Combination of ¹⁵ N Reverse Labeling and Afterglow Spectroscopy for Assigning Membrane Protein Spectra by Magic-Angle-Spinning Solid-State NMR: Application to the Multidrug Resistance Protein EmrE

Abstract

Magic-angle-spinning (MAS) solid-state NMR spectroscopy has emerged as a viable method to characterize membrane protein structure and dynamics. Nevertheless, the spectral resolution for uniformly labeled samples is often compromised by redundancy of the primary sequence and the presence of helical secondary structure that results in substantial resonance overlap. The ability to simplify the spectrum in order to obtain unambiguous site-specific assignments is a major bottleneck for structure determination. To address this problem, we used a combination of ¹⁵N reverse labeling and afterglow spectroscopic techniques that dramatically improved the ability to resolve peaks in a crowded spectrum. This was demonstrated using the polytopic membrane protein EmrE, an efflux pump involved in multidrug resistance. Residues preceding the ¹⁵N reverse labeled amino acid were imaged using a 3D NCOCX afterglow experiment and those following were recorded using a frequency-selective dephasing experiment. Our approach reduced the spectral congestion and provided a sensitive way to obtain chemical shift assignments for a membrane protein where no high-resolution structure is available. This MAS methodology is widely applicable to the study of other polytopic membrane proteins in functional lipid bilayer environments.

Introduction

Membrane proteins account for ~30% of all open-reading-frames in eukaryotic genomes (Wallin and Heijne 1998), yet only encompass 1-2% of experimentally determined structures in the protein data bank (Hong et al. 2012). Like soluble proteins, the structure and interactions of membrane proteins with other biomolecules are encoded by the primary sequence. However, unlike soluble proteins, a hydrophobic environment is required to maintain solubility and function (White 2009). Atomic resolution of membrane protein structures has been possible with X-ray crystallography (Rasmussen et al. 2011; Toyoshima et al. 2011; Krishnamurthy and Gouaux 2012; Johnson et al. 2012) and solution NMR (Kim et al. 2009; Verardi et al. 2011; Hiller et al. 2008; Zhou et al. 2008; Liang and Tamm 2007) that rely on detergent or bicelle solubilization. While these methods have been successful for several membrane proteins, crystallization and the presence of non-planar surfaces can result in structural artifacts (Chen et al. 2007; Cross et al. 2011). An alternative methodology is solid-state NMR (SSNMR) spectroscopy that is designed to use or remove the anisotropy of NMR observables and therefore does not have an intrinsic size limitation. This technique has opened the possibility to study membrane proteins under fully functional conditions such as in liposomes (Shi et al. 2009; Gustavsson et al. 2012; Cady et al. 2010; Etzkorn et al. 2007; Bhate and McDermott 2012; Linden et al. 2011; Lehner et al. 2008), aligned bilayers (Park et al. 2006; Traaseth et al. 2009; Muller et al. 2007; Vostrikov et al. 2010; Durr et al. 2007; Sharma et al. 2010; Knox et al. 2010), and in native cell membranes (Renault et al. 2012) (Gustavsson et al. 2012).

Magic-angle-spinning (MAS) has emerged as the most popular technique within SSNMR to study microcrystalline soluble proteins (Gardiennet et al. 2008; Martin and Zilm 2003; Igumenova et al. 2004; Franks et al. 2006; Bertini et al. 2010; Sengupta et al. 2012), amyloid fibrils (Bayro et al. 2012; Jaroniec et al. 2004; Tycko 2011; Li et al. 2011; Paravastu et al. 2008), and membrane proteins reconstituted in liposomes (Shi et al. 2009; Gustavsson et al. 2012; Cady et al. 2010; Etzkorn et al. 2007; Bhate and McDermott 2012; Linden et al. 2011; Lehner et al. 2008; McDermott 2009; Hong and DeGrado 2012; Sackett et al. 2010). One of the major hurdles in this methodology when applied to helical membrane proteins is the spectral crowding in traditional 2D and 3D backbone assignment experiments. This spectral overlap stems from structural heterogeneity of the sample preparation, residual protein dynamics, similarity of the residues within the primary sequence (e.g., Val, Leu, Ile, and Ala), and the degeneracy of chemical shifts for helical secondary structures found in transmembrane domains. Since spectral assignment is the major bottleneck for structure determination, the presence of overlapping resonances significantly slows or halts this process. To overcome the assignment hurdles for membrane proteins (both in MAS SSNMR and oriented SSNMR), a divide and conquer isotopic labeling strategy is often employed to reduce the number of observed spin-systems in recombinant proteins. One of the most useful is the reverse labeling approach, in which the protein is unlabeled at a few, selected residue types (Vuister et al. 1994; Heise et al. 2005; Shi et al. 2009; Krishnarjuna et al. 2011). Other methods include selective labeling, where only a few residues are labeled (Park et al. 2006; Traaseth et al. 2006) and sparse labeling that relies on specifically-labeled precursors to create predictable patterns in the protein (Hong and Jakes 1999; Castellani et al. 2002; Lian and Middleton 2001; Verardi et al. 2012; Jaipuria et al. 2012; Yang et al. 2008; Schuetz et al. 2010; Jehle et al. 2006). These sample-preparation techniques are useful and have aided in the assignment of crystalline proteins and a few membrane proteins examples using MAS (Castellani et al. 2002; Shi et al. 2009; Schmidt-Rohr et al. 2012). In addition, spectroscopic techniques have also been developed to aide in residue identification (Traaseth and Veglia 2011; Schmidt-Rohr et al. 2012). In this Article, we present the application of two spectroscopic techniques, afterglow and frequency-selective dipolar recoupling (FDR), to be used in combination with ¹⁵N reverse labeling in order to reduce spectral overlap for polytopic membrane proteins.

Materials and Methods

EmrE Expression and Purification

EmrE was expressed as a fusion protein with maltose-binding protein (MBP) in E. coli BL21(DE3) cells. The [U-¹³C,¹⁵N] sample was grown in minimal media with ¹³C glucose and ¹⁵N ammonium chloride (Sigma-Aldrich). For the RevIL EmrE sample, ¹⁵N L-Leu (120 mg/L) and natural abundance L-Ile (120 mg/L) were added to the ¹³C/¹⁵N M9 media. The fusion was bound to amylose resin (New England Biolabs) and eluted using maltose and n-dodecyl-β-D-maltopyranoside (DDM). EmrE was cleaved from MBP using tobacco etch virus protease (TEV) and purified to >95% by size-exclusion chromatography using a Superdex 200 10/300 column (GE Healthcare Life Sciences).

MAS Sample Preparation

Detergent-solubilized EmrE in DDM at a concentration of 0.85 mg/ml was added to 25 mg DMPC to give a final lipid/protein of 75:1 (mol/mol). Detergent was removed with bio-beads (75 mg/mg detergent) and incubated for 13.5 hr. The proteoliposomes were pelleted by ultracentrifugation for 1.5 hr at 142,414 × g (max force) and packed into a 3.2 mm MAS rotor using sample spacers to prevent dehydration.

Solid-State NMR Experiments

All NMR experiments were carried out using a DDR2 Agilent NMR spectrometer operating at a ¹H frequency of 600 MHz. The variable temperature unit was set to -10 °C and the MAS rate was 12.5 kHz. Based on temperature standards, the actual sample was approximately +5 °C. The transfers from ¹⁵N to ¹³CA and ¹⁵N to ¹³CO utilized SPECIFIC-CP (Baldus et al. 1998) where the ¹⁵N offset was set to 121 ppm, the ¹³CA offset to 58 ppm, and the ¹³CO offset to 176 ppm. The ¹⁵N to ¹³CA transfer used a tangent adiabatic ramp (Baldus et al. 1996) on the ¹⁵N channel, while the ¹³CO SPECIFIC-CP utilized the ramp on the carbon channel. The Δ/2π and β/2π parameters of the adiabatic cross-polarization were set to 1.2 kHz and 0.3 kHz for the ¹⁵N to ¹³CA transfer and 3.7 kHz and 0.9 kHz for the ¹⁵N to ¹³CO transfer (Franks et al. 2007). The total length of the cross-polarization time was empirically optimized and set to 4.5 msec for the ¹⁵N to ¹³CA and 4 msec for the ¹⁵N to ¹³CO transfer. All datasets used a ¹³C spectral width of 100 kHz and an acquisition time of 20 msec. The [U-¹³C,¹⁵N] sample's indirect ¹⁵N dimension was acquired with a spectral width of 3125 Hz and 28 increments and its indirect ¹³C dimension was acquired with a spectral width of 3125 Hz and 24 increments. A total of 112 scans were used for the [U-¹³C,¹⁵N] sample. The RevIL sample had an indirect ¹⁵N spectral width of 3125 Hz and 24 increments, and an indirect ¹³C dimension spectral width of 3125 Hz and 28 increments. A total of 112 scans were used for the RevIL EmrE sample. For spin diffusion experiments between ¹³C spins, the DARR condition was set to the n=1 rotational resonance condition on ¹H (12.5 kHz) and the mixing time was set to 18 msec (Takegoshi et al. 2001). The afterglow NCOCX followed the acquisition of a standard NCACX with a 50 msec DARR mixing period.

The [U-¹³C,¹⁵N] and RevIL standard NCA experiments were acquired with the SPECIFIC-CP parameters listed above and 36 and 32 points in the indirect ¹⁵N dimension, respectively, with a spectral width of 3125 Hz. Both standard NCA experiments were acquired with 128 scans. The FDR-NCA experiment used π/2 (5.5 μsec) pulses centered on ¹³CO and π (11 μsec) pulses on ¹⁵N for a total dephasing time of 1.6 msec (Traaseth and Veglia 2011). The FDR-NCA was acquired with 30 points in the indirect ¹⁵N dimension with a spectral width of 3125 Hz and 128 scans. The FDR-NCACX with DARR was acquired with 512 scans and a 25 msec DARR mixing period. All data sets were acquired with the spectral width of the direct ¹³C dimension set to 100 kHz and an acquisition time of 20 msec. The ¹³C dimension was referenced to the CH₂ resonance of adamantane (40.48 ppm) (Morcombe and Zilm 2003) and indirectly to ¹⁵N using the relative frequency ratio between ¹⁵N and ¹³C of 0.402979946 (Wishart et al. 1995).

Results and Discussion

To evaluate the challenge of assigning a polytopic membrane protein by MAS spectroscopy, we carried out experiments on the 110-residue protein EmrE. The prototype for studying the small multidrug resistance (SMR) protein family (Figure 1A), EmrE is comprised of four transmembrane domains and functions as a homo-dimer to efflux a variety of cationic and aromatic molecules by coupling transport with the proton motive force across the inner membrane of bacteria (Schuldiner 2009; Korkhov and Tate 2009). EmrE represents a typical membrane protein in that it contains an abundance of hydrophobic Leu (16) and Ile (15) residues, approximately one-third of the primary sequence and nearly all within the transmembrane domains. In total, 47% of the residues within EmrE are comprised of Leu, Ile, Gly, or Ala residues. Unlike some of the rhodopsins, which naturally form well-ordered arrays in vivo (Neugebauer et al. 1978) and are structurally homogeneous at low lipid to protein ratios (Shi et al. 2009), EmrE and other membrane proteins require dilute lipid conditions to minimize nonspecific interactions between the functional oligomeric states. For this reason EmrE was reconstituted into DMPC lipid vesicles in a 75:1 lipid to protein molar ratio.

Figure 1.

Primary sequence of EmrE with the reverse labels that will act as filters in the spectroscopic approaches highlighted. (A) Sequence of EmrE with white circles highlighting [U-¹³C,¹⁵N] labeled residues, red circles indicating [¹⁵N-Leu] residues, and blue circles showing Ile residues labeled at natural abundance. (B) The afterglow NCOCX is expected to give strong peaks for residues preceding Leu (thick red box) with weaker signals for the other spin systems within the protein (thinner red box). The FDR experiment gives a filtered NCA correlation spectrum, where only residues appearing after the unlabeled residue give rise to peaks in an NCA spectrum (dotted-blue boxes).

Standard MAS Backbone Experiments Applied to a Polytopic Membrane Protein

Figure 2 shows a 2D NCA correlation MAS spectrum obtained using SPECIFIC-CP from ¹⁵N to ¹³CA. The ¹³C linewidths for resolved peaks were measured to be ~0.8-0.9 ppm, reflecting relatively narrow resonances for a non-crystalline membrane protein in lipid bilayers. As observed within the spectrum, the ¹⁵N resolution results in spectral overlap spanning ~10 ppm for the majority of non-Gly peaks. 3D experiments such as NCACX and NCOCX with DARR mixing improved the resolution (Figure 3A; spectra in black), but alone were insufficient to resolve all spin systems given the ¹⁵N spectral congestion. These results are consistent with those reported by Reif and co-workers for [U-¹³C,¹⁵N] EmrE (Agarwal et al. 2007) and demonstrate the need for alternative approaches to supplement standard backbone experiments.

Figure 2.

Standard 2D NCA of [U-¹³C,¹⁵N] EmrE in DMPC liposomes. Due to the sequence homogeneity of EmrE (47% of the residues are either Leu, Ile, Gly, or Ala) and secondary structure similarity (Fleishman et al. 2006), the NCA spectrum is congested with few peaks fully resolved. The NCA is plotted at 7.1σ.

Figure 3.

Comparison of standard and afterglow NCOCX experiments applied to [U-¹³C,¹⁵N] and RevIL, respectively. The afterglow approach is used as an isotopic filter to aide in the assignment of EmrE. (A, B) Strip plots from the standard 3D NCOCX acquired using [U-¹³C,¹⁵N] EmrE (black peaks) and the afterglow 3D NCOCX spectrum acquired on RevIL EmrE (red peaks). Note that the afterglow 3D NCOCX was acquired with 18 msec DARR and after an NCACX experiment with 50 msec DARR. Only Leu-X spin-systems were observed in the afterglow dataset with the exception of one Gly-X peak shown in panel B (boxed region; due to overlapping resonances). At a lower contour level, the Gly-X region in the boxed region appears the same as the standard NCOCX on [U-¹³C,¹⁵N] EmrE. (C) Additional strip plots from the 3D NCOCX afterglow dataset on RevIL EmrE. The standard NCOCX strips are plotted at 6.3σ and the afterglow NCOCX is shown at 6.6σ.

In order to reduce the spectral overlap to identify individual peaks, we prepared a reverse labeled EmrE sample by supplementing the ¹³C-glucose/¹⁵N-ammonium chloride media with ¹⁵N-Leu and natural abundance Ile (referred to as RevIL). This labeling scheme (Figure 1A) resulted in complete ¹³C and ¹⁵N incorporation at all sites except Leu (¹⁵N labeled) and Ile residues, and created inter-residue pairwise connectivity that we exploited in a novel manner to identify residues preceding Leu and those following Leu and Ile in the primary sequence (Figure 1B).

Afterglow Experiments Identify Residues Preceding the Reverse Labeled ¹⁵N Amino Acid

To image the residues prior to Leu in EmrE, we used the recently developed afterglow method (Banigan and Traaseth 2012). This pulse sequence uses the residual ¹⁵N magnetization from the cross-polarization element to acquire an additional 2D NCO (or 3D NCOCX) dataset in a sequential manner following the standard 2D NCA (or 3D NCACX) acquisition period (Banigan and Traaseth 2012). Since both acquisitions are carried out on ¹³C, only one receiver is necessary and therefore can be carried out on all standard SSNMR spectrometers. The beauty of this approach is that the NCACX is identical to the standard acquisition, which results in no loss in sensitivity, and the ability to acquire a second dataset with negligible increase in data acquisition time (< 2%).

Experiments were acquired on RevIL EmrE using a 3D NCACX acquisition followed by the afterglow 3D NCOCX. The NCACX experiment utilized a 50 msec DARR mixing period while the NCOCX had an 18 msec DARR. In our original implementation, the afterglow NCO dataset gave ~32% of the sensitivity with respect to the standard NCO experiment for [U-¹³C,¹⁵N] ubiquitin (Banigan and Traaseth 2012). For the RevIL EmrE sample, Leu residues do not have ¹³C incorporation at the CA position, resulting in minimal loss of ¹⁵N magnetization during the initial NCACX experiment. For these reasons, transfers from the ¹⁵N amide of Leu to the preceding ¹³C carbonyl of residue i-1 will result in more intense peaks in the afterglow NCOCX dataset than non-Leu labeled sites. Based on the labeling pattern and primary sequence, we expected to observe 12 Leu-related spin systems (there are 3 Leu-Leu pairs and an Ile-Leu pair in EmrE that would not be observed in the RevIL sample). Overall we identified eight strongspin-system peaks (shown in Figure 3), with seven related to Leu residues. The strong peaks were defined as those having peak intensities greater than 7σ (σ is the standard deviation of the noise in the spectrum). The missing five spin systems are likely due to overlap in the 3D strips resulting from sequence redundancies (e.g., T18-L19 and T50-L51; R29-L30 and R82-L83; V69-L70 and V98-L99), but can also arise from small variations in peak intensity across the primary sequence (e.g., T_1ρ spin relaxation rates). While the storage of ¹⁵N magnetization along the z-axis for the afterglow method will have a negligible impact on signal intensities due to long T₁ times (Gopinath and Veglia 2012; Giraud et al. 2004), differences in T_1ρ can impact observed signal intensities during the cross-polarization periods. The one non-Leu spin system observed in the afterglow spectrum resulted from overlapping X-Gly residues (i.e., peak clustering gave increased intensity). This region is shown in the boxed areas in Figure 3B. Upon reduction of the noise floor in the afterglow spectrum, we observed extra peaks in this region that were also seen in the standard NCOCX spectrum acquired on [U-¹³C,¹⁵N] labeled EmrE. This is related to the discussion above that the afterglow approach when applied to [U-¹³C,¹⁵N] samples gives the same spectral features (albeit at 32% of the signal intensity) compared to the standard NCOCX experiment (Banigan and Traaseth 2012). Nevertheless, the Gly peaks in the afterglow NCOCX do not pose a problem for identifying spin systems involving Leu residues due to the absence of Gly-Leu pairs in the primary sequence and the fact that Gly residues have distinct ¹³CA chemical shifts. In the case of proteins having Gly-Leu pairs, these peaks in the afterglow NCOCX will be stronger than the background Gly signals due to the lack of ¹⁵N Leu magnetization transfer to ¹³CA.

Identification of Residues Following the Reverse Labeled Amino Acid

As seen in Figure 4A and B, the standard NCA spectra for [U-¹³C,¹⁵N] and RevIL EmrE both are substantially overlapped, which complicates data analysis. We note that a previous MAS-SSNMR study on [U-¹³C,¹⁵N] labeled EmrE reported similar spectral crowding to that we observed (Agarwal et al. 2007). With the goal of reducing this congestion, we acquired a 2D FDR-NCA experiment with the RevIL EmrE sample that was designed to identify residues following the reverse labeled sites in EmrE (i.e., Leu and Ile) (Figure 1B, blue boxes). This experiment was previously implemented with model samples that did not suffer from chemical shift overlap (Traaseth and Veglia 2011). The FDR-NCA experiment selectively recouples ¹⁵N with ¹³CO in a frequency selective manner followed by a SPECIFIC-CP transfer from ¹⁵N to ¹³CA. The result is an NCA-like spectrum, which preferentially dephases one-bond ¹⁵N-¹³CO dipolar couplings without perturbing multiple bond couplings. In contrast to the NCA spectrum for RevIL EmrE, we observed 15 clearly resolved peaks using the FDR-NCA experiment (Figure 4C). When taking into account the appearance of additional shoulders, the observed number of peaks closely matches the 21 expected resonances based on the primary sequence and isotopic labeling pattern for RevIL. Identification of the residue-types were carried out using FDR-NCA in conjunction with a short DARR mixing period (FDR-NCACX; Figure 5). The identified residue types are consistent with the known residues that follow Leu and Ile in the primary sequence (Figure 1A). Using the combination of ¹⁵N reverse labeling with the afterglow and FDR spectroscopic techniques, we were able to reduce the spectral congestion leading to the identification of discrete peaks and the assignments reported in Figures 3 and 5 for EmrE.

Figure 4.

Resolution improvement for EmrE using the reverse labeling approach in conjunction with 2D FDR-NCA experiments. (A) [U-¹³C,¹⁵N] EmrE standard NCA 2D dataset using SPECIFIC-CP (same as in Figure 2). (B) RevIL EmrE standard NCA 2D dataset using SPECIFIC-CP and acquired in a similar manner as that in panel A. (C) FDR-NCA spectrum of RevIL EmrE. The spectrum in panel C shows only peaks for residues immediately following Ile and Leu in the primary sequence. Panels A, B, and C are plotted at 7.1σ, 5.8σ, and 7.0σ, respectively.

Figure 5.

FDR-NCACX spectrum with 25 msec DARR mixing to assign residue types in the 2D spectrum. This spectrum is plotted at 7.0σ.

Implications of the Method for EmrE

The improved resolution offered by the afterglow and FDR-NCA spectra allowed us to significantly reduce the spectral congestion relative to [U-¹³C,¹⁵N] EmrE samples (Agarwal et al. 2007). Based on the spectra in Figures 3 and 5, we found that some peaks showed one apparent population while others had broader spectral features indicative of heterogeneity (i.e., multiple populations) for the ligand free form of EmrE. This is relevant in light of the recent discovery that the EmrE dimer is comprised of two exchanging asymmetric monomer populations when bound to tetraphenylphosphonium (TPP⁺) in isotropic bicelles using [¹H-¹⁵N] TROSY-HSQC solution NMR (Morrison et al. 2012). Additional studies are needed to reduce the carbon and nitrogen line widths in order to resolve the monomers in the dimer that have very similar secondary structures as depicted in the cryo-electron microscopy and X-ray structural models (Chen et al. 2007; Fleishman et al. 2006). The lack of extensive inter-monomer packing other than transmembrane-4 in the structural models suggests that the ligand-free form of the protein undergoes larger conformational fluctuations leading to a less compact structure than the TPP⁺ bound form. This finding is supported by EPR experiments carried out in liposomes (Amadi et al. 2010). Furthermore, the presence of TPP⁺ binding in the solution NMR studies (Morrison et al. 2012) may induce chemical shift dispersion of the two monomers when imaged by NMR methods that detect the isotropic component of the shift (Morrison et al. 2012). Future efforts are needed to characterize the tilt angles of the dimer helices with respect to the lipid membrane for both the apo and ligand-bound forms using oriented SSNMR (Cross et al. 2011; Traaseth et al. 2009; De Angelis et al. 2006). This combination of MAS and oriented techniques promises to reveal high-resolution and complementary structural information that cannot be easily measured from isotropic chemical shifts (Cross et al. 2011; Traaseth et al. 2009; De Angelis et al. 2006).

Conclusion

The ability to unambiguously assign membrane protein spectra using MAS will significantly advance the understanding of structure and dynamics in lipid membranes. Through a combination of spectroscopic techniques with ¹⁵N reverse labeling, our approach will be a powerful tool for obtaining unambiguous assignments for polytopic membrane proteins by MAS and reducing the time needed for data acquisition. Our method is able to selectively identify resonances that preceded and followed ¹⁵N amino acids incorporated into [U-¹³C,¹⁵N] membrane protein samples. Future improvements would be the use of inexpensive precursors such as α-ketoisovalerate and phenylpyruvate at natural abundance that will reduce costs and achieve the same isotopic labeling pattern (Rasia et al. 2012), extension of the afterglow method to additional nuclei as recently reported (Bellstedt et al. 2012), and incorporation of triple cross-polarization elements (Gopinath and Veglia 2012; Akbey et al. 2011; Herbst et al. 2008) with the afterglow experiment. Finally, using the outlined methodology, it will also be possible to select two or three isotopically labeled species at the same time as in solution NMR (Masterson et al. 2008) for detecting binary or ternary complexes using SSNMR spectroscopy.