Simultaneous pH- and oxygen-sensitive MRI of human gliomas at 3T using multi-echo amine proton chemical exchange saturation transfer spin-and-gradient echo echoplanar imaging (CEST-SAGE-EPI)

Abstract

Purpose

To introduce a new pH- and oxygen-sensitive MR imaging technique using amine proton chemical exchange saturation transfer echo spin-and-gradient echo echoplanar imaging (CEST-SAGE-EPI).

Methods

pH-weighting was obtained using CEST estimations of magnetization transfer ratio asymmetry (MTR_asym) at 3ppm and oxygen-weighting was obtained using R₂′ measurements. Glutamine concentration, pH, and relaxation rates were varied in phantoms to validate simulations and estimate relaxation rates. MTR_asym and R₂′ in normal appearing white matter (NAWM), T₂ hyperintensity, contrast enhancement (CE), and macroscopic necrosis were measured in 47 gliomas.

Results

Simulation and phantom results confirmed an increase in MTR_asym with decreasing pH. CEST-SAGE-EPI estimates of R₂, R₂^*, and R₂′ varied linearly with Gd-DTPA concentration (r₂=6.2 mM⁻¹·sec⁻¹, and r₂^*=6.9 mM⁻¹·sec⁻¹). CEST-SAGE-EPI and CPMG estimates of R₂ (R²=0.9943) and multi-echo GRE estimates of R₂^*(R²=0.9727) were highly correlated. T₂ lesions had lower R₂′ and higher MTR_asym compared with NAWM, suggesting lower hypoxia and high acidity, whereas CE tumor regions had elevated R₂′ and MTR_asym indicating high hypoxia and acidity.

Conclusion

CEST-SAGE-EPI provides simultaneous pH- and oxygen-sensitive image contrasts for evaluation of the brain tumor microenvironment. Advantages include fast whole brain acquisition, in-line B₀ correction, and simultaneous estimation of CEST effects, R₂, R₂*, and R₂′ at 3T.

INTRODUCTION

Abnormal metabolism is a hallmark of cancer (1,2). Notably, glycolysis is often enhanced in cancers, even in the presence of abundant oxygen (i.e. the Warburg effect (2), Fig 1). This form of aerobic glycolysis results in a dramatic decrease in extracellular pH due to increased concentration of lactic acid (1). This increase in extracellular acidity comes with dramatic consequences, as it can be directly linked to the degree of tumor aggressiveness (3,4) and increases tumor invasion (5,6). Interestingly, histological evidence suggests regions containing pseudopalisades, a pathological trademark of glioblastoma, are also hypoxic, express extracellular matrix proteases, and are the result of active tumor migration (7). Increased acidity within the tumor also has been shown to lead to decreased immune function (8). The acidic microenvironment in tumors is also conducive to elevated VEGF expression (9) and expression of platelet-derived endothelial cell growth factor (10), which has been shown to result in increased angiogenesis. This in turn leads to a positive feedback process leading to further tumor growth, decreasing oxygen tension, increased hypoxia, increasing glycolysis resulting in increased lactic acid, decreasing extracellular tissue pH, and more mutations and/or adaptations of the tumor genome. Thus, extensive in vitro, preclinical, and clinical evidence appears to support the hypothesis that tumor acidity and oxygen metabolism both play a critical role in gliomagenesis. There remains, however, a critical need in furthering our understanding of the role of extracellular acidosis and oxygen metabolism in human gliomas and its clinical relevance due to the lack of a robust non-invasive tool for simultaneously estimating and localizing regions of low pH and oxygen consumption. Thus, the purpose of the study was to develop a new, non-invasive MR imaging technique that can provide high-resolution images with sensitivity to tumor acidity and oxygen metabolism that can be performed on clinical MRI systems.

Fig. 1. The Warburg Effect.

A) Within normal tissues in the presence of oxygen, glucose is converted to pyruvate and then used for oxydative phosphorylation within the mitochondria. B) In normal tissues within a hypoxic environment, glucose is converted to pyruvate then to lactate or lactic acid, decreasing extracellular pH. C) In cancer cells, glucose is converted to pyruvate then to lactate and lactic acid (80–85%) and a small portion of pyruvate enters the citric acid cycle (5–15%), regardless of whether oxygen is present. This altered metabolism, also known as the Warburg Effect or aerobic glycolysis, results in increased extracellular acidity (lower pH) even when tumor tissue is well perfused.

pH-Weighted Metabolic MRI using CEST Contrast from Fast Exchangeable Amine Protons on Glutamine

In addition to glucose, glutamine is also a major source of fuel for malignant tumors (11). It is essential for cellular proliferation, tumor growth, and tumor cell survival. Glutamine is the most abundant amino acid in the body, circulating at concentrations of 0.6–0.9 mM and as high as 20 mM in tissue (11). Tumor cells often consume a significant amount of glutamine (12), acting at times like a “glutamine trap”. Glutamine demand is so high that transport systems are amplified to increase glutamine consumption (13). Glutamine has two nitrogen functional groups, an amine and an amide group, having hydrogen NMR resonance frequencies of 3.0 ppm and 3.5 ppm, respectively, compared to water protons. The chemical exchange between amine and amide protons in bulk water has been shown to be pH dependent using a new imaging method called chemical exchange saturation transfer (CEST) imaging (14). The inherently elevated concentration of glutamine within tumors further increases the available proton exchange, resulting in a higher CEST signal at 3.0ppm (15,16). The combination of increased protons (low pH) and increased glutamine makes pH-weighted MRI using CEST contrast from amine protons on glutamine particularly attractive as a non-invasive tool for assessment of microenvironment acidity within malignant brain tumors.

R₂′-Based Blood Oxygenation Level Dependent (BOLD) Imaging

The level of glycolysis depends on both the accumulation of lactic acid as well as the inefficient use of oxygen. Thus, image maps sensitive to tumor oxygen metabolism are critically important for understanding the relative level of aerobic (pathologic) versus anaerobic (normal) glycolysis. Blood oxygen level dependent (BOLD) imaging, based on the contrast mechanisms routinely used for functional imaging (fMRI), allow for non-invasive estimation of blood or tissue oxygenation. Simply stated, oxygenated blood containing oxyhemoglobin is diamagnetic, as the iron at the core is magnetically shielded from blood water, resulting in coherent MR spins and a high MR signal. Early studies from Ogawa et al.(17) confirmed this in mice breathing high concentrations of oxygen. In deoxygenated blood, the iron is exposed to blood water resulting in magnetic interference with the proton magnetic moment on water molecules. This interference results in incoherent MR spins and signal dropout in the areas of high deoxyhemoglobin concentration. The paramagnetic nature of deoxyhemoglobin enhances the effective transverse relaxation rate R₂*. Changes in tissue R₂* therefore reflect relative changes in concentration of deoxyhemoglobin, and thus an indirect measure of oxygen extraction fraction (OEF). Use of the reversible transverse relaxation rate, R₂′ = R₂* − R₂, both isolates the local susceptibility effect while compensating for R₂ changes from factors such as water content variation (18). The sensitivity of R₂′ to concentration of deoxyhemoglobin has been reported by several studies (19,20) and has been shown to correlate with the hypoxic state of tissue (21). While R₂′ does not allow for a direct measurement of OEF, previous studies have demonstrated that OEF is proportional to R₂′, after normalization to relative blood volume fraction and static dephasing expected for protons in blood at the specific magnetic field strength (22–25). Additionally, this approach has been previously used to explore oxygen metabolism in brain tumors (24,26) as well as stroke (27). It is important to note, however, that while higher measures of R₂′ may suggest higher concentration of deoxyhemoglobin, OEF, and/or local hypoxia, many other biological and/or technical influences (e.g. B₁ homogeneity, etc.) may influence this measurement.

In the current study, we introduce a new technique for obtaining fast, whole brain, non-invasive, high-resolution pH- and oxygen-sensitive MR imaging contrast using multi-echo amine proton chemical exchange saturation transfer echo spin-and-gradient echo echoplanar imaging (CEST-SAGE-EPI) on a clinical 3T MRI system. pH-weighted image sensitivity was obtained through quantification of the z-spectral asymmetry in the magnetic transfer ratio after selective saturation of the longitudinal magnetization of amine protons on glutamine at 3.0ppm (MTR_asym @ 3.0ppm), while oxygen sensitivity was obtained through quantification of R₂′ using a multi-echo spin-and-gradient-echo (SAGE) EPI readout to quantify relaxation rates R₂ and R₂^*. Sensitivity and accuracy of this approach were confirmed using custom phantoms containing both gadolinium DTPA and glutamine and different concentrations and pH. Lastly, both MTR_asym @ 3.0ppm and R₂′ for tumor and normal tissue were characterized in patients harboring diffuse or malignant gliomas.

THEORY

pH- and oxygen-sensitive MR contrast using multi-echo amine proton chemical exchange saturation transfer spin-and-gradient echo echoplanar imaging (CEST-SAGE-EPI) (Fig 2)

Fig. 2. Pulse sequence diagram for a multi-echo amine chemical exchange saturation transfer spin-and-gradient echo echoplanar imaging (CEST-SAGE-EPI) sequence.

After non-selective off-resonance CEST excitation consisting of three 100-ms Gaussian pulses, a spectral-spatial water only excitation pulse is invoked. Following excitation two gradient echo EPI trains are acquired (Echoes #1 and #2), followed by a 180-degree refocusing pulse. Lastly, an asymmetric spin echo EPI train (Echo #3) and a standard spin-echo EPI train (Echo #4) are acquired. Partial Fourier k-space acquisition strategies were used to further reduce echo times (e.g. TE₁/TE₂/TE₃/TE₄ = 14.0/34.1/58.0/92.4 ms for the current study).

The magnetization of bulk water protons undergoing chemical exchange with amine protons can be described by the Bloch-McConnell equations (28) in the form of:

$$\frac{d\mathit{M}(t)}{dt}=\mathit{X}·\mathit{M}(t)-\mathit{c}$$ [Eq. 1]

where

$$\mathit{M}=\left(\begin{array}{c}{M}_{ax}\\ M_{bx}\\ M_{ay}\\ M_{by}\\ M_{az}\\ M_{bz}\end{array}\right),\mathit{X}=\left(\begin{array}{cccccc}{C}_{2a}& k_b& -\delta a& 0& 0& 0\\ k_a& C_{2b}& 0& -\delta b& 0& 0\\ \delta a& 0& C_{2a}& k_b& -{\omega }_1& 0\\ 0& \delta b& k_a& C_{2b}& 0& -{\omega }_1\\ 0& 0& {\omega }_1& 0& C_{1a}& k_b\\ 0& 0& 0& {\omega }_1& k_a& C_{1b}\end{array}\right),\mathit{c}=\left(\begin{array}{c}0\\ 0\\ 0\\ 0\\ M_{az0}/T_{1a}\\ M_{bz0}/T_{1b}\end{array}\right);$$ [Eq. 2]

pool a and pool b are the bulk water protons and amine protons, respectively; M_az0 and M_bz0 are the equilibrium magnetizations of pool a and b, respectively; k_b is the exchange rate of protons from pool b to pool a; k_a is the exchange rate of protons from pool a to pool b as given by (M_b0/M_a0)·k_b; ω₁ is the RF pulse amplitude as given by ω₁ = γB₁(t), where γ is the gyromagnetic ratio and B₁(t) is given in μT; δa = ω−ω_a and δb = ω −ω_b, where ω is the applied RF irradiation frequency, ω_a is the bulk water resonance frequency, and ω_b is the amine proton resonance frequency; T_1a and T_1b are the longitudinal relaxation times of pool a and b, respectively; and C_1a = (1/T_1a)+k_a, C_2a = (1/T_2a)+k_a, C_1b = (1/T_1b)+k_b, C_2b = (1/T_2b)+k_b represent the sum of exchange and relaxation rates. Eq. 1 can be solved analytically to yield

$$\mathit{M}(t)=e^{\mathit{X}t}·{\mathit{M}}_{\mathbf{0}}(\mathit{X}/\mathit{c})-(\mathit{X}/\mathit{c})$$ [Eq. 3]

where M_az(t) represents the longitudinal magnetization of bulk water available for subsequent readout after CEST effects. Assuming the spoiler duration and water excitation pulse duration are negligible, M_az(t) immediately following excitation reflects the available longitudinal magnetization for subsequent readout. The chemical exchange between amine protons and bulk water protons can be characterized as a base-catalyzed process, governed by pH:

$$k_b=k_0+k_{\mathit{base}}\ast {10}^{-(14-pH)}$$ [Eq. 4]

where k₀ is the default exchange rate, k_base is the base-catalyzed rate constant, and k_b is the exchange rate of protons from the metabolite proton pool to the water pool (29). In glutamine phantoms with concentration of 50 mM, k₀ and k_base have been estimated to be 76 Hz and 5.6 Hz, respectively (16).

During the CEST imaging experiment, the RF saturation frequency ω is swept over a range of values to obtain a z-spectral dataset (30). To reduce the effects of T₁ and T₂ weighting, along with other confounding effects, the attenuation of bulk water magnetization following a saturation pulse is described by the magnetization transfer ratio (MTR), given by:

$$\mathit{MTR}(\omega )=\frac{S(\omega )}{S_0}$$ [Eq. 5]

where S(ω) is the amount of bulk water signal available after the saturation pulse with frequency ω and S₀ is the signal available without application of RF saturation. Since MTR can be affected by symmetric effects of direct water saturation and conventional magnetization transfer effects, CEST contrast is often described by the asymmetry in the magnetization transfer ratio (MTR_asym) with respect to water proton resonance:

$${\mathit{MTR}}_{\mathit{asym}}(\omega )=\frac{S(-\omega )-S(\omega )}{S_0}$$ [Eq. 6]

For amine proton CEST imaging, MTR_asym is evaluated at 3.0ppm with respect to the bulk water resonance frequency.

Estimates of R₂^′ were achieved through use of a SAGE-EPI readout consisting of two gradient echoes (TE₁ and TE₂), an asymmetric spin echo (TE₃), and a spin echo (TE₄) EPI acquisition during a single excitation event (Fig 2). The solutions for R₂^* and R₂ using the SAGE EPI were eloquently described by Schmiedeskamp et al. (31) as:

$$\mathit{A}={\mathit{Y}}^{-\mathbf{1}}\mathit{S}$$ [Eq. 7]

where

$$\mathit{S}=\left(\begin{array}{c}ln(S_1)\\ ln(S_2)\\ ln(S_3)\\ ln(S_4)\end{array}\right),\mathit{Y}=\left(\begin{array}{cccc}1& 0& -{TE}_1& 0\\ 1& 0& -{TE}_2& 0\\ 1& -1& -{TE}_4+{TE}_3& {TE}_4-2·{TE}_3\\ 1& -1& 0& -{TE}_4\end{array}\right),\mathit{A}=\left(\begin{array}{c}ln(S_0)\\ ln(\delta )\\ R_2^{\ast }\\ R_2\end{array}\right)$$ [Eq. 8]

where S_n is signal magnitude for the n^th echo and δ is the differences in residual signal differences caused by imperfectly matched slice profiles between echo trains before and after the refocusing pulse (32). Lastly, the reversible transverse relaxation rate, R₂^′, is calculated as:

$$R_2^{\prime }=R_2^{\ast }-R_2$$ [Eq. 9]

METHODS

Amine Proton Chemical Exchange Saturation Transfer Spin-and-Gradient Echo Echoplanar Imaging (CEST-SAGE-EPI)

Simultaneous acquisition of pH-sensitive information and relaxometry measures of R₂′ were performed through modification of a previously described CEST EPI sequence (16) to include a SAGE-EPI readout (Fig 2). The SAGE-EPI readout consisted of two gradient echoes (TE₁=14.0ms;TE₂=34.1ms), an asymmetric spin echo (TE₃=58.0ms), and a spin echo (TE₄=92.4ms). All phantom and human CEST-SAGE-EPI data were acquired with a CEST saturation pulse train consisting of three (3×) 100-ms Gaussian pulses with amplitude B₁=6μT, TR>10,000ms, FOV=240×217, matrix size=128×104, partial Fourier encoding =6/8, GRAPPA=3, bandwidth=1630 Hz/pixel, and 25 contiguous slices with a 4mm slice thickness. A total of 29 z-spectral points was acquired with data around +/− 3.0ppm and 0.0ppm with respect to water (from −3.5 to −2.5 in intervals of 0.1; from −0.3 to +0.3 in intervals of 0.1; and from +2.5 to +3.5 in intervals of 0.1). An additional reference S₀ scan with identical parameters and no saturation pulse was acquired with NEX=4. The total acquisition time for CEST-SAGE-EPI was 7 minutes and 30 seconds benchmarked on a 3T Siemens Prisma MR scanner (Software Versions VE11A-C; Siemens Healthcare; Erlangen, Germany).

Phantom Testing

To demonstrate the ability to simultaneously acquire pH-sensitive information along with relaxometry measures of R₂, R₂*, and R₂′, we performed CEST-SAGE-EPI, multi-echo gradient echo (ME-GRE), and multi-echo spin echo (Carr-Purcell-Meiboom-Gill, CPMG) MR acquisition in a series of 36 glutamine phantoms (100mM) with combinations of varying pH (5.0, 5.5, 6.0, 6.5, 7.0, 7.5) and gadopentetic acid (Gd-DTPA; Magnevist®) concentration (0, 0.25, 0.5, 1.0, 1.5, and 2.0mM) in 50mL falcon tubes. The 100mM glutamine solution was prepared first using distilled water. The pH in each phantom was titrated using dilute acid (0.1N HCl) and base (0.1N NaOH) solution. The desired Gd-DTPA was then added to the phantom solution and vortexed. All samples were vortexed and pH was re-evaluated prior to MRI acquisition. The ME-GRE sequence used for R₂^* quantitation was collected with TE = 20/40/60/80ms, TR=10000ms, flip angle=90°, FOV=217×240mm, matrix size=116×128, slice thickness=4mm, and pixel bandwidth=260Hz. The CPMG sequence used for R₂ quantitation was performed with TE=49/98/147/196/245/294/343/392ms, TR=10000ms, flip angle=90°, FOV=217×240mm, matrix size=116×128, slice thickness=4mm, and pixel bandwidth=260Hz. Both the ME-GRE and CPMG sequences were repeated to improve signal-to-noise ratio (SNR). All phantom experiments were physically repeated twice to ensure repeatability and compared to theoretical values using Bloch-McConnell simulations. Contrast-to-noise ratio (CNR) of MTR_asym at 3ppm between any two samples of differing pH a and b was calculated as

$$\mathit{CNR}=\frac{\mid {\mu }_a-{\mu }_b\mid }{{(({\sigma }_a^2-{\sigma }_b^2)/2)}^2}$$ [Eq. 10]

Patients

A total of 47 histologically proven glioma patients (WHO IV, N=20; WHO III, N=14; WHO II, N=13) were enrolled in the current study prior to initial surgical resection or at first recurrence. All patients provided informed written consent to have advanced imaging and this information included in our IRB-approved research database. Patient characteristics are outlined in Table 1. In addition to CEST-SAGE-EPI prior to contrast administration, all patients received the anatomic images according to the standardized brain tumor imaging protocol (BTIP)(33) including T₂-weighted fluid-attenuated inversion recovery (FLAIR) images, T₂-weighted turbo spin echo images, and diffusion weighted images with 3mm slice thickness and no interslice gap, along with parameter matched, 1mm isotropic 3D T1-weighted magnetization-prepared rapid acquisition gradient echo (MPRAGE) scans pre- and post-injection of 0.01 mg/kg Gd-DTPA.

Table 1.

Patient Demographics.

Patient ID	Age	Sex	2016 WHO Grade	IDH Status	2016 WHO Diagnosis	New or Recurrent
1	60	M	4	WT	GBM	New
2	62	M	4	WT	GBM	New
3	53	F	4	WT	GBM	New
4	49	M	4	WT	GBM	New
5	60	M	4	WT	GBM	New
6	46	F	4	MUT	GBM	Recurrent
7	53	F	4	WT	GBM	Recurrent
8	76	F	4	WT	GBM	Recurrent
9	55	F	4	WT	GBM	Recurrent
10	59	M	4	WT	GBM	Recurrent
11	60	M	4	WT	GBM	Recurrent
12	45	M	4	WT	GBM	Recurrent
13	48	F	4	WT	GBM	Recurrent
14	66	M	4	WT	GBM	Recurrent
15	53	F	4	WT	GBM	Recurrent
16	75	M	4	WT	GBM	Recurrent
17	56	M	4	WT	GBM	Recurrent
18	62	M	4	WT	GBM	Recurrent
19	75	M	4	WT	GBM	Recurrent
20	60	M	4	WT	GBM	Recurrent
21	29	F	3	MUT	AA	New
22	51	M	3	MUT	AA	New
23	64	F	3	WT	AA	New
24	56	M	3	WT	AA	New
25	62	M	3	WT	AA	New
26	69	M	3	WT	AA	New
27	36	F	3	MUT	AO	New
28	29	M	3	MUT	AO	New
29	82	M	3	WT	AA	Recurrent
30	71	F	3	WT	AA	Recurrent
31	47	F	3	WT	AA	Recurrent
32	76	F	3	WT	AA	Recurrent
33	66	M	3	MUT	AO	Recurrent
34	38	F	3	MUT	AO	Recurrent
35	38	M	2	MUT	A	New
36	41	M	2	MUT	A	New
37	29	F	2	MUT	A	New
38	30	F	2	MUT	A	New
39	22	M	2	MUT	A	New
40	48	M	2	WT	A	New
41	87	F	2	WT	A	New
42	41	M	2	MUT	O	New
43	42	F	2	MUT	A	Recurrent
44	36	F	2	MUT	A	Recurrent
45	73	M	2	WT	A	Recurrent
46	41	F	2	MUT	O	Recurrent
47	91	M	2	WT	O	Recurrent

M=Male; F=Female; IDH=isocitrate dehydrogenase; MUT=Mutant; WT=Wild Type; WHO=World Health Organization; A=Astrocytoma; O=Oligodendroglioma; AA=Anaplastic Astrocytoma; AO=Anaplastic Oligodendroglioma; GBM=Glioblastoma.

Post-Processing

Clinical post-processing of CEST-SAGE-EPI consisted of affine motion correction (mcflirt; FSL, FMRIB, Oxford, UK) and B₀ correction via the WASSR method (34) and/or creating B₀ maps using phase information from the two acquired gradient echoes (Supporting Figure S1). An integral of width 0.4 ppm was then taken around both the −3.0 and +3.0 ppm (−3.2 to −2.8 and +2.8 to +3.2, respectively) spectral points of the inhomogeneity-corrected data. These data points were combined with the S₀ image to calculate MTR_asym at 3.0 ppm as defined by Eq. 5. Estimates of T₂*, T₂, T₂′, R₂′, R₂, and R₂′ from ME-GRE or CPMG were obtained by performing a monoexponential fit to the gradient and spin echoes, respectively, whereas estimates of the same parameters were obtained using Eq. 8. Maps of R₂′ were then calculated as

$$R_2^{\prime }=R_2^{\ast }-R_2=\frac{1}{{T_2}^{\prime }}=\frac{1}{T_2^{\ast }}-\frac{1}{T_2}$$ [Eq. 11]

with a higher value of R₂′ suggesting relatively higher concentration of hemoglobin, oxygen extraction fraction, and/or hypoxia (22–26).

Data Analysis

For all data, average MTR_asym at 3.0 ppm calculated from the first (TE=14.0 ms) and second (TE=34.1 ms) gradient echoes were averaged to decrease increase signal-to-noise ratio (SNR) of the resulting MTR_asym images. The two gradient echoes were chosen instead of all echoes due to slightly higher variability in MTR_asym measurements from asymmetric spin-echo and spin-echo measurements due to the longer TE and additional signal loss from transverse relaxation. In phantom samples, regions of interest (ROIs) were drawn within each sample and the mean (μ) and standard deviation (σ) of MTR_asym at 3ppm for voxels within the sample were calculated.

When evaluating glioma patients, volumes of interest (VOIs) were drawn within normal-appearing white matter (NAWM) contralateral to the hemisphere containing evidence of tumor on T₂-weighted FLAIR images. Lesions exhibiting abnormal T₂ hyperintensity on FLAIR images (“T₂ lesions” VOIs) were manually contoured on all patients. To reduce the influence of outliers, median and median absolute deviation (MAD) of MTR_asym and R₂′ within these regions were calculated and MAD was used to define variability for all measurements. A Wilcoxon signed-rank test was used to determine whether averaging MTR_asym from two gradient echoes resulted in a decrease in healthy tissue MTR_asym variability compared with a single gradient echo. For glioblastoma patients, VOIs of gadolinium contrast enhancement (CE) were segmented using T₁ subtraction maps and a semi-automated thresholding method outlined previously(35,36). Regions of central necrosis were also delineated and examined. A paired t-test was used to determine whether R₂′ and MTR_asym differed between T₂-hyperintense lesions and NAWM. Within glioblastoma patients (WHO grade IV), a one-way repeated-measures ANOVA and Tukey’s test for multiple comparisons was used to determine whether R₂′ or MTR_asym at 3ppm differed between regions of NAWM, T₂ hyperintense lesions, regions of CE, and areas of central necrosis. An additional one-way ANOVA and Tukey’s test for multiple comparisons was used to compare R₂′ or MTR_asym at 3ppm for T₂ lesions across glioma grades II, III, and IV.

RESULTS

MTR_asym at 3.0ppm within each glutamine phantom, varying by pH, was similar across all four echoes, showing a characteristic increase in MTR_asym at 3.0ppm with decreasing pH (Fig 3A; ANOVA, P = 0.999; Comparison across echoes, P>0.99). These experimental results closely matched simulation results using measured and theoretical relaxation and exchange rates (Fig 3A). CNR was higher when averaging measurements from the two gradient echoes (TE=14.1ms & 34.1ms) compared to a single gradient echo (TE=14.1ms) in phantom samples containing the same concentration of glutamine, but varying pH. In particular, CNR was approximately 13% higher when comparing pH=7.5 to 7.0 (CNR_1&2=2.68; CNR₁=2.37), 7% higher when comparing pH=6.5 and 7.0 (CNR_1&2=5.07; CNR₁=4.74), 15% higher when comparing pH=6.0 vs 7.0 (CNR_1&2=17.2 CNR₁=15.0), and 6.5% higher when comparing pH=6.0 vs. 6.5 (CNR_1&2=6.72; CNR₁=6.31). Consistent with phantom results, median MTR_asym at 3.0ppm in NAWM across all patients was not significantly different when using a single echo and the average of two gradient echoes (P=0.31); however, the median absolute deviation (MAD), a measure of variance in the measurements, was significantly lower when averaging the two gradient echoes (P=0.003).

Fig. 3. Simulation and phantom testing.

A) CEST contrast increases with decreasing pH within a physiological pH range similarly for all four echoes in an amino acid phantom and simulation estimates. These measurements closely match simulation estimates with measured or estimated relaxation and exchange rates (dashed lines). B) R₂ and R₂^* as measured by the multi-echo sequence show a linear increase with increasing Gd-DTPA concentration. Estimates of r₂ = 6.24 ± 0.04 mM⁻¹·sec⁻¹, r₂^* = 6.86 ± 0.10 mM⁻¹·sec⁻¹, and r₂′ = 0.61 ± 0.08 mM⁻¹·sec⁻¹. C) Comparison between measured R₂ using CEST-SAGE-EPI and CPMG show strong correlation (R²=0.9943, P<0.0001), independent of pH (P=0.9915). D) Comparison between measured R₂* using CEST-SAGE-EPI and ME-GRE showing a strong correlation (R²=0.9727, P<0.0001) and no dependence on pH (P=0.2184). E) Estimates of R₂′ obtained through subtraction of sequential CPMG and ME-GRE measurements were not significantly different from those obtained using multi-echo SAGE EPI (slope = 1.26±0.1301, P=0.0555) and did not differ by pH (P=0.0533). F) Amine CEST has decreased sensitivity in the presence of contrast agents (Gd-DTPA) that shorten T₁ and T₂. Thus, edematous tumor tissue with longer T₁ and T₂ have higher CEST contrast in acidic environments. G) Coefficient of variation (COV) measured from multiple test-retest experiments is also higher in the presence of contrast agents or in environments with shorter T₁ and T₂ characteristics.

In phantom samples containing varying concentration of Gd-DTPA, CEST-SAGE-EPI estimates of R₂, R₂^*, and R₂′ varied linearly with concentration (Fig 3B), with estimates of transverse relaxivities of r₂ = 6.24 ± 0.04 mM⁻¹·sec⁻¹ (P<0.0001), r₂^* = 6.86 ± 0.10 mM⁻¹·sec⁻¹ (P<0.0001), and r₂′ = 0.61 ± 0.08 mM⁻¹·sec⁻¹ (P=0.0007). Phantom results identified a strong, significant linear correlation between R₂ measurements using CEST-SAGE-EPI and CPMG (Fig 3C; R²=0.9943, P<0.0001) and did not differ by pH (P=0.9915). Estimates of R₂ using CEST-SAGE-EPI, however, were lower than estimates of R₂ using standard CPMG measurements (slope = 0.8845± 0.006, P<0.0001 compared with slope = 1), with CEST-SAGE-EPI estimates approximately 760us lower on average than CPMG measurements (~7.1%). A strong linear correlation was also observed between R₂^* measurements obtained using CEST-SAGE-EPI and ME-GRE measurements (Fig 3D; R²=0.9727, P<0.0001) and these measurements also did not differ by pH (P=0.2184). Estimates of R₂^* were not significantly different from measurements obtained using standard ME-GRE measurements (slope = 0.9862±0.0155, P=0.3819 showing no difference between slope=1). Consistent with these results, calculated estimates of R₂′ obtained through subtraction of sequential CPMG and ME-GRE measurements were congruent with SAGE-EPI measurements (Fig 3E; slope = 1.26±0.1301, P=0.0555 showing no difference between slope=1) and did not differ by pH (P=0.0533).

Although measurements of transverse relaxation rate were not affected by pH, the relationship between MTR_asym at 3.0ppm and pH was affected by transverse relaxation rate (Fig 3F). Specifically, as transverse relaxation rates R₂ or R₂^* increased as a result of increased concentration of Gd-DTPA, the sensitivity of MTR_asym at 3.0ppm to acidic pH decreased, particularly when concentrations were higher than 1mM, corresponding to T₂ measurements of 150–170ms and T₂^* = 130–170ms. Repeated experiments showed higher coefficient of variation (COV) with higher concentrations of Gd-DTPA, or lower T₁, as well as higher COV with higher pH (Fig 3G).

Qualitatively, the T₂ hyperintense lesions in all patients exhibited heterogeneous areas of both elevated MTR_asym at 3.0ppm (acidity) and R₂′ (hypoxia). Many areas of non-enhancing tumor in all patients exhibited uncharacteristically low measures of R₂′ despite evidence of elevated acidity, suggesting non-enhancing tumor regions may be adequately oxygenated and undergoing aerobic glycolysis (e.g. Fig 1C). For example, Fig 4A illustrates a WHO grade II astrocytoma with an area of moderately elevated acidity in the medial frontal lobe; however, this region exhibited R₂′ ~50% lower than surrounding NAWM. Fig 4B illustrates a large non-enhancing, IDH mutant WHO grade III astrocytoma with sizeable regions of macroscopic necrosis, as illustrated by T₁ hypointensity. These areas of necrotic tissue exhibited both high levels of acidity and hypoxia, while surrounding non-enhancing components with largely intact blood-brain barrier had elevated acidity, as suggested by higher MTR_asym at 3.0ppm, but lower levels of hypoxia, or R₂′, compared with regions of NAWM. Figs 4C–D show patients with recurrent and newly diagnosed glioblastoma (WHO IV), respectively. Compared to the WHO grade II and III tumors with largely non-enhancing tumor with intact blood-brain barrier, patients with glioblastoma exhibiting extensive areas of contrast enhancement displayed both high acidity (MTR_asym at 3.0ppm) and hypoxia (R₂′) as well as regions of moderate acidity and oxygenated tumor. (Additional information including T₂ and T₂* maps are illustrated in Supporting Figure S2).

Fig. 4. pH- and oxygen-sensitive MR images of human gliomas.

A) A 42-year-old female with a nonenhancing recurrent WHO II IDH mutant astrocytoma (Patient #43) exhibiting a region of focal acidity (high MTR_asym at 3ppm) and low oxygen consumption (low R₂′). B) A 51-year-old male patient with a newly diagnosed nonenhancing IDH mutant WHO III anaplastic astrocytoma (Patient #22). This lesion showed large, heterogeneous regions of abnormally high and low hypoxia and acidity. C) A 53-year-old female patient with a recurrent IDH wild type WHO IV glioblastoma exhibiting a region of focal acidity and large regions of low oxygen consumption (Patient #15). D) A 53-year-old female with a newly diagnosed IDH wild type WHO IV glioblastoma (Patient #3) displaying ring enhancement, central necrosis, elevated acidity and oxygen extraction (hypoxia) within the enhancing region as well as low oxygen consumption in surrounding nonenhancing tumor.

Quantitative evaluation of various regions within these tumors confirmed these observations (Figs 5–6). Regions of T₂ hyperintense lesions exhibited a significantly lower median R₂′ (4.8 ± 0.2 sec⁻¹) compared with NAWM (6.2 ± 0.2 sec⁻¹) when pooling patients across all tumor grades (Fig 5A; Paired t-test, P<0.0001). Within T₂ hyperintense lesions, median R₂′ did not vary significantly across tumor grade (Fig 5B; ANOVA, P=0.0537; WHO II = 4.6 ± 0.4 sec⁻¹, WHO III = 4.2 ± 0.4 sec⁻¹, WHO II = 5.4 ± 0.3 sec⁻¹). In glioblastoma patients, significant differences in median R₂′ across tissue types were observed (Fig 5C; Repeated measures ANOVA, P=0.0001). In particular, median R₂′ was significantly lower in T₂ hyperintense lesions (5.3 ± 0.3 sec⁻¹) compared to both NAWM (6.3 ± 0.2 sec⁻¹; Tukey’s Test, Adjusted P=0.0078) and CE regions (10.0 ± 1.0 sec⁻¹; Tukey’s Test, Adjusted P=0.0002); whereas no difference was detected between T₂ lesions and areas of central necrosis (4.0 ± 0.5 sec⁻¹; Tukey’s Test, Adjusted P=0.1499). Additionally, median R₂′ was significantly higher in CE lesions compared with NAWM (Tukey’s Test, Adjusted P=0.0064) and necrosis (Tukey’s Test, Adjusted P<0.0001), but significantly lower in necrosis compared with NAWM (Tukey’s Test, Adjusted P=0.0064). No differences in median R₂′ were observed between recurrent or newly diagnosed tumor or normal tissues (P>0.2), nor were significant differences observed between IDH mutant and wild type tumors (P=0.12); however, IDH mutant tumors tended to have lower R₂′ compared with IDH wild type tumors.

Fig. 5. pH and oxygen-sensitive MR measurements in human glioma patients.

A) Median R₂′ measurements in normal appearing white matter (NAWM) and the T₂ hyperintense lesions in all patients, independent of tumor grade. B) Median R₂′ in T₂ hyperintense regions for different tumor grades. C) Median R₂′ in different brain regions for patients with glioblastoma (GBM). D) Median MTR_asym at 3ppm (%) in NAWM and T₂ hyperintense lesions in all patients pooled across all tumor grades. E) Median MTR_asym at 3ppm (%) in T₂ hyperintense regions for different tumor grades. F) Median MTR_asym at 3ppm (%) in different tissue types in GBM. * = P<0.05, **=P<0.01, ***=P<0.001, ****=P<0.0001.

Fig. 6. Combined pH and oxygen-sensitive MR measurements in human gliomas illustrating additional separation of tissue types.

A) MTR_asym at 3ppm (%) versus R₂′ for NAWM and T₂ hyperintense lesions in all patients pooled across tumor grades. B) MTR_asym at 3ppm (%) versus R₂′ in T₂ hyperintense regions for different tumor grades. C) MTR_asym at 3ppm (%) versus R₂′ in different tissue types in GBM.

Median MTR_asym at 3.0ppm within T₂ hyperintense lesions (1.7 ± 0.1%) were significantly higher than NAWM (0.8 ± 0.03%) when pooling all patients across grade (Fig 5D; Paired t-test, P<0.0001). Within T₂ hyperintense lesions, median MTR_asym at 3.0ppm was significantly different across tumor grade (Fig 5E; ANOVA, P=0.0297), with WHO IV glioblastoma (2.0 ± 0.2%) exhibiting significantly higher acidity compared with WHO II gliomas (1.5 ± 0.1%; Tukey’s Test, Adjusted P=0.0432). No difference was observed between WHO III gliomas (1.6 ± 0.2%) compared with other grades (Adjusted P>0.05). In a separate comparison, IDH mutant gliomas exhibited a slightly higher degree of tumor acidity compared with IDH wild type tumors when correcting for of grade (Adjusted P=0.0434). Within WHO IV glioblastoma, significant differences in median MTR_asym at 3.0ppm were observed between various tissue types (Fig 5F; ANOVA, P<0.0001). Areas of macroscopic necrosis exhibited the highest MTR_asym at 3.0ppm degree of acidity (MTR_asym at 3.0ppm = 4.4 ± 0.4%) compared with all other tissue types, including CE lesions (2.9 ± 0.1%; Tukey’s Test, Adjusted P<0.0001), T₂ hyperintense regions (2.0 ± 0.2%, Adjusted P<0.0001), and NAWM (0.8 ± 0.04%, Adjusted P<0.0001). Additionally, CE tumor exhibited significantly higher median MTR_asym at 3.0ppm compared with T₂ hyperintense lesions (Tukey’s Test, Adjusted P=0.0218) and NAWM (Tukey’s Test, Adjusted P<0.0001), while T₂ lesions showed higher median MTR_asym at 3.0ppm (Tukey’s Test, Adjusted P<0.0001).

Examination of acidity and hypoxia characteristics within T₂ hyperintense lesions pooled across tumor grade were markedly different from NAWM (Fig 6A), as T₂ hyperintense lesions tended to be more acidic but slightly less hypoxic compared with NAWM. T₂ hyperintense lesions did not show substantial separation across tumor grade (Fig 6B). Glioblastoma exhibited distinct characteristics across various tissue components (Fig 6C), with NAWM and T₂ hyperintense lesions having relatively lower acidity and hypoxia combined with CE tumor and macroscopic necrotic tissue. CE tumor exhibited a moderately high level of acidity, along with high degrees of oxygen extraction, while necrotic tissue showed low oxygen extraction yet high acidity.

DISCUSSION

High-resolution pH- and oxygen-sensitive MR imaging contrast in brain tumors can be achieved on clinical 3T MR systems by implementing a multi-echo SAGE EPI readout after off-resonance saturation or CEST preparation of amine protons (3.0ppm). Comparable techniques for CEST imaging that take advantage of multiple echo readouts are relatively limited. Sun et. al. described a sequence that utilizes a CMPG readout following CEST saturation for improving CNR in agarose phantoms by averaging images from a spin echo train with echo times between 30 and 60 ms (37). However, this sequence only measured R₂ characteristics and it could not correct for B₀ using phase information available via multiple gradient echoes. The fast EPI readout available using CEST-SAGE-EPI vastly improves acquisition times over slower, single slice or volumetric approaches (38–40). Song et. al. (41) and others have described single echo EPI approaches to speed up acquisition time; however, these studies lacked concurrent measurement of subsequent echoes that would allow for estimation of R₂, R₂*, R₂′ and B₀. Thus, CEST-SAGE-EPI has many advantages including speed, in-line B₀ mapping, whole brain coverage, and simultaneous CEST, R₂, R₂*, and R₂′ quantitation.

Mathematical simulations and phantom measurements confirmed the relationship between MTR_asym at 3.0ppm and pH reported previously (16). Measurements of transverse relaxivity for Gd-DTPA were found to be r₂ = 6.24 ± 0.04 mM⁻¹·sec⁻¹ and r₂^* = 6.86 ± 0.10 mM⁻¹·sec⁻¹, which is consistent with the literature (42,43). Measurements of R₂ and R₂* also did not vary as a function of pH. As compared to CPMG measurements, CEST-SAGE-EPI tended to underestimate R₂, consequently resulting in overestimation of R₂′; however, the magnitude of this difference averaged less than 1ms (~760us), or less than 10% difference, and increased with relaxation rate R₂. This discrepancy may be related to inaccurate modeling of imperfections during RF inversion during SAGE-EPI readout, as has been suggested in previous SAGE-EPI studies (31,32). In addition to OEF, R₂′ value may also be influenced by venous blood fraction as reported by several studies (22–25). A more direct quantification of oxygen consumption or comparison may be achieved with high-field ¹⁷O MRI by detecting metabolically produced H₂¹⁷O (44) or with ¹⁷O MRS (45).

As demonstrated previously (16), edematous tissues and cancer tissues with long T₂ tend to amplify pH-dependent CEST contrast, while tissues with shorter T₂ exhibit only a few percentage changes in contrast, even in highly acidic and highly concentrated amino acid environments. In particular, with relatively short T₂/T₂* including normal appearing white and gray matter may not have suitable sensitivity or CNR to provide meaningful pH information. Thus, the current CEST-SAGE-EPI sequence may not be immediately useful to explore functional or metabolic changes in otherwise normal neural tissues without additional considerations (e.g. higher field strengths, more averages, etc.), but it is uniquely suited for detection and assessment of brain tumor or oncologic metabolism. Although MTR_asym contrast could be altered by effects from exchangeable proton pools other than amine protons, we have shown with phantoms (Supporting Figure S3) that the low pH results in elevation of MTR_asym at 3.0ppm despite the presence of amide protons and MT effect (introduced with 5% egg-white protein) and this is comparable to previous reports (16). The predominant pH dependence of MTR_asym at 3.0ppm could be explained by the high irradiation amplitude (B₁ = 6μT) preferably saturating fast-exchanging amine protons compared when with other proton pools. Another concern with using amine CEST contrast at 3.0ppm is it could suffer from an insufficient source of exchangeable protons if glutamine concentration is too low. In addition to glutamine, other amino acids and neurotransmitters possessing a similar amine functional group have similar pH dependence of exchange rate and MTR_asym at 3.0ppm (16). Large molecules such as bovine serum albumin (BSA) were also reported to show characteristics of fast-exchanging amine protons at 3.0ppm (46). With a total of around 20–25 mM amino acids within normal neural tissues (47) and other contributions of amine proton pools, sufficient amine CEST contrast within tumor tissues is highly probable.

Combined information about both median tissue acidity (MTR_asym at 3.0ppm) and oxygen extraction (R₂′) helped to further delineate various tissue types and provide additional insights into the tumor microenvironment. Consistent with tumor biology and the Warburg effect, we observed elevated acidity in tumor tissues even when there was adequate oxygen delivery due to an intact BBB, high neovascularity, and high blood flow. In particular, we observed lower levels of hypoxia and high acidity in regions of non-enhancing tumor (T₂/FLAIR hyperintensity) (Fig 5A,D; 6A,C), which can have elevated vascularity depending on the degree of malignancy and/or grade. This was similar across tumor grade, although GBM had slightly higher levels of hypoxia compared with lower grades. In the same tumors we observed substantially higher levels of hypoxia in areas of contrast enhancement (Fig 4D; 5C; 6A,C), which is known to be the most aggressive (48) and hypoxic (49–51). These data suggest tumor acidity and oxygen consumption are both complex and spatially heterogenous, consistent with the known genetic, histopathologic, proteomic, and metabolic spatial heterogeneity (52,53). Hypoxic, non-hypoxic, and/or acidic tumors may also have differing therapeutic responses, as tumor hypoxia and acidity are both known to reduce sensitivity to chemoradiation. Thus, a tool that can co-localize areas of simultaneous pH and oxygen imbalance may allow for future therapeutic strategies to include radiation boost to regions containing the most hypoxic and acidic environments. Additionally, this tool may also allow for evaluation of direct inhibition of energetic pathways as a therapeutic target in these tumors (54–56).

In addition to tissue acidity and reduced oxygen consumption, the Warburg effect is also characterized by altered levels of glucose consumption. Dynamic glucose enhanced MRI (DGE-MRI), which selectively saturates hydroxyl protons of glucose and allows for a dynamic detection of CEST contrast during glucose uptake, has shown promising results in mapping in animals (57,58) and in human tumor patients (59–61) and could potentially add another important piece of information about the tumor microenvironment.

Mutations of isocitrate dehydrogenase (IDH1) are common in less aggressive gliomas and cause inhibition of the conversion of isocitrate to alpha-ketoglutarate, resulting in a buildup of 2-hydroxygluterate (2HG), inhibiting oxidative phosphorylation and therefore inhibiting the ability of cells to utilize aerobic glycolysis (62). Therefore, we hypothesized IDH1-mutated tumors may have lower R₂′ measurements compared to IDH1 wild-type tumors, consistent with previous work by Stadbauer et al. (63). Although we observed slightly lower R₂′ in IDH1 mutated patients compared to wild-type patients, these effects were not statistically significant. Further studies in a larger patient cohort are needed to better understand this relationship and whether pH- and oxygen-sensitive MR signatures can identify IDH1 mutant tumors.

One potential limitation of our approach is using the conventional asymmetric analysis (MTR_asym) to interpret CEST effects. Although the MTR_asym approach removes the symmetric effects from direct water saturation (DS), the resulting contrast is still influenced by a number of other effects, including asymmetric magnetic transfer (MT) effect of semi-solid tissue components (64), nuclear Overhauser enhancement (NOE) 0–5ppm upfield to water proton resonance frequency (65), and mixing effects from nearby exchangeable proton pools. Several alternative approaches have been developed to mitigate these problems, including chemical exchange rotation transfer (66), Lorentzian difference (LD) analysis (46), multiple-pool Lorentzian fit (67,68), and the three-point method (69). Further investigation of pH- and oxygen-sensitive imaging with CEST-SAGE-EPI sequence could potentially benefit from applying the mentioned methods.

Additionally, proton exchange rates are influenced by temperature, with faster exchange rates of amine protons observed at higher temperatures, due to increased kinetic energy along with increased T₂ relaxation times resulting in increased available NMR signal at a given frequency. Thus, phantom results at room temperature (~18°C) in the current studies are likely to underestimate MTR_asym when compared to in vivo comparisons (~37°C), as has been previously demonstrated (16). Thus, care should be given when interpreting absolute values of MTR_asym observed in the phantom portion of the current study.

It is important to note that in vivo R₂′ mapping using SAGE-EPI readout can be influenced by a variety of technical issues, including B₁ inhomogeneity or mismatch between excitation and refocusing slice profiles, as well as biological factors including the effects of water diffusion and multiple water compartments within the tissues (32,70). CEST contrast may also be influenced by multiple water compartments within tissues, although the water signal is thought to be predominantly from the extracellular water pool due to the very short relaxation times of bound intracellular water (71).

Technical imperfections including static magnetic field (B₀) and transmit radiofrequency field (B₁) inhomogeneity may also have led to altered image contrast. In our study, B₀ inhomogeneity was corrected with B₀ map generated from WASSR method or from multi-echo B₀ mapping; however, B₁ inhomogeneities were not compensated for. B₁ inhomogeneity under 3T is reported to be relatively small (deviation < 10–20%) and have limited effect on T₂ quantification (72) and pH-sensitive CEST contrast (16) within the variation range in the current study. Nevertheless, validation of the effect of B₁ inhomogeneity on CEST-SAGE-EPI sequence and incorporation of B₁ correction in future work may be beneficial.

A final potential limitation of this study was the relative inhomogeneity of the patient population, which contained untreated and recurrent glioma patients on a variety of therapies. A larger study that separates out specific treatments along with tumor grade is necessary to better understand their influence on our measurements.

CONCLUSION

The current study presents a new CEST-SAGE-EPI sequence for obtaining high-speed, full brain pH- and oxygen-sensitive image contrasts for brain tumor evaluation. Advantages of this technique include fast acquisition, in-line B₀ correction using phase information, whole brain coverage, and simultaneous accurate estimation of CEST effects, R₂, R₂*, and R₂′. Results in tumors showed a high degree of spatial heterogeneity and measurements were consistent with known cancer biology.

Supplementary Material

Supp figS1-3

Supporting Figure S1. Examples of B₀ correction methods available for CEST-SAGE-EPI. A) B₀ maps calculated for a single patient using phase differences from the first two gradient echoes acquired using CEST-SAGE-EPI. B) Corresponding B₀ maps for the same patient using the WASSR method. C) Resulting maps of MTR_asym at 3ppm after B₀ correction using multi-echo phase information. D) MTR_asym at 3ppm after the WASSR method for B₀ correction. E) MTR_asym at 3ppm with no B₀ correction.

Supporting Figure S2. Anatomic and relaxometry MR images of human gliomas estimated using CEST-SAGE-EPI. Complimentary T₂ and T₂* relaxometry images estimated from CEST-SAGE-EPI for the four patients illustrated in Fig. 3.

Supporting Figure S3. Phantom results for 40mM glutamine and 5% egg-white protein. Results demonstrate increasing CEST contrast with decreasing pH, similar to those reported previously with pure glutamine phantoms. This suggests fast-exchanging amine protons are the primary proton pool saturated using the current radiofrequency pulse scheme. Phantom preparation: Combined glutamine and protein phantoms with pH ranging from 6 to 7.5 were prepared by first adding glutamine into liquid egg white solution (5% protein) to reach concentrations of 40mM. The pH within the sample solutions was then titrated to 6.0, 6.3, 6.6, 6.9, 7.2, and 7.4, with acid (1N HCl) and base (1N NaOH) solutions. Scan parameters: B1 = 6uT, TR = 5000ms, TE1/TE2/TE3/TE4 = 14.0/34.1/58.0/92.4ms, FOV = 217×240mm, matrix size = 116×128, slice thickness = 4.0mm, slice number = 10, partial Fourier encoding = 6/8, GRAPPA = 3, bandwidth = 1628Hz/pixel with number of excitations or averages (NEX) = 2. A total of 36 offset frequencies were acquired: ± 2.5ppm to ± 4.0ppm with 0.1ppm interval, ± 6.0ppm and ± 20.0ppm with respect to water proton Larmor frequency. Measurements were taken at room temperature (~20°C). Error bars represent standard deviation within a region of interest within the Falcon tubes.