Abstract
Metabolic imaging produces powerful visual assessments of organ
function in vivo. Current techniques can be improved by safely
increasing metabolic contrast. The gold standard,
2-[^18F]fluorodeoxyglucose-positron emission tomography (FDG-PET)
imaging, is limited by radioactive exposure and sparse assessment of
metabolism beyond glucose uptake and retention. Deuterium magnetic
resonance imaging (DMRI) with [6,6-^2H[2]]glucose is nonradioactive,
achieves tumor metabolic contrast, but can be improved by enriched
contrast from deuterated water (HDO) based imaging. Here, we developed
a DMRI protocol employing [^2H[7]]glucose. Imaging ^2H-signal and
measuring HDO production in tumor-bearing mice detected differential
glucose utilization across baseline tumors, tumors treated with vehicle
control or anti-glycolytic BRAFi and MEKi therapy, and contralateral
healthy tissue. Control tumors generated the most ^2H-signal and HDO.
To our knowledge this is the first application of DMRI with
[^2H[7]]glucose for tumoral treatment monitoring. This approach
demonstrates HDO as a marker of tumor glucose utilization and suggests
translational capability in humans due to its safety, noninvasiveness,
and suitability for serial monitoring.
__________________________________________________________________
DMRI with [
[MATH: H72 :MATH]
]glucose allows for a safe and sensitive assessment of tumoral glucose
avidity and treatment efficacy.
INTRODUCTION
Cancer is the second leading cause of death behind heart disease
([64]1). Although substantial improvements have been made in cancer
therapeutics, early detection and treatment monitoring are essential
for increasing overall patient survival rates. Furthermore, traditional
cancer diagnostics such as biopsies and laboratory assessments are
limited by sampling bias or by insufficient sensitivity and specificity
required for complete diagnosis, posing a notable confounder for
patient health ([65]2, [66]3). There remains a pressing need for
improved cancer diagnostics for patient care. One of the primary
diagnostic methodologies is in vivo imaging, which is the only
technique that allows for an internal assessment of cancer when
physical examination is not a suitable alternative ([67]4). In
particular, metabolic imaging is a powerful modality that allows for a
direct visual assessment of real-time tumor metabolism in the human
body. Present imaging techniques can be improved through the
development of safe and noninvasive metabolic contrast approaches. ^1H
magnetic resonance spectroscopic imaging (MRSI) is a robust approach
that can generate tumoral contrast by imaging ^1H spectroscopy
metabolite signals. ^1H MRSI has shown substantial promise in breast
cancer cases where total choline is accumulated in sufficiently large
levels that distinguish cancer from healthy tissue ([68]5). The high
abundance of hydrogen allows ^1H MRSI to acquire high SNR and
resolution images but at the cost of specificity. However, without the
use of an isotope tracer, there is no straightforward way of detecting
the specific reactions that contribute to the generation of signal. The
gold standard metabolic imaging technique for cancer detection and
staging is 2-[^18F]fluorodeoxyglucose–positron emission tomography
(FDG-PET) ([69]6, [70]7), which is performed about 2.2 million times
per year in the United States ([71]8, [72]9). Although FDG-PET can
detect the presence of many cancers with extremely high sensitivity, it
is limited by the use of a radioactive glucose tracer that imposes a
risk for secondary cancer, has limited serial monitoring capacity, and
is sensitive to glucose uptake but not glycolytic utilization or
downstream metabolism ([73]10–[74]12). In addition, FDG-PET methodology
is less selective in tissues where glucose utilization is notably high
(e.g., brain).
In contrast, deuterium magnetic resonance imaging (DMRI) can generate
metabolic contrast without using ionizing radiation and has the ability
to report both tracer uptake and labeling of downstream intermediates
through metabolic activity within the tissue. The application of DMRI
in human cancer was demonstrated by the seminal work of de Feyter
et al. ([75]13), in which a human glioblastoma was imaged with the
detection of ^2H-lactate and ^2H-glutamine/glutamate (glx) generated
from a [6,6′-^2H[2]]glucose contrast agent. Given the safety of
deuterated tracers along with the ability to generate distinct tumoral
metabolic contrast within highly glycolytic brain tissue, DMRI with
[6,6′-^2H[2]]glucose has potential as a clinical approach for detecting
cancer metabolism. However, this method has limited sensitivity,
resulting in long scan times. We hypothesized that [^2H[7]]glucose
might provide increased sensitivity. Unlike partially deuterated
tracers such as [6,6′-^2H[2]]glucose, perdeuterated [^2H[7]]glucose
generates significantly more HDO and greater MRI signal that can be
used to improve spatial resolution. In combination with imaging an
earlier time frame after the administration of [^2H[7]]glucose, this
methodology can use simple MRI sequences that are substantially faster
than chemical shift imaging methods.
Most cancer cells consume more glucose and produce more lactate
relative to nonmalignant cells (the Warburg effect) ([76]14, [77]15).
By assessing differential glucose consumption, one can get a direct
readout of metabolic activity in cancer. Therefore, with
[^2H[7]]glucose, we can assess glucose utilization by measuring the
production of HDO in tumors with high sensitivity and specificity as
well as in a safe and minimally invasive manner. Focusing on the
initial kinetics and using [^2H[7]]glucose, we can detect localized
metabolism associated with the bioenergetics of glycolysis and, to some
extent, mitochondrial respiration ([78]16). We previously demonstrated
that HDO production and [^2H[7]]glucose consumption are positively and
linearly correlated in cell culture models ([79]17–[80]19). Notably,
HDO production was more sensitive to glycolytic flux than lactate, as
lactate production and glucose consumption did not correlate well at
early time points following tracer administration. In addition, HDO has
a natural abundance (NA) of 0.015%, with a readily detectable signal
before any deuterated tracer is metabolized. This makes HDO a more
robust metric, as HDO generation can be measured as an overall increase
from NA, while deuterated lactate and glx are initially undetectable.
This circumstance allows HDO spectra and images to be acquired with
higher initial signal-to-noise (SNR) than detection of lactate or glx.
Using the endogenous NA HDO signal normalization is superior to the use
of an external phantom, as there could be a voxel-by-voxel correction
for osmolality and temperature that cannot be simulated with a single
external standard in a straightforward manner.
To assess the in vivo potential of using HDO as a marker of glycolytic
flux, we imaged glucose metabolism in the rat brain. With this murine
model, we have shown that DMRI can track HDO production in a timely
manner, with a highly conserved correlation between HDO and glx
generation between 8 and 24 min after injection of [^2H[7]]glucose
([81]18). Notably, HDO maintained superior SNR in ^2H nuclear magnetic
resonance (NMR) spectra compared to lactate and glx. Thus, HDO
production is a logical in vivo biomarker for monitoring treatments
that perturb glycolysis.
Here, we examine the ability of this method to detect metabolic
responses arising from successful cancer therapy. We used Yale
University Mouse Melanoma 1.7 (Yumm1.7) cells in culture and in
subcutaneous xenografts. Yumm1.7 melanoma cells contain the common BRAF
V600E mutation and express inactive Phosphatase and Tensin homolog
(PTEN) and Cyclin-Dependent Kinase Inhibitor 2a (Cdkn2a), thus
genotypically modeling some late-stage human melanomas ([82]20–[83]22).
The BRAF V600E variant has constitutively active kinase activity that
up-regulates mitogen-activated protein kinase (MAPK) signaling and
enhances aerobic glycolysis in cancer cells ([84]20). Loss of PTEN
enhances signaling through the PI3K/Akt/mTORC1 pathway, which also
stimulates glucose metabolism ([85]23–[86]25). Cdkn2a silencing
ultimately grants deregulated cell division and genetic instability
through the deficit of p14 and p16 tumor suppressors, which cooperate
with BRAF V600E in propagating tumorigenesis. Treating BRAF-mutant
melanomas with BRAFi Mitogen-activated protein kinase kinase inhibitor
(MEKi) shows an acute suppression of glucose metabolism ([87]26,
[88]27). Therapy with dabrafenib (BRAFi) and trametinib (MEKi) has Food
and Drug Administration approval, both as individual agents and in
combination, to target this mechanism ([89]28). Reduced tumor growth
and glucose consumption has been observed in animals bearing Yumm1.7
flank tumors treated with BRAFi and MEKi compared to control animals
([90]26, [91]29). Previous studies on BRAFi and MEKi as well as other
cancers have demonstrated treatment-induced perturbations in glycolysis
by FDG-PET imaging and isotopomer analysis with [U-^13C]glucose, but
none have assessed [^2H[7]]glucose consumption and/or HDO production.
We demonstrate that imaging total deuterium (^2H) signal and
spectroscopically measuring deuterated water (HDO) production from
[^2H[7]]glucose increases tumoral metabolic contrast and readily allows
tumor detection and treatment monitoring. This result was achieved by
the implementation of straightforward gradient echo imaging and 1D in
vivo spectroscopy, both of which are simple MRI protocols that are
readily available on clinical systems. In addition, to our knowledge,
we are the first to image [^2H[7]]glucose utilization to detect tumors
and assess the efficacy of targeted therapies in vivo.
RESULTS
Established BRAFi and MEKi dose response by Western blot analysis
To assess the effectiveness of BRAFi and MEKi combinatorial treatment
on signaling, we performed a dose response from low (0.1 μM BRAFi and
0.005 μM MEKi) to high (20 μM BRAFi and 1 μM MEKi) concentrations and
performed Western blot analysis on control and treated Yumm1.7 cells in
culture. We probed the blot with phosphorylated extracellular
signal–regulated kinase (pERK) and ERK antibodies as they are
downstream targets of BRAF and MEK. Figure S1 demonstrates reduced pERK
and pERK/total ERK ratios at the low concentration with no ERK
phosphorylation signal at the medium or high concentration (fig. S1, A
to G).
BRAFi and MEKi therapeutic activity can be monitored with HDO production in
vitro
On the basis of our findings, the medium concentrations of BRAFi (1.5
μM) and MEKi (0.07 μM) were selected for use in a cell culture
[^2H[7]]glucose tracer study. Specifically, these concentrations
achieve maximal inhibition of MAPK signaling (fig. S1, A to F) with
only a modest reduction in cell number (fig. S1G). Treatment induced
alterations in signaling were matched by large changes in metabolites
in the ^2H NMR spectrum ([92]Fig. 1). Treatment significantly reduced
the rates of extracellular glucose consumption, HDO production, and
^2H-lactate production by 64, 50, and 74%, respectively ([93]Fig. 1, A
to F). Intracellular concentrations of ^2H-lactate were significantly
reduced by treatment and ^2H-glx concentrations were lower in treatment
(fig. S1H). In addition, both ^2H-lactate and HDO were found to be
highly correlated with [^2H[7]]glucose consumption, with correlation
coefficients (r) of 0.99 and 0.99, respectively (fig. S1, I and J).
Fig. 1. In vitro characterization of the effects of treatment on glucose
utilization.
[94]Fig. 1.
[95]Open in a new tab
Combinatorial treatment leads to a significant downregulation of
[^2H[7]]glucose utilization detected by HDO production in cell culture.
(A and B) Representative ^2H spectra of media samples collected across
6 hours from control and treatment groups, respectively. (C to F)
Quantitative ^2H NMR kinetic data demonstrating that treatment reduces
glucose consumption, HDO production over NA, and ^2H-lactate production
in a sensitive manner. (G) Representative schematic of expected ^2H
labeling patterns from [^2H[7]]glucose utilization. (H to M) Fractional
enrichments of major glycolytic and TCA cycle intermediates demonstrate
a significant decrease in central carbon metabolism contributing to
reduced HDO production. (N) Representative schematic of metabolic model
designed to assess changes in flux. See [96]Table 1 for a list of
abbreviations. (O) Treatment decreased flux in pathways involved with
HDO production. Linear regression analysis was used to establish
significantly different kinetic slopes. Student’s t test was used to
establish significance between control and treatment rates of glucose
utilization. Two-way analysis of variance (ANOVA) was used to establish
significance across treatments, metabolite pool sizes, specific mass
ion species, and flux measurements. Statistically significant changes
are labeled as *P ≤ 0.05, **P ≤ 0.01, and ***P ≤ 0.001. Where not
visible, SD error bars are behind symbols. ppm, parts per million.
Gas chromatography–mass spectrometry (GC-MS) analysis of control and
treated cells revealed major central disruptions that contributed to
reduced HDO and lactate production ([97]Fig. 1, G to M). Decrements in
fractional enrichment were matched by significant decreases in
metabolite pool sizes with the exception of alanine, which increased
significantly (fig. S2). Lactate-to-alanine ratios were significantly
higher in control cells (~52) compared to treatment (~9) (fig. S3A).
Fractional enrichments and kinetic rates from ^2H NMR measurements of
glucose consumption, HDO production, and ^2H-lactate production were
used as inputs to generate absolute flux measurements by using a
[^2H[7]]glucose utilization metabolic model written in the Isotopomer
Network Compartmental Analysis (INCA) software ([98]Fig. 1N and table
S2). Analysis revealed treatment-induced decreases in glycolytic flux
without substantially altering tricarboxylic acid (TCA) cycle turnover
([99]Fig. 1O).
In addition, global metabolomics analysis with multivariate statistics
demonstrated distinct differences in the metabolic profile of control
and treated cells. Principal components analysis (PCA) showed clear
separation between control and treatment groups with a PC1 of 76.9%
(fig. S4A). Hierarchical heatmap clustering demonstrated excellent
unsupervised clustering of control and treatment groups as well as
corroborated significant decreases in glycolytic and TCA cycle
metabolites caused by treatment (fig. S4B). Pathway enrichment analysis
of the top 25 metabolites that were significantly different between
control and treated cells revealed significant enrichment of several
central pathways connected to HDO production (fig. S4C). Thus, extracts
of cultured cells also demonstrate strong treatment-driven separation
by PCA.
DMRI with [^2H[7]]glucose can detect tumor metabolism in vivo
To test the in vivo applicability of this model, we injected Yumm1.7
cells into the right flanks of C57BL/6J mice, allowed for tumors to
reach 10 to 15 mm, and imaged the animals before and after injection of
[^2H[7]]glucose. Our imaging protocol was established by optimizing a
set of imaging and spectroscopy sequences that maximized the
post-injection signal acquired. Axial slices allow for a comprehensive
and simultaneous assessment of the tumor, contralateral healthy tissue,
and phantoms of HDO and [^2H[7]]glucose ([100]Fig. 2, A and B). The
optimization of imaging sequences culminated in an imaging workflow
([101]Fig. 2C) that starts with animal setup, ^1H rapid acquisition
with relaxation enhancement (RARE) contrasted anatomical images based
on body water signal that allow the clear visualization of tumors as
brighter signal, ^2H fast low angle shot (FLASH) gradient echo
pre-injection images that acquire total ^2H signal and
non-localized/localized spectroscopy that allows kinetic tracking of
the production and consumption of metabolites, ^2H FLASH post-injection
images and spectroscopy, and final ^1H images for anatomical
referencing during quantitation. ^2H images were acquired with superior
in-plane and interpolated resolutions of ~1 (1.09) and 0.27 mm^2 with
an acquisition times of ~13 min for ^2H FLASH.
Fig. 2. Development of metabolically sensitive MRI for tumor detection.
[102]Fig. 2.
[103]Open in a new tab
(A and B) Representative three-dimensional (3D) rendering of animal
positioning and imaging slice selection with respect to the ^2H surface
saddle coil and the anatomical positioning of the flank tumors. The ^2H
image within the abdomen of the animal was acquired post-injection of
[^2H[7]]glucose, which generated signal primarily in the tumor and
kidneys. (C) DMRI workflow established for the sensitive assessment of
in vivo tumor metabolism. Workflow composed of a combination of ^1H
anatomical images, ^2H pre-injection and post injection images, and ^2H
spectroscopy throughout. (D) Region of interest (ROI) mapping-based
image quantification allows for the analysis of tumor specific signal.
(E) Representative nonlocalized ^2H spectroscopy time course of tumor
bearing animal post-[^2H[7]]glucose injection (1.95 g/kg). The small
peak at 0 min arises from unsuppressed signal from the phantom.
Image quantification used region of interest (ROI) mapping to
reproducibly assess metabolic activity in all image overlays generated
([104]Fig. 2D). The mean intensities measured within each ROI were used
as inputs for total signal estimates described in detail in the
methods. Quantification of the ^2H FLASH images acquired allowed for an
average signal increase of ~2.3 within the tumor ROI, between
post-injection and pre-injection images, indicating that our method can
generate substantial metabolic contrast. Single-pulse spectroscopy
([105]Fig. 2E) demonstrated a rapid increase in HDO signal
post-injection of [^2H[7]]glucose compared to pre-injection signal.
HDO imaging can be used to assess treatment efficacy in vivo
Yumm1.7 cells were injected into the right flanks of C57BL/6J mice, and
once tumors reached ~10 mm in diameter, the mice were imaged to
establish a baseline, and then control and treatment (BRAFi, 600 μg of
dabrafenib; MEKi, 6 μg of trametinib) oral gavages were administered
for 3 days followed by another round of imaging to assess the effects
of treatment ([106]Fig. 3A). Qualitatively, images showed notably
reduced metabolic activity in treated tumors compared to control
([107]Fig. 3, B and C). Image quantification showed significant
reductions caused by treatment in tumoral measurements of ^2H signal.
Treatment led to an 84% decrease in the total ^2H signal of tumors
([108]Fig. 3D). Measurements of tumor mass and volume showed that
treated tumors decreased in size and were significantly smaller than
control tumors ([109]Fig. 3E and fig. S5). Hematoxylin and eosin (H&E)
histological analysis of paraformaldehyde-fixed and paraffin-embedded
tumors demonstrated notably reduced cell density and greater evidence
of necrosis in treated tumors compared to control ([110]Fig. 3, F to
I).
Fig. 3. Imaging [^2H[7]]glucose utilization is sensitive to tumor metabolic
activity.
[111]Fig. 3.
[112]Open in a new tab
Total deuterium imaging (^2H FLASH) sensitively detects reductions in
tumor metabolic activity. The overall experimental timeline is
represented in (A). The images presented for all animals are as
follows: ^1H RARE (T2-weighted anatomical image), ^2H pre-injection and
post-injection (total deuterium image), and overlay (merged anatomical
image with total deuterium image). (B and C) Images acquired for two
control and two treated animals, respectively, before and after
[^2H[7]]glucose administration. (D) Quantitative data showing that
imaging tumoral measurements of total ^2H signal are sensitive measures
of therapeutic efficacy in vivo. (E) Tumor growth was assessed through
the course of treatment and controls demonstrated significantly greater
growth by final mass and volume. Representative H&E-stained control
tumor at 1× (F) and 20× (G) and treated tumor at 1× (H) and 20× (I)
shows reduced cell density and greater necrosis in treated tumors. The
dashed box in 1× images highlights the area chosen for the 20× view for
representative control and treated tumors. Student’s t test was used to
establish significance between baseline/day 0, control, and treatment
animals. Statistically significant changes are labeled as *P ≤ 0.05 and
**P ≤ 0.01.
^2H spectra reveal tumor-specific decreases in metabolic activity caused by
treatment
Throughout the imaging timeline, ^2H single-pulse (nonlocalized) and
slice-localized spectroscopy were performed to track the kinetics of
glucose utilization ([113]Fig. 4 and fig. S6). Measurements of
single-pulse spectra showed overall higher, but statistically
insignificant, HDO production in baseline tumor bearing animals with no
difference observed between control and treated animals (fig. S6A).
Baseline, control, and treated tumor-bearing animals showed similar
kinetics for glucose utilization (fig. S6B). Taking the ratio of
HDO/glucose demonstrates a measurement of local HDO and
[^2H[7]]glucose. Single-pulse HDO/glucose ratios were not statistically
different across all three groups but were slightly higher in control
animals (fig. S6C). We used slice-localized spectroscopy, which allowed
for the acquisition of spectra within selective slices placed
separately on tumors and contralateral healthy tissues. The slices
placed on tumors were set to minimize the inclusion of healthy tissue
in the axial plane; however, some healthy tissue was unavoidable in the
sagittal plane as the slice extends to max field of view (FOV) in this
plane. The normalized signal intensities of slice selective
spectroscopy showed that tumor tissue demonstrated faster initial
kinetics than healthy tissue ([114]Fig. 4A). No statistically
significant differences were observed in localized glucose utilization,
although overall greater glucose levels were observed in healthy
control tissue ([115]Fig. 4B). While not detected in single-pulse
spectra, slice-localized spectroscopy allowed for the acquisition of
^2H-lactate signal in vivo. Control tumors generated the greatest
amount of lactate compared to all other groups and tumors and, overall,
were found to generate greater HDO/glucose ratios than healthy tissue
([116]Fig. 4, C and D). Furthermore, at early time points control and
treated tumors generated ~3× to 4× more HDO compared to contralateral
healthy tissue. Given that some healthy tissue was covered by the
slice-localized spectroscopy slice, all metabolites were normalized by
tumor volume. Control tumors had up to ~6× more HDO as well as
significantly greater glucose uptake/utilization and lactate production
than treated tumors ([117]Fig. 4, E to G). Although no statistical
significance was established between control and treated HDO/glucose
ratios from tumor volume–normalized measurements, control tumors
trended to generate greater ratios ([118]Fig. 4H).
Fig. 4. ^2H spectroscopy reveals treatment-induced decreases in tumor
metabolic activity.
[119]Fig. 4.
[120]Open in a new tab
Slice-localized spectroscopy demonstrates distinct kinetic profiles
between tumors and healthy tissue, and significantly higher metabolic
activity in control tumors by measurements of HDO (A), [^2H[7]]glucose
(B), ^2H-lactate (C), and HDO/glucose ratios (D). (E to H) Tumor volume
normalized data shows clear differences between the metabolic activity
observed in control and treated tumors. Data are presented as the
means ± SD of N = 3 biological replicates. Statistical significance was
established by Student’s t test and is labeled as *P ≤ 0.05, **P≤ 0.01,
and ***P ≤ 0.001. Statistical marks are presented for differences
between control and treatment tumors, other comparisons were not
significant. A.U., arbitrary unit.
[^2H[7]]glucose DMRI is highly correlated with central metabolism
To assess the fate of [^2H[7]]glucose derived ^2H, we assessed
fractional enrichments, metabolite pool sizes, and metabolic flux
through GC-MS analysis of tumors harvested after each imaging session.
Similar to cell extract samples, tumoral measurements demonstrated
significant treatment-induced decreases in m+1 and/or m+2 glycolytic
and TCA cycle intermediates including lactate, alanine, glutamate, and
succinate ([121]Fig. 5, A to F). Pool size measurements showed
significantly lower lactate in treated tumors and insignificant but
modest decreases in glutamate and succinate (fig. S2B).
Lactate-to-alanine ratios were significantly higher in control tumors
(~39) compared to treatment (~30) (fig. S3B). INCA was used to generate
a metabolic model ([122]Fig. 1N and table S2) of tumoral flux that used
metabolite fractional enrichments as inputs. Flux measurements were
derived relative to glucose import to account for tracer uptake.
Tumoral flux analysis revealed that treatment led to significantly
reduced ^2H liberation, citrate synthase (CS), and malate
dehydrogenases (MDHs) turnover along with evident decreases in
glycolytic reactions that contribute to HDO production through
triosephosphate isomerase (TPI), glyceraldehyde-3-phosphate
dehydrogenase (GAPDH)–enolase, and pyruvate kinase (PK) ([123]Fig. 5G).
Fig. 5. Tumor treatments cause robust changes to central metabolism,
correlating with overall observed HDO production.
[124]Fig. 5.
[125]Open in a new tab
(A to F) Fractional enrichment analysis revealed treatment induced
reductions in labeling of glycolytic and TCA cycle intermediates. (G)
Metabolic flux analysis demonstrates extensive decrements in glycolytic
and TCA cycle reactions involved in HDO production. (H to K) Tumoral
^2H signal and HDO measurements are strongly and positively correlated
with lactate and succinate m+1 enrichments. These results demonstrate
that the deuterium metabolic imaging treatment monitoring method
presented is highly sensitive to tumor metabolic activity. Two-way
ANOVA was used to establish significance across treatments, metabolite
pool sizes, specific mass ion species, and flux measurements.
Significant changes are labeled as *P ≤ 0.05, **P≤ 0.01, and
***P ≤ 0.001. Where not visible, SD error bars are behind symbols.
Correlation analysis was performed to establish the statistical
significance of the correlations.
Total ^2H signal imaging (^2H FLASH) was demonstrated to be highly
sensitive and correlated to treatment-induced reductions in tumor
metabolic activity ([126]Fig. 5, H to K). Measurements of total tumoral
^2H signal were positively correlated with lactate m+1 and succinate
m+1 fractional enrichments, with r values of 0.81 and 0.81,
respectively ([127]Fig. 5, H and I). Stronger correlations were
observed between tumoral HDO measurements and fractional enrichments
(lactate m+1: r = 0.89 and succinate m+1: r = 0.97) ([128]Fig. 5, J and
K). In addition, these disruptions in central carbon metabolism were
further supported by global metabolomics analysis of tumor extracts,
which showed similar trends as the cultured cells, including
treatment-induced reductions in glycolytic and TCA cycle intermediates
and related pathways (fig. S4, D to F).
[^2H[7]]glucose generates greater SNR and HDO than [6,6′-^2H[2]]glucose
Given the strong readout of HDO production from [^2H[7]]glucose
utilization observed in vivo, we compared its performance to
[6,6′-^2H[2]]glucose by injecting equal doses of each tracer (1.95
g/kg) into different cohorts of tumor bearing mice and monitoring ^2H
signal SNR and HDO production. Total ^2H signal images acquired with a
^2H gradient echo sequence demonstrates that [^2H[7]]glucose can
generate spatial maps of glucose utilization with greater signal and
spatial resolution than [6,6′-^2H[2]]glucose ([129]Fig. 6A). Images
acquired after [^2H[7]]glucose injections showed distinct spatial
localization resulting from organ-specific metabolism as observed by
signal colocalized with kidneys, abdominal organs, and flank tumors. As
a control of the spatial distribution of [^2H[7]]glucose and a proxy of
perfusion, we injected a third cohort of mice with 12.5% D[2]O (5 μl/g
body weight), a freely diffusible tracer. In contrast to
[^2H[7]]glucose images, total ^2H signal images acquired after D[2]O
injections showed no apparent spatial localization and signal appeared
to be randomly distributed across the abdomen and tumor of the animals.
ROI-based quantification of [^2H[7]]glucose and [6,6′-^2H[2]]glucose
images allowed for an SNR analysis based on the mean signal intensity
from a tumor region divided by the SD of the same noise region on the
top left hand of all images. SNR calculations demonstrated that
[^2H[7]]glucose led to ~2× greater SNR compared to [6,6′-^2H[2]]glucose
([130]Fig. 6B). Slice-localized spectroscopy acquired throughout each
imaging session showed that tumors administered with [^2H[7]]glucose
generated ~2.7× more HDO by 6 min post-injection than those
administered with [6,6′-^2H[2]]glucose ([131]Fig. 6C). Nonlocalized
single-pulse spectroscopy sampled some healthy tissue and tumoral
metabolism, reporting significantly greater HDO production from
[^2H[7]]glucose than [6,6′-^2H[2]]glucose ([132]Fig. 6D).
Fig. 6. HDO production from [^2H[7]]glucose significantly improves SNR
compared to [6,6’-^2H[2]]glucose.
[133]Fig. 6.
[134]Open in a new tab
(A) Three cohorts of animals were each injected either [^2H[7]]glucose
(1.95 g/kg), [6,6′-^2H[2]]glucose (1.95 g/kg), or 12.5% D[2]O saline
(5 μl/g). Images for each injection are shown as follows: ^1H RARE
images for anatomical referencing, pre-injection ^2H gradient echo
images, and post-injection ^2H gradient echo images. (B) ROI-based
quantification of tumoral signal mean intensity and noise standard
deviation of [^2H[7]]glucose (N = 4) and [6,6′-^2H[2]]glucose (N = 4)
images showed that [^2H[7]]glucose led to significantly greater SNR.
(C) Slice-localized ^2H spectroscopy demonstrated that tumors generated
significantly greater HDO when using [^2H[7]]glucose (N = 4) compared
to [6,6′-^2H[2]]glucose (N = 3). (D) Nonlocalized single-pulse
spectroscopy also showed greater HDO production from [^2H[7]]glucose
(N = 4) compared to [6,6′-^2H[2]]glucose (N = 3). All spectra here were
quantified by dividing each peak by the pre-injection HDO peak area
followed by a dose normalization. Student’s t test was used to
establish significance for SNR and spectroscopy data. Significant
changes are labeled as *P ≤ 0.05 and **P≤ 0.01. Where not visible, SD
error bars are behind symbols.
Table 1. List of Abbreviations.
[135]Fig. 1N Label Abbreviation Enzyme/Flux
v[2] HK Hexokinase
v[6] GPD Glycerol-3-Phosphate Dehydrogenase
v[7] TPI Triosephosphate Isomerase
v[9] GAPDH Glyceraldehyde-3-Phosphate Dehydrogenase
v[11] PK Pyruvate Kinase
v[12] ALT Alanine Aminotransferase
v[13] LDH Lactate Dehydrogenase
v[14] PDH Pyruvate Dehydrogenase
v[15] PC Pyruvate Carboxylase
v[16] CS Citrate Synthase
v[18] GDH Glutamate Dehydrogenase
v[19] GS Glutamine Synthase
v[20] SDH Succinate Dehydrogenase
v[21] FH Fumarate Hydratase
v[22] MDH Malate Dehydrogenase
[136]Open in a new tab
DISCUSSION
Our findings demonstrate that DMRI with [^2H[7]]glucose can be used as
a metabolically sensitive paradigm for cancer detection and treatment
monitoring in a safe and highly specific manner. We developed a robust
imaging protocol and demonstrated its sensitivity to tumor metabolic
activity by testing it with an established in vivo cancer model known
to acutely perturb glycolysis. We acquired images of tumoral glucose
utilization with metabolic contrast by imaging total ^2H signal and
measuring HDO production after [^2H[7]]glucose administration with
straightforward, robust, and widely applicable methods. On the basis of
our promising findings, we believe that this approach, using
alternative deuterated tracers and measuring HDO production, can be
used to assess and monitor a variety of diseases both before and during
treatment.
To discern the key biochemical consequences of BRAFi and MEKi treatment
on HDO production, we characterized the effects of treatment on
[^2H[7]]glucose utilization in Yumm1.7 cells ([137]Fig. 1 and fig. S1).
We observed that HDO production was strongly correlated with
[^2H[7]]glucose consumption, indicating that HDO can serve as a
reliable reporter of glucose utilization in vitro (fig. S1I).
[^2H[7]]glucose metabolism generates HDO through three-carbon
glycolytic intermediate flux and through the utilization of deuterated
acetyl-CoA ([138]18). Given the fast rate of glycolytic metabolism and
the relatively small intermediate pool sizes compared to the bolus of
[^2H[7]]glucose ([139]30), ^2H spectroscopy allowed for the detection
of glucose resonances with minimal contributions from intermediates. In
addition, given the numerous reactions that contribute to HDO
production, the final concentration of HDO detected in control cell
culture was 8× higher than that of ^2H-lactate, highlighting the
robustness of monitoring HDO production ([140]Fig. 1, A to F).
^2H NMR data were further substantiated by GC-MS–based fractional
enrichment analysis of Yumm1.7 cell extracts. As expected from previous
reports, the overall enrichment of glycolytic and TCA cycle
intermediates corroborates observed decreases in HDO production in a
treatment-specific manner ([141]Fig. 1, G to M). Note that unlike ^13C
tracing, ^2H tracing is sensitive to label loss through processes such
as isomerization, keto-enol tautomerization, and aminotransferase
activity. For example, ^2H-lactate enrichment can be diluted by label
loss through pyruvate keto-enol tautomerization and any pyruvate
derived from ALT activity, which liberates a ^2H of the methyl group.
Furthermore, pool sizes matched treatment-specific decrements observed
in metabolite labeling except for alanine, which was detected at higher
levels in treated cells compared to control (fig. S2).
Lactate-to-alanine ratios were significantly reduced by treatment,
revealing a metabolic reversion to the state of healthy cells and away
from Warburg metabolism (fig. S3). Absolute metabolic flux modeling
based on cell fractional enrichments and glucose utilization rates
revealed treatment-dependent decreases in reactions that directly
contribute to HDO production including TPI, GAPDH to enolase, and ALT
([142]Fig. 1, N and O). Reduced flux in TPI, LDH, and GPD is
corroborated by literature on BRAF mutant melanomas. In addition to
ERK/pERK, cMYC and hypoxia-inducible factor 1α (HIF1α) are also
downstream targets of BRAF mutant hyperactivity. cMYC and HIF1α
regulate lactate dehydrogenase (LDH) expression, and HIF1α regulates
aldolase expression, controlling the substrate pool for TPI and GPD
([143]20). Therefore, the reduction in MAPK signaling induced by BRAFi
and MEKi treatment is predicted to reduce cMYC and HIF1α activity,
culminating in reduced LDH, TPI, and GPD fluxes. The treatment-induced
reduction of ALT flux ([144]Fig. 1O) can account for decreased labeling
and increased pool size, along with an increase in unlabeled alanine.
Consistent with ^2H NMR data and fractional enrichments, flux analysis
showed decrements in the major central metabolic steps of LDH, PDH, CS,
and MDH, which together highlight decreases in glycolysis and TCA cycle
activity that contribute to the observed decrease in HDO production
([145]Fig. 1O).
Within the context of [U-^13C]glucose utilization, it has already been
demonstrated that treating BRAF V600E mutant melanomas with BRAFi and
MEKi leads to decreased glycolysis, contributing to decreased TCA cycle
labeling ([146]20, [147]26). As shown in [148]Fig. 1G, the utilization
of [^2H[7]]glucose can potentially liberate six deuterium labels as HDO
molecules through glycolysis alone, with the final release of the
remaining ^2H nuclei guaranteed through TCA cycle activity. Although
nonlocalized single-pulse spectroscopy analysis showed trends of
increased metabolic activity in control tumor bearing animals, these
increases were not significantly different from treated animals (fig.
S6). This is likely due to the nonspecific coverage of nonlocalized
pulse-acquire spectra. Single-pulse spectroscopy acquires spectra for
any tissues within the detection volume of the coil, thus these
measurements account for tumor and healthy tissue metabolic activity.
Overall, in vivo ^2H spectra generated similar line shapes to that of
[6,6′-^2H[2]]glucose. High-resolution spectroscopy can allow for the
clear distinction of [^2H[7]]glucose and [6,6′-^2H[2]]glucose with the
former having a complex superposition of peaks covering more parts per
million (ppm) than the simple singlet of [6,6′-^2H[2]]glucose. However,
the line broadening effects observed in vivo merge the multiple peaks
of [^2H[7]]glucose into what appears to be a single peak similar to
that of [6,6′-^2H[2]]glucose.
Compared to ^2H-lactate detection from [6,6′-^2H[2]]glucose, HDO
production from [^2H[7]]glucose has greater sensitivity because even
before tracer is administered, the NA HDO provides a fairly good signal
as a baseline. In the case of ^2H-lactate, a certain length of time
must elapse before enough SNR is generated to observe the resonance.
This becomes a limitation to early and fast kinetics as monitoring
lactate would be challenged by the accumulation of sufficient signal
for reliable MRI/S detection. Note most data published to date with
[6,6′-^2H[2]]glucose is acquired with oral administration of the
glucose and then a waiting period before the imaging ([149]13,
[150]31–[151]34). This time delay is needed before lactate can be
easily observed. From a metabolic standpoint, HDO is a more sensitive
readout as it is produced at more steps, whereas [6,6′-^2H[2]]glucose
can only label half of glycolytically derived pyruvate (lactate),
decreasing the overall sensitivity.
Our comparative in vivo analysis demonstrated that [^2H[7]]glucose
imaging generated ~2× more SNR than [6,6′-^2H[2]]glucose imaging for
equivalent weight of tracer administered ([152]Fig. 6B). In addition,
we show that ~2.7× more HDO can be produced from [^2H[7]]glucose
compared to [6,6′-^2H[2]]glucose ([153]Fig. 6C). The improved signal
from [^2H[7]]glucose fast imaging allowed for high metabolic contrast,
which could not be achieved with [6,6′-^2H[2]]glucose. The spatial
localization and distribution of signal was completely different when
an approximately equivalent dose of ^2H nuclei content as the
[^2H[7]]glucose injection (~2 M) was administered with 12.5% D[2]O (~7
M) into tumor-bearing animals ([154]Fig. 6A). Unlike the organ-specific
localization of signal observed in [^2H[7]]glucose images, D[2]O images
demonstrated nonlocalized signal randomly distributed across the
abdominal organs, skeletal muscle, and tumors of the animals. This
suggests that unlike the freely diffusible tracer, D[2]O,
[^2H[7]]glucose is being used and handled by specific organs to
different degrees. For example, images from [^2H[7]]glucose injections
showed negligible signal within the skeletal muscle surrounding the
spine, which is likely due to the muscle relaxant effects of isoflurane
anesthesia ([155]Fig. 6A) ([156]35). However, images after D[2]O
administration demonstrated signal within the skeletal muscle, thus
highlighting the specificity of [^2H[7]]glucose utilization ([157]Fig.
6A). The presence of D[2]O signal within the tumoral region indicates
that tumors are perfused and that the signal generated after
[^2H[7]]glucose administration depends upon perfusion, uptake, and
metabolism.
To isolate specific tissue types, we used slice-localized spectroscopy
and found not only that tumors in control animals generated greater HDO
than tumors in drug-treated animals but also that contralateral healthy
tissue exposed to drug treatment had reduced metabolic activity
compared to control healthy tissue ([158]Fig. 4, A to D). These results
suggest that treatment is affecting the metabolism of both tumors and
healthy tissues, but with a greater effect in tumors. In addition, our
method is sensitive to the metabolic differences between tumors and
healthy tissue. Tumors had a distinct kinetic profile for HDO
production that involved rapid initial HDO production that reached a
maximum level at ~5.5 min and remained stable for 1 hour ([159]Fig.
4A). In contrast, healthy tissue demonstrated linear kinetics
throughout the course of the experiments with ~5.8× less HDO produced
at early time points compared to tumors ([160]Fig. 4A). This
observation highlights the ability of our imaging methodology to
distinguish tumors from healthy tissue with high specificity and
sensitivity gained from rapid detection of high SNR HDO signal produced
from [^2H[7]]glucose.
The use of perdeuterated glucose maximizes total HDO production
compared to other deuterated glucose analogs. In terms of financial
cost, [^2H[7]]glucose is ~2× to 3× more costly than doubly and singly
deuterated analogs. There are two main approaches that can be taken to
minimize the cost of deuterated glucose tracer. One is to continue to
advance isotopic hydrogen-exchange techniques, which are more
advantageous than approaches using biotechnology such as culturing
algae in D[2]O, which require biological expertise as well as extensive
and complicated extraction processes ([161]36). Another option would be
to explore this methodology with [2,3,4,6,6′-^2H[5]]glucose which has
substrate costs estimated to be 10× less expensive than the deuteride
substrates used for [6,6′-^2H[2]]glucose and [^2H[7]]glucose synthesis
([162]37). Given the high deuterium content of [^2H[5]]glucose, it
should generate similar SNR gains as [^2H[7]]glucose while providing
identical downstream labeling information.
Another consideration regarding the feasibility of [^2H[7]]glucose
clinical imaging is the ideal dose and mode of administration. The
glucose doses used (1.95 g/kg) in this work were the same as the Yale
studies performed by de Feyter et al. ([163]13). However, other studies
in humans have been able to generate metabolic contrast with doses as
low as 0.75 to 0.8 g/kg ([164]31, [165]38). A dose-dependent imaging
study will have to be performed to establish the ideal dose. Another
aspect of administration is the mode, intravenous or oral. Oral glucose
can lead to an incretin effect, which is a greater insulin response
compared to intravenous administration. In addition, oral dosing
requires more time for the tracer to be delivered throughout the body,
which can somewhat limit the assessment of fast kinetics. Furthermore,
the assessment of metabolism after oral administration of glucose can
be complicated by contributions of the gut microbiota, which also
consume the tracer. The gut microbiome is essentially bypassed by
intravenous administration. Some of the next steps for clinical
translation of this imaging exam would be to quantitatively assess the
differences between oral and intravenous glucose administration and
which kinetics would allow for the acquisition of maximal signal. Oral
administration is the preferred method as it is straightforward and
noninvasive; however, the delayed onset kinetics may miss early kinetic
contrast generated by local HDO production. Once the administration
avenue is established, several models of cancer and therapies will have
to be assessed to build a reference of benchmarks for determining
effective treatment by quantifiable reductions in HDO production.
The biochemical characteristics of [^2H[7]]glucose utilization were
leveraged to generate striking metabolic contrast in tumors by imaging
in vivo tumor signal production with DMRI. We achieved requisite image
acquisition with simple yet powerful gradient echo imaging sequences.
These sequences allowed for high in-plane and interpolated resolutions
of 1.09 and 0.27 mm^2, respectively. Total ^2H FLASH metabolic images
allow for tumoral imaging assessing the effects of treatment with
sensitivity to tumoral metabolic activity ([166]Figs. 2 and [167]3).
Tumoral HDO signal acquired by slice-localized spectroscopy had the
strongest correlation with lactate and succinate m+1 enrichments
compared to ^2H signal, thus highlighting the specificity of our method
([168]Fig. 5). In corroboration, overall fractional enrichments, pool
sizes, flux rates, and metabolic profile changes in tumor samples
supported HDO production as a faithful reporter of tumor central
metabolic activity ([169]Fig. 5 and figs. S3 and S4). In the case of
tumors, metabolic modeling was limited to deriving relative flux
values, as opposed to absolute flux values, as another normalization
paradigm such as estimates of O[2] consumption was not available.
Unlike cultured cells, tumors displayed nonlinear kinetics in which HDO
production rose to approximately max levels within 5 min and was
maintained for 1 hour. To eliminate bias from the model, we chose to
not use the in vivo tumoral kinetic rates as inputs for deriving
absolute flux values. Instead, we derived relative flux values that
agreed with all observations about the metabolic effects of treatment.
However, these values contain some bias, as they were estimated,
assuming an arbitrary 100% glucose import rate.
Moreover, this method is sensitive to treatment induced changes in
metabolism. Analysis of pre-injection signal of day 0 and day 3 mice in
^2H FLASH images shows that our method can be performed serially with
negligible pre-injection signal contribution 3 days
post-[^2H[7]]glucose administration that can be easily accounted for
with pre-injection HDO normalization (fig. S7). The rapid rate of ^2H
signal clearance post-[^2H[7]]glucose administration may more readily
allow the serial imaging of the same patient compared to methods that
depend on the gradual incorporation of ^2H label over a multiple day
period ([170]39). With the acquisition of appropriate ^2H MRI coils for
humans, ^2H FLASH sequences can be readily implemented into existing
clinical MRI systems with minimal technical challenges as demonstrated
by ^2H FLASH imaging for treatment monitoring at 1.5 T and numerous
[6,6′-^2H[2]]glucose studies performed at 3 T ([171]32–[172]34,
[173]39). The ^2H FLASH sequence used in this work implemented total
scan times similar to clinical ^2H imaging at 3 T. The total scan time
in this study for 256 averages was ~13 min. Each acquisition took ~3 s,
but 256 averages were acquired to increase SNR. In comparison to 3 T,
acquisition at 11 T allows for images with much higher resolution.
Sequences were optimized to achieve maximal SNR; however when compared
to 3 T, the higher resolution and the shorter T[2] would offset the
benefits of a smaller coil and higher field. We have also noted through
many experiments that increasing tissue mass exerts primary control
over the SNR of experiments using isotopes to estimate metabolic flux.
A whole human brain turns over a substantially higher amount of glucose
(μmols/min) than a ~1 cm tumor in a mouse flank. In addition, the
spectral quantification of 3 T data should be as straightforward as
data acquired at high field because 11 T data presented here was
unperturbed by the minor contributions of the α and β anomeric peaks.
In this initial iteration of treatment monitoring using the
[^2H[7]]glucose contrast agent, we chose to focus on developing our
method in the most straightforward and reproducible manner. Therefore,
we selected a flank tumor model as it allowed for simple
implementation, easy detection of tumors, and ample residual tumor
tissue for downstream analysis. To further document the robustness of
this approach, future studies, can focus on an intrinsically derived
orthotopic tumor model, as it may better model metabolic differences
between tumors and adjacent non-malignant tissue as well as allow for
testing the utility of this approach for the initial detection of
tumors. Alternative tracers can also be explored, as HDO production
from [^2H[15]]octanoate has already been used as a quantitative imaging
probe for dysregulated lipid metabolism ([174]40), which is heavily
implicated in cancer.
In addition, we can improve upon our imaging methodology by
implementing chemical selectivity through fast chemical shift imaging
(CSI). CSI is a well-known and regularly used MRI experiment but
requires long acquisition times to acquire sufficient SNR. An
alternative to this technique is already under way as efforts are being
made to develop ^2H compressed-sensing CSI (CS-CSI) sequences that will
allow for the fast acquisition of chemically sensitive images by under
sampling image acquisition to achieve greater pixel averages and higher
SNR. CSI-based approaches for ^2H imaging have already shown notable
promise for generating chemically selective images with rapid
acquisition ([175]41–[176]44).
We demonstrate the highly sensitive detection of differential glucose
utilization across baseline, control, and treated tumors by measuring
total ^2H signal and HDO production. Gradient echo imaging (^2H FLASH)
generated total ^2H metabolic maps of colocalized HDO and
[^2H[7]]glucose, indicating that this method can be sensitive to
perfusion/vascularity effects and metabolic effects. For example, at
early time points (0 to 30 min), slices from the contralateral healthy
tissue generated less HDO than tumor tissue, indicating that
circulating HDO generated from the [^2H[7]]glucose agent is not driving
the hyperintensity in the tumor. In addition, our in vitro modeling
experiments, which are free of perfusion effects, show clear metabolic
effects associated with treatment. In vitro, both control and treated
cells were cultured with the same volume of media with equal
concentrations of tracer. Cell extract analysis demonstrated reduced
labeling in glycolytic and TCA cycle intermediates in treated cells.
INCA modeling analysis accounted for reduced glucose import in treated
cells and still demonstrated significantly reduced glycolytic flux in
line with reduced HDO production. This suggests that even in the
presence of perfusion effects, our model is still sensitive to
metabolic effects. Nevertheless, future studies using this model should
incorporate perfusion-based imaging to allow for the clear distinction
of the perfusion/vascularity effects of a treatment ([177]45).
Collectively, our findings show that HDO is a quantitative marker of
tumor glucose utilization in vivo, and the imaging technique has
translational potential in humans due to its safe classification,
noninvasive oral administration, and suitability for routine treatment
monitoring. Given the high in-plane (1.09 mm^2) and interpolated (0.27
mm^2) resolution achieved in this mouse study, the application of the
current methodology to humans should lead to exemplary resolution
without any substantial advancements in radio frequency coil
technology, as this study used simple linear coil configurations. The
development of human ^2H MRI coils for this technique is already
underway and will allow for a future assessment of scalability of this
approach in humans. In addition, the DMRI of HDO could lead to the
development of more personalized treatment strategies mediated through
serial imaging to inform a continuous reassessment regimen that
establishes dosing based on the treatment response of the tumor. It is
encouraging to note that imaging ^2H signal with a simple gradient echo
sequence was highly sensitive in differentiating between the metabolic
activity of control and treated tumors. This suggests that the
methodology could provide a useful alternative to FDG-PET, with
increased specificity from the detection of downstream metabolic
products, without the use of ionizing radiation.
MATERIALS AND METHODS
Experimental design
We developed a robust imaging platform and tested the method by
assessing therapeutic efficacy in a mouse flank tumor model. C57BL/6J
mice were injected subcutaneously with highly glycolytic Yumm1.7 cells,
which model late-stage human melanoma expressing constitutively active
BRAF V600E protein. In vitro characterization demonstrated that
treating Yumm1.7 cells with a combination of dabrafenib (BRAF
inhibitor) and trametinib (MEK inhibitor) significantly reduced MAPK
signaling as well as HDO production from [^2H[7]]glucose. For in vivo
treatment experiments, tumors were grown to ~10 mm in diameter, and
mice were subsequently imaged to establish a baseline followed by oral
gavage treatment with 600 μg of dabrafenib and 6 μg of trametinib for 3
days after which mice were imaged again to assess the effects of
treatment. For imaging sessions, we intravenously injected with
[^2H[7]]glucose (1.95 g/kg) and imaged the mice on an 11.1 T scanner
with a deuterium FLASH (^2H FLASH) sequence to acquire total ^2H
signal. After experiments, cells and tumors were harvested and analyzed
by GC-MS to establish pool sizes, fractional enrichment, and metabolic
models for flux measurements.
Cell lines and cell culture reagents
Yumm1.7 cells were provided by the Children’s Medical Center Research
Institute at the University of Texas Southwestern Medical Center.
Glucose-free Dulbecco’s modified Eagle’s medium/Ham’s F-12 nutrient
mixture (DMEM/F-12) was purchased from US Biological (Salem, MA, USA).
Glucose-free medium was used to administer the glucose tracer to cells.
For regular cell maintenance, 25 mM unlabeled d-glucose was
supplemented. Deuterated pyrazine ([^2H[4]]pyrazine), and d-glucose
were purchased from Sigma-Aldrich, (St. Louis, MO, USA).
[1,2,3,4,5,6,6-^2H[7]]-d-glucose ([^2H[7]]glucose) was purchased from
Cambridge Isotope Laboratories (Andover, MA, USA). Fetal bovine serum
(FBS) was purchased from Atlas Biological (Fort Collins, CO, USA).
Phosphate-buffered saline (PBS), penicillin/streptomyocin (P/S),
nonessential amino acids (NEAAs), and dimethyl sulfoxide (DMSO) were
purchased from Thermo Fisher Scientific (Waltham, MA, USA). Dabrafenib
and trametinib were purchased from Selleck Chemical (Houston, TX, USA).
Isoflurane anesthesia was obtained from Patterson Veterinary (Ocala,
FL, USA). A 0.9% sterile filtered saline was prepared from NaCl
purchased from Thermo Fisher Scientific (Waltham, MA, USA) and
constituted with 5% sterile heparin (50 USP/ml, final concentration),
to prevent clotting. Heparinized saline was used to dissolve the
[^2H[7]]glucose solution before administration to the animals. All
reagents were sterile-filtered before administration to cells or
animals.
Cell culture dose response and Western blot analysis
Yumm1.7 cells were treated with a dosing curve of combinatorial
treatment (dabrafenib BRAFi and trametinib MEKi) ranging from low to
high concentrations. The respective concentrations of BRAFi and MEKi
utilized were 0.1 and 0.005 μM for low, 1.5 and 0.07 μM for medium, and
20 and 1 μM for high. After treating the cells for 24 hours, the cells
were probed for pERK and ERK protein expression by Western blot
analysis. Cells were harvested and immediately lysed with ice-cold
radioimmunoprecipitation assay buffer supplemented with halt protease
and phosphatase inhibitors (Thermo Fisher Scientific, Waltham, MA,
USA). Protein concentrations were then established by a Bradford assay
(Bio-Rad, Hercules, CA, USA) for normalization before gel loading.
Protein-normalized cell lysates were loaded in a 4 to 20% Criterion TGX
gel (Bio-Rad, Hercules, CA, USA), and proteins were separated by
applying a constant 0.05 A to the gel. After the elution of the loading
dye front, the gel was immediately transferred to a polyvinylidene
fluoride membrane (Bio-Rad, Hercules, CA, USA) at a constant 50 V at
4°C for 1 hour. The membrane was then stained with fast green
(Sigma-Aldrich, St. Louis, MO, USA) to ensure even protein loading.
Then, the membrane was destained, washed, and blocked with 5% nonfat
dry milk dissolved in TBS-T [200 mM sodium chloride (NaCl), 30 mM
Tris-hydrochloric acid (HCl) (pH 7.6), and 0.1% Tween 20] for 1 hour.
Proteins were probed with primary pERK (mouse monoclonal, Santa Cruz
Biotechnology, Dallas, TX, USA), ERK (rabbit polyclonal, Santa Cruz
Biotechnology, Dallas, TX, USA), and α-tubulin (mouse monoclonal, Cell
Signaling Technology, Danvers, MA, USA) antibodies. Primary antibodies
were diluted [pERK, 1:200 (v/v); ERK, 1:500 (v/v); and α-tubulin,
1:5000 (v/v)] with 5% nonfat dry milk dissolved in TBS-T and
independently incubated with the blot for 18 hours at 4°C. After
incubation with each primary antibody, the membrane blot was thoroughly
washed and incubated with either secondary anti-mouse horseradish
peroxidase (HRP) antibody [1:5000 (v/v)] or secondary anti-rabbit HRP
antibody [1:10,000 (v/v)] at room temperature for 1 hour. Once probed
for secondary antibody, the blot was washed and reacted with PierceTM
ECL Western Blotting Substrate (Thermo Fisher Scientific, Waltham, MA,
USA) and then exposed to PRIMA1 autoradiography film (Midwest
Scientific, Valley Park, MO, USA). Protein expression was quantified by
densitometry analysis on ImageJ (National Institutes of Health,
Bethesda, MD, USA).
Cell culture treatment and [^2H[7]]glucose tracer administration
Yumm1.7 cells were maintained with complete growth medium composed of
DMEM/F-12 with 10% (v/v) FBS, 1% P/S, and 1% NEAA. Cell lines were
maintained at 37°C with 95% air and 5% total CO[2] in an air-jacketed
incubator (Heracell Vios 160i, Thermo Fisher Scientific, Waltham, MA,
USA). Every 2 to 3 days, complete medium was replaced, and once at 60%
confluence, cells were subcultured 1:6 into eight 100-mm cell culture
dishes. All cell lines were grown to 60% confluency and washed once
with warm PBS and incubated with 10 ml each of either DMEM/F-12 with
1.5 μM BRAFi and 0.07 μM MEKi or DMEM/F-12 with a DMSO vehicle for
18 hours. After the 18-hour time point, all DMEM/F-12 was aspirated,
cells were washed once with warm PBS and then incubated with 5-ml each
of DMEM/F-12 containing the same concentrations of BRAFi and MEKi along
with 11 mM [^2H[7]]glucose for 6 hours, withdrawing 200 μl of aliquots
at 0 min, 30 min, 2 hours, and 6 hours. Immediately after the 6-hour
time point, cells were washed, trypsinized, and harvested for
subsequent ^2H NMR and GC-MS analysis.
^2H NMR sample preparation
All media samples were processed and prepared without extraction. Each
media sample was spiked with [^2H[4]]pyrazine internal standard to a
nominal concentration of 2.5 mM. The [^2H[4]]internal standard allowed
for the quantitation of HDO, ^2H-lactate, and residual [^2H[7]]glucose
from ^2H NMR data. A total of 180 μl of cell media sample for each time
point was loaded into 3-mm NMR sample tubes for NMR analysis. Cells
were extracted with acetonitrile:isopropanol:water [3:3:2 (v/v/v)],
dried down, and further refined with acetonitrile:water [1:1 (v/v)].
Dried extracts were then reconstituted in water and transferred to 3-mm
NMR tubes for NMR analysis.
^2H NMR analysis
A Bruker Bio-Spin 18.8 T magnet system equipped with an Avance III
Console and 5-mm TXI CryoProbe and TopSpin 4.0.3 was used for ^2H NMR
data acquisition of [^2H[7]]glucose, HDO, ^2H-lactate, and ^2H-glx in
the cell and media samples. The deuterium lock channel was used for ^2H
NMR spectra acquisition at 122.79-MHz resonant frequency. An
acquisition time of 2 s and a relaxation delay of 1 s (total, 3 s of
repetition time) with a 90° pulse were used for acquisition. The 8190
complex data points were digitized with an 11-ppm spectral width and
256 scans for each of the three free induction decays (FIDs) (768
scans) for all samples. Data acquisition took place at room temperature
(25°C).
After acquisition, ^2H NMR spectra were further processed with
MestReNova v14.0.1-23284 (Mestrelab Research SL). ^2H NMR spectra were
processed by setting the exponential window function to 0.5 Hz and
increasing the zero filling of the FID to 16,384 data points performing
fourier transformation. Automatic phase and spline baseline corrections
were performed for each spectrum. All ^2H NMR spectra acquired for each
sample were aligned with respect to the [^2H[4]]pyrazine peak and
summed to account for peak shifting caused by magnetic field drift
during experimental acquisition. Concentrations of ^2H-labeled
metabolites in cell and media samples were calculated using the
internal standard peak area of [^2H[4]]pyrazine. The known
concentration of [^2H[4]]pyrazine (2.5 mM) was leveraged for the
quantification of HDO and residual [^2H[7]]glucose concentrations,
which were normalized to the total number of deuterium nuclei
responsible for the respective peak resonances.
GC-MS sample handling and analysis
GC-MS sample preparation (cell, media, and tumor samples), instrument
analysis, data processing, and metabolomics statistical analysis were
performed exactly as previously published ([178]19, [179]46–[180]49).
Metabolomic profiling was based on the metabolite panel presented in
table S1. The Thermo Scientific Single Quadrupole Mass Spectrometer
(ISQ) and Gas Chromatograph (TRACE 1310) were used for all analyses.
Fractional enrichment and metabolic modeling analysis
Mass isotopomer distributions (MIDs) were generated by integrating the
extracted ion chromatograms of each metabolite identified by GC-MS
using QuanBrowser on XCalibur 4.5 (Thermo Fisher Scientific, Waltham,
MA, USA). Metabolite MIDs were further processed by NA correction using
the INCA software (INCA 2.0, Vanderbilt University, Nashville, TN, USA)
([181]50, [182]51). INCA was also used to develop a mathematical
metabolic flux model of [^2H[7]]glucose utilization to estimate the
absolute and relative flux measurements of pathway contributions to HDO
production. These reactions are detailed in table S2. The model was
constructed to mimic the metabolic processes of glucose import,
glycolysis, TCA cycle, and key components of shuttle systems such as
the malate-aspartate shuttle and citrate-malate mitochondrial carrier.
In addition, it accounted for sources of acetyl-CoA such as fatty acid
oxidation. Specific steps generating reduced form of nicotinamide
adenine dinucleotide or FADH[2] were included, followed by the assumed
consumption of O[2] through oxidative phosphorylation. Metabolic
intermediates not identified through GC-MS analysis were omitted from
the model, for instance, flux from citrate to α-ketoglutarate was
simplified as a single step. Metabolite fractional enrichments were
calculated by inputting the mass isotopomer distribution of specific
metabolites from mitochondrial experiments and their chemical formulas
into INCA’s NA correction function, as previously described.
In vivo flank tumor imaging model
All experimental procedures on animals were approved by the University
of Florida Institutional Animal Care and Use Committee under protocol
202400000077. Male C57BL/6J mice were purchased from the Jackson
Laboratories and acclimated in the animal housing facility. For in vivo
experiments, tumors were grown to ~10 mm in diameter, and mice were
subsequently imaged to establish a baseline followed by oral gavage
treatment with 600 μg of dabrafenib and 6 μg of trametinib for 3 days
after which mice were imaged again to assess the effects of treatment.
During treatment, tumor volume was monitored by ^1H MRI (days 0 and 3)
and caliper measurements (days 1 and 2). Caliper data were quantified
by calculating the volume approximations (V = 0.5 × l × w^2) of the
tumors. For imaging sessions, animals were anesthetized with 5%
isoflurane and maintained under anesthesia until the end of the
session. Animals were tail vein catheterized and hydrated with 0.9%
saline containing 5% heparin. Animal body temperature was maintained at
37°C with heated circulating water lines, and respiration was monitored
with a respirometer for the entire extent of the imaging session. Once
shimming and pre-injection images were acquired, the animals were
injected with [^2H[7]]glucose (1.95 g/kg) for at a rate of 50 μl/min
(~2 min). ^1H anatomical MRI images were used to determine tumor volume
measurements by calculating the approximate hemi-ellipsoid volumes of
tumors (V = π/6 × a × b × c). At the end of imaging sessions, animals
were humanely euthanized, and tumors were extracted, weighed, and
stored for subsequent analysis. We present estimates of tumor volume
using caliper measurements of days without the MR imaging portion of
the paradigm and the more accurate ^1H MRI to produce the correlations
with the metabolic imaging data. While these two estimates varied
slightly in scale, the relative changes between control and treatment
tumors matched the expected changes in growth through the course of
therapy.
Another subset of animals was used to test a comparison between
[^2H[7]]glucose, [6,6′-^2H[2]]glucose, and D[2]O injections. The
imaging protocol was followed exactly as detailed above including
dosage which was kept at 1.95 g/kg for both glucose tracers. Similarly,
D[2]O experiments followed the same protocol, but imaging sessions were
concluded immediately after the ^2H FLASH image. D[2]O injections
(12.5%) were administered with 0.9% heparinized saline at a dose of
5 μl/g. All animals in this subset were imaged with a 9-mm slice.
Phantoms were used as follows: 10 mM deuterated glucose, 0.125% D[2]O,
and plain water used as the left phantom for D[2]O imaging.
MRI scanner and hardware
All animal imaging sessions were performed on an 11.1 T Magnex magnet
integrated with a Bruker Avance III HD console controlled by ParaVision
6.0.1 (Bruker Biospin, Billerica, MA). The MRI scanner was outfitted
with a RRI BFG-240/120-S6 gradient coil (120 mm in bore size), which is
capable of generating 100 mT/m with a 200-μs slew rate. A B0 map was
acquired for localized shimming on the animal abdomen by using an 85-mm
home-built, actively decoupled, linear volume ^1H transmit-receive coil
operating at 470 MHz. For deuterium spectra and image acquisition, a
home-built 26-mm diameter, ^2H half-saddle coil tuned to 72.26 MHz was
used. The ^2H half-saddle coil was placed within a custom
three-dimensional (3D) printed animal cradle suitable for abdominal
imaging. Animals were placed into the cradle such that the abdomen and
tumor region were within the midline and isocenter of the coil. Two
phantoms were 3D printed and independently filled with 20 mM
[^2H[7]]glucose and 0.125% (v/v) D[2]O. Both phantoms were
superficially attached to the exterior of the half-saddle coil and used
to confirm positioning and quality control for image and spectroscopy
quantification.
Deuterium metabolic imaging
Axial ^1H images of the abdominal cross section covering the tumor
region of mice were acquired using a FLASH gradient echo sequence
and/or a RARE spin-echo sequence. A 2D acquisition matrix size of
128 × 128 and FOV of 35 mm by 35 mm were used to acquire 10 to 14
slices (2-mm slice thickness) of mice abdomen. Before the
[^2H[7]]glucose injection, a pre-injection ^2H image capturing total
^2H signal of the mouse abdomen region covering the tumor was acquired
using a ^2H FLASH sequence, with the ^2H water (HDO) signal set on
resonance. At approximately 10 min post-injection, another ^2H FLASH
image was acquired. The following parameters were used for the
acquisition of axial ^2H images of the mouse abdomen: matrix
size = 32 × 32, FOV = 35 mm by 35 mm, slice thickness = 4 to 10 mm (set
specifically to only cover the tumor region), TR = 100 ms, 256 signal
averages, 1.416-ms time-to-echo (TE), and 30° flip angle. The
acquisition of each ^2H FLASH image took ~13 min.
Deuterium MRI
Before experiments, the performance of the ^2H coil was verified by
calibrating and optimizing the radiofrequency pulse and flip angle to
ensure maximum ^2H signal acquisition. The NA of deuterium (0.015%)
accounts for ~17 mM naturally derived HDO that appears as a single peak
in ^2H spectra. To account for system and biological variations within
each animal imaging session, the HDO signal detected in pre-injection
single-pulse acquire spectra was used to normalize all spectra
post-[^2H[7]]glucose injection. The acquisition was achieved with a
spectral width of 4 kHz, 1024 data points, a 300-ms time-to-repeat
(TR), a 60° flip angle, and 512 scans. Each single-pulse acquire
experiment lasted a total 1.28 min. Spectra were acquired throughout
the course of the imaging session. To gain tissue selectivity, we also
used slice-localized spectroscopy to acquire data from sagittal slices
(perpendicular to axial images) placed on tumor and separately on
contralateral healthy tissue. Spectral acquisition was accomplished
with a 3-kHz spectral width, 512 data points, a TR of 100 ms, a flip
angle of 60°, and 2048 averages. Slice thickness varied based on tumor
size but remained within 3.5 to 5.5 mm across all animals. The total
acquisition time for each experiment was 3.34 min. All ^2H spectra were
processed using MestReNova v14.0.1-23284 (Mestrelab Research SL). Line
broadening was set to 10 Hz, spectra were zero filled to twice the
number of data points, and spectra were baseline corrected with the
Whittaker smoother function. All spectra were line-fitted and
deconvolved to extract peak areas, which were used to quantitate the
normalized intensities of HDO, [^2H[7]]glucose, and ^2H-lactate.
Single-pulse spectra were quantified by normalizing each peak area
(HDO, [^2H[7]]glucose, and ^2H-lactate) by the pre-injection HDO peak
area. This value was then divided by the dose of [^2H[7]]glucose
(milligram) administered ([183]Eq. 1). The areas of slice-localized
spectra obtained after administration of [^2H[7]]glucose were corrected
for naturally present HDO by subtracting pre-injection values from
post-injection values. These values were then divided by the amount
(milligram) of [^2H[7]]glucose administered ([184]Eq. 2).
[^2H[7]]glucose and ^2H-lactate peak areas were directly normalized by
the amount (milligram) of [^2H[7]]glucose dosed ([185]Eq. 3). This
first quantification of slice-localized spectroscopy allowed for a
comparison of tumor and healthy tissues. To account for tumor volume
and a more direct comparison between control and treated tumors,
slice-localized spectra were quantified by a second process. In this
quantification, the pre-injection HDO peak area was normalized to tumor
volume (cubic millimeter) to account for the amount of tumor tissue
analyzed per animal since it contributes to the amount of pre-injection
HDO detected. The mean intensity of the tumor was divided by the
normalized pre-injection HDO peak area. This value was then dose
normalized to account for the amount (milligram) of [^2H[7]]glucose
administered per animal ([186]Eq. 4)
[MATH: Normalized Peak
Area (A.U.)=Peak area of metabolite of
interest (A.U.)/ Pre-injection<
/mtext> HDO peak
area (A.U.)Dose
of [H72]glucose (mg) :MATH]
(1)
[MATH: Normalized
Peak Area (A.U.)=HDO post-injection peak area − HDO pre-injection
peak areaDose of [H72]glucose (mg) :MATH]
(2)
[MATH: Normalized
Peak Area (A.U.)=[H72]glucose
or 2H-lactate post-injection peak
area (A.U.)Dose
of [H72]glucose (mg) :MATH]
(3)
[MATH: Normalized Peak
Area (A.U.)=[Peak area of metabolite of
interest (A.U.)]/[HDO p
re-injection peak area (A.U.)Tumor
volume (mm3)]Dose
of [H72]glucose (mg) :MATH]
(4)
Image processing
All image processing was performed with ImageJ (National Institutes of
Health, Bethesda, MD, USA). To coregister and overlay ^1H and ^2H
images, all ^2H images (acquired with a 32 × 32 matrix size) were
interpolated with bicubic processing to a 128 × 128 image size to match
the size of ^1H images. Signal quantitation involved gathering the mean
signal intensity within the tumor tissue region and the peak area of
the pre-injection HDO peak of slice-localized spectroscopy covering
tumor tissue. The pre-injection HDO peak area was normalized to tumor
volume (mm^3) to account for the baseline amount of HDO present in
tumor tissue. The mean intensities of tumor ROIs in ^2H FLASH images
were divided by the normalized pre-injection HDO peak area. This value
was then dose normalized to account for the amount (mg) of
[^2H[7]]glucose administered per animal. Slice-localized spectroscopy
was setup to sample the maximal tumor volume while minimizing
contributions from the healthy tissue. The HDO peak area of
slice-localized spectroscopy was chosen instead of the mean intensity
of the pre-injection tumor ROI because the HDO peak in ^2H spectra had
over 10× more SNR than the tumoral mean intensities across animals. For
the [^2H[7]]glucose and [6,6′-^2H[2]]glucose comparative study, image
SNR calculations were performed by dividing the mean intensity of each
tumor ROI by the SD of the same noise ROI placed on the top left-hand
corner of all images. The data processed for SNR calculations was
magnitude data; therefore, a correction factor (square root of 2-π/2)
was applied to the SD of noise to account for the Rician distribution
of noise ([187]52).
Histology
Upon harvesting, approximately half of each tumor was immediately
placed into a histology cassette and subsequently in a solution of 4%
(w/v) paraformaldehyde in PBS. The tumors were incubated in this
solution for ~18 hours. At the end of this time, the tumor samples were
transferred to 70% ethanol for another 24 hours and then transferred to
fresh 70% ethanol for a final 24 hours. The samples were maintained in
70% ethanol at 4°C until they were submitted to the UF Molecular
Pathology Core for further processing. The fixed tumor tissues were
paraffin embedded, sliced, stained with H&E, and imaged.
Statistical analysis
The specific statistical analysis performed for each dataset is
reported under each figure. Grouped datasets were analyzed by one-way
or two-way analysis of variance (ANOVA) with Tukey post hoc multiple
correction. Column datasets were analyzed by Student’s t test.
Correlations were analyzed by correlation matrix analysis. A P value of
0.05 or lower was considered significant. All statistics were performed
and analyzed with GraphPad Prism 9 (version 9.5.0).
Acknowledgments