Singlet‐Contrast Magnetic Resonance Imaging: Unlocking Hyperpolarization with Metabolism

Abstract Hyperpolarization‐enhanced magnetic resonance imaging can be used to study biomolecular processes in the body, but typically requires nuclei such as 13C, 15N, or 129Xe due to their long spin‐polarization lifetimes and the absence of a proton‐background signal from water and fat in the images. Here we present a novel type of 1H imaging, in which hyperpolarized spin order is locked in a nonmagnetic long‐lived correlated (singlet) state, and is only liberated for imaging by a specific biochemical reaction. In this work we produce hyperpolarized fumarate via chemical reaction of a precursor molecule with para‐enriched hydrogen gas, and the proton singlet order in fumarate is released as antiphase NMR signals by enzymatic conversion to malate in D2O. Using this model system we show two pulse sequences to rephase the NMR signals for imaging and suppress the background signals from water. The hyperpolarization‐enhanced 1H‐imaging modality presented here can allow for hyperpolarized imaging without the need for low‐abundance, low‐sensitivity heteronuclei.

Foranumber of reasons, 1 Hwould be the ideal nucleus for hyperpolarization-enhanced MRI: [34][35][36] 1. Sensitivity-1 H-detection is more sensitive than, for example, 13 Cbyafactor of % 16, and 15 Nbyafactor of % 100, if they are polarized to the same degree and are at the same abundance; 2. Equipment availability-1 Hprobes are readily available in commercial MRI scanners; 3. Isotopic abundance-Unlike 13 Co r 15 NM RI, there is no need for expensive isotopic enrichment of the samples; 4. Spatial resolution-Greater response to magnetic field gradients means higher spatial resolution can be obtained than for lower-g nuclei.
However,t here are some notable drawbacks to hyperpolarized 1 HM RI: 1. Rapid relaxation-Relaxation of hyperpolarized proton signals typically occurs with acharacteristic temporal scale on the order of seconds,w hich is not enough time to perform the complete imaging experiment from sample hyperpolarization to metabolism and detection; 2. Proton background-There is al arge proton background signal in the body because the natural abundance is % 100 %, and water and fat molecules are prevalent in the body; 3. Chemical shift dispersion-The relatively small chemical shift dispersion can make distinguishing different chemical species ac hallenge,e specially in vivo where broad NMR lines are common.
Owing to these challenges, 13 Ci sc urrently the preferred nucleus for hyperpolarization-enhanced imaging.T he relaxation times are on the order of tens of seconds; [1] thel ow (1.1 %) natural abundance means there are no significant background signals,and the chemical shift range is an order of magnitude higher than that for protons.O ne option that has been demonstrated is to store the hyperpolarized magnetization on aheteronucleus,and then transfer it onto anearby 1 Hs pin for signal read-out. [37,38] However,t his method still requires isotopic enrichment of the samples and rf probes with aheteronuclear channel.
Singlet-state NMR (see ref. [39][40][41][42][43][44][45][46][47]) offers an alternative possibility to overcome the drawbacks associated with hyperpolarized 1 Himaging. [48,49] When parahydrogen is added to an unsaturated precursor molecule,the protons remain in anonmagnetic singlet state,aslong as they remain chemically and magnetically equivalent. This state is neither directly observable in MRI, nor can it be manipulated by radiofrequency(rf) pulses.A dditionally,t he proton singlet state is immune to certain relaxation mechanisms,a nd can have am uch longer lifetime compared to proton magnetization. [39,40] Thus,t he hyperpolarization can be stored in the singlet state until the molecule undergoes ac hemical reaction that renders the protons chemically or magnetically inequivalent. This breaks the proton singlet state,and observable hyperpolarized NMR signals are released.
These favourable properties of the singlet state open up new possibilities to perform hyperpolarized 1 HM RI by injecting abiomolecule supporting asinglet state with along nuclear spin lifetime,which is converted in vivo to an NMRvisible substrate.Such an experiment would have the following advantages: 1. Until metabolism of the molecule,t he protons relax relatively slowly; 2. Theb ackground 1 HNMR signals of water and fat in the body can be suppressed with rf pulse techniques,while the nonmagnetic singlet state remains unaffected; [50][51][52][53] 3. This experiment relies on the appearance of an NMR signal rather than apeak shift, and so is insensitive to the limited chemical shift dispersion.
In this work we demonstrate "singlet-contrast magnetic resonance imaging" using fumarate,arepresentative biomolecule.U nlike previous work on this chemical system which used dissolution dynamic nuclear polarization (D-DNP) to generate singlet order, [4] we produce fumarate by chemically reacting para-enriched hydrogen gas with an acetylene precursor in D 2 O, using ar uthenium trans-hydrogenation catalyst. [15,32] Fumarate is am etabolite in the citric acid (Krebs) cycle,a nd is converted into malate by addition of aw ater molecule;areaction catalysed by the enzyme fumarase,a nd of great importance for hyperpolarizationenhanced MRI. [9][10][11][54][55][56][57] Thee nzymatic conversion to malate renders the fumarate protons chemically inequivalent. Since the fumarate protons originate from parahydrogen, the malate becomes hyperpolarized, and the resulting enhanced NMR signals are antiphase.These PA SADENA(Parahydrogen And Synthesis Allow Dramatically Enhanced Nuclear Alignment) signals can be observed by applying a4 5 8 8 rf pulse. [17] Thef ormation and metabolism of fumarate are shown in Figure 1, alongside ac omparison between PA SA-DENAa nd thermal equilibrium 1 HNMR spectra of malate. Figure 1. Top: the hydrogenation reaction used to produce fumarate [15,32] and the subsequent enzymatic conversion into [3,O-D 2 ]malate (since D 2 Oisthe solvent). Bottom: 1 HNMR spectra acquired at afield strength of 11.7 T, showing the differencebetween aPASADENA spectrum and athermal equilibriumspectrum of [3,O-D 2 ]malate. The PASADENAs pectrum has been vertically scaled down for aclearer comparison.The water peaks (marked by asterisks) are from the residual protons in the solvent, and the shift of water between spectra is atemperature effect.
Here we perform the enzymatic reaction in D 2 Ot o produce [3,O-D 2 ]malate (see Figure 1), which will henceforth be referred to as malate-D 2 for simplicity.T he deuterons are weakly coupled to the protons and are unaffected by the proton rf pulses,sothey can be ignored when we consider the spin dynamics.T his means we can treat the two protons originating from parahydrogen as an isolated two-spin system, which is convenient for this proof-of-principle demonstration. Using D 2 Oa st he solvent has the additional and important benefit of extending the fumarate proton T s ,w hich was measured to be 8sin H 2 O(see Supporting Information), but is over 40 sinD 2 O. [32] TheP ASADENAs ignals from malate-D 2 (see Figure 1) are antiphase which precludes most 1 Hi maging techniques, because the application of imaging gradients to spatially encode the magnetization would cause signal cancellation. A pulse sequence is required to convert the two-spin order into observable,in-phase magnetization, and additionally suppress signals from background magnetization arising,f or instance, from water or fat molecules present in significantly higher concentration. Forthis purpose,weutilize two versions of the out-of-phase echo (OPE) pulse sequence. [58,59] In one version, OPE-45, ah ard 458 8 pulse nonselectively excites the signals, and background magnetization is removed at the end with pulsed field gradients.I nt he other version, OPE-s90, a selective 908 8 pulse is applied to the malate-D 2 proton resonance at 2.5 ppm, which does not excite background magnetization. Thetheory of how these two pulse sequences work is given in the Supporting Information. Figure 2 illustrates the core principle of the experiment and how the pulse sequences operate.

Results and Discussion
Pulse Sequence Optimization To optimise the parameters of the two OPE sequences, experiments were performed on as ample of malate-D 2 in D 2 Oa tt hermal equilibrium (i.e.n ot generated from parahydrogen). Ap reparation rf pulse sequence known as the Sarkar sequence [60] was applied to convert the thermal equilibrium I 1z + I 2z spin order into I 1z I 2z spin order between the protons,t om imic the initial density operator in an experiment using parahydrogen. Immediately after, an OPE sequence was applied. This experiment was repeated many times,with the delay in the OPE varied to find the optimum value.T he pulse sequences and results for both the OPE-45 and OPE-s90 are shown in Figure 3.

PHIP Shuttling Experiments
To demonstrate the pulse sequences in hyperpolarized NMR experiments we used the following procedure:(1) bubble para-enriched hydrogen gas into the precursor solution to produce hyperpolarized fumarate;( 2) pneumatically shuttle the sample into an NMR tube containing fumarase in D 2 O held in an 11.7 Tmagnet;(3) apply either OPE-45, OPE-s90, or a458 8 pulse every 4sand detect the resulting NMR signal. Ther esults are shown in Figure 4. Each pulse sequence destroys most or all of the hyperpolarized spin order in malate-D 2 with each application, but the singlet order for fumarate molecules is unaffected. We observe persistent NMR signals for approximately 1m inute,w hich is possible because new molecules of malate-D 2 form between the application of each pulse sequence.T he water signal is significantly attenuated in the OPE-45 spectra, and virtually absent from the OPE-s90 spectra.
Them alate-D 2 1 Hp olarization level was estimated to be 20 %i nt he OPE-s90 experiment. This was determined by Figure 2. Ag raphicald epiction of hyperpolarized fumarate being converted into malate-D 2 ,f ollowed by the application of OPE-45 or OPE-s90 to convert the antiphase proton spin order into in-phasem agnetization, and suppress backgrounds ignals from water.The white arrows represent different types of spin order,w ith up-down arrows in the first two frames indicatingthe absence of net magnetization on the proton spins. Red and grey shading is used to represent observable and unobservable spin groups,r espectively,from different species throughout the experiment. Simulations of the resulting hyperpolarized spectra are shown on the right, using the NMR parameters given in Figure 1. Note that in principle OPE-45 and OPE-s90 produce the same integral of the proton signal.
summing the integrals of the 2.35 ppm peak during the hyperpolarized 1 Hs ignal decay and comparison with the thermal equilibrium signal integral after the hyperpolarization had fully decayed. We were unable to determine the 1 H polarization level in the OPE-45 experiment because the proton signal was not visible in the thermal equilibrium spectrum, but we were able to set al ower bound of 10 % assuming as ignal-to-noise ratio (SNR) of 1i nt he thermal equilibrium spectrum.

Imaging Results
To demonstrate singlet-contrast imaging we acquired images of ah yperpolarized reaction mixture in a1 0mm NMR tube surrounded by H 2 O. This is shown in Figure 5a. Thei maging was performed in a7Tm agnetic field. The 10 mm NMR tube initially contained fumarase in adeuterated phosphate buffer solution (PBS) at pH 7( optimal for this reaction [61] ). Thep recursor solution was hydrogenated with parahydrogen and then injected into the 10 mm NMR tube for imaging.F urther experimental details are given in the Materials and Methods section (see Supporting Information).
To image the hyperpolarized malate that formed from the metabolism, we applied OPE-s90 to selectively excite the malate protons,i mmediately followed by a9 0 8 8 rf pulse to return the magnetization back to the z-axis,and then applied af ast low-angle shot (FLASH) sequence [62] with centric reordering to acquire a6 4 64 pixel image.T he complete imaging sequence is shown in Figure 5b.T he sequence was repeated every 12 st oa cquire at rain of images as the metabolism progressed. Forc omparison, we also show nonselective 1 Hi mages acquired using the same FLASH sequence after the hyperpolarized signals had fully relaxed. The metabolites in the 10 mm tube are at such alow concentration that they are not visible in the thermal equilibrium images, but al arge water background signal is present. TheP HIP experiment was performed twice to acquire axial and sagittal images,a nd the first image from each acquisition train is shown in Figure 5c,d alongside the nonselective thermal equilibrium images.The train of images acquired in the axialorientation experiment are shown in Figure 5e.T he receiver gain was set to 101 for hyperpolarized experiments,and 1f or thermal equilibrium experiments,a nd this translates to afactor of 2difference in SNR.
Thes inglet-contrast MRI experiments shown in Figure 5 demonstrate that despite the low concentration of the metabolites present (approximately 10 3 times lower than the concentration of H 2 Omolecules), this technique can be used to suppress the water background signal and image metabolic flux. AF LASH sequence with centric reordering was used here as ac onvenient way to utilize the hyperpolarized magnetization generated by the OPE sequences,a so nly asmall fraction of magnetization is used for the acquisition of each line in k-space.H owever,o ther imaging techniques [63] such as rapid acquisition with relaxation enhancement (RARE) [64] or echo planar imaging (EPI) [65] could be utilized for this experiment. In the current work, no slice selection was applied. This can be readily achieved in the case of OPE-45 where slice selection can be done in as tandard way in the imaging part of the sequence,but is more complex for the case of OPE-s90 which uses frequency-selective excitation pulses, as the slice-selection gradient necessarily produces al arge frequency distribution, making frequencys election challenging. It would however be possible to use as lice selective 180 refocusing pulse in the OPE sequence.
In our experiments,O PE-45 and OPE-s90 show similar transfer efficiency of I 1z I 2z spin order into magnetization, and both sequences have the same theoretical efficiency.OPE-45 relies on pulsed field gradients and rf pulse phase selection to suppress background signals,w hereas OPE-s90 additionally suppresses background signals due to the frequency selectiv- Figure 3. The rf pulse sequences used in this work. Rectangular boxes represent "hard" (non-selective) pulses, rounded boxes represent "soft" (selective) pulses, and grey trapezoids represent pulsed field gradients. The malate-D 2 signal integrals are plotted on the right as afunction of the t delay.Asimulation of each experimenti sshown by the dashed blue line. The simulation and data are normalizedt o1,which corresponds to the absolute observable signal if a45 8 8 rf pulse were applied after the preparation sequence. The maximum amplitudeo fthe data is less than 1b ecause of losses due to relaxation. Error bars have been omitted as they are contained within the plot markers.
ity of its excitation pulses.Onthe other hand, the implementation of the frequency-selective pulse introduces some additional experimental challenges (e.g. requiring frequency drift of less than tens of hertz between experiments). It should be noted that alternative pulse sequences such as those in the Only ParaHydrogen Spectroscopy (OPSY) family [50][51][52][53] can be utilized to generate in-phase magnetization from antiphase spin order,b ut these methods are not explored in this work.
Ap revious hyperpolarized 1 Hi maging experiment [49] utilized in-phase magnetization generated from the PA SA-DENAs ignal under J-coupling evolution during the echo time of the imaging scheme.I nS inglet-Contrast MRI, the hyperpolarization is stored in along-lived singlet state before imaging,and the method additionally overcomes some of the limitations of that approach;i ti sm ore general, as it can be combined with any imaging scheme without synchronizing the delays to the couplings of the hyperpolarized molecule.T his allows one to easily acquire images of the hyperpolarized signal at different points in time,w hich makes real-time tracing of metabolic processes possible.A dditionally,i tdoes not require that the background and hyperpolarized 1 H signals have different relaxation properties for background suppression. [34] Experiments in this work were performed in D 2 Ot o extend the lifetime of hyperpolarized spin order,a nd to produce as pin system with just two proton nuclei in the product molecule ([3,O-D 2 ]malate). Application in biological systems means working in the presence of H 2 O, which reduces the fumarate singlet-state lifetime.T he proton T s was measured in ap rotonated phosphate buffer solution to be 8s(see Supporting Information), which presents ac hallenge for applications of this particular molecular system;t he hyperpolarized fumarate should be prepared in D 2 O, and only mixed with H 2 Oa tt he point of delivery to minimize signal losses.A sa na lternative,i th as been shown that other molecules can support proton singlet states that are relatively long-lived in protonated solvents. [66] We note that the 8s T s was measured on [1-13 C]fumarate,a nd fluctuating dipolar coupling to the nearby 13 Cs pin is an additional source of relaxation that will not be present in the unlabelled molecules used for singlet-contrast imaging.T he conversion to malate occurring in aprotonated solvent also leads to the formation of fully protonated malate,athree-spin system. In order to  The pulse sequence used to acquireh yperpolarized 1 Himages. (c,d) Ac omparison between the hyperpolarized and thermal equilibrium 1 Hi mages, with the hyperpolarized images acquired following the procedure described in the text. (e) At ime series of hyperpolarized 1 Himages. The receiver gain was set to 101 for hyperpolarized image acquisition and 1f or thermal equilibriumimage acquisition,which gives af actor of 2difference in signal-to-noise ratio (as discussedi nthe Supporting Information).
convert the initial parahydrogen-derived three-spin order into observable in-phase magnetization, the pulse sequence evolution (t)d elays need to be modified, and an overall lower transfer efficiency can be expected. We discuss the three-spin case further in the Supporting Information and provide the optimal theoretical t delays.
As imilar experiment to what has been shown here was demonstrated with DNP-polarized fumarate,t os how that long-lived spin states can be populated via D-DNP. [4] In this work we have used PHIP to hyperpolarize the proton singlet state,a nd show that this type of experiment can be used for imaging.Acomparison of PHIP with D-DNP as polarization sources is relatively straightforward for experiments in which the 13 Cspins in fumarate are hyperpolarized;the polarization levels can be compared by measuring the 13 Cmagnetization at the point of delivery.Asimilar comparison is more subtle for this proton-enhanced experiment. Thef umarate protons support four states:as inglet state and three triplet states. Thed istribution of population amongst these four states determines the maximum malate signal intensity at the point of detection. In the PHIP experiment the aim is to fully populate the singlet state by starting from para-enriched hydrogen, which would lead to ar elative malate signal intensity of 1. In the D-DNP experiment the polarization process depletes the singlet state,w hich would lead to ar elative malate signal intensity of À1/3, as discussed in the Supporting Information, as well as in ref. [4].B eyond the highest achievable signal enhancement, PHIP stands out as being significantly less expensive than D-DNP,and is able to produce boluses of hyperpolarized material at amuch higher turnover rate.

Conclusion
In conclusion, we have demonstrated an ovel type of hyperpolarized 1 Hi maging experiment, in which nuclear hyperpolarization is locked in al ong-lived singlet state until liberation by ac hemical/biological process.W eu sed paraenriched hydrogen gas to hyperpolarize the proton singlet state in the biomolecule fumarate,a nd the signals were released by enzymatic conversion to malate.I no ur experiments,u pt o2 0% proton spin polarization was observed on malate.T he released signals are antiphase when observed directly after applying an rf pulse which complicates imaging, and so we employed two pulse sequences,OPE-45 and OPE-s90, for converting the antiphase spin order into in-phase magnetization. By adding pulsed field gradients for coherence filtering,weshow that background signals from the protons in the water solvent can be effectively suppressed. We have demonstrated the method by acquiring images of hyperpolarized fumarate-to-malate metabolism over the course of am inute using OPE-s90, with effective suppression of the water background signals.T he relatively short proton singlet lifetime in water (8 s) will likely limit this specific molecular system to studying samples with high metabolic flux. We expect this imaging method to be extended to alternative PHIP systems, [67] or other nuclear spin species,f or example,