Hyperpolarization of arbitrary solutions using HYPOP; an overview. Fig. 1 illustrates how HYPOPs can be used to hyperpolarize arbitrary solutions. We describe here a first generation of HYPOP materials dubbed HYPOPs-I that feature ca. 100 nm-large aggregated polymer particles forming a monolithic material with open porosity (Fig. 1a). HYPOP-I can be impregnated (Fig. 1b) with a variety of liquids ranging from pure pyruvic acid (the gold standard in hyperpolarized metabolic imaging5) to complex aqueous solutions, such as mixtures of ligands for fragment-based drug discovery,37 or mixtures of metabolites originating from cell extracts.38 The 1H spins of the whole sample are polarized in c.a. 20 min to very high levels exceeding P(1H)> 60%. This polarization originates from the aggregated epoxy particles of HYPOP-I in which PAs are hosted, and spontaneously propagates to the frozen solution by 1H nuclear spin diffusion.39 This high 1H polarization is sheltered on low-gamma 13C nuclear spins of target molecules after brief cross-polarization contacts of a few milliseconds.9 High levels of 13C polarization exceeding P(13C)> 30% can thus be achieved in tens of minutes (represented in fig. 1d), and further stored at low temperatures (3.8 K) for hours in view of transport to a remote point of use (represented in Fig. 1e). The very long 13C relaxation times (up to 5.7 h) are attributed to the fact that paramagnetic relaxation is dramatically attenuated in the pores since 13C-13C nuclear spin diffusion towards the paramagnetic PAs is orders of magnitude slower than for 1H, and further attenuated by the low natural abundance of 13C in the HYPOP-I material. We finally show how the hyperpolarized solutions can be extracted from HYPOP-I with high efficiency leading to hyperpolarized solution of metabolites that show exceptional signal enhancements upon immediate analysis with 13C NMR (Fig. 1f).
HYPOP synthesis and characterization. A considerable number of strategies can lead to polymer networks with open porosity of various sizes,40,41 such as selective degradation of block copolymers assembled in co-continuous morphologies,42 high internal phase emulsions (HIPE),43 colloid templating,44 aerogels or open-cell foams.45 These methods also share some of the following shortcomings: i) synthetic complexity or incompatibility of polymerization strategies with the presence of stable radicals; ii) presence of additives in the materials such as surfactants; or iii) high temperature or chemical treatments incompatible with the survival of the TEMPO radicals. Other strategies, often applied to the synthesis of polymer membranes involve spinodal decomposition between a polymer and solvents through thermal transitions or changes in the solvent composition.46 The porous network can then be obtained through straightforward removal of the solvents, but often displays significant heterogeneities. We rather opted for an analogous method, where spinodal decomposition between the polymer and the solvent is caused by the polymerization process itself. This process is also relatively easy to implement using thermosetting polymers with outstanding thermo-mechanical properties such as epoxy resins. Formulations of epoxy resin, hardener and a non-reactive solvent that display initially full miscibility, and that undergoes phase separation during the curing have been previously studied to form either epoxy materials as dispersed particles,47 solids with closed porosity,48,49 and also solids with open porosities.50
Thanks to a thorough study of the formulation (nature of the non-reactive solvent and relative fraction to the epoxy resin precursors), we were able to construct an extensive phase diagram of such a system and to target network morphologies in close adequacy with our requirements. We selected conventional epoxy resin precursors leading to high Tg thermosets above 150 °C: diglycidyl ether of bisphenol A (DGEBA) and isophorone diamine (IPDA) (Fig. 2a). The mixture was cured at 373K during 24 h in presence of polypropylene glycols (PPGs) of different molar mass acting as non-reactive solvents (detailed synthesis in the “methods” section). Thorough variations of the fraction of non-reactive solvents (from 30 to 90 wt %) and of their molar mass (from 192 to 2000 g mol-1) have been tested in absence of 4-amino-TEMPO (n = 0). This results in the formation of epoxy particles aggregated in a variety of morphologies such as stable latexes, unstable suspensions, gels, monolithic networks with closed and opened porosities. The corresponding phase diagram, presented in fig. 2b, was built from a series of experiments (each dot) and the nature of porosity, opened or closed, was assessed by gravimetry after extraction of the PPG. The functionalization of this resin with commercially available 4-amino-TEMPO is straightforward and was obtained by merely mixing this commercial TEMPO derivative with IPDA in various molar ratios; the amount of DGEBA was adjusted to maintain a proper stoichiometry between amines and epoxydes (nepoxy/nNH2 = 2). Solid samples were characterized, after proper removal of PPG using solvent washes and freeze-drying, by SEM, N2 physisorption and Hg intrusion porosimetry when appropriate. All experimental details are given in the methods section and in the section 2 in the SI.
Spinodal decomposition leading to bicontinuous and homogeneous morphologies, and thus to solids with open porosity after extraction of the PPG, occurred only in a narrow range of solvent fractions (60-90 wt%), and for low molar mass of PPG. The corresponding size of aggregated epoxy particles was critically dependent on the molar mass of PPG, and varied from 1-5 μm, 100 nm and ca. 10 nm for PPG-725, PPG-400 and PPG-192, respectively (See Section 2.3 in the SI). The latter networks were obtained as transparent gels (i.e. not demonstrating extended phase separation between the epoxy and the solvent) and display structures typical of aerogels. Thus, the morphologies of the sample obtained with 85 wt% of PPG-400, showing a bicontinuous morphology composed of ca. 100 nm-large aggregated epoxy particles forming a solid with hierarchical porosity, was particularly well suited to our requirements, with a Tg estimated to be 134 °C and the absence of fracture when immersed in liquid N2. The proper curing of this particular network was monitored with in-situ rheology (Section 2.5 in SI) that enabled the identification of a first phase separation event after 3 h of reaction, followed 30 min later by the formation of a network between aggregated particles. As the storage modulus reached a plateau after about 15 h of reaction, we kept a curing time of 24 h for all samples. After subsequent extraction of the PPG and drying of the porous polymers, we found that the final weights of the solids are, within experimental error, the same as the epoxy-amine precursors. In addition, swelling tests indicate the complete absence of a soluble fraction and therefore a complete cross-linking process.
The epoxy formulation using 85 wt% of PPG-400 was modified with increasing amounts of 4-amino-TEMPO (with r = nTEMPO-NH2/nIPDA ranging from 0 to 1.06) to obtain a series of seven HYPOP-I samples that are further used for the dDNP experiments. In spite of previous results indicating complete reaction of epoxides and amines, the concentration of radicals present in the final materials cannot be directly estimated from the initial amount of 4-aminoTEMPO and was measured by Electron Paramagnetic Resonance EPR (See section 3 in SI). By comparing the effective concentrations of radicals to the concentration of 4-aminoTEMPO initially added (Table S2 in SI), we found a fairly constant and reproducible survival yield of 34 % (free radicals surviving the synthesis). This value, seemingly low, was to be expected from the long curing times (24 H) at 100 °C. This low rate is, however, acceptable in the HYPOP-I series given a) the relatively low cost of precursors and the simplicity of the synthesis and b) the large range of radical concentrations that are still attainable, up to 285 μmol g-1, which goes beyond optimal concentrations required for dDNP (see below). Incorporating such large amounts of 4-aminoTEMPO in the epoxy networks in place of the tetra-functional IPDA can yet induce significant changes in the final materials morphology due to i) the decrease of both cross-link density and Tg in the epoxy network and ii) changes of solubility parameters between epoxy and PPG that govern the spinodal decomposition and the final morphology of the materials. While an inspection of the HYPOP-I series with SEM (Section 2.4 in SI) hardly displays evidence of variations in the morphology, a more thorough analysis involving mercury intrusion porosimetry indicates noteworthy changes in the pore size distribution (Section 2.6 in SI). While all samples display similar distribution of pore sizes in the 10-100 nm range, i.e. within the interstices of the epoxy particles, micrometric pores corresponding to voids between particle aggregates are progressively disappearing when the amount of 4-amino-TEMPO is increased. Concomitantly, the total porous volume decreases from about 3.5 to 1 mL g-1. We believe this to be due to an increased compatibility of the TEMPO-rich epoxy with the PPG, and thus a more extensive plasticization and susceptibility to pore collapse upon removal of the solvent. Complementary analyses with nitrogen physisorption (Fig. S4 in SI) also confirm that these networks are essentially macroporous, with pore sizes above 50 nm.
HYPOP impregnation with aqueous solutions. While HYPOP-I samples could be completely backfilled by immersion in a few centimetres of water or aqueous solutions, their low surface energy does not allow for spontaneous capillary impregnation. We resorted to use water/ethanol mixtures (9/1 v/v) for all impregnating solutions to circumvent this issue. Swelling of the HYPOP matrix is another important parameter to consider that could affect the PA concentration either by scavenging the radical51, or simply by increasing the volume of the polymer and therefore decreasing the radical concentration and the dynamics of polarization. While a few organic solvents demonstrated extensive swelling, the water : ethanol mixtures used in this paper induced a moderate swelling of 19% (See Table S3 in SI).
In a final step, the HYPOP-I monoliths were ground and sieved into ca. 250-500 μm powders (Fig 2c). The initial polarizability of 1H spins in the HYPOP-I series was first characterized in absence of an impregnated solution to optimize the PA concentration necessary for optimum DNP conditions. In the remaining study, wet dDNP samples were obtained by slightly under-impregnating 20 mg of HYPOP-I powder with 60 μL of solutions.
1H polarization with HYPOPs. DNP was performed at 7.05 T and 1.4 K in a Bruker prototype dDNP polarizer according to a standard DNP protocol described in the “methods” section. The 1H polarization kinetics followed a conventional first-order mono-exponential increase for all samples. Corresponding final 1H polarization values and build-up rates are presented in blue in Figs. 3a and 3b, respectively. A maximal 1H polarization of about 40 % was reached for HYPOP-I samples containing radicals at concentrations between 60 and 120 μmol.g-1 while the build-up rates increased continuously with the radical concentration. Such behavior was previously observed in HYPSO materials, which demonstrated optimal polarizations at radical concentrations between 50 and 100 μmol g-1.31
After impregnation of HYPOP-I samples with 1H concentrations ca. 11 mol.L-1 (D2O/H2O/ethanol-d6 mixture (8/1/1 v/v/v)), we performed DNP experiments with the same DNP protocol which led to a similar optimal PA concentration, but significantly higher maximal levels of polarization of about P(1H)=55% (Figs. 3a and 3b in red dots). This increase in polarization to levels beyond those of dry HYPOP-1 samples was unexpected. One possible explanation for this phenomenon would be that the PA are to some extent heterogeneously distributed across the aggregated particles forming HYPOP-I, some particles exhibiting a lower 1H polarization than others and, being connected to the others through a tortuous path, are unable to polarize by long-range 1H spin diffusion. However, once the porous polymer is impregnated with a 1H containing solution, 1H spin diffusion across the whole sample is strongly facilitated, which may help to reach proper hyperpolarization in all parts of the sample. In addition, the 1H polarization kinetics diverge significantly from first-order which indicates heterogeneous build-up kinetics. This feature is typical for DNP build-up curves when polarization in the vicinity of the radicals is coupled to significant long-range spin-diffusion into radical-poor domains (the frozen solutions in our case).27 In order to provide comparable build-up rates, the 1H polarization kinetics were fitted with stretched exponentials
with β being the breadth of the distribution of build-up rates. The average build-up rate R1 is defined as:
with Γ the gamma function.52 Both dry and impregnated matrix build-up rate follow a similar trend, albeit understandably slower for impregnated HYPOP-I, than for empty materials. The optimal formulation chosen for the rest of this work, HYPOP-IA containing radicals at 95 μmol g-1, provides fast and extensive 1H polarizability to the frozen solution.
Generating and storing 13C hyperpolarization in HYPOP-IA. The HYPOP-IA sample was impregnated with a D2O:ethanol (9:1 v/v) solution also including 1M [13C]urea, 1M [1-13C]glycine, 1M [13C]sodium carbonate and 1M [13C]sodium formate (four distinct target molecules) and 150 mM of sodium ascorbate. The latter is a reducing agent that readily scavenges paramagnetic oxygen in the aqueous solution as well as the PAs at the surface of the porous polymers and thus attenuates the corresponding paramagnetic relaxations.51 The quick and efficient 1H polarization generated in HYPOP-IA and in the impregnated solution, reaching about P(1H) = 53% after a 2 min build-up time (Fig. 4a), is efficiently transferred to 13C spins by 1H→13C multiple-contact CP9 repeated every 4 minutes, and leads to a polarization of the 13C target molecules of about P(13C) ~25 % with a 7.8 min build-up time (Fig 4b). Sequence is detailed in the methods section as well as in Section 4.2 in the SI.
Yet, the major innovation in our system does not only consists in high 13C polarizability, but also and primarily in the extended hyperpolarization lifetime T1(13C). The lifetime of 13C polarization was assessed from the 13C relaxation at 3.8 K and 7.05 T (Fig. 4c), monitored over 15 hours by 13C detection with small flip angle pulses every 30 min. The pulse angle was chosen to be ∼5° prevent excessive perturbation of the hyperpolarization. The relaxation time constant T1 was estimated to be about 5.7 ± 0.1 h, taking into account the slight depletion of polarization due to the radiofrequency pulses (see section 8.4 in the SI).
Such a long T1 relaxation time is unprecedented in the case of an arbitrary solution of target molecules and paves the way towards transport of hyperpolarized frozen solutions. As we previously demonstrated transport can in principle be achieved using a helium Dewar, equipped with an assembly of permanent magnets providing a moderate magnetic field (1 T).22,27 Different strategies for fast transfers22 have been reported, which could also be implemented in the future in combination with our approach.
Dissolution and analysis of hyperpolarized solutions from HYPOP-IA. Dissolution was performed by standard dDNP methods,1 using superheated D2O (180 °C) injected directly after CP as previously described,9 in the sample holder in the polarizer (1.4 K). The impregnated HYPOP powder was flushed out of the cryostat during the process together with the hyperpolarized solution. Therefore, an additional inline filter (See section 6 in the SI) was used to retain the HYPOP powder while letting though the melted hyperpolarized solution. The extracted hyperpolarized solution was transferred into a benchtop 80 MHz NMR apparatus (Bruker F80) equipped with a dedicated injector (Bruker BioSpin prototype), coupled to a conventional 5 mm NMR tube and ready for analysis. The whole transfer (dissolution, filtration and injection in the NMR tube) up to the start of the NMR pulse sequence took approximately 1.7 s. A magnetic tunnel22 (4 mT solenoid) was implemented along the transfer capillary to prevent excessive polarization losses at low field, except at the position of the HYPOP filtration device (we believe that minor technological improvements solving this shortcoming may considerably increase the overall performance of the method).
After “dissolving” a solution of 0.95 M [1-13C] sodium acetate, 0.93 M [13C]sodium formate and 0.92 M [13C]glycine hyperpolarized in HYPOP-IA, the immediate recording of 13C spectra gave an intense 13C spectrum (Fig 5a, 1 scan). The asymmetry of the 13C signals indicates a substantial proton polarization of the J-coupled 1Hs.53 Enhancement factors were calculated (see section 8.5 in the SI) to be ca. 12’000 for 13C acetate and 6’200 for [1-13C]-formate (i.e. 2 and 1 % absolute polarization respectively). Time-resolved 13C spectra were measured every 5 seconds with 5 ° nutation pulses (Fig. 5b). T1(13C) relaxation time constant of about 90 s for sodium acetate and of about 24 s for sodium formate were observed. These values are within the range expected for 13C polarization in the absence of paramagnetic electron spins54, and demonstrate the efficient removal of HYPOP particles upon filtering. Surprisingly, 13 C-glycine signals could not be observed. We believe that the 13 C-glycine spins exhibited a complete relaxation in the low-field regions during transfer, most probably in the filtration system in absence of a magnetic tunnel.
The final polarization values reported here do not yet match state-of-the-art values obtained with conventional dDNP sample formulation (ca. 50 %9), and a lengthy work of optimization is certainly still pending. However, this work illustrates the great potential of dDNP using HYPOPs as polarization generation, storage and transport matrices which paves the way to a widespread use of dDNP in NMR or MRI experiments. Further technological improvements on the transfer line (e.g. magnetic tunnel over the whole line) still need to be implemented and may help to further increase the performance of the method.