Microsecond Motion of the Bacterial Transporter EmrE in Lipid Bilayers

Abstract

The bacterial transporter EmrE is a homo-dimeric membrane protein that effluxes cationic polyaromatic substrates against the concentration gradient by coupling to proton transport. As the archetype of the small multidrug resistance family of transporters, EmrE structure and dynamics provide atomic insights into the mechanism of transport by this family of proteins. We recently determined high-resolution structures of EmrE in complex with a cationic substrate, tetra(4-fluorophenyl)phosphonium (F₄-TPP⁺), using solid-state NMR spectroscopy and an S64V-EmrE mutant. The substrate-bound protein exhibits distinct structures at acidic and basic pH, reflecting changes upon binding or release of a proton from residue E14, respectively. To obtain insight into the protein dynamics that mediate substrate transport, here we measure ¹⁵N rotating-frame spin-lattice relaxation (R_1ρ) rates of F₄-TPP⁺-bound S64V-EmrE in lipid bilayers under magic-angle spinning (MAS). Using perdeuterated and back-exchanged protein and ¹H-detected ¹⁵N spin-lock experiments under 55 kHz MAS, we measured ¹⁵N R_1ρ rates site-specifically. Many residues show spin-lock field dependent ¹⁵N R_1ρ relaxation rates. This relaxation dispersion indicates the presence of backbone motions at a rate of about 6000 s⁻¹ at 280 K for the protein at both acidic and basic pH. This motional rate is three orders of magnitude faster than the alternating access rate but is within the range estimated for substrate binding. We propose that these microsecond motions may allow EmrE to sample different conformations to facilitate substrate binding and release from the transport pore.

Graphical Abstract

Introduction

Protein dynamics are crucial for myriad biological processes, including substrate binding and recognition and conformational changes required to bring key functional groups into proximity. To understand the mechanism of action of proteins, an atomic-level description of both the equilibrium ground-state structure and the transient excited-state structures is required ^1–3. In the case of multidrug transporters, dynamics are important for the proteins to interconvert between multiple conformations of the transport cycle. Small multidrug resistance (SMR) transporters efflux quaternary ammonium compounds and other lipophilic cations out of bacterial cells against their concentration gradients to confer resistance to antiseptic and antibiotic compounds ⁴. To achieve this, these transporters alternately open to each side of the membrane to move substrates across the lipid bilayer ⁵. Importantly, SMR transporters efflux a wide variety of cytotoxic compounds with different shapes, sizes, and chemical properties ^6–7. Thus, these proteins need to change both the sidechain and backbone conformations to accommodate the diverse substrates ^8–9. The conformational plasticity that enables this substrate promiscuity ^10–11 has made it difficult to fully characterize the structure and dynamics of these small transporters.

Insights into the structure, dynamics, and mechanism of action of SMR proteins can be gained by studying the archetype member of this family of transporters, EmrE. Found in the inner membrane of the gram-negative bacteria Escherichia coli ^12–13, EmrE is implicated in pH and osmotic stress responses of bacteria, biofilm formation, and resistance to toxic polyaromatic cations ^14–17. Biophysical studies have shown that the transport mechanism of EmrE is complex: The protein can couple the export of toxic substrates to proton import, but can also perform substrate-gated proton uniport, and potentially drug uniport or proton-drug symport ^{18 19}. This mechanistic complexity, together with the substrate promiscuity, implies the existence and dynamic interconversion of many conformational states of the protein. Although structural understanding of EmrE has advanced rapidly in recent years, the available structures only represent a limited set of functionally relevant states.

Recently, we discovered an S64V mutant of EmrE that possesses the transport activity of the wild-type protein but exhibits slower alternating access ²⁰, thus providing a good system for high-resolution structure determination. Simultaneously, we developed a high-sensitivity ¹H-¹⁹F dipolar coupling solid-state NMR technique to measure inter-atomic distances up to about 2 nm in a high-throughput manner ²¹. These two advances allowed us to determine the high-resolution structures of EmrE in lipid bilayers using solid-state NMR. The structures were solved for the protein complexed to a fluorinated analog of the classical substrate, 4-fluoro-tetraphenylphosphonium (F₄-TPP⁺). The complex was studied at acidic pH (pH 5.8) ²² and basic pH (pH 8.0) ²³, which correspond to the proton-bound and putative periplasm-facing state versus the proton-unbound and putative cytoplasm-facing state, respectively. These structures provided important insights into the structural basis for the weaker substrate affinity at low pH, when EmrE simultaneously binds proton and substrate.

Here we report site-specific dynamics of F₄-TPP⁺-bound S64V-EmrE in DMPC bilayers at both low and high pH. We measured ¹⁵N R_1ρ relaxation rates and their dispersion to determine the rates and amplitudes of protein backbone motions on the microsecond (μs) to millisecond (ms) timescales ^24–28. ¹⁵N relaxation dispersion under MAS is similar to R_1ρ relaxation in solution, but the former is additionally sensitive to dynamics with rates of tens of kilohertz due to their dependence on the MAS frequency. We conduct high-sensitivity ¹H-detected ¹H-¹⁵N 2D correlation experiments ^29–31 under 55 kHz MAS to obtain residue-specific ¹⁵N R_1ρ relaxation rates of the protein. Our results indicate a global motion with an exchange rate of ~6000 s⁻¹, with varying exchange amplitudes for different TM helices and between different pH. We discuss how these motions may be associated with substrate binding and release ³² as well as the intrinsic fluctuations of the protein to recognize and accommodate diverse substrates.

Materials and Methods

Membrane protein samples

The expression, purification, and membrane reconstitution of S64V-EmrE have been described recently ^{20, 22–23}. Samples used in this study were prepared from the recent structure determination studies. Briefly, ¹³C, ²H, ¹⁵N (CDN)-labeled S64V-EmrE was expressed in deuterated media containing 2.5g/L d₇-²H, ¹³C-labeled glucose, 1 g/L ¹⁵NH₄Cl and CDN-labeled ISOGRO. The protein was purified using Ni-NTA affinity column chromatography, followed by thrombin cleavage of the His-tag and size exclusion chromatography. The CDN-labeled EmrE was reconstituted into d₅₄-DMPC (Avanti Polar Lipids) liposomes in protonated buffer by detergent dialysis using 0.5% octyl-glucoside (OG). Thus, all amide hydrogens and other labile hydrogens were back-exchanged to protons. The molar ratio of the protein monomer to DMPC lipids (P : L) was 1 : 25. The proteoliposomes were incubated with excess solid F₄-TPP⁺ at room temperature for at least 16 hours. Excess F₄-TPP⁺ was then removed by microcentrifugation (20,000 x g, 5 min). Proteoliposomes were pelleted at 100,000 x g and 4°C for 2 h in an ultracentrifuge, then dried to 40% water by mass in a desiccator. Samples were centrifuged into 1.3 mm MAS rotors. Two samples were used in this study: a pH 5.8 sample containing 0.9 mg S64V-EmrE in a total of ~3.9 mg proteoliposomes, and a pH 8.0 sample containing 0.9 mg S64V-EmrE in ~3.6 mg proteoliposomes.

Solid-State NMR experiments

All ¹⁵N R_1ρ experiments were conducted on a 600 MHz (14.1 T) Bruker AVANCE III HD spectrometer using a 1.3 mm HCN probe. The samples were spun at 55 kHz, and the sample temperature was maintained at 280 K to match the temperature at which chemical shifts were assigned ^22–23. This sample temperature was reached by setting the bearing gas to 253 K and keeping the water ¹H chemical shift δ_H2O at 4.94 ppm, based on the empirical relation T_eff (K) = 96.9 x (7.83 – δ_H2O) ³³. Additional ¹⁵N R_1ρ data were measured on the low-pH sample at a sample temperature of 305 K, using a bearing gas temperature of 275 K. ¹H chemical shifts were internally referenced to match the DSS-referenced solution-state ¹H chemical shifts of the protein. ¹⁵N chemical shifts were externally referenced to the ¹⁵N peak of N-acetylvaline at 122.0 ppm on the liquid ammonia scale.

¹⁵N R_1ρ relaxation dispersion data of EmrE were measured using a dipolar 2D hNH ¹⁵N spin-lock (SL) pulse sequence (Fig. 1a). The experiment starts with ¹H-¹⁵N cross polarization (CP), followed by a ¹⁵N SL pulse of duration τ_SL. The relaxation rates were measured for different SL field strengths ν₁. A 180° ¹H pulse was applied in the middle of the ¹⁵N SL period to suppress ¹H-¹⁵N J and dipolar coupling. No high-power ¹H decoupling was necessary during the spin lock because the sparse ¹H network combined with a fast MAS frequency of 55 kHz largely averages the ¹H-¹⁵N dipolar coupling ^{24, 28}. The ¹⁵N SL was followed by ¹⁵N chemical shift evolution (t₁) under ¹H WALTZ decoupling and ¹³C J decoupling by a 180° ¹³C pulse. The ¹⁵N magnetization was then stored along the z axis for 200 ms, during which water suppression was achieved using the MISSISSIPPI sequence ³⁴ at an rf field strength of 15 kHz. The ¹⁵N magnetization was next read out by a 90° pulse and transferred back to ¹H by CP for detection under 10 kHz WALTZ decoupling on both the ¹⁵N and ¹³C channels. For the two ¹H-¹⁵N CP steps, we applied the zero-quantum (ZQ) condition of 0.7ν_r for the ¹⁵N rf field (38.5 kHz) and 1.7ν_r for the ¹H rf field (93.5 kHz). The ¹⁵N rf field strength was ramped 70–100% for ¹H-¹⁵N forward CP and 100–70% for ¹⁵N-¹H reverse CP. The CP contact times range from 800 to 1100 μs. For the hard pulses, we used rf field strengths of 83.3 kHz for ¹H, 50 kHz for ¹⁵N and 62.5 kHz for ¹³C.

Figure 1.

2D ¹⁵N R_1ρ relaxation NMR for investigating microsecond motion of membrane-bound EmrE. (a) Pulse sequence diagram of the 2D hNH resolved ¹⁵N spin-lock experiment. (b-e) 2D hNH spectra of F₄-TPP⁺-bound EmrE in DMPC bilayers at different pH under different spin-lock field strengths. (b) Spectra of the pH 5.8 sample measured with ¹⁵N τ_SL values of 0.5 ms and 80 ms at a spin-lock field strength of 2 kHz. (c) Spectra of the pH 5.8 sample measured with τ_SL values of 0.5 ms and 80 ms at a spin-lock field strength of 12.5 kHz. (d) Spectra of the pH 8.0 sample measured with ¹⁵N τ_SL values of 0.5 ms and 80 ms at a spin-lock field strength of 2 kHz. (e) Spectra of the pH 8.0 sample measured with ¹⁵N τ_SL values of 0.5 ms and 80 ms at a spin-lock field strength of 12.5 kHz. Residues in subunit A and B are denoted by a letter A and B after the residue number.

The rf field strength (ν₁) of the ¹⁵N SL pulse was carefully calibrated using nutation experiments. To correct for differential rf heating between experiments with different τ_SL values ³⁵, we applied a constant-energy (CE) compensation pulse with duration τ_CE after ¹H detection. The sum of τ_CE and τ_SL is set to 80.1 ms for all the experiments within each R_1ρ decay series.

For the pH 5.8 sample at 280 K, we measured the 2D hNH spectra at eight ¹⁵N SL mixing times (0.5, 2.5, 5.0 10.0, 20.0, 40.0, 60.0, and 80.0 ms) for each SL field strength. Eight ¹⁵N SL field strengths (2.0, 3.0, 4.0, 5.0, 6.0, 7.5, 10.0, and 12.5 kHz) were used. Therefore, a total of 64 2D spectra were measured. For the pH 8.0 EmrE sample at 280 K, the τ_SL values were 0.5, 1.0, 2.5, 5.0, 10.0, 20.0, 40.0, and 80.0 ms and the ¹⁵N SL field strengths were 2.0, 2.5, 3.0, 4.0, 5.0, 6.0, 7.5, 10.0, and 12.5 kHz. We also measured the ¹⁵N R_1ρ of the pH 5.8 sample at a higher temperature of 305 K. For this condition, we chose five mixing times (0.02, 1.0, 3.0, 6.0, 12.0 ms) and four ¹⁵N SL field strengths (2.0, 4.0, 8.0, and 12.0 kHz). For all 2D spectra, ¹⁵N chemical shift evolution was measured to 45.0 ms using 300 complex t₁ data points. ¹H detection was acquired to 25.0 ms using 2500 complex data points. 2D hNH spectra were zero-filled to 4096 complex points in the direct dimension and 1028 complex points in the indirect dimension, then Fourier-Transformed and phase corrected in the TopSpin software. A QSINE apodization with an SSB parameter of 3 was applied to both dimensions.

Extraction of ¹⁵N R_1ρ relaxation times

The ¹⁵N R_1ρ relaxation modulated 2D hNH spectra were processed using a combination of the Topspin processing function “serial” and Python scripts. Peak intensities were integrated in Topspin by defining an integral region (intreg) file for a reference spectrum, then using the same file for all other 2D spectra using serial processing commands. The integral values were extracted from Bruker’s int2D file using a custom Python script along with the ¹⁵N chemical shift of each peak for subsequent correction of the measured R_1ρ relaxation rates to on-resonance R_1ρ values. The integrals of each peak were normalized to the maximum intensity for that residue within a relaxation time series for a specific spin-lock field strength. Uncertainty in the normalized integrals was estimated from the signal-to-noise ratio (SNR) of the peaks and propagated using 2σ errors:

$${2\mathrm{*}\sigma }_{I/I_0}=2\mathrm{*}\frac{I}{I_0}\mathrm{*}\sqrt{{\left(\frac{1}{SN{R}_I}\right)}^2+{\left(\frac{1}{SN{R}_{I_0}}\right)}^2}$$ (1)

Here $I$ and $I_0$ are the integrated intensities of the observed peak and its maximum value in each series. $SN{R}_I$ and $SN{R}_{I0}$ are the respective signal-to-noise ratios. The observed relaxation decay rates for each peak and each spin-lock field strength were fit to a single-exponential decay function using the SciPy optimization module in Python to extract $R_{1\rho ,obs}$ ³⁶:

$$I\left(t\right)=A{e}^{-R_{1\rho ,obs}t}$$ (2)

We estimated the uncertainties (${\sigma }_{R_{1\rho }}$) in the fitted ${\mathrm{R}}_{1\mathrm{\rho }}$ parameters using a Monte Carlo analysis as described before ³⁷. Briefly, 1000 datasets were simulated by multiplying a random number drawn from a standard normal distribution, $p\left(x\right)=\frac{1}{\sqrt{2\pi {\sigma }^2}}{e}^{-\frac{{\left(x-\mu \right)}^2}{2{\sigma }^2}}$ with $\mathrm{\mu }=0$ and $\mathrm{\sigma }=0.3$. These parameters were chosen such that nearly all random numbers fall between −1 and 1. This random number was multiplied with the normalized integral error, ${2*\sigma }_{I/I_0}$ (Eqn. 1), and the resulting value was added to the measured intensity $I/I_0$ to generate the simulated datasets. This procedure randomizes each normalized integral within its 2σ error. These 1000 Monte Carlo datasets were then fit with the same single-exponential decay function using SciPy optimization. The standard deviation of the fit parameters to the Monte Carlo datasets was used as the ${\sigma }_{R_{1\rho }}$ value, and the 2 * ${\sigma }_{R_{1\rho }}$ errors were reported.

The experimentally measured $R_{1\rho ,obs}$ values were converted to on-resonance ${\mathrm{R}}_{1\mathrm{\rho }}$ rates ³⁷ according to:

$$R_{1\rho }=[R_{1\rho ,obs}-{\cos }^2\theta \cdot R_1]/{sin}^2\theta$$ (3)

where $\mathrm{\theta }$ is the angle between the effective ¹⁵N spin-lock field and the z-axis. This angle is related to the spin-lock field strength ${\mathrm{\omega }}_1$ and chemical shift offset $\mathrm{\Omega }$ of each peak by:

$$\tan \theta ={\omega }_1/\mathrm{\Omega }$$ (4)

In equation (3), an ¹⁵N $R_1$ relaxation rate of ~0.2 s⁻¹ was common for solid proteins ^38–39. This ${\mathrm{R}}_1$ term is a small correction compared to $R_{1\rho ,obs}$, which range from 2.3 to 45.9 s⁻¹ for 0.5 and 99.5 percentile values for the EmrE samples studied here.

The ¹⁵N R_1ρ rates as a function of ${\mathrm{\omega }}_1$ were fit to the two-state Bloch-McConnell equation ^{37, 40–41}:

$$R_{1\rho }\left({\omega }_1\right)=R_{1\rho ,0}+{\varphi }_{ex}{k}_{ex}/\left({\omega }_1^2+k_{ex}^2\right)$$ (5)

Here $R_{1\rho }^0$ is the rotating-frame spin-lattice relaxation rate in the limit of infinitely strong ${\mathrm{\omega }}_1$ and $k_{ex}$ is the rate constant of exchange. ${\varphi }_{ex}$ is the exchange amplitude, and depends on the populations of the two states, $p_A$ and $p_B$, and the isotropic chemical shift difference $\mathrm{\Delta }\delta$ between them according to ${\varphi }_{ex}=p_A{p}_B\mathrm{\Delta }{\delta }^2$. Because populations and chemical shift differences are highly correlated, we obtain their product ${\varphi }_{ex}$ from the fitting. We conducted a global fit of the R_1ρ dispersion of all residues using a single $k_{ex}$ value and variable ${\varphi }_{ex}$ using the LmFit Python package. This was done by setting $k_{\mathrm{e}\mathrm{x}}$ to be equal for all residues. The $k_{\mathrm{e}\mathrm{x}}$ was bound to a physically meaningful range of 0 to (2π∗600)², where the maximum value corresponds to a 10-ppm ¹⁵N chemical shift difference on the 14.1 Tesla magnet, and p_A = p_B = 0.5. Fit convergence was tested by attempting 10 global fits with random initial values of R_1ρ,0, ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$, and $k_{\mathrm{e}\mathrm{x}}$. Uncertainties in the Bloch-McConnell fitting were estimated using the same Monte-Carlo method as described above using 200 simulated datasets, and 2σ errors were reported. As an alternative approach, we also tested the fitting of the R_1ρ dispersion data by allowing different $k_{\mathrm{e}\mathrm{x}}$ values for different residues.

The ¹⁵N R_1ρ relaxation data and bar diagrams of R_1ρ and ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ values were plotted using the Matplotlib program in Python ⁴². Protein structural views of R_1ρ and ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ values were generated in PyMol ⁴³ using a Python-PyMol script that uploaded the R_1ρ and ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ values as NumPy arrays ⁴⁴ into the b-factors of the coordinate files. We used the ‘Putty’ representation and the ‘spectrum b’ option for coloring the values. Residues for which no data was available were set to exceedingly small values of 0.1 and were colored gray.

Intrinsic Tryptophan Assays

Purified S64V-EmrE were reconstituted into DMPC/DHPC isotropic bicelles (q=0.33) as described before. Purified EmrE was reconstituted into liposomes as described above using an EmrE : DMPC molar ratio of 1 : 75. Bicelle stock (2 x) was prepared separately at pH 5.8 and pH 8.0. Each bicelle stock had 300 mM DMPC in the assay buffer (100 mM MES pH 5.8 or 100 mM Bicine pH 8 and 20 mM NaCl) and was incubated at 45°C for 1.5 hrs, then 100 mM DHPC was added, and the solution was incubated for another hour before being subjected to 3 freeze-thaw cycles. F₄-TPP⁺ was prepared at a maximal concentration of 1 mM in assay buffer with 1 x bicelle stock at either pH 5.8 or pH 8.0. 1.3 mM S64V-EmrE in DMPC/DHPC bicelles at pH 7.0 was diluted with 2 x bicelle stock and 1 x assay buffer to create separate stocks of 40 μM EmrE dimer at pH 5.8 and pH 8.0. 50 μl of this protein solution was mixed with F₄-TPP⁺ and assay buffer (200 μl total volume) in black 96-well flat-bottom plates and incubated at room temperature for one hour. Samples for pH 5.8 and 8.0 were prepared separately and each F₄-TPP⁺ concentration was run in triplicate. Endpoint fluorescence was determined using a TECAN Spark with an excitation wavelength of 280 nm (15 nm bandwidth) and an emission wavelength of 340 nm (20 nm bandwidth). The measurement integration time was 40 μs with 50 flashes. The Z-position and gain were determined automatically by the TECAN instrument from the A1 position.

Results

Quantitative studies of EmrE dynamics have so far focused on millisecond-to-second dynamics that are associated with rate-limiting steps in the transport process. To investigate whether the protein undergoes faster internal motions, and whether the protonation states of the two E14 residues affect the protein dynamics on faster timescales, we measured residue-specific ¹⁵N rotating-frame spin-lattice relaxation (R_1ρ) rates of the EmrE-TPP complex. We used CDN-labeled and 100% back-exchanged S64V-EmrE bound to F₄-TPP⁺ in DMPC bilayers, and conducted ¹H-detected 2D hNH correlation experiments with an ¹⁵N spin-lock period (Fig. 1a). ¹H detection dramatically increases the sensitivity of the R_1ρ measurements ⁴⁵. The experiment was conducted at 280 K to match the temperature at which the ¹H and ¹⁵N chemical shifts had been measured and assigned.

Under MAS, ¹⁵N R_1ρ rates depend on the spectral densities $\mathrm{J}\left(\mathrm{\omega }\right)$ at the sum and difference frequencies of the spin-lock field strengths ${\mathrm{\omega }}_1$ and MAS frequency ${\mathrm{\omega }}_{\mathrm{r}}$: ^{26, 28, 37, 46–47}.

$${\mathrm{R}}_{1\mathrm{\rho },\mathrm{N}\mathrm{H}}=\frac{1}{20}{\left[\frac{{\mathrm{\mu }}_0}{4\mathrm{\pi }}\frac{\mathrm{\hslash }{\mathrm{\gamma }}_{\mathrm{H}}{\mathrm{\gamma }}_{\mathrm{N}}}{{\mathrm{r}}_{\mathrm{N}\mathrm{H}}^3}\right]}^2\left[\frac{2}{3}\left(\mathrm{J}\left({\mathrm{\omega }}_1+2{\mathrm{\omega }}_{\mathrm{r}}\right)+\mathrm{J}\left({\mathrm{\omega }}_1-2{\mathrm{\omega }}_{\mathrm{r}}\right)\right)+\frac{4}{3}\left(\mathrm{J}\left({\mathrm{\omega }}_1+{\mathrm{\omega }}_{\mathrm{r}}\right)+\mathrm{J}\left({\mathrm{\omega }}_1-{\mathrm{\omega }}_{\mathrm{r}}\right)\right)+3\mathrm{J}\left({\mathrm{\omega }}_{\mathrm{N}}\right)+\mathrm{J}\left({\mathrm{\omega }}_{\mathrm{H}}-{\mathrm{\omega }}_{\mathrm{N}}\right)+6{\mathrm{J}(\mathrm{\omega }}_{\mathrm{H}})+6\mathrm{J}({\mathrm{\omega }}_{\mathrm{H}}+{\mathrm{\omega }}_{\mathrm{N}})\right]$$ (6)

$$R_{1\rho ,N_{CSA}}=\frac{1}{45}{\left({\sigma }_{\parallel }-{\sigma }_{\mathrm{⟂}}\right)}^2\left[\frac{2}{3}\left(J\left({\omega }_1+2{\omega }_r\right)+J\left({\omega }_1-2{\omega }_r\right)\right)+\frac{4}{3}\left(J\left({\omega }_1+{\omega }_r\right)+J\left({\omega }_1-{\omega }_r\right)\right)+3J\left({\omega }_N\right)\right]$$ (7)

For ${\mathrm{\omega }}_1/2\mathrm{\pi }$ values of 2–12.5 kHz and ${\mathrm{\omega }}_{\mathrm{r}}/2\mathrm{\pi }$ of 55 kHz, motions with rates of 55–132 × 10³ s⁻¹ modulate the ¹⁵N-¹H dipolar coupling and ¹⁵N chemical shift anisotropy (CSA) to cause R_1ρ relaxation. Therefore, the ¹⁵N R_1ρ rates measured under our experimental conditions are sensitive to motions on the timescale of 8–18 μs. The effects of microsecond-timescale motion on dipolar- and CSA-driven ¹⁵N R_1ρ relaxation rates under MAS for perdeuterated and back-exchanged proteins have been numerically simulated and experimentally validated on model proteins ^{24, 26, 28}.

Fig. 1b–d show representative 2D hNH spectra of the EmrE-TPP complex at pH 5.8 and pH 8.0, measured using varying ¹⁵N spin-lock durations τ_SL and spin-lock fields ν₁. Inspection of well-resolved peaks reveals that the relaxation rates are site-specific. For example, the T18A intensity decayed more rapidly than G67A under 2.0 kHz ¹⁵N spin-lock (Fig. 1b), indicating that T18 in the TM1 helix of subunit A has enhanced dynamics compared to G67 in the TM3 helix. Different residues also exhibit different relaxation dependence on the spin-lock field strength. For example, T18A relaxation is strongly field-dependent, with much slower relaxation under 12.5 kHz spin-lock than under 2.0 kHz spin-lock (Fig. 1c). In contrast, G67A relaxation has a weaker field dependence. Thus, T18A is mobile on the 80–500 μs timescale that is probed by the ¹⁵N spin-lock (2–12.5 kHz). The fact that the observed relaxation dispersion is residue-specific indicates that it does not result from coherent effects but reflects real differences in the motional rates and amplitudes of the residues. We also detected motional differences between the two subunits. For example, G26 of subunit A has slower relaxation than G26 of subunit B (Fig. 1b–e). This difference is present at all the spin-lock field strengths and at both acidic and basic pH, indicating that the motion is asymmetric between the two subunits. Finally, the high-pH spectra show more pronounced intensity decays with τ_SL than the low-pH spectra, indicating that the high-pH EmrE-TPP complex is more dynamic. For example, G67B and R82A retain intensities after 80 ms spin-lock in the low-pH complex (Fig. 1b, c) but lost the intensities by this time in the high-pH sample (Fig. 1d, e).

To obtain the rates and amplitudes of the EmrE motion, we extracted the intensity decays of the cross peaks at all spin-lock field strengths. These decays fit to single exponential functions well, giving ¹⁵N R_1ρ relaxation rates (Fig. 2). Many residues such as I62A and G9B (Fig. 2a, d) show significant differences in the R_1ρ rates between the low-pH and high-pH samples, with the high-pH complex showing faster decays for all sites. Significant relaxation dispersion is observed for some of the residues. For example, at high pH, T18A has an ¹⁵N R_1ρ of 14.1 s⁻¹ under 2.0 kHz spin-lock but only 6.1 s⁻¹ under 12.5 kHz spin-lock (Fig 2c). Other residues such as I62A and I62B show little relaxation dispersion (Fig. 2a). This confirms that the observed R_1ρ dispersion does not result from coherent contributions to ¹⁵N relaxation but reflects conformational motions for some of the backbone amides. After correcting for ¹⁵N chemical shift offsets, the largest pH dependence of R_1ρ relaxation is found in the TM3 helix and in the loop connecting the TM3 and TM4 helices (Fig. S1). On average, the high-pH complex exhibits faster relaxation than the low-pH complex. For example, at 2 kHz spin-lock, the average R_1ρ value for residues G57A-S75A in TM3 was 10.7 s⁻¹ at low pH and increased to 16.3 s⁻¹ at high pH. The TM3 helix contains the S64V mutation, which causes a pronounced reduction of the alternating access rate ²⁰. Thus, motion of the TM3 helix is important for opening and closing the substrate-binding site.

Figure 2.

Representative ¹⁵N R_1ρ decay curves of EmrE at high and low pH at different spin-lock field strengths. In each panel, the pH 5.8 data (red) and pH 8.0 data (blue) are overlaid and compared, and the R_1ρ rates are given. Shaded areas represent the 2σ uncertainty around the best fit curve based on Monte Carlo simulations. (a) I62A in TM3. (b) I62B in TM3. (c) T18A in TM1, one helical turn from the crucial E14 residue. (d) G9B in TM1. Residues in subunit A and B are denoted by a letter A and B after the residue number.

To visualize the site-resolved ¹⁵N R_1ρ relaxation in the protein, we color-coded the measured ¹⁵N R_1ρ rates into the antiparallel asymmetric dimer structure of TPP-bound EmrE (Fig. 3). For clarity, the two subunits, as defined by their NMR chemical shifts, are shown separately, and the high- and low-pH structures are oppositely oriented to illustrate how the two subunits swap conformations to switch between the open-in and open-out states. It is worth noting that the high- and low-pH structures do not strictly correspond to the putative open-in and open-out conformations in the cell. The cytoplasmic pH is maintained at 7.4–7.6 in E. coli while the periplasmic pH is close to the external pH ⁴⁸. Laboratory media used for many drug resistance assays is pH 7.4, while E. coli is found in a wide variety of pH in the gastrointestinal tract, urinary tract, intracellular bacterial communities, and other environments. Therefore, whether the cytoplasm is more acidic or more basic than the periplasm depends on the environment. What is clear is that proton binding lowers the affinity of the protein for small-molecule substrates and enhances their off-rate, which is consistent with the more peripheral location of F₄-TPP⁺ in the low-pH structure. The residue-specific ¹⁵N R_1ρ rates (Fig. 3) indicate that the fastest relaxing residues are located at the ends of the TM helices and in loops near the lipid-water interface. For example, S24A, F27A, A52B, Y53A, I58B, S75A, G80B, R82B and L83A exhibit fast relaxation. Increasing the pH caused TM3 residues and the C-terminal half of TM1 to exhibit the largest increase in the R_1ρ relaxation.

Figure 3.

Residue-specific ¹⁵N R_1ρ rates of F₄-TPP⁺-bound EmrE at 280 K, color-coded onto the protein structures ^22–23. These R_1ρ rates were measured at a spin-lock field strength of 2 kHz. (a) ¹⁵N R_1ρ rates at pH 5.8. (b) ¹⁵N R_1ρ rates at pH 8.0. The two structures are shown open to opposite sides of the membrane to illustrate the open-in and open-out conformations. While the E. coli cytoplasm is homeostatically maintained near pH 7.4–7.6, the periplasmic pH varies with the external environment and may be lower or higher than the cytoplasmic pH. The fastest-relaxing residues (red) are observed at the loops connecting the TM helices while the slowest relaxing residues (blue) are observed in the middle of the TM1 and TM3 helices. The high-pH complex exhibits faster R_1ρ relaxation than the low-pH complex.

While the ¹⁵N R_1ρ relaxation is sensitive to 10⁵ s⁻¹ motions (Eqn. 6, 7), the R_1ρ relaxation dispersion across spin-lock field strengths of 2 to 12.5 kHz is sensitive to motions with rates of 10³–10⁴ s⁻¹. Using the Bloch-McConnel two-state exchange model, we fit the rf-field dependence of the R_1ρ rates to Lorentzian functions. This fitting allowed the extraction of the populations of the two states, the chemical shift difference between them, and the exchange rate $k_{\mathrm{e}\mathrm{x}}$ (Eqn. 5). Using the offset-corrected relaxation rates, we fit the measured relaxation dispersion using a global $k_{\mathrm{e}\mathrm{x}}$ for all residues. Representative fits are shown in Fig. 4, and the full dataset is shown in Fig. S2. The global $k_{\mathrm{e}\mathrm{x}}$ value is 6.4 ± 1.3 × 10³ s⁻¹ for the low-pH complex and 5.9 ± 1.7 × 10³ s⁻¹ for the high-pH complex. Thus, the global exchange rate is mostly independent of pH. In comparison, different residues show different exchange amplitudes. Residues such as I62A exhibits little dispersion at low pH while residues such as K22B display significant dispersion at high pH. The exchange amplitudes are pH-dependent. For example, I62A displays no relaxation dispersion at low pH but large dispersion at high pH. In comparison, residues such as E14A show similar R_1ρ dispersion at both low and high pH.

Figure 4.

Representative ¹⁵N R_1ρ relaxation dispersion curves of F₄-TPP⁺-bound EmrE in DMPC bilayers. The pH 5.8 data are shown in red while the pH 8.0 data are shown in blue. The extracted R_1ρ° value and ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ value are given in each panel. Residues in subunit A and B are denoted by a letter A and B after the residue number. (a) E14A. (b) I62A. (c) W63A. (d) T18B. (e) K22B. (f) S43B. Best-fit ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ and R_1ρ° values were obtained using the two-state Bloch-McConnell model with a global exchange rate $k_{\mathrm{e}\mathrm{x}}$. Shaded areas represent the 2σ uncertainty around the best fit curve based on the Monte Carlo simulations.

The fact that the global exchange rates $k_{\mathrm{e}\mathrm{x}}$ are similar between the low-pH and high-pH complexes suggest that the outward-facing and inward-facing complexes undergo the same microsecond motion. Indeed, when we attempted to fit $k_{\mathrm{e}\mathrm{x}}$ residue-specifically, we obtained unphysically large variations in both $k_{\mathrm{e}\mathrm{x}}$ and ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ values for neighboring residues (Fig. S3), supporting the view that the microsecond motion observed here is global. To evaluate whether the two subunits of the protein may have different motional rates, we tested fitting the ¹⁵N R_1ρ dispersion separately for the two subunits, assuming a single $k_{\mathrm{e}\mathrm{x}}$ value for each subunit. For the pH 5.8 sample, the separate fitting did not yield significantly different $k_{ex}$ values compared to the full-protein fit (Fig. S4a, c): monomer A gave a $k_{\mathrm{e}\mathrm{x}}$ of 6.2 ± 2.4 ×10³ s⁻¹ whereas monomer B yielded a $k_{\mathrm{e}\mathrm{x}}$ of 6.6 ± 1.8 ×10³ s⁻¹. Both values are within the uncertainty of the full-protein $k_{\mathrm{e}\mathrm{x}}$ of 6.4 ± 1.5 s⁻¹. In comparison, separate-monomer fit of the high-pH data yielded distinct exchange rates of 1.3 ± 0.5 ×10³ s⁻¹ for monomer A and 10.1 ± 1.5 ×10³ s⁻¹ for monomer B, which differ significantly from the full-protein $k_{\mathrm{e}\mathrm{x}}$ of 5.9 ± 1.8 ×10³ s⁻¹ (Fig S4b, d). However, these separate-monomer fits yielded unphysically large exchange amplitudes for monomer A that would correspond to chemical shift differences of at least 10 ppm. These results suggest that the two monomers do not have significantly different microsecond motions. To obtain additional, more site-specific, dynamics information, measurements such as NERRD dispersion near the MAS frequency would be required ^{46, 49}.

Fig. 5 shows site-specific exchange amplitudes for the low-pH and high-pH EmrE-TPP complex based on the global exchange rates. The largest exchange amplitudes are observed in the TM3 helix, with S75A at pH 5.8 and S72A at pH 8.0 exhibiting the highest amplitudes. At high pH, many monomer B residues display large exchange amplitudes. For example, T18, K22, and G35 in the C-terminal half of TM1 to the N-terminal end of TM2 show enhanced ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ values. This region of the protein was previously found to have significant disorder in the high-pH structure ^22–23. In the high-pH complex, residues at the N-terminal end of TM3B and the C-terminal end of TM3A, such as I58B, S72A and S75A, exhibit large exchange amplitudes, consistent with the fact that these residues are also disordered in the structure.

Figure 5.

Site-resolved exchange amplitude ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ at 280 K, obtained from best fit of the ¹⁵N R_1ρ relaxation dispersion curves. (a) pH 5.8 data. (b) pH 8.0 data. The global $k_{\mathrm{e}\mathrm{x}}$ value of each sample is shown. Grey shaded bars indicate the positions of the interhelical loops. (c) ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ values color coded onto the pH 5.8 structure. (d) ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ values color coded onto the pH 8.0 structure. The protein is shown open to either side of the membrane to illustrate the different possible states, although the low- and high- pH structures do not necessarily correspond to the open-out and open-in structures, respectively.

These ¹⁵N R_1ρ measurements were carried out below the DMPC phase transition temperature. To investigate if the protein dynamics change significantly in the liquid-crystalline phase, we repeated the ¹⁵N R_1ρ experiments at 305 K for a smaller number of spin-lock field strengths on the low-pH sample. As expected, the high-temperature spectra have lower sensitivity (Fig. 6a) due to motional averaging of the ¹⁵N-¹H dipolar couplings. Nevertheless, we resolved many residues in the 2D hNH spectra, which allow us to obtain the R_1ρ rates for many residues in the TM1–3 helices. The R_1ρ rates are larger at 305 K than at 280 K, as seen by the fact that the remaining intensities in the 12.0 ms spectrum at 305 K (Fig. 6a, right) are much lower than the intensities of the 80.0 ms spectrum at 280 K (Fig. 6b, right). Intensity analysis showed that R_1ρ rates increased by an order of magnitude at the high temperature (Fig. 6c, left). The offset-corrected R_1ρ rates, mapped onto the structure (Fig. 6d), show similar trends as the low-temperature data: the most dynamic segments lie between TM1 and TM2 helices and at the ends of the TM3 helix. We also analyzed the ¹⁵N R_1ρ dispersion at 305 K. Global fitting using the Bloch-McConnel model yielded an exchange rate of 9.4 ± 3.8 × 10³ s⁻¹ for the low-pH complex, which is 1.5-fold faster than the rate of 6.4 × 10³ s⁻¹ found at 280 K (Fig. 6c). The exchange amplitudes also increased at high temperature. This is exemplified by S72A, whose ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ value increased from 3.2 ± 0.8 ×10⁵ rad²s⁻² at 280 K to 1.6 ± 1.3 ×10⁶ rad²s⁻² at 305 K (Fig. 6c, right). Interestingly, residues such as G8B, which did not show a significant ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ at 280 K, exhibit a large exchange amplitude at 305 K (Fig 6c). When mapped onto the protein structure, the high-temperature exchange amplitudes (Fig. 6e) show similar spatial variation as the low-temperature case, indicating that the same microsecond motions are observed below and above the membrane phase transition.

Figure 6.

Temperature dependence of the ¹⁵N R_1ρ relaxation of the low-pH EmrE-TPP complex. (a) 2D hNH spectra measured at 305 K, where the DMPC bilayer is in the liquid-crystalline phase. The spectra were measured with an ¹⁵N spin-lock field strength of 2 kHz. (b) 2D hNH spectra measured at 280 K. (c) Left: R_1ρ decay curves for representative residues at 305 K (orange) and 280 K (magenta); Right: relaxation dispersion curves. Both R_1ρ relaxation and relaxation dispersion increase with temperature. (d) 305 K ¹⁵N R_1ρ rates at 2 kHz spin lock at 305 K, color-coded onto the low-pH structure. Some of the fast-relaxing residues are indicated. (e) 305 K exchange amplitudes ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$, color-coded onto the low-pH structure. The ${\mathrm{\varphi }}_{\mathrm{e}\mathrm{x}}$ were obtained from best fit of the R_1ρ dispersion using a global $k_{\mathrm{e}\mathrm{x}}$ of 9.4 × 10³ s⁻¹.

These ¹⁵N R_1ρ data indicate that TPP-bound EmrE undergoes a global backbone motion at a rate of ~6000 s⁻¹ at both acidic and basic pH. The largest-amplitude motions occur in the C-terminal half of TM1 through the TM1-TM2 loop into the N-terminal portion of TM2, and in the TM3-TM4 loop. Since the water-exposed loop residues may have enhanced hydrogen exchange with the solvent compared to membrane-embedded helices, it is difficult to exclude a contribution of hydrogen exchange to the measured ¹⁵N R_1ρ rates at the experimental temperature of 280 K. However, the exchange amplitudes vary significantly for different loops and between high and low pH. For example, the TM3-TM4 loop in monomer A has much smaller exchange amplitudes at low pH than at high pH, and this loop also has distinct exchange amplitudes between monomer A and monomer B. These observations imply that hydrogen exchange cannot account for all the relaxation dispersion, and microsecond motion must be present. Within the helices, the TM3 helix, which contains the S64V mutation and several functionally important ligand-interacting residues such as W63, exhibits larger motional amplitudes than the other helices, and is more dynamic at high pH than at low pH. The motional rate $k_{\mathrm{e}\mathrm{x}}$ increases only moderately from 280 K to 305 K, indicating that this dynamics is relatively insensitive to the DMPC phase transition.

Discussion

The ¹⁵N R_1ρ relaxation rates and dispersion measured here provide new insights into the site-specific motions undertaken by substrate-bound EmrE. The surprising finding is that the S64V mutant, despite its slow alternating access of only 0.6 s⁻¹ at 45°C ²⁰, undergoes backbone exchange at a rate of about 6000 s⁻¹ at 280 K, which increases to 9400 s⁻¹ at 305 K. Therefore, EmrE undergoes global microsecond motions that are three orders of magnitude faster than the alternating-access motion. This microsecond motion occurs with similar rates for the high- and low-pH complexes. Previous measurement of 2D ¹H-¹⁵N TROSY spectra of apo EmrE in isotropic DMPC/DHPC bicelles found exchange rates of 500 s⁻¹ at 45°C ¹⁰, with an activation energy of 117 kJ/mol. This data led to the conclusion that EmrE is conformationally plastic, and the fast motion in the absence of ligand is not rate-limiting in the transport cycle. The current data, measured on TPP-bound EmrE, found even faster backbone motions, implying that the conformational plasticity of EmrE is independent of the substrate but is intrinsic to the protein.

We hypothesize that this microsecond motion is intrinsic to the EmrE structure and topology ⁵⁰, and may prime the protein for successful binding and recognition of the substrate, and allow the protein to overcome the energy barrier to interconvert between inward- and outward-facing states. Substrate transport across the membrane requires several microscopic steps: substrate binding, protein conformational change that switches binding site accessibility from one side of the membrane to the other, and substrate release. Proton-coupled transport additionally requires coordination of these steps between the transported small molecule and the driving ion, proton. Microscopic rate constants for all of these steps have been estimated for EmrE transport of TPP⁺ from a combination of NMR and stopped-flow studies ^51–53, and have been used to build a kinetic model of transport ³². Inspection of these estimated rate constants reveals that the on- and off-rates of proton and the small molecule are the only rates in the 10³-10⁴ s⁻¹ range, matching the motional timescale probed by the current ¹⁵N R_1ρ experiments.

The apparent substrate-binding affinity of EmrE is highly dependent on the identity of the substrate ⁵² and the pH ¹⁹. Stopped-flow data ⁵¹ showed that substrate on-rates are diffusion-limited whereas substrate off-rates vary with the small molecule being released. We used intrinsic tryptophan fluorescence to confirm that the fluorinated substrate, F₄-TPP⁺, has high affinity for S64V-EmrE (Fig. S5). While the protein concentration is constrained to the micromolar regime to maintain bicelle structure and EmrE dimerization, the quenching of intrinsic tryptophan fluorescence occurs at sub-micromolar concentrations. The affinity cannot be determined quantitatively in this titration regime ⁵⁴, but is similar to the affinity of hydrogenated TPP⁺ to wild-type EmrE, which ranges from a few nanomolar to a few hundred nanomolar depending on pH, temperature and the membrane-mimetic environment. Given the high affinity, the off-rate is expected to be slow, on the order of 10 s⁻¹ as found for TPP⁺. The reported on-rate for TPP⁺ is pH-dependent, ranging from 10⁵ M⁻¹s⁻¹ at low pH to 10⁷ M⁻¹s⁻¹ at high pH ⁵¹. As a result, the apparent substrate affinity is lower at low pH, requiring higher substrate concentration to saturate binding. While it is difficult to estimate the effective substrate concentration in the solid-state NMR rotor where there is little bulk solvent, substrate concentrations in the range of mM to tens of mM are used to prepare the samples. This would put the F₄-TPP⁺ on-rate in the range of 10³ s⁻¹, which is the same timescale as the dynamics observed here. For TPP⁺-bound wild-type EmrE, the proton off-rate is estimated to be 1600 s⁻¹, and proton release is tightly correlated with the small-molecule on-rate ⁵¹. This similarity suggests that the motion detected by these ¹⁵N R_1ρ experiments may report on the opening of the transporter loop regions to allow substrate entry.

Even in the substrate-free form, EmrE is dynamic, with rates of 300–500 s⁻¹ between 37°C and 45°C ¹⁰. This conformational plasticity is thought to be important for promiscuous recognition of many substrates by this transporter ^{10–11, 52}. Structurally, this implies that the substrate-binding pocket of the protein must be spacious to permit binding of differently shaped substrates. In turn, this means that the substrate should be dynamic. Indeed, F₄-TPP⁺ exhibits motionally averaged ¹⁹F NMR spectra ^22–23. Residues known to be important for TPP⁺ recognition are clustered in the TM1-TM3 helices, in close proximity to E14. Chemical shift differences between S64V-EmrE bound to TPP⁺ or F₄-TPP⁺ highlight those residues that directly interact with the substrate and sense the impact of even the small difference in hydrogenation versus fluorination of the substrate ²². Extensive mutagenesis of EmrE in the past twenty years also implicated additional residues in substrate binding and specificity. These residues are located near the ends of the TM helices and loops, in a pattern similar to the residues that undergo conformational exchange in the R_1ρ data. This again suggests that the microsecond dynamics plays a key role in permitting substrate entry and egress.

The pH insensitivity of the exchange rates ($k_{\mathrm{e}\mathrm{x}}$ values of 5.9–6.4 ×10³ s⁻¹) obtained from the ¹⁵N R_1ρ relaxation dispersion data is unexpected, since the structures of F₄-TPP⁺-bound S64V-EmrE are distinct between low pH and high pH ^22–23. TPP⁺ is positioned closer to the open side of the transport pore at low pH ²² but near the middle of the helical bundle at basic pH ²³. The distinct substrate positions suggest that F₄-TPP⁺ is primed for exit at acidic pH whenever the loops open sufficiently, whereas substrate entry to the high-pH complex may be limited by the movement of the substrate in the pore to access the deeply embedded E14. The pH insensitivity of the protein dynamics is echoed by the pH insensitivity of the substrate motion. 2D ¹⁹F-¹⁹F exchange spectra showed that F₄-TPP⁺ reorients at similar rates of 137–207 s⁻¹ in the high-pH complex and 38–209 s⁻¹ in the low-pH complex ²³ at 280 K. These data consistently indicate that the internal motions of the EmrE-substrate complex are not strongly controlled by the protonation state of E14, but are intrinsic to the protein. We hypothesize that the structural differences between the low and high-pH complexes and the rapid reorientation of TPP⁺ in the binding pocket require a high attempt frequency by the protein to interconvert between the inward-facing and outward-facing states. This high-frequency motion should be pH independent. Second, although most structural and kinetic studies focus on the role of E14 and the TM helices of EmrE, emerging evidence ^{18, 55–56} implicates that the C-terminal tail may play a role in regulating access to the E14-binding site. It is possible that the microsecond dynamics measured here may partly reflect pH-independent motion of the C-terminal tail in regulating substrate transport.

The existence of fast protein internal motions to accomplish slower functional processes is common to many membrane proteins. For example, proton conduction by the influenza M2 protein occurs at rates of 10–1000 s⁻¹, but is accompanied by histidine-water proton exchange at rates of 10⁵ s^{−1 57–58}. The rate-limiting step in proton conduction by M2 is the rearrangement of the four-helix bundle between two conformations, which are exclusively controlled by pH ^59–60.

Conclusion

In summary, the current data demonstrate that ¹H-detected ¹⁵N R_1ρ relaxation dispersion experiments under fast MAS are an effective approach for obtaining site-specific information about microsecond timescale motions in membrane transporters. Our data show that substrate-bound EmrE undergoes microsecond motions at both acidic and basic pH. These motions have a global rate of ~6000 s⁻¹ at 280 K at both pH. This rate is three orders of magnitude faster than the alternating access rate, and is best attributed to the protein dynamics to allow substrate binding and release. Among the four TM helices, the TM3 helix has the largest motional amplitude, indicating that this helix plays a key role in substrate recognition.