Effects of contrast agents on relaxation properties of 31P metabolites

Phosphorous MR spectroscopy (31P‐MRS) forms a powerful, non‐invasive research tool to quantify the energetics of the heart in diverse patient populations. 31P‐MRS is frequently applied alongside other radiological examinations, many of which use various contrast agents that shorten relaxation times of water in conventional proton MR, for a better characterisation of cardiac function, or following prior computed tomography (CT). It is, however, unknown whether these agents confound 31P‐MRS signals, for example, 2,3‐diphosphoglycerate (2,3‐DPG).

Purpose: Phosphorous MR spectroscopy ( 31 P-MRS) forms a powerful, non-invasive research tool to quantify the energetics of the heart in diverse patient populations. 31 P-MRS is frequently applied alongside other radiological examinations, many of which use various contrast agents that shorten relaxation times of water in conventional proton MR, for a better characterisation of cardiac function, or following prior computed tomography (CT). It is, however, unknown whether these agents confound 31 P-MRS signals, for example, 2,3-diphosphoglycerate (2,3-DPG).

Methods:
In this work, we quantitatively assess the impact of non-ionic, low osmolar iodinated CT contrast agent (iopamidol/Niopam), gadolinium chelates (linear gadopentetic acid dimeglumine/Magnevist and macrocyclic gadoterate meglumine/ Dotarem) and superparamagnetic iron oxide nanoparticles (ferumoxytol/Feraheme) on the nuclear T 1 and T 2 of 31 P metabolites (ie, 2,3-DPG), and 1 H in water in live human blood and saline phantoms at 11.7 T. Results: Addition of all contrast agents led to significant shortening of all relaxation times in both 1 H and 31 P saline phantoms. On the contrary, the T 1 relaxation time of 2,3-DPG in blood was significantly shortened only by Magnevist (P = .03).
Similarly, the only contrast agent that influenced the T 2 relaxation times of 2,3-DPG in blood samples was ferumoxytol (P = .02). Conclusion: Our results show that, unlike conventional proton MR, phosphorus MRS is unconfounded in patients who have had prior CT with contrast, not all gadolinium-based contrast agents influence 31 P-MRS data in vivo, and that ferumoxytol is a promising contrast agent for the reduction in 31 P-MRS blood-pool signal.

K E Y W O R D S
contrast agent, magnetic resonance spectroscopy, phosphorus-31

| INTRODUCTION
Cardiac MR (CMR) imaging is the gold-standard noninvasive technique used to diagnose heart disease. It can quantify cardiac structure, function, viability as well as myocardial metabolism directly. 1 The latter is commonly probed by in vivo phosphorus MR spectroscopy ( 31 P-MRS), which allows the assessment of high-energy metabolites, for example, adenosine triphosphate (ATP) and phosphocreatine (PCr). In particular, the PCr/ATP ratio remains a biomarker of high interest, as it changes in most cardiac pathologies, [2][3][4][5] predicts mortality in patients with dilated cardiomyopathy, 2 and decreases even in systemic diseases, such as obesity 6 and type-II diabetes. 7 31 P-MRS is commonly combined with other imaging modalities, both MR and non-MR based, for a more comprehensive cardiac characterisation. Cine imaging, providing cardiac volumes and function, 8 and late gadolinium enhancement (LGE) assessment of tissue viability 9 are the most commonly used cardiac MR methods. In order to clearly resolve any necrotic or fibrotic tissue, 10 LGE scans rely upon the extravasation of gadolinium contrast agents, which reduce the longitudinal relaxation time (T 1 ) of the water signal. 11 As this effect persists for a length of time that is variable depending upon (unknown) pathology, LGE scans are typically performed after all other proton ( 1 H) CMR sequences, in order to avoid any potential confounding effects of gadolinium on the native contrast. Furthermore, as the effect of gadolinium on 31 P metabolites is not well characterized, as a potentially unnecessary precaution 31 P-MRS data sets are also acquired before LGE. From non-MR imaging techniques, contrast-enhanced computed tomography (CT) is frequently used in cardiology, for example, for coronary artery imaging. CT contrast agents are radiopaque and not based on gadolinium, but rather iodine. Still, they shorten the T 1 relaxation time, as well as the transverse (T 2 ) relaxation times of water signal in 1 H MRI. [12][13][14] These effects can influence the interpretation of native contrast in subsequent MR scans and can last for up to 24 h after the CT scan. 13 Again, any potential confounding effects on 31 P-MRS remain unknown.
Although it is possible to postpone contrast enhanced examinations until after the 31 P scan, this delay might be scientifically/logistically undesirable and negatively affect patient comfort. Therefore, our aim was to determine the effects of gadolinium-and iodine-based contrast agents on the relaxation times of 31 P metabolites in order to evaluate potential influence of prior contrast enhanced scans on the quantification of the cardiac PCr/ATP ratio. In addition, we have investigated an iron oxide nanoparticle agent, ferumoxytol, which is an iron replacement product that is licenced and used to treat iron deficiency anaemia. However, it has recently been shown that ferumoxytol has a T 2 shortening effect on 1 H signals and may act, therefore, as a negative-contrast agent in MR examinations, in an off label use. 15 This could help suppress the signals originating in the blood pool, for example, 2,3-diphosphoglycerate (2,3-DPG) that contaminate cardiac 31 P-MRS spectra and, especially at lower fields, obscure the inorganic phosphate (Pi) resonance in the cardiac spectra. 4,16 2 | METHODS All measurements were performed using a vertical bore 11.7T MRI system (Magnex Scientific/Varian DDR2) equipped with a dual-tuned 1 H/ 31 P radiofrequency, transmit/ receive birdcage coil (20 mm diameter, Rapid Biomedical) under temperature-controlled conditions at 21 ± 1°C (Bruker BVT3000). This temperature was chosen as a compromise between physiological state, isothermal homogeneity, and blood viability (which is maximized at lower temperatures).
Fresh human blood was obtained from four healthy volunteers via antecubital fossa venepuncture and stored in 4 ml aliquots in an ehtylenediamine tetraacetic acid (EDTA) buffer containing 1.8 mg EDTA per milliter of blood (Vacutainer, BD Healthcare). All experiments performed on human tissue adhered to the Declaration of Helsinki and Caldicott principles, and all volunteers gave informed consent following local institutional procedures and ethical review. All samples were manually shimmed on a third-order shim set to obtain an adequate linewidth of water signal, typically ~30 Hz, desirably ≤ 50 Hz and always ≤ 100 Hz (~0.1-0.2 ppm). A flip angle power calibration was performed before each set of acquisitions on a given phosphate buffered saline (PBS) or saline sample, and this calibration was also performed on each separately drawn tube of blood. T 1 relaxation times were measured on both channels ( 1 H and 31 P) in one vial sequentially after sampling using a pulse-acquire progressive saturation sequence with eight repetition times (TRs) ranging from 0.11 s, to 12 s for 31 P and 13 TRs ranging from 0.007 s to 6 s for 1 H. Other measurement parameters were as follows: 50 µs 90° rectangular pulse, 16 averages, 10 kHz bandwidth. 512 dummy scans were used before each set of averages, effectively acting as a spoiler between TRs. T 2 relaxation times were computed via the Carr-Purcell-Meiboom-Gill (CPMG) 17 pulse sequence (with τ = 2.9, 5.8, 11.5, 23, 46, 92 ms; 64 averages on 31 P in blood, 1 on 1 H with 16 dummy scans, with a TR chosen to be far longer than T1, that is, 7 s TR for 1 H, 50 s TR for 31 P in phantoms; 15 s in blood.). Short rectangular radiofrequency (RF) pulses were again used in the canonical CPMG phase cycling scheme, that is, 50 µs 90°x followed by a chain of 100 µs 180°y pulses. Phase cycling was performed between averages as described previously to ameliorate imperfect refocusing. 18 Owing to signal-to-noise ratio (SNR) limitations of short TR and long TE data primarily, only the dominant signals in their | 1807 respective spectra, that is, either the sum of the two 2,3-DPG signals (which have a near-identical T 1 owing to the symmetry of the molecule 19 ) or the water signal, were used for relaxation times calculation.
To assess the effects on other metabolites, high-SNR, partially saturated data were also acquired with short TR and a 90° flip angle. In particular, a 0.1 s TR and 16 averages were used for 1 H spectra and 0.75 s TR and 512 averages for 31 P spectra. All other parameters were identical to the T 1 measurements.
All measurements were performed before and after the addition of a clinically indicated dose of either an iodinated contrast agent, iopamidol (1.5 mL/kg, ie, 80 µL/vial); ferumoxytol (4 mg/kg, ie, 6 µL/vial); Dotarem (Gd-DOTA) or Magnevist (Gd-DPTA; both 0.2 mL/kg; ie, 10 µL/vial given the near-identical relaxivity of the two agents at high field 20 ), and all experiments were finished within 2 h after blood sampling. A separate experiment showed no changes in blood 31 P-MRS signals over this duration. The T 1 and T 2 experiments were repeated for each contrast agent using normal saline (Baxter IV Viaflo Sodium Chloride 0.9% w/v) and PBS (1X Dulbecco's PBS without Ca 2+ or Mg 2+ powder [Biochrom AG/Sigma Aldrich] containing 137 mM NaCl, 2.7 mM KCl, 8 mM of Na 2 HPO 4 and 1.5 mM of KH 2 PO 4 ) filled tubes, for comparison with literature. At least four vials of blood, saline and PBS were used for each contrast agent.
Spectra were quantified by fitting either a single peak in 1 H and 31 P PBS spectral data, or separate multiple Lorentzian lineshapes in the 31 P blood spectra using the OXSA toolbox. 21 Both T 1 and T 2 relaxation curves were fitted in Matlab (Mathworks). The chemical shift between Pi and α-ATP signals from the high SNR data sets was used to calculate pH of the blood samples directly. 22 The pH in saline and PBS samples was determined using a calibrated Jenway 3510 pH meter with a temperature-corrected microelectrode placed directly in the vial. After quantification, the timecourses of spectral amplitudes were fitted to the functional forms In both cases, a represents the initial signal amplitude, and c a small offset (typically ≪ 1% of a). This parameter is included to compensate for an inadvertent tendency of AMARES to fit a peak to the noise amplitude if no peak is present, which is particularly problematic in the low SNR regime of 31 P MRS.
An unpaired, unequal variance t-test was used to compare computed values with and without contrast agent, with P < .05 considered statistically significant.

| RESULTS
All calculated relaxation times are given in Table 1, and depicted in Figure 1. Figure 2 depicts representative fits of T 1 and T 2 relaxation curves of blood signals. Significantly shorter T 1 relaxation times (by 71-83%) of blood water signal were observed after adding any of the investigated contrast agents except iopamidol (P = .490). The T 1 relaxation time of 2,3-DPG was significantly shortened (by 25%) only when Magnevist was added to the blood sample (P = .033). The only contrast agent that influenced T 2 relaxation times in the blood samples was ferumoxytol (P = .017 and P = .048 for both 2,3-DPG and water, respectively).
In normal saline, the native 1 H T 1 was significantly shortened after mixing with any of the investigated contrast agents. Both gadolinium-based contrast agents shortened T 1 by over T A B L E 1 Relaxation times (T 1 and T 2 ) of 2,3-DPG and water measured in human blood and saline 90%, while iopamidol and ferumoxytol by 14% and 81%, respectively. As expected, analogous results were observed in 1 H spectra of PBS, with T 1 shortening by over 95% with gadolinium-based contrast agents, by 88% with ferumoxytol and by 22% using iopamidol. Similar shortening effects were observed on the 1 H T 2 relaxation times of saline and PBS, where each of the contrast agents caused a decrease in T 2 by a minimum of 67%. However, unlike in the blood data, the 31 P F I G U R E 2 Representative exponential fits of T 1 (left) and T 2 (right) relaxation times of 2,3-DPG (top), water in blood and in saline (middle), and PBS water and 31 P signals (bottom) are depicted for each contrast agent Normalized signal / -relaxation times in PBS were significantly shortened by addition of any contrast agent, including Iopamidol (P = .029 for T 1 and P = .035 for T 2 ). More details can be found in Table 1. Saturated, high-SNR 31 P spectra did not show any visible differences in any metabolite signals after the addition of contrast agents, as can be seen in the representative example in Figure 3A. Similarly, there were no significant differences between the fitted signals of the 31 P blood metabolites preand post-addition of any of the contrast agents investigated (Figure 4). In contrast, high-SNR 1 H spectra of human blood show the significant influence of the investigated contrast agents ( Figure 3B), with a significant increase in water signal intensity observed for Magnevist and Ferumoxytol. The pH of free blood in the vials was 7.4 ± 0.2 throughout the acquisition. No contrast agent effected the blood pH significantly with mean values of 7.5 ± 0.1 for iopamidol, 7.4 ± 0.1 for Dotarem, 7.4 ± 0.1 for Magnevist and 7.4 ± 0.2 for ferumoxytol. Similar results were found in PBS (free PBS pH 6.2; PBS + iopamidol pH 6.2; PBS + Dotarem 6.2; PBS + Magnevist pH 6.2; PBS + ferumoxytol pH 6.3) and saline F I G U R E 3 Representative spectra acquired from whole human blood before and after addition of contrast agents. A, 31 P-MRS data show that the magnitudes of 2,3-DPG, PDE, γ-ATP, and α-ATP are all unaffected by any contrast agent. B, 1 H-MRS data show that water signal is significantly increased after adding either of the gadolinium contrast agents as well as after ferumoxytol

| DISCUSSION
As contrast agents form an invaluable part of many clinical imaging protocols, it is desirable to understand their effects on subsequent scans to avoid misinterpretation. However, the mechanisms of contrast agents are diverse, physically complex and often difficult to predict. In this study, we have experimentally investigated the influence of several different contrast agents on 31 P and 1 H MRS resonances in human blood and saline. We found differences in their effect on the relaxation times of saline, PBS, and blood signals, and more importantly that effects observed in 1 H signals are not directly transferrable to 31 P data.
Both investigated gadolinium-based contrast agents caused a significant shortening of the 1 H T 1 of water signal in blood and in saline and PBS phantoms, which is in accordance with the literature. [8][9][10] This chelate-dependent relaxivity has previously been demonstrated also for several non-proton nuclei, 23 including inorganic phosphate in solution. 24 This is in good agreement with our results in PBS where the T 1 relaxation times of the 31 P signal were shortened after adding gadolinium-based contrast agents. However, this relaxation mechanism is only expected to occur in freely diffusible small molecules in solution, which permit a proton exchange mediated relaxation mechanism between molecule and agent. In the case of 31 P-MRS in freshly sampled blood, which simulates the in vivo situation, metabolite signals are localized within the cell, and these contrast agents are extracellular. Thus, the relaxation agent cannot interact with the metabolite; hence, a direct first-sphere exchange mechanism would not predict a decrease in T 1 after addition of a gadolinium-based contrast agent. While our Dotarem results support this hypothesis, as the T 1 remained unchanged, we observed the shortening of the 31 P T 1 of the 2,3-DPG blood signal using Magnevist. This might be explained by the different structure of these two gadolinium-based substances: Dotarem is a macrocyclic chelate and Magnevist a linear one. Similar effects have been recently observed in the 13 C signal of hyperpolarised pyruvate, where linear, open-chain contrast agents reduced the T 1 of [1-13 C] pyruvate by up to 62%, while macrocyclic agents only by up to 25%. 25 In our case, this could be potentially related to the reported ability of 2,3-DPG to directly change the biophysical membrane properties in erythrocytes, pointing toward a biophysical interaction between 2,3-DPG and cell membrane. 26,27 Thus, by an interaction of the agent to the extracellular membrane alone, allowing Magnevist to effect 2,3-DPG via an exchange mechanism that is possible because of the different solvation spheres of the two gadolinium agents considered, forming a second-order relaxation effect. Albeit significant, this T 1 shortening effect of Magnevist on 31 P signal in blood observed in our current work was only minor (25%) in comparison to the effect on 1 H water signal (83%). Furthermore, neither of the gadolinium contrast agents affected the T 2 of water, nor 2,3-DPG, in human blood. Importantly, while future metabolic studies should take into account that linear chain contrast agents could potentially influence the T 1 of 31 P metabolites, the effect of macrocyclic contrast agents on T 1 of in vivo 31 P-MRS data might be potentially ignored.
While iopamidol is a widely used CT contrast agent not typically used in MRI, it has slow clearance kinetics, and hence, can still be present in the body 24 hours after a contrast enhanced CT scan. We have observed significant shortening of both T 1 and T 2 relaxation times in saline and PBS solutions, which is in good agreement with literature. 12,14,28 However, these effects are highly concentration dependent, with negligible impact in tissues when the contrast agent is diluted. 13,29 Our results support this hypothesis, as after adding the prescribed clinical dose of Iopamidol (~2% concentration of contrast agent), it was not possible to observe any effect on the blood 1 H relaxation times. Similarly, no effect was observed on the 31 P signals in fresh human blood. We, therefore, conclude that differences in perfusion kinetics dominate, and that any potential confounding caused by prior CT would be pathology dependent.
Ferumoxytol is known to be an MR negative-contrast agent and, thus, was expected to have a significant 1 H T 2 shortening effect. This was observed for both 1 H as well as 31 P blood signals, with blood water T 2 decreasing from 8.12 ± 5.64 ms to 2.14 ± 0.43 ms (P = .048) and 31 P 2,3-DPG T 2 reducing from 8.03 ± 1.06 ms to 6.26 ± 0.47 ms (P = .017), respectively. Furthermore, a shortening of the T 1 relaxation time after the addition of Ferumoxytol was observed in blood 1 H signal as well as in saline phantoms, but not in the 31 P blood pool signal. This discrepancy is most probably again caused by the intracellular location of 31 P metabolites: while the comparatively far-field alterations to magnetic field homogeneity caused by Ferumoxytol can affect the 31 P metabolites bound within cells shortening their T 2 , there is no direct interaction between the contrast agent and 31 P metabolites to affect their T 1 . However, the T 2 shortening effect, together with the long blood pool phase of ferumoxytol suggest that it might be of great use in suppressing blood pool signals in 31 P-MRS. 30 All our experiments were performed at ultra-high field, which could be considered a limitation of our study, as relaxation times vary with the external magnetic field. We note also that the experiments were not performed at 37°C, which will further quantitatively alter the measured relaxation times and contrast agent relaxivities, approximately linearly with temperature, 31,32 but argue that this compromise was necessary to the success of the experiment, as evidenced by the stability of the pH of the blood. We note that a simple phantom experiment would not represent the in vivo situation well and, as our data show, would results in incorrect conclusions. Therefore, in order to secure sufficiently high SNR and scan times short enough to ensure the ex vivo viability of small blood samples, the use of ultra-high field, constant slightly reduced temperature, and RF coils matched to the size of the sample vials was essential. We also note that, in particular, the T 2 and, for larger 31 P metabolites also the T 1 , shortens with increasing field strength. 33 Therefore, although quantitatively likely to be different in magnitude, these effects of the tested contrast agents will exist at clinical field strengths, and may well be more pronounced at lower fields. 34 31 P-MRS is affected distinctly by different contrast agents to conventional proton MRS/MRI, and in particular is not subject to the potentially confounding effects from Dotarem (Gd-DOTA) or iopamidol that may have been administered separately. This fact both aids protocol planning within an MR session, and additionally enables greater flexibility in multiparametric investigations. Moreover, we have provided the novel demonstration that ferumoxytol does significantly reduce the 31 P T 2 of 2,3-DPG in blood, which may be exploited by future studies to supress unwanted blood signal in cardiac 31 P MRS.