Backbone Amide ¹⁵N Chemical Shift Tensors Report on Hydrogen Bonding Interactions in Proteins: A Magic Angle Spinning NMR Study

Abstract

Chemical shift tensors (CSTs) are an exquisite probe of local geometric and electronic structure. ¹⁵N CST are very sensitive to hydrogen bonding, yet they have been reported for very few proteins to date. Here we present experimental results and statistical analysis of backbone amide ¹⁵N CSTs for 100 residues of four proteins, two E. coli thioredoxin reassemblies (1–73-(U-¹³C,¹⁵N)/74–108-(U-¹⁵N) and 1–73-(U-¹⁵N)/74–108-(U-¹³C,¹⁵N)), dynein light chain 8 LC8, and CAP-Gly domain of the mammalian dynactin. The ¹⁵N CSTs were measured by a symmetry-based CSA recoupling method, ROCSA. Our results show that the principal component δ₁₁ is very sensitive to the presence of hydrogen bonding interactions due to its unique orientation in the molecular frame. The downfield chemical shift change of backbone amide nitrogen nuclei with increasing hydrogen bond strength is manifested in the negative correlation of the principal components with hydrogen bond distance for both α-helical and β-sheet secondary structure elements. Our findings highlight the potential for the use of ¹⁵N CSTs in protein structure refinement.

1. INTRODUCTION

Hydrogen bond (HB) is one of the important noncovalent interactions stabilizing folded proteins [1]. NMR spectroscopy has long been used in the identification of hydrogen bonds in peptides and proteins through a number of NMR parameters. In particular, solution NMR signatures include chemical shift changes of the nuclei participating in HB interactions, reduced amide proton exchange rates, and indirect scalar or J-couplings mediated across hydrogen bonds to identify residues involved in hydrogen bonds. Among these, chemical shift perturbations are the easiest to measure directly using multidimensional homonuclear and heteronuclear correlation spectroscopy, as chemical shift parameters are highly sensitive to the local changes in the electronic environment due to HB interactions.

While solution NMR methods yield only the isotropic chemical shifts, in solid-state NMR (SSNMR) chemical shift tensors (CSTs) can be recorded. Chemical shift anisotropy (CSA) can be measured for multiple sites in a site-specific manner using magic angle spinning (MAS) NMR spectroscopy, where a chemical shift recoupling sequence is introduced into multidimensional experiments [2]. Yet, the literature on CSTs in proteins is sparse [3; 4] and the general rules governing the dependence of CST principal components on HB interactions remain to be established. For instance, the downfield change in the isotropic chemical shift due to the deshielding of amide protons involved in hydrogen bonds has been very well documented in the literature [5; 6; 7]. Recently our group reported site-specific amide proton CSA measurements for CAP-Gly domain of the mammalian dynactin, using the RN-symmetry based sequences [4]. The results revealed that indeed there is a high degree of correlation between the downfield shift of the principal components of the amide proton CST and the hydrogen bond length. Direct ¹H–¹⁵N dipolar recoupling measurements have also been sometimes used in the literature to identify hydrogen bonds in terms of the elongation of N-H bond vector [8].

There have also been reports in the literature on the sensitivity of heteronuclear CS parameters to HB interactions in peptides and amino acids. In particular, the CSTs of both carbonyl carbon and amide nitrogen nuclei are as sensitive to HBs as amide protons. In an early SSNMR study of L-alanine containing short peptides, Asakawa et al. showed that as the hydrogen bond strength increases with decreasing distance between the nitrogen and oxygen atom (R_N···O), the principal component δ ₂₂ of the carbonyl carbon shows a large downfield shift accompanied by a similar trend in the isotropic chemical shift [9]. In a more detailed SSNMR study of CSA of carboxyl groups of amino acids, Gu et al. demonstrated the unique sensitivity of δ ₁₁ and δ ₂₂ to the protonation state and HB interaction of the CO group, respectively [10]. In a DFT study of model dipeptides, Walling et al. showed that, in addition to carbonyl atoms, backbone amide ¹⁵N CST parameters are also sensitive to hydrogen bonds [11]. In particular, they showed that both δ ₁₁ and δ ₃₃ exhibit downfield shift in both α-helical and β-sheet conformation, while δ ₂₂ displays upfield shift in β-sheet conformation. This trend is consistent with ¹⁵N CST calculations performed on other model systems such as benzamide [12]. In contrast, a downfield shift of δ ₂₂ with increasing R_N···O and negligible sensitivity to HBs in both δ ₁₁ and δ ₃₃ was observed for the ¹⁵N nuclei of imidazole groups in histidines by Wei et al. [13], which is in agreement with results from ¹⁵N solution NMR studies of imidazoles [14]. Altough the magnitudes of the principal components of ¹⁵N CST are very sensitive to HB interactions, the orientation of the ¹⁵N CSTs in the molecular frame has been reported to be generally unperturbed on hydrogen bonding [15].

Despite the vast amount of theoretical and experimental-SSNMR studies of hydrogen bond effect on both ¹⁵N and ¹³C CST parameters on amino acids and model peptides, the corresponding experimental investigations in large systems, such as proteins are lacking. In this work, we present experimental SSNMR results and a statistical analysis of backbone amide ¹⁵N CST parameters for a total of 100 residues belonging to four proteins, two E. coli thioredoxin reassemblies ((1–73-(U-¹³C,¹⁵N)/74–108-(U-¹⁵N) and 1–73-(U-¹⁵N)/74–108-(U-¹³C,¹⁵N)), dynein light chain 8 LC8, and CAP-Gly domain of the mammalian dynactin. The ¹⁵N CST parameters measured by the ROCSA symmetry based recoupling are interpreted in terms of their sensitivity to HB interactions of the residues. Our results reveal unequivocal correlation between the principal components of the backbone ¹⁵N chemical shift tensors and hydrogen bonding interactions, underscoring the potential for use of ¹⁵N CSTs in protein structure refinement.

2. EXPERIMENTS AND METHODS

2.1. Protein samples preparation.

Preparation of solid-state NMR samples of E. coli thioredoxin reassemblies [1–73-(U-¹³C,¹⁵N)/74–108-(U-¹⁵N) and 1–73-(U-¹⁵N)/74–108-(U-¹³C,¹⁵N)], U-¹³C/¹⁵N DLC8, and U-¹³C/¹⁵N CAP-Gly domain of the mammalian dynactin by controlled precipitation was reported previously [16; 17; 18]. Approximately 10 mg of each sample was packed into a 3.2 mm Varian MAS rotor and sealed with an upper spacer and a top spinner.

2.2. SSNMR spectroscopy.

All SSNMR spectra presented in this work were acquired at 14.1 T on a narrow bore Varian InfinityPlus spectrometer operating at Larmor frequencies of 599.8 MHz for ¹H, 150.8 MHz for ¹³C, and 60.8 MHz for ¹⁵N. The instrument was outfitted with a 3.2 mm T3 MAS probe. The MAS frequency was 10 kHz for all experiments, and was controlled to within ±1 Hz by a Varian MAS controller. The sample temperatures were kept in the range 0 ^oC to –15 ^oC. The temperature reported includes a MAS frequency-dependent correction determined experimentally by using PbNO₃ as the temperature sensor [19]. ¹⁵N chemical shifts were referenced with respect to NH₄Cl used as an external referencing standard following the standard protocol [20]. The ¹⁵N CSA recoupling was achieved by the symmetry-based ROCSA method [21]. The pulse sequence for the 3D NCA-ROCSA experiment is shown in Figure 3. The excitation frequencies were set at 122.4 ppm for ¹⁵N and at 55.5 ppm for ¹³C. Eighteen ROCSA points were acquired with the dwell time equal to one rotor period, and 64 ¹⁵N t₂ points with acquired with the dwell time of 120 μs. 128 scans were added to record each point in the indirect dimensions of the 3D spectra. ¹H 90 pulse width was 2.78 μs. The contact time for the ¹H-¹⁵N CP was 1.6 ms. The ¹H radio frequency field strength was 51 kHz, the ¹⁵N field was linearly ramped 70–100% with the center of the ramp being 41 kHz. The ROCSA sequence was used with a window τ_a of 3.29 μs, and 42.8 kHz rf irradiation was employed. The Z-filter pulses were placed after the ROCSA block; the rf field strength was 50 kHz. The Z-filter delay was set equal to one rotor period. XY-16 decoupling [22] on the ¹³C channel was performed with the rf field strength of 50 kHz. 100 kHz CW decoupling was performed on the ¹H channel during ROCSA. SPECIFIC-CP [23] for ¹⁵N-¹³CA transfer was utilized. The ¹³C field was tangentially ramped 90–100% with the center of the ramp being 35 kHz; the rf field strength for ¹⁵N was 25 kHz; the contact time was 6.2 ms. 100 kHz CW decoupling was applied during SPECIFIC-CP. The Z-filter pulses were incorporated on the ¹³C channel immediately after SPECIFIC-CP; the rf field strength was 67.9 kHz. 90 kHz TPPM decoupling [24] was applied during the t₃ evolution; the TPPM pulse width was 5.4 μs.

Figure 3.

Illustration of the orientation of the backbone amide ¹⁵N CS tensor in the peptide plane (molecular frame). The bond lengths and bond angles are not to scale.

2.3. ¹⁵N ROCSA lineshape simulations.

Numerical simulations of the ROCSA lineshapes were performed on a single ¹⁵N spin using SIMPSON [25] with repulsion 320 powder angle set [26] and 36 γ-angles. An ideal ¹H-¹⁵N cross polarization and proton decoupling during ROCSA evolution was assumed. The ROCSA lineshapes were fitted using MINUIT with the SIMPLEX method with three adjustable parameters: i) reduced anisotropy (δ_σ); ii) asymmetry parameter (η_σ); and iii) exponential line-broadening parameter to model transverse relaxation during ¹⁵N ROCSA evolution. The uncertainties in the fitting parameters were determined using the Monte Carlo method as described in the literature [27]. The principal components of the ¹⁵N CS tensors in the standard convention (δ₁₁ ≥ δ ₂₂ ≥ δ ₃₃) were determined using the relations, ${\delta }_{11}={\delta }_{iso}+{\delta }_{\sigma }$, ${\delta }_{22}={\delta }_{iso}-{\delta }_{\sigma }(1-{\eta }_{\sigma })/2$, and ${\delta }_{33}={\delta }_{iso}-{\delta }_{\sigma }(1+{\eta }_{\sigma })/2$.

2.4. Hydrogen bond analysis.

The hydrogen bond information of residues in the E. coli thioredoxin reassembly was inferred from its 1.8 Å crystal structure (PDB ID 2TRX) [28]. For LC8, a 2.8 Å resolution crystal structure (PDB ID 2PG1) was used [29]. MAS NMR structure was used for CAP-Gly domain of the mammalian dynactin (PDB ID 2M02) [30]. In all cases, an upper cutoff distance of 3.5 Å was set for R_N···O. Residues exhibiting significant dynamic averaging of ¹⁵N CSA due to internal motions in the solid-state samples, such as G21, R73, and I75 of E. coli thioredoxin [31], were excluded from this analysis.

3. RESULTS AND DISCUSSION

3.1. Sensitivity of backbone amide ¹⁵N CST to presence of hydrogen bonds.

As shown in Figure 1B for E. coli thioredoxin reassembly (U-¹⁵N 1–73/¹³C,¹⁵N 74–108), the resolution of the NC chemical shift correlation plane of the 3D ROCSA-NCA spectrum is high permitting the extract the ROCSA lineshapes corresponding to the resolved peaks. In total, we have recorded ROCSA lineshapes for 100 residues in E. coli thioredoxin reassembly (U-¹³C/¹⁵N 1–73/¹⁵N 74–108, U-¹⁵N 1–73/¹³C,¹⁵N 74–108), dynein light chain 8 (LC8), and CAP-Gly domain of mammalian dynactin. The corresponding CST parameters are summarized in Table S1. Representative experimental and best-fit ROCSA lineshapes for E. coli thioredoxin reassembly (U-¹⁵N 1–73/¹³C,¹⁵N 74–108) and LC8 are shown in Figure 2.

Figure 1.

A) 3D ¹⁵N-¹³CA-ROCSA pulse sequence for site-specific measurement of the backbone amide ¹⁵N CSA lineshapes. B) The first 2D ¹⁵N-¹³CA plane of the 3D ¹⁵N-¹³CA-ROCSA spectrum of E. coli thioredoxin reassembly (U-¹⁵N 1–73/¹³C,¹⁵N 74–108).

Figure 2.

Representative backbone amide ¹⁵N ROCSA lineshapes of A) E. coli thioredoxin reassembly (U-¹⁵N 1–73/¹³C,¹⁵N 74–108) and B) dynein light chain 8, LC8.

Downfield shift of the isotropic component of the CS tensor of amide protons and amide nitrogens is known to be a characteristic marker of HB interactions [5; 6; 7]. This trend is confirmed in our results by the comparison of the statistical average value of the isotropic chemical shifts of the hydrogen-bonded ¹⁵N nuclei versus those that do not participate in hydrogen bonds. As shown in Table 1, the average isotropic chemical shift δ _iso, of the hydrogen bonded ¹⁵N nuclei is around 120.6 ppm while that of the non-hydrogen-bonded ¹⁵N is 118.9 ppm. The difference of 1.7 ppm is significant enough for the direct detection of the nuclei participating in H-bonds from the 1D ¹⁵N CPMAS spectra or 2D correlation spectra, such as NCA or NCO.

Table 1.

Statistical Averages of the Principal Components of ¹⁵N^H Chemical Shift Tensors for Hydrogen-Bonded and Non-Hydrogen-Bonded Residues of E. coli Thioredoxin Reassemblies, LC8, and CAP-Gly.

Type	δiso (ppm)	δσ (ppm)	ησ	δ11 (ppm)	δ22 (ppm)	δ33 (ppm)
HB	120.6 ± 6.6	95.8 ± 5.1	0.23 ± 0.08	216.3 ± 8.9	83.8 ± 8.6	61.5 ± 6.6
NHB	118.9 ± 7.1	93.6 ± 7.0	0.24 ± 0.09	212.5 ± 8.8	83.3 ± 10.6	61.0 ± 8.4
Difference	1.7	2.2	–0.01	3.8	0.5	0.5

HB – hydrogen-bonded residues, NHB – non-hydrogen-bonded residues. The values presented here are the statistical average ± standard deviations.

The hydrogen bond interaction increases the width of ¹⁵N CSA lineshapes, which can easily be measured from the width of the ROCSA patterns. The reduced anisotropy, δ_σ, of the ¹⁵N CST exhibits the same trend as the isotropic chemical shifts. Specifically, the average δ_σ value of hydrogen bonded residues is 95.8 ppm while δ_σ of residues not forming hydrogen bonds is 93.6 ppm. Although the difference of 2.2 ppm is relatively small, it is statistically significant and exceeds the typical experimental errors associated with modern CSA recoupling sequences designed for MAS frequencies exceeding 10 kHz, such as ROCSA and RNCSA [3]. Unlike δ_iso and δ_σ, the asymmetry parameter of the ¹⁵N CSA tensor (η_σ) is not sensitive to hydrogen bond interaction. The average η_σ values of both hydrogen-bonded and non-hydrogen-bonded ¹⁵N nuclei are essentially the same 0.2.

More insight into the sensitivity of the individual components of the CSA tensor to the hydrogen bonding interactions can be gained from the analysis of the principal components using the standard convention, (δ₁₁, δ₂₂, δ₃₃). It is evident from Table 1 that the most downfield shifted component δ₁₁ is the most sensitive to the presence of hydrogen bonds. The average δ ₁₁ value for hydrogen-bonded ¹⁵N nuclei is 216.3 ppm, while it is 212.5 ppm for non-hydrogen-bonded residues. The component with intermediate shielding, δ_22, and the most shielded component δ₃₃ of the ¹⁵N CSA tensor show relatively minor downfield shift due to the presence of HB interactions. The average value of δ₂₂ for hydrogen-bonded residues is 83.8 ppm and 83.3 ppm for the non-hydrogen-bonded residues. The corresponding values for δ₃₃ are 61.5 and 61.0 ppm.

The result that the δ ₁₁ component of the ¹⁵N CSA tensor has the highest sensitivity to hydrogen bonding is in perfect agreement with theoretical studies of ¹⁵N shielding tensors on hydrogen bonded amino acids and peptides [15; 32]. In these studies, it has been shown that the orientation of the ¹⁵N CS tensor relative to the molecular frame is such that the δ₁₁ component lies in the peptide plane with the angle between the N–H bond vector and the δ₁₁ component being around ca. 10–20 degrees (Figure 3). The almost parallel orientation of δ₁₁ component with respect to the N–H bond vector gives it the highest perturbation due to hydrogen bond interaction. The δ₂₂ component which shows weak sensitivity to hydrogen bond interactions lies almost perpendicular to the peptide plane. The equally weakly sensitive δ₃₃ component lies in the peptide plane with almost perpendicular orientation to the N–H bond vector.

It is important to mention that the backbone amide ¹⁵N CS tensors are not only sensitive to hydrogen bonds, but are also dependent on secondary structure, nature of the amino acid, and internal dynamics of backbone. This fact is evident in the large standard deviations associated with each of the three principal components for both hydrogen-bonded and non-hydrogen-bonded residues as presented in Table 1. Nevertheless, despite the limited number of ¹⁵N CSA lineshapes available so far and analyzed in this work, we do observe a clear downfield shift in δ₁₁ due to hydrogen bond interactions.

3.2. Correlation of backbone amide ¹⁵N CST components with isotropic chemical shift.

In the case of amide proton CSA parameters, Tjandra et al. reported the correlation of the three principal components with the isotropic chemical shift as an essential metric for the sensitivity of amide proton CSA parameters to HB interactions [5]. Recently, in our work on MAS NMR measurements of backbone amide proton CSTs of CAP-Gly domain of the mammalian dynactin, the correlation of the three principal components with the isotropic chemical of the backbone amide protons was verified to be correct [4]. In order to examine whether the same trend is observed for the backbone amide ¹⁵N CST parameters, in Figure 4 we have plotted the three ¹⁵N principal components versus the isotropic chemical shift. Similar to the findings for the backbone amide proton CSAs, ¹⁵N CS principal components also follow a linear relationship with the isotropic chemical shift, with a strong positive correlation indicated by positive values of Pearson correlation coefficients of 0.73, 0.84, and 0.79 for δ₁₁, δ₂₂, and δ₃₃ respectively. These results suggest that that one can expect to see a similar positive correlation of the three amide ¹⁵N CSAs with hydrogen bond strength or a negative correlation with hydrogen bond distance. This assertion is corroborated in the following section.

Figure 4.

Correlation of the backbone amide ¹⁵N CST principal components with isotropic chemical shift for the four proteins under investigation, E. coli thioredoxin reassemblies, LC8, and CAP-Gly. The correlation is determined the Pearson correlation coefficients (R_P) with values greater than 0.70 indicating a strong positive correlation of the three principal components with ¹⁵N isotropic chemical shift.

3.3. Correlation of backbone amide ¹⁵N CST with hydrogen bond length.

Ab initio theoretical and experimental studies have shown that CS tensor values of nuclei involved in hydrogen bond correlate well with hydrogen bond length values [7; 12]. In particular, the downfield shift of nuclei of the donor group with increasing hydrogen bond strength or decreasing hydrogen bond length (R_N…O or R_H…O) due to increased deshielding has been well investigated [6; 7]. In this section, we present the statistical correlation of backbone amide ¹⁵N CS principal components with the distance between the amide ¹⁵N atom in the donor group and the oxygen atom in the acceptor group. As mentioned in the previous section, in addition to HB interactions backbone amide ¹⁵N CST values are dependent on several other parameters such as secondary structure, local conformation and internal dynamics. In order to decouple the sensitivity of amide ¹⁵N CST on secondary structure, we have performed two separate correlation analyses, one for α-helical residues, and the other for β-strand residues. The results are shown in Figure 5. Out of the 100 residues we have analyzed from the four proteins, 65 are hydrogen bonded and 35 are non-hydrogen bonded. Among the hydrogen-bonded, 21 are α-helical residues and 32 are located in β -strands.

Figure 5.

Correlation of the hydrogen-bonded backbone amide ¹⁵N CST principal components with hydrogen bond distance R_N…O for A) α-helical and B) β-sheet structure, for the four proteins under investigation, E. coli thioredoxin reassemblies, LC8, and CAP-Gly. The downfield shift of all three ¹⁵N CST principal components, notably in δ ₁₁, with decreasing R_N…O is indicated by the negative Pearson correlation coefficients (R_P) for both α-helical and β-sheet secondary structures.

In Figure 5, the correlation of the principal components of backbone amide ¹⁵N CST is plotted against R_N…O for both α-helical and β -strand residues. First, a negative correlation as determined by the Pearson correlation coefficient is evident in all the graphs. This statistical observation is consistent with the fact that the downfield chemical shift change of backbone amide ¹⁵N CST values correlate with decreasing hydrogen bond lengths as reported in literature [13]. Among the two secondary structure types, for α-helices, the δ₁₁ component exhibits the highest sensitivity to the hydrogen bonding distance with R_P of –0.19. The δ₂₂ component is somewhat less sensitive with R_P of –0.14, and δ₃₃ has the smallest or almost no correlation (R_P equal to –0.04). This trend is in very good agreement with the assertion that the least-shielded component δ₁₁ acquires maximum sensitivity due to its collinear orientation with the N-H bond vector.

Interestingly, for β-strands, the δ₃₃ component, which lies in the peptide plane with almost perpendicular orientation to the N-H bond vector, has the maximum negative correlation with R_P of –0.25 while the δ₁₁ component is somewhat less sensitive with R_P equal to –0.16. The δ₂₂ component shows the weakest correlation with R_P equal to –0.05. The less negative correlation of δ₂₂ component with R_N…O for β-sheets is consistent with earlier reports that δ₂₂ component is shifted upfield or shows positive correlation with increasing hydrogen bond strength or decreasing R_N…O [11]. However, the decreased sensitivity of δ₁₁ and the increased sensitivity of δ₃₃ to hydrogen bond length in β -strands is somewhat surprising. It would be interesting to explore this further through quantum chemical calculations of the hydrogen-bonded backbone amide ¹⁵N chemical shift tensors as a function of the secondary structure type.

As shown in Figure 5, the magnitudes of the Pearson correlation coefficients are in the range of 0.04–0.19, which indicate only a modest correlation of backbone amide ¹⁵N CST with hydrogen bond length. Despite these modest values of R_P, a clear trend of δ_ii (ii = 11, 22, 33) linearly decreasing with R_N…O is evident for both α-helical and β-strand secondary structure types.

4. CONCLUSIONS

We have presented experimental SSNMR data and statistical analysis of site-specific backbone amide ¹⁵N CS tensor components of 100 residues of four different proteins, E. coli thioredoxin, DLC8 and CAP-Gly domain of the mammalian dynactin. Our results reveal that the least-shielded component δ ₁₁ of the backbone amide ¹⁵N is very sensitive to hydrogen bond interaction, displaying a significant downfield shift. The downfield chemical shift change of backbone amide nitrogen nuclei with increasing hydrogen bond strength is manifested in the negative correlation of the principal components with hydrogen bond distance for both secondary structure elements, α-helices and β -sheets. Our results are consistent with the theoretical studies of ¹⁵N shielding tensors on model compounds and short peptides. In order to improve this analysis further, one needs to record a larger database of ¹⁵N chemical shift tensors, which would encompass multiple proteins and different secondary structure types. These studies are currently ongoing in our laboratory.