The effects of nitroxide structure upon ¹H Overhauser dynamic nuclear polarization efficacy at ultralow-field

¹H Overhauser dynamic nuclear polarization (ODNP) is a hyperpolarization method that transfers the higher electron spin polarization to target proton spins.¹ In contrast to other common dynamic nuclear polarization (DNP) mechanisms (solid effect, cross effect, and thermal mixing), the classical Overhauser effect allows for the hyperpolarization of liquids.² In recent years, there has been a renaissance in terms of ODNP.² While it was also shown that ODNP works at high magnetic field strengths (>1 T),^3,4 the main reason for the growing interest in this method was the construction of novel low-field (<0.5 T) and ultralow-field (ULF, <10 mT) MRI scanners.^5–7 In this field range, the radio frequency (RF) needed for transferring spin order from electrons to protons via ODNP¹ lies in the 100 MHz range. Thus, a sufficient penetration depth for larger samples is enabled, making in vivo hyperpolarization experiments with small animals possible. The continuous in vivo hyperpolarization is one of the main advantages of ODNP over other techniques such as parahydrogen-based hyperpolarization, where the hyperpolarization process takes place outside of the subject.⁸ It led to the development of in vivo Overhauser MRI (OMRI). Several groups already performed OMRI experiments on animals using trityl radicals^9,10 or nitroxides such as carboxy-PROXYL (PCA)¹¹ and TEMPO¹² in mM concentrations.

Several trityl radicals show no hyperfine splitting of the electron energy levels, and the toxicity of some hydrophilic derivatives, e.g., Oxo63, is sufficiently low.¹³ However, they are difficult to synthesize or to obtain commercially. Nitroxides, e.g., TEMPO or PCA, are easily available, but the magnitude of the maximal enhancement is lower than that of trityl free radicals because of hyperfine splitting.¹⁴ For more than half a century, nitroxides have been exploited as spin probes and spin labels in biophysics and biomedical research. Specific spin probes have been designed for imaging of extracellular pH,¹⁵ thiols,¹⁶ and enzymatic activities.¹⁷ The unique redox properties of nitroxides make them a useful tool for the investigation of oxidative stress.^11,18 Unlike large trityls, small-sized nitroxides show a high ability to permeate cells, tissues, and even the blood–brain-barrier, thus enabling brain imaging.^19–22 

Those kinds of small-size free radical molecules or spin probes are being investigated in respect to their ODNP properties since Hausser and Stehlik discovered the ODNP effect in liquids.²³ Not only the enhancement of the proton magnetic resonance (MR) signal was reported but also other nuclei, such as ²H, ¹³C, or ¹⁹F, could be hyperpolarized.²⁴ For OMRI, so far only ¹H in vivo hyperpolarization was implemented.

The main drawbacks of nitroxides as free radical probes for in vivo applications are their fast chemical reduction to diamagnetic compounds by biogenic reductants and/or their rapid excretion.^25,26 There are several general pathways to address the problem. Small spin probes can be incorporated into different macromolecular or supramolecular structures, which can prevent rapid reduction and clearance. Unfolded proteins, polyelectrolytes, polymers, surface lipid vesicles, and folded globular proteins have been already investigated.^27–32 The attached nitroxide produces ODNP enhancement within 10 Å distance.^28,29 Depending on the macromolecule, the magnitude of the enhancement can be reduced by a limited access of the solvent to the radical and by an increase in rotational correlation time. Thus, polymeric structures that reduce solvent accessibility and radical mobility can only be efficient as delivery systems (targeted or not), releasing small spin probes spontaneously or in response to certain biochemical processes, e.g., receptor interaction. Some examples of free nitroxide release from spin-labeled biopolymers have been published.^33,34 Experiments with spin probes attached to different heparins or tobacco mosaic viruses have been performed. Solvent accessibility and radical mobility seem sufficient in these experiments, and they provide promising candidates for in vivo studies.^35,36 While toxicity varies strongly for different spin probes,^13,36–38 the inclusion of spin probes into macromolecules could additionally render this problem negligible. As an alternative, sterically shielded nitroxides, such as tetraethyl-substituted nitroxides, with a higher resistance to bioreduction can be used.^39,40 To retard excretion, the nitroxides can be designed to permeate or even accumulate in tissues, living cells, or cellular compartments.^22,41

While multiple radicals have been investigated in previous studies,^24,42–49 a lot of promising nitroxide spin probes have never been examined in regard to their ODNP suitability.

The main goal of this work is to investigate nitroxide radicals with different chemical properties in order to compare their ODNP related characteristics. This may give insight into the underlying physical mechanisms in order to provide a foundation for future spin probe design.

All of the presented substances are stable in aqueous buffered solution. While biocompatibility, toxicity, and stability are equally important for biological and medical applications, we do not regard these features here since it is possible to manipulate these biological properties by inclusion of the spin probes in macromolecular structures. Instead, the focus of the study lies on their physical suitability for ODNP and how changes in the chemical structure affect ODNP related properties.

When assessing different free radicals in respect to their suitability for ODNP applications, the most crucial physical property is their ability to transfer electron spin polarization to the target nuclear spins, thus hyperpolarizing the sample.

The enhancement factor E = 〈I〉/I₀, comparing the hyperpolarized signal 〈I〉 to the non-hyperpolarized signal I₀, is one of the most important parameters. The achievable enhancement depends among others on the magnetic polarization field B_p, power of the RF-pulse P, hyperpolarization buildup time T_HP, and radical concentration.¹⁴ These dependencies are specific for each type of free radical.

In practice, in particular for in vivo applications, additional limitations to RF-power P, T_HP, and spin probe concentration apply. The optimal free radical exhibits a high magnitude of enhancement |E(P)| at low RF-power P and a short T_HP. The two most fitting parameters for an easily comprehensible comparison of spin probe performance and suitability are E_max and P_1/2. While E_max states the maximal enhancement achievable with a certain spin probe, P_1/2 as the RF-power needed to reach half of E_max expresses how easily the maximal enhancement is reached.

The efficiency of the local polarization transfer is characterized by the coupling factor ξ, which is independent of spin probe concentration and RF-power P. It is specific for each type of free radical spin probe and is defined as⁵⁰ 

with the electron and proton Larmor frequencies ω_e and ω_H. The leakage factor

describes how the dipolar relaxation mechanisms at B_p affect the relaxation time T_1,Bp of the sample, compared to T_1,0,Bp in a sample without free radical spin probes.

The electron spin resonance (ESR) saturation factor s(P) describes how well the electron spin transition is saturated for the specific sample,

The maximal saturation s_max depends on the spin probe concentration and has a maximum value of $13≤smax≤1$ for ¹⁴N or $12≤smax≤1$ for ¹⁵N nitroxides.^50–52 P_1/2 defines the RF-power P needed to reach half of the maximal saturation s_max, and it is inversely proportional to the longitudinal and transversal electron spin relaxation time.^53,54 At the limit of P → ∞, Eq. (1) determines the maximal enhancement E_max for a given sample as

Inserting Eq. (3) into Eq. (1) and substituting s_max via Eq. (4) yield an expression for the power dependent enhancement factor,

For OMRI applications, spin probes with a combination of high |E_max| and low P_1/2 are desirable.

The saturation of the ESR transition is crucial for the effective transfer of spin polarization. A RF-field is used to increase the population of higher energy states of the coupled spin system. For nitroxides, we find additional hyperfine splitting, originating from the interaction of the electron spin with the nitrogen nuclear spin.^14,44,55 In a continuous wave (cw)-Overhauser experiment with nitroxide radicals, there are one resonant and one or two non-resonant electron transitions, depending on the nitrogen isotope. Intermolecular or intramolecular interactions of the electron spin with other electron or nuclear spins can lead to a mixing of the resonant and non-resonant energy levels and/or broadening of the spectral lines. Therefore, these interactions also have an influence on s_max.^50,52 This mixing and broadening can increase s_max at the cost of a higher P_1/2.

Heisenberg spin exchange occurs when two nitroxide radical molecules with opposing spins collide and exchange spin polarization. This will lead to a mixing of the respective hyperfine states. Molecular collision rates increase with radical concentration. Heisenberg exchange leads to a broadening of ESR lines. According to Armstrong and Han, the effect is significant above 0.5 mM and the broadening effect is dominant above 3 mM radical concentrations.^52,55

In a nitroxide radical molecule, hyperfine splitting of the electron energy levels occurs not only due to interactions with the nitrogen spin. Coupling to other spins within the molecule (generally proton spins) adds additional hyperfine lines. If the energy difference of the additional hyperfine structure is smaller than the ESR linewidth, it is unresolved in the spectrum and can lead to inhomogeneous broadening of the linewidth.^56,57

The tumbling rate of the molecules in solution affects the spin–spin interactions. In a rapidly tumbling regime, molecular motion averages anisotropic interactions and only isotropic contributions are relevant, thereby affecting the ESR linewidth. If spin probes are immobilized (e.g., by inclusion into a macromolecule), the reduced tumbling rate leads to a line broadening in the ESR spectrum. Tumbling and translational diffusion also affect the interaction rate of free radicals with target protons, which has an effect on E_max.^47,50,58

Figure 1 shows the chemical structure of the measured spin probes. Besides commercially available tetramethyl-substituted nitroxides (M4, M6–M8), which are well established in many ODNP related publications^12,43,46,47 and provide a point of reference for other compounds, we also tested some bioreduction resistant, tetraethyl-substituted nitroxides (E1–E4). Other compounds such as M1–M5 look promising due to their small size (e.g., tumbling rate and permeability). Substitution of one or both geminal methyls of M2 with cyclic structures (C1, C2) provides a comparison to all other compounds with non-cyclic alkyl substituents. 3,4-Dicarboxy-proxyl (M9) is a product of intracellular hydrolysis of corresponding acetoxymethyl esters, which are well-known brain-targeted probes.⁵⁹ Deuterated and ¹⁵N-labeled versions of some of these compounds (M4, M6, M9, E3) also provide promising results. Finally, some biradicals (B1–B3) are added to the list of investigated spin probes.

Table I lists the investigated spin probes and their measured ODNP properties. The spin probes were dissolved in PBS (Phosphate Buffered Saline), and the pH was adjusted to 7.3. For TEMPO (M4), a 2 mM concentration leads to the highest |E_max| at a moderate power level P.¹² In this range, the radical concentration is large enough for an efficient polarization transfer, without exhibiting a dominant Heisenberg exchange rate, which would increase P_1/2. For good comparability, the same radical concentration of 2 mM was used for all compounds, leading to a 1 mM molecular concentration for biradicals and 2 mM for monoradicals.

The transition with the lowest frequency, which is equivalent to the peak with the lowest B_p in the ODNP spectrum (where the ESR-frequency is held constant), was used for all characterizations. According to Guiberteau and Grucker,¹⁴ this line exhibits the highest |E_max|.

The measured parameters characterizing the ODNP efficiency of the spin probes are E_max and P_1/2 [see Eq. (5)]. Additional measurements of the longitudinal relaxation times T_1,Bp and T_1,0,Bp enable to separate the contribution of the leakage factor f on E_max and the product of the coupling constant ξ and the maximal saturation factor s_max. The influence on T_1,Bp at 2 mM concentration was similar for all measured spin probes, leading to leakage factors f between 0.65 and 0.82.

By inserting f into Eq. (4), the product ξ · s_max can be calculated. In order to determine ξ, the spin probe concentration can be varied and fitted to theoretical models.^50,51 Pulsed electron–electron double resonance (ELDOR) experiments can be performed,⁶⁰ or one can approximate s_max ≈ 1 for very high spin probe concentrations or tethered nitroxides.⁵² We only determined the product of ξ · s_max, partially due to physical constraints such as the maximal RF-power and the limited amount of free radical compounds available to us, but mostly because we are more interested in a qualitative comparison of the compounds at the mentioned concentration.

Perdeuterated ¹⁵N-labeled nitroxides showed the best results, exhibiting high |E_max| at moderate RF-power; compound M9dn, a perdeuterated and ¹⁵N-labeled version of 3,4-dicarboxy proxyl (M9), showed the lowest P_1/2, while M6dn, a deuterated and ¹⁵N-labeled version of 4-hydroxy-TEMPO (M6), showed the highest |E_max| (E_max = −126.2) of the measured compounds. Of the unmodified nitroxides, di-tert-butyl (M3) showed the lowest P_1/2 and M2 exhibited the highest |E_max|. Generally, smaller (lower molecular weight) molecules seemed to perform better, most likely because of a faster tumbling rate and translational diffusion.

Since they offer the possibility of increasing radical concentration without changing the molecular concentration, we also investigated several biradicals. The biradicals with high enough spin–spin coupling of the electrons showed significant line broadening in the ODNP spectrum. More RF-power is needed to saturate the ESR transitions, and P_1/2 is increased significantly [see Fig. 2(a)].

Comparing the maximal enhancement of the different compounds, it becomes clear why TEMPO (M4) and PCA (M7) are commonly used materials for ODNP experiments. While TEMPO, often used for in vitro experiments, exhibits a relatively high |E_max| of 101.9, PCA has a lower |E_max|, but with a lower P_1/2 of 5.3 W, and is thus better suited for low power applications such as in vivo experiments.

In the following, we discuss in detail the various factors that influence the most important physical properties of nitroxides as spin probes in ODNP applications.

Figure 2(b) shows a trend for the monoradicals (dark blue and red data points), correlating E_max and the MW, where molecules with higher MW have lower |E_max| values. For this evaluation, the MW of the ionic form is used for all compounds that exist protonated or unprotonated in the buffered solution at pH = 7.3 to improve comparability. Special cases such as biradicals, isotope-enriched derivatives, or an unsaturated bicyclic compound can deviate from the trend.

The correlation of E_max with the molecular weight is consistent with the theoretical considerations of the local dynamics of the radicals in solution.^50,58 Translational diffusion of the molecules is influenced by the MW. Lower weight leads to faster diffusion, increasing ξ and |E_max|.⁴⁷ 

In contrast, Fig. 2(a) indicates no clear correlation between the MW and P_1/2. Whether this correlation is not existing or just superimposed by other effects cannot be determined from the data obtained.

The P_1/2 values of the monoradicals (with the exception of C1, which will be discussed below) show a clear correlation with the linewidth (FWHM) of the ODNP spectrum [see Fig. 2(c)]. In accordance with the theoretical considerations above, a broadened linewidth FWHM lowers s_max and increases P_1/2, independent of the underlying cause. The biradicals are not included in Fig. 2(c) and are discussed separately below.

The effects defining the ESR linewidth are influencing the ODNP linewidth in a similar manner.^14,61 The lower boundary of the measured FWHM is additionally limited by the finite homogeneity of the polarizing field. M9d and M9dn may provide examples of this limit in Fig. 2(c).

Previously, it was shown that the linewidth and the number of electron spin resonance lines of the nitroxides are influencing E_max.⁶² Figure 2(d) indicates that FWHM is dependent on multiple influences. As described in the theoretical section “Tumbling and translational diffusion”, the tumbling rate of the molecule affects E_max as well as the linewidth. Other effects, such as inhomogeneous broadening due to intramolecular spin coupling, seem to affect FWHM much more than E_max. Compounds E1, E2, E3, and E3d provide good examples for this effect, showing a small variance of E_max, but a broad range of FWHM.

Pyrrolidines with two geminal ethyl groups at positions 2 or 5 of the heterocycle are known to show high hyperfine coupling (∼0.2 mT) with the methylene hydrogens of these groups in their X-band ESR spectra. In low-field ESR or ODNP, this coupling is not resolved and a strong line broadening is observed in the literature and in our measurements.^63,64 Due to the repulsion of geminal ethyl groups with a substituent in the neighboring position (positions 3 or 4) of the ring, less averaging occurs, leading to more pronounced broadening. Comparing E1–E3, this broadening seems to be influenced by the type and number of substituents at positions 3 and 4. However, this influence needs further investigation.

The influence of the geminal substituents becomes especially apparent when comparing similar compounds, where only the geminal methyls and ethyls are exchanged. This is shown in the ODNP spectra of Fig. 3. Here, the FWHM of E2 (FWHM_{left peak} = 104 ± 5 µT) is more than twice when compared to the methyl group counterpart M7 (FWHM_{left peak} = 51 ± 1 µT).

For monoradicals, the data in Figs. 2(a) and 2(c) show a correlation of the type of substituent next to the nitroxide group with P_1/2. In general, compounds with methyl substituents (M1–M9) exhibit lower P_1/2 than the compounds E1–E3 with ethyl substituents. Only E4 (P_1/2 = 6.2 W) has a lower P_1/2 than M6 (P_1/2 = 7.1 W).

In Figs. 2(c) and 2(d), the results for E3d and E4 further illustrate the unresolved coupling caused by the geminal ethyl substituents. As will be discussed below, deuterium has less hyperfine coupling than standard methylene hydrogens.⁵⁷ Radical E4 is based on a 1,2,3,4,5,6-hexahydrocyclopenta[c]pyrrole ring system with a nearly planar geometry, where averaging of ethyl group conformations can occur and therefore no large couplings are observed. As a result, E3d and E4 show much narrower lines and E_max and P_1/2 are improved in comparison to other tetraethyl compounds. A similar spectral peculiarity was earlier reported for sterically shielded imidazolidine nitroxides.⁶³ The nitroxide C1 has a methyl substituent on one side and a spiro-2-(carboxy)cyclopentane moiety on the other side of the nitroxide group. P_1/2 = 4.8 W of C1 fits into the methyl substituent group and could not be expected for a nitroxide with relatively broad lines. The deviating behavior of C1 might be related to relaxation, to fast proton exchange by the carboxy group near the radical, or to the remaining geminal methyl group. However, the dispirocyclic nitroxide dicarboxylic acid C2 did not show any deviation from the general trend, and its ODNP characteristics are similar to those of the tetraethyl compounds. The observed phenomenon deserves further investigation.

Compared to tetramethyl analogs, the four ethyl groups adjacent to the N–O moiety could increase the distance of approach between the free radical electron spin and the targeted protons of water. This would affect the coupling constant ξ and in turn lower |E_max|. The influence of bulky substituents upon the accessibility of the nitroxide group has been studied previously.^39,65,66 These studies showed lower solvent accessible surface areas and a higher volume steric shielding parameter (V_ss) for sterically shielded nitroxides. However, there is no evidence that sterical hindrance affects the access of small molecules, such as water, to nitroxide oxygen. Solvation with water implies an interaction of the polar nitroxide moiety with water protons. The interaction efficacy of sterically hindered nitroxides with water was already demonstrated by the similarity of T₁-relaxivities for organic radical contrast agents based on sterically shielded and tetramethyl-substituted nitroxides.⁶⁷ In our study, the sets of |E_max| data points for tetramethyl- and tetraethyl-substituted nitroxides exhibit a smooth transition, and the observed changes in |E_max| are sufficiently explained by the influence of MW. An effect through a change of accessibility/distance of approach is not observed.

Comparing the pyrrolidine and piperidine based tetramethyl compounds (Fig. 2), the ring structure does not seem to significantly affect E_max. If the ring structure has an influence on E_max, it is not visible in the data since the correlation of E_max with the molecular weight remains the dominating factor.

However, there is a difference in P_1/2. The unmodified tetramethyl pyrrolidines exhibit P_1/2 ≤ 5.3 W and FWHM ≤ 51 µT, while the structurally similar piperidines all have P_1/2 ≥ 6.1 W and FWHM ≥ 55 µT. Similarly, P_1/2 ≤ 3.1 W was observed for all perdeuterated and/or ¹⁵N-labeled pyrrolidines and P_1/2 ≥ 3.5 W was observed for all perdeuterated and/or ¹⁵N-labeled piperidines.

This effect could be explained by an increase in intramolecular interaction within piperidines, which could be caused by an increase in the number of atoms per molecule or by a change of conformational flexibility of the ring. As discussed in the theoretical considerations on saturation and linewidth, any effect that leads to a broadening of lines in the ODNP spectrum will also increase P_1/2.

We measured multiple (per)deuterated and/or ¹⁵N-labeled variants of DCP (M9) and of TEMPOL (M6), marked with “d” and/or “n,” respectively. Additionally, we measured a perdeuterated TEMPO M4 (M4d) and a partially deuterated tetraethyl compound E3 (E3d).

Both variations as well as their combination show a positive effect on the performance of the compound compared to their unmodified counterparts (see Table II).

All deuterated nitroxides show lower FWHM and P_1/2 as well as higher |E_max| than their corresponding non-deuterated analogs despite the latter having smaller molecular weights.

In nitroxide radicals, the hyperfine splitting caused by the interaction of the free electron with the nitrogen atom depends on the nitrogen isotope. The six energy levels of the naturally abundant ¹⁴N nitroxide radicals are reduced to four levels in ¹⁵N-labeled nitroxides.^14,61 Therefore, ¹⁵N-labeled nitroxide radicals exhibit two ESR lines, while samples with unmodified nitroxides (predominantly ¹⁴N) have three ESR lines. This leads to an increased theoretical s_max and therefore improved E_max for ¹⁵N-labeled nitroxides, compared to their unmodified versions. Additionally, a reduced number of hyperfine states reduces the amount of mixing between them, thus resulting in narrower lines and an improved P_1/2 for the ¹⁵N-labeled compounds. Furthermore, ¹⁵N-labeled nitroxide radicals have longer electron relaxation times than the corresponding unlabeled counterparts,⁶⁸ which also reduces P_1/2.⁶⁹ 

The positive effect of (per)deuteration on P_1/2 is a result of the reduced intramolecular coupling. Deuterium has lower hyperfine coupling than hydrogen, reducing the unresolved hyperfine splitting, which in consequence narrows the ESR linewidth and improves P_1/2.⁵⁷ 

Interestingly, the magnitude of the improvement by deuteration or ¹⁵N-labeling on E_max and P_1/2 differs greatly between the different nitroxides and no clear pattern could be determined. Still, deuteration and ¹⁵N-labeling of sterically shielded nitroxides certainly are promising ways to improve the ODNP properties of reduction-resistant probes for in vivo applications.

Here, we compare a monoradical (E4) with two biradicals (B1 and B2) of a similar molecular structure. In addition, we measured B3, which is a rigid biradical with a greater intramolecular distance between the two radical groups.

Since the compared compounds differ significantly in MW, any possible biradical-related effect on E_max is likely concealed by the dominant correlation to the MW.

Figure 4 depicts their respective ODNP spectra. While the monoradical E4 shows a three-peak ODNP spectrum as expected and a P_1/2 below 7 W, the biradicals B1 and B2 show P_1/2 greater than 60 W (see Table I). The ODNP spectra of the latter two biradicals show an extreme broadening of the peaks to a degree that the three peaks are not clearly distinguishable anymore.

We attribute this effect to intramolecular electron spin–spin interactions of the radicals, which in turn leads to inhomogeneous broadening of the electron transitions.

In accordance with this, B3 (biradical) shows a P_1/2 value of 16.5 W. It has a greater distance between the radicals within one molecule, leading to a weaker intramolecular electron spin–spin interaction and thus less broadening of the ODNP spectra.

In order to obtain a deeper understanding of the performance of biradicals in ODNP experiments, a more quantitative study of the magnetic hyperfine-, dipolar-, and g-tensors as well as the exchange interaction of the two unpaired electron spins would be necessary. Such a study was done for the rigid and narrow linewidth biradical TEMPO-bis-ketal, which showed comparable enhancement to monoradicals at 9.2 T.^46,49

The presented spin probes contain a range of compounds that are ionic in aqueous solution at pH 7.3, in particular compounds containing acid or amino groups and compounds E4, B1, and B2 containing a quaternary ammonium group. One could expect a reduced Heisenberg exchange because of repulsion between the ionic nitroxides. Within our data, however, we cannot find a correlation of the ionic charge of a spin probe with their E_max or P_1/2.

The goal of this work is to better understand the various factors that influence the physical properties of free radical spin probes. We investigated a broad range of 26 different nitroxides and their physico-chemical parameters to help in predicting and optimizing the suitability of free radical spin probes for in vivo ODNP applications.

For such applications, a high |E_max| by hyperpolarization is required. While a narrow ESR linewidth seems to facilitate a better E_max, our data suggest that the molecular weight of the spin probe has the greatest impact on its maximal enhancement, indicating a correlation to the tumbling rate and translational diffusion of the spin probe. E_max can be improved additionally by using deuterated and ¹⁵N-labeled derivatives of a chosen compound.

To avoid excessive sample or tissue heating, the transmitted RF-power will be limited in most applications. In such cases, a low P_1/2 ensures sufficient hyperpolarization. A narrow ESR linewidth correlates with low P_1/2 values and can be achieved by limiting intermolecular and intramolecular interactions. Therefore, strong spin–spin interactions in some biradicals can render them useless. Unresolved hyperfine splitting from geminal ethyl substituents next to the nitroxide group can negatively impact the ODNP performance of such sterically shielded nitroxides. Using ring systems with a more planar geometry seems to circumvent this problem but needs further investigation. Pyrrolidines with a five-membered heterocyclic structure showed better P_1/2 than six-membered piperidines, but this finding also needs further investigation. Deuteration and ¹⁵N-labeling of the compounds improved FWHM and P_1/2 for all examples in our study and generally seem to be an effective way of boosting the ODNP-performance of nitroxide spin probes. However, depending on the specific case, a more complex synthesis may be a disadvantage of these modified compounds.

In a nutshell, the ideal nitroxide radical would be a light weight, deuterated ¹⁵N-pyrrolidine monoradical, with neighboring methyl substituents and a narrow linewidth ODNP spectrum.

However, further investigations are needed since we are not looking at the biological context at this stage. Any compound will have to compromise between stability, low molecular concentration, long retention, and low RF-power. While compounds with neighboring methyl groups would need less power for the same level of hyperpolarization, the ones with ethyl groups are expected to be more stable and resistant against reduction in biological environments.⁷⁰ Small monoradicals are usually rapidly excreted, and thus, their short retention in blood might lower their suitability for in vivo ODNP applications. Nanosized delivery systems or the coupling of small radicals to nanosized structures might offer a solution to the problem and can help to protect the radicals against biological inactivation. When coupling multiple small spin probes to such nanosized structures, decoupling of the electron spin–spin interactions of the radicals within one molecule will help to prevent inhomogeneous broadening. Such incorporation in larger structures could also impact the tumbling and translational diffusion of the spin probes. This could lower |E_max| at a given RF-power P but could be mitigated by ensuring a flexible linker between the small radical and nanosized structure to allow for the unhindered movement of the small spin probe.

While technical challenges for ODNP applications differ for different magnetic fields, the coupling factor ξ also changes with the field.² This change is, however, gradual, and therefore, the presented results still provide a point of reference for spin probe selection and spin probe design for applications operating at fields above the ULF-regime.

Ultimately, our study can help to investigate the functionalization of free radicals by incorporation into macromolecules or carrier molecules. The chemical processes involved in such functionalization often require specific chemical features of the radical molecules. Our results provide a list of possible candidates and give a good starting point for these types of experiments.

Compounds M4 and M6–M8 were purchased from MERCK. Nitroxides M2,⁷¹ M3,⁷² M5,⁷¹ M4d,⁷³ M6n,⁷⁴ M6d,⁷⁴ M6nd,⁷⁴ M9,¹⁶ M9n,¹⁶ M9d,¹⁶ M9nd,⁷⁵ E2,^39,76 and E3⁶⁴ were prepared according to literature procedures. The synthesis of nitroxides M1, E4, and B1–B3 is described in the supplementary material. Preparation of compounds C1, C2, E1, and E3d will be published elsewhere. Analytical data for the in-house synthesized nitroxides are shown in the supplementary material.

The solubility of the different nitroxides was not specifically measured but differed slightly during sample preparation. With pH adjustment and sometimes the use of an ultrasonic bath, all samples were soluble in 2 mM concentration at room temperature in PBS.

All measurements were performed at the ULF-MRI setup previously described in Ref. 7. The superconducting quantum interference device (SQUID) based sensor is the heart of the system. In combination with a gradiometric pickup coil, the intrinsic noise level of this SQUID based magnetic field sensor lies in the femtotesla range. The static magnetic field, generated by a tetracoil with a battery-driven current source, was set to B₀ = 92 ± 0.8 µT for all measurements presented here. The polarization field was generated by a Helmholtz coil with a diameter of 166 mm. The RF-signal for the ODNP excitation is transmitted by a resonator consisting of a single loop of a 1 mm thick copper wire with a diameter of 16.1 mm. It is tuned to ω_e = (120 ± 1) MHz.

The whole setup is placed inside a multilayered magnetic- and RF-shielded chamber. All electrical lines needed for the experiments are fed through relays and/or pi-filters in order to suppress amplifier noise during acquisition of the MR signal.

The sample containers consist of modified microcentrifuge tubes and have a volume of 1.2 ml. The geometry was identical for all measurements to ensure comparability across samples. Larger sample volumes yielded a greater signal-to-noise ratio (SNR), but a limited RF-penetration depth leads to larger systematic errors. With the 1.2 ml samples, a relatively uniform saturation of the whole sample could be achieved.

Ahead of specific characterization measurements, two preliminary measurements have to be performed.

First, the polarization field B_p has to be set to the optimal field strength, where the RF-coil frequency corresponds to the hyperfine splitting transition of the nitroxide radical.

In order to determine the appropriate value, the B_p field strength is varied, while all other parameters are fixed. For each B_p step, the nuclear magnetic resonance (NMR) spectrum is recorded via a simple free-induction decay (FID) sequence (see Fig. 5). The area under the peak of each NMR spectrum as a function of B_p is divided by the non-hyperpolarized area under the peak of the respective field strength, which forms an ODNP spectrum of the sample and shows the number of usable transitions from hyperfine splitting as well as the respective line shapes. Each peak is fitted with a Lorentzian function to determine the FWHM from the fit result.

Second, the duration of the hyperpolarization pulse has a critical impact on the sample polarization. By sweeping the polarization time t_Bp, instead of B_p, and setting B_p to the previously determined optimum of the left ODNP peak, the hyperpolarization buildup time T_HP at the polarization field strength can be obtained, by fitting the area under the peak of the NMR spectra as a function of t_Bp to an exponential buildup function.

For the further measurements, a compromise was made between low measurement durations and a high polarization level. Therefore, the hyperpolarization time t_Bp is set to t_Bp,90 = 2.3 · T_HP, where the sample reaches 90% of the maximum polarization.

In order to characterize a sample, the leakage factor f, the product of the coupling constant and the maximal saturation factor ξ · s_max, the maximal theoretically possible enhancement E_max, and the power P_1/2 needed to reach 0.5 · E_max have to be determined.

The leakage factor f is determined from the longitudinal relaxation times T_1,Bp and T_1,0,Bp at the polarization field strength B_p according to Eq. (2).

In order to determine ξ · s_max, E_max, and P_1/2, the power dependency of the enhancement factor has to be measured. This is done by varying the RF-power. The area under the peak of the NMR spectra can be fitted to a model following Eq. (5), which determines E_max,measured and P_1/2. Since we use the hyperpolarization time t_Bp,90, the measured E_max,measured is scaled up to 100% saturation afterward. Inserting E_max into Eq. (4) gives ξ · s_max.

All errors presented in Table I represent the statistical errors, determined from the respective fit results.

In Ref. 50, a measurement intensive method to correct for heating effects is presented, altering the theoretical description above. Our sample volumes are in the ml range. The samples are air stream cooled, and a long repetition time TR is used, in order to avoid heating effects and to reduce the systematic error. Therefore, we are not correcting for heating effects.

We present the characterization of ODNP properties for 26 different nitroxide radicals. The majority of these are novel and bespoke compounds, synthesized at NIOCH SB RAS for this very purpose. Only limited quantities of these substances were available for measurements at our ULF-MRI system.

We found a decrease in P_1/2 and an increase in relaxation times within 2 days of sample preparation (dissolving the radical compound in PBS) when repeating measurements of the same sample. This indicates a decrease in the spin probe concentration^12,44 by degeneration, ruling out repeated measurements of the same samples. The presented samples were therefore prepared and characterized within a day. The degeneration was unexpected for these samples. We suppose that interactions of the radicals with the plastic composition of the modified microcentrifuge tubes as sample containers are the most likely reason. However, further investigation is needed to verify this claim.

The degeneration of samples introduces a systematic error, which depends on the compound’s stability and the age of the sample. Since the radical stability is not the focus of this paper and other methods such as ESR are better suited for such tests, we used exemplary data from the commercially obtained compounds to estimate this error. The data suggested a change of ≤10% of the measured parameters after 2 days and up to ≤25% after 2 weeks of sample preparation.

To assess the precision, we used the commercially obtained compounds (radicals M4, M6, M7, and M8) to repeatedly prepare and measure samples of the same radicals for n = 3 times (see Table II of the supplementary material). The statistical errors (standard deviation) of those repeated measurements were between ±2.1 and ±13 for E_max, ±0.05 and ±0.41 W for P_1/2, and ±0.53 and ±1.21 µT for FWHM.

The supplementary material provides a list of the measured samples, providing their chemical structure, their chemical name with common abbreviations (if applicable), and a reference for their synthesis or the source from where they were purchased. Additionally, it contains the dataset of the repeated measurements of commercially available compounds, from which the precision of the characterization measurements was determined. The supplementary material further describes the synthesis of samples M1, E4, B1, B2, and B3 and contains infrared- and NMR-spectroscopy data of the synthesized samples and gas chromatography–mass spectrometry analyses of compounds M1 and M2.

The financial support of the ERA.Net RUS+ project (No. ST2017-382), NanoHyperRadicals (including RFBR 18-53-76003-ERA-A and BMBF, FKZ: Grant No. 01DJ18009), and the Shanghai Municipal Science and Technology Major Project (Grant No. 2019SHZDZX02) is acknowledged. The authors thank the Multi-Access Chemical Research Center SB RAS for spectral and analytical measurements of nitroxides (supplementary material). P.F. acknowledges the financial support from the Cusanuswerk e.V. The authors thank the DEAL project for promoting open access to research publications.

The authors have no conflicts to disclose.

The data that support the findings of this study are available within the article and its supplementary material, as well as from the corresponding author upon reasonable request. Data on the chemical synthesis of the used compounds are available in the supplementary material, as well as from I. A. Kirilyuk upon reasonable request.