Imaging brain deoxyglucose uptake and metabolism by glucoCEST MRI

Abstract

2-Deoxy-D-glucose (2DG) is a known surrogate molecule that is useful for inferring glucose uptake and metabolism. Although ¹³C-labeled 2DG can be detected by nuclear magnetic resonance (NMR), its low sensitivity for detection prohibits imaging to be performed. Using chemical exchange saturation transfer (CEST) as a signal-amplification mechanism, 2DG and the phosphorylated 2DG-6-phosphate (2DG6P) can be indirectly detected in ¹H magnetic resonance imaging (MRI). We showed that the CEST signal changed with 2DG concentration, and was reduced by suppressing cerebral metabolism with increased general anesthetic. The signal changes were not affected by cerebral or plasma pH, and were not correlated with altered cerebral blood flow as demonstrated by hypercapnia; neither were they related to the extracellular glucose amounts as compared with injection of D- and L-glucose. In vivo ³¹P NMR revealed similar changes in 2DG6P concentration, suggesting that the CEST signal reflected the rate of glucose assimilation. This method provides a new way to use widely available MRI techniques to image deoxyglucose/glucose uptake and metabolism in vivo without the need for isotopic labeling of the molecules.

Introduction

The rate of glucose uptake and its conversion into subsequent metabolites are important biomarkers of cellular function. Since the seminal work of Sokoloff et al,¹ isotopically labeled 2-deoxy-D-glucose (2DG) has been established as a way to measure glucose metabolism; this is based on the observation that 2DG enters cells by the same transporters as glucose (e.g., mostly GLUT-1 and GLUT-3 in the brain). It is phosphorylated by hexokinase into 2DG-6-phosphate (2DG6P) but only minimally metabolized further via glucose-6-phosphate dehydrogenase in the oxidative pentose phosphate pathway, and glucose-6-phosphate isomerase in the glycolytic pathway, because of the lack of a hydroxyl group on carbon atom 2 (C2). As the amount of glucose-6-phosphatase that catalyzes the hydrolysis of 2DG6P to 2DG is low in mammalian brain, and having low membrane permeability 2DG6P becomes trapped in brain cells for many hours.² By quantifying the amount of 2DG and 2DG6P, the glucose metabolic rate can be estimated via compartmental modeling.³

To enable the detection of 2DG and 2DG6P, a variety of methods using isotopically labeled 2DG have been developed over several years. ¹⁴C-2DG was the first to be applied to ex vivo imaging of animal brain with high-resolution autoradiography.^{1 18}F-labeled fluorodeoxyglucose (FDG) has been developed to enable in vivo imaging of glucose metabolism by positron emission tomography;⁴ thus it has been widely used to assess cellular function and metabolic activity, especially for the diagnosis and evaluation of cancer and neurodegenerative diseases, such as Alzheimer disease.^{5, 6} However, the positron range limits the spatial resolution of images to ∼1 mm and the radioactivity mitigates repeated use, and therefore longitudinal studies. Nuclear magnetic resonance (NMR) has also been used to detect 2DG, 2DG6P, and other sugars. The resonances of ¹³C-labeled 2DG and its metabolite, 2DG6P, are well resolved in ¹³C NMR spectra and thus enable estimation of the metabolic rate of glucose in vivo.^{7, 8} These earlier studies showed that high doses (up to 0.5 g/kg) of 2DG are well tolerated by animals and reveal reaction kinetics similar to those of ¹⁴C-2DG experiments. The phosphorylation of 2DG makes it detectable by ³¹P NMR, thereby indirectly assessing glucose metabolism.⁹ However, even at such high doses of 2DG, the sensitivity of in vivo NMR is still insufficient to allow spatial mapping of glucose metabolism.

Chemical exchange saturation transfer (CEST) is an approach in NMR spectroscopy that may be used to amplify hexose detectability.¹⁰ By exchanging proton magnetization between hydroxyl groups and water, low-concentration metabolites can be detected via abundant water protons using ¹H magnetic resonance imaging (MRI). A study of glycogen metabolism showed that the hydroxyl groups on glucose and glucose residues in glycogen can be detected by CEST in perfused liver;¹¹ and in vitro glucose release from starch can be measured similarly.⁹ As glucose, 2DG, and 2DG6P have common core structures, 2DG and 2DG6P might be detected in the same way. In the present work, we show for the first time the results of in vivo imaging of glucose, 2DG, and 2DG6P in rat brains using CEST MRI.

Materials and Methods

In Vitro Studies

2DG and 2DG6P (Sigma, Singapore, Singapore) were prepared at different concentrations in phosphate-buffered saline (PBS) containing 10% D₂O for NMR field/frequency locking. Experiments were conducted on an NMR spectrometer (Bruker Biospin, Karlsruhe, Germany) with an Avance III console running at 400.13 MHz for ¹H, with the sample temperature set to 37°C. Chemical exchange saturation transfer z-spectra were acquired using the same presaturation settings as for the in vivo experiments (5 seconds relaxation delay and 5 seconds continuous presaturation of 1.5 μT amplitude) with offset range varying from + to −1,600 Hz decremented in 40 Hz steps. The magnetization reference was measured with an offset frequency of +6,000 Hz.

For determination of the effect of cellular internalization on the CEST signal, human red blood cellswere obtained by venipuncture from the cubital fossa of a healthy donor. The cells were centrifugally washed three times (10 minutes, 3,000 × g, 4 °C) in isotonic saline consisting of 154 mmol/L NaCl, 290 mOsmol/kg. The buffy coat was removed by vacuum pump aspiration. Cells were bubbled for 15 minutes with CO to convert the hemoglobin into a stable diamagnetic state. The cells were incubated with or without 50 mmol/L 2DG. Two further washes were performed with PBS buffer. The hematocrit was measured and adjusted to 0.35 by suspending the cells in PBS buffer. For the determination of the effect of the natural proteinaceous fluid (blood plasma) on the CEST signal, 8 mL of blood was centrifuged (10 minutes, 3,000 × g, 4 °C), and the plasma was then removed from the top of the cell layer leaving it in the intercellular spaces. Ten percent v/v of PBS constituted in D₂O was added to the sample for NMR field/frequency locking.

In Vivo Experiments

Animal experimental protocols were approved by the Institutional Animal Care and Use Committees of the Biomedical Sciences Institutes (Agency for Science Technology and Research, Singapore). Male Wistar rats (320 to 400 g) were fasted for 24 hours before each experiment. The rats were initially anesthetized with isoflurane (5% for induction, 3% during preparation) in a mixture of air and O₂ gases (ratio 5/3). The animals were orally intubated and mechanically ventilated (TOPO, Kent Scientific, Torrington, CT, USA) with a respiration rate of 55 b.p.m., inspiration pressure ∼9.1±0.1 mm H₂O, and inspiration ratio of 35%. A tail vein and the tail artery were cannulated for injection of 2DG or glucose and for monitoring the blood-gas concentrations, respectively. Then the animals were secured on an MRI-compatible cradle with ear bars and a bite bar to prevent head motion. The end-tidal CO₂ was monitored (SurgiVet V90043, Smiths Medical, Dublin, OH, USA) and the rectal temperature was maintained at 37°C by a feedback-controlled air heater (SAII, Stony Brook, NY, USA). Pancuronium bromide (1.5 mg/kg) was injected at the beginning of the time course to prevent any movement of the animal. The appropriate isoflurane level (1.0%, 1.5%, or 2.0%) and hypercapnia were adjusted before starting data acquisition. The hypercapnia study was conducted by adding 1.8% of CO₂ in the total gas supply to the animal.

Except for the studies in which various 2DG doses were used, 0.5 g/kg 2DG dissolved in saline was injected as a bolus after at least three baseline imaging scans. In the case of glucose injection, either a bolus of 1 g/kg of D- or L-glucose (Sigma) was injected, or a bolus of 1 g/kg D-glucose, followed by continuous infusion of varying rates of D-glucose, was given so that the blood glucose concentration was maintained at ∼400 mg/dL. Blood gases and glucose were measured by using a blood gas analyzer (ISTAT, Abbott, Abbott Park, IL, USA).

In vivo MRI measurements were conducted on a 9.4T magnet (Agilent Technologies, Santa Clara, CA, USA). A volume coil (Rapid Biomedical GmbH, Rimpar, Germany) was used for RF transmission, and a custom-designed surface coil of 1.5 cm diameter was positioned on the top of the animal's head and served as the receiver. Field homogeneity of <10 Hz deviation was obtained with volume-selective second-order shimming. A train of Gaussian pulses of 1.5 μT amplitude and 58 milliseconds duration was applied for saturating proton-spin populations at different frequency offsets. For glucoCEST, a saturation duration of 4.8 seconds, 33 offset frequencies within ±4 p.p.m., and a magnetization reference at +6,000 Hz were used. To correct for frequency offset due to B₀ inhomogeneity and variation during the scanning, CEST imaging was interleaved with WASSR imaging.¹² For WASSR, 0.1 μT pulse amplitude and 33 offset frequencies within ±1 p.p.m. were used. The images from both scans were acquired by single-shot spin-echo echo-planar imaging covering a slice that crossed the somatosensory area of the brain with a thickness of 2 mm, a field of view of 32 × 32 mm², an imaging matrix of 64 × 32, repetition time=10 seconds, and echo time=22 milliseconds. The scan times were 1.2 and 10.2 minutes for the WASSR and CEST scans, respectively, making a temporal resolution of ∼11.5 minutes. Cerebral blood flow (CBF) was measured using an optimized flow-sensitive alternated inversion recovery (FAIR) arterial spin labeling with repetition time=20 seconds, echo time=21 milliseconds, averaging=4, and multiple values of TI: viz., 50, 200, 500, 800, 1,400, 1,700, 2,000, 3,000, 4,000, and 5,000 milliseconds.¹³

In vivo ³¹P MRS studies were conducted on the same MR scanner as described above, using a ¹H/³¹P double-tuned surface coil (Rapid Biomedical) for transmit and receive. Each spectrum was acquired via a hard 90° pulse and a spectral width of 10 kHz. A recycle time of 20 seconds with 32 transients was used, giving an acquisition time of 10.5 minutes that was comparable to the CEST temporal resolution. No spatial localization was used for these experiments, as it had been shown that the contribution to the spectra from metabolites outside the brain was negligible.¹⁴

Data Analysis

CEST imaging data were processed pixel by pixel using custom written software in Matlab (Mathworks, Natick, MA, USA). After correcting the data for B₀ offset using the WASSR experiment, the z spectrum was interpolated using a cubic spline. The MTRasym was estimated from the expression:

where S₀ was the signal at an offset of 6,000 Hz. The glucoCEST signal was calculated as the integral of the MTRasym within 1.00±0.25 p.p.m. Each signal time course was obtained by averaging the glucoCEST signal in the brain and subtracting from it the mean of the three baseline scans.

The CBF, f, was estimated from the expression:

where T₁ and T_1app were obtained by nonlinear least squares fitting of the expression to the non-selective and slice-selective inversion recovery data, respectively,¹⁵ and λ is the blood–tissue–water partition coefficient assumed to be 0.9 mL/g.

³¹P MRS were processed in TopSpin 3.0 (Bruker). An exponential line broadening factor of 20 Hz was applied before Fourier transformation of the data. The resonance at ∼7.2 p.p.m. was assigned to 2DG6P. As the line widths of the phosphorus resonances did not change during the time course, the amount of metabolite was estimated from the calibrated signal amplitude. The amount of 2DG6P was estimated by normalizing to the amplitude of the γP-adenosine triphosphate resonance, although absolute quantification of concentration was not done.

Results

In Vitro Studies

To explore the extent to which the CEST effect was dependent on 2DG concentration, z-spectra (see e.g., Chapman et al¹⁶) were recorded from samples with 2DG and 2DG6P concentrations from 0 to 100 mmol/L, at pH 7.4 and 37°C. Figure 1 shows the high-resolution ¹H NMR z-spectra of 2DG and 2DG6P. The magnetization transfer asymmetry ratios (MTRasym) are given in the insets of the figures. For both molecules, the exchange of the hydroxyl protons with water was detected by selective RF irradiation between 0.5 and 1 p.p.m. with a maximum asymmetry that was concentration dependent. The MTRasym at 1 p.p.m. increased linearly with the concentration of 2DG and 2DG6P below 80 and 100 mmol/L, respectively, and converged on a ‘saturation type' of asymptote (Figure 1C). We also noted that the exchange effect was higher for 2DG6P than 2DG, for which MTRasym was typically ∼20% higher.

Figure 1.

Chemical exchange saturation transfer z-spectra of: (A) 2-deoxy-D-glucose (2DG); and (B) 2DG-6-phosphate (2DG6P) with concentrations (mmol/L) of 100 (red), 75 (blue), 50 (purple), 25 (green), 12.5 (gray), and 6.25 (black) in phosphate-buffered saline buffer (pH=7.4) measured at 37°C on the 400.13 MHz nuclear magnetic resonance spectrometer. The insets are the corresponding MTRasym plots. (C) The MTRasym of 2DG and 2DG6P at 1 p.p.m. increased linearly with concentration until ∼75 mmol/L and 100 mmol/L, respectively.

In Vivo Studies

The feasibility of in vivo applications was demonstrated by dynamic CEST imaging of rat brains after intravenous injections of 2DG or glucose. Physiologic parameters were adjusted and maintained within each normal range before beginning each experiment with: pO₂=191.1±28.6 mm Hg (mean±standard deviation (STD); n=36); pCO₂=31.8±3.5 mm Hg; and pH=7.36±0.04. The baseline blood glucose concentration was 82±13 mg/dL (5.2±0.75 mmol/L) after 24 hours fasting.

The glucoCEST signal was calculated from the integral of MTRasym at 1.00±0.25 p.p.m. Figure 2A shows maps of relative changes in glucoCEST signal in rat brains before and after bolus injections of 0.5 and 1.0 g/kg 2DG, respectively. Significant uptake of 2DG was seen in the cortical and thalamic regions after the injection of both doses of 2DG. The signal was sustained for ∼40 minutes and then slowly declined over more than 1 hour. To ascertain if the observed signal was related to glucose metabolism, 2DG6P was measured by in vivo ³¹P MRS under the same conditions. After injection of 2DG, a new resonance appeared in the ‘sugar–phosphate' chemical shift region (∼7.2 p.p.m.) (Figure 2B). As the amount of γP-adenosine triphosphate is normally maintained constant in the brain under physiologic conditions,¹⁴ the amplitude of the 2DG6P peak was normalized to the amplitude of γP-adenosine triphosphate as an index of the amount of 2DG6P present.

Figure 2.

(A) Dynamic glucoCEST images of rat brains before and after a 1.0 (top) and 0.5 (bottom) g/kg 2-deoxy-D-glucose (2DG) injection showed high signal increases in the cortex and thalamus. The image intensity represents relative glucoCEST signal change from the baseline (color scale between ±100%). (B) Examples of ³¹P magnetic resonance spectra of a rat brain before and ∼38 minutes after injection of 1.0 g/kg 2DG. Especially note the resonance assigned to 2DG-6-phosphate (2DG6P) at 7.2 p.p.m.

Figure 3 compares the change in glucoCEST signal with 2DG6P from ³¹P MRS over the brain after injection of 0.5 and 1.0 g/kg 2DG. 2-Deoxy-D-glucose uptake was readily detected by glucoCEST immediately after its injection. The change in the integral of MTRasym reached a maximum of 1.33±0.19 (mean±standard deviation; n=6) after injecting 0.5 g/kg of 2DG and was sustained for ∼30 minutes and then decreased to near the baseline at ∼80 minutes. The glucoCEST signal after injecting 1.0 g/kg 2DG had a similar trend but with a maximum at 1.85±0.29 (n=6), which was ∼40% higher than with the lower dose (P<0.05; Student's t-test). Consistent with the literature, the 2DG6P resonance reached a maximum at ∼30 to 40 minutes and then slowly decreased (Figure 3B).⁸ Compared with 0.5 g/kg of 2DG, the maximum of the 2DG6P concentration under the 1.0 g/kg dose was 17.3% higher. To evaluate the signal contribution from blood glucose, this was measured in the arterial blood in the same time interval (Figure 3C); it increased to 272±24 (15.1±1.3 mmol/L; n=6) and 428±34 mg/dL (23.8±1.9 mmol/L; n=6) after 0.5 and 1.0 g/kg 2DG, respectively. After the initial surge, the glucose concentration fell and remained constant at 183±18 mg/dL (10.2±1.0 mmol/L) after 0.5 g/kg 2DG injection; whereas in the 1.0 g/kg 2DG group, the concentration gradually fell to 300±26 mg/dL (16.7±1.4 mmol/L) at the end of the time course.

Figure 3.

Time courses of: (A) glucoCEST signal; (B) 2DG-6-phosphate measured by ³¹P magnetic resonance spectra and (C) arterial blood glucose after injection of 1.0 (red) and 0.5 (blue) g/kg 2-deoxy-D-glucose. The glucoCEST signal represents the difference of MTRasym integral around 1.0 p.p.m. from the baseline signal. Error bars denote standard error of the mean.

To explore whether the glucoCEST signal reflected 2DG uptake and metabolism, time-dependent responses to bolus or continuous injection of D-glucose, which can be transported and metabolized rapidly, or L-glucose, which has limited blood–brain barrier (BBB) permeability,¹⁷ were compared (Figure 4). D-glucose uptake was readily detected by glucoCEST immediately after a bolus injection of 1 g/kg for which the signal reached a maximum of 0.77±0.19 (n=4). However, the signal fell to the baseline immediately afterwards (Figure 4B) while the blood glucose concentration fell to 304±30 mg/dL after reaching a maximum of 524±69 mg/dL (n=6) (Figure 4C), suggesting rapid metabolism and disappearance of detectable hexose. When the blood glucose concentrations were maintained at 464±46 mg/dL (n=3) throughout the experiment, the glucoCEST signal remained constant at ∼0.96±0.6 (n=6). The signal rapidly fell when the infusion was stopped. When the BBB-impermeable L-glucose was injected, no CEST signal increase was seen (Figure 4B), indicating glucoCEST signal was not due to high concentration of glucose in plasma.

Figure 4.

(A) Time series of glucoCEST images of rat brain under constant infusion of D-glucose (top), after bolus injection of 1 g/kg D-glucose (middle), and after bolus injection of 1 g/kg L-glucose (bottom). The image intensity represents relative glucoCEST signal change from the baseline. (B) Time courses of glucoCEST signal under the above three injection conditions in (A). The signal represents the difference of MTRasym integral around 1.0 p.p.m. from the baseline signal. (C) Arterial blood glucose with constant infusion (blue) and bolus injection (red) of D-glucose. Error bars denote standard error of the mean.

As 2DG uptake and metabolism are indicative of the overall energy utilization of a tissue, we investigated whether the glucoCEST signal might reflect changes in cerebral activity by modulating it with anesthesia and CO₂. Isoflurane is known to suppress cerebral glucose metabolism and neural activity, while elevating CBF via systemic acidification.^{18, 19} Figure 5A shows the glucoCEST signal under 1.0%, 1.5%, and 2.0% isoflurane. The maximum signal decreased only slightly at 1.5%, but the decrease was dramatic under 2.0% isoflurane. The signal decayed more slowly than under 1.0% isoflurane suggesting delayed 2DG6P production (Figure 5B) under the latter conditions. Figure 5C shows the kinetics of 2DG6P change under the same levels of isoflurane measured by ³¹P MRS. In a manner similar to glucoCEST, no difference in 2DG6P production was observed between 1.0% and 1.5% isoflurane, but the response was significantly delayed and reduced under 2.0% isoflurane.

Figure 5.

(A) Examples of glucoCEST images under 1.0%, 1.5%, and 2.0% isoflurane (top to bottom) anesthesia. Time courses of (B) glucoCEST signal, and (C) 2DG-6-phosphate measured by ³¹P magnetic resonance spectra after injecting 0.5 g/kg of 2-deoxy-D-glucose under 1.0% (blue), 1.5% (red), and 2.0% (green) isoflurane. Error bars denote standard error of the mean.

As isoflurane also increases CBF, which may affect 2DG delivery rate to brain cells, and therefore uptake and clearance, we evaluated the influence of CBF on glucoCEST using mild hypercapnia achieved with ∼1.8% (v/v) CO₂ added to the gas mixture. The CBF was increased from 111±12 mL (100 g/minute) using normal gas composition to 223±29 mL (100 g/minute) under hypercapnia. Figure 6A shows that the glucoCEST signal did not change under normal and hypercapnic conditions. This conclusion was further supported by the ³¹P MRS spectra recorded under the same conditions (Figure 6B); these showed that the rate of production of 2DG6P was not altered.

Figure 6.

Time courses of: (A) glucoCEST signal; and (B) 2DG-6-phosphate measured by ³¹P magnetic resonance spectra after injecting 0.5 g/kg of 2-deoxy-D-glucose (2DG) under hypercapnia (red) and normocapnia (blue). (C) The MTRasym of 50 mmol/L 2DG in phosphate-buffered saline (blue), of human erythrocytes (red blood cells (RBCs); gray), of 50 mmol/L 2DG mixed with RBCs (green), 50 mmol/L of 2DG in RBCs (red), and 50 mmol/L of 2DG in plasma (orange). Error bars denote standard error of the mean.

Influence of pH and Cellular Uptake

It is known that pH influences the rate of exchange of protons on hydroxyl groups and therefore might alter the glucoCEST signal intensity.²⁰ As both isoflurane and hypercapnia can cause acidosis,^{21, 22} we evaluated their influence on blood and cellular pH. ³¹P MRS enables the estimation of intracellular pH from the chemical shift of the inorganic phosphate (Pi) resonance or that of 2DG6P.^{23, 24} For all the experimental conditions, the brain pH estimated using the Pi resonance was constant at 7.15±0.03, and there was no change before and after injection of 2DG. However, a gradual decrease in blood pH was seen immediately after injection of 2DG; it changed from 7.37±0.05 before 2DG injection to 7.25±0.05 after injection. Similarly, for the injection of 2DG, a fall in pH from 7.41±0.05 to 7.3±0.02 was seen in the case of L-glucose injection, while all other blood-gas parameters were unaltered. Blood pH did not change with D-glucose injection.

To assess the influence of cellular internalization of 2DG on the glucoCEST signal, human erythrocytes were mixed with 2DG (Figure 6C). Human erythrocytes have a high rate of glucose transport^{22, 23} but relatively low hexokinase activity;^{24, 25} hence, the majority of any added 2DG in the mmol/L concentration range used would have been free rather than present as 2DG6P, even over several hours. With 50 mmol/L 2DG (concentration corresponding to an injection of 0.5 g/kg of 2DG in the animal) similar but largely reduced MTRasym was observed. The maximum of the MTRasym spectrum was shifted slightly from 0.5 p.p.m. toward 1.2 p.p.m. The MTRasym at 1 p.p.m. was reduced from 7.6% in PBS to 2.2% in the erythrocytes.

To evaluate the influence of the natural proteinaceous biologic fluid (blood plasma) on glucoCEST signal, plasma was extracted from rats and mixed with 2DG. The plasma mixed with 50 mmol/L 2DG had a slightly higher pH of 7.8 compared with erythrocytes and PBS. The MTRasym spectrum in plasma showed a peak at ∼0.25 p.p.m. and was much lower than that in erythrocytes at offsets>0.5 p.p.m. (Figure 6C). The MTRasym at 1 p.p.m. was ∼0.5%.

Discussion

In the present work, we detected 2DG and 2DG6P by glucoCEST MRI via the exchange of magnetization between the naturally abundant ¹H atoms on 2DG and 2DG6P and water. This method does not require any isotopically labeled molecules, neither a radioactive nor a ¹³C-enriched compound. Furthermore, the signals from 2DG and 2DG6P are amplified through the detection via the dominant water signal, thus enabling very sensitive imaging of the dynamics of 2DG uptake and metabolism. The glucoCEST signal declined with reduced metabolism under increased isoflurane concentrations but did not change with changes in CBF (in the ranges studied); this occurred in a manner similar to the 2DG6P measured by ³¹P MRS. Significantly, lower glucoCEST signals after D- and L-glucose injection, even when blood glucose concentrations remained high, suggested that the signal was not contributed by extracellular glucose. This suggested that the glucoCEST signals reflected 2DG uptake and metabolism and were not being confounded by changes in CBF or blood glucose concentrations.

The cerebral glucoCEST signal was also seen with injection of D-glucose, although the signal was much lower than with 2DG. When injected as a 1 g/kg bolus, the peak glucoCEST signal under 2DG injection was ∼2.4 times of that under D-glucose injection. Notably, the glucoCEST signal of a D-glucose bolus was not discernable from the baseline immediately after the initial surge in signal. Considering similar CEST z-spectra of 2DG, D-glucose, and L-glucose (see Supplementary Figure S1), the fast glucoCEST signal decay after D-glucose injection could have been due to: (1) faster reduction of blood glucose concentrations due to rapid insulin secretion and hence reduction of both extracellular and intracellular concentrations; (2) metabolism of D-glucose that eliminated the metabolites detectable by glucoCEST, as opposed to 2DG, which was trapped in the cells; and (3) the sensitivity of glucoCEST was insufficient to distinguish glucose and its metabolites at these concentrations. These three possibilities are discussed next.

In general, the glucoCEST signal arises from 2DG/D-glucose and their metabolites that are present in any of the three compartments, viz., intravascular, extravascular–extracellular, and intracellular. Typical blood volume and extracellular spaces are<5%²⁵ and 10% to 15%²⁶ of rat brain tissue volume, respectively. Therefore, the plasma and extracellular contribution should be small unless the concentrations were extremely high. As the glucose transporter would not be saturated at the concentrations used,²⁷ the glucose concentration in the extracellular compartments should reflect the blood glucose value and its temporal change.²⁸ To clarify whether the glucoCEST signal was due to the extracellular glucose, three different experiments were carried out. First, blood glucose concentration was clamped above the peak level achieved with 1 g/kg 2DG by continuous infusion of D-glucose (Figure 4). The results showed that even with the huge concentrations of extracellular sugar, the glucoCEST signal of D-glucose was only 50% of that after 2DG injection. Therefore, we concluded that the detected high signal achieved with 2DG injection largely reflected the amounts of 2DG and 2DG6P that had accumulated in the cells, rather than in the extracellular space.

Second, the contribution to the CEST signal from the intravascular compartment was further clarified by injecting L-glucose that has very low, if not negligible, BBB permeability.¹⁷ Considering the difficulty in obtaining a CEST signal from flowing blood, as predicted, no signal enhancement was seen (Figure 4B). Together with the fact that the temporal changes of the glucoCEST signal under both 2DG and D-glucose injection did not correlate with blood glucose concentration, the intravascular contribution to the overall glucoCEST signal was concluded to be minimal.

Third and finally, the contribution of extracellular and intracellular glucose to the glucoCEST signal depends on their volume fraction, and their sensitivity to the glucoCEST phenomenon. As glucose transport across the BBB was not saturated under our conditions and GLUT-1 and GLUT-3 can be upregulated under hyperglycaemic conditions,²⁹ the transport of 2DG is facilitated so it should not have accumulated in the extracellular compartment. Therefore, the extracellular concentration might only be prominent at the initial time points where overwhelming influx of 2DG is present. Based on a ¹H MRS study of anesthetized rat brain, the extracellular–intracellular volume faction of glucose is 19%, and this value does not change in hyperglycemia.³⁰ The contribution of extracellular glucose would be ∼20% if the sensitivity of the glucoCEST experiment is similar for each compartment. The influence of cellular internalization of 2DG and its presence in the extracellular fluid on the glucoCEST signal were further evaluated by measuring MTRasym of 2DG in erythrocytes and plasma, respectively. Our results showed a largely attenuated glucoCEST signal at 1 p.p.m. in both samples with higher signal in erythrocytes than in plasma (Figure 6C). Therefore, the contribution to the overall glucoCEST signal favours the intracellular compartment; hence, the glucoCEST signal would be surmised to be dominated by the signal from sugars in the intracellular compartment. As the milieux of erythrocytes and plasma are obviously different from those in brain cells and their extracellular space, further study of this type of experiments will provide useful insight. Recently, Chan et al³¹ reported that most of the glucoCEST signals seen in tumors after injecting D-glucose to be of extracellular origin, while another recent study by Samuel et al³² provided evidence from positron emission tomography and ¹³C NMR suggesting an intracellular origin of glucoCEST using D-glucose. Therefore, the compartmental origin of the glucoCEST signal is still a subject of future research.

In addition, by doubling the injected dose of 2DG, a limited increase of glucoCEST and 2DG6P signals, but similar temporal decay, was seen (Figure 3). The decayed glucoCEST signal, despite an elevated blood glucose concentration, suggested a decreased metabolic rate of 2DG, as the ‘flux determining' step for 2DG metabolism is hexokinase rather than the glucose transporters.²⁷ The high circulating glucose concentrations in the blood achieved after a single bolus of 2DG indicated that 2DG uptake and metabolism in the body was slowed down;³³ and this is consistent with the gradual fall of the glucoCEST signal. However, D-glucose can be transported and metabolized rapidly after injecting a high-concentration bolus,³⁴ and so this would cause faster elimination of metabolites, as detected by the glucoCEST method.

By manipulating the cerebral metabolic rate with isoflurane levels, the glucoCEST signal showed a plateau under 2% isoflurane that was similar to the relative 2DG6P concentration estimated by ³¹P MRS, as compared with a decrease under 1% isoflurane (Figure 5). Such similar responses suggested that the glucoCEST signal had a contribution from the 2DG6P present. Although both glucoCEST and ³¹P MRS showed similar trends under isoflurane and hypercapnia, a discordance between the time courses of glucoCEST and the 2DG6P (measured by ³¹P MRS) was seen. While the glucoCEST signal decreased to the baseline ∼80 minutes after 2DG injection, the 2DG6P concentration remained high, consistent with the literature.⁹ This discrepancy most probably arises from the different sources of the signal recorded in each method: the glucoCEST signal is a superposition of signals from 2DG, 2DG6P, and its metabolites, so the signal does not directly reflect the 2DG6P concentration but rather is more sensitive to the total amount of 2DG and 2DG6P in a given region of tissue. In relation to both ¹⁴C-2DG and ¹³C-2DG studies, 2DG in the brain decays exponentially after a bolus injection, while 2DG6P concentration peaks around 15 to 30 minutes then decays gradually.^{8, 35} This is consistent with our observation that the glucoCEST signal remained at a high level in the first 30 minutes before the 2DG6P level began to decrease (Figure 3A). The more rapid fall in the glucoCEST signal compared with the 2DG6P signal in ³¹P MRS spectra can be ascribed to several factors, including the sensitivity to detection, anesthesia and further metabolism of 2DG6P. First, the more rapid decrease to baseline compared with that reported in the ¹³C MRS literature⁸ may be due to restricted in vivo sensitivity of glucoCEST in our experiments; specifically, the glucoCEST signal arises from exchange of protons on the 2DG and 2DG6P molecules with that on water, and this rate is sensitive to the molecular environment including pH, temperature, and the presence of other solutes. Unlike the in vitro conditions, the presence of other solutes in vivo will alter the –OH exchange properties and hence the CEST signal. To better determine the amount of 2DG6P, we also conducted high-resolution ¹³C spectroscopy in brain extracts at several time points after 2DG injection. In the preliminary experiment (data not shown), we found a very high amount of ¹³C-2DG6P at 50 minutes, which decreased to near the level of 5 minutes at 80 minutes after 2DG injection, which suggested that the 2DG6P may decrease faster than that measured by 31P MRS. We also observed a new peak at ∼184.2 p.p.m. at the 80 minutes time point, which may represent the additional metabolites generated.

By taking advantage of the unique properties of red blood cells, we showed that CEST signal was extensively attenuated when 2DG was taken up by red blood cells in suspension (Figure 6C); this attenuation is partly due to the reduced longitudinal relaxation time.³⁶ Human red blood cells have a high rate of glucose transport but relatively low hexokinase activity and hence the majority of any added 2DG in the mmol/L concentration range would be present inside the cell suspension rather than free outside if the packing density (hematocrit) is high.

Another important difference is that our studies were conducted with anesthetized animals while the majority of ¹⁴C-, ¹³C-2DG, or FDG studies have been done with awake animals. It is well documented that anesthesia suppresses cerebral glucose metabolism, and this was evident in our experiments under different levels of isoflurane. One major impact of anesthesia is the reduced production of 2DG6P. As shown by McDougal et al using ¹⁴C-2DG,³⁷ the proportion of 2DG6P in the total amount of 2DG+2DG6P was 89% at 48 minutes after injection in the awake mouse brain. However, in anesthetized mouse, the proportion was reduced to 57% at the same time point. Therefore, the total amount of 2DG and 2DG6P will be lower and more signals may come from 2DG instead of 2DG6P in anesthetized animals, and hence the observed glucoCEST signal would be lower than, although not dominated by, that of 2DG6P.

Furthermore, the loss of glucoCEST signal at later time points could be because of further metabolism of 2DG6P so that –OH exchange was reduced while the metabolites were still detectable in the ³¹P spectra. Indeed, unlike the usual misunderstanding, 2DG6P does not accumulate but could be metabolized through the pentose phosphate and other pathways. It is found that 2DG6P is metabolized to 2DG-1-phosphate and 2DG-1,6-phosphate, which can account for 20% of the metabolites in the brain.³⁸ As metabolism of 2DG6P at tracer dose is different from that at the dosages used in this study, additional metabolic products from 2DG and 2DG6P could be indicated in studies of ¹⁹FDG. It is found that 2-fluoro-2-deoxy-D-mannose-phosphate is another major metabolite generated in the rat brain, and 2-fluoro-2-deoxy-D-6-phosphoglucon is also detected when high dose of ¹⁹FDG (∼0.3 g/kg) was injected.³⁹ In another study, FDG-6-phosphate comprised less than 50% of the total ¹⁹F-substrates at 90 minutes after injection of 0.15 g/kg of ¹⁹FDG;⁴⁰ this suggests that 2DG6P does not dominate the CEST signal, especially at later time points. Besides, a small proportion of 2DG will be incorporated into glycogen as has been reported in studies using either ¹⁴C-2DG⁴¹ or ¹⁹FDG.⁴⁰ Therefore, the other sugar phosphates may have similar chemical shifts as 2DG6P, leading to the possibility of overestimating 2DG6P from ³¹P NMR, but contributing less CEST effects because of a change in the –OH group of these metabolites of 2DG6P. High-resolution ¹³C NMR of brain extract would be a method to determine the amount of various 2DG metabolites under the similar experimental condition. Together with in vitro CEST z-spectra measure of all metabolites, we would be able to clarify their contributions to the detected glucoCEST signal. Because of the extensive work needed on preparing each metabolite and characterizing their chemical shifts, this is beyond the scope of this paper.

A major challenge in refining the glucoCEST method will be its quantification. Conventionally, the rate of glucose metabolism can be estimated by using compartmental modeling of the tracer kinetics. This approach requires estimates of the arterial 2DG concentration and at least the total concentration of 2DG and 2DG6P in the relevant tissues. The MTRasym has an almost linear relationship with concentrations within the physiologic range (Figure 1). However, its different sensitivity to 2DG and 2DG6P, as well as to the source of the signal from intracellular and extracellular compartments, makes estimation of the overall concentration imprecise. Spectral information might have been used to estimate the relative proportions of 2DG and 2DG6P. However, the MTRasym spectra of 2DG and 2DG6P are very similar (Figure 1), thus restricting the ability to resolve the two spectra by line-shape fitting or spectral deconvolution.

In the present study, 0.5 g/kg of 2DG was used to achieve good sensitivity for detection. We calculated the pixel-wise signal-to-noise ratio (SNR) of glucoCEST imaging as the change of MTRasym at the first post injection time point, divided by the standard deviation of three baseline time points in each pixel. Thus, the mean SNR from the cortical area of the brain at two 2DG dosages was: 23.0±7.0 (mean±mean±standard deviation, n=6) at 1.0 g/kg 2DG; and 14.9±2.3 at 0.5 g/kg 2DG. With such SNR, there is potential to achieve higher-resolution images than the currently used 1 μL voxel volume. Compared with the maximum in-plane resolution of ∼1.2 mm with a slice thickness larger than that on current micro-positron emission tomography scanner, the glucoCEST MRI can provide comparable or better resolution. The high doses of 2DG used herein may cause changes in whole animal physiology, as 2DG competes with glucose for tissue entry throughout the body. 2-Deoxy-D-glucose dosages of 0.5 g/kg have been shown to be well tolerated in rats and they had no permanent effect on behavior.⁴² In humans, 2DG of up to 0.3 g/kg have been used without adverse effects.⁴³ Nevertheless, further reduction of the 2DG dose is desirable. With linear extrapolation from our experimental data, a dose as low as 0.25 g/kg may still be readily detectable. Dosage may also be reduced by increasing the sensitivity to detection; hence by using a higher magnetic field strength, the sensitivity would be enhanced because the spectral effects of proton exchange are increased when the system enters the ‘slow-exchange regime'. In addition, a higher magnetic field strength would give an increase in SNR.

Potential ‘metabolic problems' due to high 2DG doses could be obviated by injecting D-glucose instead of 2DG. However, D-glucose is rapidly metabolized, leading to the rapid decline of the glucoCEST signal (Figure 4). To maintain a high glucoCEST signal, high concentrations of glucose will need to be continuously infused. In addition, cellular metabolism is likely to be altered by such large doses of monosaccharides. However, elevated glucose concentrations can be achieved through intraperitoneal injection of glucose; this route elicits a smaller insulin response than if glucose is administered via other routes.⁴⁴

In conclusion, glucoCEST enables imaging 2DG and its metabolite, 2DG6P, in vivo. The images represent metabolic maps of the brain activity with respect to glucose uptake and metabolism. A major advantage of the technique is its ability to detect a low concentration of each metabolite via the dominant ¹H MR water resonance. The approach also makes possible the implementation of a glucose/deoxyglucose-detection method on those MRI scanners that have only a ¹H channel. The method opens up new avenues for mapping glucose uptake and metabolism in vivo for the diagnosis and prognosis of clinical conditions such as cancer and neurodegenerative diseases.

The authors declare no conflict of interest.