High-Sensitivity Detection of Nanometer ¹H-¹⁹F Distances for Protein Structure Determination by ¹H-Detected Fast MAS NMR

Abstract

Protein structure determination by solid-state NMR requires the measurement of many inter-atomic distances through dipole-dipole couplings. To obtain multiple long-range distance restraints rapidly and with high sensitivity, here we demonstrate a new ¹H-detected fast magic-angle-spinning (MAS) NMR technique that yields many long distances in a 2D-resolved fashion. The distances are measured up to ~15 Å, with an accuracy of better than 10%, between ¹H and ¹⁹F, two nuclear spins that have the highest gyromagnetic ratios. Exogenous fluorines are sparsely introduced into the aromatic residues of the protein, which is perdeuterated and back-exchanged to give amide protons. This ¹H-¹⁹F distance experiment, termed 2D HSQC-REDOR, is demonstrated on the singly fluorinated model protein, GB1. We extracted 33 distances between 5-¹⁹F-Trp43 and backbone amide protons, using 2D spectral series that were measured in less than 3 days. Combining these ¹H-¹⁹F distance restraints with ¹³C-¹⁹F distances and chemical shifts, we calculated a GB1 structure with a backbone RMSD of 1.73 Å from the high-resolution structure. This ¹H-detected ¹H-¹⁹F distance technique promises to provide a highly efficient tool for constraining the three-dimensional structures of proteins and protein-ligand complexes, with not only precise and fast measurements, but also access to truly long-range distances.

Graphical Abstract

Solid-state NMR (SSNMR) is an excellent tool for determining the three-dimensional structures of complex biological macromolecules such as membrane proteins and fibrillar proteins ¹. A necessary component of SSNMR-based structure determination is the measurement of inter-atomic distances using nuclear-spin dipole-dipole couplings. To determine the three-dimensional fold and oligomeric structure of proteins and macromolecular assemblies, distances in the 1-2 nm range are crucial ². When ¹³C-¹³C and ¹³C-¹⁵N distances are measured, as in most NMR structural determination studies ¹, the distance restraints are typically limited to ~7 Å or less due to the low gyromagnetic ratios of ¹³C and ¹⁵N spins. Longer distance restraints of ~10 Å can be measured when dipolar couplings between different nuclear spins with higher gyromagnetic ratios such as ¹⁹F and ³¹P ^3–5 are recoupled under MAS using the rotational-echo double-resonance (REDOR) technique ⁶. However, these REDOR experiments are typically conducted in a 1D fashion, thus limiting the number of distances that can be measured in each experiment. Moreover, detection of low-frequency nuclei limits the spectral sensitivity, thus requiring extensive signal-averaging times of weeks to even months. As a result, although long-distance measurement by SSNMR is possible, so far such measurements have been inefficient, making fast and high-sensitivity detection a bottleneck in accessing distances in the nanometer range.

Here we demonstrate the combination of fast MAS, ¹H detection, and ¹H-¹⁹F REDOR to measure a large number of long-range distances in a 2D-resolved fashion within 2-3 days. The 2D spectra have high sensitivity due to ¹H detection, distances can be measured to ~15 Å due to the high gyromagnetic ratios of ¹H and ¹⁹F spins, and the measured distances have a high accuracy of ~10% due to the simplicity of heteronuclear spin dynamics. The choice of ¹H-¹⁹F heteronuclear distance measurement circumvents the difficulties in measuring homonuclear ¹H-¹H distances, where the dense ¹H network in biomolecules causes dipolar truncation ⁷ and relayed transfer ^8–9. Proton dilution by perdeuteration followed by back H/D exchange ¹⁰ still leaves uncertainties in ¹H-¹H distance measurements due to residual relayed polarization transfer, while ultrafast MAS of ~100 kHz for protonated samples ^11–15 requires 3D and 4D correlation experiments for ¹H chemical shift assignment ^16–18. The ¹H-¹⁹F distance approach demonstrated here circumvents these difficulties, and complements the recently introduced homonuclear ¹⁹F-¹⁹F distance measurements ^19–21. It also complements a 2D-resolved ¹³C-¹⁹F REDOR technique ²², which has a shorter distance upper limit of ~1 nm. The combination of ¹H and ¹⁹F for distance measurement was previously reported for the slow MAS condition ^23–24, which suffered from short ¹H T₂ relaxation times and the lack of possibility for high-sensitivity ¹H detection.

The pulse sequence for this 2D-resolved ¹H-¹⁹F distance experiment (Fig. 1) contains a ¹H-¹⁹F REDOR module ^{6, 23} inserted into a cross-polarization (CP) based 2D ¹H-¹⁵N heteronuclear single-quantum coherence (HSQC) sequence immediately before ¹H detection. Two 2D spectra are measured back-to-back, one with the ¹⁹F pulses off (S₀) and the other with the ¹⁹F pulses on (S). The S₀ spectrum gives relaxation-compensated intensities for the ¹H-¹⁵N correlation peaks while the S spectrum gives lower intensities for protons that are close to the ¹⁹F probe. The difference spectrum (ΔS) between the two shows the amide protons that are close to the ¹⁹F spin. We demonstrate this technique using ¹³C, ¹⁵N, ²H-labeled GB1 that contains a single fluorinated residue, 5-¹⁹F-Trp43 (5F-W43 CDN-GB1). The protein was fully back-exchanged in H₂O to obtain amide protons. The fluorinated Trp was incorporated into the protein using glyphosate before the onset of protein expression, to suspend aromatic amino acid synthesis in E. coli ²⁵. Unlabeled Tyr and Phe were also added, thus the 2D spectra do not contain ¹H-¹⁵N cross peaks of the aromatic residues. The single 5F-W43 served to dephase all ¹⁵N-bonded amide protons.

Figure 1.

¹H-detected ¹H-¹⁹F HSQC-REDOR pulse sequence. The REDOR block is inserted between the ¹⁵N evolution (t₁) and ¹H detection (t₂) periods. A MISSISSIPPI period after the ¹⁵N evolution suppresses the solvent ¹H signals ²⁶.

2D H^N-¹⁵N correlation spectra of 5F-W43 CDN-GB1 (Fig. S1) measured at 40 kHz MAS and a nominal temperature of 273 K, which corresponds to an estimated sample temperature of ~303 K, show well-resolved peaks with ¹H linewidths of 0.1–0.3 ppm. These ¹H and ¹⁵N chemical shifts are nearly identical to those of non-fluorinated GB1 (Table S1), indicating that single-site fluorination does not perturb the protein structure. Control spectra measured with the ¹⁹F pulses off show the full intensities of H^N-N correlation peaks (Fig. 2A) modulated only by T₂ relaxation of the protons. Turning on the ¹⁹F pulses reduced the intensities of ¹⁹F-proximal amide protons, thus the difference spectrum (ΔS) displays H^N signals close to ¹⁹F. At short mixing times such as 1.8 ms (Fig 2B), only a few correlation peaks are observed that correspond to H^N atoms that are 4-6 Å from the ¹⁹F. At longer mixing times such as 7.5 ms (Fig. 2C), many more correlation peaks appear in the difference spectrum, including amide protons that are 10–14 Å from the ¹⁹F probe. The difference intensity, after dividing by the control intensity to account for T₂ relaxation (ΔS/S₀ = 1 − S/S₀), indicates the strength of the dipolar coupling and hence the ¹H-¹⁹F distances. Importantly, these 2D spectra have very high signal-to-noise ratios of 100-200 : 1 due to ¹H detection and fast MAS, thus giving precise distance-dependent REDOR dipolar dephasing.

Figure 2.

Representative 2D ¹H-¹⁹F HSQC-REDOR spectra of 5F-W43 CDN-GB1. The spectra were measured under 40 kHz MAS at 273 K. Indicated distances are from PDB 2LGI structure. (A) Control S₀ spectrum at 0.6 ms mixing, showing most H^N-N cross peaks. (B) Difference spectrum ΔS at 1.8 ms. (C) Difference spectrum at 7.5 ms mixing. (D) H^N T₂ relaxation times at 40 kHz (black) and 30 kHz MAS (red). Faster MAS gives longer relaxation times and thus higher sensitivity for the 2D ¹⁹F-dephased spectra.

The cross peak intensities in the control spectra decay with increasing mixing times due to ¹H^N T₂ relaxation. We found that the residue-specific ¹H^N T₂ values (Fig. 2D) nearly doubled from 30 kHz MAS (average 4.2 ms) to 40 kHz MAS (average 7.2 ms), indicating that faster MAS lengthens the coherence lifetime, by averaging residual ¹H-¹H dipolar couplings. Long ¹H^N T₂ times are essential for measuring weak ¹H-¹⁹F dipolar couplings that encode longer distances, since longer REDOR mixing times of ~10 ms are required to detect these weak couplings. Fig. S2 shows that the difference spectra measured at 40 kHz MAS have many more cross peaks than the 30 kHz spectra for similar mixing times. Thus, the increased ¹H^N T₂ by fast MAS compensates for the additional pulse imperfections arising from the larger number of rotation periods needed to reach the same mixing time.

To quantify the ¹H-¹⁹F distances, we investigated how the large ¹⁹F chemical shift anisotropy (CSA) affects ¹H-¹⁹F dipolar dephasing. We simulated the REDOR dephasing, S/S₀, as a function of mixing time (Fig. S3). At a slow MAS frequency of 10 kHz, the REDOR curves show noticeable dependences on the orientation of the ¹⁹F chemical shift tensor relative to the ¹H-¹⁹F internuclear vector. But this orientation dependence vanishes at MAS rates of 40 kHz and above, so that the ¹H-¹⁹F dipolar dephasing depends chiefly on distances. This is consistent with the observation for ¹³C-¹⁹F REDOR ²² and represents a significant benefit of fast MAS for REDOR experiments. The magnitude of the ¹⁹F CSA also affects the REDOR dephasing, by elevating the S/S₀ values; but similarly this effect is diminished by fast MAS and can be rigorously accounted for by numerical simulations. In fitting the measured REDOR dephasing below, we also take into account finite ¹⁹F pulse lengths and pulse imperfections ^{22, 27}.

We extracted ¹H-¹⁹F distances from the mixing-time-dependent S/S₀ intensities of the 2D HSQC-REDOR spectra. Representative REDOR decays for several residues are shown in Fig. 3 for the 40 kHz dataset; the complete set of decay trajectories is given in Fig. S4. The high sensitivity of the ¹H-detected 2D spectra gave rise to very precise S/S₀ values, with uncertainties of less than 0.02 in the dephasing values (Fig. 3a). A wide range of dipolar dephasing trajectories are observed: residues such as I6 and V54 show rapid dephasing to zero within ~5 ms while residues such as V39 manifest slow dephasing and high S/S₀ values of ~0.6 even at 10 ms. Therefore, the technique is sensitive to the distance distribution in the protein.

Figure 3.

H^N-¹⁹F REDOR dephasing, measured under 40 kHz MAS, for representative residues of GB1. Best-fit simulations (solid lines) are shown, together with simulated curves for distances that are ±0.5 Å from the best fit (dashed lines). Residues with shorter T₂ relaxation times such as I6 may have their signals disappear from some of the spectra at longer mixing times, in which case fewer REDOR intensity data points are detected. The high sensitivity of the ¹H-detected spectra yielded error bars that are typically less than 0.02, which are smaller than the points in the plot, as shown for the I6 data as an example.

We extracted best-fit distances from the minimum RMSD between the simulated and measured S/S₀ intensities (Fig. S5). Since fast protein motions weaken the ¹H-¹⁹F dipolar couplings, to increase the accuracy of the distance restraints, we semi-quantitatively accounted for residual motion of the fluorine-bearing indole of W43 by analyzing the measured ¹⁹F chemical shift anisotropy (CSA). The ¹⁹F spinning sideband spectrum of 5F-W43 ²² indicates a CSA order parameter of 0.87, which is used to approximate the scaling factor for the ¹H-¹⁹F dipolar couplings. A more exact analysis of the motional scaling factor is not possible without detailed knowledge of the geometry of motion and the relative orientation of the ¹H-¹⁹F vector to the indole ring. Since distances depend on the dipolar coupling as ω_d⁻³, a dipolar scaling factor of 0.87 means that the true distances may be up to 4.6% shorter than the best-fit distances. Thus we applied a uniform scaling factor of 95.4% to the measured distances (Table S2). These motionally adjusted distances, although approximate, show excellent agreement with the distances in the high-resolution GB1 structure (PDB code: 2LGI) (Fig. 4A, Fig. S6): the correlation plot has a slope of 1.1, a correlation coefficient of 0.89 and an RMSD of 1.1 Å between the measured and predicted distances. We observed ¹H-¹⁹F distances up to 1.5 nm, which is the upper limit expected for this model protein, since the overall dimension of the protein is about 1.5 x 1.4 x 2.9 nm. Based on the 2D dataset obtained at a nominal temperature 273 K (which corresponds to an estimated sample temperature of ~303 K), the measured distances are on average 0.66 Å longer than the distances in the high-resolution structure, indicating that the measured ¹H-¹⁹F dipolar couplings are still weaker than expected even after taking into account the indole motion. We attribute this slight lengthening of the measured distances to residual motion of the protein backbone, which may occur on the microsecond timescale to escape detection in the ¹⁹F CSA spectra ²⁸. This hypothesis is supported by the fact that the distance accuracy improved at a lower sample temperature of ~275 K (Fig. S6C). Among all measured distances, Q2 and A20 have shorter distances than predicted by the structure. We attribute this effect to intermolecular dipolar coupling, since Q2 and A20 lie at one end of the ellipsoid-shaped protein, in closer contact with the neighboring protein than other amide protons (Fig. 4D). L5 and K31 show longer measured distances than predicted by the structure, which can be attributed to partial overlap of these resonances in the spectra.

Figure 4.

(A) Comparison of the ¹H^N-F distances measured at 273 K under 40 kHz MAS with predicted distances in the high-resolution structure (PDB ID: 2LGI). (B) Measured ¹H^N-F distances (dashed lines) in 5F-W43 (pink sphere) labeled GB1. (C) ¹H-¹⁹F and ¹³C-¹⁹F distance-restrained GB1 structure (blue), superimposed with the high-resolution SSNMR structure (red) ²⁹. (D) A20 and Q2 are located at one end of the protein, and have similar H^N-F distances of 12-13 Å to 5F-W43 of the same molecule versus to 5F-W43 of a neighboring molecule. The simultaneous coupling to two fluorines explains the apparent shortening of the measured distances compared to the known intramolecular distances.

Altogether, we measured 33 distances from 5F-W43 to amide protons (Fig. 4B). Among these, 5 are shorter than 0.7 nm, 10 are 0.7–1.0 nm, whereas 18 are 1.0–1.5 nm. Thus, the majority of the measured ¹H^N-¹⁹F distances are sufficiently long to constrain the three-dimensional structure of the protein. We calculated the GB1 structure using the 33 ¹H-¹⁹F distances measured here, the 47 ¹³C – ¹⁹F distances reported recently ²², and the known ¹³C and ¹⁵N chemical shifts of the protein ³⁰. The structure ensemble (Fig. S7) has a backbone RMSD of 1.24 Å (Table S3) and the best structure agrees with the high-resolution GB1 SSNMR structure with a backbone RMSD of 1.73 Å (Fig. 4C, Table S4). Importantly, the protein fold is correctly captured with a minimal number of constraints, with the main deviation being the spacing between two of the four β-strands. Therefore, ¹H-¹⁹F REDOR not only gives crucial qualitative inter-atomic proximities but also quantitative distance restraints for the three-dimensional structure.

We envision that this ¹H-¹⁹F HSQC-REDOR technique will facilitate SSNMR-based structure determination by providing accurate, long-range distances in a high-throughput manner. Each 2D ¹H-¹⁹F HSQC-REDOR spectrum of GB1 was measured in about 3 hours with signal-to-noise ratios of 100-200 : 1 for the cross peaks, and a full series of spectra were measured in 2-3 days (Table S5). Thus this technique can be applied to proteins that are several times larger than GB1 while still retaining sufficient sensitivity. Similar to all ¹H-detected fast MAS experiments, the key requirement for multiplexed ¹H-¹⁹F distance measurement is sufficient resolution of the ¹H and ¹⁵N chemical shifts in the 2D spectra. This ¹H-¹⁹F distance technique is complementary to the recently introduced ¹³C-¹³C resolved ¹³C-¹⁹F distance experiment ²², and yields more distance restraints than ¹⁹F-¹⁹F 2D experiments ^20–21. Aromatic residues in proteins, particularly the sparse tryptophan (Trp) residues, can be fluorinated during protein expression without perturbing the structure ^31–32. Lipid-exposed Trp residues on the surfaces of membrane proteins can allow the investigation of ligand binding ³³ and intermolecular assembly ^34–35, whereas Trp residues in the core of globular proteins can provide crucial constraints for the three-dimensional structure. The ability to measure distances to ~2 nm opens the possibility of elucidating protein conformational changes, protein-ligand binding, and macromolecular binding interfaces where ¹⁹F labeling of one partner can be combined with orthogonal labeling of the other partner ⁴. With the increased sensitivity from ¹H detection and the increased distance reach, this ¹H-¹⁹F distance technique should significantly enhance the capability of SSNMR-based structure determination.