The hyperpolarized HDO-based buffers that are at the heart of this experimental strategy were produced using D-DNP systems operating at ~ 1.2 K and magnetic fields of either 3.35 T (Oxford HyperSense™) or 6.7 T (Bruker BioSpin). The general procedure is as follows (for details see the Experimental section): (i) Hyperpolarization of a frozen H2O solution containing a paramagnetic polarization agent and a cryo-protectant is achieved by means of off-resonance microwave irradiation. (ii) After building up the polarization for 2–3 h, the hyperpolarized H2O is dissolved with a burst of superheated neat or buffered D2O to produce a mixture containing 1–4% HDO (depending on the system and experimental parameters), which is then injected in 1–3 s into an NMR tube waiting in a high-field detection spectrometer, where it mixes in-situ with the target protein solution. (iii) Chemical and ‘magnetic’ exchange processes (transient Overhauser effects) between hyperpolarized HDO and the protein transfer 1H-hyperpolarization to the protein, thereby selectively enhancing NMR signals of residues with favorable solvent interactions (Kadeřávek et al. 2018; Szekely et al. 2018; Kurzbach et al. 2017; Olsen et al. 2016; Kim et al. 2017).
Signal intensities for all experiments are tabulated in the Supplementary Information. In most cases, even after extensive signal averaging, the signals in thermal equilibrium spectra for the D-DNP samples were too weak for a reliable determination of the enhancement factor ‘ε’ (often defined as the ratio between per-scan signal amplitudes in the hyperpolarized spectrum and the corresponding spectrum acquired after return of the system to thermal equilibrium). Therefore, to quantify the signal intensities obtained with hyperpolarized HDO, we instead report signal-to-noise ratios (SNR) for the hyperpolarized spectra. (As an approximate lower bound for ε, one can thus assume that ε ≥ SNR for instances where the corresponding reference signal remains below the limit of detection).
Hyperpolarized high-resolution 13C-detected 2D NMR
In previous work, we demonstrated how solvent hyperpolarization can be combined with rapid 2D 1H-15N HMQC spectroscopy (Schanda et al. 2005) of folded proteins and small IDPs (Kadeřávek et al. 2018; Szekely et al. 2018; Kurzbach et al. 2017; Olsen et al. 2016; Kim et al. 2017). However, analysis of proton-detected hyperpolarized 2D spectra of larger proteins or IDPs is sometimes challenging, as signal overlap and broadening due to exchange, radiation damping, and paramagnetic relaxation may hinder spectral analysis. To address these limitations, we expand our approach here to include 13C-detected 13C-15N correlation experiments (HN-CON, see pulse sequence in Fig. S1) (Gil et al. 2013; Bertini et al. 2011). In these experiments, chemical and magnetic exchange transfer proton hyperpolarization from HDO to protein backbone amide (and side-chain) sites. The amide hyperpolarization is then transferred by selective INEPT to the neighboring 15N nuclei, and, after an 15N evolution period, transferred onward to the adjacent 13CO spins for detection (Gil et al. 2013). The amide proton polarization is thus continuously replenished in successive scans by the hyperpolarized HDO pool, while 1H-detection is avoided.
Direct 13C detection offers the added advantage that the chemical shift dispersion in the 13C dimension is much larger compared to 1H, while at the same time the penalties associated with 1H detection are reduced.
Application to a folded protein: Ubiquitin
A hyperpolarized HN-CON spectrum obtained for ubiquitin via this strategy is shown in Fig. 1a (red), overlaid with a conventional HN-CON spectrum of ubiquitin obtained at thermal equilibrium in a 90:10% H2O:D2O phosphate buffered solution (blue). The hyperpolarized spectrum was acquired in 69 s at 37 °C. Both spectra were obtained in a magnetic field of 14.1 T (600 MHz).
Only if chemical and magnetic exchange from HDO is sufficiently fast will enhanced signals be observed (Kadeřávek et al. 2018; Szekely et al. 2018; Kurzbach et al. 2017; Olsen et al. 2016; Kim et al. 2017; Nucci et al. 2011; Otting 1997; Modig et al. 2004). In particular, the signal enhancementfor an amide site should be increased by faster exchange rates on the one hand, as replenishment of hyperpolarized amide protons between successive scans is improved, while on the other hand the improvement will be attenuated by losses due to proton exchange and relaxation during evolution delays and detection. Thus, as expected, the hyperpolarized spectrum contains a reduced subset of the signals seen in the corresponding conventional spectrum. Figure 1b maps the observed signals onto the crystal structure of ubiquitin, revealing that some residues in the hydrophobic core are also detected, indicating that hyperpolarization may in some cases be transferred from the solvent-exposed surface to the protein’s inner core.
The set of peaks observed depends on many variables, in particular on the overall SNR characteristic of the experiment in question, the delay between successive detections, and as noted above, the hyperpolarization transfer efficiency through proton exchange, nuclear Overhauser effects (NOE), or exchange-relayed NOE. While the latter three polarization transfer pathways do not critically depend on the pulse sequence or nucleus used for detection, the SNR of course does. However, if the underlying HYPEX/HyperW exchange and NOE polarization transfer mechanisms are operating as in our earlier proton-detected work (Kadeřávek et al. 2018), one should anticipate a qualitative agreement between the set of residues observed in the hyperpolarized proton-detected 1H/15N HMQC spectrum reported there and in the 13C-detected spectrum we obtain here. And indeed, the HN-CON results confirm these expectations, as all peaks observed with HN-CON detection were also observed in the hyperpolarized HMQC. Note that in the HN-CON, we do not see all of the peaks seen in the HMQC spectrum; we see instead only the set of peaks which showed the highest SNR in the proton-detected experiment. The reduction in the number of detected residues is due to several factors, such as the additional coherence transfer steps in the HN-CON experiments and the intrinsically lower sensitivity of 13C detection.
Moreover, as seen in Fig. 1a, despite the rapid injections and brief experimental durations used, the linewidths in the directly-detected dimension of the hyperpolarized HN-CON (10–15 Hz) are comparable to or narrower than those in its conventional counterpart, and approximately half of those obtained previously using 1H-detection (25–30 Hz in hyperpolarized 1H/15N HMQC) (Kadeřávek et al. 2018), indicating that 13C-detection reduces radiation damping and/or decoherence due to rapid proton exchange and paramagnetic relaxation – factors that limited resolution in earlier proton-detected hyperpolarized studies of protein systems (Kurzbach et al. 2017).
Application to an IDP: Osteopontin and its complex with heparin
Figure 2a displays a hyperpolarized 13C-15N spectrum of a 220-residue truncation mutant of OPN. Figure 2b shows a thermal equilibrium reference spectrum collected in a 10% deuterated buffer, and Fig. 2c shows a comparison of the hyperpolarized and reference spectra. The hyperpolarized spectrum was obtained in 55 s at 37 °C and pH 7.4 at 14.1 T (600 MHz for 1H). As for ubiquitin, the hyperpolarized and conventional HN-CON spectra show nearly identical linewidths. In contrast to the folded ubiquitin, however, a markedly larger fraction of the peaks observed in the conventional spectrum of OPN are also present in the hyperpolarized spectrum. This is not surprising, given the greater solvent exposure expected for an IDP, which would be consistent with particularly efficient chemical exchange and polarization transfer.
A strikingly different spectral response is observed when OPN is monitored after mixing with its ligand heparin (Clemente 2016). The heparin interaction site spans OPN residues 100 to 180. Upon binding of the large 17 kDa ligand, much of the IDP is in direct contact with heparin, and thus likely becomes somewhat shielded from the solvent. Indeed, when probed by hyperpolarized HN-CON after addition of heparin, the number of detected OPN peaks drops significantly, and the subset of observed residues changes markedly as well (Fig. 2d).
Due to signal overlap, numerous peaks observed in both the conventional and hyperpolarized spectra of OPN cannot be assigned with 2D HMQC or HN-CON data alone, making it difficult to determine the true spectral ‘binding footprint’ for this system. However, the fact that the pronounced drop in observed signals is not restricted to the binding site suggests that screening by the ligand alone is not the only determinant of changes in polarization transfer efficiency in the complex. Complex formation would also be expected to affect solvent exposure and polarization transfer of newly-sequestered residues substantially, perhaps resulting in a sparse enhancement pattern not unlike that seen for ubiquitin.
Hyperpolarized three-dimensional NMR
While the improvements in resolution provided by the hyperpolarized HN-CON relative to hyperpolarized 1H-15N HMQC experiments can be substantial, particularly for samples such as OPN (Mateos et al. 2019), in many instances 2D NMR cannot reduce signal overlap sufficiently to permit residue-specific analysis. To further reduce spectral crowding and to widen the scope of our approach, we therefore implemented hyperpolarized 3D 1H-13C-15N correlation experiments—in particular, the BEST-HNCO experiment shown in Fig. S2 (Lescop et al. 2007).
Methodological considerations
Given a final solvent deuteration level of 96–98% after dissolution and mixing, typical proton hyperpolarization lifetimes in these experiments are less than 2 min (T1 (1H) ≈ 20 s). To permit detection of two indirectly encoded dimensions under this time constraint, the use of non-uniform sampling (NUS) (Zawadzka-Kazimierczuk et al. 2012; Mayzel et al. 2014) is mandatory. In the present case, after empirical optimization, 9% NUS was chosen to provide a compromise between spectral resolution along ω1 (13C) and ω2 (15N), while allowing us to complete the 3D acquisitions within the time restriction of the HDO polarization lifetime. Poisson gap sampling (Hyberts et al. 2012) was used to generate NUS schedules, and a total of 96 FIDs were recorded in 65–75 s for a nominal digital resolution of 44 × 89 × 352 Hz in ω3, ω2 and ω1 respectively. The hyperpolarized HNCO spectra presented here were acquired in 65 s (OPN + Heparin) or 75 s (ubiquitin, OPN) at 18.8 T, in 96% deuterated buffer, at pH 7.4 and 37 °C.
The full 3D spectra were reconstructed using hmsIST (Hyberts et al. 2012) in NMRPipe (Delaglio et al. 1995), and all spectra were cross-checked against conventional thermal equilibrium spectra to identify potential spectral reconstruction artifacts. While many alternative NUS strategies are available (Kazimierczuk et al. 2009, 2010), the HMS implementations of sampling and reconstruction were chosen for the proof-of-concept experiments described here, as they are readily available to the NMR community and robust in our hands (https://gwagner.med.harvard.edu/intranet/hmsIST/).
Ubiquitin
As for the selectively hyperpolarized HN-CON experiments described above, we first tested the hyperpolarized HNCO using ubiquitin as a representative folded protein. Figure 3a displays the projection of the 1H-15N plane of a hyperpolarized 3D HNCO spectrum collected in just over one minute overlaid onto a conventional 1H-15N 2D HSQC of ubiquitin detected in 2 h (in 90:10 H2O:D2O, at pH 7.4 and 37 °C). Representative planes of the hyperpolarized HNCO spectrum are shown in Fig. 3b. While the conventional 2D reference spectrum has better resolution than its hyperpolarized 3D counterpart, as expected, the subset of peaks obtained in the hyperpolarized HNCO spectrum are in good agreement with their counterparts in the conventional 2D spectrum, and the linewidths in the NUS 3D are near the nominal resolution expected for an equivalent ‘fully-sampled’ 75 s acquisition (15N nominal resolution: 89 Hz (~ 1.1 ppm); 15N observed resolution: 1–2 ppm). Figure 3c maps the detected residues (in red) onto the ubiquitin crystal structure. This set of detected signals corresponds to the one obtained by the hyperpolarized HN-CON (Fig. 1b), as all peaks observed in the HN-CON are also observed in the HNCO. Notably, as was seen previously in the hyperpolarized HMQC of ubiquitin (Kadeřávek et al. 2018), significant enhancements are observed not only for solvent-exposed surface residues, but also for several residues in the protein core.
Osteopontin and its ligand heparin
Figure 4a, b shows representative planes of the hyperpolarized HNCO spectra of the heparin-free and -bound states of OPN (magenta and orange peaks, respectively), overlaid on the corresponding conventional thermal equilibrium HSQC spectra (green and blue, respectively). Figure 4c illustrates the residue-selectivity of the hyperpolarization in the presence and absence of heparin. The signal overlap in conventional 2D spectra of this system would make a transfer of assignments and spectral analysis particularly challenging (see also the Supporting Information). Here, extension of hyperpolarized experiments to encompass a third spectral dimension provides a significant improvement in resolution, and is also able to highlight several features of the OPN-heparin interaction.
Notably, in the presence of 2 equivalents of the ligand, traces of a second small population were seen, which we attribute to the unbound IDP. This could not be observed in conventionally detected spectra of the complex. This is illustrated in Fig. 4c, d, where spectral signatures of both the free and bound forms of OPN are detected for residue T176, while in the corresponding conventional spectrum, only the bound form is seen. If we assume that free OPN is not screened by the ligand, and thus more readily enhanced, solvent-based hyperpolarization would be expected to be particularly effective in highlighting the small population of free OPN even in the presence of a much larger population of the bound state. For residue K158 this effect is indeed observed, as only the free form is seen in the presence of heparin. While the pH and temperature used here (7.4, 37 °C) differ from those used in previous studies of this system, which may affect the visibility of certain residues, in any case the ability to detect such small populations should prove valuable for the characterization of sparsely-populated states.
Due to substantial overlap, the resolution in hyperpolarized 1D or 2D spectroscopy would not be sufficient to distinguish these two signals (see Fig. S9 in the Supporting Information for representative 1D projections). In the 3D experiments, however, these peaks are well isolated from other resonances and thus could be unambiguously assigned. As can be seen in Fig. 4c, the 15N chemical shifts of the two T176 signals are in good but not perfect agreement with those observed for the two peaks in this region of the static reference spectra. While we cannot exclude the possibility that alternative OPN conformations giving a similar but slightly shifted spectral signature could be present in the sample, the presence of distinct T176 signals in the 3D spectrum is consistent with the presence of both a bound and an unbound OPN species following heparin addition. Thus, by combining the improved sensitivity provided by hyperpolarized water with the improved resolution afforded by 3D detection, signatures of even small populations of minor states can be discerned.