unrestricted. Specifically, we produced four trajectories of 600 ns for the state with protonated Asp41 and Asp184, two trajectories of 600 ns and two trajectories of 1 μs for the Na⁺-bound state with protonated Asp184 with state, two trajectories of 1 μs for the Na⁺-bound state and Asp184 deprotonated, and one trajectory of ∼120 ns for a Na⁺-bound state with protonated Asp41 and Asp184.

Free Energy of Selectivity.

MD simulations were also used to evaluate the free-energy difference between the proposed Na⁺-bound state and a hypothetical K⁺-bound configuration. This free-energy value was evaluated in two different cases: In one case the geometry of the binding site was restricted to that observed in the crystal structure; in the other case, the binding site was permitted to freely adapt to K⁺. In the former case, the conformational restriction was implemented through a set of protein–protein distance restraints, each imposed with a flat-harmonic potential of force constant k_F. Three different calculations were carried out with k_F values of 120, 30, and 7.5 kcal⋅mol⁻¹⋅Å⁻². A fourth calculation was carried out with no conformational restraints, as mentioned. In each case, the Free-Energy Perturbation module of NAMD was used to induce the alchemical transformation of the Na⁺ ion bound to the protein into K⁺ and, concurrently, of a K⁺ ion in solution into Na⁺, recording the resulting free-energy change. (The transformed ion in solution was kept away from the protein and membrane using a boundary potential, so that |z| > 35 Å.) The process was then reversed and the free energy recomputed. Each transformation was made gradually, using a parameter λ that scales up the radius of the ion, in 50 consecutive simulations of 300 ps each (the initial 100 ps were discarded as equilibration time). Mean values from the forward and backward transformations are reported along with their difference. To describe the change in the geometry of the binding site upon transformation of Na⁺ into K⁺, we calculated the time average of the abovementioned protein–protein distances, $\langle d_i\rangle$, for both the Na⁺ and K⁺, and quantified $\mathrm{\Delta }D=\left(1/N\right){\displaystyle {\sum }_{i=1}^N\left|\mathrm{\Delta }{d}_i\right|}$, where $\mathrm{\Delta }{d}_i=\langle d_i\left({\text{Na}}^{+}\right)\rangle -\langle d_i\left({\text{K}}^{+}\right)\rangle$ and N is the number of distances, as well as its variance $\mathrm{\Delta }\mathrm{\Delta }D={\left[\left(1/N\left(N-1\right)\right){\displaystyle {\sum }_{i-1}^N{\left(\left|\mathrm{\Delta }{d}_i\right|-\mathrm{\Delta }D\right)}^2}\right]}^{1/2}$.

Note Added in Proof.

The companion manuscript by Claxton et al. (59) describes experimental spectroscopic measurements for NorM-VC, a close homolog of PfMATE, that are consistent with the conclusions of this computational study—namely that there exists a binding site in the N-lobe of this class of prokaryotic transporters with dual specificity for Na⁺ and H⁺, and that a H⁺-dependent site also exists in the C-lobe.