Author Correction: Biochemical phosphates observed using hyperpolarized 31P in physiological aqueous solutions

The original version of the Supplementary Information associated with this Article contained an error in Supplementary Figure 2 and Supplementary Figure 5 in which the 31P NMR spectral lines were missing. The HTML has been updated to include a corrected version of the Supplementary Information.

R ecent advances in the dissolution-dynamic nuclear polarization (d-DNP) and para-hydrogen-induced polarization (PHIP)-based hyperpolarization techniques have revolutionized the field of liquid-state nuclear magnetic resonance (NMR) spectroscopy. By providing signal enhancements of up to four orders of magnitude, such methods enable the detection of small amounts of materials as well as detections made on very short time scales. The d-DNP technology, pioneered by Ardenkjaer-Larsen et al. 1 , has been predominantly focused on metabolic work using hyperpolarized 13 C sites 2 . A few studies looked at metabolites labeled with 15 N 3, 4 . In the field of contrast material, hyperpolarized yttrium showed beneficial relaxation properties 5,6 and water protons were also hyperpolarized [7][8][9][10] . d-DNP studies with 107,109 Ag 11 , 6 Li 12, 13 , 29 Si 14 , and 19 F 15 have been reported as well. However, to date, to the best of our knowledge, the phosphorus nucleus ( 31 P), which is 100% naturally abundant and plays an important role in chemistry and life, had not been seen in a d-DNP-driven hyperpolarized state in aqueous solutions. One study reported the ability to hyperpolarize the 31 P nuclei of adenosine triphosphate (ATP) in the solid state using DNP 16 . However, this study was not followed-up by detection of this hyperpolarized state in solution. Since the T 1 of the 31 P nuclei of ATP in aqueous solutions is very short, on the order of 1 s 17 , it would be highly challenging to visualize ATP in a hyperpolarized state in solution. In our hands, this was indeed impossible (data not shown).
PHIP is a method of creating highly polarized nuclear spins in solution utilizing the spin order of para-hydrogen 18 . The polarization transfer occurs following a hydrogenation reaction of an unsaturated molecular precursor with the para-hydrogen molecule. A variant of the PHIP approach uses a reversible interaction with the para-hydrogen molecule and is termed signal amplification by reversible exchange (SABRE) 19 . Both techniques offer orders of magnitude increase in the NMR signal for multinuclei. To date, we have seen a couple of reports of hyperpolarized 31 P in solution-done by the SABRE method 20,21 . However, the 31 P hyperpolarization achieved in these studies was of molecules that are not water soluble and do not present any biomedical relevance.
Phosphate-containing compounds are of key importance in biology and medicine and the 100% natural abundance of their NMR-active 31 P nucleus make them affordable, attractive targets. Here, we show that observation of hyperpolarized 31 P in important phosphate-containing molecules is indeed possible with the d-DNP methodology under physiologically relevant conditions. We present a potential application of this capability for fast pH sensing. We note that there are numerous important phosphate-containing metabolites in biology and medicine. Further research into their metabolism and bodily distribution is likely to benefit from the approach described herein.

Results
Inorganic phosphate (Pi) is observed in a hyperpolarized state in aqueous media. To explore the hyperpolarized state of Pi in solution, the following formulation was prepared: 120.1 µmol KH 2 PO 4 , 15.1 µmol OXO63 radical, 5.4 mmol glycerol, and 33.1 mmol H 2 O. A vitrification assay 22 showed that this formulation indeed formed a glass upon rapid freezing to cryogenic temperature (liquid nitrogen). About 100-120 mg of this formulation were placed in a sample cup and polarized for 1.5-3 h with a microwave frequency of 94.097 GHz, a power of 100 mW, and a temperature of 1.40-1.46 K at a 3.35 T magnetic field using a d-DNP setup. Then the formulation was quickly dissolved in 4 ml of aqueous media at a temperature of 170°C and a pressure of 10 bar. The dissolved hyperpolarized solution was directly injected to a 10-mm NMR tube in a 5.8 T NMR spectrometer and 31 P spectra were continuously recorded. The hyperpolarized signals were recorded starting about 16 s from the start of the dissolution process.
When dissolved in D 2 O, the Pi signal was clearly visible for about 50 s across 50 scans, each acquired with a 10°nutation angle. The T 1 of this hyperpolarized site was about 15 s ( Fig. 1) and the pH of the hyperpolarized media was 5.1. To explore the effect of a protonated solvent and the addition of osmotic pressure at physiological levels, the same experiment was repeated  Fig. 1 Summary of T 1 relaxation time constants for hyperpolarized phosphates in aqueous media. The average value of n measurements is given and the error bars represent the standard deviation. The hyperpolarized decay curves were recorded at room temperature. The concertation of the hyperpolarized phosphate, the pH, and the number of measurements for each condition was as follows: KH 2 PO 4 in D 2 O: 0.94 ± 0.14 mM, pH 5.1, n = 2; KH 2 PO 4 in saline and citrate-Tris buffer: 0.43 ± 0.07 mM, pH 4.5, n = 2, includes two experiments with N 2 bubbling and two experiments without N 2 bubbling. K 2 HPO 4 in D 2 O: 1.80 ± 0.90 mM, pH 8.1, n = 5 K 2 HPO 4 in pure H 2 O: 2.95 ± 0.03 mM (n = 3), 34.93 mM (n = 1), pH 8.1 (altogether four experiments; one experiment was performed with a 2 M concentrated formulation of Pi, resulting in a high concentration in the NMR tube). K 2 HPO 4 in physiological saline: 1.14 ± 0.02 mM, pH 8.1, n = 2 PCr in D 2 O: 2.17 ± 0.71 mM, pH 6.5, n = 2; PCr in H 2 O: 2.78 ± 0.00 mM, pH 7.4, n = 2; PCr in D 2 O with phosphate buffer: 1.20 ± 0.17 mM, pH 7.4, n = 2. Pi inorganic phosphate, PCr phosphocreatine with water instead of D 2 O. The dissolution medium consisted of 90% physiological saline and 10% D 2 O with 100 mg/l EDTA, which were combined in the NMR tube with 5 ml of citrate-Tris buffer (altogether 300 mOsm and a pH of 4.5). Both the dissolution solvent and the citrate-Tris buffer were bubbled with N 2 for 1 h prior to use. The hyperpolarized Pi signal was again clearly visible but the T 1 appeared to be shortened to about 6 s ( Fig. 1). This result could be expected as T 1 shortening by protonated solvents and by the addition of osmotic pressure to the solution was previously seen for other hyperpolarized compounds 23,24 . To check for possible effects of oxygen depletion, the experiment was repeated twice more in fresh solutions without N 2 bubbling. The resulting T 1 of the hyperpolarized Pi was similar at 6 s. Therefore, the results with and without N 2 bubbling were combined and are presented in Fig. 1.
A basic environment prolonged the T 1 of hyperpolarized Pi. We further explored the effect of pH of the solid-state formulation (and the resulting hyperpolarized media) on the production and maintenance of the hyperpolarized state. To this end we used K 2 HPO 4 to prepare the formulation for polarization instead of KH 2 PO 4 , keeping all other conditions the same. When dissolved in D 2 O, this formulation resulted in a pH of 8.1. This change led to an improvement in the ability to observe the hyperpolarized state of Pi, as the T 1 in solution increased to about 29 s (Fig. 1). Figure 2 demonstrates a typical experiment done at these conditions. The hyperpolarized signal of Pi in D 2 O is clearly visible for more than 2 min across more than 120 acquisitions, each acquired with a 10°nutation angle. The thermal equilibrium spectrum of this sample is shown in Fig. 2b. The enhancement factor for the hyperpolarized Pi signal at these conditions was calculated to be 11,364 ± 438 (at 5.8 T, n = 2, 1.5 h of polarization), equivalent to 9.3 ± 0.3% polarization. In addition, we observed two other hyperpolarized signals downfield from Pi, likely corresponding to contaminations of the K 2 HPO 4 batch.
The effect of water protons on the hyperpolarized Pi decay.
To explore the effects of solvent (water) protonation in the basic condition, the experiment was repeated using pure water for dissolution. The T 1 of the hyperpolarized Pi decreased twofold to about 15 s (Fig. 1). This effect can be explained by inter-molecular dipolar interaction between the water protons and the hyperpolarized 31 P nucleus. The effect of dissolution osmolarity was also tested (using physiological saline as the dissolution medium, ca. 300 mOsm), and the T 1 was found to be shortened slightly to about 13 s (Fig. 1). Despite this T 1 shortening effect of using water instead of D 2 O and using iso-osmotic solutions, the T 1 of the basic formulation was still comparable to the T 1 of the acidic formulation in D 2 O. We conclude that the basic formulation is preferential for preparing Pi solutions for hyperpolarization.
Media components affect the Pi T 1 relaxation rate too. Thus far we investigated the T 1 of Pi under varied solution conditions in a hyperpolarized state. These conditions included protonation of the solvent, mixture pH, and osmotic pressure. However, the T 1 of Pi could be sensitive to other medium components as well.
To explore this aspect further we performed a preliminary investigation comparing two different types of media (water and citrate-Tris media). This investigation was carried out at thermal equilibrium and is summarized in Supplementary Note 3 and Supplementary Figs. 3 and 4. Briefly, we have found a surprising T 1 prolongation of the citrate-Tris media on the T 1 of Pi at the acidic to neutral pH values (1 ≤ pH and lower than 7). The T 1 values determined at thermal equilibrium (in both types of media) are in agreement with the T 1 values determined in the hyperpolarized state (which are summarized in Fig. 1).
Hyperpolarized Pi reports on instantaneous pH changes. The chemical shift of Pi is known to be sensitive to pH at the most physiologically relevant pH ranges 25 (Supplementary Note 2 and Supplementary Fig. 1). To illuminate an aspect of biomedical significance, we decided to demonstrate that hyperpolarized Pi can report on instantaneous pH changes. To this end we carried out a couple of experiments in which the hyperpolarized Pi signal underwent through varying and controlled pH regimes within one hyperpolarization decay time and was recorded continuously. The first experiment is described below and the second is shown in Supplementary Note 4 and Supplementary Fig. 5. In the experiment depicted in Fig. 3, a basic formulation of Pi was hyperpolarized as described in the "Methods". The hyperpolarized sample was dissolved in 4 ml of medical grade saline solution with 10% D 2 O and immediately transferred from the d-DNP device to the NMR spectrometer. The pH of this solution composition was determined independently prior to the hyperpolarized experiment to be 8.1. About 9 s after the appearance of the hyperpolarized signal, the pH in the sample tube was quickly acidified by means of a quick injection of 0.5 ml of citrate-Tris buffer of pH 4.5. In response, the continuously recorded 31 P spectra show two spectra with broad lines attributed to air bubbles introduced during mixing and then the following spectra show the hyperpolarized 31 P signal at a new chemical shift representing the new pH of 4.8, determined in the same sample immediately after the end of the hyperpolarized decay. This result demonstrates the potential for instantaneous pH sensing of aqueous solutions using hyperpolarized Pi. Since Pi is the ultimate pH sensor for biology, this result brings hope that hyperpolarized Pi can be used for clinical imaging of pH.

P hyperpolarization of phosphocreatine (PCr) is feasible.
PCr is a phosphate-containing molecule of key biochemical importance. Its main biochemical activity relates to the creatine kinase (CK) reaction, which serves to regenerate ATP in cells that rapidly consume ATP such as brain and muscle cells. To test the ability to observe the PCr phosphate in a hyperpolarized state, the following formulation was prepared: 25.4 μmol PCr, 1.8 μmol OXO63, 0.9 mmol glycerol, 3.6 mmol D 2 O. About 40-80 mg of this formulation were polarized in the same conditions described above. Indeed, we were able to detect the hyperpolarized state of the 31 P nucleus in the PCr molecule in aqueous media as well. hyperpolarized site was about 11 s (Fig. 1). The CK protein was added to the NMR tube to test the ability to observe not only the PCr phosphate in a hyperpolarized state but also in the presence of the enzyme that may bind it and therefore may affect the decay of the hyperpolarized state. The effects of solvent protonation and osmolarity were investigated for PCr as well and the results are summarized in Fig. 1. The T 1 of PCr did not appear to be significantly affected either by the water protonation or the osmolarity of the media. An important property of the PCr phosphate is that its 31 P chemical shift is much less sensitive to pH changes in the physiological range, because its pKa is below the physiological range, at pH 4.3 25 .

Discussion
Magnetic resonance (MR) imaging provides the most exquisite anatomical soft tissue details and therefore contrast in medical imaging that relies on MR-and can be fused to conventional magnetic resonance imaging (MRI) images-is preferential. Hyperpolarized compounds have the potential to serve as molecular imaging probes on clinical imaging and to provide unique contrast mechanisms that will report on tissue functionality without ionizing radiation. Here, we capitalized on the d-DNP technique to create an unexplored type of molecular imaging probes for MR, i.e., hyperpolarized phosphates. We demonstrated the ability to hyperpolarize the 31 P nucleus in Pi and PCr, which warrants research into many more endogenous phosphates.
We note that a key factor in the success of hyperpolarization of phosphates for use in solutions is a relatively long T 1 in solution. This requirement applies to any spin hyperpolarized MR molecular imaging probe. Indeed, Pi and PCr have a relatively long T 1 in aqueous solutions at 5.8 T compared to other phosphates; although in the d-DNP field these are considered among the shortest T 1 that have been amenable for investigation. A detailed investigation of 31 P T 1 in aqueous media is provided in Supplementary Notes 1-3 Another key factor for the utilization of hyperpolarized phosphates is the actual percent polarization achievable in solution. This polarization level is intimately dependent on the solution state T 1 of the particular nucleus in the specific molecule, solvent, mixture, temperature, pressure, and magnetic field throughout the path of the hyperpolarized media to the spectrometer and within it. For d-DNP studies, the polarization level in solution is primarily dependent on the polarization level achieved in the solid state, which is dependent on the formulation, microwave irradiation frequency, temperature, magnetic field, and the duration of polarization. Nevertheless, the following is a brief review of previously reported polarization levels in solution. We have demonstrated here 31 P polarization levels of 9.3%. This polarization level is within the same order of magnitude of 13 C polarization achieved on d-DNP studies (e.g., 5.5% 26 , 4.2% 23 , and 24% 27 ) and is higher than the 31 P polarization level achieved using para-hydrogen-based experiments (2.3% 20 , 0.3% 21 ). Other examples of polarization levels previously documented in solutions are 1.3-9.1% for 19 F 15 , and 0.009 10 to 2% 9 for 1 H. To demonstrate one potential biochemical application of hyperpolarized Pi, we carried out experiments in which the hyperpolarized Pi reported on the pH of the microenvironment in a rapid pH change process, with a temporal resolution of 1 s. pH sensing in intact tissues in living subjects is a long sought target with immediate diagnostic and therapeutic implications in various diseases and conditions 25 . Pi has long been recognized as a promising probe for imaging tissue pH 28,29 , as (1) its chemical shift is sensitive to changes of pH in the range of 6.4-7.6, which is the relevant physiological range (Supplementary Note 2 and Supplementary Fig. 1), (2) it is a non-toxic endogenous substrate present in high concentrations physiologically (0.3-1.1 mM in blood 30 ), (3) it requires no expensive isotopic labeling as 31 P is 100% naturally abundant, and (4) its gyromagnetic ratio is high (γ 31P /γ 1H = 0.40). However, the clinical relevance of Pi for pH sensing has been limited by two factors. First, the signal-to-noise ratio (SNR) of in vivo 31 P spectra is low due to the relatively long T 1 relaxation time of Pi combined with the inherently low SNR ratio of thermal-equilibrium 31 P MR, dictating acquisition times that are too long to be clinically useful. Second, previous studies have shown that the endogenous Pi signal appearing on in vivo 31 P spectra predominantly originates from the intracellular space 29 , which is not a sensitive indicator of disease, unlike extracellular pH that has been shown to vary in a large number of important diseases and to be a marker of tissue changes in response to ischemia 31 , malignancy 32 , and metastasis 29 . The approach described herein surmounts these two obstacles: (1) the d-DNP method provides a very high SNR and thus allows detection of lower quantities (which can be translated to higher spatial and temporal resolution on imaging); (2) unlike endogenous Pi, the hyperpolarized Pi developed here could serve as a potential reporter of extracellular pH, as, depending on transport kinetics, it will likely be primarily located in the extracellular space during the short T 1 limited window of observation.
The following rough calculation is provided to give an estimate for the potential utility of Pi in vivo, considering the signal enhancement demonstrated herein. The concentration of Pi in the hyperpolarized media measured here was ca. 1 mM. We shall consider the case in which such a concentration is reasonably achieved in the tissue. The detected volume in our probe was about 1.3 ml. Let us consider a potential spatial resolution of 5 × 5 mm 2 and a slices thickness of 5 mm. This results in voxels with a volume of 125 mm 3 . Thus, in the volume detected in the current experiment (1.3 ml), there are about 10 such potential voxels. All of the hyperpolarized studies presented here show an SNR of more than 100 (per 1 mM), even after ca. 15-20 s that would be required for circulation through the vasculature.
Therefore, it appears that an in vivo observation is warranted with a potential SNR higher than 10 (per 1 mM per 5 × 5 × 5 mm 3 voxels). Many more factors should be taken into account: one factor that will increase the SNR on imaging is the fact that one actually excites each voxel many times, depending on the acquisition sequence, thereby increasing the SNR of each voxel correspondingly. Factors that may reduce the potential signal are (1) a reduced sensitivity of the in vivo radio frequency (RF) coils in a pre-clinical MRI compared to the high resolution NMR spectrometer and 10-mm probe used here, and (2) possible T 1 shortening by blood and tissues. The sensitivity of RF coils can be optimized and should be checked back-to-back with the relevant coils. The main shortening effect in blood and tissues is likely due to oxygen. In our hands, oxygen did not dramatically shorten the T 1 of hyperpolarized Pi (Supplementary Note 5). In short, in vivo experimentation is warranted keeping in mind that each part of the detection system (sequences and hardware) should be optimized for the detection of hyperpolarized 31 P nuclei.
We note that NMR chemical shifts are relative properties and therefore a chemical shift standard would be required for hyperpolarized Pi to actually serve as such a reporter on tissue pH. However, the SNR of the hyperpolarized Pi is much larger than the thermal equilibrium SNR meaning that other endogenous phosphates, which could potentially serve as references for chemical shift, are unlikely to be observed simultaneously with the hyperpolarized Pi. One potential solution to this problem would be to generate a map of relative differences of all of the observed chemical shifts. Another potential solution would be to combine another hyperpolarized phosphate with a chemical shift that is insensitive to pH. Hyperpolarized PCr which was developed here is a potential candidate for this purpose as its chemical shift is insensitive to pH at the physiological pH range. Thus, by combining two agents with different chemical shifts and different chemical shift pH sensitivities, one could accurately determine the pH of the microenvironment.
We note that using the d-DNP technique 1 , it was previously shown that detection of hyperpolarized bicarbonate and carbon dioxide provides a measure of tissue pH 12,33 . However, the particular tissue compartment inspected in this way is inherently dependent on the hyperpolarized substrate administered to the subject (pyruvic acid 33 or bicarbonate 12 ) and the kinetics of membrane transport and metabolism in the specific tissue inspected 31 .
Here, we propose that hyperpolarized Pi could serve as a robust pH sensor for tissue pH. The data suggest that hyperpolarized Pi could be imaged with high SNR on a time scale of seconds (calculation above). In addition, co-polarization and dissolution of Pi with PCr can provide an accurate internal standard for chemical shift.
The solid-state polarization, and therefore the SNR of hyperpolarized phosphates in solution, could likely be improved by further formulation trials. However, we note that in the current state of the art of d-DNP instruments, it is not possible to monitor the solid-state build-up of 31 P polarization. This makes formulation development for 31 P DNP cumbersome and costly as the polarization build-up can only be monitored after dissolution of the hyperpolarized sample. Further studies are required to robustly characterize the solid-state build-up time constants of Pi and PCr and the factors that may affect them. Such studies are currently underway in our laboratory.
In summary, we have observed here a DNP-driven hyperpolarized state of 31 P in aqueous solutions. Using this technology, this state was observed in two biochemically important molecules, Pi and PCr. Factors such as solvent de-protonation, a basic pH, and media composition increased the T 1 of Pi and therefore also increased the time duration at which this signal could be observed. While the comparably short T 1 remains a challenge, this work warrants further research in other important phosphate-containing biological molecules that could potentially shed light on fundamental biochemical and physiological processes.

Methods
Chemicals. K 2 HPO 4 , KH 2 PO 4 , and PCr (as disodium salt) were purchased from Sigma-Aldrich, Rehovot, Israel. The OXO63 radical (GE Healthcare, UK) was obtained from Oxford Instruments Molecular Biotools (Oxford, UK). The citrate-Tris medium used here contained 100 mM citrate, 10 mM KCl, and 95 mM TRIS, all purchased from Sigma-Aldrich.
DNP spin polarization and dissolution. Spin polarization and fast dissolution were carried out in a d-DNP set-up (HyperSense, Oxford Instruments Molecular Biotools, Oxford, UK). A microwave frequency of 94.097 GHz was chosen based on the work of Reynolds et al 16 . In this report, a solid-state polarization of 31 P was achieved when the microwave irradiation frequency was set at 0.017 GHz below the microwave frequency required to achieve solid-state polarization of the 13 C nucleus of pyruvic acid. In our spin polarizer, the latter frequency corresponds to 94.114 GHz.
The dissolution process was performed as previously described 1 and detailed in the results section. Briefly, 40-120 mg of the 31 P containing formulation which were placed in a the polarization sample cup were quickly dissolved in 4 ml of superheated aqueous media (170°C and 10 bar). The dissolved hyperpolarized solution was directly injected to a screw cup 10-mm NMR tube in a 5.8 T NMR spectrometer via a Polytetrafluoroethylene (PTFE) line of about 2.4-m length with 6 s of He (g) chase. 31 P spectra were continuously recorded immediately at the start of the dissolution process. The hyperpolarized signals appeared in the spectra at about 15-16 s from the start of the dissolution process (meaning that the dissolution process and the chase of the media into the NMR tube occurred within about 15-16 s). 31 P NMR. 31 P NMR spectroscopy was performed in a 5.8 T NMR spectrometer (RS2D, Mundolsheim, France), using a 10-mm broad band NMR probe. Unless otherwise stated, the chemical shift scale of the spectra presented herein was calibrated based on a separate measurement of an ATP standard sample (127 mM in D 2 O, pH 7), carried out prior to each 31 P acquisition, calibrating the α-ATP signal to −10.03 ppm 34 .
Processing and data analysis. Spectral processing was performed using MNova (Mestrelab Research, Santiago de Compostela, Spain). Integrated intensities were calculated either with MNova or with DMFIT 35 .
Determination of the T 1 of the hyperpolarized sites was performed by curve fitting of the signal decay to the following equation: M t ð Þ ¼ M o Ã e Àt T 1 À Á Ã cos θ t TR ð Þ , in which TR, the time between excitations, and θ, the nutation angle of excitation, are known. Curve fitting was performed using Matlab (Mathworks, Natick, MA, USA). The absolute enhancement factor was determined by comparing the maximal integrated signal intensity obtained under hyperpolarized conditions to the integrated signal intensity of the same sample at thermal equilibrium acquired with the same nutation angle under fully relaxed conditions. In order to compare the integrated intensities from the thermal and hyperpolarized acquisitions the same spectral acquisition parameters (spectral width, number of points, receiver gain) and processing (apodization, zero-filling) were used and the thermal equilibrium was corrected for number of scans.
Data availability. The data that support the findings of this study are available from the corresponding author on request.