Rapid Measurement of Long-Range Distances in Proteins by Multidimensional ¹³C-¹⁹F REDOR NMR under Fast Magic-Angle Spinning

Abstract

The ability to simultaneously measure many long-range distances is critical to efficient and accurate determination of protein structures by solid-state NMR (SSNMR). So far, the most common distance constraints for proteins are ¹³C-¹⁵N distances, which are usually measured using the rotational-echo double-resonance (REDOR) technique. However, these measurements are restricted to distances of up to ~5 Å due to the low gyromagnetic ratios of ¹⁵N and ¹³C. Here we present a robust 2D ¹³C-¹⁹F REDOR experiment to measure multiple distances to ~10 Å. The technique targets proteins that contain a small number of recombinantly or synthetically incorporated fluorines. The ¹³C-¹⁹F REDOR sequence is combined with 2D ¹³C-¹³C correlation to resolve multiple distances in highly ¹³C-labeled proteins. We show that, at the high magnetic fields that are important for obtaining well resolved ¹³C spectra, the deleterious effect of the large ¹⁹F chemical shift anisotropy (CSA) for REDOR is ameliorated by fast magic-angle spinning (MAS) and is further taken into account in numerical simulations. We demonstrate this 2D ¹³C-¹³C resolved ¹³C-¹⁹F REDOR technique on ¹³C, ¹⁵N-labeled GB1. A 5-¹⁹F-Trp tagged GB1 sample shows the extraction of distances to a single fluorine atom, while a 3-¹⁹F-Tyr labeled GB1 sample allows us to evaluate the effects of multi-spin coupling and statistical ¹⁹F labeling on distance measurement. Finally, we apply this 2D REDOR experiment to membrane-bound influenza BM2 transmembrane peptide, and show that the distance between the proton-selective histidine residue and the gating tryptophan residue differs from the distances in the solution NMR structure of detergent-bound BM2. This 2D ¹³C-¹⁹F REDOR technique should facilitate SSNMR-based protein structure determination by increasing the measurable distances to the ~10 Å range.

Introduction

Long-range inter-atomic distances are valuable restraints in NMR-based structure determination. In magic-angle-spinning (MAS) solid-state NMR, the rotational-echo double-resonance (REDOR) experiment (Gullion and Schaefer 1989; Pan et al. 1990) is one of the most robust methods for recoupling heteronuclear dipolar interactions and quantifying distances. Since its introduction nearly 30 years ago, REDOR has been applied to many spin-1/2 and quadrupolar nuclei (Cegelski 2013). In biomolecular systems, the most commonly exploited nuclear spins are ¹H, ¹⁹F, ³¹P, ¹³C, ¹⁵N, ²H, among which ¹H and ¹⁹F hold the most promise for giving long-range distances because of their high gyromagnetic ratios. Indeed, REDOR experiments between ¹⁹F and other nuclei such as ³¹P and ¹³C have been exploited (Goetz et al. 1999; Graesser et al. 2007; Kim et al. 2006; Mani et al. 2006a; Mani et al. 2006b), and REDOR between ¹H and other spins have also been demonstrated (Ghosh and Rienstra 2017; Hong and Schmidt-Rohr 2013; Sinha and Hong 2003; Wi et al. 2004). For example, ¹H-¹⁹F REDOR has been used to measure a distance of 7.7 Å in a model peptide at MAS frequencies of < 10 kHz (Wi et al. 2004), where ¹H-¹H homonuclear decoupling was used to lengthen the ¹H transverse relaxation time to observe sufficiently weak dipolar dephasing.

Recently, we showed that biosynthetic incorporation of a small number of fluorine atoms into proteins allows the measurement of multiple ¹⁹F-¹⁹F distances to ~1.6 nm (Roos et al. 2018). These sparse ¹⁹F spins also serve as probes to dephase the surrounding ¹³C and ¹⁵N nuclei through commuting dipole-dipole interactions, and hence should permit the measurement of many ¹³C-¹⁹F and ¹⁵N-¹⁹F distances. However, at the high magnetic fields that are commonly used for protein structure determination, the chemical shift anisotropy (CSA) of aromatic fluorines can be as large as ~50 kHz, which easily exceeds the ¹³C-¹⁹F dipolar coupling of interest by two orders of magnitude. The effects of this large CSA on REDOR have not been investigated. When fast MAS is used to reduce the number of ¹⁹F spinning sidebands and to obtain high-resolution ¹³C and ¹⁵N spectra, the fraction of the rotor periods occupied by the REDOR 180° pulses increases, which slows down dipolar dephasing (Jaroniec et al. 2000; Sack et al. 1999). Moreover, under fast MAS, a large number of ¹⁹F pulses are required to reach sufficiently long mixing times to measure weak dipolar couplings and long distances. The cumulative pulse imperfections are expected to reduce the accuracy of the REDOR technique. These potential difficulties need to be investigated and addressed before ¹³C-¹⁹F REDOR can be practically useful at high magnetic fields under fast MAS.

Here we demonstrate a 2D ¹³C-¹³C resolved ¹³C-¹⁹F REDOR experiment to measure multiple long-range distances in proteins. We use the 56-residue GB1 (Franks et al. 2005) and the tripeptide formyl-MLF (Hong and Griffin 1998; Rienstra et al. 2002) as our model compounds, and show that ¹³C-¹⁹F REDOR is effective at spinning rates of 25–40 kHz, for aromatic ¹⁹F CSA’s up to ~40 kHz at the magnetic field strength of 14.1 Tesla. We show that ¹³C-¹⁹F distances can be quantified with reasonable precision using a simple two-spin model, as long as the fluorine spins are well separated from each other. Finally, we apply this 2D ¹³C-¹³C resolved ¹³C-¹⁹F distance experiment to membrane-bound influenza B M2 transmembrane peptide (Williams et al. 2017; Williams et al. 2016), and show that ¹³C-¹⁹F dipolar dephasing between the proton-selective histidine and the gating tryptophan indicates significantly different sidechain structures from the micelle-bound BM2 structure.

Materials and Methods

Preparation of fluorinated peptides and proteins

Five fluorinated samples were used in this study. Two formyl-MLF peptides with different labeling schemes were used: the first sample contains 4-¹⁹F-Phe and ¹³CO-labeled Leu (Wi et al. 2004) and was used for 1D ¹³C-¹⁹F REDOR experiments, while the second sample (Biopeptek Pharmaceuticals, Malvern, PA) contains CF₃-Met, uniformly ¹³C, ¹⁵N-labeled Leu, and 4-¹⁹F-Phe, and was used for measuring ¹⁹F chemical shift anisotropies (Roos et al. 2018). Two ¹³C, ¹⁵N-labeled GB1 samples were used for measuring 2D REDOR-RFDR spectra: one sample contains 3-¹⁹F-labeled Tyr3, Tyr33, and Tyr45 (3F-Tyr GB1), while the other sample contains 5-¹⁹F labeled Trp43 (5F-Trp GB1). The fifth sample is the influenza B virus M2 (BM2) transmembrane (TM) peptide (residues 1–33), which contains uniformly ¹³C, ¹⁵N-labeled S12, A17, H19, G26, and 5-¹⁹F-Trp23 (Williams et al. 2017; Williams et al. 2016). The peptide was bound to a virus-mimetic membrane (VM+) (Cady et al. 2011; Luo et al. 2009) at pH 5.5 at a peptide/lipid molar ratio of 1 : 13.3.

¹³C, ¹⁵N and fluorinated GB1 samples were expressed in BL21 (DE3) E. coli cells. After cells were grown to OD₆₀₀ = 0.5 in M9 media containing ¹³C-labeled glucose and ¹⁵N-labeled ammonium chloride, we added glyphosate to a final concentration of 1 g/L to suppress the pentose phosphate pathway for aromatic amino acid synthesis (Lehninger et al. 1993). Immediately after glyphosate addition, we added aromatic amino acids with different combinations of fluorinated residues. For the 3F-Tyr-labeled GB1, 50 mg each of 3F-L-Tyr, unlabeled L-Phe and unlabeled L-Trp were added to the M9 media to give a final concentration of 100 mg/L. For 5F-Trp labeled GB1, 50 mg each of 5F-L-Trp, unlabeled L-Tyr and L-Phe were added. Each aromatic amino acid was first dissolved in 5 mL M9 media before being added to the culture, in order to facilitate rapid uptake of the amino acids by the cells. Following incubation for 30 minutes, when OD₆₀₀ reached 0.6, we added 0.5 mM isopropyl β-D-thiogalactoside (IPTG) to initiate protein expression. Proteins were purified by size-exclusion chromatography using a HiLoad 26/60 Superdex 75 prep grade column (GE) in pH 7.0 phosphate buffer containing 100 mM NaCl. The protein was then dialyzed to remove sodium chloride, then concentrated to 30 mg/mL using centrifugal concentrator devices (Millipore) with a 3,000 Da molecular weight cut-off under 5,000 g at 4°C. The concentrated protein was crystallized in pH 5.5 phosphate buffer using 2-methyl-2, 4-pentanediol (MPD) and isopropanol 2 : 1 v/v (Franks et al. 2005). The 3F-Tyr incorporation level was found to be ~95% by ESI and MALDI-TOF mass spectrometry, and the 5F-Trp incorporation level was verified by ESI-MS to be >95%.

For 3F-Tyr labeled GB1, each tyrosine is mono-fluorinated at the position ortho- to the hydroxyl group; however the 3-¹⁹F and 5-¹⁹F positions are indistinguishable during incorporation into the protein, thus the two positions are equally statistically occupied to 50% by a ¹⁹F. For simplicity, we still designate this sample as 3F-Tyr GB1, although either the 3- or 5- position is fluorinated.

Solid-state NMR experiments

All solid-state NMR experiments were carried out on a Bruker Avance III HD spectrometer operating at a magnetic field of 14.1 Tesla, corresponding to Larmor frequencies of 600.10 MHz for ¹H, 564.66 MHz for ¹⁹F, and 150.90 MHz for ¹³C. A 1.9 mm HFX MAS probe with a maximum MAS frequency of 42 kHz was used. Chemical shifts were externally referenced to the 38.48 ppm CH₂ resonance of adamantane on the tetramethylsilane scale for ¹³C, and to the −122.1 ppm signal of 5-¹⁹F-Trp on the CF₃Cl scale (Dürr et al. 2008). Typical ¹H radiofrequency (rf) field strengths were 71.4 kHz for excitation, 10 kHz for low-power WALTZ-16 decoupling, and 100–130 kHz for high-power TPPM decoupling (Bennett et al. 1995). RF field strengths for ¹⁹F were 100 kHz for direct-polarization (DP) experiments and 71 kHz for REDOR 180° pulses. ¹³C hard-pulse field strengths were 62 – 71 kHz, while the ¹³C Gaussian 180° pulse field strengths were 1.56 – 3.13 kHz. Recycle delays were 1.5 – 2.0 s for CP-based experiments and 7.0 – 9.0s for DP experiments. Most REDOR spectra were measured under 25 kHz MAS, but additional GB1 spectra were also measured under 40 kHz MAS. ¹H – ¹³C cross polarization (CP) used a spin-lock time of 1 ms, a 70 – 100% ramp on the ¹³C channel, with Hartmann-Hahn matching to centerband condition (62.5 or 71.4 kHz) in the middle of the ramp (85%).

1D ¹³C-¹⁹F REDOR experiments were conducted on 4F-Phe labeled formyl-MLF to investigate the effectiveness of ¹³C-¹⁹F REDOR in the presence of large ¹⁹F CSAs. A ¹³C 180° pulse of 7 μs was applied to refocus the ¹³C chemical shift during the REDOR mixing period (Fig. 1A). The ¹⁹F carrier frequency was placed on resonance with the −115.9 ppm 4F-Phe peak to maximally invert the 4-¹⁹F-Phe signal. The ¹⁹F rf field strength was 62.5 kHz. ¹⁹F 180° pulses used XY-4 phase cycling while the ¹³C 180° pulse used the EXORCYCLE scheme to compensate for pulse flip-angle errors (Sinha et al. 2004). At 25 kHz MAS, ¹H TPPM decoupling was 120 kHz during the REDOR period and 100–120 kHz during acquisition to maximize ¹³C T₂ and to obtain high-resolution ¹³C spectra.

Figure 1.

Pulse sequences used in this study. (A) 1D broadband ¹³C-¹⁹F REDOR, conducted at 25 kHz MAS with high power ¹H TPPM decoupling during REDOR and ¹³C detection. Filled and open rectangles represent 90° and 180° pulses, respectively. (B) 2D REDOR-RFDR experiment. The ¹³C selective Gaussian 180° pulse in the middle of the REDOR period uses EXORCYCLE phase cycling. High-power ¹H TPPM decoupling was used for 25 kHz MAS. In comparison, the 40-kHz MAS data for GB1 was measured under low-power (10 kHz) WALTZ decoupling.

2D ¹³C-¹³C REDOR-RFDR experiments were implemented using the pulse sequence in Fig. 1B. A frequency-selective REDOR period was inserted before ¹³C t₁ evolution. The ¹³C Gaussian 180° pulse length was 320 μs for the GB1 spectra, and 240 μs and 160 μs for the BM2 Cα- and CO-selective spectra, respectively. The MAS rates of 25 and 40 kHz were sufficiently high that ¹³C coherences outside the frequency-selected region have long T₂ relaxation times. These undesired coherences were removed by phase-cycling the ¹³C Gaussian 180° pulse following the EXORCYCLE scheme. The ¹³C RFDR (Bennett et al. 1998) mixing period was 2 ms to detect one-bond cross peaks, and the ¹³C signals were detected after a Hahn echo to obtain undistorted baselines. Two 2D experiments, without (S₀) and with (S) ¹⁹F recoupling pulses, were conducted to measure the T₂-corrected REDOR dipolar dephasing, S/S₀. ¹⁹F recoupling field strengths were 71 kHz in the REDOR-RFDR experiments. For 5F-Trp GB1, REDOR mixing times of 6.4, 9.6, and 10.8 ms were measured, while for 3F-Tyr GB1, REDOR mixing times were 6.4 and 8.4 ms. For the influenza BM2 sample, REDOR mixing times were 4.0 and 6.4 ms. ¹H decoupling field strengths of 100–130 kHz were used for experiments at 25 kHz MAS. 2D REDOR-RFDR spectra of GB1 were acquired with 64 – 128 scans per t₁ point and 512 t₁ points. BM2 spectra were acquired with 256 scans and 144 t₁ increments.

REDOR simulations

REDOR simulations were performed using the SIMPSON software (Bak et al. 2000). Powder averaging used the REPULSION320 scheme with 64 γ angles (Bak and Nielsen 1997). Simulations to explore the effects of CSA on REDOR dephasing employed a two-spin system with a 6 Å distance, corresponding to a dipolar coupling constant of 132 Hz. The ¹⁹F CSA used in the REDOR simulations were measured from the 1D ¹⁹F spinning sideband spectra using TopSpin’s Solids Lineshape Analysis module (Fig. S1) (Roos et al. 2018). All REDOR simulations used finite ¹⁹F and ¹³C 180° pulses that match the experimental pulse lengths.

For 4F-Phe formyl-MLF, ¹⁹F pulse angle imperfection was taken into account by adding a scaling factor to the ¹⁹F rf field strength to give effective flip angles of 180° to 145° in 5° increments. The simulations for different ¹⁹F pulse angles were added using weights that match a normal distribution centered at 180° with a standard deviation of 15°. The weighted average S/S₀ curves were calculated for ¹³C-¹⁹F distances of 4.0 to 9.0 Å in 0.5 Å increments. To average over a distribution of distances, flip-angle distributed simulations for different distances were added using coefficients from a normal distribution centered at 6.5 Å with a standard deviation of 1.5 Å. The data for 5F-Trp were analyzed using the same flip-angle distribution, but without distance distribution. Data for 3F-Tyr GB1 and BM2 samples did not consider a distribution of flip angles, since complicating factors of multiple ¹⁹F dephasers and lower spectral sensitivity obscure these effects.

The random uncertainties ${\sigma }_{S/S_0}$ in the REDOR S/S₀ values at each mixing time were propagated from the signal-to-noise ratios (SNR) of the S₀ and S spectra using the equation:

$${\sigma }_{S/S_0}=\frac{S}{S_0}{\left(\frac{1}{{\left(SN{R}_S\right)}^2}+\frac{1}{{\left(SN{R}_{S_0}\right)}^2}\right)}^{1/2}$$ (1)

Root-mean-square deviations (RMSDs) between the measured and simulated REDOR S/S₀ values were calculated in MATLAB. Any distances whose simulated REDOR curves agree with the experimental data with an RMSD below the average experimental uncertainty are considered to be consistent with the data. In cases where the S/S₀ values at the longest mixing time are relatively high and do not distinguish multiple simulated REDOR decay curves, the upper bound of the measured distance cannot be determined.

Results

¹⁹F chemical shift anisotropies in proteins and their influence on ¹³C-¹⁹F REDOR dephasing

The main difference between REDOR of low-frequency spin-1/2 nuclei and ¹³C-¹⁹F REDOR is the large isotropic and anisotropic chemical shifts of the ¹⁹F spins. In combination with the large gyromagnetic ratio of ¹⁹F, this results in a very wide ¹⁹F spectral width, which is difficult to invert fully unless strong rf fields are applied. As an example, Fig. 2 shows the ¹⁹F MAS spectra of GB1 and formyl-MLF, which include both aromatic fluorines and a trifluoromethyl (CF₃) group. Under 25 kHz MAS, significant sideband intensities are detected for 3F-Tyr residues in GB1, and only at 40 kHz MAS are most sidebands suppressed, as shown for 4-¹⁹F-Phe in formyl-MLF. These spectra show a large isotropic shift dispersion of ~100 ppm between CF₃ and CF groups: in formyl MLF the Met CF₃ peak resonates at −38.9 ppm, with a full-width at half maximum (FWHM) of 0.53 ppm, while the 4F-Phe peak resonates at −115.9 ppm with a FWHM of 1.4 ppm. This isotropic chemical shift difference corresponds to 43.5 kHz at 14.1 Tesla. The 3F-Tyr peaks in GB1 resonate even more upfield than 4F-Phe, at about −135 ppm. Therefore ¹³C-¹⁹F REDOR experiments must either use strong ¹⁹F radiofrequency pulses for broadband inversion or employ frequency-selective approaches. ¹⁹F spinning sideband spectra measured under slow MAS rates indicate a chemical shift anisotropy (CSA) parameter (δ_zz-δ_iso) of −59 ppm for 4F-Phe, while Met CF₃ has a motionally averaged CSA of +17 ppm (Fig. S1). For the three 3F-Tyr residues in GB1, the isotropic shift range is small, about 3 ppm (Fig. 2A), and the linewidths were 0.3 ppm for this microcrystalline protein. Y3 and Y45, which are located in two β-strands, show CSAs of −75 ppm while the α-helical Y33 has a motionally averaged CSA of −58 ppm.

Figure 2.

¹⁹F DP MAS spectra of GB1 and formyl-MLF. (A) 3F-Tyr GB1 spectrum measured under 25 kHz MAS. The α-helical Y33 peak is well resolved from the β-sheet Y3 and Y45 signals. The Y33 sidechain is mobile, as manifested by weaker sideband intensities compared to the Y3 and Y45. The solid-state NMR structure of GB1 (PDB: 2LGI) is shown, with 3F and 5F positions (orange and magenta) both highlighted. (B) Formyl-MLF spectrum measured under 40 kHz MAS. The 4-¹⁹F-Phe sideband intensities are largely suppressed at this MAS rate. The solid-state NMR structure of formyl-MLF is shown (PDB: 1Q7O).

To investigate how the large ¹⁹F CSAs affect REDOR dephasing, and to determine the optimal MAS rate and rf field strength required to obtain undistorted REDOR curves, we simulated ¹³C-¹⁹F REDOR curves for various combinations of ¹⁹F CSA, MAS rate, and ¹⁹F rf field strengths (Fig. 3). When the MAS rate is smaller than the ¹⁹F CSA, the REDOR S/S₀ values shift up compared to the ideal values, and this deleterious effect is amplified by weak ¹⁹F rf fields. These observations indicate incomplete recoupling for internuclear vectors that are oriented roughly parallel to the magnetic field. Increasing the MAS frequency to 50 kHz (Fig. 3B) resulted in lower S/S₀ values and hence more ideal dipolar dephasing, with minimum S/S₀ values close to 0. Under this condition, weaker ¹⁹F rf pulses only slow down the dipolar dephasing moderately, due to the known finite pulse-length effect (Jaroniec et al. 2000). Fig. 3C–F further illustrate why fast MAS is important for overcoming the influence of the large ¹⁹F CSA on REDOR dipolar dephasing. Under slow MAS, the REDOR dephasing curve does not have a monotonic dependence on the CSA, but deviates significantly from the ideal curve when the CSA exceeds the MAS rate by 2-fold. Increasing the MAS rate to above the CSA removes most of this effect and restores nearly ideal REDOR dephasing, other than the finite pulse-length effect.

Figure 3.

Simulated ¹³C-¹⁹F REDOR curves for varying ¹⁹F rf field strengths, ¹⁹F CSA and MAS rates. All simulations are for a 6.0 Å distance with a dipolar coupling constant of 132 Hz. (A) When the MAS frequency is smaller than the ¹⁹F CSA, weaker ¹⁹F rf field strength results in less dipolar couplings and higher REDOR S/S₀ intensities. (B) When the MAS frequency exceeds the ¹⁹F CSA, weaker ¹⁹F rf fields shift the REDOR minimum to longer times but do not raise the S/S₀ values. (C) Under slow MAS, large ¹⁹F CSAs significantly increase the S/S₀ values and shift the time of the minimum S/S₀ value to longer times, even when strong ¹⁹F rf fields are applied. (D) When the MAS frequency exceeds the ¹⁹F CSA and when high rf fields are applied, the REDOR S/S₀ values are relatively invariant to the CSA. (E) For a fixed ¹⁹F rf field of 70 kHz and a small ¹⁹F CSA of 10 kHz, MAS frequencies from 12.5 to 50 kHz cause little difference in REDOR dephasing. (F) For a fixed ¹⁹F rf field strength of 70 kHz and a large ¹⁹F CSA of 30 kHz, a slow MAS rate of 12.5 kHz causes incomplete REDOR dephasing, while 25 kHz and 50 kHz MAS restores REDOR dephasing to the expected minimum, but with different rates. (G, H) Simulated ¹³C-¹⁹F REDOR curves for distances from 4 Å to 10 Å under 25 kHz MAS (G) and 40 kHz MAS (H). A ¹⁹F CSA of 30 kHz and a ¹⁹F rf field strength of 70 kHz were used in the simulations.

Taken together, these simulations show that fast MAS is important for producing accurate REDOR dephasing curves at high magnetic fields. An aromatic ¹⁹F CSA of −75 ppm, which is found for Y3 and Y45 in GB1, corresponds to a frequency of −42 kHz on the 600 MHz NMR spectrometer. Therefore, MAS frequencies above ~35 kHz are advisable. In practice, this consideration is moderated by the fact that fast MAS requires more ¹⁹F pulses to reach the long mixing times necessary for measuring weak dipolar couplings, and the cumulative pulse imperfection will reduce the performance of the REDOR experiments.

¹³C-¹⁹F REDOR experiment of formyl-MLF

To investigate the effectiveness of ¹³C-¹⁹F REDOR under fast MAS, we measured 1D ¹³C-¹⁹F REDOR spectra of ¹³CO-Leu, 4-¹⁹F-Phe labeled formyl-MLF (Hong and Griffin 1998; Rienstra et al. 2002) under 25 kHz MAS. With the ¹⁹F carrier frequency on resonance with the 4F-Phe peak, the Leu ¹³CO signal shows dipolar dephasing to ~0.4 by 11.2 ms (Fig. 4A, B). Based on the structure of formyl-MLF (PDB code: 1Q7O), Leu ¹³CO is 6.5 Å from 4-¹⁹F of Phe, which corresponds to a coupling constant of 104 Hz. The ideal REDOR curve for this coupling, in the absence of large ¹⁹F CSA, should exhibit a minimal S/S₀ at ~17 ms. The measured data show much higher S/S₀ values, indicating the effects of the ¹⁹F CSA and pulse imperfections. We took into account the ¹⁹F CSA and ¹⁹F pulse-length imperfections when simulating the REDOR curves. To account for rf field inhomogeneity and pulse flip-angle errors, we assumed a symmetric Gaussian distribution of ¹⁹F flip angles centered at 180° with a standard deviation of 15° (Sinha et al. 2004). However, even with this flip-angle distribution, single-distance simulations still did not reproduce the experimental data well. Since the formyl-MLF structure (PDB: 1Q7O) shows moderate disorder in the Phe ring position due to a distribution of χ₁ torsion angles, we next adopted a Gaussian distribution of ¹³CO-¹⁹F distances centered at 6.5 Å with a standard deviation of 1.5 Å. With this distance distribution, the simulated REDOR curve agrees well with the measured dephasing, as shown in Fig. 4B.

Figure 4.

1D ¹³C-¹⁹F REDOR data of 4F-Phe labeled formyl-MLF under 25 kHz MAS. (A) Representative REDOR S₀ and S spectra, measured at mixing times of 1.6 ms (left) and 11.2 ms (right). (B) Measured S/S₀ decay (red circles) of Leu ¹³CO as a function of mixing time. The PDB structure 1Q7O shows the average distance of 6.5 Å between 4-¹⁹F Phe and Leu CO. The simulated REDOR curve (black) uses a Gaussian distance distribution as shown in the inset, and takes into account the ¹⁹F CSA, ¹⁹F finite-pulse lengths, and a pulse flip-angle distribution of 15° around 180°.

2D ¹³C-¹³C resolved ¹³C-¹⁹F REDOR spectra of GB1

3F-Tyr GB1 represents a good model system for multiply fluorinated proteins in which distances to the ¹⁹F spins need to be resolved in 2D or 3D correlation spectra. We measured 2D ¹³C-¹³C REDOR-RFDR spectra using the pulse sequence of Fig. 1B to demonstrate this approach. The triply labeled 3F-Tyr GB1 exhibits three levels of complexity, which are expected to be common in fluorinated proteins. First, the ¹⁹F isotropic chemical shift range is relatively small, only 3 ppm. Second, both 3F and 5F are statistically equally incorporated, thus distances to both Hε1 and Hε2 need to be considered in the distance analysis. Third, if a Tyr sidechain undergoes fast reorientation, the REDOR dipolar dephasing will be partially averaged. Below we show that the three Tyr residues in GB1 exhibit both static and dynamic averaging, but we can still obtain distances with reasonable precision due to the sparseness of the ¹⁹F incorporation, the long distances of the fluorines to many ¹³C spins, and direct knowledge of sidechain motion from 1D ¹⁹F spectra.

We first investigated whether fluorination of three Tyr residues perturbs the protein structure. A 2D ¹³C-¹³C RFDR spectrum with 4.8 ms mixing gave nearly identical ¹³C chemical shifts (Fig. S2, S3) as non-fluorinated GB1 (Franks et al. 2005; Wylie et al. 2011), with a root-mean-square deviation of only 0.53 ppm (Table S1), indicating that biosynthetic incorporation of three fluorines does not perturb the global structure. Fig. 5 shows the 2D ¹³C-¹⁹F REDOR-RFDR spectra of GB1, measured under 25 kHz MAS with a REDOR mixing time of 8.4 ms and an RFDR mixing time of 2.08 ms. The shorter RFDR mixing time was used to obtain only one-bond correlations to reduce spectral congestion. The ¹³C carrier frequency was placed at 56.2 ppm, thus the soft Gaussian pulse selectively refocuses the Cα chemical shifts in the indirect dimension. The 2D S₀ control spectrum has a different intensity distribution from the 2D RFDR spectrum with REDOR due to the ¹³C selective inversion, while T₂-induced signal loss was negligible at the REDOR mixing time of 8.4 ms. After subtracting the dephased (S) spectrum (data not shown) from the S₀ spectrum, we obtained a considerably simplified 2D difference spectrum (ΔS), which exhibits only signals of carbons that are in close proximity to the fluorinated Tyr residues. Fig. 5B shows two representative ω₁ cross sections from the S₀ and ΔS spectra. In the 52.5-ppm cross section, the A24 Cα-CO cross peak and the M1 Cα-CO cross peak exhibit S/S₀ values of 0.73 ± 0.11 and 0.35 ± 0.08, respectively, indicating sizeable dipolar coupling of A24 and M1 Cα to the nearest fluorine, which is 3F-Y3. When multiple cross peaks for the same Cα site are observed, such as the A34 Cα-CO and Cα-Cβ cross peaks, which both represent dipolar dephasing to the same carbon of A34 Cα, we used the average S/S₀ value to obtain more accurate dipolar dephasing values. For example, the A34 Cα-CO and Cα-Cβ cross peaks show an average S/S₀ value of 0.75 ± 0.03 at 8.4 ms mixing. Similar 2D spectra measured with a shorter REDOR mixing time of 6.4 ms under both 25 kHz and 40 kHz MAS are shown in Fig. S4.

Figure 5.

2D REDOR–RFDR spectra of 3F-Tyr-labeled GB1 measured at 25 kHz MAS. The REDOR mixing time was 8.4 ms and the RFDR mixing time of 2.08 ms. (A) Control spectrum (S₀) showing all signals and difference spectrum (ΔS) showing only signals of ¹³C spins that are close to a fluorinated Tyr. (B) Representative ω₁ cross sections from the S₀ and ΔS spectra. The difference spectrum shows fewer peaks, indicating that the experiment can serve as a spectral editing method.

To convert the cross peak intensities in the difference spectrum to quantitative ¹³C-¹⁹F distances, we plotted the S/S₀ values at 6.4 and 8.4 ms mixing times for representative cross peaks (Fig. 6). The Tyr residue that caused the dephasing of a specific Cα atom is assigned based on the known GB1 structure. Clearly different S/S₀ values are observed for different residues, indicating that the experiment is sensitive to different ¹³C-¹⁹F distances. When fitting the measured REDOR decays, we took into account the ¹⁹F CSAs and the finite ¹⁹F pulse length. We chose not to include pulse imperfections in the simulation for this 3F-Tyr labeled GB1, since the multiple fluorines in this protein cause stronger dipolar dephasing to each ¹³C, thus offsetting the flip-angle imperfection. In other words, we expect these two systematic uncertainties to partially cancel each other. We simulated the measured decays assuming a two-spin system, thus the extracted distances are apparent values for the statistically mixed fluorination positions 3F (Tyr Hε1) and 5F-Tyr (Hε2). Based on the minimum RMSD between the experimental and simulated REDOR intensities, we obtained best-fit distances of 5.3 Å to 7.7 Å for the cases shown in Fig. 6.

Figure 6.

Representative Cα REDOR S/S₀ decays of 3F-Tyr-labeled GB1 from 2D REDOR-RFDR spectra. (A) Measured S/S₀ decays for indicated ¹³C-¹⁹F spin pairs. Two-spin simulations that include ¹⁹F CSAs and finite pulse-length effects are overlaid with the data, with the best-fit simulations indicated in black. (B) RMSD between the measured and simulated REDOR dephasing. Horizontal blue lines correspond to the average uncertainty of the S/S₀ values propagated from the experimental spectra, below which are distances that agree with the measured REDOR intensities.

Since the ¹³C-¹⁹F REDOR experiment on the triply 3F-Tyr labeled GB1 is complicated by multi-spin effects and the statistical incorporation of 3F and 5F, we prepared a second fluorinated GB1 containing a single dephasing spin at 5F-Trp43. 2D REDOR-RFDR spectra were measured under 25 kHz MAS (Fig. S5), similar to the conditions for 3F-Tyr GB1. Fig. 7 shows representative ¹³C-¹⁹F dipolar dephasing in this sample. REDOR simulations for this sample include not only ¹⁹F CSA and finite-pulse length effects but also ¹⁹F flip-angle distributions of 15°, as used for formyl-MLF, giving best-fit distances indicated in Fig. 7.

Figure 7.

Representative Cα-F S/S₀ decays of 5F-Trp43-labeled GB1, extracted from 2D REDOR-RFDR spectra measured under 25 kHz MAS. Single-distance REDOR simulations incorporated the ¹⁹F CSA, finite pulses, and a flip-angle distribution of ±15°.

Fig. 8 summarizes the results of these REDOR experiments on the two fluorinated GB1 samples. For 3F-Tyr GB1, we color-code residues in the protein structure according to the extent of ¹³C-¹⁹F dipolar dephasing measured at the mixing time of 8.4 ms (Fig. 8A). It can be seen that a number of residues near Y3 and Y45 are strongly coupled to the fluorines, with S/S₀ values below 0.55 (red) and between 0.55 and 0.80 (orange). For example, A20, V21, M1 and Q2 are strongly dephased by Y3, while D46 and T51 are significantly dephased by Y45 (Fig. 8B). In the 3F-Tyr-GB1 sample, ten residues belong to the category of having significant difference intensities and hence have short distances to fluorines. Another 12 residues exhibit S/S₀ values greater than 0.80 (grey), indicating either longer distances or proximity to a dynamic fluorine. Selected cross sections for these residues are shown in Fig. S6. We found that residues near Y33, which has a motionally averaged ¹⁹F CSA (Fig. S1), show weaker REDOR dephasing, suggesting motional averaging of the ¹³C-¹⁹F dipolar coupling. When motion can be excluded based on the ¹⁹F CSA, then even residues with limited dipolar dephasing provide important structural information, by giving lower-bound distances to the ¹⁹F atoms. Another 28 residues of GB1 do not have fully resolved signals in the control 2D S₀ spectrum and the difference ΔS spectrum, thus do not allow distance quantification (cyan). Although GB1 is a microcrystalline protein, at the long REDOR mixing time used, ¹³C T₂ relaxation reduces spectral sensitivity, thus the spectra were processed using window functions that favor higher sensitivity, resulting in lower resolution than a simple 2D RFDR spectrum. Finally, the six aromatic residues have ¹³C at natural abundance and thus are excluded from the distance analysis. Fig. 8C highlights the positions of residues that are strongly dephased by 5F-Trp43. In this singly fluorinated sample, it can be clearly seen that many residues within 6–9 Å of 5F-Trp43 show significant dipolar dephasing, thus verifying the effectiveness of the 2D REDOR-RFDR technique for measuring distance constraints in the hydrophobic core of proteins.

Figure 8.

¹³C-¹⁹F distance measurement and quantification in GB1. (A) Solid-state NMR structure of GB1 (PDB code: 2LGI), showing the positions of the three Tyr residues. Residues whose carbons show significant REDOR dephasing at 8.4 ms are color-coded. Cyan residues have S₀ signals that are not sufficiently resolved. (B) Zoomed-in regions around Y3 and Y45, showing several residues with significant ¹³C-¹⁹F dipolar dephasing. For A20 Cα, V21 Cα, and T51 Cα, distances to the 3F (magenta) and 5F (orange) positions of Y3 and Y45 differ significantly, thus the measured REDOR dephasing is an average of two distances. In comparison, D46 Cα has similar distance of 4.8 and 5.3 Å to Y45 3F and 5F. (C) GB1 structure showing the position of 5F-Trp43. Residues that exhibit significant ¹³C-¹⁹F dephasing are indicated. (D) Comparison of measured and predicted ¹³C-¹⁹F distances. The y-axis shows the shortest distance of a carbon to one of the three 3F-Tyr residues and to 5F-Trp43. The measured distances that are assigned to Y33 and W43 were corrected for sidechain dynamics using the measured CSA order parameters. Residues are color-coded according to the fluorinated aromatic residue that dephases the carbon. Magenta symbols indicate two residues (V29 and A24) with similar distances to Y3 and Y33. The measured and predicted distances correlate with a slope of 1.12. (E) Measured Tyr-dephased ¹³C-¹⁹F distances with associated random uncertainties due to spectral sensitivity. Large distance upper-bounds result from limited sensitivity of some of the cross peaks at long REDOR mixing times, which do not permit exclusion of long distances. This random uncertainty is expected to be significantly offset by systematic uncertainties. A slope-1 line guides the eye for the ideal correlation between measured and predicted distances.

Fig. 8D plots all measured ¹³C-¹⁹F distances against the distances read off from the GB1 structure (PDB code: 2LGI). For the Tyr-labeled sample, we use the average distance to 3F and 5F to represent the PDB distances. The further away a carbon is from the fluorinated Tyr, the better the approximation, while for carbons close to a Tyr, the two distances can differ significantly if the aromatic ring is asymmetrically positioned relative to the carbon (Fig. 8B). Y33 has a motionally averaged ¹⁹F CSA (Fig. S1), suggesting that this sidechain may undergo rapid two-site jumps. For ¹³C-labeled residues whose dipolar dephasing is attributed to Y33, we used the CSA reduction factor of 0.76 as the approximate order parameter for the ¹³C-¹⁹F dipolar coupling, and correspondingly shortened the apparent measured distances by a factor of 0.76 ^1/3 = 0.91. For 5F-Trp43 labeled GB1, the ¹⁹F sideband spectrum (Fig. S1) indicates a CSA of +46.6 ppm, which is scaled by a factor of 0.87 from the rigid-limit CSA of +53.7 ppm for amino acid 5F-Trp (Roos et al. 2018). We thus decreased the measured apparent distances by 0.87^1/3 = 0.95 to obtain truer distances. With these corrections, we obtained a slope of 1.12 for the correlation and a root-mean-square deviation of 0.88 Å between the measured and predicted distances, indicating overall good agreement with the GB1 structure (Table S2).

Fig. 8E shows the random uncertainties of the measured distances at 95% confidence intervals (2σ) for the 3F-Tyr GB1 sample, propagated from the sensitivities of the S₀ and S spectra. Large upper-bound distance error bars are seen for a number of residues, reflecting the fact that the measured S/S₀ values for various cross peaks at long REDOR mixing times are sufficiently high or have sufficiently large uncertainties that they do not exclude longer distances. An example is shown for T49 Cβ (Fig. 6), whose S/S₀ values correspond to a best-fit distance of 7.7 Å, in excellent agreement with the expected distance in the GB1 structure, but the sensitivity is insufficient to cap the distance upper limit. In reality, since systematic uncertainties such as flip-angle imperfections, finite pulse-length effects and incomplete ¹H decoupling all increase the measured S/S₀ values, the true S/S₀ value is unlikely to be higher than the measured value, therefore the distance upper limits, after accounting for both random and systematic error sources, should be less than those shown in Fig. 8E.

Although REDOR imperfections tend to increase the apparent observed distances, it is interesting to note that many measured distances are shorter than the predicted distances based on the PDB structure. One source of this discrepancy is the simplifying assumption of a two-spin model in the REDOR simulations, since in reality a particular carbon can be simultaneously dephased by two or more ¹⁹F spins. To investigate the extent to which the two-spin model can bias the distance quantification, we simulated three-spin REDOR curves for a ¹³C coupled to one fluorine spin that is 6 Å away and a second fluorine spin that is 7 – 11 Å away, with an arbitrary angle of 90° between the two C-F vectors. Fig. 9 shows that when the second C-F distance is 10–11 Å, then the three-spin REDOR curve is indistinguishable from the two-spin REDOR curve for 6 Å. This is expected since the second dipolar coupling is so much weaker than the first coupling that it causes little change to the dephasing curve of the strong coupling. When the second distance is 7 or 8 Å, then the three-spin REDOR curve can be well fit by two-spin REDOR curves for a distance of 5.6 Å or 5.8 Å for mixing times up to 12 ms. Therefore, the systematic error introduced by the two-spin approximation is ±0.4 Å for mixing times shorter than 12 ms. For mixing times longer than 12 ms, the two-spin REDOR curves deviate significantly from the three-spin curves. This suggests that the REDOR dephasing at long mixing times can in principle be used to determine whether a carbon is similarly distanced from two fluorines or is much closer a single fluorine than all other fluorines (Tang et al. 2007).

Figure 9.

Simulated ¹³C-¹⁹F REDOR curves to investigate the applicability of two-spin simulations to approximate three-spin systems containing one carbon and two fluorines (F1 and F2). The three-spin simulations used a C-F1 distance of 6 Å and a variable C-F2 distance from 7 Å to 11 Å, with an arbitrary angle of 90° between the two vectors. The two-spin REDOR curve approximates the three-spin REDOR curve for mixing times up to ~12 ms when the C-F2 distances are longer than 7 Å. These simulations used a ¹⁹F CSA of −75 ppm, a ¹⁹F rf field strength of 71 kHz, and an MAS rate of 25 kHz.

Application of the 2D REDOR-RFDR experiment to the influenza BM2 TM peptide

To investigate the applicability of this REDOR-RFDR experiment to biologically relevant samples, we measured the 2D REDOR-RFDR spectra of the influenza BM2 TM peptide (Fig. 10). The peptide contains 5-¹⁹F-W23 and several ¹³C, ¹⁵N-labeled residues, and was reconstituted into a virus-mimetic lipid bilayer and measured using REDOR mixing times of 4.0 and 6.4 ms. Both Cα-selective (75–35 ppm) and CO-selective (185–165 ppm) 2D spectra were measured (Fig. 10A). Among the ¹³C, ¹⁵N-labeled residues, only H19 Cα shows difference intensities (red) at Cα-CO and Cα-Cβ cross peaks, while S12 and G26 difference signals are absent, indicating that these two residues are far from 5-¹⁹F-W23. The ω₁ cross section at 174.7 ppm (Fig. 10B) illustrates the benefit of the 2D experiment for spectral editing: while the ω₁ chemical shift includes contributions from the CO sites of three residues (S12, H19, and G26), the difference spectrum shows only the H19 CO-Cα cross peak. Based on the measured S/S₀ values at 4.0 ms and 6.4 ms, we obtained a best-fit H19 CO – W23 5F distance of 6.2 Å (Fig. 10C). A weak difference cross peak is also observed for A17 Cα-Cβ, which fit to a distance of 7.5 Å, but the uncertainty indicates that this distance should be interpreted as a lower bound.

Figure 10.

2D resolved ¹³C-¹⁹F REDOR data of membrane-bound BM2 TM peptide containing 5F-Trp23. The spectra were measured under 25 kHz MAS with a REDOR mixing time of 6.4 ms and a RFDR mixing time of 2.0 ms. (A) 2D S₀ spectrum (black) overlaid with the ΔS spectrum (red). Only H19 shows significant difference intensities, indicating that only H19 is near 5-¹⁹F-W23. (B) Selected cross sections of the 6.4 ms REDOR-RFDR 2D spectra. H19 CO and Cα both show significant dipolar dephasing while S12 CO and A17 CO have negligible dephasing. (C) Measured S/S₀ values of H19 and A17 CO from the CO-selective 2D spectra, overlaid with simulated REDOR curves. The RMSD between the measured and simulated S/S₀ values are shown below the data. Best-fit simulations give a distance of 6.2 Å between H19 CO and W23 5-¹⁹F. The shallow minimum in the A17 CO RMSD curve suggests a distance of ~7.5 Å from 5-¹⁹F-W23. (D) Solution NMR structure of DHPC-bound BM2 (PDB code: 2KIX), showing average intra-helical distances of H19 CO, G26 CO, and A17 CO to 5-¹⁹F-W23. The H19-W23 distances are too long compared to the data, while the G26-W23 distances are too short compared to the data.

These REDOR dephasing values suggest different sidechain structures and packing of BM2-TM in lipid bilayers compared to DHPC detergent micelles. First, the solution NMR structure of DHPC-micelle bound BM2 (PDB code: 2KIX) indicates an average intra-helical G26 ¹³CO – W23 5-¹⁹F distance of 6.8 Å (Wang et al. 2009) and an average nearest-neighbor inter-helical distance of 8.0 Å. These distances are expected to cause significant dipolar dephasing of G26 by 5F-W23, thus the absence of a difference cross peak for G26 CO strongly suggest that the W23 sidechain structure in bilayer-bound BM2 differs from its structure in the detergent-bound peptide. Second, the measured H19 CO distance of 6.2 Å to 5-¹⁹F-W23 is significantly shorter than the average intra-helical distance of 7.9 Å and the average inter-helical distance of 8.3 Å in the solution NMR structure. This observation further supports the notion that the W23 rotameric structure and the helix-helix interface structure in membrane-bound BM2-TM may differ from those of detergent-bound BM2.

Our previous SSNMR experiments showed that membrane-bound BM2-TM proton channels have different pK_a’s for the proton-selective H19 from the pK_a’s of the proton-selective H37 in influenza A M2 (Williams et al. 2016). This difference was proposed to result from the presence of a second histidine, H27, in BM2, which is located at the membrane-water interface (Williams et al. 2017). With two titratable histidines in the BM2 TM domain, the sidechain structure of the proton-selective H19 may differ from that of H37 in AM2. The ¹³C-¹⁹F distance data shown here will help to elucidate the structural differences between AM2 and BM2 proton channels, to understand how the second histidine impacts proton conduction, and how the membrane environment affects the oligomeric structure of this family of canonical proton channels.

Conclusion

The present data show that multiple ¹³C-¹⁹F distances in the 5–10 Å range can be measured rapidly in fluorinated proteins and peptides under fast MAS in high magnetic fields. The 2D REDOR-RFDR ¹³C-¹⁹F experiment provides site-resolved distances between ¹³C-labeled residues and a few fluorinated residues in a protein. Our data indicate that the 2D S₀ and S spectra can be measured reliably to give difference spectra, whose cross peak intensities reflect distances after taking into account ¹⁹F CSA and the finite pulse-length effect under fast MAS. The precision of the extracted distances chiefly depends on whether multiple fluorines are close together in space to cause multi-spin dipolar dephasing. In GB1 the three 3F-Tyr residues are separated by 6.7–16.3 Å from each other (calculated as the average of all 3F/5F combinations for Tyr pairs). For structurally unknown proteins, these inter-fluorine distances can be independently measured using 2D ¹⁹F-¹⁹F spin diffusion or dipolar recoupling experiments (Roos et al. 2018). For GB1, we assigned the ¹⁹F spins that dephased each carbon based on the known structure of the protein. For de novo structure determination, the ¹⁹F dephasers can be assigned biosynthetically, by introducing one ¹⁹F at a time, by selective inversion of one of the ¹⁹F signals, or by a joint and iterative analysis of the ¹⁹F-¹⁹F and ¹³C-¹⁹F dipolar coupling data.

This 2D ¹³C-¹³C resolved ¹³C-¹⁹F REDOR technique can readily incorporate other dipolar recoupling or spin diffusion schemes for ¹³C-¹³C correlation, and can be extended to 3D, for example adding a ¹H dimension, to further enhance site resolution. The 2D ¹³C-¹³C resolved ¹³C-¹⁹F distance measurement approach is conceptually similar to the 2D ¹³C-¹⁵N resolved ¹⁹F-¹⁵N REDOR technique, which has been demonstrated at 11.7 Tesla (500 MHz ¹H Larmor frequency) under 11 kHz MAS (Graesser et al. 2007). The main advantage of the ¹³C-¹³C resolved ¹³C-¹⁹F technique is the 1.4-fold longer distances compared to ¹⁵N-¹⁹F distances for the same coupling strength. A practical benefit is that the REDOR-RFDR experiment can be conducted using a triple-resonance HFC probe while the ¹³C-¹⁵N resolved ¹⁵N-¹⁹F REDOR requires a quadruple-resonance HFCN probe. It is worth noting that for low-sensitivity biological macromolecules, even a single 2D REDOR-RFDR difference spectrum, measured at one REDOR mixing time, already gives powerful structural constraints about the ¹⁹F-proximal residues in proteins. Finally, for proteins in which Tyr residues are the targets of fluorination, labeling both 3-¹⁹F and 5-¹⁹F will be advantageous over single 3-¹⁹F or 5-¹⁹F labeling (Ravichandran et al. 2013), by increasing the number of dephasing spins and simplifying distance analysis.