1H detection and dynamic nuclear polarization-enhanced NMR of Aβ1-42 fibrils.

Several publications describing high-resolution structures of amyloid-β (Aβ) and other fibrils have demonstrated that magic-angle spinning (MAS) NMR spectroscopy is an ideal tool for studying amyloids at atomic resolution. Nonetheless, MAS NMR suffers from low sensitivity, requiring relatively large amounts of samples and extensive signal acquisition periods, which in turn limits the questions that can be addressed by atomic-level spectroscopic studies. Here, we show that these drawbacks are removed by utilizing two relatively recent additions to the repertoire of MAS NMR experiments-namely, 1H detection and dynamic nuclear polarization (DNP). We show resolved and sensitive two-dimensional (2D) and three-dimensional (3D) correlations obtained on 13C,15N-enriched, and fully protonated samples of M0Aβ1-42 fibrils by high-field 1H-detected NMR at 23.4 T and 18.8 T, and 13C-detected DNP MAS NMR at 18.8 T. These spectra enable nearly complete resonance assignment of the core of M0Aβ1-42 (K16-A42) using submilligram sample quantities, as well as the detection of numerous unambiguous internuclear proximities defining both the structure of the core and the arrangement of the different monomers. An estimate of the sensitivity of the two approaches indicates that the DNP experiments are currently ∼6.5 times more sensitive than 1H detection. These results suggest that 1H detection and DNP may be the spectroscopic approaches of choice for future studies of Aβ and other amyloid systems.

Amyloid fibrils are highly stable protein deposits found in β-sheet conformations and are notoriously recognized as disruptive agents to cellular function in over 40 human diseases (1, 2). Alzheimer’s disease (AD) is the most pervasive of all known plaque-related diseases and is associated with the presence of amyloid-β (Aβ) peptides in the extracellular space of the brain (3–6). As of 2021, there are ∼6.2 million people in the United States living with Alzheimer’s dementia and ∼50 million worldwide (7), and there is as of yet no cure available for AD. In order to address this epidemic, it is essential that we learn as much as possible about the formation and structure of Aβ plaques, including the detailed features of their catalytic surface, in order to design and develop appropriate treatments to limit the propagation of aggregates and the generation of toxic forms.

Aβ is derived from the C-terminal region of the amyloid precursor protein (APP), a membrane protein in neuronal cells, via proteolysis by β- and γ-secretase (8, 9). One of the principal challenges in rationalizing AD etiology is Aβ’s diversity in peptide length, mutations, and posttranslational modifications (10). Their low solubility renders solution NMR ineffective, and high-resolution diffraction analyses have thus far been restricted to shorter peptides with all or most residues being ordered in the fibril core structure (11). Cryogenic electron microscopy (cryo-EM) has made strides in resolution in fibril studies within the past decade (12–18), but faces challenges studying with atomic-level detail due to polymorphism and heterogeneity in the fibril macroassemblies. Studying the individual and collective roles of amyloids at atomic resolution therefore requires alternative, high-resolution, high-throughput techniques for structural analysis. Magic-angle spinning NMR (MAS-NMR) was introduced as a technique with the potential to address these problems (19, 20). Recent technical advances (21, 22) and progress in sample preparation (23) have vastly improved the sensitivity and resolution of the spectra (24). Accordingly, there are now publications describing high-resolution structures of Aβ (25–29) and other amyloid (12, 16, 30–35) fibrils based on distance and torsion angle constraints derived from MAS experiments.

To date, all of the known NMR structures of amyloid fibrils were determined using constraints obtained from 13C/15N MAS spectra, which are inhomogeneously broadened and therefore feature well-resolved lines at low spinning frequencies (<25 kHz) (36). However, resolution often remains insufficient for in-depth analysis, and the experiments require relatively large amounts of peptide and extensive signal acquisition periods. Two relatively recent additions to the repertoire of MAS NMR experiments—namely, 1H detection and dynamic nuclear polarization (DNP)—promise to circumvent these issues by reducing signal acquisition times or, alternatively, the amount of protein required for the experiment (37). 1H detection offers a factor of (γH/γS)3/2 gain in sensitivity, where S is usually a low γ-spin (38–40) such as 13C or 15N. In these two cases it is possible to achieve a factor of ∼8 or ∼32 gain in sensitivity, respectively. Importantly, 1H detection also introduces an additional spectral dimension and therefore significantly increases the resolution. In parallel, DNP offers a general approach to enhancing sensitivity by factors of ∼100, dramatically reducing signal acquisition times (by ∼104). It does so by exploiting the high spin polarization of unpaired electrons (of gyromagnetic ratio γe ∼660 times larger than γH) of a paramagnetic polarizing agent to enhance sensitivity by a theoretical factor of γe/γH. (41–44) Furthermore, DNP experiments are conducted at ∼100 K, thereby increasing the Boltzmann polarization and sensitivity by another factor of ∼3 over experiments conducted at ambient temperature (45).

While these arguments are well established for MAS NMR in many systems, it is less obvious that they are applicable to amyloid samples because spectra of amyloids are known to be broad for a variety of reasons, such as sample purity and polymorphism. Furthermore, 1H-detected NMR at moderate MAS frequencies (∼20 to 60 kHz) needs to be coupled to different levels of deuteration to ensure high sensitivity and narrow linewidths (42). Accordingly, deuteration with partial reprotonation of the amide or Hα sites has been implemented in pioneering studies on Aβ1-40 at 20 kHz MAS (46), HET-s(218–289) (47), and D76N-β2m at 60 kHz MAS (48). In addition, selective protonation in Aβ1-40 fibril methyl groups at 18 kHz MAS has led to highly resolved 1H-detected 13C correlations (49). In deuterated samples, however, the amount of potentially available structural information is significantly reduced, which can impair high-resolution structure determinations. The advent of 0.7 mm MAS rotors that achieve ωr/2π >110 kHz attenuates 1H-1H dipole couplings and allows direct acquisition of multidimensional 1H data without requiring deuteration (50). Furthermore, the spectra provide assignments and structural information. While a proof-of-concept application of this approach was demonstrated on fully protonated highly regular prion fibrils (51, 52), it is not clear whether this methodology is generally applicable and extendable to the detection of resolved inter- and intramolecular contacts in complex amyloid assemblies.

In parallel, our MAS DNP studies on M0Aβ1-42 (28, 32, 53) report significant broadening of the NMR lines at cryogenic temperatures, which was attributed to distributions of conformations trapped at low temperature and is therefore inhomogeneous in origin. The loss of resolution associated with the MAS DNP methodology is a major obstacle for the detailed structural study of uniformly labeled amyloid samples. Concurrently, reports of well-resolved spectra at high fields and spinning frequencies suggest that the broadening is homogeneous (54–56). The advent of DNP instrumentation operating at high magnetic fields (18.8 T) and faster MAS (ωr/2π = 40 kHz) provides an approach to alleviate this limitation by attenuating homogeneous couplings (57). However, this comes at the expense of the enhancement factor, potentially compromising the capacity to carry out expeditious multidimensional and multinuclear correlations. Moreover, NMR spectra of amyloid fibrils are known to suffer from additional debilitating broadening associated with their heterogeneous character (sample purity, polymorphism, etc.), which may mitigate the benefits of high magnetic fields.

In this work, we show that high resolution and sensitivity are possible for fibrils of M0-Aβ1-42. Notably, we demonstrate rapid resonance assignment and site-resolved detection of numerous site-specific internuclear proximities on submilligram sample quantities via 1H-detected NMR at ωr/2π ∼110 kHz and high field (23.4 T/1,000 MHz for 1H) at room temperature and 13C-detected DNP MAS NMR at ωr/2π = 40 kHz and high field (18.8 T/800 MHz for 1H) at low temperature. While both 1H detection and DNP afford increased sensitivity, we estimate, using approaches outlined by Ishii and Tycko (40), that DNP, with our current ε = 22, yields a factor of ∼6.5 higher sensitivity. These results therefore illuminate possible paths for the rapid structure elucidation of amyloid fibrils available in limited quantities.

Results and Discussion

Under physiological conditions (pH ∼7 to 8), M0-Aβ1-42 fibrils formed in vitro contain two filaments, and each filament is a C2 symmetric dimer, in which K16-A42 forms the fibril core while D1-H14 is disordered and likely solvent exposed (28, 29, 35). Twelve residues of each monomer are buried in the hydrophobic core, while the fibril surface has a mixed hydrophobic, hydrophilic, and charged character with the A42 C-terminal carboxyl group and K28 sidechain forming a so-called salt bridge. Recently, Wickramasinghe et al. identified a new form of Aβ1-42 using 1H detection on brain-seeded fibrils (58). Likewise, a very different structure of Aβ1-42 was published by Gremer et al. (15). However, these latter fibrils were grown at pH = 2 in 30% CH3CN, which likely accounts for the disparity. We can thus say that M0-Aβ1-42 fibrils that have been formed under physiological conditions and without brain plaque seeding have a reproducible structure and are therefore a well-characterized system useful to develop novel techniques using MAS NMR.

1H-Detected Aβ Spectra.

Fig. 1 shows two-dimensional (2D) 1H,15N and 1H,13C fingerprints of fully protonated U-13C,15N-labeled M0-Aβ1-42 fibrils recorded at ωr/2π = 110 kHz on a 1-GHz spectrometer. These spectra represent dipolar-driven correlations that reveal interactions between nuclei that have little or no dynamics (59). Consequently, our existing 13C/15N assignments for M0-Aβ1-42 fibrils (24) led us to quickly identify many Hα and HN resonances from the fibril core without the need for further correlations (60). These spectra feature both excellent resolution and sensitivity for all amide and aliphatic protons in the backbone and sidechains. Signals have linewidths of ∼0.2 to 0.3 ppm, which are larger than those observed under similar conditions on microcrystalline proteins (50) but nonetheless guarantee a good dispersion and limited overlap. Most importantly, the resonances feature very long coherence lifetimes (e.g., bulk T2s of ∼70 and 40 ms were measured for 15N and 13Cα, respectively). This allows the acquisition of two three-dimensional (3D) experiments for sequential backbone assignment based on the detection of HN protons, namely (H)CANH and (HCO)-CA-(CO)NH that correlate the intra- and interresidual CA to the HN-N pair (60), as well as 1H-detected (H)NCAHA and 13C-13C TOCSY-HSQC correlations for sidechain resonance assignments (51). These spectra required only ∼0.5 mg of sample and a few days of data acquisition to yield the complete identification of backbone and sidechain signals (∼92% completeness) within the fibril core, consistent with the existing 13C/15N assignments for M0-Aβ1-42 fibrils (28).

Fig. 1.

High-resolution 1H-detected spectra recorded on fully protonated U-13C,15N-labeled M0-Aβ1-42 at 111.111 kHz MAS. (Top) 2D 13C-1H CP-HSQC; (Bottom) 2D 15N-1H CP-HSQC.

At the same time, ωr/2π = 110 kHz MAS approaches the limit at which highly flexible regions appear in dipolar-based spectra. As a result, we observed that Gly37 and Ala42 exist in two distinct states, which are visible in the 2D CP-HSQC spectrum (in red in Fig. 1). These instances of polymorphism were not detectable in our previous experiments at 20 kHz MAS, suggesting that they derive from backbone flexibility in the C-terminal region. Likewise, several more resonances appear that have no apparent correlation to neighboring nuclei and likely arise from the N-terminal residues of the fibril or polymorphs of the fibril core (61).

Ultimately, 1H detection has the potential to provide direct access to 1H-1H distances for structure calculations, as shown in microcrystalline proteins (62), viral assemblies (50), and membrane-embedded proteins (63, 64). Here, we implemented radio frequency driven recoupling (RFDR) mixing into 3D HN and HC dipolar correlation schemes [(H)NHH or (H)CHH sequences], with representative slices from these datasets in Fig. 2. Even with shorter mixing times, RFDR transfers polarization between protons that are further apart than typical one- or two-bond 13C-13C distances that we typically achieve at moderate spinning frequencies. As a result, these spectra contain correlations between sequential or near-neighbor residues but also a number of unambiguous long-distance contacts up to 4 to 5 Å. These experiments can alternatively be performed as 13C-edited variants, to increase the spectral resolution through the indirect evolution of two carbon dimensions in the 3D 13C, 13C-edited 1H-1H RFDR experiment (Fig. 2). Although the additional pair of forward and backward H-C cross-polarization transfers causes a twofold drop in sensitivity, it allowed identification of a large number of unambiguous contacts listed in .

Fig. 2.

Representative restraints from the (A) 3D (H)NHH, (H)CHH and (B) (H)C(HH)CH spectra that show long-range 1H-1H contacts corresponding to the intramolecular structure of M0-Aβ1-42, highlighted in red. Asterisks mark diagonal peaks. (A) The NHH and CHH experiments establish contacts between F20H and I32γ12 and A30α, between A30α and F20ζ F19β3 and I42δ1, and from F19α to I32γ2/γ12/δ1. (B) The CCH experiments establish contacts between G33α and V36β and V36γ1 and between I41β and G29α. (C) Display of intramolecular contacts found in the spectra shown in Fig. 3 on the lowest energy NMR structure of M0Aβ1-42 (Protein Data Bank ID 5kk3) (28). The contacts illustrated in this figure are listed in the legend for Fig. 3.

Fig. 3.

(A) DNP enhanced 13C spectra of 1,3-13C2/2-13C/15N-labeled M0Aβ1-42 illustrating and enhancement of ε = 22 using (B) bis-nitroxide polarizing agent M-TinyPol. (C) DNP-enhanced 13C-13C CORD-RFDR spectra of 1,3-13C2/2-13C/15N-labeled M0Aβ1-42. The resolution in the spectrum is comparable to that obtained at ambient temperatures, ∼0.6 ppm as indicated for I31Cβ-Cα. In red we show the long-range contacts that correspond to the intramolecular monomer structure of the fibril. (D) DNP-enhanced NCO and NCA (Right) spectra acquired using M-TinyPol as the polarizing agent. ωr/2π = 40 kHz and T = 115 K.

Overall, while faster spinning would still be required to make clear and quantitative interpretation of these 1H-1H distance measurements, these contacts map the aliphatic sidechains within the hydrophobic “steric zipper,” some of which are illustrated in Fig. 2. One of the most important was to confirm the position of F19 and F20 on the inside of the β-strand as shown in Fig. 2. This conformation was observed in the 13C-13C spectra but is unusual, and an independent observation is important.

13C DNP Experiments.

A full determination of fibril architecture additionally requires measuring intermolecular interactions among different monomers, and the 3D spectra described above contain additional cross-peaks between pairs of residues at the interface (e.g., H K16–Mε M35 Hα L17–Hγ L34 or H L17–Mε M35). However, the repetitive structure of the assembly does not allow assigning the experimental cross-peak intensity as intermolecular restraints in an unambiguous fashion. Current protocols to address this problem rely on mixed samples obtained from differently isotopically labeled species (52, 65, 66). Such approaches have been successfully applied to describe interstrand arrangements and intermonomer contacts, but the necessary dilution of the labeled spins further reduces sensitivity and consequently requires even larger sample quantities. In these situations, MAS DNP at low temperature has a unique potential to complement 1H-detected NMR studies at room temperature.

To this end, we combined high-field DNP (800 MHz/527 GHz) with MAS in a 1.3-mm rotor, which permits spinning frequencies up to 40 kHz. In addition, we used a suitably designed water-soluble polarizing agent, M-TinyPol (67), and a selectively labeled 1:1 mixture of 1,3-13C-glycerol/U-15N- and 2-13C-glycerol, U-15N-enriched M0Aβ1-42. Here, the specific labeling scheme offers nearly the full range of labeled spins as that of a uniformly labeled sample but results in considerably sharper linewidths due to the elimination of the J couplings between neighboring 13C nuclei (28, 68–70). The resulting enhancement factor (ε = 22, Fig. 3) is one of the highest reported to date at this magnetic field on a biomolecular sample and reduces a week’s worth of experiment time to approximately one-third of an hour, enabling the rapid acquisition of 2D and 3D MAS NMR experiments with only ∼1 mg of sample. Furthermore, this modest enhancement yields a sensitivity ∼6.5 times larger than that available from 1H detection at room temperature. Accordingly, improvements in polarizing agents will increase this number. Details of the calculation of this estimate are included in .

Importantly, a 2D 13C-13C CORD-RFDR spectrum (Fig. 3) features resolved cross-peaks from the fibril core with linewidths of ∼0.6 ppm, on par with the best results reported for DNP (56, 71–74) and equal to the linewidths we observed in spectra acquired at room temperature. In the carbonyl region, a few long-range contacts that reflect both inter- and intramolecular contacts also appear in the carbonyl region (highlighted in red in Fig. 3). For example, I32-Cγ1/V18-CO and G37-Cα/V39-CO are consistent with the monomer structure, while L17-Cβ/M35-CO corresponds to the dimer interface. DNP enhanced NCO and NCA spectra (Fig. 3) also display very good resolution in the 15N dimension and feature all the signals from residues in the 15 to 22 range. Notably, the intensity is lower for residues 22 to 25, corresponding to a loop, which is more dynamical at room temperature. This is consistent with a broader distribution of shifts corresponding to the freezing of multiple conformations. The spectra additionally reveal several peaks that were not present in the equivalent experiment at room temperature (24). These likely arise because cryogenic conditions required by DNP quench any dynamics that would otherwise interfere with the 1H decoupling and thus attenuate peak intensities (75, 76). Concurrently, the microwave enhancement could potentially bring out minor polymorphs as well as the otherwise disordered or dynamic N-terminal region, opening new windows to study the extent of the fibrils’ heterogeneity.

The excellent resolution and sensitivity suggest that resolved site-specific contacts could be monitored by combining these 2D datasets in 3D experiments. A pair of 3D NCACX and NCOCX spectra can be acquired by appending a 13C-13C CORD-RFDR mixing to NCO and NCA modules. Fig. 4 shows for the fragment I31-V36 how the joint analysis of these spectra yields both sequential linking of the resonances and contacts between 13CO and 13Cα/13Cβ across the interface of two stacked molecules. On the one hand, this demonstrates the possibility to perform an independent backbone sequential assignment of 13C and 15N resonances directly under cryogenic conditions, extending a strategy proposed by Sergeyev et al. (74). Notably, joint analysis of the NCACX and NCOCX pair allows us to associate each 15N frequency with those of the 13Cαs of both the same and the previous residue. This is illustrated in Fig. 4, where residues I31-I32-G33 and L34-M35-V36 feature a clear sequential connectivity along the backbone. On the other hand, given the mixed labeling, we can easily highlight long-range contacts that reflect intermolecular proximities. For example, among the strips displayed in Fig. 4, we observe CA/CO correlations for G33, L34, and V36, which is consistent with the parallel-in-register architecture. The same strategy also allows us to detect unambiguous correlations such as, e.g., L17-Cβ/M35-CO, corresponding to the dimer interface. Overall, these DNP experiments show that combining high-field, fast-spinning, up-to-date biradical polarizing agents, and specific labeling schemes, produces a wealth of structural information in amyloid fibrils not available in other experiments. These data advantageously complement the capacity of the 1H-detected experiments described earlier.

Fig. 4.

Selected cross-sections from DNP-enhanced NCOCX (blue) and NCACX (red) spectra of 1,3-13C2/2-13C/15N-labeled M0-Aβ1-42. Green circles indicate signals that reflect the interdimer contacts and the parallel-in-register arrangement of the fibrils. Asterisks mark diagonal peaks.

Conclusions

The possibility of directly recording and assigning resolved resonances throughout the backbone and sidechains is key for obtaining a large set of contacts in amyloid fibrils. Here we have shown that this information can be accessed using only submilligram sample amounts and without the need for deuteration from high-field 1H-detected NMR with 110 kHz MAS and using high-field MAS DNP. Our results demonstrate the significant benefits of these methodologies in the context of amyloid studies and suggest the potential to perform high-throughput analyses of different amyloid constructs without requiring preparation of large quantities of protein or even to study unlabeled peptides derived from tissue samples available in limited quantities.

Additionally, the high-spectral resolution provided by high-spinning frequencies and large signal enhancements reported here lays the groundwork for implementing 1H detection and DNP in a combined fashion for amyloid research in the near future. The ultimate purpose of our exploration is to determine whether combining 1H detection and high-field DNP may be feasible for studying fibrils. Together with the ε ∼22 enhancement from M-TinyPol, 1H-detected DNP at low temperature may bring a factor of several thousandfold increase in sensitivity, which will likely improve with further development of heterobiradical polarizing agents. Indeed, 1H-detected DNP experiments have recently been accomplished on histidine at 65 kHz MAS and 100 K at high field (77). Extending this work to amyloid fibrils at natural abundance is thus a challenge that is within reach, and these possibilities will enhance our understanding of molecular events underlying Alzheimer’s disease and improve the perspectives for therapeutic invention.

Materials and Methods

Sample Preparation.

The biosynthetic preparation of U-13C/15N M0Aβ1-42 fibril samples was accomplished by expression of the M0Aβ1-42 peptide in Escherichia coli in M9 minimal medium supplemented with 15N-ammonium chloride and 13C6-glucose and purification of the peptide from inclusion bodies using ion exchange and size exclusion chromatography, followed by a second size exclusion step to isolate pure monomer prior to spontaneous fibril formation under quiescent conditions. This procedure was used previously (24, 28, 35) and is adapted from Walsh et al. (23). The 1,3/2-13C glycerol/15N mixture used in the DNP experiments was prepared by expressing the peptide in E. coli in two different M9 media, one containing 1,3-13C2 glycerol and the other with 2-13C glycerol. The peptides were purified independently but then combined as monomers after the final size exclusion step, using the absorbance of the collected monomer peak to determine the concentration of each sample before preparing a 1:1 mixture of the two 13C-labeled variants. Fibrils were formed by incubating the mixture under quiescent conditions, at pH = 8.

1H Detection Experiments.

For the 1H-detected experiments, U-13C,15N-M0Aβ1-42-labeled fibrils were resuspended in water and directly centrifuged at 165,000 × g for 15 h at 12 °C into the NMR rotor using a device provided by Giotto Biotech, similar to those described in the literature (78, 79).

All spectra were acquired at the static field of 23.5 T (1,000 MHz 1H) with a three-channel (HCN) 0.7-mm Bruker MAS probe and at the static field of 18.8 T (800 MHz 1H) with a four-channel (HCND) 0.7-mm Bruker MAS probe. The sample was spun at 111.111 kHz, and we maintained the temperature using a Bruker cooling unit (BCU III) with a regulated dry N2 gas directed at the rotor. The temperature detected by the sensor (260 K) at the point where bearing, drive, and variable temperature gases are mixed, corresponds to roughly 285 ± 5 K inside the 0.7-mm rotor. On both probes, high-power pulse durations were 2.5, 3.5, and 4.5 μs for 1H, 13C, and 15N, respectively. 1H, 13C, and 15N chemical shift data are available in the Biological Magnetic Resonance Data Bank.

The pulse sequences follow, with no modifications, those described in Barbet-Massin et al. (60), Andreas et al. (50) and Stanek et al. (51). For the dipolar-based 15N,1H and 13C,1H CP-HSQC experiments, 1H-15N and 1H-13C CP transfers were optimized around nutation frequencies of 5/4 ωr and 1/4 ωr, respectively, for proton and 15N (or 13C), with a 10% linear ramp applied on the 1H channel. For 3D CANH and (H)(CO)CA(CO)NH (80), 13C-15N CP was performed with a 10% tangent ramp applied on the 15N frequency at 2/5 ωr, while the 13C nutation frequency was kept around 3/5 ωr. The H(C)CH and (H)CCH TOCSY sequences were implemented with composite 13C pulses applied with a low nutation frequency of one-fourth the MAS frequency (∼27.5 kHz).

In 3D (H)NHH, (H)CHH, and (H)C(HH)CH spectra, 1H-1H RFDR recoupling (81–83) was applied after the first back-CP at a 1H RF frequency of 100 kHz, yielding 1H-1H contacts resolved using the shift of 15N or aliphatic 13C.

In all experiments, low power WALTZ-16 decoupling of 10 kHz was employed for heteronuclear decoupling, while 13C decoupling during acquisition was performed with DIPSI-2 irradiation of γB1/2π = 20 kHz. The MISSISSIPPI scheme (84) without the homospoil gradient was used for 100 ms in order to suppress the water signal.

Spectra were apodized in each dimension with 60° shifted squared sine bells (“qsine 3” in Bruker TopSpin), and zero filled to at least twice the number of points in the indirect dimensions. Where linewidths are reported, no apodization was applied for the reported frequency. Acquisition parameters specific for each spectrum can be found in .

Experimental Parameters for DNP.

For the DNP MAS experiments, we resuspended M0Aβ1-42 fibrils prepared from a mixture of 1,3-13C-glycerol and 2-13C-glycerol–labeled samples in 25 μL of DNP juice (d8-glycerol/D2O/H2O 60/30/10) containing 10 mM of M-TinyPol. The suspension was left in the refrigerator overnight and packed into a 1.3-mm ZrO2 rotor the following morning by centrifugation for 3 h at 120,000 × g and 4 °C.

DNP MAS NMR spectra were recorded on a Bruker Avance III wide bore spectrometer, operating at 18.8 T and equipped with a triple resonance H/X/Y 1.3-mm low-temperature MAS probe. The sample was irradiated with high-power microwaves at a frequency of 527 GHz generated by a gyrotron that was operating continuously during the DNP experiments (stability of better than ±1%). A microwave power of 22 W was used (measured at the bottom of the probe), with a field position we adjusted for optimal cross-effect. The MAS rate was 40 kHz. Sample temperature, estimated from a sensor placed inside the stator that measures the temperature of gas flows around the rotor, was about 115 K (the variable temperature unit was 95.0 K, bearing 94.5 K, drive 127.8 K). High-power pulse durations were 2.5, 3, and 6 μs for 1H, 13C, and 15N, respectively. For all experiments, 1H-15N and 1H-13C CP transfers were optimized around nutation frequencies of 100 kHz for proton and 60 kHz for 15N or 13C, with a 50% linear ramp applied on the 1H channel. CP transfers between 13C and 15N were performed with a 10% tangent ramp applied on the 15N frequency at 15 kHz, while the 13C nutation frequency was kept around 25 kHz. The 13C-13C CORD-RFDR experiment was recorded as described by Hou et al. (85), with a mixing time of 104 ms. In the NCOCX and NCACX experiments, we employed the CORD-RFDR mixing scheme for an identical duration.

Spectra were apodized in each dimension with 60° shifted squared sine bells (qsine 3 in Bruker TopSpin) and zero filled to at least twice the number of points in the indirect dimensions. Details about the used acquisition and processing parameters are given in below.