Considering whole‐body metabolism in hyperpolarized MRI through 13C breath analysis—An alternative way to quantification and normalization?

Hyperpolarized [1‐13C]pyruvate MRI is an emerging clinical tool for metabolic imaging. It has the potential for absolute quantitative metabolic imaging. However, the method itself is not quantitative, limiting comparison of images across both time and between individuals. Here, we propose a simple signal normalization to the whole‐body oxidative metabolism to overcome this limitation.


INTRODUCTION
Hyperpolarized [1-13 C]pyruvate MRI is an emerging clinical metabolic imaging tool. It allows for direct measurement of specific metabolic pathways by imaging both the substrate itself and the subsequent breakdown products simultaneously. [1][2][3][4] Currently, the most used substrate is [1-13 C]pyruvate, which is taken up by the cells and enzymatically converted to [1-13 C]lactate by lactate dehydrogenase, [1-13 C]alanine by alanine transferase, and [1-13 C]bicarbonate by pyruvate dehydrogenase, converting pyruvate to 13 CO 2 and subsequently to [1-13 C]bicarbonate by carbonic anhydrase. These enzymatic pathways sit at a particularly important junction, controlling the balance between the anaerobic and aerobic metabolism in both health and disease. [5][6][7][8][9] Hyperpolarized [1-13 C]pyruvate MRI has been used widely preclinically to investigate both acute and chronic kidney disease conditions. [10][11][12][13][14][15][16][17][18][19][20][21] Despite these advantages, hyperpolarized MRI is-like conventional MRI-difficult to absolutely quantify. [22][23][24][25] A common problem in quantitative imaging of perfusion or metabolism is the absence of an accurate arterial input function for quantification. 26,27 This has been shown to be less of an issue for hyperpolarized agents. 28 However, the simple ratiometric approach is unable to account for differences between whole-body conditions, such as body composition, feeding status, age, and metabolic disease states such as fever, diabetes, and thyroid diseases. This has led to attempts to standardize the patient's and animal's physiology before imaging; however, as we migrate toward clinical utility, it is important to consider alternative solutions to this problem.
It is well known that whole-body metabolism varies across individuals due to physiological factors such as genetic variability, 29 and pathological conditions such as diabetes 12,30 and liver disease. [31][32][33] Individual metabolism varies based on the time of day, season, alterations in circadian rhythm, and feeding state. [34][35][36] Such variations mask the disease-related metabolic changes that we wish to observe, and it cannot accurately be compared with signals of previous images on the same subject or with images from different subjects. Hence, variation in whole-body metabolism calls for a method to correct for this in hyperpolarized MRI signals.
Hyperpolarized 13 C MRI relies on the injection of molecules enriched with stable isotopes, like those used in 13 CO 2 breath tests. Such breath tests rely on measuring the 13 C content in CO 2 of expired air before and after ingestion of a 13 C-labeled substrate by benchtop infrared (IR) devices. 31 The 13 CO 2 breath test is an easily accessible and widely used clinical and research tool. Clinically, the test is predominantly used for helicobacter pylori detection using orally administered 13 C-urea. 37,38 The method is also used to assess liver function in patients with nonalcoholic fatty liver disease and hepatocellular carcinoma, using 13 C-ketoisocaproate and 13 C-methiacin. [31][32][33] The method has further been proven to be effective in identifying patients with early-stage diabetes and type 2 diabetes by measuring 13 CO 2 after oral administration of a 13 C-labeled oral glucose tolerance test. [39][40][41] In this study, we propose using a simple 13 C breath test for normalizing the hyperpolarized MRI signals in kidney images to whole-body pyruvate-to-bicarbonate metabolic activity, encompassing global pyruvate uptake and oxidation ( Figure 1). Specifically, we measured 13 CO 2 in expiratory air before and shortly after imaging hyperpolarized [1-13 C]pyruvate. Breath samples were analyzed for 13 C content and used for signal normalization. We questioned whether a whole-body metabolic difference exists between two porcine renal ischemia cohorts and whether normalization using the expired air reduces signal ratio variances within the cohorts.

Study cohort
This study complied with institutional and national guidelines and was approved by the Danish Animal Inspectorate before initiation. The study contains retrospective MRI data from 11 pigs (female, Danish landrace, 38-42 kg, 12 weeks old). One cohort (N = 6) was exposed to a global ischemic reperfusion injury (gIRI) in the form of a total unilateral renal artery occlusion. Four of the pigs had their renal artery occluded for 2 h, and two of them for 45 min. Here, those pigs are pooled into one group. The pigs were scanned 7 days after reperfusion. 42 The other cohort (N = 5) was exposed to a partial ischemic reperfusion injury (pIRI) in the form of an occlusion of one branch of the left renal artery for 1 h. After 90 min of reperfusion, the pigs were scanned.

Animal handling
The pigs were sedated with an intramuscular Zoletile mixture (25 mg/mL tiletamine, 25 mg/mL zolazepam, 20 mg/mL zylazine, 100 mg/mL ketamine, 10 mg/mL butorphanol; 1 mL/10 kg) before arrival at the experimental facility. The pigs were canulated for venous access, intubated, and then connected to a mechanical respirator, ventilating the pig with 35%-40% oxygen for a target CO 2 of 5 kPa (± 1 kPa). The pigs were anesthetized with propofol (3 mg/kg followed by 4-8 mg/kg/h) and fentanyl (0.01 mg/kg followed by 0.035 mg/kg/h) and bladder  13 C-CO 2 is assumed to be produced in linear proportion to the other three metabolites. 13 C-CO 2 is disposed and sampled through alveolar ventilation. Peak concentration of 13 C-CO 2 in alveolar air is found to be approximately 5 min after [1-13 C]pyruvate injection. (4) Regional metabolism is traced as metabolite/pyruvate ratios.
catheterized. Arterial and venous sheaths were placed in the iliac artery and vein for pyruvate administration and monitoring. Vital parameters, expired CO 2 , and arterial gas values were monitored before and after the pig was transported to the scanner, and after the injection of hyperpolarized [1-13 C]pyruvate. The pigs were transported from the operating room to the MR scanner under manual ventilation. Mechanical ventilation and invasive blood pressure monitoring was then reapplied.

Imaging
The pigs were injected with 625 mg (0.14 mmol/kg for a 40-kg pig) of a hyperpolarized [1-13 C]pyruvate agent processed using an on-site polarizer (SpinLab; GE Healthcare) and dissolved in saline to a final volume of about 24 mL. Hyperpolarized MRI was performed on a 3T system (MR750; GE Healthcare) equipped with commercial 13 C coils (16-channel receive coil with volume transmit [RAPID Biomedical] or 20 cm transmit/receive Helmholz loop-pair [PulseTeq Limited]). Images were acquired using metabolite selective excitation followed by spiral readouts. 43 The flip angles applied were 8 • /70 • or 8 • /90 • for pyruvate/metabolites, respectively. Voxel sizes were 6 × 6 × 20 mm 3 or 8 × 8 × 20 mm 3 . In the global ischemic reperfusion injury gIRI cohort, the time resolution was 2 s, whereas it was 1 s for pyruvate and 3 s for metabolites in the pIRI cohort. Five minutes after injection, the second breath sample was extracted. After examination, the pigs were euthanized with a pentobarbital overdose.

CO 2 breath test
A baseline breath sample was taken 5 min before pyruvate injection. Breath was sampled by attaching a 60-mL syringe connected with a breath collection bag to the endotracheal tube. The breath collection bags were placed on a HeliFan spectrometer (IRMS; Helifan Plus; Fischer ANalysen Instrumente, Leipzig, Germany) for analysis right after breath sampling, according to the manufacturer's instructions. The amount of 13 CO 2 / 12 CO 2 in the expired air is represented by the dimensionless Δ value, given (in , where R = 13 CO 2 / 12 CO 2 , and where the natural-abundance reference value R reference is defined by the manufacturer and provided by experimental calibration standards. From this equation and two breath samples, the software calculates the difference between the two unitless Δ values after a given time interval following injection and the Δ value of the baseline. This is referred to as the 13 C-Delta over baseline: 13 C-DOB = Δ after − Δ baseline . The initial increase in Δ after the injection of the hyperpolarized 13 C-pyruvate represents the whole-body uptake and oxidative metabolic rate of the injected 13 C-pyruvate that is exhaled as CO 2 , 39,40 whereas the second phase of the curve is dominated by elimination (Figure 1.3).
It was established in a pilot study (Figure 1.3) that a baseline sample before injection of the [1-13 C]pyruvate and 5 min after the injection is sufficient to capture the upstroke of Δ-curve in the expired air following 13 C-pyruvate injection. Hill et al. 28 showed that a simple unitless ratio of the integral over time of a given metabolite and pyruvate (the so-called area under the curve ratio) correlates with the mono-exponential production rate constant of that particular metabolic reaction (k px ), which is itself widely reported to correlate with both underlying enzyme activity and other kinetic model fits. [44][45][46][47] A simple unitless expression can therefore be written to convey the fraction of metabolic activity for a given metabolite compared with the total overall activity (accounted for by the later expired 13 CO 2 ) as the product, as follows: Therefore, a simple expression for the apparent exhaled 13 CO 2 normalized rate constant (k px ) with units (mmol/min) can easily be constructed by multiplying the absolute pyruvate amount injected (in mmol) divided by the time interval from injection of the hyperpolarized 13 C-pyruvate to the final gas sampling (Δt = 5 min) as follows:k For simplicity, the constant pyruvate concentration and the 5-min time span is omitted in the following analysis, as it was constant for all subjects scanned.

Data analyses and statistics
The metabolite-to-pyruvate ratios were computed as the ratios of the integrals under the temporal signal curves of the individual metabolites. Regions of interest in the kidneys were identified and drawn manually. Data were analyzed and visualized in PRISM GraphPad 9.1.3 or R.
The Delta over baseline (DOB) 13 CO 2 levels in breath were analyzed using a nonparametric Mann-Whitney test, and metabolite/pyruvate ratios were compared between the groups using unpaired parametric t-tests. The coefficient of variance describes differences in variance between two or more sets of observations. Mean signal ratios and coefficients of variance were calculated for the three signal ratios in both cohorts, both for the ischemic reperfusion (IR) kidneys and for the contralateral (CL) kidneys. Coefficients of variance were compared using the R package "cvequality," implementing the approach of Feltz and Miller. 48 Normality was assessed using QQ-plots and Shapiro-Wilk's test, and a p-value of <0.05 was considered statistically significant.

Differences between cohorts
We found a significant difference in 13 C-DOB between the two cohorts (p < 0.001) (Figure 2A). The mean lactate/pyruvate ratio was 0.527 in the gIRI pigs and 0.300 in the pIRI pigs (difference = 0.228; SEM = ±0.056; p = 0.0006). The mean alanine/pyruvate ratio was 0.27 in the gIRI pigs and 0.31 in the pIRI pigs (difference = −0.033; SEM = ±0.035; p = 0.1371). The lactate and alanine production was associated with the 13 C enrichment ( 13 C-DOB) of the expired air in the pIRI group (p = 0.028) and gIRI group (p = 0.025), respectively, but not for both groups nor overall (Figure 2A). Arterial lactate, pCO 2 , and pO 2 were monitored at baseline and shortly before initiation of reperfusion and compared between cohorts. The pO 2 was significantly higher in the pIRI group than the gIRI group both at baseline (pIRI vs. gIRI: 41.6 vs. 20.32, p < 0.0001) and during ischemia (pIRI vs. gIRI: 43.2 vs. 18.93, p < 0.0001). The pCO 2 was significantly higher in the gIRI group at baseline (pIRI vs. gIRI: 4.99 vs. 5.51, p = 0.028), but not during ischemia (pIRI vs. gIRI: 4.97 vs. 5.04, p = 0.647). Arterial lactate did not differ significantly between the two groups at baseline (0.980 vs. 0.967, p = 0.941) nor during regional ischemia (0.800 vs. 1.133, p = 0.142).
Metabolite ratios before and after normalization using the expired 13 CO 2 were calculated for both the gIRI and pIRI groups for both the IR and CL side ( Figure 2B,C).

F I G U R E 2
(A) Bottom: Difference in 13 C-Delta over baseline (DOB) in the two cohorts, highlighting distinct whole-body metabolism of pyruvate between them. The cohorts were compared with an unpaired parametric t-test (p < 0.0001). Middle: Lactate/pyruvate ratios of the subjects as a function of 13 C-DOB. Top: Alanine/pyruvate ratios of the subjects as a function of 13 C-DOB. (B) Three-way analysis of variance (ANOVA) comparing raw and normalized-to-13 C-DOB (DOBratio) alanine/pyruvate signals in the ischemic reperfusion (IR) and contralateral (CL) sides in the global ischemic reperfusion injury (gIRI) and partial ischemic reperfusion injury (pIRI) groups, respectively. (C) Three-way ANOVA comparing raw and DOBratio lactate/pyruvate signals in the IR and CL sides in the gIRI and pIRI groups, respectively.
Within the lactate/pyruvate ratio on the IR side, the difference in mean values between the gIRI and the pIRI groups increased from 40% before normalization to 70% after normalization. On the CL side, the mean difference increased from 42% to 70%. Within the alanine/pyruvate ratio on the IR side, the difference increased on the IR side from −12% to 42%. On the CL side, differences increased from 12% to 55%.

Variance within cohorts
Coefficients of variance were calculated in the two cohorts for the lactate/pyruvate and alanine/pyruvate ratios before and after normalization to exhaled 13 CO 2 . Interestingly, bicarbonate was only observable in the gIRI group. Coefficients of variance of raw and normalized signal ratios in both models were pooled and compared (Table 1). In the alanine/pyruvate and bicarbonate/pyruvate ratios, four of four and two of two kidneys had a reduction in coefficients of variance, respectively, whereas only two of four were reduced in the lactate signals. In the contralateral kidneys, the coefficient of variance of alanine/pyruvate decreased from 29% to 16% across the two cohorts (p = 0.07). The contralateral lactate/pyruvate, ipsilateral alanine/pyruvate, and ipsilateral lactate/pyruvate coefficients of variance were unaltered (p > 0.5 for all).

DISCUSSION
There are two main findings in this study. First, normalizing the hyperpolarized signals to exhaled 13 CO 2 reduced the coefficients of variance of the alanine/pyruvate ratio, whereas the lactate/pyruvate ratio was largely unaltered. This finding is particularly important, as it might indicate that the lactate exchange reaction observed in the kidney is dominated by the interventions, and second that the extent of the damage is inversely correlated with the overall metabolic capacity. The other significant finding was that this normalization augmented the differences between the large, global injury and the small, local injury, showing its potential to remove unwanted whole-body contributions to hyperpolarized MRI. We found a significant difference in 13 C-DOB in the gIRI and pIRI groups ( Figure 1A), subsequently leading to big differences in normalization factors between the two groups. Ischemic injuries as the ones used in this study lead to an increase in regional lactate production, altering the local lactate/pyruvate signal ratio without significantly altering whole-body metabolism. Hence, big differences in metabolism in cohorts cause a problem when examining organs with different regional conditions, which potentially leads to a too-heavy normalization of the lactate/pyruvate ratio, as these might only be vaguely associated. Hence, our data indicate that the relatively larger lactate production in the gIRI group compared with the pIRI group is due to the difference in time and intervention severity, whereas the oxidative metabolic state is more likely originating from the differences in overall handling of the pigs at the time of examination, and as such represents a true clinical situation, with patients with vastly different overall metabolic profiles.
The underlying cause of the difference in the overall metabolic profile between the two cohorts might be explained by other factors than the renal injury, such as animal handling, anesthesia level, respirator settings, preparation time spent in the operating room, and overall time in anesthesia and/or the time from the injury to the scan (i.e., recovery time in the gIRI group). This is seen in the arterial gas analyses showing a higher p-CO 2 in the gIRI group, while exhaled 13 CO 2 was bigger in the pIRI group, as this discrepancy is likely due to differences in respirator settings. Differences in respirator flow oxygen content likewise appear in the difference in p-O 2 between cohorts. These variations are why normalizations to metabolic state can be needed, even in animal studies, when comparing different cohorts.
It is important to note that because the study only consists of healthy animals with an operative intervention, it is not possible to tell how much of the differences found is caused by physiological variance between cohorts and how much is caused by the actual difference in intervention types. Furthermore, the regional or organ-specific connection to the whole-body metabolism is likely complex and dependent on factors such as health status; therefore, this calls for further studies using this new approach. Conditions such as diabetes mellitus, hypothyroidism or hyperthyroidism, and simple fever are examples that affect whole-body metabolism; such conditions may be normalized to reduce variance in a cohort. Of course, these normalizations should be performed carefully, and one might risk masking a true metabolic difference. Another limitation of this study is that the intervention severity varies within the gIRI group in occlusion time. Both interventions, however, are still considered more severe than those of the pIRI pigs.
Existing literature on the breath-sampling method regards it as a diagnostic tool and stresses the need for fasted state in subjects, as variation in feeding states, and hence overall metabolism, leads to reduced test sensitivity. [31][32][33][39][40][41]49,50 In this study we use the method to normalize a metabolic signal, which is partially influenced by feeding state. Hence, fasted state is intuitively unnecessary, as metabolic variance caused by different feeding states is one of the factors for which we normalize. Currently there is no consensus of whether subjects must be fasting before being subjected to hyperpolarized MRI. 10,51 Hence, this method could contribute to deeming fasted subject state unnecessary in hyperpolarized MRI.
In this study, two factors influence the raw signal: whole-body metabolism and regional conditions. It is impossible to assess how much each factor contributes to the signal. Future studies should aim to examine cohorts with healthy kidneys and big variance in whole-body metabolism, as this accentuates the effect of the normalization. A possible method of inducing metabolic variance is intravenously administered metabolic stressors. 52,53 The proposed method is easily incorporated into hyperpolarized MRI workflow; thus, further studies in a clinically relevant situation is both feasible and necessary to determine whether this method has a role in clinical use. As breath-test analysis is an already established clinical routine analysis, this supports the use of the combination in hyperpolarized examinations and reduces the barrier for adoption into the clinical workflow.
It is very likely that the method would prove useful in situations where the literature normal control data pool is low, such as in children who are known to be metabolically different from adults, thus making direct comparisons difficult. The ability to quantitatively compare data acquired under very different situations would also have huge research implications, as this would open the door to interstudy cohort comparison and meta-analysis studies in hyperpolarized MRI.