Frequency-Selective Heteronuclear Dephasing and Selective Carbonyl Labeling to Deconvolute Crowded Spectra of Membrane Proteins By Magic Angle Spinning NMR

Abstract

We present a new method that combines carbonyl-selective labeling with frequency-selective heteronuclear recoupling to resolve the spectral overlap of magic angle spinning (MAS) NMR spectra of membrane proteins in fluid lipid membranes with broad lines and high redundancy in the primary sequence. We implemented this approach in both heteronuclear ¹⁵N-¹³C^α and homonuclear ¹³C-¹³C dipolar assisted rotational resonance (DARR) correlation experiments. We demonstrate its efficacy for the membrane protein phospholamban reconstituted in fluid PC/PE/PA lipid bilayers. The main advantage of this method is to discriminate overlapped ¹³C^α resonances by strategically labeling the preceding residue. This method is highly complementary to ¹³C′_i-1-¹⁵N_i-¹³C^α_i and ¹³C^α_i-1-¹⁵N_i-1-¹³C′_i experiments to discriminate inter-residue spin systems at a minimal cost to signal-to-noise.

Introduction

Magic angle spinning (MAS) and oriented NMR experiments have been widely employed to determine the structure, topology, and dynamics of biomacromolecules [1–3]. These methods are particularly suited for integral membrane proteins, whose association with lipid vesicles limits the use of liquid-state NMR techniques. In the past years, there has been an outburst of MAS methodologies for resonance assignments and distance measurements in membrane protein samples [1, 4–13]. Unlike oriented solid-state NMR (SSNMR), which requires a high-degree of orientation with respect to the magnetic field [14–21], MAS sample preparations are more straightforward, making this technique popular in membrane protein structural biology. The strategy to obtain structural parameters for membrane protein structures by MAS is similar to that defined for liquid-state NMR, with the first step involving the tedious and lengthy process of resonance assignments followed by distance and torsion angle measurements. The NMR experiments carried out on micro-crystalline membrane protein samples give highly resolved NMR spectra [22–25], although the lack of a fluid membrane protein environment is a concern, and risks (as with X-ray crystallography) to offer conformational snapshots that may not reflect the physiological conditions of proteins in membranes.

In contrast, the membrane protein samples obtained in fluid lipid membranes [26] display significant line broadening and overlap of resonances, particularly in transmembrane (TM) segments, composed of highly redundant hydrophobic amino acids types (Leu, Ala, Gly, Val, Ile, and Phe) [27]. The helical nature of the majority of membrane proteins, the redundancy of the residue-types, and the line broadening caused by static and dynamic disorder limit the resolution in the ¹⁵N and ¹³C spectra. Only in selected cases do the linewidths enable sequence-specific assignments for non-crystalline uniformly ¹³C and ¹⁵N samples [28–31]. A number of approaches have been proposed to address these challenges: (1) main-chain assignment [32] to identify specific residues based on ¹³C/¹³C correlation spectra [33], (2) solid-phase peptide synthesis that introduces a minimal number of labeled residues at a time [34–38], and (3) selective [25, 39] or reverse biosynthetic labeling [40–43] to achieve pair-wise distributions. The second and third methods substantially reduce the number of resonances in the spectrum, making sequential assignments possible. Nevertheless, these labeling approaches used in combination with multiple bond transfers pulse sequences have been optimized for protein fibers, crystalline proteins, or lipid-depleted membrane protein samples and do not perform well for membrane proteins functionally reconstituted in fluid lipid membranes.

An elegant way to resolve complex spectra in a uniformly labeled background is to use rotational echo double resonance (REDOR) difference spectroscopy [44–46]. This experiment is analogous to the spin-echo difference spectroscopy developed for solution NMR [47], which has been applied to deconvolute spectra containing three different isotopically labeled species [48, 49]. Similar approaches have been utilized in solid-state NMR to measure short- and long-range distances between atoms. Since MAS causes I-S dipolar couplings to average to zero over a complete rotor cycle, 180° recoupling pulses on the I-spin applied during the rotor cycle reintroduce the dipolar couplings. By subtracting the spectra obtained with and without recoupling pulses, only S resonances that have a measurable dipolar coupling with I spins are observed. Significant efforts have been made to achieve frequency-selective REDOR [44] and TEDOR [50] experiments for spectral filtering, heteronuclear magnetization transfer, as well as distance/dihedral angle measurements [7, 24, 51–60]. One of these methods, the frequency-selective dipolar recoupling (FDR), has been implemented in 1D spectra showing applicability to model amino acid compounds [57, 58]. This sequence uses 90° pulses instead of 180° recoupling pulses to I spins that makes the average Hamiltonian offset dependent [57].

Here, we combined the FDR approach with carbonyl-selective labeling to obtain unambiguous assignments in ¹⁵N/¹³C^α and ¹³C^α/¹³C^X 2D spectra. This new approach complements the existing experiments for sequential assignment (¹⁵N-¹³C′, ¹⁵N-¹³C^α, ¹³C^α-¹⁵N-¹³C′, ¹³C′-¹⁵N-¹³C^α, etc.) of crowded MAS spectra of membrane proteins. When extended in the third dimension, this approach will increase the spectral resolution for membrane proteins with high primary sequence redundancy, allowing the measurement of inter-residue dipolar correlations.

Results

Figure 1 shows heteronuclear and homonuclear pulse sequences used to correlate ¹⁵N-¹³C^α (t₁-t₂) and ¹⁵N-¹³C^α-¹³C^X (t₁-t₂-t₃) with the FDR block. Following cross-polarization from ¹H to ¹⁵N, the FDR train of 90° pulses is applied after each rotor cycle (125 μsec) at the ¹³C′ frequency (177 ppm). In the middle of rotor cycles, 180° pulses are applied to the ¹⁵N channel in order to refocus chemical shift evolution, and allow for dipolar recoupling to occur between ¹³C′ and ¹⁵N [57, 58]. The pulses utilized MLEV-4 [61] phasing cycling to ¹⁵N (S spins) and XY-4 [62] to ¹³C (I spins).

Figure 1.

Pulse sequences used to obtain (A) 2D FDR-¹⁵N-¹³C^α and (B) 3D FDR-¹⁵N-¹³C^α-¹³C^X-DARR spectra. The 90° pulses on ¹³C are phase cycled with XY-4 [62] and applied at the end of each rotor cycle, while those 180° pulses applied to ¹⁵N are applied in the middle of the rotor cycle with MLEV-4 phase cycling [61]. Phase cycling is as follows: ₁ {x,-x}, ₂ {y}, ₃ {x,x,y,y,-x,-x,-y,-y}, ₄ {y,y,-x,-x,-y,-y,x,x}, ₅ {-y,-y,x,x,y,y,-x,-x}, _rec {x,-x,y,-y,-x,x,-y,y}. Phases ₂ and ₃ are adjusted by 90° to generate phase-sensitive data in t₁ and t₂, respectively.

In proteins, the covalent ¹³C′-¹⁵N peptide bond is ~1.3 Å, whereas the two-bond distance is 2.5 Å, giving maximum dipolar couplings of 1.4 and 0.2 kHz, respectively. The theoretical average Hamiltonian dephasing curves for the two distances are plotted in Figure 2 using analytical Bessel functions [63]. Note that the dephasing of FDR scales $1/\sqrt{2}$ with respect to the REDOR experiment [58]. From the theoretical curves, it is evident that ¹⁵N magnetization with one-bond ¹³C′ dipolar couplings is dephased after 1.5 – 2.0 msec (ΔS/S₀ > 0.9), whereas the magnetization for the two-bond couplings is not dephased (ΔS/S₀ < 0.1). Following the FDR block, ¹⁵N chemical shift is evolved during t₁ under ¹H TPPM decoupling [64]. The next step uses SPECIFIC-CP [65, 66] to transfer magnetization from ¹⁵N to ¹³C^α. Therefore, the FDR selectivity filter combined with the ¹⁵N-¹³C^α transfer gives a 2D heteronuclear correlation spectrum of ¹⁵N-¹³C^α filtered by the preceding ¹³C′ (Figure 1A). When 90° pulses are applied to ¹³C′, the magnetization of the ¹⁵N spins covalently attached to ¹³C′ will be selectively dephased (filtered-out), while that of the ¹⁵N spins attached to ¹²C′ will be retained. Subtracting the two experiments results in a spectrum similar to that obtained from the ¹³C′-¹⁵N-¹³C^α pulse sequence with double SPECIFIC-CP transfers [28, 67, 68].

Figure 2.

FDR dephasing ¹⁵N{¹³C} curves for one-bond (1.3 Å) and two-bond (2.5 Å) ¹³C′-¹⁵N distances. The dotted line indicates the maximum dephasing for one-bond transfer, showing less than 0.1 dephasing for the two-bond curve. The FDR dephasing data is scaled by $1/\sqrt{2}$ with respect to the REDOR experiment [44, 58].

The results from the FDR-¹⁵N-¹³C^α 2D spectra on microcrystalline NAVL are shown in Figure 3. For these experiments, NAVL was labeled [U-¹³C,¹⁵N]-Leu/Val with the acetyl group at natural ¹³C abundance. The MAS rate was set to 8 kHz and used 2 msec of dipolar dephasing. The experiment with no ¹³C 90° pulses (reference spectrum) gives two peaks from the two ¹⁵N-¹³C^α bonds in NAVL (Figure 3B). In contrast, the experiment with dephasing gives one intense resonance that is 85% of the intensity of the reference spectrum and one that was reduced to 12% intensity (Figure 3A). The difference spectrum (dephased - reference spectra) gives one intense peak that was assigned to the ¹⁵N resonance covalently bonded to ¹³C′ (i.e., the valyl peak). These experiments demonstrate the usefulness of the FDR scheme to discriminate between one- and two-bond ¹³C′-¹⁵N dipolar couplings in heteronuclear multidimensional correlation experiments.

Figure 3.

FDR-¹⁵N-¹³C^α spectra for NAVL. Spectra were acquired with (FDR, panel A) and without ¹³C 90° pulses (reference, panel B). A total dephasing time of 2 msec was used with an MAS rate of 8 kHz. (C) Subtraction of the FDR spectrum from that of the reference. Note that in the subtracted spectrum, the noise increases by a factor of $\sqrt{2}$.

Filtering one- and two-bond ¹³C′-¹⁵N dipolar couplings can also be achieved using the ¹³C′-¹⁵N-¹³C^α experiment [28, 67, 68]. This experiment utilizes two SPECIFIC-CP [65] elements to transfer magnetization from ¹³C′ to ¹⁵N and then from ¹⁵N to ¹³C^α. To compare the two approaches, we optimized all parameters in each experiment and directly tested the efficiencies with respect to the ¹H-¹³C cross-polarization and ¹⁵N-¹³C^α experiments using NAVL (Figure 4). The integrated intensities from these spectra are shown in Table 1. We found that the FDR approach gives 36–41% (for 2.0 and 1.5 msec dephasing times) of the signal with respect to the ¹H-¹³C cross-polarization experiment for the peak at 61.0 ppm (resonance not dephased), while the largest peak in the ¹³C′-¹⁵N-¹³C^α experiment (peak at 56.3 ppm) gives 22% intensity retention. Relative to the ¹⁵N-¹³C^α spectrum, FDR gives 68–76%, while the ¹³C′-¹⁵N-¹³C^α experiment gives 49% efficiency, an improvement in signal-to-noise of 38–55%. Another added benefit of our method is the relative ease of setting up the FDR experiment, which requires calibration of 90° and 180° pulses, while the ¹³C′-¹⁵N-¹³C^α experiment requires an additional SPECIFIC-CP transfer, which is sensitive to instrumental instabilities [88].

Figure 4.

Comparison of ¹H-¹³C cross-polarization, ¹⁵N-¹³C^α SPECIFIC-CP, ¹³C′-¹⁵N-¹³C^α double SPECIFIC-CP, and FDR-¹⁵N-¹³C^α (2.0 msec dephasing time) experiments. The integrated intensities from these spectra are shown in Table 1.

Table 1.

Transfer efficiencies calculated from integrated 1D NAVL spectra for the ¹⁵N-¹³C^α, ¹³C′-¹⁵N-¹³C^α, and FDR-¹⁵N-¹³C^α experiments in Figure 4 expressed as a percentage from that of the cross-polarization (¹H-¹³C) experiment. Data are normalized with respect to each peak in NAVL. The FDR experiment was conducted using total dephasing times of 1.5 and 2.0 msec. Errors for all values are less than 0.1%.

Experiment	Peak 1 (Cα = 61.0ppm)	Peak 2 (Cα = 56.3ppm)
Cross-Polarization	100%	100%
N-Cα	53.3%	44.5%
C′-N-Cα	5.6%	22.0%
FDR-N-Cα (1.5 msec)	40.7%	9.6%
FDR-N-Cα (2.0 msec)	36.3%	6.3%

We also applied the FDR-¹⁵N-¹³C^α experiment to the integral membrane protein phospholamban (PLN). The monomeric mutant of PLN (AFA-PLN, C36A/C41F/C46A) was synthesized by solid-phase FMOC chemistry with residues 30–33 labeled [U-¹³C,¹⁵N], resulting in four ¹⁵N-¹³C^α bonds and three ¹³C′-¹⁵N peptide bonds. The protein was reconstituted in 8/1/1 (w/w/w) egg phosphatidylcholine/phosphatidylethanolamine/phosphatidic acid (PC/PE/PA) lipid bilayers at a lipid/protein ratio of 60/1. This ratio is higher than the typical membrane protein samples utilized for MAS spectroscopy and is necessary to maintain the protein structural and functional integrity [69, 70]. We acquired 1D (Figure 5) and 2D spectra (Figure 6) of the reference and FDR sequence. From these spectra, it is clear that the most upfield peak in the reference spectrum is the only signal that does not change in intensity (Figures 5A and 5B, respectively). This is more apparent from the difference spectrum in Figure 5C, which shows only three peaks. The 2D version of the sequence is shown in Figure 6 with a comparison to a standard ¹⁵N/¹³C^α spectrum obtained with SPECIFIC-CP. It is straightforward to assign the missing peak to Asn³⁰. The incomplete dephasing of the signals in the FDR experiment can result from a small offset of the ¹³C transmitter from the carbonyl carbons, isotopic dilution in the sample, or protein dynamics that can scale the dipolar couplings.

Figure 5.

FDR-¹⁵N-¹³C^α spectra for AFA-PLN labeled [U-¹³C,¹⁵N] at residues 30–33. 1D spectra were acquired with (FDR, panel A) and without ¹³C 90° pulses (reference, panel B). A total dephasing time of 2 msec was used with an MAS rate of 8 kHz. (C) Subtraction of the FDR spectrum from the reference.

Figure 6.

2D FDR-¹⁵N-¹³C^α spectra for AFA-PLN labeled [U-¹³C,¹⁵N] at residues 30–33. (A) Difference spectrum for AFA-PLN^30-33 using similar parameters as that in Figure 5. (B) ¹⁵N-¹³C^α spectrum of AFA-PLN^30-33.

The 3D version of the FDR-¹⁵N-¹³C^α-DARR pulse sequence is shown in Figure 1B. Prior to t₂ evolution, this pulse sequence is identical to the 2D version reported in Figure 1A. Following SPECIFIC-CP, ¹³C^α chemical shifts are evolved during t₂, and then allowed to mix under the dipolar assisted rotational resonance (DARR) condition. This gives a 3D spectrum of ¹⁵N-¹³C^α-¹³C^X (t₁-t₂-t₃). The advantage of this method is to discriminate overlapped C^α resonances by strategically labeling the preceding residue (e.g., ¹³C′). The 2D spectra without evolution of ¹⁵N is shown in Figure 7, and demonstrates that the FDR pulse element filters out more than 80% of the signals from the reference spectrum.

Figure 7.

FDR-¹³C^α-¹³C^X-DARR spectra for NAVL. (A) Spectra were acquired with (FDR, panel i) and without ¹³C 90° pulses (reference, panel ii) using the pulse sequence in Figure 1B. A total dephasing time of 2 msec was used with an MAS rate of 8 kHz. (iii) Subtraction of the FDR spectrum from that of the reference. A 100 msec DARR mixing time was used. Panels B and C show the 1D traces from the 2D spectra in panel A.

Discussion

Sequence-specific assignments in MAS spectra of membrane proteins in fluid lipid membranes are a challenging endeavor. Several SSNMR studies on membrane proteins have reported residue-specific assignments that fail to give the atomic resolution provided for in crystalline or synthetically labeled samples. To address these problems, we propose an approach that is based on carbonyl-selective labeling combined with frequency-selective dipolar recoupling experiments and SPECIFIC-CP. These experiments will be useful to obtain specific assignments, while retaining better signal to noise (~68–76%) with respect to a 2D ¹⁵N-¹³C^α experiment using SPECIFIC-CP than the ¹³C′-¹⁵N-¹³C^α experiment (~49%).

Selective labeling strategies have become useful to reduce the complexity of spectra and the assignment process in general. A common procedure is to grow bacteria and overexpressing target proteins in unlabeled growth media enriched with one or more ¹³C and ¹⁵N amino acids [71]. Alternatively, one can use reverse labeling [40], where unlabeled amino acids are added to the isotopically enriched growth media. Both give a significant reduction in spectral complexity. More advanced strategies involve the use of labeled metabolic precursors (Reviewed in [71]). This approach has been primarily used for reducing the number of ¹³C-¹³C scalar couplings [41, 43, 72], while maintaining a significant amount of ¹⁵N and ¹³C labels in the protein. The most robust protocols have been developed for labeling ¹³C′, ¹³C^α, and ¹³C^β with the goal of performing ¹³C solution NMR relaxation studies without the interference of covalently attached ¹³C nuclei [72, 73]. In particular, ¹³C′ labels can be incorporated in several different residues without the need to reduce the labeling of the ¹³C^α (reverse labeling + labeled amino acids). The ideal use of our approach from a sensitivity perspective is to label all ¹³C^α in the protein and incorporate ¹³C′ labels in several different residue types. This can be accomplished using 2-¹³C-glucose [73] and supplementing the growth medium with [U-¹³C] labeled leucine. This application would allow i+1 residues with respect to leucine to be filtered out in the FDR-NC^α experiment, and clearly identified in the difference spectrum, which would simplify the assignment. Further advancements will be to combine carbonyl-selective labeling with combinatorial approaches used to assign overlapped spectra in solution NMR [74–79]. In particular, one can apply the FDR approach to ¹³C^α and ¹³C^β atoms to achieve frequency selective filters for assignment, or can use any of the labeling strategies discussed above to achieve pair-wise ¹³C′-¹⁵N labels to simplify the spectrum.

Conclusions

The approach presented here is a sensitive complement to the sequential ¹³C′-¹⁵N-¹³C^α experiments [28, 67, 68]. In particular, our method gives 38–55% more signal than the ¹³C′-¹⁵N-¹³C^α experiment. It allows one to select ¹²C′-¹⁵N-¹³C^α patterns while filtering out the ¹³C′-¹⁵N-¹³C^α patterns, with a sensitivity ~68–76% of the standard ¹⁵N-¹³C^α spectrum obtained with SPECIFIC-CP. In addition, our method allows for the possibility of identifying chemically identical proteins with differential labeling in a single sample. Frequency selective methods combined with novel isotope incorporation such as the method described in this paper and others [1, 41, 68, 80–83] will be useful to making sequence-specific assignments in fluid membrane protein samples and other samples with non-crystalline linewidths.

Experimental

Sample Preparation

N-acetyl-valyl-leucine was synthesized using solid-phase synthesis from FMOC-[U-¹³C,¹⁵N]-Leu and FMOC-[U-¹³C,¹⁵N]-Val (Sigma-Aldrich, Isotec). The N-terminus of valine was acetylated by acetic anhydride at natural ¹³C abundance. AFA-PLN was labeled at residues 30–33 with 99% [U-¹³C,¹⁵N]-Asn, Leu, Phe, and Ile (Sigma-Aldrich, Isotec) using solid-phase synthesis as described previously [16]. Labeling was checked by solution NMR in dodecylphosphocholine micelles [70] and mass-spectrometry. PLN (~2 mg) was solubilized in 500 μl trifluoroethanol and added to 15 mg egg PC/PE/PA lipids (8/1/1, w/w/w; Avanti Polar Lipids, Albaster, AL) in chloroform. The solvent was dried under a stream of nitrogen gas and placed in a vacuum dessicator overnight to remove residual organics. Vesicles were prepared by adding 2 ml buffer (20 mM HEPES pH 7.0, 100 mM potassium chloride, 10 mM dithiothreitol, and 3 mM sodium azide) to the lipid/protein and flash frozen 3–4 times in liquid nitrogen. Samples were then spun for 30 min at 200,000 x g in a benchtop ultracentrifuge. The pellet was transferred to a 3.2 mm thin-walled rotor with sample spacers (Agilent, Fort Collins, CO) to avoid dehydration.

Solid-State NMR Spectroscopy

NMR experiments were performed using a VNMRS spectrometer operating at a ¹H frequency of 600 MHz, with a 3.2 mm bioMAS probe [84], an MAS spinning rate of 8 kHz (υ_R), and a temperature of 20 °C. All experiments utilized cross-polarization from ¹H (61 kHz) to ¹⁵N (45 kHz) satisfying the n = 2 condition for 1.0 msec (PLN) or 2.0 msec (NAVL). The FDR sequence used 5.5 μsec 90° pulses (45.5 kHz) applied to ¹³C′ and 11 μsec 180° pulses (45.5 kHz) centered on the ¹⁵N resonances. The ¹³C pulses were phase-cycled using XY-4 [62], while those to ¹⁵N were cycled using MLEV-4 [61]. A total of 12 (1.5 msec) or 16 (2.0 msec) rotor periods were employed to achieve dipolar dephasing from ¹³C′. The ¹⁵N dimension had spectral width of 2 kHz and was evolved for 12 and 50 increments for PLN and NAVL experiments, respectively. Transfers from ¹⁵N to ¹³C^α were achieved by moving the ¹³C carrier frequency to ¹³C^α (59.2 ppm) and performing SPECIFIC-CP [65], which is related to the double cross-polarization experiment [66]. Spin-lock was applied to ¹⁵N at $\frac{5}{2}\frac{{\omega }_R}{2\pi }(20\text{kHz})$ and ¹³C^α at $\frac{3}{2}\frac{{\omega }_R}{2\pi }(12\text{kHz})$ or ¹³C′ at $\frac{7}{2}\frac{{\omega }_R}{2\pi }(28\text{kHz})$. The SPECIFIC-CP time was 5–6 msec utilizing either a linear (±10% γB₁) or adiabatic ramp ( $\frac{\mathrm{\Delta }}{2\pi }=\sim 2\text{kHz},\frac{\beta }{2\pi }=\sim 0.6\text{kHz}$) [89] to ¹³C^α or ¹³C′ for reducing the sensitivity to instrumental instability [88]. The adiabatic ramp was used only to acquire spectra in Figure 4 in order to achieve the optimal transfers for the ¹³C′-¹⁵N-¹³C^α experiment (two SPECIFIC-CP transfers). The direct ¹³C spectral width was 100 kHz and was acquired for 20 msec under TPPM [64] ¹H decoupling at 100 kHz. For the NAVL 2D experiments, 8 scans were collected with a recycle delay of 4 sec. The ¹³C-¹³C FDR experiment used an indirect ¹³C spectral width of 16 kHz, 50 increments, and a DARR [85] mixing time of 100 msec set to n = 1 rotary resonance condition. The 2D FDR-¹⁵N-¹³C^α PLN experiment used 2,048 scans with a recycle delay of 2 sec. The 1D FDR experiment used 20,000 scans with a recycle delay of 2 sec. The CH₂ resonance of adamantane was referenced to 40.48 ppm [86] and indirectly to ¹⁵N using the relative frequency ratio between ¹⁵N and ¹³C of Ξ = 0.402979946 [87].

ABBREVIATIONS

SSNMR

solid-state NMR

MAS

magic angle spinning

PLN

phospholamban

FDR

frequency-selective dipolar recoupling

DARR

Dipolar Assisted Rotational Resonance