Ultra-low field ¹³C MRI of hyperpolarized pyruvate

Abstract

Medicine is evolving beyond therapy largely predicated on anatomical information and towards incorporating patient-specific molecular biomarkers of disease for more accurate diagnosis and effective treatment. The complementary combination of hyperpolarization by spin-lock induced crossing signal amplification by reversible exchange (SLIC SABRE) and low field magnetic resonance imaging (MRI) can enable accessible metabolic imaging to advance personalized medicine. Hyperpolarized ¹³C-enriched pyruvate has demonstrated promise for imaging metabolism in cancer, heart disease and neurodegenerative disorders; however, broader clinical adoption awaits validated clinical indications, and is further constrained by the cost and limited availability of current hyperpolarization technology. Parahydrogen-based polarization techniques, paired with low-cost high-performance MRI at millitesla fields, offer a means of broadening the reach of metabolic imaging. Here we show results demonstrating in situ hyperpolarization of pyruvate at 6.5 mT by SLIC SABRE, followed by immediate readout without field cycling or sample shuttling. We achieve ¹³C signal enhancements several million times above thermal equilibrium at 6.5 mT, corresponding to polarization levels of approximately 3%. Leveraging this enhancement, we perform ¹³C MRI and acquire NMR spectra with resolution sufficient to distinguish chemical shifts between pyruvate isotopomers. These results show a viable pathway towards accessible metabolic imaging with hyperpolarized ¹³C MRI at ultra-low field.

Hyperpolarized ¹³C-enriched pyruvate has demonstrated promise for imaging metabolism in cancer, heart disease and neurodegenerative disorders; however, broader clinical adoption is constrained by the cost and limited availability of current hyperpolarization technology. Here, the authors report in situ hyperpolarization of pyruvate at 6.5mT by SLIC SABRE, demonstrating its application in low-cost high-performance ¹³C MRI at milli Tesla fields with sufficient resolution to distinguish chemical shifts between pyruvate isotopomers.

Introduction

While changes in cellular metabolism are associated with many diseases, the majority of diagnostic imaging only measures macroscopic anatomical changes that occur after disease progression or treatment response. This paradigm is especially prevalent in cancer, where altered energy metabolism fuels unrestricted cell growth and change in tumor volume is the dominant measure of disease progression¹. In addition to helping clinicians better delineate tumor boundaries, metabolic imaging can aid cancer treatment by grading tumor malignancy and monitoring treatment response². To enable agile personalized medicine, there is a need for fast, accessible and nonionizing metabolic imaging.

Magnetic resonance imaging (MRI) is the premier imaging modality for soft tissue contrast with high spatial resolution, free from ionizing radiation. Beyond anatomical information, MRI also offers insight into brain activity, the microstructure of tissues and blood perfusion by providing functional, diffusion and dynamic contrast modalities.

MRI can also perform molecular imaging via magnetic resonance spectroscopy (MRS) and chemical exchange saturation transfer (CEST). These techniques have shown promise for imaging metabolic biomarkers in cancer but the inherent sensitivity limits of MRI reliant on thermal Boltzmann polarization of nuclear spins have hampered widespread clinical adoption of these techniques, which are technically challenging to implement within clinically acceptable time frames^3–6.

An emerging alternative is hyperpolarized MRI, which overcomes the sensitivity limits of MRI by boosting the nuclear polarization of a sample of spins many orders of magnitude above Boltzmann equilibrium. This process enables metabolic imaging within seconds following injection of a hyperpolarized metabolite, where the downstream metabolic products are detected in real time with chemical shift imaging. ¹³C MRI for metabolic imaging is in over 50 clinical trials for translation and has demonstrated utility for imaging metabolic hallmarks of disease, especially cancer^7–10. Pyruvate is the most widely utilized metabolite, due to its central role in several metabolic pathways and its favorable relaxation properties. ¹³C pyruvate has shown promise as a hyperpolarized MRI contrast agent for measuring the metabolism of cancers of the prostate^11,12, pancreas¹³, liver¹⁴, kidney^15,16, breast^17,18, and brain^19,20.

However, clinical research with hyperpolarized ¹³C MRI has been largely confined to a small number of centers worldwide. To date, there remains insufficient evidence that hyperpolarized ¹³C MRI improves diagnosis or guides treatment decisions across established clinical pathways. Establishing validated clinical indications is therefore the primary requirement for widespread adoption. In parallel, the cost and limited availability of current hyperpolarization platforms—predominantly based on dissolution dynamic nuclear polarization (DNP)—constrain the scale and throughput of the clinical and translational studies needed to generate this evidence²¹. Polarizers using dissolution DNP typically hyperpolarize samples by transferring spin polarization from electron spins at magnetic fields of several Tesla and temperatures of a few Kelvin. This is followed by rapid dissolution and sample extraction by the injection of high-pressure superheated water into the polarizer where it can freeze, blocking or breaking the fluid path. These practical limitations lead to slow throughput and high costs for DNP-based workflows (e.g., ~1 h per sample for commercial systems costing > $2M), restricting the breadth of research required to validate clinical utility and define robust use cases for ¹³C MRI.

Parahydrogen-induced polarization (PHIP) offers an affordable and fast alternative to DNP. Polarizers using PHIP methods can be built at a cost of $10k and generate hyperpolarized samples within a few seconds^22–24. Spin order from the parahydrogen (pH₂) singlet state is used to boost the MRI signal of molecules either by direct hydrogenation of unsaturated precursors^25,26 or via a reversible chemical reaction in a process called signal amplification by reversible exchange (SABRE)²⁷. PHIP techniques lagged DNP technology for generating hyperpolarized ¹³C metabolites, as reaching comparable ¹³C polarization levels in nontoxic metabolite solutions proved challenging. Preclinical polarizers based on PHIP side arm hydrogenation (PHIP SAH) are now commercially available²⁸ and clinical devices are on the horizon. The PHIP SAH approach involves costly and difficult development of individually optimized chemical reaction pathways between precursor-metabolite pairs. Consequently, pyruvate is the only metabolite currently available from this first generation of commercial PHIP SAH polarizers. In comparison, SABRE offers a more flexible process where the hydrogenation step is reversibly carried out with the aid of the SABRE catalyst. The reversible and repeatable SABRE process also offers opportunities outside single-shot delivery of biocompatible solutions, enabling experiments that benefit from many repetitions of the hyperpolarization step over time. Recent advances in SABRE and especially spin-lock induced crossing SABRE (SLIC SABRE) have demonstrated high polarization of ¹³C metabolites^29–31 and compatible methods for sample purification^32–34. These breakthroughs position SLIC SABRE as a fast, flexible, accessible method poised to offer unprecedented access to ¹³C metabolic imaging.

Combining these new breakthroughs in fast, accessible and low-footprint hyperpolarization with the complementary advantages of ultra-low field (ULF) MRI opens a pathway to broad adoption of metabolic imaging. ULF MRI is the perfect partner technology to SLIC SABRE for achieving the ultimate of broadening the impact of metabolic MRI. To date, the inaccessibility of metabolic MRI has been compounded by the drawbacks of high-field MRI, where costs and exacting siting requirements restrict universal access. ULF MRI is a flexible, cost-effective alternative that can be deployed in environments where high-field MRI is impossible^35–37. ULF MRI has demonstrated utility in the diagnosis of stroke^38,39, hydrocephalus⁴⁰ and neurodegenerative disease⁴¹ and has the potential to enable transformative portable screening paradigms^42–44.

The limitation of ULF MRI is the inherently low sensitivity of inductive detection of weakly polarized spins. Hyperpolarization methods decouple spin polarization from thermal equilibrium at the detection field and offer a way to overcome the SNR limits of ULF MRI³⁵. Hyperpolarized ¹³C offers the additional advantage of background-free contrast. A further motivation for pursuing SABRE hyperpolarization at low field is that it can be combined with fields matching conditions for the generation of spin polarization. For example, 6.5 mT is the ideal field to perform ¹H-SABRE hyperpolarization of molecules, such as pyridine and pyrazine^27,45. Adding ¹³C SLIC SABRE hyperpolarization expands the tool chest, uniquely positioning our platform at 6.5 mT as a test bed for exploring hyperpolarization dynamics in the mT regime.

Here we show rapid hyperpolarization of ¹³C-enriched pyruvate in an ULF open-access MRI scanner. Within 10 s, SLIC SABRE generates enhancements in ¹³C polarization of pyruvate at 6.5 mT of approximately 6 orders of magnitude. The resulting high-resolution spectra are sufficiently narrow to detect chemical shift and J-coupling effects and reveal features indicating both catalyst-bound and free hyperpolarized pyruvate species. We characterize the build-up and T₁ decay dynamics of SLIC SABRE at 6.5 mT for pyruvate and show the optimum spin-lock parameters for generating transverse ¹³C magnetization on pyruvate from singlet order. Leveraging this optimization, we demonstrate ¹³C MRI in two modes—a single-shot sequence utilizing one hyperpolarization step compatible with an in vivo imaging paradigm and a multiple-shot sequence for interrogating emergent polarizer dynamics.

These results demonstrate the flexibility and utility of SLIC SABRE, especially for MRI sequence development when deployed as a hyperpolarization technique that can repeatedly re-polarize a sample within the sensing region for iterative optimization. Our results motivate future in vivo experiments performing ¹³C MRI at ultra-low field and have the potential to enable accessible metabolic imaging to accelerate the transition towards personalized medicine.

Results

SLIC SABRE enhances ¹³C signal by > 10⁶× at ULF

We begin our results showing hyperpolarized pyruvate enhancement above thermal equilibrium at 69.5 kHz, alongside diagrams of the SABRE reaction and experimental setup, in Fig. 1. The reversible chemical exchange reaction that underpins the SLIC SABRE results shown here is depicted in Fig. 1B. At its heart is the iridium-based SABRE catalyst that acts to reversibly bind both hydrogen and pyruvate, facilitating the transfer of spin order from pH₂ to ¹³C driven by SLIC. Bulk hyperpolarization is achieved through a repeated cycle of fresh pH₂ binding, transfer of spin order to unpolarized pyruvate, and the release of the orthohydrogen and spin-polarized pyruvate. For this cycle to result in significant polarization of ¹³C pyruvate free in solution, there are four broad requirements: Firstly, a supply of parahydrogen to fuel hyperpolarization and to displace orthohydrogen by continuous bubbling. Secondly, appropriate concentrations of precatalyst, stabilizing ligand and substrate in solution. Thirdly, correct radiofrequency (RF) spin-lock frequency and amplitude. Fourthly, and finally, favorable chemical exchange rates for both pH₂ and pyruvate. Appropriate sample composition has been extensively characterized previously⁴⁶ and we selected a solution composition of pyruvate, DMSO and SABRE precatalyst in deuterated methanol in suitable concentrations determined by previous work, described in “Materials and Methods” below. Our experimental setup, providing control over the other 3 requirements—pH₂ delivery, sample temperature and RF—is shown in Fig. 1A and described in the “Methods” section. Briefly, pH₂ is bubbled through the SABRE sample inside the 6.5 mT scanner via a capillary that ends in the bottom of the sample tube. The pH₂ flow rate is controlled by a gas handling circuit outside the RF-shielded enclosure of the scanner. RF excitation is performed by a solenoid coil wrapped around a hollow former, through which coolant flows to control the temperature of the SABRE solution.

Fig. 1. SLIC SABRE hyperpolarizes pyruvate at 6.5 mT.

A Experimental setup schematic, showing a front-on view of the inside of the 6.5 mT MRI scanner and an expanded cross-section of the variable-temperature 69.5 kHz NMR probe. Temperature and pH₂ flow are controlled by cooling water and hydrogen gas circuits that extend outside the RF-shielded scan room. B Chemical exchange between the Ir-based organometallic complex and free pH₂ and pyruvate in solution facilitates bulk hyperpolarization. Dashed red arrows indicate the spin-lock driven transfer of spin order. C Pulse sequence diagram of a generalized SLIC SABRE experiment. During steady pH₂ bubbling a spin-lock pulse is applied on resonance with ¹³C at an amplitude corresponding to an effective field close to the J-coupling between the hydrides bound to the Ir complex. Magnetization builds up in the transverse plane along the vector of the spin-lock in the frame rotating at the ¹³C Larmor frequency. Following the spin lock, magnetization is immediately detected as a free-induction decay. D Spectra of hyperpolarized deuterated [1-¹³C]pyruvate and thermal [1-¹³C]acetic acid. The pyruvate spectrum was acquired in a single 32 s acquisition following 120 s SLIC SABRE with 24 sccm pH₂ flow through a sample containing 11.4 μmol pyruvate. Insert shows the [1-¹³C]acetic acid spectrum acquired with 1024 averages of a 45^∘ pulse every 15 s applied to a 31.4 mmol sample.

The general sequence for a SLIC SABRE experiment is shown in Fig. 1C. Following catalyst activation, with the sample cooled to 4.4 °C and pH₂ bubbling continuously, the application of a spin-lock pulse at the ¹³C frequency with an amplitude corresponding to the J-coupling between the hydride groups and ¹³C nucleus of the bound pyruvate substrate causes net magnetization to build up in the transverse plane, aligned with the spin-lock vector in the frame rotating at the ¹³C Larmor frequency. An example frequency spectrum of SLIC SABRE hyperpolarized [1-¹³C]pyruvate-d₃ is plotted in Fig. 1D, alongside an inset showing the thermal signal from a [1-¹³C]acetic acid sample.

The signal enhancement here is an estimate calculated by comparison to a larger, highly concentrated ¹³C-enriched acetic acid sample, described in detail in Supplementary Section 1. This enhancement represents an increase in ¹³C polarization of 6.2 × 10⁶ for deuterated pyruvate. This corresponds to ¹³C polarization of approximately 3%. While this ¹³C polarization is lower than has been achieved with SABRE under other conditions, here we employed a low pH₂ flow rate to minimize solvent boil-off and displacement from the NMR coil, allowing for experiments with a single SABRE sample over several hours. Hyperpolarization at higher pH₂ flow rate is anticipated to yield ¹³C polarization values in excess of 10%^32–34.

Pyruvate ¹³C imaging with SLIC-shot MRI

Having achieved in situ hyperpolarization of pyruvate at 6.5 mT, we move to show results leveraging SLIC SABRE- induced hyperpolarization for ¹³C MRI of pyruvate at 6.5 mT. Of the pyruvate molecules investigated, [1-¹³C]pyruvate-d₃ was identified as the best for exploring ¹³C MRI, as it has the longest T₁ and reaches the highest in situ SLIC SABRE polarization (see Supplementary Sections 2, 3, and 5 for SLIC SABRE optimization and spin relaxation details). Envisaging future in vivo metabolic imaging, we begin by demonstrating ¹³C MRI in a single-shot mode in Fig. 2, where hyperpolarization is generated at the beginning of the experiment and stored for sampling later. The imaging sequence we developed for our ULF scanner is an adaptation of the 3D balanced steady-state free precession (bSSFP) sequence we have previously tailored to maximize SNR and contrast at ULF^35,47. The sequence diagram in Fig. 2 highlights the addition of a SLIC SABRE hyperpolarization step that precedes the repeated bSSFP spatial encoding block. To preserve the transverse magnetization generated by the spin-lock, a 90^∘ pulse with orthogonal phase is applied immediately following SLIC to store magnetization along the longitudinal axis. To sample that stored magnetization steadily, a series of pulses of increasing tip angle follows, each of which tips part of the remaining magnetization back into the transverse plane⁴⁸. We term this method SLIC-shot bSSFP, noting that the sequence is more pedantically “bSSFP-like”, with steady state magnetization a goal requiring careful sampling of the remaining available magnetization rather than simply reached after many pulses with a freely precessing magnetization vector, as is more typical in bSSFP.

Fig. 2. SLIC-shot ¹³C MRI at 6.5 mT.

Top row: 512 × 64 × 5 3D ¹³C MRI of [1-¹³C]pyruvate-d₃ following a 120 s SLIC pulse with 24 sccm pH₂ bubbling and 50% undersampling. Middle row: left, MRI sequence using a single SLIC-shot to generate ¹³C hyperpolarization, followed by a bSSFP-like sequence with a variable tip angle to sample the hyperpolarized signal in a near-steady fashion. Center, the region inside the solenoid of the variable temperature NMR probe in schematic cross-section. Right, Co-registered ¹³C and ¹H 3D MRI images, with the ¹H image a single slice of 3D bSSFP at 1.6 mT and the ¹³C image from above overlaid. Bottom row: 3D bSSFP of ¹H performed by ramping B₀ down to 1.6 mT to bring protons in the cooling water of the variable temperature probe on resonance at 69.5  kHz. All MRI data were zero-filled by a factor of 2 in the readout, and first phase encode directions and normalized in units of the noise floor by dividing the signal in each voxel by the standard deviation of the noise in a signal-free region. The peak signal-to-noise ratio (SNR_peak) of the ¹³C MRI is 293, while the mean SNR inside the region of the NMR tube in the central slice is 204.

The ¹³C images in Fig. 2 were acquired with SLIC-shot bSSFP in 17 s following 120 s of SLIC SABRE hyperpolarization. To make experiments reliably repeatable, and as there is relatively little magnetic susceptibility distortion at ULF, pH₂ bubbling was maintained throughout at 24 sccm. This comes at the cost of minor motion artifacts, most pronounced in the second phase encode direction. The pyruvate sample is approximately 12 μmol in 400 μL of methanol. The scale bar is estimated based on the 3.8 mm inner diameter of the 5 mm NMR tube.

To provide context and create an image in analogy to an anatomical background, we also show a ¹H image of the variable temperature NMR probe cooling circuit acquired by ramping B₀ down to 1.6 mT to bring protons on resonance at 69.5 kHz. We coregister the central slices from the ¹³C and ¹H MRI data to show the hyperpolarized pyruvate solution in place in a 5 mm high-pressure NMR tube, inside the 10 mm water-filled tube in thermal contact with the coolant-filled variable temperature NMR probe body.

While in vivo metabolic imaging requires fast imaging sequences that efficiently sample the available nonrenewable magnetization, the design of polarizer and reactor technology able to produce larger volumes of hyperpolarized material efficiently can also benefit from slower imaging sequences that reveal emergent dynamics. With this secondary focus, we developed a gradient echo (GRE) sequence leveraging multiple spin lock hyperpolarization steps, as shown in Fig. 3. The multi-SLIC-shot MRI sequence incorporates a SLIC pulse for each phase encoding step within a 3D GRE sequence, drawn in Fig. 3A. We note that there are no hard pulses in this imaging sequence as the spin-lock generates transverse magnetization directly. This initializes each line of k-space encoding with a steady ¹³C magnetization, eliminating artifacts from underlying signal variation without the need for a variable tip angle approach or k-space filtering. Example imaging results from the multi-SLIC-shot sequence, displayed in Fig. 3B, capture the standing pattern of hydrogen bubbling at a flow rate of 24 sccm. As with the single SLIC-shot MRI results, there are motion artifacts apparent, especially in the 2nd phase encode direction. However, characteristics of the in situ polarization method can be identified that show nonuniform polarization within the NMR tube and sample being forced out of the detection coil, highlighting opportunities to further improve hyperpolarization and sensitivity.

Fig. 3. Multiple SLIC-shot ¹³C MRI at 6.5 mT.

A MRI sequence for multiple SLIC-shot 3D gradient echo. A SLIC pulse is applied to generate fresh hyperpolarized signal for every line of k-space. B 576 × 64 × 5 3D ¹³C MRI of [1-¹³C]pyruvate-d₃ with the multi-SLIC-shot sequence in (A) and 50% undersampling, 10 s SLIC per shot, and 24 sccm pH₂ bubbling. As the parahydrogen bubbling creates a constantly renewing helical pattern of bubbles in the tube, that shape is captured in the ¹³C MRI image despite the 27-min total imaging time. All MRI data were zero-filled by a factor of 2 in the readout, and first phase encode directions and normalized in units of the noise floor by dividing the signal in each voxel by the standard deviation of the noise in a signal-free region. The peak signal-to-noise ratio (SNR_peak) is 613, while the mean SNR inside the region of the NMR tube in the central slice is 270.

High-resolution spectroscopy resolves J-coupling and chemical shift at ULF

Looking forwards from our proof of principle ¹³C MRI demonstration towards accessible metabolic imaging at ULF, we present high-resolution pyruvate spectra acquired at 6.5 mT in Fig. 4. While our ¹³C MRI results showcase an important step towards hyperpolarized ULF ¹³C metabolic imaging, in vivo studies tracking metabolism will require a means of distinguishing downstream products after injection. Conventionally, in the few-Tesla magnetic field regime, chemical shift readily separates the metabolic products of pyruvate from each other in frequency. In comparison, as chemical shift is proportional to B₀, the splittings in spectra acquired at ULF are dominated by J-coupling. For example, the chemical shift separation between [1-¹³C]pyruvate and [1-¹³C]lactate is 12.2 ppm, corresponding to 0.9 Hz at our field. However, the hyperpolarized ¹³C spectra shown in Fig. 4 resolve chemical shift between pyruvate species and highlight the opportunities for CSI and spectroscopy in the mT regime.

Fig. 4. SLIC SABRE enabled NMR spectroscopy at 6.5 mT.

¹³C spectra of [1-¹³C]pyruvate-d₃ (blue), [1-¹³C]pyruvate (red), [2-¹³C]pyruvate (yellow) and natural abundance pyruvate (green). Splittings are dominated by J-coupling to the protons on the methyl group, creating a quartet of peaks for each ¹³C site. All ¹³C-enriched spectra were acquired in a single shot following 120 s of SLIC. The natural abundance spectrum was acquired with 64 averages of 20 s SLIC. SLIC frequency and amplitude were selected to maximize the integrated signal for each spectrum. All spectra were acquired at 24 sccm pH₂, and 4.4 ± 0.2 °C and are normalized and offset for clarity.

Deuterated [1-¹³C]pyruvate has the simplest spectrum, with a single primary peak with a full width at half maximum (FWHM) of 0.46 Hz. The fine features to either side of the primary peak are from a small fraction of pyruvate with a protonated methyl group. Protonated pyruvate exhibits a quartet of peaks associated with J-coupling to the methyl group protons. This splitting is absent from the deuterated pyruvate spectrum, where the J-coupling to the spin-1 deuterons is much weaker. The FWHM of the individual sharp peaks in the protonated spectrum is 0.16 Hz. We attribute the fine details in our [1-¹³C]pyruvate spectra to imperfect enrichment and/or exchange of protons and deuterons occurring over time⁴⁹. To investigate other potential explanations, we performed additional NMR spectroscopy at 11.7 T on SABRE solutions after measurement at ULF, detailed in Supplementary Section 4.

The [2-¹³C]pyruvate spectrum has the same quartet of peaks as [1-¹³C]pyruvate, split further apart by stronger J-coupling between the ¹³C and methyl protons. We attribute the additional broad features to catalyst-bound pyruvate. These features are most apparent when inspecting the inner peaks of the [2-¹³C]pyruvate spectrum, but we think it likely features associated with bound pyruvate are present and unresolved in other spectra. The bound pyruvate peaks are broader and chemically shifted downfield from the peaks we attribute to free [2-¹³C]pyruvate, whereas bound [1-¹³C]pyruvate is expected to shift upfield, and to a lesser degree³⁰. The peaks associated with free pyruvate are narrowed by comparatively rapid molecular tumbling. The distinction is clearer to see on [2-¹³C]pyruvate, as the chemical shift difference between bound/free pyruvate is greater, stronger J-coupling helps separate all features, and there is a lower fraction of free hyperpolarized pyruvate. We note the separation between the center of [1-¹³C]pyruvate and [2-¹³C]pyruvate spectra is from the different chemical shift experienced at these carbon positions.

The spectrum of natural abundance pyruvate helps to intuit the separate contributions from individual ¹³C spins that occur, even in small impurities in the enriched spectra. A convolution of the [1-¹³C]pyruvate and [2-¹³C]pyruvate spectra accounts for all the features present in the spectrum from the unenriched sample, with the relative weighting of each contribution a function the ratio of the enhancements of each species after 20 s of SLIC SABRE. To account for the nearly one hundred times lower ¹³C signal in the natural abundance sample, signal averaging was performed, and a shorter acquisition duration employed to reduce the overall experimental time to a reasonable 30 min. Although our primary focus here is upon understanding the hyperpolarization mechanisms and characteristics of pyruvate spectra at ULF in the context of metabolic imaging, we take the opportunity to suggest that the ability to collect J-coupling resolved spectra of natural abundance ¹³C molecules at ULF with μmol quantities of material could be highly advantageous for applications in chemical screening.

Discussion

Our results have demonstrated the utility of ULF MRI as a test-bed for SABRE ¹³C hyperpolarization experiments, broadened our understanding of hyperpolarization mechanisms and optimal conditions for the mT regime, shown pulse sequence development leveraging ¹³C signal enhancements >10⁶ for 3D MRI of pyruvate, and reported chemical shift resolution and spin-lattice relaxation times compatible with spectroscopic imaging in vivo. Taken altogether, these findings demonstrate the clinical potential of ULF ¹³C MRI for the first time. We now discuss the advantages of the mT regime, the challenges and opportunities facing imaging metabolism at ULF and the necessary steps ahead for clinical translation.

SABRE in the mT regime

While previous results have shown the flexibility of SLIC SABRE by applying the method in the microtesla^50–52 and tesla^53–55 regimes, our results show there are clear advantages to operating at millitesla magnetic fields. SLIC SABRE is our chosen hyperpolarization method primarily because it generates reproducible ¹³C enhancement in situ^56–58. Further advantages of performing SLIC SABRE at mT magnetic fields, compared to the T and μT regimes, are discussed in more detail in Supplementary Section 6. More generally, the cost-effectiveness of SABRE hyperpolarization and its compatibility with 6.5 mT are our key motivations for pursuing applications of SABRE methods. 6.5 mT in particular is the ideal field to perform SABRE hyperpolarization of ¹H on molecules, such as pyridine and pyrazine, as this field strength is close to matching the level anti-crossing condition. This opens up opportunities for hyperpolarized imaging schemes at 6.5 mT that leverage hyperpolarization across multiple nuclei in an agile and flexible platform^27,45. Indeed, work from our laboratory and others’ has already demonstrated hyperpolarization of spins beyond ¹H and ¹³C is feasible and has utility for in vivo applications^29,59,60.

Further work will also leverage the advantages specific to our system. High homogeneity in B₀ and B₁ combine to make an excellent platform for exploring novel hyperpolarization schemes and transfer of polarization between nuclei. While the elegant simplicity of spin-lock pulses for generating hyperpolarization has driven the discoveries presented in this manuscript, the modest polarization reached in these initial results suggests there is significant potential for improvement. To quantify the current sensitivity of our system, we consider the average SNR of ~200 achieved in our SLIC-shot ¹³C MRI experiment performed with ~12 μmol ¹³C. Assuming SNR scales linearly with ¹³C concentration, at our current ¹³C polarization, our results imply we could visualize ¹³C above our system noise floor at concentrations down to 0.12 μmol. We note that the need for discrimination of metabolic products and faster readout to minimize T₁ relaxation and maximize temporal resolution will require tradeoffs in absolute sensitivity to ¹³C. We hope to utilize the flexibility of our research system’s hardware and pulse sequence programming to implement nonintuitive schemes for polarization transfer that do not rely on matching spin energy levels to anti-crossings⁶¹. The favorable characteristics of our system also position us to explore generating long-lived spin states on ¹³C tracers that can be read out via proton magnetization for longer effective metabolic imaging timescales and greater sensitivity^62,63.

Challenges and opportunities for ULF ¹³C MRI

Leaving aside the advantages of in situ hyperpolarization in the ULF regime, the imaging platform we have showcased here is entirely compatible with a more traditional in vivo imaging paradigm where a polarizer operates outside the scanner and hyperpolarized samples are shuttled inside for imaging. In time, further optimization of in situ SLIC SABRE may yield polarization levels comparable with those achieved by polarizers operating at other field strengths, but we envisage the speediest path to in vivo work at ULF will incorporate an external polarizer into an experimental method that uses ULF for imaging alone. Using an external polarizer will address the need for larger boluses of polarized ¹³C metabolites and incorporate automated depressurization and purification processes for rapid transfer. Recent advances have made progress on each of these steps. PHIP-SAH preclinical polarizers that can reliably deliver 2 mL doses of ~20% polarization at 70–180 mM concentration are available commercially²⁸. SABRE-based prototypes have reached similar performance levels, crucially overcoming hurdles associated with the removal of the iridium-based SABRE catalyst^32,33 and moving away from methanol solvent towards more readily extracted acetone mixtures³⁴. A key future challenge is scaling up parahydrogen-based technology towards human clinical applications where larger doses at higher polarization and concentration are routinely delivered alongside more stringent quality control requirements. For example, most sites performing clinical trials with hyperpolarized pyruvate deliver doses of 220–280 mM pyruvate in volumes >40 mL and >15% ¹³C polarization⁶⁴. Despite clinical translation gaps, parahydrogen-based techniques show remarkable promise and have succeeded in preclinical demonstrations. Future process engineering and device automation should enable scale-up of the technology.

In this subsection, we now turn to the challenges and opportunities specific to ULF ¹³C MRI. While a lack of chemical shift is advantageous for in situ SLIC SABRE hyperpolarization, it is less convenient when considering the challenges faced performing chemical shift imaging at ULF⁶⁵. For example, it will preclude the use of interleaved spectrally selective 3D bSSFP sequences^28,33. However, the spectra we have presented demonstrate that while challenging, CSI at 6.5 mT is by no means impossible. The ¹³C linewidths we have reported here are sufficiently narrow to encourage the pursuit of chemical shift imaging at 6.5 mT, despite reduced chemical shift in absolute frequency terms at low magnetic field. The 0.46 Hz linewidth we observe in deuterated pyruvate corresponds to 6.6 ppm at 69.5 kHz, which suggests that resolving the chemical shift of 12.2 ppm between [1-¹³C]pyruvate and [1-¹³C]lactate, while challenging, is achievable. More encouraging, resolving the 136.8 ppm separation between [2-¹³C]pyruvate and [2-¹³C]lactate is clearly practicable. It is necessary to acknowledge that the linewidths reported here are measured within a 5 mm NMR tube rather than across a several-cm volume of interest in vivo. However, our shims can be further optimized, and our spectra are broadened by the contribution from catalyst-bound pyruvate, which will not be present in purified samples. Operating at low field also offers the opportunity to narrow ¹³C spectra of protonated molecules by employing proton decoupling pulses in vivo, as the higher specific absorbance rates that forestall such methods at T fields are near negligible in the mT regime.

Both MRI sequence and hardware development will be necessary to fulfill the potential of ULF metabolic MRI. To improve temporal resolution, we will accelerate MRI acquisition and increase gradient performance. For the former, the implementation of methods honed at high field, such as compressed sensing and deep learning, offers a pathway for future work, so long as the specific noise characteristics at low field are taken into account⁶⁶. For the latter, increased gradient efficiency is readily achievable for future in vivo applications by upgrading our hardware, applying methods already demonstrated for efficient gradient and shim design⁶⁷. Gradient strength is particularly critical to improve resolution for a low gamma nucleus at low field, and a factor of 4 improvement upon the 1  mT/m gradients used for our proof of principle ¹³C MRI here is feasible, especially for a preclinical demonstration. Long term, however, the optimum choice of field strength for ULF metabolic imaging will be informed by the maximum gradient strength and field homogeneity achievable at an accessible cost. Other hardware development will include a more robust field lock, able to operate during MRI scanning, and dedicated ¹H/¹³C coils to remove the need for the impractical field cycling method that we utilized in this work to produce co-registered ¹H/¹³C images.

A challenge for all hyperpolarized MRI, regardless of field strength, is T₁ relaxation. While the highest 45 s T₁ value for pyruvate measured in our SABRE solutions at ULF is shorter than pyruvate T₁ times measured at high field, our T₁ is still sufficiently long to probe pyruvate-lactate conversion if replicated in vivo. As our values were measured under continuous pH₂ bubbling conditions, convection, susceptibility fluctuations and SABRE-related hyperpolarization effects may bias the extracted relaxation times. As such, our reported T₁ represents the effective relaxation with continuous bubbling rather than the intrinsic decay constant in the millitesla regime. Further, we expect purified pyruvate solutions to exhibit longer T₁ times, particularly if D₂O is utilized as the solvent³³ and rigorous care is taken to optimize the solution composition and exclude dissolved O₂: recent work has shown that pyruvate T₁ times approaching 4 minutes are achievable at millitesla fields⁶⁸. Future preclinical work will not only need to address sample composition but incorporate automation of purification, transfer and injection steps to minimize T₁ relaxation between polarization and imaging. There are also opportunities to extend effective T₁ at ULF with long-lived singlet states. This takes advantage of the strong coupling regime at ULF to enable long-lived singlet states to persist until read out through the application of a spin-lock or trains of pulses practicable in vivo at ULF due to lower SAR and high field homogeneity⁶². Beyond extending the effective time over which hyperpolarized metabolites maintain spin-order in vivo, schemes leveraging long-lived singlets offer readout methods that can distinguish pyruvate and lactate based on their different ¹³C J-couplings, and proton-only sensing that can further broaden the reach of hyperpolarized metabolic imaging by alleviating the need for specialized ¹³C hardware⁶³.

Considering the clinical translation of hyperpolarized ¹³C MRI at low field, we acknowledge that a clinical indication remains to be established. However, cost-effective parahydrogen-based hyperpolarization methods will accelerate research to establish clinically validated use cases for ¹³C MRI. The decisive next step will be clinical validation at high field through further clinical trials^64,69. Following high-field validation, a compatible low-field use case must leverage the advantages of portable, low-cost MRI to expand access to ¹³C MRI.

In conclusion, parahydrogen-driven hyperpolarization techniques are uniquely positioned to take the extensive body of work from the dissolution DNP community on metabolic imaging and amplify its impact by making the modality widely accessible in the clinic. Leveraging this technology alongside ULF MRI presents an excellent opportunity for each technology to complement the other: hyperpolarization can bypass the SNR challenges inherent to Boltzmann polarization at low field, and presents contrast mechanisms in a regime where methods developed for high-field MRI don’t work. ULF MRI can simultaneously increase the reach of metabolic imaging to clinical paradigms, including population screening and remote locations. While challenges associated with spin relaxation and resolving subtle chemical shifts at low field are present, there is no fundamental limitation barring this technology. SABRE offers a fast, flexible and accessible method for hyperpolarizing ¹³C metabolites under a range of conditions to accelerate the translation of this technology from physics advance, through preclinical tool, toward clinical application. In the ULF regime, where affordable and portable MRI scanners operate, there is significant potential to enable next-generation molecular imaging with point-of-care scanners.

Materials and methods

Sample preparation

Solutions for SABRE were prepared with 30 mM of either [1-¹³C] or [1-¹³C]-d₃ pyruvate, 6 mM [IrCl(COD)(IMes)] precatalyst and 20 mM DMSO in 450 μL of methanol-d₄. DMSO is added to modulate the exchange rate of the pyruvate as previously described^46,70. Precatalyst was synthesized following established methods^70–72. Chemicals were purchased from Sigma-Aldrich and stored refrigerated under argon. All liquids were degassed with argon to displace dissolved paramagnetic oxygen, which otherwise acts to relax pH₂ to H₂. SABRE solutions were sonicated to ensure all precatalyst was dissolved.

pH₂ generation and delivery

pH₂ was generated by flowing hydrogen gas over iron oxide hydroxide at 27 K at a rate of 2 slm and stored in aluminum cylinders at 480 psi before use within 24 h. The pH₂ generator system is described in Supplementary Section 5.

To run a SLIC SABRE experiment, a SABRE solution was placed in a 5 mm high-pressure NMR tube (Wilmad) at 6.5 mT and bubbled with pH₂ at a pressure of 95 psi and rate controlled by the combination of a mass flow controller (MFC, Alicat) and back pressure regulator. The pH₂ delivery circuit, including a detailed schematic, is shown in Supplementary Section 5. A period of tens of minutes’ pH₂ bubbling facilitates activation of the catalyst precursor by hydrogenation of the COD ligand. Twelve standard cubic centimeters per minute was found to be a reliable rate of pH₂ delivery for the majority of experiments, providing excellent enhancement whilst limiting sample evaporation.

Temperature control

To modulate the chemical exchange rate of pyruvate binding and releasing the SABRE catalyst, the 5 mm sample tube was cooled by being placed in a 10 mm water-filled NMR tube in thermal contact with the walls of a temperature-controlled 10 mm inner diameter NMR probe. The NMR probe was home-built using a 3D-printed hollow resin former, through which automotive coolant was pumped by a circulating chiller (Huber). The chiller was situated outside the RF-shielded room of the MRI scanner and connected to the temperature-controlled probe with 1/4” insulated tubing. The cooling circuit is described in Supplementary Section 5. Temperature was monitored with a fibreoptic thermometer (Osensa) placed in the 10 mm tube of water to account for the offset between the chiller setpoint and temperature at the sample inside the warm MRI scanner.

NMR hardware

All measurements were performed in a custom-built MRI scanner with a Tecmag Redstone console described previously³⁵. This scanner consists of a biplanar electromagnet, operating at 6.5 mT magnetic field, powered by a direct current (DC) supply (DANFYSIK 854 T) and biplanar magnetic gradient coils. Active shimming is achieved by applying DC offsets to the gradient coils via gradient amplifiers (AE Techron 7794MRL). The high B₀ homogeneity and field-locked stability of this scanner are particularly advantageous for spin-lock experiments^73–75. A B₀ field-frequency lock maintains proton resonance frequency within ±0.25 Hz, and the scanner is shimmed to achieve a linewidth of deionized water of better than 0.5 Hz. The work described here used a 2-layer solenoid coil, wound about the 10 mm inner diameter 3D-printed hollow former using 5/39/42-AWG20 Litz wire with fluorinated ethylene propylene insulation (New England Wire Technologies). The coil was tuned to 69.5 kHz with an external resonator board in parallel-tune, series-match configuration. The tuned and matched coil had a q-factor of 22, corresponding to a bandwidth of 3 kHz, determined by an S21 measurement about 69.5 kHz with a vector network analyzer and an untuned pick-up coil weakly coupled to the NMR probe. To achieve the low power and stability needed for spin-lock pulses of a few μT B₁ amplitude, the power amplifier was bypassed and RF pulses used directly from the console synthesizer. To maintain the fidelity of the low-power SLIC pulses, measurements employed an active CMOS-based transmit/receive switch (Mini-Circuits) with low power loss and low phase distortion.

Following catalyst activation and sample cooling, a typical SLIC SABRE experiment, shown in Fig. 1C, involved the application of a spin-lock pulse followed by immediate FID acquisition, with pH₂ bubbling constant throughout. Reliable and repeatable SLIC parameters on resonance at 69.5 kHz were 10 s spin-lock pulse duration, 10 Hz spin-lock amplitude and 12 sccm pH₂ flow rate. Experiments exploring individual parameters—detailed in Supplementary Sections 2 and 5—used these values as a starting point for controlling variables not under investigation. The free induction decay acquired following the spin-lock pulse was converted to a frequency spectrum via fast Fourier transform and phase corrected to bring the real part of the signal purely into the absorption mode. The real part of the spectrum was then integrated between ±50 Hz. The spectra shown in Figs. 1 and 4 were zero-filled by a factor of 2.

MRI

Imaging was carried out with the same variable temperature NMR probe as the spectroscopy measurements. For ¹³C MRI sequences, the field lock was applied during SLIC SABRE but blanked while imaging gradients were applied.

The SLIC-shot ¹³C MRI sequence is a 3D bSSFP sequence adapted with the addition of spin-lock and 90^∘ pulses prefacing a bSSFP gradient block with a variable tip angle. The SLIC-shot MRI sequence used an echo time (T_E) of 54 ms, a repetition time (T_R) of 108 ms and a 512 × 64 × 5 matrix undersampled by 50%. The variable tip schedule was calculated following the iterative method described in ref.⁴⁸ with an initial tip angle of 1^∘. Total duration for acquiring the 160 lines of k-space pulse was 17.3 s.

¹H MRI was performed by ramping the electromagnet down to 1.6 mT to bring protons on resonance at 69.5  kHz to match the NMR probe tuning. The ¹H MRI used a bSSFP sequence with the same T_E, T_R and undersampled matrix as the ¹³C SLIC-shot MRI, a tip angle of 45^∘, and 128 averages for a total imaging time of 37 min. The field lock was not used at 1.6 mT as it was far outside its frequency bandwidth. Shim values were not re-optimized for 1.6 mT.

The multi-SLIC-shot ¹³C MRI sequence was adapted from a gradient echo sequence with the substitution of a spin-lock pulse for the standard hard pulse. The multi-SLIC-shot scan used T_E = 87 ms, T_R = 121 ms and a 10 s spin-lock for each line of 50% undersampled 576 × 64 × 5 matrix. This equates to a total sequence duration of 27  min.

Image reconstruction applied DC offset and ADC turn-on spike corrections, zero-filled k-space by a factor of 2 in the readout and first phase encode directions and performed a 3D inverse Fourier transform. Normalization in units of the noise floor was performed by dividing the signal in image space by the standard deviation of a region of interest comprising the bottom left 1/8 of the FOV.

Author contributions

Conceptualization: T.B., T.T., and M.S.R. Methodology: T.B., S.J.M., M.P., E.C., P.T., S.S., N.K., E.C., T.T., and M.S.R. Investigation: T.B., H.B., N.K., and D.K. Visualization: T.B., D.E.J.W. Funding acquisition: T.T., E.C., and M.S.R. Project administration: D.E.J.W., T.T., and M.S.R. Supervision: T.B., T.T., D.E.J.W., and M.S.R. Writing-original draft: T.B. Writing-review and editing: all.

Peer review

Peer review information

Communications Chemistry thanks Eleonora Cavallari and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information. Raw data are available from the corresponding author upon reasonable request.

Competing interests

M.S.R. is a founder and equity holder of Hyperfine Inc. M.S.R. is an equity holder of DeepSpin GmbH. M.S.R. also serves on the scientific advisory boards of ABQMR, Synex Medical, Nanalysis, and O2M Technologies. E.Y.C. is a co-founder and equity holder of XeUS Technologies LTD. All other authors declare no competing interests.

Supplementary information

The online version contains Supplementary material available at 10.1038/s42004-026-01971-2.