Determination of oxygen relaxivity in oxygen nanobubbles at 3 and 7 Tesla

Objective Oxygen-loaded nanobubbles have shown potential for reducing tumour hypoxia and improving treatment outcomes, however, it remains difficult to noninvasively measure the changes in partial pressure of oxygen (PO2) in vivo. The linear relationship between PO2 and longitudinal relaxation rate (R1) has been used to noninvasively infer PO2 in vitreous and cerebrospinal fluid, and therefore, this experiment aimed to investigate whether R1 is a suitable measurement to study oxygen delivery from such oxygen carriers. Methods T1 mapping was used to measure R1 in phantoms containing nanobubbles with varied PO2 to measure the relaxivity of oxygen (r1Ox) in the phantoms at 7 and 3 T. These measurements were used to estimate the limit of detection (LOD) in two experimental settings: preclinical 7 T and clinical 3 T MRI. Results The r1Ox in the nanobubble solution was 0.00057 and 0.000235 s−1/mmHg, corresponding to a LOD of 111 and 103 mmHg with 95% confidence at 7 and 3 T, respectively. Conclusion This suggests that T1 mapping could provide a noninvasive method of measuring a > 100 mmHg oxygen delivery from therapeutic nanobubbles. Supplementary Information The online version contains supplementary material available at 10.1007/s10334-022-01009-3.


Introduction
Tumour hypoxia is a predictor of disease progression, treatment failures, and metastatic potential in multiple types of cancer [1][2][3][4]. Strategies for treating hypoxia have included methods such as blood transfusions, hypoxia-selective drugs, and direct oxygen delivery [5]. The primary methods for oxygen delivery have generally focused on: hyperbaric oxygen or high oxygen content breathing therapy, oxygen-generating materials, and oxygen-carrying materials [6][7][8]. More recent approaches to hypoxia mitigation have taken advantage of developments in drug-delivery systems to deliver oxygen to the tumour. Of these drug-delivery approaches, therapeutic and diagnostic applications of nano-and micromaterials are playing increasingly important roles [9], such as micro and nanobubbles which can be used to encapsulate oxygen [10][11][12][13][14][15][16]. McEwan et al. [13] have shown impressive results using microbubble-delivered oxygen for improving sonodynamic therapy in pancreatic tumours, Eisenbrey et al. [17] have successfully increased breast tumour oxygenation levels in vitro by 20 mmHg, and Owen et al. [16,18] have demonstrated that oxygen nanobubbles enhanced tumour response to sonodynamic therapy. These results have illustrated the potential of oxygen loaded nano-and microbubbles to deliver oxygen to the tumour microenvironment, reduce tumour hypoxia, and improve treatment outcomes [13,16].
However, despite the enhanced therapeutic responses, statistically significant changes in tumour oxygenation were not always observed when this was directly measured [13,17,18]. This makes it extremely difficult to establish the underpinning mechanism(s) by which the micro/nanobubbles are promoting therapeutic effects. A key challenge has been that the methods available for measuring changes in oxygenation have multiple limitations. For example, Owen et al. [18] used a single fibre-optic probe and were only able to probe one spot in the tumour. Since tumours are highly heterogeneous in terms of perfusion, this resulted in a very high variability in the measurements. Alternatively, techniques may only be suitable for measuring either dissolved oxygen or oxygen bound to haemoglobin and thus not able to detect oxygen encapsulated within bubbles. For example, Eisenbrey et al. were unable to detect any change in oxygenation using photoacoustic imaging, but this is unsurprising since it relies on comparing ratios of oxygenated and deoxygenated haemoglobin, whereas the delivery of oxygen using microand nanobubbles is independent of haemoglobin transport [19]. Therefore, to continue effective research on promising oxygen carriers and hypoxia reduction, robust methods for measuring oxygen delivery from these carriers are needed.
The ideal method should be noninvasive and quantitative, allowing tumour oxygen measurements to be obtained before, during, and after treatment. As magnetic resonance imaging (MRI) is a widely clinically available, non-ionizing, noninvasive imaging technique, it is very well suited for determining tissue oxygenation. Molecular oxygen is paramagnetic [20][21][22][23][24][25][26][27], and it has been demonstrated experimentally repeatedly that the relaxation rate of the imaged material, R 1 , is linearly proportional to the concentration or partial pressure of oxygen (PO 2 ) [21,[28][29][30][31][32][33][34][35][36][37][38][39][40]: where R 1Ox is the relaxation rate with oxygen added, R 10 is the relaxation rate without oxygen, C is the concentration or partial pressure of oxygen, and r 1Ox is the relaxivity of oxygen in the imaged material (whose units depend on the oxygen measurement used for C). Since the partial pressure of oxygen (PO 2 ) is a common measurement in biomedicine and clinical applications, in this manuscript, we report C as PO 2 in mmHg and r 1Ox as s −1 /mmHg.
This linear, reproducible relationship between PO 2 and R 1 has been used to infer oxygen levels in vitreous fluid as a noninvasive alternative to the highly invasive oxygen electrodes used to measure retinal hypoxia [28,30,39,41], bladder urine [41] and urine in the renal pelvis to create a noninvasive detection of renal dysfunction [42], cerebrospinal fluid [33,41], brain tissue [43,44], and blood, including foetal blood [36,45,46]. Therefore, we hypothesized that this method could also be applicable as a noninvasive method for measuring oxygen delivery from these nanocarriers.
However, changes in R 1 are not specific to oxygen: the R 1 of the material can be affected by the temperature [39,41], pH [47], protein concentration [30], field strength [48], and numerous other changes in the chemical environment [49]. Therefore, unless the specific PO 2 -R 1 relationship was measured in the material of interest previously (with consistent temperature, scanner protocol, and protein concentration), absolute oxygen levels cannot be inferred from R 10 [41], only a relative change in oxygenation, since the y-intercept of the linear relationship is not known. The nanobubbles used for oxygen delivery in the studies discussed above are small lipid-based particles, ~ 100 nm in diameter, within an acidic (pH 2.3) solution. The relaxivity of oxygen in this nanobubble mixture has not yet been measured, however, analogous work by Thompson et al. demonstrated a measurable change in oxygenation in a phantom before and after destruction of oxygen-filled microbubbles [50]. Thompson et al. found that the linear relationship between PO 2 and R 1 in a phantom containing microbubble solutions of varying PO 2 showed a relaxivity of 0.0003 s −1 /mmHg at 7 T. With a similar objective, Vatnehol et al. measured the relaxivity and limit of detection of oxygen at 3 T in "oxygen-enriched" water intended for therapeutic oxygen delivery [37].
We have built upon their work in the experiments in this manuscript to investigate whether MRI T 1 mapping is a suitable technique to examine oxygen delivery from such carriers. Therefore, we hypothesized that this method could also be applicable as a noninvasive method for measuring oxygen delivery from these nanocarriers and that the additional substances in the nanobubble solution would not interrupt the linear relationship between R 1 and PO 2 in the solution. To examine this, we performed two separate experiments to find the relaxivity and limit of detection of oxygen in this nanobubble mixture in two experimental scenarios: a preclinical 7 T MRI and clinical 3 T MRI.

Phantom design
For the 7 T experiment phantom, seven vials containing different solutions were prepared: oxygen-filled nanobubbles, nitrogen-filled nanobubbles, air-filled nanobubbles, water, oxygenated water, and a half oxygen-filled and half nitrogenfilled nanobubble mixture. One 0.5 ml syringe was filled per solution and sealed using Cristaseal wax (ProSciTech Pty Ltd, Australia). A seven-chamber custom-built holder was used to hold seven 0.5 ml syringes in the scanner for the duration of the phantom experiment.
For the 3 T experiment phantom, glass vials (10 ml) containing different solutions were prepared: five vials of airfilled nanobubbles, five vials of oxygen-filled nanobubbles, five vials of water, and five vials of water sparged with 100% oxygen. All nanobubble and water solutions were prepared according to the method listed in the section below. The vials were placed in a cylindrical vessel with saline solution surrounding them for the duration of the phantom experiment.
Lastly, to examine the effect that temperature may have on the experiment, three phantom tubes from a calibration phantom (Eurospin II TO5 phantom, Diagnostic Sonar LTD, Livingston, Scotland) were used in the scanner and a Variable Flip Angle (VFA) T 1 map acquired five times, while monitoring the temperature. The three phantom tubes had nominal T 1 values of 830, 1020, and 1350 ms.

Nanobubble preparation
All nanobubble solutions were prepared according to the following method: lecithin and citric acid were obtained from Special Ingredients (Chesterfield, Derbyshire, UK). Glycyrrhizic acid and glycerol were obtained from Sigma-Aldrich Ltd (Gillingham, Dorset, UK). Oxygen and nitrogen cylinders were obtained from BOC gases (Guildford, Surrey, UK). The nanobubble solutions were prepared according to Owen et al. (2016). To create one 100 ml bottle of the solution, glycyrrhizic acid (0.5 mg/ml), lecithin (3 mg/ml), citric acid (5 mg/ml) and glycerol (0.0125 ml/ml) were mixed with 100 ml of freshly boiled, filtered deionized water. The solution was then stirred for 30 min, while it cooled to room temperature. The vial was then immediately sealed and mechanically agitated for 30 s. To produce oxygen-laden nanobubbles, 5 ml of the solution was transferred to a glass vial and sparged with oxygen gas, for 3 min. To produce nitrogen or air-laden nanobubbles, the solution was sparged with nitrogen gas or air for 3 min. The nanobubble size distribution measurements were ascertained using a nanoparticle tracking analyzer (NTA) (Nanosight NS300, Malvern, PA) and all size distributions were consistent with results published by Owen et al. [16].

MRI acquisition details
The 7 T imaging experiments were performed using a 7.0 T 210 mm horizontal bore VNMRS preclinical imaging system equipped with 120 mm bore gradient insert (Varian Inc. Palo Alto, CA, USA). A variable flip-angle T 1 map [51] was calculated using non-linear least squares from 3D RF and gradient spoiled gradient-echo sequence with 16 [52]. The 16 flip angles were chosen according to the Ernst angle for a large range of values of tissue T 1 -using more flip angles allows for a more accurate fit for the calculation of T 1 . T 1 values were measured and averaged in a region of interest (ROI) placed lengthwise along each syringe to produce a mean T 1 value for each syringe.
The 3 T imaging experiments were performed using a 3.0 T Siemens Prisma MRI (Erlangen, Germany). The VFA T 1 maps were calculated from a linear fit of [51] 3D gradient-echo images with five different flip angles (TR = 4.1 ms, TE = 1.23 ms, FA = 3, 6, 9, 12, 15, slice thickness = 3 mm). The Modified Look-Locker Inversion Recovery (MOLLI) [53] and Shortened Modified Look-Locker Inversion Recovery (ShMOLLI) T 1 maps were calculated from a non-linear fit using a MOLLI T 1 -mapping sequence (TR = 3.5 ms, TE = 1.05 ms, 11 inversion times, FA = 35, slice thickness = 8 mm), and Shortened Modified Look-Locker Inversion Recovery (ShMOLLI with the acquisition details: (ShMOLLI_192i protocol, TR = 371.84 ms, TE = 1.01 s, FA = 35, slice thickness = 8 mm, 7 inversion times TI = 100, 1100, 2100, 3100, 4100, 180, 260 ms). Each type of T 1 map was acquired multiple times, repeated in the following order: ShMOLLI 1, MOLLI 1, VFA 1, ShMOLLI 2, MOLLI 2, VFA 2, ShMOLLI 3, MOLLI 3. The ShMOLLI T 1 -mapping method produced the lowest standard deviation within each vial ROI and was, therefore, used for the relaxivity and limit of detection (LOD) calculations (see Supplementary Figure S1). Although multiple T 1 maps were acquired for this experiment, the ShMOLLI T 1 map acquired closest to the time point of the collection of the oxygen measurement (ShMOLLI 3) was used to calculate the following relaxivity and LOD results. This decision was made for two reasons: (1) the ShMOLLI 3 and MOLLI 3 data was acquired closest to the time the oxygen measurements were made, and thus provide the T 1 measurement most closely corresponding to the measured PO 2 if oxygen levels were leaking or decreasing over time; and (2) the ShMOLLI T 1 map was chosen over the MOLLI T 1 map due to an artefact corruption on the MOLLI T 1 map. The T 1 measurement from each repetition of the 3 ShMOLLI T 1 maps was compared in Supplementary Figure S2.
The resulting T 1 maps used for analysis are shown in Supplementary Figure S3. Note that the scale bars highlights the size differences between the phantoms, and therefore the field of view shown in each image: the 7 T phantom used small 0.5-ml syringe vials to fit into a small-bore preclinical scanner, while the 3 T phantom used much larger 10 ml glass vials within a large saline bucket. The resolution of the ShMOLLI image is 1.7 mm × 1.7 mm, and the resolution for the VFA slice displayed was 0.5 mm × 0.5 mm.

Oxygen measurements
Following the 7 T scans, the remaining solution in each of the seven larger vials was measured using a fibre-optic oxygen sensor. All PO 2 measurements were carried out using a PreSens OXY Mini-fibre-optic oxygen meter in combination with SP-PSt3-NAU oxygen sensor spots inside sealed glass vials (PreSens Precision Sensing GmbH, Regensburg, Germany). All oxygen measurements were calibrated and carried out at room temperature according to PreSens manual with oxygen-free water and air-saturated water, as performed by Owen et al. [16,18].
Following the 3 T scans, the phantom was carefully removed from the scanner, and one by one, each vial was quickly opened and poured into a vial with a PreSens sensor sticker with the PreSens tip recording the PO 2 measurement (PreSens OXY Mini fibre-optic oxygen meter). The PO 2 was recorded for 30 s, recording one measurement per second. All oxygen measurements were calibrated and carried out at room temperature. The oxygenated samples were recorded first, then the non-oxygenated samples, to reduce the time over which the oxygen levels could be decreasing due to exchange with the room air.

Calculation of limit of detection
The limit of detection of oxygen in the solutions can be calculated by the following equation [54]: where b is the slope of the regression line (relaxivity, or r 1Ox ), SD is the standard deviation of the intended in vivo situation, and F is a factor set to 2.2 for LOD calculations with 95% confidence intervals. This method is used by Vatnehol et al. to calculate the limit of detection of dissolved oxygen in water [37].

Oxygen relaxivity
The phantoms and resulting T 1 maps can be seen in Supplementary Figure S3. We measured a strong linear relationship (R 2 = 0.97 and R 2 = 0.97) between PO 2 and R 1 in the nanobubble solution in vitro at 7 and 3 T, respectively, shown in Fig. 1. The relaxivity (r 1Ox ) at 7 and 3 T was 0.00024 s −1 / mmHg and 0.00057 s −1 /mmHg, respectively, and the corresponding upper and lower confidence intervals and P values are listed in Table 1.

Limit of detection
The SD in the LOD equation represents the standard deviation of the method in the intended tissue. Therefore, for the 7 T preclinical setting, the mean of eight reported standard deviation values of R 1 from the intended experimental tissue-a preclinical tumour model-was calculated, which was approximately 0.011 s −1 [55]. Using the measured relaxivity at 7 T (0.000235 s −1 /mmHg) and Eq. 1, the resulting LOD (95% confidence) at 7 T in a preclinical tumour model is 103 mmHg. For the 3 T clinical setting, the mean of 11 reported standard deviation values of R 1 from clinical tumour ROI was calculated, which was approximately 0.029 s −1 [56]. Using the measured relaxivity at 3 T (0.00057 s −1 /mmHg) and Eq. 1, the resulting LOD (95% confidence) at 3 T in a clinical tumour is 111 mmHg. Of course, it is likely that measurements from T 1 mapping methods will be less precise in vivo than in phantomstherefore, in vivo measurements for SD were used in these LOD calculations.

Nanobubbles and R 1
In comparison to water, the nanobubble solution contains several attributes that may affect R 1 , in particular, the presence of the lipid-based particles, and the acidity (pH 2.3) of the solution. It was apparent from both the 7 and 3 T experiments (Fig. 2) that the nanobubble solution alonewithout oxygen added-produced an increase in R 1 of 0.03 and 0.07 s −1 compared to water; equivalent to increasing the water PO 2 by 230 and 130 mmHg (at 7 and 3 T, respectively). The experiment measuring the effect of temperature on R 1 showed that the R 1 of all three tubes decreased as temperature increased, and a linear regression fit strongly (R 2 = 0.83, 0.90, 0.91). The change in R 1 per change in temperature was − 0.037, − 0.027, and − 0.020 s −1 /°C for the three tubes, respectively (see Supplementary Figure S4), and a steeper slope corresponded to a higher baseline R 1 , and vice versa. The corresponding upper and lower confidence intervals and P values are listed in Supplementary Table S1. This means that from just a 0.5 °C fluctuation in temperature, in R 1 would change by 0.018 s −1 , which would correspond to a 32 and 77 mmHg inaccuracy in PO 2 estimation, at 3 and 7 T, respectively. The same data are also provided in terms of T 1 in Supplementary Figure S5, where a linear regression fit T 1 vs temperature data strongly as well (R 2 = 0.83, 0.90, 0.91) with a slope of 20, 23, and 29 ms/°C for the three tubes, respectively. These results are consistent with findings by Zhang et al. where the temperature sensitivity increased with baseline T 1 , following closely a second-order polynomial (see Fig. 2 in Zhang et al.) [57].

Discussion
The relationship between PO 2 and R 1 has been used to noninvasively infer oxygen levels in vitreous fluid [28,30,39,41], urine [41,42], cerebrospinal fluid [33,41], brain tissue [43,44], and blood [36,45,46]. Therefore, we hypothesized that this method could also be applicable as a noninvasive method for measuring oxygen delivery from these nanocarriers. To examine this, we performed two separate experiments to find the relaxivity and limit of detection of oxygen in this nanobubble mixture in two experimental scenarios: a preclinical 7 T MRI and clinical 3 T MRI.

Relaxivity
These experiments reproduced the linear R 1 -PO 2 relationship expected (R 2 = 0.97), and showed that the PO 2 of the nanobubble solution could be estimated using T 1 mapping at 7 and 3 T. It was a possibility that the composition of the nanobubble solution would interrupt this linear R 1 -PO 2 relationship or induce image artefacts. Fortunately, this was not found to be the case.
The r 1Ox in the nanobubble solution was 0.00024 and 0.00057 s −1 /mmHg at 7 and 3 T respectively. In these experiments, the r 1Ox was stronger at the lower field strength, which agrees with other reports from phantom Fig. 2 The mean R 1 and PO 2 values of the vials of water and oxygenated water (black crosses), in comparison to nonoxygenated nanobubbles (red cross), at A 7 T and B 3 T. The contents of the non-oxygenated nanobubble solution resulted in an R 1 that was 0.03 and 0.07 s −1 higher than water, the equivalent effect as increasing the water PO 2 by 230 and 130 mmHg (at 7 and 3 T, respectively) measurements of r 1Ox being greater at lower field strengths [58][59][60]. The comparable experiments by Thompson et al. [50] found an r 1Ox of 0.0003 s −1 /mmHg at 7 T, which within the upper 95% confidence interval of our measurement performed at 7 T (0.00031 s −1 /mmHg). Therefore, even though the composition of the microbubble solution and nanobubble solutions differ considerably (see Thompson et al. [50] for the microbubble formula), the r 1Ox seems to be almost independent of the nano or microbubble composition, which is a notable strength of this method.

Limit of detection
Using our measured r 1Ox and published values of the expected standard deviation in R 1 within preclinical and clinical tumour tissue, the resulting 95% confidence interval LOD calculated indicates that a 103 mmHg (4.56 mg/L) and 111 mmHg (or 7.04 mg/L) increase in oxygen is required to reliably detect a change in R 1 from oxygen delivery (in a preclinical and clinical setting, respectively). These are both slightly lower than the 8.5-12.3 mg/L limit of detection estimated by Vatnehol et al., however, for their calculations, the authors use the standard deviation in the hepatic portal vein as that is their material of interest [37]. Of course, this raises the most important point of this discussion: these LOD calculations assume that there are no other potential T 1 changing effects present other than an increase in oxygen-which, in the body, is an often unrealistic assumption, especially in a voxel containing blood. The other main endogenous paramagnetic material in the body is deoxyhemoglobin, and the R 1 of blood decreases linearly as oxygen saturation increases (i.e. as deoxyhemoglobin is converted to oxyhemoglobin, which is diamagnetic) [34,61,62]. Only following full oxygen saturation (SO 2 > 99%) will the R 1 of blood increase linearly due to the increase in the PO 2 of the plasma being the dominant remaining effect [29,34,36,45,62]. Therefore, if an oxygen delivery method results in an increase in oxygen saturation of the blood, it is likely to induce a negative change in R 1 first, until it delivers enough oxygen to reach full saturation [62]. In fact, this is precisely what was observed by Vatnehol et al. when they performed their intended oxygen delivery experiment in humans and measured a negative R 1 change in the blood [63]. However, for a voxel that does not contain 100% blood, such as normal tissue, it has been demonstrated that the R 1 of the tissue will increase linearly with the level of dissolved oxygen in the tissue [43]. Since the effect from changing deoxyhemoglobin levels will only pertain to the percent volume in the voxel that is occupied by blood, then the remaining 90-98% of voxel volume will be dominated by R 1 change of the tissue. In addition to deoxyhemoglobin, iron levels present in the tissue would affect R 1 as well, and it can accumulate depending on the disease and organ. Therefore, while the LOD calculations in this manuscript do not account for confounding effects on R 1 , they are still useful to obtain an estimation of whether the detection of nanobubble oxygen delivery by R 1 measurement is feasible.
The expected PO 2 change in the tumour tissue following the administration of nanobubbles has been reported to be approximately 30 mmHg, albeit this was measured using a method that was likely underestimating the PO 2 [16]. This is much lower than both the LODs calculated in this experiment (103 and 111 mmHg), however this LOD calculation is stating that the two peaks of the distributions are separated by 2.2 standard deviations, which is considerable. Therefore, although the expected oxygen delivery is below the measured LOD, this could still be a feasible method for detecting the expected change in PO 2 . In addition, this measurement describes the LOD for a single voxel, which is different from the detection power in a large ROI-due to the repeated number of voxel measures in a large ROI, it would be feasible to detect a much smaller difference between two large ROIs that would otherwise not be realistic in a single voxel. Furthermore, the limit of detection will depend not only on the tumour type, and in particular its T 1 values at a particular magnetic field, but also the site of the tumour-such as the head and neck region versus the pelvic region-which will affect the SNR characteristics, and may be a caveat to this estimate.
In addition, the standard deviation within the tumour ROIs used is a combination of both tissue heterogeneity and T 1 mapping measurement variability, and as T 1 mapping methods continue to improve, it is possible that the standard deviations could decrease significantly and oxygen detection could be made on a voxel-wise level. Therefore, although the resulting LODs calculated from this experiment are greater than the expected PO 2 change, it is possible that as T 1 mapping techniques become more stable, assessment of spatial variation of PO 2 may then be possible compared to region averaging which is needed currently, enabling T 1 mapping to be a suitable technique to examine oxygen delivery from these nanobubbles.
Lastly, the LOD measurement describes the sensitivity of this technique to detecting PO 2 changes in an individual. As a paired measurement in a study across multiple subjects at 3 T, however, assuming that a similar standard deviation as previous clinical R 1 measurements [56], then a study of 10, 13, or 15 participants would yield 80%, 90%, or 95% power, respectively (alpha 0.05) [64].

Additional confounding effects
In comparison to water, the increased lipid content and acidity of the non-oxygenated nanobubble solution resulted in an R 1 that was higher than water. The magnitude of this effect was the equivalent of increasing the water PO 2 by 230 and 130 mmHg (at 7 and 3 T, respectively). Thus, it is likely that increasing the nanobubble concentration in a water or saline phantom solution would increase R 1 , making it difficult to use that change in signal to infer a change in PO 2 . This problem is similar to that reported by Vatnehol et al. [63], where the delivery of a dissolved oxygen water drink caused a dilution that changed R 1 in both the control drink and the oxygen drink. It is possible that this dilution would be a negligible effect from the nanobubbles once spread out in the circulatory system, however, it could make oxygen nanobubble delivery indistinguishable from non-oxygenated nanobubble delivery. Since the percent volume of nanobubble solution would be far below 10% of the blood volume and the tissue voxel would contain below 10% blood volume, and since the low pH would be neutralized in the bloodstream, it is reasonable to conclude that this confound would not be as substantial as the R 1 change from the PO 2 change within the tissue dominating 85-98% of the voxel volume, assuming a range of fractional blood volume in tumours of approximately 2-15% [65].
There are additional remaining factors that can affect R 1 . Experiments using R 1 to measure PO 2 in vitreous fluid have demonstrated the strong effect of temperature on the R 1 measurement [39,41]. Unfortunately, this is not insignificant-Zaharchuk et al. reported that based on their phantom experiments, they found a potential error ± 19 mmHg for physiologic temperature fluctuations of ± 1 °C [41]. We believe this drift in R 1 values shown in Supplementary Figure S2 is due to a combination of two factors: (1) the temperature of the vials increasing due to absorption of heat from RF field exposure, and (2) a slight decrease in PO 2 in the vials as oxygen may be able to slowly leak out of the vial during the duration of the scanning session. Both of these effects independently would cause a decrease in R 1 . This possible decrease in oxygen throughout the scanning session is why ShMOLLI 3, the ShMOLLI acquired closest to the time of oxygen measurement, was used for the analysis.

Experiment limitations
The gold-standard T 1 mapping method to use would have been inversion recovery, since Look-Locker techniques such as ShMOLLI are known to underestimate the true T 1 and VFA will overestimate the true T 1 (seen in Supplementary Figure S1) [66]. However, these experiments were intended as preliminary work for preclinical and clinical experiments, and therefore, it was more appropriate to use the actual experimental set up and scanning techniques that could be used in mice and human volunteers and/or patients. In comparison to phantoms, scanning live subjects introduces issues such as motion from breathing and limitations to scan time, and therefore the faster VFA method and much faster ShMOLLI method were more appropriate than inversion recovery. Lastly, although the T 1 might be over-or underestimated, the relative change in R 1 (and therefore r 1Ox ) should remain the same in each method.
In addition, the 95% confidence interval of the fit of the slope of the measurements was ± 30% at 7 T, and ± 16% at 3 T, which at first glance, suggests that the T 1 measurements from the 7 T scans were less reliable. However, upon further examination of the underlying data, we believe this discrepancy is due to the oxygen measurements rather than the T 1 measurements. For the two data points with the highest PO 2 in the 7 T data, the PO 2 measurements have larger error bars-this is due to those glass vials showing a larger range of PO 2 during the PreSens measurement time due to being disturbed while being moved from the scanner. Therefore, it is possible that the true PO 2 during the time of T 1 measurement was slightly different than it is showing there, which would result in a slightly tighter alignment of the data points and more narrow CI range. However, the T 1 measurements from the 7 T data were of relatively lower quality than T 1 measurements from the 3 T data, which remains a limitation of this study.
A final noteworthy observation is that adjusting the concentration of nanobubble solution in each vial would also alter the R 1 , as shown by the difference between the 'water' vial and 'nanobubble' vial in Fig. 2. We did not perform measurements at multiple concentrations of nanobubbles, so we are unable to provide additional data on this, but we hypothesise that R 1 would decrease change linearly with increasing nanobubble concentration. To clarify, in this experiment, any solution containing nanobubble solution contained the same concentration of nanobubble solution, and the oxygen levels were altered via the gas used, not by the nanobubble concentration.

Conclusion
We measured a strong linear relationship (R 2 = 0.97) between PO 2 and R 1 in the nanobubble solution, and measured the relaxivity of oxygen in the nanobubble solution to be 0.00057 s −1 /mmHg and 0.000235 s −1 /mmHg, and established that a 111 and 103 mmHg increase in oxygen is required to detect a change in R 1 with 95% confidence at 7 and 3 T, respectively. This suggests that T 1 mapping could provide a noninvasive method of measuring a > 100 mmHg oxygen delivery from oxygenated nanobubbles in therapy.