Introduction

Hyperpolarization techniques have emerged as powerful tools in nuclear magnetic resonance (NMR). They enable the amplification of NMR signals by several orders of magnitude beyond those achievable under thermal equilibrium conditions1. Despite NMR widespread use, its improvement has been necessary because NMR suffers from inherently low sensitivity. Dissolution dynamic nuclear polarization (dDNP) is the leading technique to polarize liquids for in vivo MRI diagnosis2. However, this method’s high cost and low throughput are considered significant drawbacks.

Among other hyperpolarization techniques, signal amplification by reversible exchange (SABRE) has garnered significant attention due to its versatility and efficiency3,4,5,6,7,8,9. SABRE enables the utilization of hyperpolarized substrates while preserving their chemical composition. It achieves this by temporarily binding parahydrogen (pH2) and a target to an Ir-catalyst. Such spin order is transferred through the resulting scalar coupling network10,11,12,13,14. Subsequent dissociation facilitates the accumulation of polarized substrate in solution (Fig. 1). This process can be used to rapidly analyze biofluids with a micromolar accuracy15,16 or subsequent in vivo imaging4,5. SABRE has made significant advances during the last few years, enabling the large-scale polarization of pyruvate and its derivatives to more than 10% 13C polarization values4,17. Unlike dDNP, pH2-based techniques are scalable and offer a short duty cycle. The pursuit of hyperpolarization techniques represents a compelling avenue for pushing the boundaries of NMR and expanding its utility in various medical, scientific, and industrial applications.

Fig. 1: Schematic diagram for SABRE hyperpolarization.
figure 1

The pH2 and substrate (S) molecules used here transiently bind to a SABRE complex of the form [Ir(H)2(IMes)S3]Cl. At the high magnetic field, spin-spin interactions together with RF pulses drive polarization transfer from parahydrogen (pH2, vermillion) into the substrate, resulting in the hyperpolarization of both equatorial substrate ligands (blue) and, ultimately, the free substrate (vermillion) and formation of normal hydrogen (nH2, black). (Py pyridine, MNZ metronidazole, 4AP 4-aminopyridine, NAM nicotinamide, ACN acetonitrile).

Full size image

From the very first description of the SABRE effect, great efforts were put into understanding the interplay between the chemical exchange, lifetimes of transient complexes, and nuclear spin interactions - much of the SABRE effect is determined by these factors18,19,20,21,22. As the spin interactions are not easy to change, many efforts focused on modifying the exchange and lifetimes. The mainstream way of doing this is to vary the composition of the mixture, solvent, temperature, and ligands surrounding the Ir center23,24,25,26,27,28,29.

The SABRE complex lifetimes can be assessed indirectly by observing the effect of chemical exchange on polarization transfer efficiency30,31. Alternatively, a more direct measurement can be achieved using proton (1H) exchange spectroscopy (EXSY)32,33,34,35. Following this, in a previous study36, we have compared the performance of 1H 1D SEXSY37,38,39 and 2D EXSY32,34,35 NMR spectroscopic techniques and the robustness of the extracted exchange rate constants for the most widely used SABRE organometallic complex (IrIMes). However, these techniques exploit the thermal 1H NMR spectra with a lower spectral dispersion than heteronuclear NMR40,41,42. For example, measuring acetonitrile and metronidazole SABRE complexes’ lifetimes with 1H NMR was impossible because of the marginal chemical shift differences between the corresponding free and catalyst-bound substrate species. Both compounds are common SABRE targets, and the latter one is an antibiotic and prospective hypoxic radiosensitizer whose MRI properties were recently explored43. 15N-enhanced EXSY also uses the long relaxation of 15N compared to 1H, better-separating exchange and relaxation time periods, allowing for a more precise measurement of relatively slow exchange.

Our study aims to quantify the lifetimes of SABRE-active Ir complexes whilst improving the precision of such estimates. Furthermore, we want to increase the scope of the systems where such analysis is possible, especially, for example, in congested spectra where the selective excitation of 1H resonances of interest is impossible; such situations are common at low magnetic fields. To achieve this aim, we modified a high-field 15N hyperpolarization scheme that has been used before for both Crabtree’s catalyst32 and N-heterocyclic carbene complex studies40. The detection of the insensitive 15N nucleus is facilitated by polarization transfer from the pH2-derived hydride ligands via a phINEPT+ spin order transfer (SOT) sequence44. This process boosts the sensitivity of 15N EXSY and allows to rapidly acquire the spectra benefiting from the hyperpolarization and chemical shift dispersion of 15N.

When such INEPT-type sequences are used for pH2-based heteronuclear NMR signal enhancement32,40, up to 50% X-nuclei polarization has been predicted theoretically45,46,47. The efficiency of this approach, though, can be improved further if frequency-selective radiofrequency (RF) pulses are used instead of hard pulses to excite the protons originating from pH214,46,48. Such an approach is not applicable when these protons are chemically equivalent, which is the case for symmetric Ir complexes that only yield PHIP due to magnetic inequivalence associated with the AA′X or AA′XX′ type spin system created with 15N labeling49. More recently, the ESOTHERIC SOT pulse sequence has been introduced for hydrogenative PHIP50. This SOT sequence offered up to 100% X-nuclear polarization using only broadband RF pulses.

Here, we compare the efficacy of the ESOTHERIC and phINEPT+ approaches for 15N SABRE and use them to elucidate substrate exchange rates. Experimentally, the ESOTHERIC approach provided up to 35% higher polarization levels compared to phINEPT+. Using the larger signal enhancement of SABRE-ESOTHERIC, the lifetimes of the SABRE-active Ir complexes were studied with greater accuracy.

Results and discussion

Spin order transfer sequences

phINEPT+ and ESOTHERIC sequences (Fig. 2A, B) were used to transfer polarization from the pH2-derived hydride ligands of the SABRE catalyst to the 15N nuclei of the bound substrate. Although here only symmetric spin systems were studied, resulting in two chemically equivalent hydrides IrHH, the fact that these nuclei are strongly magnetically nonequivalent results in an averaging of the singlet state, leaving imbalanced observable antiphase \({\hat{I}}_{{{{\rm{z}}}}}{\hat{S}}_{{{{\rm{z}}}}}\) spin order51,52. The discussed pulse sequences are designed to transfer this type of polarization to heteronuclei.

Fig. 2: SOTs for SABRE hyperpolarization and measurement of chemical exchange.
figure 2

phINEPT+ (A) and ESOTHERIC (B) SOT sequences and the modifications allowing investigation of chemical exchange (C, SABRE-INEPT or SABRE-ESOTHERIC). The phINEPT+ (A) and ESOTHERIC (B) sequences convert the spin order of the pH2-derived hydride ligands into transverse (X) magnetization on the 15N nuclei of the bound substrate. Adding two 90° 15N RF pulses enables the observation of chemical exchange between bound and free substrates during the mixing time τe between the pulses (C). The blue square in (C) should be substituted with an appropriate SOT sequence: here phINEPT+ or ESOTHERIC. The total phINEPT+ time ttot = 2(τ1 + τ2) and the total ESOTHERIC time ttot = 2(2τ1 + τ2).

Full size image

Adding two 90° 15N pulses after the SOT block allows for observing the chemical exchange of the substrate (Fig. 2C). Here, the first 90° pulse after the SOT block converts the transverse net 15N magnetization into longitudinal magnetization (as was suggested for l-phINEPT+)40. During the following free evolution interval τe, some of the equatorial substrate ligands dissociate from the complex. The second 90° pulse samples the longitudinal magnetization of the free and bound substrate53. By variation of τe, the exchange kinetics are monitored, which can be analyzed to find the lifetime of the SABRE complex. We refer to these sequences as SABRE-INEPT and SARBE-ESOTHERIC.

SABRE with phINEPT+ and ESOTHERIC

The SOT efficiency in SABRE experiments is not trivial to predict due to a complex interplay of the chemical exchange and spin-spin interactions12,31. Therefore, first, we estimated the optimum parameters for phINEPT+ and ESOTHERIC for polarization transfer from pH2-derived hydrides to the equatorially bound substrate (eS) on the SABRE catalyst. For this, we experimentally varied the two intervals τ1 and τ2 (Fig. 2A, B) and measured the 15N signal of the bound 15N-labeled substrates at 288 K for 15N-Py, 15N-NAM, 15N-4AP, and 15N-ACN and at 267 K for 15N3-MNZ (Fig. 3). At 288 K, 15N3-MNZ did not provide high polarization values, as will be evident below, due to the rapid dissociation rate—one of the fastest among the tested substrates. Therefore, we cooled the sample down to 267 K where the dissociation rate sufficiently slowed down, enabling polarization of 0.8% at optimal SOT parameters. We used a 1 ms time grid to assess the optimal conditions that required from 25 to 81 experiments per map, with 453 hyperpolarization experiments in total. The optimal SOT parameters found through this approach were used in all other experiments, including different temperatures for 15N3-MNZ and other substrates: this is justifiable because although exchange rates are changing significantly as a function of temperature, the spin-spin interactions—the driving force for SOT—do not change significantly in the studied range.

Fig. 3: Performance of phINEPT+ and ESOTHERIC SOT sequences.
figure 3

Experimentally measured 15N polarization of [15N]pyridine (15N-Py, A), 4-amino[15N]pyridine (15N-4AP, B), [15N]nicotinamide (15N-NAM, C), [15N]acetonitrile (15N-ACN, D), and [15N3]metronidazole (15N3-MNZ, E; the signal of 15N-3 site of the symmetric [Ir(IMes)(H)2(MNZ)3] complex with resonance at 203 ppm was analyzed) as a function of τ1 and τ2 delays of phINEPT+ (A–E top) and ESOTHERIC (A–E bottom) SOT sequences at 9.4 T. The measurements were performed at 267 K for 15N3-MNZ and 288 K for the other substrates. phINEPT+ and ESOTHERIC 2D maps were jointly normalized for each substrate (i.e., the signals should not be compared between the different substrates). 15N NMR spectra with the maximum polarization of the substrate at the equatorial site and the corresponding τ1 and τ2 values are given for phINEPT+ (F) and ESOTHERIC (G). At optimum conditions, ESOTHERIC provided higher polarization than phINEPT+. e and f stand for equatorial and free substrates.

Full size image

ESOTHERIC always provided from 3 to 35% higher polarization than phINEPT+ (Fig. 3). For example, a maximum 15N polarization of 0.96% was achieved with phINEPT+ (τ1 = 11 ms, τ2 = 9 ms, ttot = 40 ms) and 0.99% with ESOTHERIC (τ1 = 7 ms, τ2 = 6 ms, ttot = 40 ms) for 15N-ePy (Fig. 3A). The largest relative signal gain of 35% was achieved for 4AP: phINEPT+ provided 0.14% (τ1 = 14 ms, τ2 = 9 ms, ttot = 46 ms), while ESOTHERIC yielded 0.19% (τ1 = 8 ms, τ2 = 3 ms, ttot = 38 ms) 15N polarization. 1% of 15N polarization corresponds to a signal enhancement factor of 2584 at 9.4 T. Although the ESOTHERIC sequence has more RF pulses and time intervals, the total duration of the phINEPT+ sequence is not shorter than that of ESOTHERIC, when both use the optimal settings for the associated chemical system (Fig. 3): an additional positive feature of ESOTHERIC sequence.

The spectral lines of hyperpolarized molecules measured immediately after phINEPT+ and ESOTHERIC have distorted line shapes due to polarization of zero and higher-order quantum coherences, which are populated due to numerous spin–spin interactions in the system. These coherences can be suppressed by adding a corresponding quantum filter54. In the following experiments with SABRE-INEPT and SABRE-ESOTHERIC (SOT scheme in Fig. 2C), we used a pulsed field gradient to filter out quantum coherences of higher order, which was found to be sufficient to provide distortionless lines (Fig. S1, SI).

As mentioned, ESOTHERIC is expected to give twice the signal intensity of phINEPT+, however, only at specific conditions and for a three-spin-½ system. Here, the spin systems are more complex (>4 spins), and additional interactions with other magnetic nuclei (e.g., 1H or 15N in our case) in the substrates destructively modulate SOT. The application of selective refocusing pulses can improve the efficiency of both sequences since they reduce the system to an effective four-spin system (two hydrides IrHH and two equatorial 15N nuclei); however, it was not tested here.

Note that experiments for all compounds were conducted at ambient pressure except for experiments with MNZ. This is because preliminary tests at ambient pressure with MNZ revealed the necessity for a long activation period, exceeding one hour and did not provide sufficient signal gain for reliable measurements. Using a higher pressure (6.6 bar), a common practice to increase activation rate and polarization yield55, the sample was activated within 40 minutes, and both phINEPT+ and ESOTHERIC provided close to 0.8% 15N polarization.

After activation, we conducted experiments with the same sample for up to 5 h. During the course of the experiment, the sample volume decreased by less than 10% thanks to the low temperatures used. Still, this leads to additional errors when evaluating the ligand exchange rate constants. To avoid this, two solutions were proposed in literature (not used here): adding a sample reservoir56 that keeps the sample volume constant or saturation of dry pH2 with methanol that significantly slows down evaporation29,57. Note that the latter approach works thanks to the slow para-to-ortho conversion of H2 in clean solvents and tubes58.

Transferring polarization to the free substrate

Both ESOTHERIC and phINEPT+ SOT sequences create transverse net magnetization on the equatorial substrate ligands. Chemical exchange can therefore transfer a bound response to signal for free substrate. However, to measure such a process beyond the decay time of transverse magnetization (T2*) it is necessary to add a 90o RF pulse that converts the transverse magnetization of the bound substrate to its longitudinal magnetization. The subsequent chemical exchange transfers it to the free substrate that can be measured by applying another 90° pulse after the appropriate free evolution interval τe between the two pulses (Fig. 2C).

The chemical exchange was monitored using SABRE-ESOTHERIC (or SABRE-INEPT) for the same five compounds at several temperatures, using optimized SOT parameters (intervals τ1 and τ2). At τe = 0 s, the polarization of the free substrate is zero, and that of the bound form is at its maximum. Increasing the time delay between the last two pulses encodes polarization transfer from the bound to the free substrate. For example, the signal of initially hyperpolarized equatorial 15N-NAM (15N-eNAM) with a chemical shift of 255.7 ppm was decreasing while that of free 15N-NAM (15N-fNAM) at 302.3 ppm was initially increasing as a function of τe (Fig. 4). After about 1 s, visible polarization reaches its maximum on the free form and then decreases due to relaxation.

Fig. 4: Dissociation kinetics of equatorially bound NAM.
figure 4

Normalized 15N signals (A) and 15N NMR spectra (B) of free (15N-fNAM, vermillion) and equatorial (15N-eNAM, blue) 15N-NAM as a function of free evolution interval τe at 293 K and 9.4 T. 15N signals were enhanced with ESOTHERIC with τ1 = 7 ms and τ2 = 5 ms. Kinetics in A were fitted with a shared biexponential decay function (eigenvalues analysis, resulting in two decay rates of k = (4.63 ± 0.07) s−1 and R = (0.07 ± 0.002) s−1), using models CSS2 ↔ CSS + S and CSS2 ↔ S236. The slower rate constant R corresponds to the effective relaxation rate, while the faster rate constant k corresponds to the effective exchange rate.

Full size image

Exchange rates obtained with SABRE-ESOTHERIC

All kinetics for longitudinal magnetization of free (Mf) and equatorial (Me) substrates measured with SABRE-ESOTHERIC (one example for NAM is given in Fig. 4) were fitted using global fitting for exponential decay constants R and k in Eq. 1 (see Ref. 36 for details).

$${M}_{{{{\rm{e}}}},{{{\rm{f}}}}}={A}_{{{{\rm{e}}}},{{{\rm{f}}}}}{e}^{-{Rt}}+{{B}_{{{{\rm{e}}}},{{{\rm{f}}}}}}^{-{kt}}+{C}_{{{{\rm{e}}}},{{{\rm{f}}}}}$$
(1)

where e, f stand for equatorial and free substrates, giving eigenvalues of two-site exchange: R = 0.072 ± 0.002 s−1 and k = 4.63 ± 0.07 s−1 for exchange of NAM at 293 K (Fig. 4A). As a result, we have two decay constants: the smaller one (almost temperature independent) attributed to the effective relaxation rate constant R and the larger one (strongly temperature dependent) attributed to the exchange rate constant k which lets us calculate the dissociation rate kd using Eq. 236.

$${k}_{{{{\rm{d}}}}}=\frac{k}{0.5+\frac{\left[{{{{\rm{C}}}}}_{{{{\rm{S}}}}}{{{{\rm{S}}}}}_{2}\right]}{\left[{{{\rm{S}}}}\right]}}$$
(2)

The concentration ratio \(\frac{\left[{{{{\rm{C}}}}}_{{{{\rm{S}}}}}{{{{\rm{S}}}}}_{2}\right]}{\left[{{{\rm{S}}}}\right]}\) was calculated from the 1H spectrum. For \(\frac{\left[{{{{\rm{C}}}}}_{{{{\rm{S}}}}}{{{{\rm{S}}}}}_{2}\right]}{\left[{{{\rm{S}}}}\right]}=12.24\), this yields kd = 8.0 ± 0.1 s−1.

Additionally, the resulting data were fit to two chemical exchange models developed in the earlier work36: the more complete model CSS2 ↔ CSS + S and the simplified model CSS2 ↔ S2. In this example (Fig. 4A), both models yielded the same kd within the error margin kd = 10.0 ± 0.3 s−1.

Dissociation rates for all substrates were calculated using these three models (Table S3, SI) and plotted as functions of inverse temperature together with the weighted mean value across the models (Fig. 5). The mean dissociation rate constants were further fitted using Eyring equation (Eq. S8, SI).

Fig. 5: Substrate dissociation rate constants extracted from SABRE-ESOTHERIC kinetics as a function of temperature.
figure 5

Dissociation rate constants for 15N-Py (black), 15N-4AP (reddish purple), 15N-NAM (bluish green), 15N-ACN (yellow), and 15N3-MNZ (orange) were obtained using model CSS2 ↔ CSS + S (circles), model CSS2 ↔ S2 (triangles), and using eigenvalues analysis (squares). The weighted mean kd values (stars) were calculated using Eq. S6 (SI). The mean kd values were fitted using Eyring equation (line).

Full size image

The dissociation rates for the five substrates under study were significantly different. 15N-ACN and 15N3-MNZ exchanged significantly faster than the pyridine derivatives 15N-Py, 15N-4AP, and 15N-NAM. It has been predicted that, for pyridine derivatives, in the absence of steric effects, the lower the pKa of the conjugate acid, the stronger the ligand binding59. ACN and MNZ do not belong to the six-membered heterocycle family; the higher charge density of six-membered heterocycles leads to stronger interactions between charged entities (Ir+ and the ring) and slows down dissociation rates. In addition, MNZ has additional steric hindrances. A rationale for the rapid loss of 15N-ACN (the smallest molecule in the group) needs further evaluation, although these data are consistent with other studies, as shown in Ref. 60. To confirm the influence of the reagent structure, a larger pool of substrates, including five-membered rings with 1–3 nitrogens, should be tested.

With SABRE-ESOTHERIC, the lifetimes obtained from the CSS2 ↔ CSS + S and CSS2 ↔ S2 models were almost equal but differed from those derived from the eigenvalues analysis. Comparison with the previously published 1H SEXSY data36 for Py, 4AP, and NAM showed mixed results. For Py, SABRE-ESOTHERIC typically provided greater dissociation rate constants than 1H SEXSY with up to a 2.4-fold difference (Table S3). For 4AP and NAM, the values obtained using these two approaches were closer.

Due to the negligible 1H chemical shift variation of MNZ and ACN upon coordination with the Ir catalyst, we could not measure their exchange rates with 1H EXSY before. Here, however, we were able to determine these values because of the larger 15N chemical shift difference of 45–60 ppm between equatorial-bound and free substrates.

In the case of 15N3-MNZ, several Ir complexes are present in the SABRE mixture, as confirmed by 15N and 1H hyperpolarized NMR (Figs. S8 and S9, SI). The major complexes are [Ir(IMes)(H)2(MNZ)3], [Ir(IMes)(H)2(MNZ)2Cl], and [Ir(IMes)(H)2(MNZ)2(CD3OD)] species29. The precise assessment of the substrate exchange, considering all these complexes, requires a more detailed analysis of the exchange mechanism and, thus, will be a subject of our future studies. Here, we simplified the matter by estimating phenomenological exchange rates between all hyperpolarized equatorial-bound 15N3-MNZ species and free 15N3-MNZ. Note, however, that the τ1 and τ2 delays in SABRE-ESOTHERIC SOT sequence were optimized for the polarization transfer in [Ir(IMes)(H)2(MNZ)3] species (see Fig. S10). Therefore, this species has the primary contribution to the overall observed substrate dissociation flux. The fitting of the total dissociation flux of MNZ with models CSS2 ↔ CSS + S and CSS2 ↔ S2 was not accurate, which is reasonable considering these simplifications. However, the curves fit well with a simple two-exponential decay (Fig. S6, SI). Therefore, only this approach was used to estimate the MNZ dissociation rate constants, enthalpy and entropy of activation.

Enthalpy ΔH ‡ and entropy ΔS ‡ of activation

Activation enthalpies ΔH‡ and entropies ΔS‡ were calculated using the dissociation rate constants kd (Fig. 6, Table S5, SI). The better signal-to-noise ratio resulting from the SABRE-ESOTHERIC approach enabled greater accuracy in the ensuing rate constants and, therefore, more accurate thermodynamic data to be obtained (Table S5, SI). The ΔH‡ values obtained were ~70–80 kJ/mol for all substrates under study (Table S5, SI). This value reflects the bond strength as the process of ligand loss is dissociative. The smaller substrates Py and ACN showed ΔS‡ of ca. 60 J/(mol K) according to eigenvalues analysis (Table S5). 4AP and NAM had ΔS‡ of ca. 40 J/(mol K), and MNZ had ΔS‡ = 29 J/(mol K) (although it should be noted that variation of ΔS‡ values between the CSS2 ↔ CSS + S and CSS2 ↔ S2 models, eigenvalues analysis and mean kd values was quite high).

Fig. 6: Comparison of estimated enthalpy and entropy values.
figure 6

Enthalpies ΔH‡ (A) and entropies ΔS‡ (B) of activation for the corresponding IrIMes complexes with 15N-Py, 15N-4AP, 15N-NAM, 15N-ACN, and 15N3-MNZ as a substrate. ΔH‡ and ΔS‡ were obtained from the dissociation rate constants obtained with the model CSS2 ↔ CS S + S (bluish green), with the model CSS2 ↔ S2 (reddish purple), eigenvalues analysis (yellow), mean kd values (orange), and from DFT calculations (blue). The values (except those obtained from DFT calculations) are obtained by fitting kd (Table S3, Fig. S7, SI) to the Eyring equation and are given in Table S5.

Full size image

To assess the measured enthalpies of activation, we computed dissociation pathways and their energies using density functional theory (DFT) quantum chemistry simulations at the B3LYP-D4/def2-TZVP level of theory61,62,63,64 including DLPNO-CCSD(T)/def2-TZVP single point energies with implicit solvation. The ligand dissociation potential energy surface did not exhibit an identified transition state (Fig. S12, SI). Hence, we used the computed dissociation enthalpies to estimate ΔH‡ (Fig. 6, Table S6, SI). A qualitative agreement is observed between the computed dissociation energies and the enthalpies of activation derived from the measured rate constants.

ACN exhibited the smallest dissociation energy of 73 kJ/mol, MNZ had a higher dissociation energy of 82 kJ/mol, and Py, 4AP, and NAM fall into a group of similar values of 104, 99, and 106 kJ/mol (Table S6). Thus, the actual enthalpies seem to be overestimated in the computational values, very likely due to approximations related to the implicit description of solvation (for computational values see Supplementary Discussion: Quantum chemical calculations).

In recent work65, rate constants kd of Py and NAM dissociation were found to be kd = 10.8 s−1 and kd = 7.86 s−1 at 298 K, respectively. These values were determined using the 1H EXSY NMR exchange spectroscopy technique. Corresponding transition-free Gibbs energy at 298 K for losing Py and NAM was estimated to be approximately 67.05 and 68.22 kJ/mol using the Eyring plots of the dissociation rates65. DFT estimates are 45.48 kJ/mol for Py and 68.22 kJ/mol for NAM65. Here, DFT failed to replicate the similarities between the two substrates measured experimentally.

In our study, we estimated the exchange rate constants of Py and NAM at 298 K to be 24.32 and 15.68 s−1, respectively, with experimentally estimated activation enthalpies of 84 ± 4 and 79 ± 1 kJ/mol and DFT calculated values of 104 and 106 kJ/mol, respectively. Our energies are somewhat higher than those reported before65, however, the level of theory applied herein successfully reproduces similarities in activation energies with only a slight offset.

The difference between exchange rate constants measured here and in Ref. 65 is almost two-fold, likely as a result of the different definitions of the exchange rate. Based on our previous work36, we define the kd value in such a way that the lifetime of the Ir complex is given by 1/kd. Because the exchange proceeds via a dissociative or SN1-type mechanism and two equatorial substrates dissociate at the same rate, the individual dissociation rate of a chosen single ligand is half the dissociation rate of any of the two ligands. Therefore, when comparing the results presented here with the literature values, special attention should be paid to the description of the exchange models. After correcting for this factor, the dissociation rate constant values are similar.

Hyperpolarization of naturally abundant (n.a.) pyridine

To assess the hyperpolarization efficiency, we attempted to hyperpolarize unlabelled 400 mM pyridine (the natural abundance of 15N isotope is about 0.4%) with a 4 mM IrIMes catalyst. We experimentally transferred polarization to equatorial pyridine using both phINEPT+ and ESOTHERIC pulse sequences (Fig. 7A) and measured exchange rates (Fig. S11, SI) that gave deviation of up to 20% compared to values measured with labeled 15N-Py (Table S3). The maximum achievable polarization for equatorially bound pyridine was ~7.2% using phINEPT+ and ~17.5% using ESOTHERIC: ESOTHERIC provided twice the signal than phINEPT+. This is because the system was reduced effectively from a four-spin system consisting of two IrHH protons and two 15N nuclei of two equatorial substrates to the system with only one 15N nucleus in the equatorial plane. In this case, ESOTHERIC performance is optimal, while phINEPT+, at best, can give 50% of ESOTHERIC polarization values. To confirm our measurements, we simulated SOT performance for three- and four-spin systems without chemical exchange to estimate performance in the case of n.a. Py (Fig. 7C) and 15N-Py (Fig. 7D). The data qualitatively agree with experimental observation.

Fig. 7: Comparison of phINEPT+ and ESOTHERIC SOT for 15N-labeled and unlabeled substrate.
figure 7

Performance of phINEPT+ and ESOTHERIC SOT sequences for pyridine (n.a.—A, 15N—B) and simulations for three- (C) and four- (D) spin-½ systems. phINEPT+ and ESOTHERIC 2D maps were jointly normalized for each panel. Rectangles on C, D highlight the corresponding experimentally measured sections. phINEPT+ and ESOTHERIC SOT simulations have a maximum of 27.7% and 66.9% polarization for three-spin-½ systems (C) and 21.2% and 21.8% polarization for four-spin-½ systems (D): ESOTHERIC is 3 times more efficient for three-spin-½ system than for four-spin-½ system. Parameters used for the simulations were: JHH = –8 Hz, JNHt = –22 Hz, and JNHc = 0 Hz. To consider exchange during SOT, all polarization values were multiplied by the exp(−kd × ttot) value, with kd = 2.7 s−1 and ttot is the total time of the SOT sequence, which accounts for the fact that only a fraction of molecules that survived whole SOT sequence step contribute to the final NMR signal; alternatively, one can use an explicit simulation of chemical exchange and spin evolution20,21.

Full size image

Conclusions

The approach based on polarization transfer to heteronuclei and then the encoding of an exchange rate illustrated here for 15N reflects an exciting addition to the toolkit of mechanistic chemists. We compared the performance of broadband phINEPT+ and ESOTHERIC SOT sequences for transferring nuclear spin polarization from the parahydrogen-derived hydride ligands of the Ir SABRE catalyst to the 15N nuclei of the coordinated substrates examplified by 15N-labeled pyridine, 4-aminopyridine, nicotinamide, acetonitrile, metronidazole and none labeledpyridine. ESOTHERIC was shown to be superior, providing higher 15N polarization values, especially in the none labeled case. The created hyperpolarization allowed the subsequent measurement of substrate exchange rates by 15N NMR. The chemical exchange models CSS2 ↔ CSS + S and CSS2 ↔ S2 described in the literature36 were used to achieve this. More precise data fitting provided by strong 15N NMR signal enhancement allowed us to estimate the dissociation rate constants more robustly and to measure enthalpies of activation with higher accuracy than in the earlier work. The utility of 15N detection offers much better chemical shift dispersion to delineate the free and catalyst-bound species in SABRE mixtures compared to conventional 1D 1H SEXSY or 2D 1H EXSY techniques. As a result of the greatly enhanced spectral resolution, it became possible to measure the dissociation rate constants for acetonitrile and metronidazole. The reported methodology broadly applies to various substrates suitable for SABRE hyperpolarization.

Through this methodology, it would be possible to evaluate ligand dissociation kinetics in the wide-ranging number of transition metal66 and metal-free67 species detected through PHIP. The suggested approach is not limited to 15N and can be used for other magnetic nuclei. For example, one can study dissociation of phosphine ligands (which are common in organometallic catalysis) using 31P NMR. Furthermore, as 31P is essentially 100% abundant, labeling is not needed, and it may be possible to extend these approaches to normal NMR without hyperpolarizsation using regular INEPT. Exchange of carbon-containing ligands may be studied as well, in this context, the 1% natural abundance suggests hyperpolarization will be necessary. For example, recently, we measured J-coupling constants between IrHH protons and coordinated [1-13C]pyruvate using a modified INEPT sequence14. A slight modification of the sequence presented here will allow one to gain insight into the chemical kinetics of pyruvate exchange during SABRE. This approach is also not limited to RF SOT-driven hyperpolarization. With demonstrated 15N polarization values achieved in ultra-low field experiments68, kinetics measurements even on low-field benchtop NMR systems should be possible. As the accuracy of this approach has been shown to be high, the results will be significant.

Methods

Chemicals

Perdeuterated Ir precatalyst [Ir-d22] = [IrCl(COD)(IMes-d22)] was synthesized according to Ref. 55 (IMes = 1,3-bis(2,4,6-trimethylphenyl)imidazol-2-ylidene, COD = 1,5-cyclooctadiene). [15N]Pyridine (15N-Py, 486183, Sigma-Aldrich), [15N]acetonitrile (15N-ACN, 487864, Sigma-Aldrich), and methanol-d4 (441384, Sigma-Aldrich) were used as received. 4-Amino[15N]pyridine (15N-4AP) was synthesized according to Ref. 69. [15N]Nicotinamide (15N-NAM) was synthesized in a two-step reaction from unlabeled NAM (72340, CAS: 98-92-0, Sigma-Aldrich) via the corresponding Zincke salt followed by nitrogen exchange with 15NH4Cl (299251, CAS: 39466-62-1, Sigma-Aldrich)70,71. [15N3]metronidazole (15N3-MNZ) was synthesized according to Ref. 72.

Sample preparation

The samples were prepared by mixing one of the five SABRE substrates (15N-Py, 15N-NAM, 15N-4AP, 15N3-MNZ, or 15N-ACN) at a concentration of 40, 80, 80, 80, and 120 mM, respectively, with 4 mM of [Ir-d22] in 1.2 mL of methanol-d4.

Parahydrogen enrichment

pH2-enriched hydrogen gas with a 93% fraction of pH2 was prepared by passing high-purity hydrogen over hydrated iron(III) oxide at 25 K using a pH2 generator similar to the one in Ref. 73. For 52% content, pH2 was prepared using an in-house built liquid nitrogen system described before in Ref. 74.

phINEPT+, ESOTHERIC, SABRE-INEPT, and SABRE-ESOTHERIC experiments

All experiments with hyperpolarization were carried out on a Bruker 400 MHz spectrometer using a 5 mm BBO probe. Typically, a 600 µL sample was transferred to a screw-cap 5 mm NMR tube (507-TR-8, Wilmad Lab). The NMR tube was then attached to a low-pressure system, which enabled pH2 bubbling through the solution using 0.2-0.3 bar overpressure with respect to the ambient pressure75, and placed inside the spectrometer. First, we flushed the sample with pH2 for 30 min to activate the catalyst. In the case of 15N3-MNZ, 400 µL of the sample was instead transferred to a high-pressure 5 mm NMR tube (524-PV-7, Wilmad Lab) connected to a high-pressure system operating at 6.6 bar (similar to the one used in Ref. 76). The 15N3-MNZ sample was bubbled with pH2 for 40 min to activate the catalyst precursor. Then, a series of experiments with one of the SOT sequences: phINEPT+, ESOTHERIC, SABRE-INEPT, or SABRE-ESOTHERIC, were carried out. One experiment consisted of (1) 10 s period of relaxation delay followed by (2) 2 s of WALTZ-16 on 1H channel, then (3) 10 s of pH2 bubbling through the sample, followed by (4) a delay of 1.5 s for bubbles to disappear, followed by (5) one of the SOT sequences (Fig. 2). The following parameters of RF pulses were used: for 1H, the frequency of the RF pulse was −22.79 ppm in resonance with hydride protons, and the duration of hard 90° RF pulse was 11.575 µs; for, 15N the frequency was in resonance with the corresponding equatorial substrate and the duration of hard 90° RF pulse was 21.25 µs.

Data processing

All spectra were analyzed using spectral data analyzing software Bruker TopSpin (4.1.4), MestReNova (14.2.2), and Origin (2021). SABRE-INEPT and SABRE-ESOTHERIC data were modeled using Origin or MATLAB (R2021a) as described in the text as well as in the previous publication36.