Imaging of sugar-based contrast agents using their hydroxyl proton exchange properties

Abstract

The ability of CEST MRI to detect the presence of millimolar concentrations of non-metallic contrast agents has made it possible to study, noninvasively, important biological molecules such as proteins and sugars, as well as drugs already approved for clinical use. Here, we review efforts to use sugar and sugar polymers as exogenous contrast agents, which is possible based on the exchange of their hydroxyl protons with water protons. While this capability has raised early enthusiasm, for instance about the possibility to image D-glucose metabolism with MRI in a way analogous to PET, experience over the past decade has shown that this is not trivial. On the other hand, many studies have confirmed the possibility to image a large variety of sugar analogues, each with potentially interesting applications to assess tissue physiology. Some promising applications are the study of (i) sugar delivery and transport to assess blood brain barrier integrity, (ii) sugar uptake by cells for their characterization (e.g. cancer vs healthy), as well as (iii) clearance of sugars to assess tissue drainage for instance through the glymphatic system. To judge these opportunities and their challenges, especially in the clinic, it is needed to understand the technical aspects of detecting the presence of rapidly exchanging protons through the water signal in MRI, especially as a function of magnetic field strength. We expect that novel approaches in terms of MRI detection (both saturation transfer and relaxation based), MRI data analysis, and sugar design will push this young field forward in the next decade.

1. INTRODUCTION

Gadolinium-based contrast agents (GBCAs) have been used clinically for MRI since the late 1980s. As these agents provide specificity and sensitivity by visualizing anatomical and physiological information not available from standard imaging, they have been proven to be extremely valuable for diagnostics. The most common uses for GBCAs in neuroimaging are associated with visualizing disruption of the blood-brain barrier in tumors and enhancement of vessels in MR angiography. However, they can also be employed to assess physiological parameters, such as blood flow, blood volume, and permeability. Today, GBCAs are administered in about 25% of all MRI examinations¹ with approximately 30 million yearly injections worldwide.² While these GBCAs are generally considered safe, they have been linked with cases of nephrogenic systemic fibrosis,^3,4 especially in patients with renal diseases. It has also been reported that Gadolinium⁵ accumulates in the brain^6–9 and the bones¹⁰ after repeated use. Although no association has been established between clinical findings and these deposits, the regulatory agencies are now issuing warnings for their use. As a consequence, the development of new contrast agents and methods not dependent on Gd use is warranted. A breakthrough towards that goal was achieved by Balaban and co-workers,^11,12 who suggested to use compounds with exchangeable protons (e.g. hydroxyl -OH, amide -NH, amine -NH₂) as MRI contrast agents. They showed that sensitivity-enhanced MRI experiments can be designed to image these millimolar concentration compounds by labeling the exchangeable protons using radiofrequency (RF) and measuring the repeated label transfer to water¹². Exchange also affects the transverse relaxation time of water^13,14. In recent years, the interests and efforts for developing exchange-sensitive MRI techniques and exchange agents have increased strongly, leading to a new field of study. Methods include chemical exchange saturation transfer (CEST) MRI,^12,15 chemical exchange sensitive spin-lock (CESL) MRI¹⁶, and even simple T₂-relaxation sensitive MRI^13,17–19 of exogenous exchange agents. Unfortunately, the development and regulatory approval of agents typically takes many years, starting with toxicity assessment and extensive trials. As such, the initial development of CEST agents for human use has focused on agents already approved for other imaging methods or drugs already used in the clinic. Examples are the computed tomography (CT) agents iopamidol^20–23 and iopromide,^24,25 salicylic acids,^26,27 several cancer drugs,²⁸ and different sugars^29,30 and sugar polymers.

In this review we focus on sugars that have been suggested as exogenous CEST contrast agents in the current MRI literature. Thus, we include for instance the polymer family of dextrans,^31–35 but not glycogen.^36,37 It is important to realize the difference between sugars and conventional contrast agents, namely that some of them are physiologically relevant. For instance, abnormal D-Glc transport and/or utilization in the brain could indicate early neuropathology; cancer cells exhibit aerobic glycolysis, where D-Glc is converted to lactate; hypoglycemia and deficiency of a major glucose transporter (GLUT1) of the blood-brain barrier (BBB) may lead to epilepsy.³⁸ Also, a reduction in cerebral glucose metabolism has been reported as one of the earliest signs of Alzheimer’s disease.³⁹ We therefore start the review with an overview of the physiological properties of the different sugars (Section 2). Exchange-based MRI detection in principle offers an opportunity to report on the delivery, transport and metabolism of D-Glc and other types of sugars without the use of either radioactive or metallic labelling. However, while this potential has stirred a lot of excitement, it has to be realized that separation of these three stages of uptake is not trivial and care should be taken not to assume that D-Glc based CEST MRI provides information similar to fluorodeoxyglucose (FDG) based positron emission tomography (PET).

In addition to physiological properties, the magnitude of the MRI effects depends on the MR spectral properties of the exchangeable protons in the agents (Section 3). Knowledge of these is crucial for optimizing the experimental parameters. MR methodological aspects such as pulse sequences, field dependence, and quantification are addressed in Section 4. Importantly, MRI and CEST tend to use acronyms for methods. With respect to such nomenclature, if not already specified in the original paper, we will follow the usual CEST MRI convention of listing either the proton group (amide, amine, etc) or the compound itself before the detection method. For instance, exchange-based saturation studies of D-glucose (D-Glc) would be glucoCEST or glucoCESL, while this would be dexCEST or dexCESL for dextran, etc. For dynamic studies, the nomenclature will be agent based and the detection method deleted, e.g. dynamic glucose enhanced (DGE) will apply to CEST, CESL and contrast-enhanced (CE) relaxation measurements. Section 5 summarizes some promising examples of current preclinical and clinical applications of sugar-based contrast agent studies in the brain, cancers, stroke, and Alzheimer’s disease (AD). Challenges and opportunities associated with using sugars in clinical imaging are described in Section 6.⁴⁰

2. PHYSIOLOGICAL PROPERTIES OF SUGAR ANALOGUES

Figure 1 gives a summary of sugars that have been studied at present using exchange MRI (Fig, 1A). Not surprisingly, the first sugars studied as contrast agents in vivo were D-Glc,^{29,30,41–54} and other single-ring hexoses, such as 2-deoxy-D-glucose (2-DG),^55,56 3-o-methyl-D-glucose (3-OMG),^57–60 and D-glucosamine (D-GlcN).^61,62 Some artificial sweeteners (sucralose, Slc, and maltitol, Mlt) and glucose dimers (Maltose, Mls) and polymers (dextrans, dex) have also been investigated. Of all of these, until this year, only D-Glc had been studied in humans,^{45–48,52–54,63–66} the reason being that it is already approved as a drug for intravenous glucose tolerance testing. A simple three-compartment model^41,67,68 can describe the biodistribution of sugars in most tissues: (1) vasculature (v), (2) extravascular extracellular space (EES, (e)) and (3) intracellular (c). As an example, the characteristics of tissue delivery, uptake and metabolism of the above sugars in tumors are shown in Fig. 1B. In other tissues, additional parameters may have to be accounted for, especially the BBB for transport into the brain, and the predominance of aerobic metabolism in most cell types in healthy tissue. From a simple organizational point of view, to understand the imaging findings in terms of compartmental signal origin, we can separate the sugars in four types: (i) Sugars transported into the cell and fully metabolized (D-Glc), (ii) those that are taken up and, within the period of dynamic contrast measurement, phosphorylated only (FDG, 2-DG, D-GlcN, GlcNAc), (iii) those are just taken up and accumulate within the limited scan period (3-OMG and xylose), and (iv) those that, similar to GBCAs, are not transported into the cell (L-Glc, Slc, Maltitol, Mls, dextrans).

Fig. 1.

(A) Structures of sugars studied as potential contrast agents by exchange-based MRI. (B) Overview of compartmental location of sugars (S) and their metabolic products (if any) in tumors, and the rate constants (k) for transport or metabolism. Color indicates relative contrast contribution (Darker color shows increased contribution, white indicates negligible sugar signal). For D-glc, the concentrations in vascular space and EES are comparable, but due to lower pH (See Fig. 2), the EES has a higher signal contribution. Intracellular signal is very small to negligible due to rapid phosphorylation and glycolysis. For 3-OMG and xylose, the signal is predominantly due to trapped intracellular compound, but xylose has very slow metabolism. For FDG, 2-DG, D-GlcN and GlcNAc, the phosphorylated product is the main component measured, but D-GlcN and GlcNAc are metabolized slowly. For L-Glucose and dimers or larger sugars, the agents occupy only plasma in blood. When they enter the EES, the concentration depends on k₁. For equal concentration, the signal in EES is higher due to the lower pH. Figure 1B is an extended version of a model presented in Chan et al.⁴¹ using physiological information from many references in Sections 2.1–2.5. This illustration, which assumes the presence of BBB disruptions, was derived from the model for D-Glc transport and metabolism in normal brain by van Zijl et al⁶⁸, which in turn was constructed based on the Sokoloff model for ¹⁴C-DG uptake for autoradiography.⁶⁷ Some parts were reproduced with permission from Chan et al. Magn Reson Med 2012;68(6):1764–1773.⁴¹

2.1. D-Glc

D-Glc is the natural substrate for most tissues. After delivery and facilitated transport into the cell, it undergoes glycolysis to pyruvate, which is further metabolized aerobically in the tricarboxylic acid (TCA)-cycle to glutamate, glutamine and other derivatives.^69,70 Under ischemic conditions, anaerobic glycolysis is activated with the end product being lactate. Despite the availability of millimolar amounts of substrate, the concentration of phosphorylated glucose products and pyruvate in the cell are in the micromolar range,⁷¹ which currently is too low to be detected by MR, indicated by a white color in Fig. 1B. In the healthy cell, the total amount of glutamate is constant,⁷² and no signal change is expected in the MR spectrum. Lactate accumulates during tissue ischemia or in actively metabolizing tumors, but the OH group of lactate resonates very closely to water and is not easily detectable in CEST, CESL or CE-T₂ relaxation experiments at clinical field strengths. Therefore, the main signal from D-Glc originates from the vasculature and the EES, which in case of a lack of a tissue barrier (BBB or other) will have the same concentration of D-Glc concentration. The reason the EES has a darker indication for more signal in Fig. 1B is that the pH in the EES of tumors is lower than in blood due to lactate exported from the cell. Because OH exchange for sugars is base-catalyzed in the physiological range (above pH = 6),^73,74 the exchange rate reduces with increased acidity. As OH exchange is very fast on the MR timescale, OH signal is merging with the water signal in the Z-spectrum. The reduced pH edges the OH protons to a slower exchange regime,⁷⁵ leading to increased detectability at lower pH for these fast exchanging protons (see Section 3 and Fig. 2C for an illustration of the effect of pH changes on the water saturation in the presence of a sugar contrast agent). In normal blood, the D-Glc concentration is about 5 mM and in intravenous (i.v.) DGE studies in humans, this is elevated to about 10–20 mM, depending on the amount (typically about 25–35 grams)⁷⁷ and timing of infusion and the insulin response. D-Glc is considered safe and has regulatory approval for other purposes (e.g. i.v. glucose tolerance testing,⁷⁸ which has led to enthusiasm in the CEST field for using D-Glc as an alternative to FDG-PET. However, as shown in Fig. 1B, the D-Glc signal origin is different from FDG and its uptake signal includes the sum of delivery, transport and phosphorylation. Unfortunately, there is only a single signal to be studied and separation into the different contributions is not trivial. On the other hand, some recent applications, e.g., in AD models (see below) are showing promises^79,80, and the field has to be further developed to understand the possible applications.

Fig. 2: Structures and spectral features of D-Glc (A-C) and 3-OMG (D) and exchange rates of hydroxyl protons in D-glucose (E).

(A) 11.7T ¹H NMR spectrum of D-Glc(300 mM, pH 3.5 in H₂O) at 4 °C and 37 °C. At higher temperature, the visibility of the hydroxyl protons reduces strongly due to the increased exchange rate, while the opposite will occur in Z-spectra. Assignments of the resonances for the two D-Glc anomers (B) are also given. (C) Z‐spectra of 20mM glucose solution in PBS buffer (37°C) acquired at 14.1 T as a function of pH and B₁. At pH = 6.2, some resonances can still be distinguished, but the increased exchange rates at higher pH (E) cause a merging of the signals with the water resonance. (D) Anomer structures and 11.7T ¹H NMR spectrum of 3-OMG (100 mM, pH 5.77 in H₂O) at 4 °C, 24 hours after dissolving α−3-OMG, showing the two anomers with similar resonance frequencies as the D-Glc anomers. (E) Estimated exchange rates for four resonance groups in the Z-spectra as a function of pH using a multi-B₁ fits of the data in (C). Figure parts reproduced with permission from Zhou and van Zijl. Prog Nucl Magn Reson Spectrosc 2006;48:109–136⁷⁶; Zaiss et al, NMR Biomed 2019;32(9):e4113⁵¹; Rivlin et al. Magn Reson Med 2014;72(5):1375–1380.⁵⁷ Assignment from the first paper was adjusted with recent knowledge from Zaiss et al.

2.2. FDG and 2-DG

Contrary to D-Glc, the metabolic pathways of 2-DG and FDG^41,49,55,81 stop after the first step of glycolysis (phosphorylation), after which 2-DG-6-phosphate (2-DG-6P) and FDG-6P are trapped intracellularly (Fig. 1B), which provides a long imaging window.^{49,55,56,81–83} Therefore, by measuring their accumulation over a prolonged period of time thus removing the influence of delivery and transport (generally one hour in FDG-PET), these sugar products can be used as an indicator of glucose metabolism.^{49,55,56,81–83} Because of the glycolytic inhibition property of these sugars, they have been considered as a drug for cancers.^84–86 Unfortunately, these deoxyglucose analogues are toxic with LD50 of 8 g/kg in mouse i.v.⁸⁷

2.3. D-GlcN and GlcNAc

D-GlcN and GlcNAc are amino monosaccharides that are transported into the cell via the glucose transporters GLUT1 (same affinity as D-Glc) and GLUT2 (~20 times more affinity)⁸⁸. Once in the cell, these sugars are phosphorylated and slowly metabolized.^89–91 As a consequence, GlcN-6P and GlcNAc-6P will accumulate in the cell making them a useful contrast agent, especially for tumor cells where they are often overexpressed. Rivlin and Navon^61,62 therefore suggested that these are good agents for human study, especially in view of their low toxicity (both per os (P.O.) and i.v.^92,93). They are also available over the counter as a nutritional supplement, as well as a drug for treatment of osteoarthritis⁹⁴ and inflammatory bowel disease.⁹⁵ Rivlin et al. also pointed out that high oral doses (5–15 g/kg P.O.) are well tolerated in humans^89,96 and that concentrations over 10 mM can be produced a few hours after oral administration, with a half-life time of many hours.^89,96,101 This long time window and the fact that neither D-GlcN nor GlcNAc show significant effects on blood insulin and glucose levels,⁹⁷ making them very suitable even for diabetic populations.⁹⁸ Both compounds also cross the BBB.^99,100

In an exciting recent first human study, using a dose of 7.5g GlcN-sulfate in 150 mL per person, Rivlin et al. demonstrated the possibility of D-GlcN to highlight breast tumors.¹⁰¹

2.4. 3-O-Methyl-D-Glucose (3-OMG) and D-xylose (D-Xyl)

3-OMG has a similar transport mechanism as D-Glc but cannot be phosphorylated by hexokinase. Therefore, it can be used as a substrate for studying glucose transport into the brain and tumors^{58–60,102,103} and since it accumulates after transport, to identify the location of tissues with increased uptake.⁵⁷ A typical dose of 3-OMG to study nutrient absorption has been 30 mg/kg in pediatric patients¹⁰⁴ and up to 2.5 g in adult.¹⁰⁵ 3-OMG is in general regarded as non-toxic, as it is not phosphorylated. It was found that it plays a role in regulating the thioredoxin-interacting protein.¹⁰⁶ The typical allowed dose for D-glucose is 500 mg/kg⁷⁷ and has been used up to 35 g in adults.¹⁰⁷ In animal studies, when given at the same concentration as D-Glc, 3-OMG showed an approximate doubling of contrast.⁵⁹ Therefore if 3-OMG is to become a useful contrast agent it has to be proven that it can be given at a higher dose than in previous human studies, i.e. close to that of D-Glc.

Recently, xylose (D-Xyl), a plant sugar that has been used as a diabetic sweetener in food products, was added to the list of potential exchange-based agents by Wang et al.¹⁰⁸ It has transport properties similar to D-Glc, but minimal metabolism in human tissues including the brain.^109,110 It has already been approved by the Food and Drug Administration (FDA) for human use at a dose of 25 g P.O. in malabsorption testing¹¹¹ and at doses of up to 1.5 g/kg in children. Combinations of 25 g P.O. and 10 g i.v. infusions have also been used.^111,112 While inducing an insulin response, D-Xyl barely affects blood glucose levels and actually has shown some benefits in regulation blood glucose levels in patients with diabetes.^113,114 Xylose can be transported across BBB, and was detected in rat brains using an on-resonance spin-lock approach¹⁰⁸. Based on the lack of metabolism in brain, we added D-Xyl in the same row as 3-OMG in Fig. 1B, but we could not find any literature on D-Xyl in tumors to verify this.

2.5. L-Glc, Slc, Maltitol, D-Maltose and dextrans

L-Glucose (L-Glc) is an enantiomer of D-Glc⁴⁹ that cannot be metabolized. L-Glc transport in vivo differs from D-Glc in that it is transported across the placenta less actively¹¹⁵, while transport across the BBB was found to be less than even for the polymer inulin.¹¹⁶ It is taken up only in tracer amounts by cancer cells.^117,118 Using CESL and CEST, L-Glc contrast could not be detected in the brain.^49,55 As such, it has been used in animals to evaluate the BBB leakage in brain tumors.^49,55 Physiologically, it decreases hematocrit and blood viscosity.¹¹⁹ Thus, any interpretation of the dynamical changes caused by L-Glc should be cautious in view of these physiological effects of L-Glc. It should also be mentioned that L-Glc has a laxative effect and has been used as a colon-cleansing agent in adults before colonoscopy (24 g in 8 oz of water).¹²⁰ This is not optimal for use as a contrast agent.

Slc (Splenda) and maltitol are artificial sweeteners that have been tested as exchange agents by Bagga and coworkers.^121,122 They are safe for their intended use as sweeteners¹²³. Slc could reduce or perturb the intestinal microbiota^124,125 and have other metabolic effects.¹²⁶ Similarly, maltitol could have an effect on gastrointestinal microbiota.¹²⁷ Further investigations on their physiological effects other than their intended use could help to navigate their potential imaging applications in humans.

The glucose polymers D-maltose¹⁸ and dextrans^32,34,35,129 have also been suggested as exchange-based contrast agents. D-Maltose is a 1–4 linked disaccharide that is not taken up or metabolized by the cell.¹⁸ Dextrans are 1–3 or 1–6 linked polymers of glucose that cannot be transported or metabolized by mammalian cells. While limited studies were found related to D-maltose, it could lead to an increase in specific extracellular enzymes, such as glucoamylase.¹²⁹ High molecular weight dextrans have been used in clinical studies for several decades with an excellent safety profile.¹³⁰ Both dextran-70 at 6%, 4 mL/kg and low molecular weight dextran-1 are safe, which have been used widely for hypersensitivity inhibition and islet autotransplantation in children.^105,131 Despite some risk for anaphylaxis for larger size molecules, dextrans have a long established safety profile which could be promising for repurposing them as contrast agents in humans.

These compounds do not cross the BBB and are generally not metabolized by mammalian cells and as such are expected to be excellent substitutes for GBCAs for the goal of studying tumor enhancement, BBB breakdown and even angiography. Except for L-Glc, they have low toxicity and are expected to be suitable for both oral or i.v. delivery.

3. MR SPECTRAL PROPERTIES OF SUGAR ANALOGUES

The hydroxyl resonances of cyclic hexoses such as D-Glc (Fig. 2A) and its analogues (2-DG, 3-OMG, D-GlcN, GlcNAc) can be visualized in NMR spectra for solutions in H₂O at low pH and low temperature, such as shown at 11.7T for D-Glc in Fig. 2A and 3-OMG in Fig. 2D. The exchangeable proton resonances of the two enantiomers in solution (e.g. for D-Glc in Fig. 2B) appear over the approximate ranges of 0.6–0.8 ppm, 1.1–1.7 ppm, 2.1–2,2, and 2.8–2.9 ppm relative to the water resonance assigned to 0 ppm for reference (Figs. 2A, D). For glucose polymers with linkage in the 1-position or analogues without a hydroxyl group in the 1-position, the spectral range is generally limited to the 0.5 – 1.8 ppm range. Increasing the temperature (Fig. 2A) and the pH and buffering the solution will make these resonances invisible in the ¹H NMR spectrum due to exchange broadening, but they remain visible in the water saturation spectrum (Z-spectrum, Fig. 2C). At low pH in PBS (Fig. 2C), these Z-spectra^{36,41,42,49,51,56,57,59–62,81–83,102,103,132} show similar ranges for the resonance maxima as the NMR spectra, but these saturation signals are broadened and appear to be pushed to higher frequency when increasing B₁, an effect due to the interfering effects of direct water saturation and, in vivo, the semi-solid magnetization transfer effect. Not unexpectedly, based on the number of protons involved, the maximum signal appears around 1.2 ppm for most glucose analogues which has become a popular offset frequency to be used in DGE experiments (see below). When increasing the pH to physiological levels, all signals start to merge with the water resonance due to faster exchange (Figs. 2C, E). Even at the high field of 14.1T the spectrum is in the intermediate exchange regime: Δω rad /s ≈ k_ex (Hz), while it moves closer to the fast exchange regime at clinical field strengths such at 3T. Thus, under physiological conditions, the hydroxyl protons of sugars are typically detected over the 0–2 ppm range in the Z-spectrum, with features depending on pH and the static field strength B₀ and the RF strength B₁ used. In vivo, the Z-spectrum is complicated with potential contributions from other hydroxyl protons, for instance from glycogen³⁶ present at high concentration in liver and muscle, or myoinositol in the brain.¹³³ Other exchangeable protons, e.g. from amine protons in glutamate in the brain¹³⁴ may also contribute. Fortunately, when using sugars as contrast agent, we study the dynamic difference time curve of the Z-spectrum and these other interfering resonances can be (partly) subtracted out. As seen above from Fig, 1, different sugars have specific transport and utilization in vivo, so they can reflect different physiology or pathology when used as unique CEST contrast agents. Moreover, the sensitivity of the Z-spectrum to exchange rate may be exploited to better visualize pathological microenvironments, such as the acidic tumor microenvironment (EES ~pH 6.5–6.8) where the fast exchanging hydroxyl protons slow down, increasing their detectability in tumors.

Fig. 3.

Simplified illustration of pulse sequence preparation periods commonly used for DGE MRI. (A) continuous wave-CEST: on clinical scanners, usually continuous wave saturation is not possible due to hardware and SAR limits. Instead, a pseudo-continuous wave saturation can be achieved by applying RF irradiation through alternating transmitting channels or using a pulse-train approach in which short delays in between saturation pulses are added. (B) on-resonance spin lock, which means the 90 pulses are on resonance to the water frequency. In practice, adiabatic pulses are helpful to avoid B₁ inhomogeneity. (C) on-resonance variable delay multi-pulse (onVDMP): the binomial pulse pair is on resonance. The delay, t_mix, should be minimized to increase saturation efficiency in DGE MRI. Phase cycling of binomial pulses can be used to reduce image artifacts caused by B₁ inhomogeneity. (D) Carr-Purcell-Meiboom-Gill (CPMG) sequence: t_mix values of 6 ms and 111 ms have been used to measure the DGE signal in the brain parenchyma and CSF at 3T, respectively,¹⁹ while a t_mix of 10ms was used to assess T₂ effects with DGE.¹⁷

4. METHODOLOGY

4.1. Exchange sensitized acquisition

The exchange rate k_ex of the hydroxyl protons in D-Glc or its derivatives under physiological conditions is more than 3 kHz (Fig. 2E) and in addition to using chemical exchange label transfer, sugars also reduce the water proton transverse relaxation time, T₂.^13,17,135 This mechanism, which also underlies the CESL technology, is based on signal dephasing due to exchange between the frequency of the solute proton and the water frequency (see below for the equations). Exchange-based contrast has been obtained by either the acquisition of Z-spectra or by dynamic-glucose enhanced MRI,¹³⁶ where typically signal intensities at specific offset frequencies related to sugars are acquired at a temporal resolution of a few seconds or slower, depending on the number of slices and the image resolution. Any pulse sequence that is designed to be sensitive to exchange transfer (CEST, CESL) or T₂ relaxation can be used for DGE MRI. Numerous approaches, including both off-resonance and on-resonance, have been investigated (see Fig. 3 for some examples), with the “on-resonance” terminology relating to the water proton frequency. CEST with full Z-spectral acquisition is the most suitable technique for both static and dynamic acquisitions since it may offer some distinction of contributions based on chemical shift and allows frequency referencing to correct for possible B₀ changes due to field drift or subject motion. The on-resonance techniques are only suitable for dynamic studies since no chemical shift information is encoded. The field strength (B₀) and saturation parameters (B₁, pulse shape and length, inter-pulse timing) dictate the detection sensitivity of D-Glc and its analogues in vivo (Fig. 2).⁷⁵

A key difference between DGE approaches and studies designed to detect endogenous exchangeable species is that the CEST contrast agent (glucose or its derivatives) is given intravenously during the MRI scan. The change in signal can therefore be attributed to the effect of the injected contrast agent. In most studies, quantification of DGE signal is performed by taking the difference between post-injection scans at each time point and the average of the pre-injection scans, followed by analyzing the dynamic time difference curve. The first DGE studies used classical saturation transfer parameters, namely long, low B₁ off-resonance RF irradiation (continuous wave, CW; or pseudo continuous wave) lasting between 1–2 s. Contrary to classical CEST MRI, where the irradiation frequency is changed, a single fixed irradiation frequency was used in early DGE MRI studies to allow the detection of dynamic signal changes with higher temporal resolution. As mentioned above, since there is an expected signal maximum around 1.2 ppm at lower pH and all signals merge at higher pH, it has been common to use this frequency offset. A few DGE studies at ultrahigh fields (11.7T, 9.4T and 7T) used this approach.^44,45 At lower field strength, such as 3T, the frequency separation between protons at 1.2 ppm and bulk water becomes smaller (~150 Hz) and saturating at this frequency will result in profound direct water saturation effects. In addition, a signal coalescence effect of hydroxyl protons with water protons occurs due to being in the intermediate to fast exchange regime, which removes any frequency specificity for the hydroxyl protons. Fortunately, all exchangeable protons downfield of water collectively still result in an asymmetry of water spectrum.⁵³ Figure 4 illustrates this effect at 7T and 3T. Under such conditions, it is no longer crucial to set the irradiation frequency at 1.2ppm and an offset judged on a trade-off between maximal CEST signal and water direct saturation can be determined for the B₁ used. An initial in vivo human DGE study at 3T used a frequency offset of 2 ppm.⁵³ The use of a single frequency instead of asymmetry analysis makes the signal change sensitive to symmetric changes in the water signal after injection, such as a change in T₂ or T₁. T₁ does not change upon infusion,¹⁷ but T₂ decreases enhance the DGE effect. Possible changes in S₀, e.g. due to magnetic susceptibility changes or water content changes affect the results of both approaches. In the current literature, the effects of these contributions have not yet been analyzed.

Fig. 4. Simulation of the effect of peak coalescence between hydroxyl and water protons on the DGE signal difference at different fields.

The concentration of D-Glc was assumed to be increased by 10 mM, and the water T₂ reduced by 1% due to the glucose infusion. The difference spectra were calculated by taking the difference between the pre-infusion and the post-infusion Z-spectra at (A) 7T and (B) 3T. For a more comprehensive simulation, taking into account of the effect of B₁ and different tissue types, please refer to Fig 1 of Xu et al. Magn Reson Med. 2020;84(1):247–262.⁵³

An alternative way to avoid the coalescence effect while at the same time sensitizing to rapid chemical exchange is to use on-resonance techniques.¹³⁷ The on-resonance CESL technique (Fig. 3B) has been used both in pre-clinical^43,49,58 and human studies at various magnetic fields.^46,47,52 During the spin-locking period (TSL), the decay of the water signal is governed by the T₁ relaxation rate in the rotating frame, R_1ρ(1/T_1ρ). The presence of fast exchanging protons with a frequency different from water protons will increase the R₂ and R_1ρ due to exchange-mediated signal dephasing:^17,43

$$R_2=R_{2,0}+R_{\mathit{\text{ex}}}$$ [1a]

$$R_{1\rho }=R_{\mathit{\text{eff}}}+R_{\mathit{\text{ex}}}^{\mathit{\text{SL}}}$$ [1b]

In the case of on-resonance spin lock, R_eff = R_2,0, where R_2,0 is the transverse relaxation rate of water protons without chemical exchange contribution; R_ex is the exchange induced relaxation rate. Assuming a proton pool at a single offset and that the exchange effect of the administered sugar will dominate the change in relaxation, the changes in R₂ and in on-resonance R_1ρ upon glucose injection can be expressed as:^17,43,138

$$\Delta R_2=\Delta R_{\mathit{\text{ex}}}+\Delta R_{2,0}\approx \Delta R_{\mathit{\text{ex}}}=\frac{\Delta p·k_{\mathit{\text{ex}}}·\Delta {\omega }^2}{\Delta {\omega }^2+k_{\mathit{\text{ex}}}^2}$$ [2a]

$$\Delta R_{1\rho }=\Delta R_{\mathit{\text{ex}}}^{\mathit{\text{SL}}}+\Delta R_{2,0}\approx \Delta R_{\mathit{\text{ex}}}^{\mathit{\text{SL}}}=\frac{\Delta p·k_{\mathit{\text{ex}}}·\Delta {\omega }^2}{\Delta {\omega }^2+{\omega }_1^2+k_{\mathit{\text{ex}}}^2}$$ [2b]

where Δp is the ratio of hydroxyl proton concentration of the exogenous agent to water proton concentration, k_ex is the exchange rate (in Hz) and Δω is the chemical shift between the hydroxyl protons and water protons (in rad/s). ω₁ = γB₁ is the amplitude of the spin lock field (in rad/s). Notice that the rotating frame exchange-based relaxation enhancement increases with reduced spinlock field to the limit of being equal to the transverse relaxation rate change without RF irradiation. When measuring a spin-lock based Z-spectrum, the observed signal change needs to be normalized to the signal without spin lock preparation. The normalized relative signal change can be approximated as: ΔS_rel ≈ ΔR_ex·TSL, and given a fixed spin-locking time (TSL), the signal is proportional to change in exchangeable proton concentration, Δp.⁶⁴

When comparing detection sensitivity between CEST enhancement (~ Δp·k_ex·α, in which $\alpha \approx {\omega }_1^2/[{\omega }_1^2+k_{\mathit{\text{ex}}}^2])$ and the relaxation enhancements, it is important to note that all methods have a proportionality to Δp·k_ex. This is multiplied by a method-specific ratio that is 1 or less, which includes MR acquisition or spectral parameters (e.g. RF strength and chemical shift difference). Thus maximum effect size should be similar between these approaches. These latter terms can be used to optimize sensitivity with respect to acquisition or agent properties for ${\omega }_1^2$ and Δω² respectively. Depending on the agent, the relative magnitude of these terms versus the exchange rate determines the enhancement. Notice that while CEST signal can be optimized with B₁, the relaxation methods have an inherent B₀ dependence in the term Δω, which leads to optimal detection of D-Glc under physiological conditions at a field strength about 11.7T (~ 5 times larger signal than at 3T)¹⁷. In addition, when studied at the D-Glc frequency between 1–2 ppm, the Z-spectral changes during the first few minutes in the dynamic scan should be specific for D-Glc and include an additional effect due to the transverse relaxation rate increase.

An issue with ultra-high field human scanners (≥7T) is the presence of strong B₁ and B₀ field inhomogeneities. Using conventional hard 90° pulses for the flip-down flip-up pulses surrounding TSL can cause severe image artifacts and reduced locking efficiency due to a mismatch between the excited magnetization and the spin locking field. Therefore carefully designed adiabatic pulses that are immune to B₁ field inhomogeneities are important for in vivo DGE studies at ultra-high field.⁶⁴

Another technique that has been used in DGE studies is the on-resonance variable-delay-multi-pulse (onVDMP) method (Fig. 3C),¹³⁹ composed of a train of binomial pulses interleaved by short delays. The term “variable delay” stemmed from the original off-resonance version of the sequence, a special version of the frequency-labeled exchange (FLEX) transfer method,^140,141 where the mixing time (delay) between saturation periods was varied to serve as an exchange rate filter.^142,143 Since fast exchange causes the frequency label to transfer rapidly during the mixing time, the delay between pulses can be as short as possible within specific absorption rate limit.¹⁴⁴ The signal resulting from the onVDMP sequence is:

$$S=S_0{\text{e}}^{-t_{\mathit{\text{on}}}{R}_{\mathit{\text{on}}}}$$ [3]

where t_on is the length of the onVDMP pulse train; R_on is the relaxation rate during onVDMP which is sensitive to exchange, which can be described using relaxation theory.¹⁴⁵ The R_on is very similar to R_1ρ except that the water magnetization is being kept predominantly along the longitudinal direction by the binomial pulses.

With glucose infusion:

$$S+\Delta S=S_0{e}^{-t_{\mathit{\text{on}}}\left(R_{\mathit{\text{on}}}+\Delta R_{\mathit{\text{on}}}\right)}=S·e^{\left(-t_{\mathit{\text{on}}}·\Delta R_{\mathit{\text{on}}}\right)}$$ [4]

the normalized signal change is:

$$\frac{\Delta S}{S}=e^{\left(-t_{\mathit{\text{on}}}·\Delta R_{\mathit{\text{on}}}\right)}-1$$ [5]

When −t_on·ΔR_on ≪ 1, this can be approximated as:

$$\frac{\Delta S}{S}=-t_{\mathit{\text{on}}}·\Delta R_{\mathit{\text{on}}}$$ [6]

Similar to on-resonance spin lock technique, the onVDMP method provides higher sensitivity in detecting glucose signal¹⁴⁶ and is sensitive to B₀ and B₁ field inhomogeneities. The latter can be mitigated by using a simple phase cycling scheme.^79,80 Inverting the phases of some binomial pulses improves the robustness of onVDMP against small shifts in water resonance frequency and B₁ inhomogeneity as demonstrated by a recent report.⁸⁰ At ultra-high magnetic field, onVDMP DGE imaging showed improved signal sensitivity compared to CW-CEST,¹⁴⁶ and equivalent sensitivity to onCESL.⁸⁰

It is important to point out that while exchange sensitive on-resonance techniques detect all fast exchanging protons, this lack of specificity is generally not an issue with DGE experiments since the sugar is given as a contrast agent. The change in signal that is detected following the injection can be attributed to the change introduced by glucose (concentration, pH, osmolarity and eventually metabolism). It might be worth noting that in the later part of the DGE dynamic curve (>5 min), the signal may have contributions from D-Glc metabolic products, such as lactate in tumors. Some rapidly exchanging products may contribute more weighting in onVDMP and onCESL images than in CW-CEST images.

T₂ relaxation can also be exploited to detect the change in sugar concentration.^{13,17,135,136} Standard fMRI techniques such as blood oxygen level-dependent (BOLD) MRI can be used to detect changes in certain regions of the brain following oral glucose intake or IV glucose injection.^107,147 A T₂ specific measurement for DGE would use a Carr-Purcell-Meiboom-Gill (CPMG) pulse train (Fig. 3D), which employs of a series of refocusing 180° pulses to preserve T₂ effect while reducing the effect of inhomogeneities. When used as preparation, the refocusing series is surrounded by two 90° pulses of opposite phase. A very recent paper compared the CPMG, onVDMP and onCESL in a single DGE session in a mouse study.¹⁹ The CPMG method showed the highest change of DGE signal in brain parenchyma upon glucose infusion, but showed high level of fluctuations in signal in the CSF, most likely due to through-plane flow.

When choosing a suitable technique for DGE experiments, considerations need to take into account magnetic field strength, inhomogeneities of the B₀ and B₁ fields, the targeted anatomy (eg. brain parenchyma vs CSF), length of the dynamic study, hardware and SAR limitations, and for clinical scanners, sequence availability.

4.2. Quantification

Analysis of static Z-spectra for sugar studies is similar to that for other fast exchanging compounds⁷⁵ and not discussed here. For dynamic studies, the quantification of the DGE hemodynamics has some resemblance to tracer kinetic contrast agent methods, but several important details are different. Firstly, glucose is not really a tracer due to the large quantity that gets injected. Secondly, the infusion period is much longer than what is used typically for other contrast agents (e.g. 1–4 minutes for glucose compared to 3–4 seconds for GBCA). While DGE is being further established, investigators have used some simple approaches to at least get an impression of how the signal is changing. A common one is to calculate the area under the curve (AUC) of the DGE signal response, either by normalizing the signal to S₀ or to the average baseline signal (S_base) and then subtracting the measured signal at each time point from this normalized reference signal. This difference signal is then integrated:^{44,45,48,53,59}

$$\mathit{\text{AUC}}=\frac{1}{n}{\Sigma }_1^n\frac{S_{\mathit{\text{base}}}}{S_0 \mathit{\text{or}} S_{\mathit{\text{base}}}}-\frac{S\left(t_n\right)}{S_0 \mathit{\text{or}} S_{\mathit{\text{base}}}}=\frac{1}{n}{\Sigma }_1^n\frac{\Delta S\left(t_n\right)}{S_0 \mathit{\text{or}} S_{\mathit{\text{base}}}}$$ [7]

To compare between studies or time periods of integration, the signal difference can be normalized based on the number of acquisitions (n). Challenges in kinetic modelling of DGE MRI are not only related to the large volume injected and the slow infusion. Contrary to Gd agents, D-Glc is transported over the BBB and blood CSF barrier in normal tissue. The main issue for quantitative analysis then is the availability of only a single exponential-like tissue response curve that reflects the convolved effects of D-Glc delivery, tissue uptake and metabolism. The theory for this effect has been described previously for ¹³C-glucose uptake studies using a rapid hyperglycemic clamp in which blood D-Glc was ramped to maximum within a minute and kept constant, basically providing a rectangular arterial input function (AIF)¹⁴³ to simplify the analysis.⁶⁸ There have been some disagreements on whether intracellular phosphorylated glucose compounds contribute, but older literature has shown that their concentrations are in the micromolar range in normal tissue and not relevant as a signal contribution.⁶⁸ Also, the D-Glc concentration reduces per transport step with a factor of 4–5, i.e. from blood to EES and from EES to intracellular. In tumors, when the BBB is leaky, the EES concentration will be comparable to that in blood, but the intracellular concentration decays fast¹⁴⁸ due to fast metabolism⁴¹ (Fig. 1B). Another potential issue is that in tracer kinetic modeling one usually has to convert signal to concentration. While D-Glc CEST signal and thus CEST-based DGE signal changes have been confirmed to be proportional to concentration (Δp term in the equations above) in phantoms, such a conversion is complicated at higher concentrations (> 20–30 mM) and higher B₁ values as water saturation becomes limited due to back exchange of saturated protons. Furthermore, in vivo factors such as pH differences between tissue compartments, competing background magnetization transfer effects, as well as contributions from changes in the transverse relaxivity of D-Glc due to OH proton exchange.^13,17 For the relaxation methods, proportionality to concentration should extend to higher concentrations. Another difference from conventional contrast agent methods is that the response to D-Glc may vary between individuals due to differences in insulin response and glucose metabolism.⁴⁸ Thus, AIF measurements resulting from the DGE experiments cannot be generalized and the retrieval of individual AIFs will probably be needed to quantify hemodynamic parameters.

5. APPLICATIONS

Exchange-based imaging of D-Glc and its analogues provides an approach to study alterations in glucose uptake and utilization, in particular in the brain where D-Glc transport and metabolism are tightly regulated and abnormalities could indicate pathological status. Moreover, the exchange-based signal is dependent on the tissue environment, especially pH, that may be affected under specific physiological and pathological conditions. For example, tumors have an acidic pH in the EES, leading to higher exchange-based effects as the exchange rate slow-down allows more efficient labeling. Many interesting studies in cancers, stroke, Alzheimer’s disease, placenta and kidney have been published, especially in animals, but also in humans. Currently published exchange-based studies using D-Glc, 2-DG, and 3-OMG are summarized in Table 1, which includes animal models, field strength, pulse sequence and acquisition method used, as well as the frequency offset employed for dynamic scans.^{19,41–56,64–66,79–83,108,146,150–153} Due to the abundance of studies, we will just give a few examples that should be representative of current options and capabilities for physiological studies.

Table 1.

Exchange-based studies using D-glucose (D-Glc), deoxyglucose (2-DG) and 3-o-methyl-D-glucose (3-OMG) as contrast agent.

Organ	Condition	Species	B0 (T)	Method, offset frequency	Dose (g/kg)	Study
D-glucose
Brain	Healthy	C57BL/6 mouse	3	glucoCEST CPMG/onVDMP/onCESL DGE (on-resonance)	2.5, 1.25, 0.63	Huang J, et al. 202119
		Rat	9.4	CESL, DGE (on-resonance)	0.075, 0.5, 1.0	Jin T, et al. 201443
		Human	7	glucoCEST DGE (1.2 ppm)	~0.38	Knutsson L, et al. 201848
			3	glucoCEST DGE (1.0 ppm, 2.0 ppm)	~0.38	Seidemo A, et al. 202154
	AD APP/PS1	C57BL/6	11.7	glucoCEST onVDMP, DGE (on-resonance)	3.125	Chen L, et al. 202180
	Tau	mouse	3	glucoCEST onVDMP, DGE (on-resonance)	~2.5	Huang J, et al. 202079
	Aβ protein	Rat	7	glucoCEST, 0.9 ppm	~0.0012	Chen P, et al. 2021149
	Cancer U87EGFRvIII	SCID mouse	11.7	glucoCEST DGE (1.2 ppm)	~2.5	Xu X, et al. 201544
	U87EGFRvIII		11.7	glucoCEST DGE, (1.2 ppm)	~2.5	Xu X et al. 2019150
	9L glioma		11.7	glucoCEST, onVDMP, DGE (1.2 ppm) DGE (on-resonance)	~2.5	Xu X et al. 2019146
	glioma	Human	7	glucoCEST DGE (1.2 ppm)	~0.38	Xu X, el al. 201545
	glioblastoma			CESL DGE, (on-resonance)	0.26	Paech D, et al. 201746
	glioblastoma			CESL DGE (on-resonance)	~0.31	Schuenke P, et al. 201747
	gliosarcoma			CESL DGE (on-resonance)	~0.31	Schuenke P, et al. 201764
	glioblastoma			CESL DGE (on-resonance)	~0.31	Boyd PS, et al. 202066
	glioblastoma		3	CESL DGE (0.6,0.9, 1.2,1.5 ppm)	0.3	Herz K, et al. 201952
	glioma			glucoCEST DGE (2.0,1.5 ppm)	~0.38	Xu X, et al. 202053
Breast	Cancer MDA-MB-231/MCF-7	Nude mouse	11.7	glucoCEST, CW-CEST (0.8–2.2 ppm)	2.5	Chan KWY, et al. 201241
	-TS/A	BALB/c mouse	9.4, 7, 3	CESL (1.2 ppm)	3	Zaiss M, et al. 201951
	-4T1	BALB/c mouse	7	glucoCEST, CW-CEST (1.2 ppm)	3	Capozza M, et al. 2021151
Colon	Cancer -LS174T -SW1222	Nude mouse	9.4	glucoCEST, (0.5–4 ppm)	~0.17	Walker-Samuel S, et al. 201342
Head/Neck	Cancer	Human	3	glucoCEST, (1.12–5.03 ppm)	20 mL, unknown	Wang J, et al. 201663
Kidney	Renal transplant	Rat	9.4	glucoCEST, (1.0–2.4 ppm)	2.68	Kentrup D, et al. 2017152
Lymph/Prostate	Cancer	Human	3	glucoCEST DGE, (2–3 ppm)	0.34	Kim M, et al. 201950
Placenta	Healthy	CD-1 mouse	11.7	glucoCEST onVDMP, DGE (on-resonance)	2.5	Wu D, et al. 2018153
		Human	3	glucoCEST, (0.75–4 ppm)	0.77	Luo J, et al. 201965
2-Deoxy-D-glucose
Brain	Healthy	Rat	9.4	glucoCEST, DGE (1.0 ppm)	1.0, 0.5	Nasrallah FA, et al. 201355
				CESL, DGE (on-resonance)	1, 0.5, 0.25	Jin T, et al. 201656
	AD	Mouse	7	glucoCEST, DGE (1.2 ppm)	0.5	Tolomeo D et al. 201883
	Cancer 9L glioma	Rat	9.4	CESL, DGE (on-resonance)	1, 0.5, 0.25	Jin T, et al. 201849
			7	CESL, on-resonance	1	Zu Z, et al, 201481
Breast	Cancer DA3	BALB/C mouse	7	glucoCEST DGE (1.2 ppm)	2	Rivlin M, et al. 201382
3-o-methyl-D-Glucose
Brain	Cancer U87 glioma	SCID mouse	11.7	glucoCEST, (1.2 ppm)	3	Sehgal AA et al. 201859
	glioblastoma	Rat	9.4	CESL (on-resonance)	1.5	Zu Z et al. 2018132
	Stroke	Wistar rat	9.4	CESL, (on-resonance)	1, 4	Jin T, et al. 201858
Breast	Cancer DA3-D1-DMBA-3	BALB/C Mouse	7	glucoCEST, (1.2 ppm)	1.5	Rivlin M, et al. 201457
	D1-DMBA-3	BALB/C mouse	7	glucoCEST, (1.2 ppm)	0.45, 0.57, 0.7, 0.75, 1, 1.5, 3.0	Rivlin M, et al. 2018102
	MDA-MB-231	RNU rat	7	Deuteration MRS, 3.5 ppm	0.89	Hartmann B et al. 2021154
Skin	Cancer B16-F10	C57BL/6 mouse	7	glucoCEST, (1.2 ppm)	1.5, 3	Anemone A et al. 202160
NA	In vitro 3-OMG solution	NA	11.7	glucoCEST, 1.2, 2.1, 2.8 ppm	NA (20 mM)	Rivlin M, et al. 2018103

NA Not applicable.

5.1. Cancer

D-Glc is a primary source of energy for cells and a precursor for the synthesis of various important metabolites. Tumors, such as gliomas, prostate and breast carcinomas have a high reliance on anaerobic glycolysis and display a concomitant increased production and cellular export of lactate, which results in a decreased pH in EES.¹⁵⁵ Moreover, leaky vasculature and upregulation of the glucose transporters GLUT1 and GLUT3 may also occur as a consequence of aggressiveness.⁵

GlucoCEST contrast was first demonstrated in two independent studies in tumor mouse models after administration of D-Glc.^41,42. In a study of a mouse breast tumor models, a lower glucoCEST contrast was observed in MDA-MB-231 tumors compared to MCF-7 tumors, which was explained tentatively by the authors to be due to the higher utilization of glucose in the aggressive tumors (MDA-MB-231),⁴¹ which in view of the below considerations may not be correct. Interestingly, differences between these two tumors were not observed in FDG-PET and contrast-enhanced MRI, leading to the conclusion that different mechanisms of contrast play a role. In the other early study, Walker-Samuel et al.⁴² demonstrated that glucoCEST could distinguish two types of colon tumors (LS174T and SW1222). These results correlated with FDG-PET uptake findings. However, correlation need not imply causality. Based on the different signal origins for D-Glc and FDG (Fig 1B) and the low concentration of phosphorylated glucose derivatives in tumor cells, it should be kept in mind that PET and glucoCEST provide intrinsically different information.

A great advantage of exchange-based MRI over MRS is that it allows acquisition at a higher spatial and temporal resolution, allowing the acquisition of dynamic data similar to those when using GBCAs. As a first example, DGE MRI using D-Glc has been shown to generate valuable information about D-Glc uptake kinetics in the brain and brain tumors. A first study of mice with glioblastoma⁴⁴ showed that the dynamic curves could clearly distinguish tumor from healthy tissue both in terms of maximal signal intensity and AUC, similar to DCE results in the same animals. The differences between tumor and contralateral normal brain in DGE reflect the perfusion properties and BBB breakdown^43–45 and the enhanced tumor signal were attributed to the relative low pH microenvironment of tumors (~6.5–6.8 pH) and high blood volume in tumors (Fig. 1B). To understand this, it is important to know that the intracellular concentrations of D-Glc and its phosphorylated products have been reported to be negligible (micromolar concentrations) in tumor cells¹⁵⁶ and the brain⁶⁸ and that lactate is not well detectable with CEST, especially at frequencies above 1ppm. So fast metabolism to lactate should not increase the DGE signal but actually delay reaching its maximum and further lead to reductions when substrate runs out. A recent study by Jin et al.⁴⁹ using multiple sugar monomers provides important insight into the signal origin for DGE studies and great support for the principles outlined in Fig. 1B. They compared DCE using GdDTPA (Fig. 5A) with DGE curves for D-Glc, 2-DG and L-Glc at the same infusion concentration (Fig. 5B) for a rat 9L glioma model. The results in Fig. 5B, show several relevant features. (i) the D-Glc and L-Glc peak maxima at about 4–5 min (Fig. 5B) are equal within error. This is experimental confirmation that most of the CESL-MRI signal comes from the EES due to low concentrations of intracellular sugar and its phosphorylated products (see above^{69, 156,157}). As L-Glc is not preferentially transported by GLUT,^49,56 its signal increase after infusion is a good indicator of that in EES and vasculature after BBB breakdown (cf. Fig. 1B). This is further confirmed by the time curves of DCE and L-Glc based DGE being comparable within error (Figs. 5D, E). (ii) the D-Glc curve increases slightly slower than the L-Glc curve (Figs. 5B,D), which we attribute to transport into the cell and metabolism; (iii) the maximum CESL (ΔR_1ρ) signal change for 2-DG infusion is several times that of D-Glc (Fig. 5B)⁴⁹, attributed to signal from the intracellular accumulation of 2-DG-6P (cf. Fig. 1B) adding to that of vasculature and EES; (iv) the D-Glc signal decreases at later time points (Fig. 5B), attributed to continuous metabolism by the tumor to lactate, a metabolic product not easily detected by CESL⁴⁹ or CEST at low field,²⁹ due to a small chemical shift difference with water. Overall, the results of Jin et al.⁴⁹ confirm the principles outlined in Fig. 1B, but these authors are also careful to point out that the relative contribution in the EES may be higher for CEST than for CESL because CESL is less sensitive to pH changes. The same study of Jin et al.⁴⁹ also showed a small uptake of L-Glc in contralateral brain (Fig. 5C), presumed normal, on which the authors did not comment, but which we attribute to contributions from blood vessels in the voxel. Contralateral brain showed reductions in 2-DG and D-Glc uptake by a factor of about two (Fig. 5C) relative to tumor (Fig. 5B). In this case, L-Glc should be due mainly to blood vessels and D-Glc to vessels + EES. However, compared to tumor EES, normal brain EES has reduced D-Glc concentration (about 1/4 of the blood value⁶¹) due to facilitated transport over the BBB. This is also most likely one of the causes of the uptake curve of D-Glc in normal brain being slower than in the tumor (Fig. 5F), but not the only one as the half-time of D-Glc uptake into the brain is about 1–2 min^68,158,159. Actually, the maximum of the DGE curve has moved to about 15 min, possibly due to the contribution of glutamate metabolism, which is being produced on this time scale. The fact that the uptake signal in normal brain is not a factor of 4 less could be due to contributions of the metabolic products of the TCA cycle (glutamate, glutamine) contributing to the non-specific on-resonance CESL and on-resonance VDMP signal, but in general these are turned over and maintained at constant levels in the brain^160–162. So the reason for this factor is less clear, and will have to be investigated further.

Fig. 5.

Comparison of time courses (DCE and ΔR_1ρ based DGE) for contrast agent uptake in a 9L-glioma rat model. (A) DCE after GdDTPA injection (n = 4). (B,C) DGE time courses for 0.25 g/kg of 2-DG, 0.25 g/kg of L-Glc, and 1 g/kg of D-Glc (scaled down by a factor of 4 to correspond to 0.25 g/kg D-Glc) to allow visual comparison. The initial parts of the curves were fitted to Sigmoid functions. (B) ROI in the tumor, showing equal fast early delivery for all sugars. L-Glc should reflect signal in extracellular space (Vessels + EES, See Fig. 1B). D-Glc and L-Glc delivery lead to similar maximum signal, indicating D-Glc similar compartmental origin. However, D-Gl reaches maximum signal slower, which we attribute to transport into the cell and fast metabolism to lactate (not detected). This is supported by D-Glc signal reducing further when substrate runs out. 2-DG signal reflects extracellular and intracellular space, continuously increasing due to build-up of the phosphorylated product. (C) ROI in contralateral brain, assumed to be normal. Both 2-DG and D-Glc signal are about halved compared to tumor, while L-Glc signal is small (noisy), but not negligible. In this case, L-Glc should be due mainly to blood vessels and D-Glc to vessels + EES, but normal brain EES has reduced D-Glc (factor of about 4 due to facilitated transport). (D-F) normalized time course comparisons. (D) DCE has the fastest time to peak, while 2-DG has the slowest. (E) Comparison of the 10 to 90% rise time for DGE MRI and CESL with L-Glc, D-Glc, and 2-DG injections. *: p<0.05, **: p<0.01. (F) The D-Glc time course in tumor is faster than normal brain, attributed to BBB breakdown. Reproduced, with permission from Jin et al. Magn Reson Med. 2018;80(2):488–495⁴⁹; Parts (B), (C), and (F) were kindly generated upon our request by Dr. Tao using the original data.

Among a few other studies that have also investigated the compartmental contribution of glucoCEST signal,^56,151 DGE was recently used to study D-Glc utilization changes under the glucose deprivation treatment (mTOR inhibitor) in a mouse glioblastoma tumor model. With minimal intracellular contribution due to the inhibition of hexokinase activity by mTOR, the DGE contrast increased under the treatment, which supports the notion that DGE detects the accumulation of glucose in the vascular and EES compartments (Fig. 1B).

Dextran is a clinically used natural polysaccharide that is a polymeric form of glucose. It is available in multiple molecular sizes (5.9 nm – 77.0 nm),³² and each dextran particle has a high number of hydroxyl protons. Liu et al. have demonstrated that dextrans of molecular weight ranging from 10 kD to 2000 kD generate CEST contrast at 0.9 ppm.³² The Z-spectrum of dextran also showed a different T₂ contribution compared to D-glucose solution, which could be attributed to the different exchange environment in forms of particles. Dextrans of different sizes have been used to characterize the vascular permeability of different tumor regions (Figs. 6A–C),³² by injecting a low-molecular weight (10 kD) dextran first (t = 0) and a heavier (70 kD) after clearance of the first (t = 40 min). The 70 kD agent occupied a tumor area similar to a Gd-enhanced study of the same rat, while the 10 kD agent showed penetration over a larger area. In another study,³⁴ this high molecular weight dextran was used to monitor vascular disruption therapy. The same group used dextran with an added prostate specific membrane antigen (PSMA) targeting ligand to show promise in identifying tumors with high PSMA expression (PSMA⁺, PC3-PIP cells) through selective retention as a function of time.³¹ This is illustrated in Figs. 6D,E, where the contrast in PSMA overexpressed and control (PSMA⁻, PC3-flu cells) tumors initially had comparable contrast (Fig. 6E), which changed to only PSMA+ tumors being highlighted at longer time points.

Fig. 6. Characterization of vessel permeability using multiple dextran sizes (A-C) and PSMA targeted dextran CEST MRI (D,E).

(A) Anatomical T2w images and dynamic dexCEST MRI showing distinctive tumor uptake of 10 kD and 70 kD dextran (administered at t = 0 and t = 40 min, resp) in terms of spatial distribution and pharmacokinetics. (B) Zoomed view of a Gd contrast-enhanced image, permeability-sensitive (K^trans) map, and corresponding ΔMTR_asym maps at 17 and 79 minutes, respectively, in the same rat. (C) Dynamic change of the dexCEST signal in the two ROIs (marked in zoomed view of T2w image on the left). (D) Urea conjugated dextran for PSMA targeting Changes in the dynamic CEST signal in PSMA+ and PSMA– tumors. T2-weighted (T2w) image (left) and dynamic CEST maps (right) at 1 ppm after the injection of 375 mg per kg urea–Dex10 (injection volume = 100 μl). (E) Mean changes in the CEST signal in PSMA+ and PSMA– tumors in one of the mice for which time dependence was measured (left). The error bars are the standard errors of the CEST signal of all the pixels in each tumor. All CEST images were acquired using a 1.8 μ T and continuous-wave pulses that were 3 s long. Average CEST signal in the tumor for five mice before and 1 h after the injection of urea–Dex10 (middle). Error bars are standard deviations of the CEST signal of all five tumors. Column scatter plots (right) show the mean changes in CEST signal as quantified by ΔMTR_asym (1 h) in each tumor type (n = 5 and n = 3 for urea–Dex10 and non-targeted Dex10, respectively). *P < 0.05 (Student’s t-test two-tailed and unpaired) Figure parts reproduced with permission from Li et al. Magn Reson Med 2018;79(2):1001–1009.³²; Liu et al. Nat Biomed Eng 2017;12(1);977–982.³¹;

Multiple sites have implemented DGE MRI for assessing brain tumor patients at 3T and 7T, with both glucoCEST^45,53 and glucoCESL^{46,47,52,64,66} being applied. Interestingly, DGE maps showed partly similar enhancement in tumors when compared with DCE-MRI using GBCAs. Due to the difference in the compartmental contributions in DGE vs DCE, other than BBB permeability, DGE may provide useful and unique information to indicate key events in tumors, such as altered glucose uptake, perfusion and metabolism, for cancer diagnosis/staging, follow-ups and monitoring treatment response. In the consideration of compartmental contrast and cancer pathology, imaging sugars enables the assessment of glucose uptake and utilization in both tumor regions and normal brain regions,^{19,34,44–46,49,52,54,64,66,150,163} where glucose is transported and metabolized. FDG-PET mainly detects the elevated glucose uptake in tumors. GBCA DCE-MRI detects the BBB breakdown in tumors. The unique information provided by the DGE MRI uptake signal in terms of the delivery, transport and metabolism of D-Glc may facilitate the assessment of different types of cancers, especially in non-Gadolinium-enhancing brain tumors.

5.2. Alzheimer’s disease

Alzheimer’s disease (AD) is a progressive neurodegenerative disease that leads to neuronal cell death and brain atrophy. Early pathology includes glymphatic dysfunction, hypometabolism and hypoperfusion.^83,164–168 Plaques will then be formed by two well-known mechanisms related to beta amyloid and tau proteins.¹⁶⁹ Moreover, reduced glymphatic clearance was found to associate with the formation of plaques in AD.^170–172

CSF forms a major part of the glymphatic system (Fig. 7A), Huang et al. and Chen et al. recently developed a DGE method using onVDMP to detect glucose in both the CSF and parenchyma^79,80 to facilitate the assessment of glymphatic system in AD. Huang et al., using a mouse model of amyloid plaques (APP/PS1) at 3T, reported a three-fold decrease in the glucose clearance (μ_out) in CSF of young (6 months old) AD mice compared to age-matched wide-type mice¹⁰⁰ (Fig. 7B), while little or no observable plaques were found in the brain. This reduced clearance was also observed in the cohorts of old mice (16 months old, Fig. 7B). Moreover, the glucose uptake maximum signal (S_max), which is the convolved signal increase due to effects of delivery, transport and metabolism, was lower in old AD mice compared to young AD mice (Fig. 7B). In the parenchyma, an increase in the rate of glucose uptake (μ_in) was observed in old AD mice (Fig. 7C) compared to young AD mice. Interestingly, this was not observed in the WT cohorts, which may provide essential parameters for the identification of AD from normal aging. Another point to note is that the measured μ_in of WT parenchyma from this study was comparable with the D-Glc uptake rate measured by van Zijl et al. in the cat brain using ¹³C MRS.⁶⁸ These results indicate that DGE MRI generates multiple parameters to detect changes of both glucose uptake and clearance in AD brain with early pathology, which has the potential to facilitate the identification of AD from normal aging.

Fig. 7. DGE MRI of Alzheimer’s disease.

(A) Schematic diagram of the glucose transportation in the brain. Glucose from the artery enters parenchyma brain and CSF through glucose transporters in the periarterial space, including glucose transporters in blood-CSF barrier (BCSFB) and BBB. A large portion of CSF recirculates to the parenchyma brain and finally drains interstitial fluid (ISF) clearance. μ_in, uptake rate; μ_out, clearance rate. (B) Experimental (solid line) and fitted (dashed line) parenchymal DGE curves for WT (6 months, n = 5; 16 months, n = 5) and APP/PS1 (6 months, n = 5; 16 months, n = 5) mice. (C) Experimental (solid line) and fitted (dashed line) CSF DGE curves for WT (6 months, n = 5; 16 months, n = 5) and APP/PS1 (6 months, n = 5; 16 months, n = 5) mice. (D) Dynamic AUC maps for WT and Tau mice over the 27.5 min DGE experiment, in both parenchyma and CSF. The brain tissue maps were recorded using the onVDMP sequence with 12 binomial pulse pairs (parenchyma), while the CSF maps were obtained with 100 binomial pulse pairs (CSF). Each dynamic AUC map was calculated by averaging 10 successive DGE images, which lead to a time window of 150 s. The averaged time-resolved DGE curves for the (E) cortex and (F) CSF of WT and Tau mice. Figure parts reproduced with permission from Huang et al. Sci Adv 2020;6(20):eaba3884⁷⁹ and Chen et al. J Cereb Blood Flow Metab 2021;41(5):1013–1025.⁸⁰

Chen et al., using a mouse model with tauopathy at 7–8 months old (Fig. 7D), by fitting the initial part of the uptake curves to a simple exponential (DGE(t) = S_max[1−exp{−rate·t}])), observed a reduction in both the amount (S_max) and rate of glucose uptake in parenchyma (Fig. 7E) and in CSF (Fig. 7F). These two studies indicate that imaging CSF and parenchyma with DGE MRI enables the assessment of glymphatic function in AD. In addition to the above studies using D-Glc,^79,80 Tolomeo et al. reported that a reduction of 2-DG uptake in an amyloid plaque AD mouse model was detected by DGE.⁸³ As mentioned in section 2, some studies have found a competitive inhibition of glycolysis by 2-DG, thus these reported CEST contrast changes in AD should be distinctive from cancers.^84–87

These early DGE data for the glymphatic system show promise for the possibility to identify early AD pathology, such as a slow-down of D-Glc clearance in the CSF. D-Glc or 2-DG CEST could assess changes in the transport and clearance kinetics of these sugars in vivo potentially related to hypometabolism and hypoperfusion in AD. Conventional DCE-MRI has been applied to study the glymphatic system, however the use of GBCAs with intrathecal administration is a concern, in particular in renal compromised and vulnerable populations.¹ Thus, CEST imaging of sugars in the brain may provide unique information for early AD diagnosis.

5.3. Other applications (Placenta, kidneys)

Wu et al. applied DGE MRI to assess mouse placental functioning related to D-glucose utilization.¹⁵³ A decrease in glucoCEST contrast was observed for inflamed placenta, possibly due to the decrease in GLUT activity. This could provide a non-invasive approach to study placental disorders, by assessing the glucose uptake of the placental tissue or the fetal-maternal exchange. Another study by Luo et al. reported the feasibility of using DGE MRI to map D-glucose in human placenta using MTR_asym.⁶⁵ Some issues remain to be addressed for DGE in placenta, such as motion.

Another interesting study is to identify acute renal allograft rejection using glucoCEST.¹⁵² GlucoCEST contrast in MTR_asym increased during acute rejection, which could be attributed to the increase in cellular uptake and vascular permeability. This allowed the identification of rejection with high sensitivity and specificity in a rat model.

In summary, several promising applications of glucoCEST have been demonstrated in both preclinical and clinical settings, especially in assessing abnormality of D-glucose-related events. In order to apply glucoCEST for diagnosis and monitoring treatment, challenges, such as a lack of kinetic models for glucose uptake and utilization, motion artifacts and choice of B₀, need to be addressed to bring glucoCEST from initial research studies to wider clinical applications.

6. CHALLENGES AND OPPORTUNITIES

6.1. Physiological Challenges

Like all other contrast agents, there is also a concentration dependency of the CEST effect since the number of solute protons is directly proportional to the effect size.¹⁵ Currently, there is no consensus which D-Glc concentration to use in human experiments. For example, some investigators have used a bolus injection dose of 50mL of 50% w/v dextrose^45,48,53,54, others 100 mL of 20%w/v dextrose.^46,47 Instead of a bolus injection, Kim et al. used 20% dextrose with a clamp infusion where the blood glucose level was raised to a stabilized hyperglycemic level.⁵⁰ In this particular study, the approach required up to 12 min to reach the targeted plasma glucose level, which is time consuming compared to the other studies where the infusion durations ranged from 30 s to 4 min.

Depending on the D-Glc concentration by weight and the total infusion amount, some physiological issues may occur. One potential problem is the osmolarity effect induced by high concentration D-Glc solutions, which may affect the observed signal and also cause trauma to the vein and result in thrombophlebitis when infused rapidly.^50,54,173 Seidemo et al.⁵⁴ investigated two different D-Glc infusion durations, 1.5 min and 4 min, and concluded that longer infusion duration was preferred because it minimized the side effects while still producing a sufficient change in blood glucose levels. Not only can the thrombophlebitis risk be minimized with longer infusion duration, but the participants in this study also experienced reduced sensory effects when using the longer infusion. Infusion sensory effects that can be felt during D-Glc infusion include a sugary taste in the mouth, warmth, or pulsating feeling at the injection site and in the head and crotch, as well as a need to urinate.

The fasting blood glucose level within the normal range is 70 mg/dL (3.9 mM) to 125 mg/dL (7 mM),¹⁷⁴ which we have used as baseline range for our experiments^45,174 prior to D-Glc infusion. However, some medications such as steroids in tumor patients can cause elevated blood glucose levels, which will prevent the subjects from being eligible. Patients with diabetes mellitus, sickle cell disease, and low blood iron levels are also currently excluded from glucoCEST experiments. The exclusion of these patient groups will hamper the general applicability of glucoCEST and more research is needed to evaluate if some of these patient groups can be included in the future.

6.2. Technical Challenges

Even though several studies have successfully shown a DGE signal change when D-Glc or its analogues have been infused, a major challenge is the rather small effect size in humans. Xu et al. reported a D-Glc based DGE effect of 2–6% in the Gd-enhancing region of a glioma rim at 7T,⁴⁵ which reduced to 0.5–1.5% at 3T.⁵³ The main reason for the difference in effect size with field strength is that the hydroxyl protons exchange rapidly with the surrounding water protons (k_ex of approximately 1.5–15 kHz).⁵¹ As discussed in the technical section, when going to lower field, the exchange regime moves from slow-intermediate to intermediate-fast on the MR timescale, leading to increased coalescence with the water resonance²⁷ and a reduction in detectability of the hydroxyl CEST effect (Fig. 4).⁵⁰ While higher field strengths are preferred due to the increased separation between the hydroxyl protons and water protons, ultra-high human field human scanners (≥7T) come with other problems, such as increased inhomogeneity in the B₀ field and the RF transmit field (B₁₊). Even though B₀ and B₁⁺ shimming methods have been developed to mitigate these effects,^175–177 in order to have impact, the methods ultimately need to be available at clinical field strengths (i.e. 3T and lower).

Partial volume effects (PVE) are common in MR images due to the limited resolution used to achieve adequate SNR. In DGE the voxel volume usually ranges to 14 to 48 mm³ depending on pulse sequence, field strength and target volume which leads to different amounts of PVE. A major problem with PVE in DGE, that is not related to the usual mixing of different tissues, is that the glucose infusion also leads to some ventricular swelling. This swelling, that is in the order of 1–2%,⁵⁴ can cause additional CSF to be mixed with tissue during the acquisition, thus creating erroneous hyperintensities due to the large D-Glc concentration in CSF or hyper/hypointensities due to CSF pulsation.

Image hyper- and hypointensities can also occur due to patient movement that causes local B₀ changes in tissue and concomitant shifts of the resonance frequency. Zaiss et al. demonstrated that, when using difference images obtained at a single frequency, motion and B₀ shifts can signal changes on the order of ±1%,¹⁷⁸ the same effect size as reported from DGE studies at 3T (Figures 8A–F). The hyperintensities, caused by a field shift, can therefore be mistaken as real CEST effects. These artifactual CEST effects have been dubbed as “pseudo CEST effects”. Performing standard motion correction algorithms will reduce these pseudo CEST effects,⁵² however, when applying these, there is a risk that true DGE signal can be mistaken for motion and altered erroneously. This warrants for more advanced motion correction post-processing methods or the use of volumetric navigators before the saturation period in the acquisition.¹⁷⁹ Despite the problems with motion, DGE remains a promising technique, as illustrated in Figures 8G–R. These results from Xu et al.⁴⁵ obtained at a 7T scanner in a brain tumor patient (Figure 8G–J) clearly show that the rim around the necrotic region in a tumor enhances rapidly (Figure 8H), providing contrast similar to DCE, whereas the tumor center and surrounding tissue enhance more slowly (Figure 8I,J), providing information not available in DCE. At 3T the issue with motion becomes more imperative due to the smaller CEST effect. However Herz et al⁵² and Xu et al.⁵³ demonstrated that, despite these challenges (Figure 8K–R), a DGE signal increase was still observable at 3T in patients with brain tumors. Additionally, in some cases, the increased DGE signal in the tumor corresponded to the Gd enhanced location, particularly in the later stages of the DGE imaging (Figure 8N,R).

Fig. 8. Examples of human DGE studies.

(A-F) DGE images of a brain tumor patient without glucose injection at 3T show the effect of motion creating so called “pseudo-CEST effects”. The difference maps between pre-motion and post-motion were generated to simulate DGE contrast (A,D) together with B₀ shifts (B,E). Both hyper- and hypointensities are seen in the tumor region in the DGE images, similar to the Gd contrast-enhancement (C). This demonstrate that motion artifacts in DGE in tumor areas can be mistaken as uptake signal. (F) ROI analysis revealed pseudo CEST effect in tumor sub-regions. DGE imaging in a brain tumor patient at 7T. Gd-T1w image (G) and DGE based AUC images (H-J) for different time periods relative to the start of infusion at 7T. The DGE shows a faster uptake in the rim around the necrotic region and with a slower uptake in the tumor center and surrounding tissue. (K - R) DGE imaging in brain tumor patient at 3T. First row: Gd-T1w images (K) and DGE based AUC images (M,N) for different time periods relative to the start of infusion. The increased DGE signal corresponds with the enhanced Gd location, particularly toward the later phase of the DGE imaging. Second row: T2 FLAIR (O), Gd-T1w (P), Gd-T1w with overlaid ΔDGEρ map (Q) and the ΔDGEρ map (R). These early images support potential for DGE MRI if higher SNR can be achieved. Figure parts reproduced with permission from Zaiss M, et al. J Magn Reson. Jan 2019;298:16–22;¹⁸⁹ Xu et al. Tomography. 2015;1(2):105–114;⁴⁵; Xu et al. Magn Reson Med. 2020;84(1):247–262.⁵³; Herz et al. Magn Reson Med. 2019; 82(5):1832–1847⁵²

Another challenging issue before DGE MRI can be used clinically is the acquisition time. Papers have reported scan times ranging from 15 to 60 min with temporal resolutions ranging from 5.3 to 42 seconds.^{45–48,50,53,54,66} Not only are longer scan times more prone to motion artefacts, but they can also make the patient feel more uncomfortable and anxious. Faster acquisition techniques such as simultaneous multi-slice imaging^180,181 and compressed sensing^182,183 that reduce the scan duration are available. However, reaching typical MRI scans of 5 minutes is not possible when a long baseline is needed (previous literature has reported baseline times of 3 minutes) for the purpose of SNR and when the infusion duration is recommended to be 4 minutes for the patient to be comfortable physiologically due to the amounts of sugar needed.

6.3. Opportunities

Based on the above considerations, increased SNR/CNR is the most important opportunity to be pursued for the field to be moved forward. This is in principle possible through the design of new pulse sequences with increased sensitivity, such as for instance pulsed approaches (e.g. the onVDMP method) in which a larger fraction (broader chemical shift range) of fast-exchanging protons is labelled rapidly and repeatedly (i.e. after exchange). In addition, recent papers indicate that reduced amounts of GBCAs can be used with properly trained artificial intelligence approaches^184,185, which one would think should also be possible for DGE.

The simplest approach for increasing effect size is increasing the amount of sugar infused, but that is very limited. More promising, the use of alternative sugars with different physiological properties provides promising opportunities. Currently, only D-Glc has been used in human glucoCEST experiments. However, there is a large interest in other sugars such as 3-OMG and D-GlcN that accumulate in the tumor. It has for instance been shown in studies by Rivlin et al. 2014 and Sehgal et al. 2019,^57,59 that, when used at the same concentration, 3-OMG induces a CEST contrast effect that is larger than that of D-Glc with up to twice the contrast enhancement in tumor tissue.⁵⁹ Although no toxicology studies of 3-OMG have been reported, mice or rats given various doses of 3-OMG have not shown any physiological or behavioral changes.¹⁸⁶ Since 3-OMG has been used to study intestinal permeability in humans^104,105 it is probable that 3-OMG can be translated into the clinic as a glucoCEST agent if suitable concentrations can be used. Very recently, the first human studies of D-GlcN demonstrated the possibility to highlight breast tumors.¹⁰¹

7. CONCLUSIONS

CEST imaging of D-Glc and other sugars in principle provides a minimally-invasive way to detect their related delivery, transport and metabolism in vivo, without the need of radioactive or metallic labeling. Recent compartmental studies based on glucose-related physiology and pathology help to elucidate the potential contributions to the glucoCEST signal, but the availability of a single uptake curve complicates interpretation. Despite this, several animal studies are showing the possibility to detect alterations of sugar transport/clearance and/or uptake in cancers, Alzheimer’s disease, placenta and during kidney allograft rejection. Also, several small sugars such as 3-OMG and D-GlcN may be available for human studies soon. Remaining challenges are many but expected to be addressable when methodological advances become available. We are therefore cautiously optimistic that such advances in MRI detection and data analysis as well as availability of more sugar analogues will allow this young field to steadily move forward in the coming years.

ABBREVIATIONS

(c)

Intracellular

2-DG

2-Deoxy-D-glucose

3-OMG

3-orthomethyl-D-glucose

AD

Alzheimer’s disease

AIF

Arterial input function

AUC

Area under the curve

BBB

Blood brain barrier

BOLD

blood oxygen level-dependent

CESL

Chemical exchange sensitive spin-lock

CEST

Chemical exchange saturation transfer

CPMG

Carr-Purcell-Meiboom-Gill

CSF

Cerebral spinal fluid

CT

Computed tomography

CW

Continuous wave

DCE

Dynamic contrast enhanced

DGE

Dynamic glucose enhanced

D-Glc

D-glucose

D-GlcN

D-glucosamine (2-amino-2-deoxy-D-glucose)

D-Xyl

D-Xylose

EES, (e)

Extravascular and extracellular space

FDG

Fluorodeoxyglucose

GBCA

Gadolinium based contrast agent

Gd

Gadolinium

GdDTPA

Gadolinium diethylenetriamine penta-acetic acid

GlcNAc

N-acetyl-D-glucosamine

glucoCESL

glucose CESL

glucoCEST

glucose CEST

GLUT

Glucose transporter

ISF

Interstitial fluid

i.v.

Intravenous

k_ex

Exchange rate

K^trans

Volume transfer constant

L-Glc

L-glucose

Mls

Maltose

Mlt

Maltitol

mTOR

Mechanistic target of rapamycin kinase

MTR_asym

Magnetization transfer ratio with asymmetric analysis

onVDMP

on-resonance variable delay multi-pulse

PET

Positron emission tomography

p.o.

per os

PSMA

Prostate specific membrane antigen

PVE

Partial volume effect

RF

Radiofrequency

SAR

Specific absorption rate

S_base

Average baseline signal

Slc

Sucralose

TCA

Tricarboxylic acid

TSL

Time of Spin-lock (duration)

(v)

Vascular

WT

Wild-type