We first assessed whether C-terminal tail truncation affected the ability of EmrE to confer resistance to toxic substrates, the well-established primary function of this transporter. Growth assays of MG1655-ΔemrE E. coli cells expressing WT-, Δ107-, or E14Q-EmrE show that all strains grow well in the absence of toxic compounds (Figure 2B). In these assays, uninduced leaky expression from a low copy number plasmid (p15 origin) with a pTrc promoter keeps transporter expression relatively low (Spreacker et al., 2022). In the presence of ethidium bromide, a substrate commonly used to assess the activity of EmrE and other multi-drug efflux pumps, a functional transporter is required for survival (Figure 2C). This confirms the known resistance activity of WT-EmrE, that E14Q-EmrE is non-functional, and that Δ107-EmrE is properly expressed to the inner membrane, folded and functionally able to confer resistance to toxic compounds in a manner comparable to the WT transporter.

There is a small but reproducible growth defect for cells expressing Δ107-EmrE in the absence of exogenous substrate (Figure 2A, B). This defect becomes apparent around 5 hr, the point at which available fermentable sugars in LB media are depleted, increasing dependence on the proton motive force for energy production (Baev et al., 2006). A similar time-dependent growth inhibition is observed for WT-EmrE in the presence of harmane, and we have previously shown that this substrate triggers uncoupled proton leak through EmrE (Spreacker et al., 2022). Thus, while ∆107-EmrE competently performs the proton-coupled drug antiport necessary to confer resistance to toxic substrates, it is detrimental to E. coli in the absence of known small molecule substrate in a manner suggestive of proton leak.

To directly test this hypothesis, we measured proton leak in proteoliposomes. The pH-sensitive dye pyranine was encapsulated inside proteoliposomes at pH 6.5, and the liposomes were then diluted 100-fold into pH 7.5 buffer. If protons leak out of the liposome (down the proton concentration gradient), the internal pH will rise and pyranine fluorescence will increase. We compared the fluorescence, normalized to time zero, of proteoliposomes with WT, ∆107, or E14Q-EmrE. WT-EmrE proteoliposomes show a gradual increase in internal pH over time (Figure 3A, solid black), which is faster than the pH change for E14Q-EmrE proteoliposomes (Figure 3A, solid gray). This is consistent with a small amount of proton leak through WT-EmrE and a role for E14 in mediating leak. ∆107-EmrE proteoliposomes show a much faster rise in internal pH in this assay (Figure 3A, solid red). Repeating the experiment with the protonophore CCCP in the external buffer results in rapid proton leak for all proteoliposome samples (Figure 3A, dotted lines), and the results match the timescale and amplitude of proton leak observed for the ∆107-EmrE proteoliposomes. Thus, C-terminal tail truncation causes rapid proton leak through EmrE.

Initiation of the assay by dilution complicates measurement of the fluorescence baseline, obscuring rapid changes in the few seconds between dilution and initial fluorescence read. We repeated the assay with side-by-side dilution of proteoliposomes (internal pH 6.5) into pH 6.5 buffer (baseline), or pH 7.5 buffer (transport). This baseline normalization reveals a rapid internal pH change (Figure 3B). The empty liposome control is flat, showing that the liposomes do not leak without EmrE. However, the signal is non-zero (Figure 3B, black arrow), likely due to residual exterior pyranine. The time 0 fluorescence of WT-, E14Q-, and ∆107-EmrE proteoliposomes is much higher (Figure 3B, blue arrow), indicating an additional rapid change in the internal pH in the presence of protein. While the encapsulated pyranine is protected from the direct impact of external pH change, proton transport (leak) through EmrE will change internal pH. The transmembrane pH gradient may also affect EmrE itself, altering the pKa of key residues (E14, H110) and causing rapid release of protons inside the liposome, or cause a rapid burst phase of leak as the transporter transitions from a symmetric-pH conformation to an asymmetric-pH conformation.

To further assess the initial rapid pH change, we used solid supported membrane electrophysiology (SSME) to measure ∆pH-driven current in proteoliposomes since this technique provides a continuous readout as a gradient is applied. Many proteoliposomes can be adsorbed onto the gold-coated sensor, enabling highly sensitive detection of electrogenic transport. The same lipid to protein ratio was used as in the pyranine assay, and reported values are an average of three independently prepared sensors per mutant to account for variability in liposome adsorption onto sensors. The liposomes are first equilibrated with external buffer identical to the interior, and then a different external buffer is rapidly washed over the liposomes to create a transmembrane pH gradient while recording is in progress. The capacitive current is measured (Figure 3C, Figure 3—figure supplement 1, left A, B) and integrated to yield the total transported charge in response to the applied gradients (Figure 3D, Figure 3—figure supplement 1, right, Figure 3—figure supplement 2C). In the absence of drug, empty liposomes have minimal charge movement as expected for minimal proton leak. However, both WT- and Δ107-EmrE have measurable current in the presence of a pH gradient, and ∆107-EmrE has consistently higher leak under all pH conditions (Figure 3—figure supplement 1). The consistency of the SSME and pyranine assay results establishes the validity of the assay for comparing WT- and ∆107-EmrE proton leak and the ability to perform multiple assays with the same proteoliposome sensors to compare flux at different absolute pH or gradient magnitudes (Figure 3—figure supplements 1 and 2).

Both WT- and ∆107-EmrE also show increased net charge movement at higher absolute pH (Figure 3—figure supplement 1). In symmetric pH environments, the only amino acid side chains with pKa values near neutral pH are E14 (pKa 6.8 ± 0.1 and 8.5 ± 0.2 at 25°C) and H110 (6.98 ± 0.01, 7.05 ± 0.02) (Thomas et al., 2018; Morrison et al., 2015). ∆107-EmrE is lacking Histidine (H110), so any proton binding/release from H110 that contributes to the capacitive current will be absent in ∆107-EmrE, but the net charge transport (Figure 3D, Figure 3—figure supplement 1, right, Figure 3—figure supplement 2C, D) is greater for ∆107-EmrE than for WT-EmrE. This rules out a simple model where proton binding/release from H110 accounts for the fast proton flux. Any conformational change involving the C-terminal tail that contributes to the capacitive current should also decrease as average pH increases and H110 protonation and net charge decrease. However, the opposite pH dependence is observed for WT-EmrE, indicating that this is unlikely to be a major contributor to the SSME current. Furthermore, any rapid release of protons from E14 upon gradient formation will be decreased at high pH as the initial protonation state, and thus the number of protons that can be released, is reduced. Thus, the increased charge movement at higher absolute pH must be due to increased proton leak since there is no other substrate present in these experiments. Figure 3—figure supplement 2 shows the transient current and net transported charge at low pH, high pH, and under conditions where a drug gradient drives transport. These assays show multiphasic behavior, and close examination of the data with different lipid to protein ratios under each experimental condition further distinguishes pre-steady state (proton release, conformational change) and steady state (transport) processes that contribute to net charge movement. A model where movement of the C-terminal tail regulates access to E14 and C-terminal truncation alters this gating process would be consistent with the observed currents and their pH dependence (Figure 3C–E, Figure 3—figure supplements 1 and 2), as well as the longer-timescale change in intraliposome pH (Figure 3B). We note that the 1-s SSME traces do not reach equilibrium as the current is not zero and net transported charge is still changing at the end of the assay (Figure 3D, Figure 3—figure supplements 1, right and 2), but do provide insight into the dead time of the pyranine assay (Figure 3B) and match the relative magnitude of the observed burst phase. Altogether, this data supports a role for the C-terminal tail as a secondary gate that minimizes proton leak through WT-EmrE in the absence of substrate.

We next used solution NMR to assess the impact of C-terminal tail truncation on the structure and dynamics of EmrE, since this will impact gating and transport. Due to the asymmetric structure of EmrE, the two subunits have unique chemical shifts. As EmrE undergoes alternating access, the two subunits swap conformations, resulting in exchange between AB and BA dimer topology (Morrison et al., 2015). The rate of the alternating access exchange process affects the NMR line shape, resulting in distinct sets of peaks for each subunit when exchange is slow, line broadening as the rate increases, and eventually coalescence into a single set of peaks at the average chemical shift when exchange is fast. Thus, the appearance of simple 2D ¹H-¹⁵N TROSY-HSQC spectra can provide significant insight into both the structure and dynamics of the transporter under different conditions.

In the absence of drug, Δ107-EmrE is in fast-intermediate exchange at both pH values, with only one set of peaks and significant line broadening (Figure 4A). However, the spectra are distinct, with slightly more line broadening at high pH. This indicates that protonation of E14 still affects the overall structure of ∆107-EmrE, and alternating access is slightly slower at high pH. WT-EmrE has similar fast exchange behavior at low pH, but increasing pH results in a significantly slower rate of alternating access (Figure 4—figure supplement 1; Morrison et al., 2015).

Upon addition of a tight binding drug-substrate, TPP⁺ to ∆107-EmrE at low pH, two distinct peaks become visible for each residue (Figure 4B, red). This confirms that the poor spectral quality of the substrate-free spectrum was due to protein motion and not degradation or aggregation, and that alternating access is significantly slower with TPP⁺ bound. The similarity of this spectrum of ∆107-EmrE bound to TPP⁺ at low pH with the spectrum of WT-EmrE under the same conditions (low pH, TPP⁺ bound) also provides additional evidence that the general structure of Δ107-EmrE remains intact and the binding site has undergone minimal perturbation upon truncation of the last four amino acids (Figure 4—figure supplement 1A).

At high pH, TPP⁺-bound ∆107-EmrE shows significant line broadening and partial coalescence of the two distinct sets of peaks in the NMR spectrum, indicating that the rate of alternating access is faster (Figure 4B, blue). We quantitatively measured the rate of alternating access as a function of pH for TPP⁺-bound ∆107-EmrE with ¹H-¹⁵N TROSY-ZZ-exchange NMR experiments (Li and Palmer, 2009). In this experiment, a delay is inserted in the pulse sequence between recording the ¹⁵N and ¹H chemical shifts, such that a conformational exchange during this delay will result in the appearance of cross-peaks with the ¹⁵N chemical shift of the original state and ¹H chemical shift of the final state (Figure 4C, D). By comparing the intensity of these cross-peaks relative to the auto-peaks as a function of the delay time, we can determine the rate of alternating access (Miloushev and Palmer, 2005; Figure 4E, Figure 4—figure supplement 2). The alternating access rate for TPP⁺-bound Δ107-EmrE is 4 ± 1 s^–1 at low pH, and 17 ± 3 s^–1 at high pH. There is greater scatter in the peak intensity ratio at high pH due to enhanced exchange with water for residues on the open face of the transporter, which reduces the peak intensity. However, there is no overlap between low pH and high pH, clearly demonstrating a significant change in alternating access rate for ∆107-EmrE with pH. TPP⁺-bound WT EmrE has the same rate of alternating access as ∆107-EmrE at low pH, but does not vary significantly with pH (Morrison and Henzler-Wildman, 2014). Thus, truncation of the C-terminal tail alters the pH dependence of alternating access for ∆107-EmrE in both the absence and presence of drug substrates, supporting a role for this region in regulating the pH-dependent conformational dynamics of EmrE.

To further investigate how the C-terminal tail of EmrE may interact with other regions of EmrE and gate access into the transport pore, we carried out MD simulations of substrate-free WT-EmrE and ∆107-EmrE in a DMPC lipid bilayer. The protonation states were set to simulate a pH between 7.0 and 8.0, where only E14^A with the higher pKa is protonated. We used an NMR structure determined with TPP⁺ (PDB: 7JK8) as the initial structural model (Shcherbakov et al., 2021). Since this NMR structure does not include the C-terminal tail, we modeled it with CHARMM-GUI (Lee et al., 2016). The systems were first equilibrated at constant temperature 310 K and 1 bar pressure for 400 ns, while position restraints were gradually released. After this equilibration, the RMSD of the protein compared with the initial structure plateaued. The production simulations were run for another 1000 ns, and MD trajectories were output every 0.1 ns. With the same starting structure, we ran three parallel replicas to ensure consistency (Figure 5, Figure 5—figure supplements 1 and 2).

A critical prerequisite of proton transport is a water wire, either transient or long-lasting, that allows protons to transport through the Grotthuss hopping mechanism (Agmon, 1995). In the ∆107-EmrE MD simulations, we identified a water chain (aka ‘water wire’) not seen in WT simulations that connects E14 at the primary binding site to bulk water. It enters the protein from the open side near R106^A, passing through the triad of A61^A, I68^B, and I71^B, and then comes into the primary binding site (Figure 5C). To quantitatively understand the connectivity of this water wire over time, we calculated the length for the shortest water path S for each frame in the trajectory with graph theory. This length is defined such that smaller values for S reflect better connectivity of the water molecules, as described in more detail in the methods section (Li and Voth, 2021a). The logarithm of the shortest path, log(S), is plotted for all simulation systems (Figure 5A). For ∆107-EmrE, there is a leak state characterized by a much smaller log(S) where the water wire is very well connected, while for WT-EmrE, log(S) is consistently large. The existence of this water chain in ∆107-EmrE is consistent with the enhanced proton leak observed experimentally. This newly found water wire starts very close to the C-terminus and is distinct from the ligand entry path (Jurasz et al., 2021).

In the initial structure of WT-EmrE, the C-terminus is floating in bulk water. After equilibration, we observed the tail coming closer and interacting with the protein in all three replicas (Figure 5—figure supplement 3). The tail has two notable interactions with other parts of the protein, the first of which is a salt bridge between D84 and R106. The second involves the carbonyl group of the C-terminus, which forms hydrogen bonds with T56 and occasionally forms a salt bridge with K22 (Figure 5—figure supplement 4). When these interactions occur, the tail moves near the water chain. Examining characteristic snapshots from the simulation trajectories illustrates the effect of the C-terminal tail on water wire formation. Figure 5B shows the worst-hydrated snapshot of WT-EmrE determined by the highest S. Figure 5C shows the best-hydrated snapshot of ∆107-EmrE determined by the lowest S. Despite the overall dryness in the channel of WT-EmrE, there were a few rare moments where a transient water wire formed, characterized by a sudden drop in water path length, and Figure 5D shows the best-hydrated snapshot from that simulation. In WT-EmrE, the water wire is broken at the triad of A61^A, I68^B, and I71^B (Figures 5B and 6D), suggesting that these three hydrophobic residues may act as a bottleneck for the water wire.

To test this hypothesis with statistically meaningful results, we calculated the distances between the closest pairs of side-chain hydrogens among these three residues for the whole trajectory. For WT-EmrE, the minimum hydrogen–hydrogen distances of A61^A–I68^B and A61^A–I71^B are 2.9 ± 0.6 and 2.8 ± 0.6 Å, respectively, and for ∆107-EmrE, these distances increased to 5.9 ± 1.0 and 5.4 ± 1.2 Å (Figure 6A). The I68^B–I71^B distance does not differ significantly. These increased distances suggest that A61^A is moving away in the ∆107 variant, opening a pore for the water wire to form. For comparison, the diameter of a water molecule is measured to be around 2.7 Å (Schatzberg, 1967). This means a triangle larger than 5.4 Å in size is required for a water molecule to fit inside. Additionally, the most hydrated snapshot in WT-EmrE showed A61^A takes a conformation more similar to ∆107-EmrE and very different from dry WT-EmrE. This confirms the role of A61^A rotation in controlling the water wire formation.

Experimental testing of this hypothesis by mutagenesis is complicated by the small size and antiparallel topology of EmrE, as many residues play multiple functional roles, and mutation of any of these residues will perturb not only the proposed hydrophobic gate (A61^A, I68^B, and I71^B) but also the close packing necessary to close the transporter on the opposite face of the membrane where A61^B, I68^A, and I71^A are located. Prior scanning mutagenesis replacing A61, I68, and I71 with alanine, valine, glycine, or cysteine (Amadi et al., 2010; Mordoch et al., 1999; Wu et al., 2019) did not impact the ability to confer resistance to common EmrE substrates, such as ethidium, acriflavine, or methyl viologen. However, the mutation to tryptophan severely impaired ethidium resistance (Lloris-Garcerá et al., 2013), and mutation of any of these residues to cysteine impacts substrate binding (Amadi et al., 2010), demonstrating that these residues are functionally important.

The potential of mean force (PMF) for explicit proton transport (see Methods) shows the free energy change of the system as a function of a particular collective variable (CV) or reaction coordinate. It can provide additional information beyond structural snapshots of a reaction, which are only incomplete samples of the ensemble. For example, the existence of a water chain does not necessarily mean good proton conductance, but by contrast, the PMF for explicit proton transport (including Grotthuss proton shuttling) can provide key thermodynamic and kinetic information about the transport process (Li and Voth, 2021a; Ilan et al., 2004). However, free energy sampling involving proton transport is also intrinsically complicated. First, proton transport involves chemical bond breaking and formation, which is beyond the capability of classical MD, so expensive quantum mechanics/molecular mechanics may be required. Second, when an excess proton is solvated in the water, the net positive excess charge defect arising from the presence of the excess proton can be delocalized in the water network. It is not possible to define which proton is exactly the ‘excess proton’ as Grotthuss shuttling dynamically rearranges these definitions. To address these issues, we have developed a method called Multiscale Reactive MD (MS-RMD) (Kaiser et al., 2024). It can model bond forming and breaking involving excess proton shuttling at a computational cost near classical MD. In this approach, one can also conveniently define a ‘center of excess charge’ (CEC) to describe the location of the excess positive charge defect.

We carried out umbrella sampling with MS-RMD that describes the proton transport from E14^B to the bulk water in WT-EmrE. The CV ‘x’ is defined similarly as in reference (Li and Voth, 2021b) as the distance between the glutamate oxygen to the CEC, mapped along a vector that aligns with the proton transport direction (see Methods). The resulting PMF (Figure 6B, Figure 6—figure supplement 1) indicates a deep well near x = 0 Å, where the proton is on the glutamate, and a transition state near x = 10.0 Å. A conformational snapshot from the transition state shows the CEC (Figure 6C, orange sphere) is in close proximity to the pore defined by I68 and A61 (Figure 6D). The validity of this PMF calculation can be further supported by the pKa calculation from this PMF. The resulting pKa for this E14 is 7.1, close to the experimental value of 6.8 ± 0.1 at 25°C (or 7.0 ± 0.1 at 45°C) (Morrison et al., 2015). This supports the hypothesis that the bottleneck of proton transport is this hydrophobic gate.

Prior NMR pH titrations of WT-EmrE in the absence of drug-substrate revealed that the two essential E14 residues in the asymmetric homodimer have distinct pKa values (Morrison et al., 2015), reflecting their unique structural environments. Upon binding TPP⁺, one of the E14 residues is protected from protonation and no longer titrates, while the other (E14^A) retains a pKa of 6.8 ± 0.1, similar to the drug-free state (Morrison et al., 2015). The only other titratable residue previously identified in WT-EmrE is the C-terminal histidine (Thomas et al., 2018), which also has a pKa near neutral pH in WT-EmrE. Since H110 was removed by truncation in ∆107-EmrE, we expected only one protonation event in NMR pH titrations of TPP⁺-bound ∆107-EmrE (Figure 7A), corresponding to the one E14 residue that remains titratable when substrate is bound. A single protonation event would normally result in linear change in peak position from the chemical shift of the protonated state to the chemical shift of the deprotonated state over the course of the titration. This is because proton on-/off- is almost always in the fast-exchange limit for NMR, resulting in the observation of a single peak at the population-weighted average chemical shift at each titration point. However, several peaks exhibit distinctly curved titration paths with transitions in two different pH ranges (Figure 7—figure supplement 1) for TPP⁺-bound ∆107-EmrE. Plotting the chemical shift of well-resolved peaks as a function of pH yields titration curves that can be fit using standard pKa equations. In this case, the data is well fit with a global 2 pKa model yielding apparent pKa values of 5.6 ± 0.2 and 7.1 ± 0.2. The higher of these two pKa values is close to the E14^A pKa in TPP⁺-bound WT-EmrE, and the residues most sensitive to this protonation event are found lining the transport pore near E14^A, supporting the assignment of this pKa to E14^A (Figure 7D).

This leaves the lower pKa unaccounted for. Possibilities include other acidic residues in the loops, E25 or D84, or the C-terminal carboxylate itself (now at R106 in the ∆107-EmrE construct). The residues most sensitive to this lower pKa include R106 and the tail of subunit A, the TM1–TM2 loop of subunit B, and TM3–TM4 loop of monomer B, all of which are on the same ‘open’ face of EmrE (Figure 7C). Residues E25 and D84 are located in these loops and have previously been suggested to be part of a secondary binding site for substrates like TPP⁺ (Glaubitz et al., 2000). There is no evidence that these residues titrate in this pH range in WT-EmrE or in other mutants for which we have carried out NMR pH titrations (Robinson et al., 2017; Morrison et al., 2015; Wu et al., 2019), making it unlikely that E25 or D84 titrate in this pH range in the full-length transporter. However, truncation of the C-terminus in ∆107-EmrE could alter the structure, environment, and pKa of these residues. Indeed, the hydrogen bond between R106 and D84 observed in the WT-EmrE simulations (Figure 5B, D) is broken when the tail is truncated in ∆107-EmrE (Figure 5C), and D84 and R106 have some of the largest pH-dependent chemical shift changes. There are relatively few experimental reports of the pKa of the terminal carboxylate in proteins, but it has been reported to have a pKa as high as 5.9 for the partially buried C-terminus of subunit c of F₁F₀ ATP synthase (Grimsley et al., 2009). MD simulations show the C-terminal carboxylate in WT-EmrE hydrogen bonds with T56 (TM 2–3 loop) and K22 (TM 1–2 loop) on the open face of the transporter (Figure 5B, D), but the C-terminus in ∆107-EmrE no longer interacts with these residues (Figure 5C). Examining the residues that sense this lower pKa shows that the TM3–TM4 and TM1–TM2 loops on the open face have larger chemical shift changes associated with the low pKa protonation event than those loops on the closed face of EmrE, while the C-terminal tail residues in both subunits detect the lower pKa. Comparison with the MD simulations shows that the residues involved in the hydrogen bond networks anchoring the C-terminal tail over the pore align well with the full list of residues that are sensitive to the lower pKa in the NMR titrations, including T56_A (TM2–TM3 loop on the ‘open’ face), V15–G17, I37, Y40, V69, and S72–L73. Thus, although this lower pKa is likely an artifact of tail truncation, this data experimentally supports the importance of the interactions between the tail and the rest of EmrE identified in the MD simulations as important for disrupting the water wire and occluding the E14-binding site.

The well-established function of EmrE is proton-coupled antiport of toxic substrates, leading to toxin efflux and drug resistance. Recently, we discovered that some substrates, such as harmane, instead trigger uncoupled proton uniport, leading to ∆pH dissipation and defects in NADH production and growth in E. coli (Spreacker et al., 2022), essentially causing susceptibility rather than resistance. We suspected the enhanced proton leak observed through Δ107-EmrE and this harmane-triggered proton leak might have common elements in their underlying mechanism. Using SSME, we first compared the inherent proton leak through WT- and Δ107-EmrE in the absence of substrate. For the same ∆pH driving force, Δ107-EmrE has ≈3-fold greater proton leak than WT-EmrE (Figure 8A, D). However, upon addition of 16 μM harmane, there is a large increase in proton leak through WT-EmrE and a small increase in leak through ∆107-EmrE, such that this substrate triggers identical total leak through either transporter (Figure 8B, D, Figure 8—figure supplements 1 and 2). This SSME-detected proton leak increases with harmane concentration and is saturable in both WT- and ∆107-EmrE (Figure 8C, Figure 8—figure supplements 1 and 2). If harmane acts as an allosteric regulator of the transporter that can unlock the secondary gate, then a saturating amount of harmane will result in the maximal signal for the WT transporter as observed. This was also confirmed in a pH-detected liposomal assay where addition of harmane decreases the magnitude of the pH change upon addition of CCCP to WT-EmrE containing proteoliposomes relative to empty liposomes (Figure 8—figure supplement 3). In Δ107-EmrE, C-terminal truncation removes the majority of the secondary gate and key residues in the allosteric site, rendering proton leak comparably independent to harmane (Figure 8, Figure 8—figure supplements 1 and 2).

To test this theory in the native organism, we conducted in vivo growth assays with WT- and Δ107-EmrE in the presence of harmane. MG1655-ΔemrE E. coli cells constitutively expressing WT-, E14Q-, and Δ107-EmrE from a plasmid were grown in the presence of 25 μM harmane. The cells expressing E14Q grew equally well in the presence or absence of harmane, as the mutation of the primary binding site prevents proton binding in the transport pore and abolishes any proton leak (Figure 8E). In the presence of harmane, the difference in growth between Δ107- and WT-EmrE is eliminated, with significant growth defect for both constructs relative to E14Q-EmrE (Figure 8E, F, Figure 8—figure supplement 4). This in vivo data exactly matches the in vitro SSME and pyranine transport assays, demonstrating the importance of the C-terminal tail in gating and allosteric regulation of EmrE in vivo.

Each EmrE construct was cloned into the pWB vector (Spreacker et al., 2022), a low copy number plasmid vector with a p15A origin and pTrc promoter, and transformed into MG1655-ΔemrE E. coli. For experiments, LB plates were streaked and grown overnight at 37°C. In the morning, single colonies were picked to inoculate liquid LB cultures at 37°C. Once liquid cultures reached log phase growth, they were diluted back to an OD600 of 0.2 and further diluted 20-fold into microplates with LB media containing the indicated amount of substrate. Growth in microplates at 37°C was monitored for 15 hr using a TECAN Spark or BMG-Labtech microplate reader at OD700 (Ethidium) or OD600. Reported growth curves and final ODs are mean values of two biological replicates containing technical triplicates, with errors calculated using the standard deviation of the mean.

Protein expression utilized BL21 (Gold) DE3 E. coli transformed with a pET15b plasmid containing the respective EmrE construct, with cells grown in M9 minimal media. Protein was solubilized in decyl maltoside (DM) detergent and purified using immobilized nickel chromatography and size exclusion chromatography as previously described (Morrison and Henzler-Wildman, 2014).

Triple resonance backbone walk experiments were acquired for backbone assignment of TPP⁺-bound ∆107-EmrE at pH 5.5 and 45°C using a sample with 1.25 mM ²H,¹⁵N,¹³C ∆107-EmrE and 16 mM TPP⁺. TROSY-HNCA, TROSY HNcoCA, and TROSY-HNCACB experiments were acquired on a 900 MHz Bruker Avance III NMR spectrometer equipped with a TCI cryoprobe, and TROSY HNCO, TROSY-HNcaCO experiments were acquired on a 750 MHz Bruker Avance III NMR spectrometer equipped with a TCI cryoprobe. Amide assignments were transferred to other pH values using pH titrations. 2D TROSY-HSQC and TROSY-selected ZZ-exchange spectra of ∆107-EmrE at pH 5.5 or 8.5, and TPP⁺-bound ∆107-EmrE at pH 5.5 or 7.7, were acquired on an 800 MHz Varian VNMRS DD spectrometer equipped with a 5-mm cryoprobe at 45°C using samples with 0.7–1 mM ²H,¹⁵N ∆107-EmrE using standard pulse sequences with gradient coherence selection. 70% of the backbone resonances of TPP⁺-bound Δ107-EmrE were assigned at pH 5.5 by combining standard triple resonance experiments (TROSY-HNCA, TROSY-HNCACB, TROSY-HNCO, and TROSY-HN(CO)CA) with ZZ- exchange data. For NMR pH titrations, identical samples were prepared at the extreme pH values, and the two samples were gradually mixed to create intermediate pH values, ensuring constant protein, lipid, and salt concentrations across the titration.

To analyze the ZZ-exchange experiments, peak intensities were fit using the nlinls function in nmrPipe to accurately extract peak parameters. Residues for analysis were chosen that had all four peaks (two auto peaks, I_AA and I_BB, and two exchange cross-peaks, I_AB and I_BA) resolved in the 2D planes. Exchange with water reduces the peak intensity of the auto and cross-peak from the open face of the transporter at high pH, resulting in greater scatter for the high pH data. The peak intensity ratio was calculated using the method developed by Miloushev and Palmer, 2005 the Palmer lab:(1)Peak intensity ratio=IABIBAIBBIAA−IABIBA=k2t2\begin{document}$$\displaystyle Peak \ intensity \ ratio=\frac{I_{AB}I_{BA}}{I_{BB}I_{AA}- I_{AB}I_{BA}}=k^{2}t^{2}$$\end{document}

Calculation of this peak ratio cancels out initial peak intensity and intrinsic relaxation rates to first order and depends on the mixing time (t) of the ZZ-exchange experiment in a simplified manner as shown in the equation above. Since the forward and reverse rate constants are identical for EmrE in bicelles (Morrison and Henzler-Wildman, 2014), there is only a single rate constant for alternating access, k.

All data were acquired on a TECAN spark instrument. The excitation wavelength was 465 nm (35 nm bandwidth) and the emission wavelength was 530 (25 nm bandwidth). The excitation spectrum maximum of pyranine shifts from 400 to 450 nm as pH increases, so with a constant 465 nm excitation wavelength, the observed fluorescence signal will increase as pH increases. The number of flashes was set to 30 to reduce well-to-well measurement time. To minimize instrument integration time, replicates were allowed to equilibrate for the full 30 min, and an average of the Z-position and gain recorded by the instrument was used as manual input for the reported assays. Liposome stocks with an internal buffer concentration of 100 mM MOPS, 20 mM NaCl, and 1 mM pyranine, pH 6.5. Aliquots were first pipetted into the plate, which was then input into the instrument, and the assay was started to perform instrument checks, at which point the instrument was paused. The plate was ejected, and 198 μl of 100 mM MOPS, 20 mM NaCl pH 7.5 buffer was pippeted into the well containing the liposomes and returned into the instrument to begin recording as soon as possible. Conditions with CCCP contained 1 μl of CCCP at 200 μg/ml on the opposite side of the well for a final concentration of 1 μg/μl. No gradient conditions were diluted into 198 μl of 100 mM MOPS, 20 mM NaCl pH 6.5 buffer. Reported data are average values of three replicate wells recorded for 30 min each to minimize well-to-well measuring times, with error bars representing the standard deviation of the mean.

All SSME data were acquired on a Nanion SURFE2R N1 instrument. Liposome aliquots were thawed, diluted fourfold, and briefly sonicated. 10 μl of liposomes were added to prepare 3 mm sensors according to a standard protocol (Thomas et al., 2021). For comparison of different mutants, sensors were prepared side-by-side for all variants (including all replicates) on the same day using a single batch of sensors to ensure maximum similarity in proteoliposome loading onto the sensor. While results obtained with different batches of sensors prepared on different days show similar results in terms of relative leak between variants, the absolute value varies from batch to batch and day to day. Thus, while ∆107-EmrE was always leakier than WT-EmrE, the absolute flux through the WT- or ∆107-transporter varied between batches of sensors prepared. Data was not averaged or compared across different batches of sensors. Equivalence of the SSME data and pyranine assay demonstrates the success of this approach. Prior to experiments, sensor capacitance and conductance values were obtained to ensure sensor quality. For all experiments, both internal and external buffers contained 50 mM MES, 50 mM MOPS, 50 mM bicine, 100 mM NaCl, and 2 mM MgCl2, with the pH and drug concentration as indicated for each dataset. For data acquisition, sensors were equilibrated with internal buffer, and transport was initiated by perfusion of the external buffer before re-equilibration with the internal buffer. Signals were obtained by integrating the current during perfusion of the external buffer, with the final 100 ms of the initial buffer equilibration used as the baseline. Reported data are average values of data recorded from at least three separate sensors, with error bars representing the standard deviation of the mean.

Liposomal transport assays were performed as previously described (Robinson et al., 2017). Briefly, 1 ml aliquots with internal buffer (50 mM MOPS pH 7, 100 mM KCl) were thawed and extruded the day of the experiment as described above. The samples were run over 2 PD-10 spin columns (Cytiva) equilibrated in external buffer (50 μM MES pH 6 with 1 mM KCl and 99 mM NaCl) following the manufacturer’s spin protocol. Samples were then diluted to 1.5 ml in external buffer. Eluted samples were added to 2 ml cuvettes with a stir bar, and a microelectrode was inserted and allowed to equilibrate. The pH was monitored in real time by a WINDAQ DI-710 from DataQ at a rate of 100 per second. Aliquots of valinomycin and CCCP at 1 mg/ml in 100% DMSO were thawed and diluted by half in external buffer to better match the pH. During the recordings, valinomycin was added to a final concentration of 1 μg/ml to create a ΔΨ, harmane to a concentration of 100 µM, CCCP to a concentration of 1 μg/ml as a control, and 50 nmol of HCl was added for quantification.

All MD simulations were conducted with GROMACS 2020.4 (Abraham et al., 2015). Simulation inputs were generated by CHARMM-GUI membrane bilayer builder (Lee et al., 2016). The protein was solvated by 162 DMPC molecules, and 40 mM NaCl was added to the water to neutralize the system. The system was coupled to a Nose–Hoover thermostat (Nosé, 1984; Hoover, 1985) and a Parrinello–Rahman barostat (Parrinello and Rahman, 1981), at 310 K and 1 bar, respectively. The system was minimized, then equilibrated with position constraints gradually releasing, as by the default setting of CHARMM-GUI. Then, the system was further equilibrated for 400 ns without constraint, where the RMSD plateaued, and the box sizes were stable. For one simulation, we observed lipids penetrating the protein in this 400 ns equilibration, so we added another 100 ns simulation with backbone constraints to further equilibrate the membrane before releasing these constraints again. Figures were rendered with ChimeraX. The hydrogen distances were analyzed with PLUMED (The PLUMED consortium, 2019), by computing the softmin of all hydrogen–hydrogen distances between two sidechains with β = 500.

The water path length calculation was implemented in an in-house modification of PLUMED. The algorithm can be briefly described as follows: Each water oxygen is considered a node in a graph, and the distance for each edge connecting two nodes is determined by a function that is close to 1 when the oxygen-oxygen distance is smaller than r0 and grows rapidly when it is larger than r0. This r0 is set to 3 Å, which is the typical distance between the oxygens of hydrogen-bonded water. A more detailed discussion can be found in reference (Li and Voth, 2021a). Then, for each frame in the trajectory, the shortest path is found for the graph. We used the two oxygens of E14B at the starting point and the midpoint of Cα of R102A and G57A as the destination.

The simulations were run with the LAMMPS MD engine, and the umbrella sampling was carried out as implemented in PLUMED (The PLUMED consortium, 2019; Thompson et al., 2022). The codes were co-compiled with RAPTOR, a plug-in to model proton transport reactions (Mordoch et al., 1999). The source code of RAPTOR is available at https://github.com/uchicago-voth/raptor (Teng, 2024). The starting structure was taken from the classical MD simulation of WT-EmrE. A water molecule was protonated at the mouth of the channel, and steered MD was then used to create initial configurations at different CV values. A total of 43 umbrella windows spanning from CV = 0.0 Å to 15.0 Å were used (Figure 6—figure supplement 1), with a varying restraint force constant of 80–15 kcal/mol/Ų. The CV is defined as(2)x=dOC⋅ePT\begin{document}$$\displaystyle x=d_{OC}\cdot e_{PT}$$\end{document}

where d_OC is a vector pointing from the closer glutamate oxygen to the CEC and e_PT is a unit vector of the direction of proton transport. Each umbrella window was equilibrated for 1 ns, and then the production run was for 2 ns. The PMF was reconstructed with the weighted histogram analysis method (WHAM, version 2.1.0) (Grossfield, 2013). In these simulations, the Cα of the residues at least 10 Å away from the path and those in TM1–TM2 were restrained to its initial coordinate with a 2.4 kcal/mol/Ų harmonic potential to ensure the bias force does not unrealistically distort the protein conformation.