Introduction

Biomolecular NMR experiments often demand enhancing the sensitivity of low-γ nuclei by polarization transfers from protons, as well as using polarization transfers to correlate different nuclear species with one another.^1-6 Both the sensitivity enhancing and the multinuclear correlation aspects of these polarization transfers play central roles in tackling the structures and dynamics of proteins and nucleic acids, and often involve bonded ¹H/¹⁵N spin pairs belonging to amide, amine, or imino groups.^6-8 These polarization transfers are usually achieved by a series of pulses and RF-free evolution periods, as in the INEPT pulse block used in ¹H−¹⁵N HSQC experiments.⁹ Here, 90° and 180° hard pulses with suitably interspersed free-evolution delays convert the initial longitudinal proton magnetization H_z to either single- or two-spin operators containing N_x elements, via the conversion of initial states into multi-spin 2H_±N_z intermediates. This kind of transfers, however, may suffer considerable losses in the presence of fast solvent exchanges, as under these conditions the transverse multi-spin components will rapidly decohere by interchanges with the H_z^water magnetization. It has been shown that, in such instances, polarization transfers via J-driven cross-polarization—so-called J-CP—can serve as a more robust scheme for transferring polarization from labile protons to heteronuclei.^10-12 As INEPT, CP also proceeds from single-spin proton (H_x^N) to single-spin nitrogen (N_x) polarization via two-spin elements.^{13, 14} Unlike what happens in INEPT, however, this transfer is gradual and continuous. Hence, if the CP ¹H spin-locking field is designed strong enough to capture both the ¹H^N and ¹H^water states, exchange-driven transfers from H_x^water to H_x^N can be leveraged to further boost the heteronuclear polarization transfer process. It has thus been shown that even when solvent exchange rates k_ex are much larger than the heteronuclear J couplings, the H_x^N→N_x CP process can proceed nearly to completion if given long enough contact times.¹⁵ Such CP strategies can then be used to improve the sensitivity of protein 2D NMR experiments, by replacing the initial H_x^N→N_x INEPT-based transfer with a Hartmann-Hahn (HH) contact block.¹⁶

CP's ability to execute polarization transfers even under fast k_ex≫J exchanges demands that the labile and water ¹Hs be simultaneously spin-locked, while fulfilling the HH match.¹⁷ This demand can be difficult or impossible to fulfill in bioNMR experiments, where the use of high fields leads to large ¹H^N−¹H^water chemical shift differences, and the use of cryogenically-cooled probes limits the γB₁ fields that can be achieved on the low-frequency channels. For the particular setup to be considered here, the maximum safe ¹⁵N field that can be deposited without causing RF arcing is γ_NB_1,N/2π≤3.5 kHz; the HH γ_HB_1,H=γ_NB_1,N condition thus requires setting the ¹H RF γ_HB_1,H/2π to ≈3.5 kHz, a field that is insufficient for providing an effective, simultaneous spin-locking of the water and amide/amine/imino resonances. Paradoxically, considerably stronger γ_HB_1,H fields are achievable in high-field cryogenically-cooled probes, but these are incompatible with the HH-imposed demands.

A similar scenario, where a strong γ_HB_1,H field is needed for efficient spin-locking but these strengths cannot be matched by the low-γ RF field hardware, arose decades ago in solid-state NMR.¹⁸ Strong RF fields are then needed on the ¹Hs because of the substantial T_1ρ shortening arising when γ_HB_1,H was smaller than the dipolar ¹H line width,¹⁹ but these strong fields are hard to match for the low-γ frequency channels. Although these problems were to some extent superseded by the advent of small-diameter magic-angle-spinning coils,²⁰ a number of experiments were proposed to alleviate these conflicting demands. The present study explores and adapts some of these experiments to the optimization of ¹H^N→¹⁵N transfers in aqueous systems. We find that the polarization transfer efficiencies of these adapted experiments are comparable to that of current J-CP Schemes under slow exchange and on-resonance irradiation conditions, but they act over broader ¹H chemical shift ranges and retain their efficiencies as the solvent exchange rates become faster. These performances are experimentally verified on small biomolecules, on rapidly exchanging lysine protein side chains, and on intrinsically disordered protein (IDP) samples, in a series of tests that also included ¹H→¹⁵N→¹³C transfer experiments.²¹

Results and Discussion

Figure 1 summarizes the various pulse sequences explored in this study. They include: (a) CP sequences based on continuous wave (CW) irradiations on all channels with fields applied on resonance at the highest γ_HB_1,H=γ_NB_1,N conditions possible, or with the ¹⁵N spin-lock applied off-resonance in order to carry out the HH match at a higher γ_N effective field as aided by the offset (which can then be matched by the ¹H channel thanks to the faster nutation rates achievable in the latter). (b) Phase-modulated schemes based on widely used offset-compensated DIPSI spin-locking pulses,²² or on a mismatch-optimized I−S transfer MOIST proposal,²³ fulfilling either γ_HB_1,H=γ_NB_1,N RF as limited by the probe-imposed maximum γ_NB_1,N, or exploiting off-resonance effects. (c) A novel CP scheme that we propose based on looped, wideband, uniform rate, smoothly truncated (WURST) ¹H pulses, capable of preserving both the H^N and H^water magnetizations spin-locked over large chemical shift bandwidths, while still fulfilling the γ_HB_1,H=γ_NB_1,N condition. (d) Time-averaged precession frequency (TAPF) CP schemes,²⁴ which can match larger γ_HB_1,H fields to lower γ_NB_1,N by unbalanced alternations of the ¹H spin-locking phase.

To the best of our knowledge, the last two sequences are new within the context of solution state NMR, and hence deserve a short explanation. The TAPF scheme was designed to match the large γ_HB_1,H needed for an effective solid state spin-lock with the lower γB₁ fields achievable by low-γ nuclei like ¹⁵N, while still retaining an effective <γ_HB_1,H>=γ_NB_1,N HH match. This is carried out by imposing an uneven 180° alternation in the phase of the ¹H spin-locking field B_1,H(t), which per unit duration τ applies the ¹H RF with a phase +x for a time τ₊ and with a phase -x for a time τ₋. For sufficiently short overall cycle times τ, the ensuing ¹H field becomes then effectively scaled by a factor κ=(τ₊−τ₋)/τ. In the present study, we relied on a cycle whose unit involved a +x/-x/+x phase alternation with each phase of an equal t₊= t_- duration; this lead to a total unit lasting a total time τ=2 t₊+t₋, and to a κ=(2 t₊−t₋)/(2 t₊+t₋)=1/3 scaling. This allowed us to apply ≈10 kHz for the ¹H spin-locking, while matching ¹⁵N fields of ≈3.3 kHz while fulfilling the HH condition.

The second novel scheme, L-WURST CP, is also inspired by a solid-state NMR experiment: broadband adiabatic inversion CP (BRAIN-CP).²⁵ Whereas in BRAIN-CP, the swept pulses are applied on the low-γ nuclei in order to polarize them while covering large bandwidths associated to anisotropies that are much larger than γB₁, these pulses are here applied on the ¹H channel. Their aim is thus not to receive polarization, but rather to ensure the spin-locking of both ¹H^N and ¹H^water magnetizations throughout the CP process—despite using RF fields set to match ≈γ_NB_1,N that would be too weak to achieve a normal ¹H spin-lock. These ¹H pulses, however, are chosen here as adiabatic sweeps with a WURST shape,²⁶ that can then spin-lock magnetizations along a wide range of targeted ¹H chemical shift offsets. As these swept pulses will only transiently fulfill the HH condition, the efficiency of the polarization transfer—particularly in the presence of exchanges—will be relatively low. To compensate for this, the sequence relies on looping the WURST pulses multiple times, while sweeping the RF in opposite directions. By repeating N times this RF-driven sweeps of the H^N and H^water magnetizations from the +z direction and back, this sequence takes advantage of a flux of spin locked magnetization from the solvent to the exchanging labile protons—hence preserving the efficiency of the ¹H→¹⁵N process—even if at the expense of substantially longer contact times than a conventional CP (a feature that is also shared by the BRAIN-CP experiment).

Figure 2 employs an alanine-based model to examine the resilience of the experiments depicted in Figure 1, in terms of their broadband-ness under conditions of slower and faster exchanges (298 and 333 K). These experiments were optimized so as to fulfill the HH demands and were measured as a function of the central ¹H carrier offset. Intensities in this and in all remaining plots of this study are shown normalized to the ¹⁵N amplitude arising upon applying a DIPSI-2-based CP²² at the highest available powers (γB₁/2π≈3.4 and 3.1 kHz for the 498 MHz and 1 GHz NMR experiments respectively) and at 298 K, as this was considered the most robust and common scenario for conventional CP-based solution NMR transfers. In Figure 2a, the DIPSI scheme with γB₁/2π=3.4 kHz shows good sensitivity when applied at δ(¹H)≈6–8 ppm, but it drops substantially as the offset moves away and water's spin-locking is lost. This drop becomes more severe as the spin-locking field decreases from 3.4 to 2.0 kHz. Also, MOIST variants are significantly degraded by the combined action of exchange plus ¹H off-resonance. By contrast, the novel L-WURST and TAPF-CP variants show high broadbandness—the former owing to its use of swept pulses, and the latter due to its stronger instantaneous γ_HB_1,H fields. The use of WURST, in particular, allows one to extend the coverage all the way up to the 14 ppm region demanded by imino protons. Supporting Figure S1 extends these comparisons to data measured at other temperatures, where similar trends are observed. Unlike what happens when changing the ¹H offset, changing the ¹⁵N offset over a standard chemical shift range makes little difference on the signals measured by these experiments except for MOIST-CP (Supporting Figure S2). It was also found that different cycle times and arrangements for the +x/-x alternation, make little difference in the overall performance of the TAPF-CP (data not shown). What did introduce a difference in the TAPF-CP performance was adding a linear ramp in the amplitude of the RF pulses, yet another provision commonly implemented in solid-state CP NMR.²⁷ For instance, it was found that by ramping the amplitude of the ¹⁵N CW throughout a 10 % range of the nominal HH demands, alanine's TAPF-CP maximum signal intensity did not change—but the optimized contact time went from 400 ms in the constant amplitude case, to 240 ms in the ramped amplitude one. None of the other variants exhibited similar changes regarding signal intensity and/or optimal contact time with ramping. Amplitude ramping therefore appears beneficial to alleviate a potential complication of the TAPF sequence, related to its need for relatively long contact times.

The situation is both similar and different when comparing the various sequences at 1 GHz (Figure 2b). The CP efficiency of DIPSI is markedly better than all counterparts at lower temperatures, presumably thanks to its shorter contact times. However, the efficiencies of all sequences become significantly lower upon increasing temperature—even when compared to the efficiency drops measured at 498 MHz. We ascribe this to the loss of the water and H_N signals during the spin-lock period: for instance, a 20 % signal loss was observed for these ¹H resonances when applying L-WURST-CP at 1 GHz NMR over a 300 ms contact time, whereas negligible losses were observed in these peaks at the lower field. Particularly marked are the ca. 90 % losses exhibited by DIPSI-CP under these high-field, fast-exchange conditions. Such losses go beyond the aforementioned T_1ρ effects. As nearly identical behaviors were observed when experiments were repeated under dematched or spatially-limited samples, these losses are unlikely to arise from radiation damping and/or RF inhomogeneity effects; we are investigating whether interferences between the exchange process and DIPSI's composite-pulse cycle could explain them. By contrast, the new CP methods demonstrated relative resilience under these high-field, fast-exchanging conditions: as at lower fields, the L-WURST-CP showed then the best compromise between sensitivity and ¹H bandwidth.

Figure 3 uses the same L-alanine model to examine the resilience of the experiments depicted in Figure 1, against the effects of chemical exchange. These were evaluated by measuring ¹H→¹⁵N polarization transfer efficiencies for alanine's ¹⁵N signal in water as a function of temperature, within the RF and thermal limits of our cold probehead. At all exchange conditions, the INEPT performance is markedly poorer than that of all CP counterparts; these limitations become even more evident as temperature increases. All CP variants give similar performances at the lowest temperature when applied on-resonance on the 42 ppm ¹⁵N signal; although going off-resonance in the ¹⁵N channel enables the use of larger B_1,H fields,^{28, 29} the overall CP performance tends to be poorer—probably as a result of the sinθ scaling brought about in the efficiency of the HH recoupling upon varying the θ= spin-locking axis. Also notable are the decreased performances of phase-alternating sequences like MOIST and DIPSI as exchange rates increase, perhaps reflecting interferences between stochastic exchange and coherent phase-shifting processes that begin to act on similar timescales.

As mentioned, a H^water→H^N→¹⁵N “conveyor” of spin-locked magnetizations is important for enabling an efficient polarization transfer to the ¹⁵N, despite the onset of fast chemical exchanges. Figure 4 explores this aspect of the CP process with three variants introduced in Figure 1, as clarified by Liouville-space simulations. These numerical calculations accounted for chemical exchange, spin relaxation, and RF waveforms centered on-resonance with the labile proton;¹⁵ in addition, a 4 ppm offset between the H^N and the water resonance (498 MHz) was considered. The introduction of this offset, akin to that arising in alanine, drops the maximum achievable ¹⁵N enhancement from the ≈100 % value arising in the absence of chemical shifts (Figure 7 in ref. [15]), to ca. 60 %. Importantly, the effects of the solvent exchange on the ¹⁵N buildup are similar for all the pulse sequences, confirming the preservation of the water repolarization process—which adds to the broadband characteristics of the novel L-WURST-CP and TAPF-CP experiments illustrated above. Also visible in these plots are the rapid modulations introduced on the water by TAPF's phase switches, as well as the adiabatic inversions introduced by L-WURST-CP that continuously invert the proton magnetizations from +z to -z and back to +z. This becomes somewhat of a drawback when this approach is applied in the absence of exchange, as these inversions need to be then coordinated with the J-driven CP oscillations in order to maximize the effects of the H^N →¹⁵N transfer. By smoothing the latter oscillations, the onset of solvent exchanges removes this need for a precise timing. Furthermore, although these simulations focus on two-spin systems, qualitatively similar considerations are expected to hold—even if with slightly modified contact times—for more complex NH₂ and NH₃ systems.

Figure 5 explores the efficiency of CP processes based on the DIPSI, L-WURST and TAPF variants, as applied to the direct ¹⁵N detection of lysine side-chain in unfolded α-synuclein and in a drkN SH3 fragment under majority-folded conditions. All ¹H→¹⁵N transfer experiments were recorded at 498 MHz (¹H), and the former sample was examined at two temperatures. Lysine side chains were chosen as these residues are known for having fast solvent exchange rates (k_ex≥100 s⁻¹).^{33 1}H→¹⁵N transfers were thus explored in ¹⁵N-labeled α-synuclein and in an ¹⁵N-labeled fragment of SH3, containing a total of 14 and 5 lysine residues, respectively. These sidechains are also mobile, and hence only single resonances were observed in their 1D ¹⁵N NMR spectra. The general behavior observed for the residues in these two proteins is very similar, with the rapid solvent exchanges precluding the observation of INEPT-derived signals but not of the CP counterparts. Somewhat along the lines displayed by the alanine data at 333 K, TAPF provides the highest transfer efficiency for these fast exchanging groups.

Novel CP strategies for polarizing ¹⁵N sites bound to solvent exchanging protons should open possibilities for enhancing triple-resonance experiments involving transfers to ¹³C that are J-coupled to the ¹⁵N.^{21, 31} Demonstrations of such concatenated CP (CCP) blocks have been presented for both non-exchanging³² and exchanging systems.¹⁵ In such cases, the first CP is optimized for performing a ¹H→¹⁵N transfer, and is followed by a second CP process implementing the ¹⁵N→¹³C transfer. Figure 6a shows pulse sequences adapted to test into CCP the performance of the novel CP blocks introduced in Figure 1; Figure 6b examines the resulting experiments when applied to ¹³C/¹⁵N-enriched urea, together with ¹H→¹⁵N→¹³C transfers based on conventional CP-CP and INEPT-INEPT (both of these refocused) sequences. These tests measured the carbonyl signal of this compound when placed in water, at three different temperatures. As expected from the ¹⁵N measurements, the INEPT-based sequences show poor performance, while all CP-based ones provide detectable signals in all temperature ranges. In parallel to what was shown in Figure 3, CCP's efficiency decrease with temperature; a similar observation was reported by Lopez et al.,³⁴ whereby the intensity of CP-based HNCO experiments dropped as a function of temperature, but was always better than INEPT-based HNCO experiments. Notice how the new CP versions maintain their efficiency despite the onset of faster chemical exchanges at higher-temperatures. Similar experiments were used to target the lysine side chains of PhoA4, an intrinsically disordered protein. These sidechains are mobile and, in parallel with what was noticed in the ¹⁵N-detected experiments, basing the initial CP on TAPF provided the best sensitivity among all CCP variants. Supporting Figure S4 presents additional results obtained upon applying these CCP strategies to enhance the carbonyl region signals of two IDPs, via polarization transfers starting from the amide protons. While also here CCP shows better efficiency than INEPT-based counterparts at all temperatures, the heterogeneous nature of the samples makes a detailed interpretation more challenging than in the cases introduced in Figure 6: in many instances slowly-exchanging residues provide the DIPSI-based transfer with the best sensitivity; however, this sequence's deterioration with increasing temperatures is clearly more marked than those of the L-WURST and TAPF variants.

Conclusion

CP-based experiments are generally superior to INEPT-based counterparts when targeting labile protons, This is thanks to additional exchange-driven polarization H^water→H^N transfers, that will allow the primary H^N→N^x transfer to proceed. Taking advantage of these chemical exchanges requires ensuring a simultaneous spin-locking of both the H^N and H^water magnetizations, something which is difficult to achieve while under Hartmann-Hahn matching due to the limited fields available. The present study examined schemes that can provide broad ¹H effective spin-locks covering wider chemical shift ranges, even while fulfilling the ¹⁵N-imposed HH match. The simplest option involved adding an off-resonance onto the ¹⁵N, leading to larger effective fields, and hence to the possibility of utilizing the extra power available in the ¹H RF channel. This route proved disappointing, both when confronted with ¹H off-resonance effects and with rapid chemical exchanges. Alternatives were thus sought in solid-state NMR experiments; in particular on schemes relying on adiabatic pulses that will perform spin-locking over a range of offsets defined by their sweep ranges, and on time-alternated phase-flipped schemes that will match strong γ_HB_1,H fields to lower γ_NB_1,N fields on an “average” sense. These alternatives showed improvements vis-à-vis CW- or DIPSI-based CP schemes, both in terms of the range of operational ¹H offsets, and of resilience to chemical exchanges. Their performance, however, is still far from optimal in a number of aspects. One is still a limited chemical shift range; indeed, though expanded, the TAPF range is still insufficient to cover imino protons resonating at ca. 14 ppm. This range—and even wider ones—can be easily covered by the L-WURST-CP, but the efficiency per unit time of this sequence is limited, necessitating multiple loops and, in turn, long contact times for its operation. Still, these new solids-inspired schemes seem to present improved solutions to this long-standing problem, providing broadbanded ¹H effective fields which can potentially be applied to numerous double and triple resonance experiments involving fast exchanging labile protons at physiological conditions.

Conflict of interest

The authors declare no conflict of interest.