A 3D MAS NMR experiment utilizing through-space ¹⁵N-¹⁵N correlations

Abstract

We demonstrate a novel 3D NNC magic angle spinning (MAS) NMR experiment that generates ¹⁵N-¹⁵N internuclear contacts in protein systems using an optimized ¹⁵N-¹⁵N proton assisted recoupling (PAR) mixing period and a ¹³C dimension for improved resolution. The optimized PAR condition permits the acquisition of high signal-to-noise 3D data that enables backbone chemical shift assignments using a strategy that is complementary to current schemes. The spectra can also provide distance constraints. The utility of the experiment is demonstrated on an M₀Ab_1–42 fibril sample that yields high-quality data that is readily assigned and interpreted. The 3D NNC experiment therefore provides a powerful platform for solid-state protein studies and is broadly applicable to a variety of systems and experimental conditions.

Graphical Abstract

Protein structure determination using magic angle spinning (MAS) NMR spectroscopy invariably begins with chemical shift assignments, where each distinct resonance is associated with a specific nuclear site. Resonance assignment is typically accomplished using 2D and 3D experiments where magnetization is transferred among ¹H/¹³C/¹⁵N nuclei using dipole recoupling experiments designed to reintroduce the coupling attenuated by MAS. While this initial step is of crucial importance, it is often tedious and challenging. Homonuclear dipolar recoupling in 2D and 3D assignment experiments predominantly utilizes ¹³C-¹³C magnetization transfer, due to the prevalence of ¹³C nuclei in biological samples, and to a favorable gyromagnetic ratio of ¹³C. In principle, it is also possible to use ¹⁵N-¹⁵N couplings for assignments and structural studies, an approach that is appealing because of the excellent resolution of the ¹⁵N dimension. Accordingly, previous approaches employed proton driven spin diffusion (PDSD) techniques to generate ¹⁵N-¹⁵N correlation spectra of three model peptides^1–2, three model proteins^3–5, and more recently, oriented membranes⁶. Additionally, a ¹⁵N-¹⁵N spectrum obtained using an early version of proton assisted recoupling (PAR) was reported⁷. Despite the fact that these initial results are very promising, ¹⁵N-¹⁵N techniques have not evolved to be part of the standard repertoire of MAS protein experiments, primarily for three reasons: (1) the sensitivity in the PDSD and the initial PAR spectra is low and does not permit extension to higher dimensions; (2) cross peaks for long-rang contacts are often missing from the spectra; and (3) the mixing times are 2–5 s and therefore the total experimental time is long. Here we describe a novel 3D NNC experiment (depicted in Figure 1A) that utilizes an optimized ¹⁵N-¹⁵N PAR protocol that yields excellent ¹⁵N sensitivity, and therefore easily extends to a third directly detected ¹³C dimension to further increase the resolution. Furthermore, the PAR mixing sequence is short (20 ms) and the spectra display an abundance of cross peaks.

Figure 1:

The 3D NNC pulse sequence is shown in A) with phase cycling: ϕ₁ = 02, ϕ₂ = 0022, ϕ₃ = 1, ϕ₄ = 2002, ϕ₅ = 0123, ϕ₆ = 2002 0220, ϕ_rec = 0123 2301. The CP, PAR, and SPECIFIC-CP blocks are color coded to arrows in (B) illustrating the corresponding magnetization transfer. Additionally, B) contrasts the mode of mixing for the NNC experiment, and commonly used protocols (NCACX, NCOCX, CONCA), illustrating their complementary information due to reliance on separate contacts. C) visualizes the internuclear distances and corresponding dipole coupling values for a range of ¹⁵N-¹⁵N contacts in α-helices and parallel and anti-parallel β-sheets. The residue interval label (i.e. i ± 1, i ± 2, etc.) is color coded to the lines/arrows illustrating transfer. The ¹⁵N-¹⁵N inter-strand β-sheet distances are given as approximate averages.

As mentioned above, 3D MAS experiments usually utilize ¹³C-¹³C correlations (generated using RFDR^8–9, DARR¹⁰, PDSD^11–12, or PAR¹³ mixing, amongst others) that are resolved with a ¹⁵N dimension. The transfer from ¹⁵N to ¹³C is typically based on SPECIFIC-CP^14–16 or other ¹⁵N-¹³C transfer techniques^17–19. This combination leads to a suite of experiments used for backbone resonance assignments that includes a combination of NCA/NCACX/NCACB^20–25, NCO/NCOCX^21–25, and CONCA/CANCO^23–25. The 3D NNC, however, allows for the acquisition of non-redundant and complementary information that is otherwise inaccessible and thus facilitates comprehensive assignment strategies when combined with other data sets. This unique sequence shares features with previous ¹⁵N-¹⁵N correlation experiments in liquids^26–27 as well as solid, deuterated samples at high MAS frequencies²⁸. Additionally, direct ¹⁵N detection has recently been used for proteins in solution experiments²⁹. However, the NNC is distinguished from the above experiments since it utilizes through-space ¹⁵N-¹⁵N dipole coupling to mediate coherence transfer and generate homonuclear contacts. However, ¹⁵N homonuclear dipolar recoupling remains difficult to implement in non-model samples such as M₀Ab_1–42 (as opposed to N-f-MLF-OH³⁰ or GB1³¹), the 42 amino acid protein that forms toxic fibril species associated with Alzheimer’s disease^32–46. While this system is well suited for solid state NMR analysis because of its monodispersity and small molecular size (enabling high sensitivity), it presents many difficulties for collecting high-quality data as it is significantly more structurally heterogeneous, and less static than any of the usual model systems. It is for this reason that we have chosen M₀Ab_1–42 to test the efficiency of the NNC protocol.

To demonstrate the importance of PAR mixing we compared the signal-to-noise ratio, the number of observed cross peaks, and maximum distance observed for three available ¹⁵N-¹⁵N homonuclear mixing schemes: RFDR (t_mix = 12.8 ms), PDSD (t_mix = 4 s), and PAR (t_mix = 20 ms). These results are illustrated in Figure 2. We found that ¹⁵N-¹⁵N mixing with the first-order recouping sequence RFDR^8–9 (Figure 2A), yields no cross-peaks due to the small homonuclear dipolar couplings <50 Hz (Figure 1C). Using ¹⁵N-¹⁵N PDSD^11–12, a second-order recoupling sequence, we observed a limited number of cross peaks (Figure 2B). However, in addition to the minimal quantity of cross-peaks, this experiment requires t_mix = 4 s, and all the cross-peaks are assigned to distances <3.5 Å (i±1). In contrast, the ¹⁵N-¹⁵N PAR experiment, also a second-order recoupling sequence, shows significantly more cross-peaks than the PDSD, and distances of up to 6.8 Å are observed. These are apparent in Figure 2C. Figure 2D compares 1D traces from the RFDR, PDSD, and PAR spectra, clearly showing the superior signal intensity given by PAR, and the presence of additional cross peaks that are attributed to long-range contacts (which are not present in either the RFDR or PDSD spectra). It is also important to note that the total acquisition time for the 2D PAR in Figure 3 was around 21 h, while the PDSD required 39 h, nearly twice as long.

Figure 2:

2D ¹⁵N-¹⁵N homonuclear correlation spectra on M₀Aβ_1–42. A) gives a comparison of 1D traces extracted from the 2D spectra in B)-D) as indicated by a gray line, with a comparison of measured signal-to-noise (S/N) and total experiment times. Panels B)-D) show a τ_mix = 12.8 ms ¹⁵N-¹⁵N RFDR (green), a τ_mix = 20 ms ¹⁵N-¹⁵N PAR (red), and a τ_mix = 4 s ¹⁵N-¹⁵N PDSD (blue), respectively. Cross-peak assignments are shown.

Figure 3:

3D NNC^α projections of M₀Aβ_1–42. One CN projection (CN₁) shows one peak per residue correlating the nitrogen and the C^α of the same amino acid i. The second CN projection (CN₂) is identical to CN₁ plus additional peaks due to NN-mixing. For each C^α/N pair (which forms the diagonal peak) there are peaks that correspond to nearby ¹⁵N sites, most prominently the i±1 backbone amide ¹⁵N nuclei. The third projection (NN) is identical to a 2D ¹⁵N-¹⁵N PAR spectrum showing the backbone nitrogen of residue i as diagonal and other peaks (mostly backbone nitrogen of residues i±1) as cross-peaks. The 3D spectrum was recorded on a Bruker Avance III 800 MHz spectrometer at ω_r/2π = 20 kHz with temperature set to 277 K. A more detailed assignment is shown in Figure S1.

The 2D ¹⁵N-¹⁵N spectra utilize an optimum PAR matching condition, enabling this innovative approach to backbone resonance assignments on disease-relevant M₀Ab_1–42. The optimum condition was obtained using a high-throughput protocol to evaluate magnetization transfer across an array of conditions, as described in detail in the Supporting Information. While the PAR mixing period in the NNC experiment generates ¹⁵N-¹⁵N contacts that are both short-range (between neighboring residues) and long-range (between non-neighboring residues), only the short-range contacts are useful for sequential assignment. Thus, the PAR mixing time can be adjusted to maximize short-range over long-range contacts. Long-range contacts are undesirable for sequential assignment purposes as they may augment spectral congestion and/or obfuscate the assignment process. However, long-range contacts are essential for structural characterization, and can be extracted from NNC spectra collected at longer mixing times. The NNC spectra shown here were collected with a PAR t_mix=20 ms, which yields predominantly short-range contacts in M₀Ab_1–42, with only minor long-range contacts that are easily distinguished by significantly weaker intensities.

The NNC pulse sequence (Figure 1A) consists of three transfer steps: ¹H-¹⁵N CP, ¹⁵N-¹⁵N PAR mixing, and ¹⁵N-¹³C SPECIFIC-CP, where the magnetization is selectively transferred to either the Ca or the CO nuclei to generate either an NNC^a (see Figures 2 and 3) or an NNC^O (see Figures S2 and S3) spectrum. Transfer efficiencies for the ¹⁵N-¹⁵N PAR mixing alone (i.e. in comparison to ¹H-¹⁵N CP and ¹⁵N-¹³C SPECIFIC-CP without ¹⁵N-¹⁵N PAR mixing) are shown in Figure S4 for N-f-MLF-OH and M₀Ab_1–42. The remaining ¹⁵N magnetization after ¹⁵N-¹⁵N PAR mixing depends on the mixing time and relaxation rates. After 20 ms of PAR mixing, 53% and 20% of the magnetization remain for N-f-MLF-OH and M₀Ab_1–42, respectively. Besides the ¹⁵N-¹⁵N PAR mixing, the total efficiency of the 3D NNC experiment depends on the heteronuclear transfer, which is coupled to the selectivity and also the sample and the experimental conditions chosen (vide infra).

A comparison of NNC^a and NNC^O for M₀Ab_1–42 shows the NNC^a spectrum to have superior resolution due to larger chemical shift dispersion in the C^a spectral region than the CO spectral region. However, the diagonal of the NNC^O spectrum contains additional information as it displays correlations between the ¹⁵N of residue i and the ¹³CO of residue i-1, whereas the diagonal of the NNC^a shows correlations only between nuclei within the same residue Figure 1B illustrates the internuclear correlations obtained in an NNC experiment vs. commonly used ¹³C-¹³C based experiments, and Figure 1C shows internuclear distances (and corresponding dipole coupling values) in standard α-helix as well as parallel and anti-parallel β-sheet structures for the ¹⁵N-¹⁵N contacts observed with the 3D NNC sequence.

A distinct advantage of the NNC experiment is the increased resolution gained from the inclusion of a 3^rd (¹³C) dimension. The 2D ¹⁵N-¹⁵N PAR spectrum shown in Figure 3 displays efficient magnetization transfer, but limited resolution. However, both the 3D NNC^a as well as the 3D NNC^O spectra have immensely improved resolution over their 2D counterparts.

Figure 3 displays 2D projections for each of the three unique faces of the NNC cube. The CN₁ projection is virtually identical to an NCA experiment²² or the C^a region of a short-mixing ZF-TEDOR experiment⁴⁷. Hence, CN₁ shows one peak per residue correlating the nitrogen and the C^a of the same amino acid. The CN₂ projection is identical to CN₁ plus additional peaks due to ¹⁵N-¹⁵N mixing. For each C^a/N pair (which forms the diagonal peak) there are peaks that correspond to nearby ¹⁵N sites, most prominently the i±1 backbone amide ¹⁵N nuclei. The third projection (NN) is identical to a 2D ¹⁵N-¹⁵N PAR spectrum showing the backbone nitrogen of residue i as a diagonal peak, as well as other coherences (mostly backbone nitrogen of residues i±1) as cross-peaks. The correlations between these three projections and the information therein is shown as example for residue A30 of M₀Ab_1–42 in Figure 2 and in greater detail in Figure S1.

For resonance assignment, each diagonal peak can be arranged in a strip plot (Figure 4) that displays the sequential walk. Each strip shows the diagonal peak (blue) and one or more cross-peaks (red), unless overlapping with other peaks. Typically, the strongest cross-peaks are of i±1 residues. Occasionally, i±2 residues can be present (as discussed earlier), for example between N27 and G29 (6.6 Å, D_NN=4.3 Hz), I31 and G33 (6.8 Å, D_NN=3.93 Hz), as well as G33 and M35 (5.8 Å, D_NN=6.3 Hz).

Figure 4:

Strip plot of the 3D NNC^α of M₀Aβ_1–42. Each strip displays the diagonal peak of residue i (blue) and several cross-peaks (red). The most prominent cross-peaks at 20 ms PAR mixing are from residues i±1, although cross-peaks from residues i±2 can be present (*). The sequential backbone walk is indicated by dashed lines. Colors do not indicate sign and are solely used as visual aids.

In summary, the 3D NNC experiment presented here provides a novel and compelling approach to protein backbone assignments by generating unique data not otherwise accessible. Typically, ¹³C-¹³C mixing schemes are used to correlate consecutive residues in proteins while using ¹⁵N to resolve spectral crowding. The 3D NNC experiment, however, generates ¹⁵N-¹⁵N correlations and uses ¹³C for spectral decongestion. Application of the PAR mixing scheme coupled with a careful optimization of the PAR fields provides efficient ¹⁵N homonuclear magnetization transfer, enabling high signal-to-noise data in 3D settings. While this initial demonstration serves as a showcase of the NNC experimental protocol, variations are readily envisaged with different heteronuclear transfer techniques to accomplish the ¹⁵N-¹³C transfer including adiabatic passage⁴⁸, Lee-Goldburg decoupling⁴⁹, TEDOR¹⁸, PAIN-CP¹⁹, and RESPIRATION-CP¹⁷, amongst others. These may offer improved selectivity and/or efficiency depending on the sample and the experimental conditions (i.e. spinning frequency and magnetic field). Furthermore, the NNC sequence naturally lends itself to other experiments of higher dimensionality that include additional transfer steps (e.g. a 4D NNCC sequence) and ¹H detection. As experimental NMR transitions to higher magnetic fields and faster sample spinning, we expect the NNC protocol to have a continued relevance as PAR has previously shown excellent functionality at higher spinning frequencies (i.e. 65 kHz)⁵⁰. Finally, while the NNC experiment was developed for sequential assignments, we anticipate that it will provide long-range ¹⁵N-¹⁵N contacts (using longer mixing times) that are highly valuable for structural characterization. We foresee this approach to be broadly applicable to a large variety of samples including protein crystals and fibrils, nucleic acids, as well as membrane proteins and sedimented proteins.