FULL PAPER
Separation and Quantification of Lactate and Lipid at1.3 ppm by Diffusion-Weighted Magnetic ResonanceSpectroscopy
Anna M. Wang,1,2 Gilberto K. K. Leung,3 Karrie M. Y. Kiang,3 Danny Chan,4
Peng Cao,1,2 and Ed X. Wu1,2,5*
Purpose: To separate the spectrally overlapped lactate and
lipid signals at 1.3 ppm using diffusion-weighted magnetic res-onance spectroscopy (DW-MRS) based on their large diffusiv-
ity difference.Methods: DW-MRS was applied to the gel phantoms contain-ing lactate and lipid droplets, and to the rat brain tumors. Lac-
tate and lipid signals and their apparent diffusion coefficientswere computed from the diffusion-weighted proton spectra.
Biexponential fitting and direct spectral subtraction appro-aches were employed and compared.Results: DW-MRS could effectively separate lactate and lipid
signals both in phantoms and rat brain C6 glioma by biexpo-nential fitting. In phantoms, lactate and lipid signals highly cor-related with the known lactate concentration and lipid volume
fractions. In C6 glioma, both lactate and lipid signals weredetected, and the lipid signal was an order of magnitude
higher than lactate signal. The spectral subtraction approachusing three diffusion weightings also allowed the separation oflactate and lipid signals, yielding results comparable to those
by the biexponential fitting approach.Conclusion: DW-MRS presents a new approach to separate
and quantify spectrally overlapped molecules and/or macro-molecules, such as lactate and lipid, by using the diffusivitydifference associated with their different sizes or mobility
within tissue microstructure. Magn Reson Med 000:000–000,2016. VC 2016 Wiley Periodicals, Inc.
Key words: magnetic resonance spectroscopy; diffusion
weighting; spectral overlapping; molecular size; lactate; lipid;C6 glioma
INTRODUCTION
Spectral overlapping is a common problem that affects
clinical and preclinical magnetic resonance spectroscopy
(MRS) and hampers identification of important bio-
markers for diagnosis and prognosis. Diffusion-weighted
MRS (DW-MRS) is a noninvasive tool to probe the diffu-
sion of metabolites (1–3). At short diffusion time regime,
diffusion properties can be largely dependent on the
molecular size. Therefore, it may be feasible to separate
spectrally overlapped compartments with greatly differ-
ent molecular sizes, such as lactate and lipid, by apply-
ing diffusion weighting.Lactate is usually undetectable by MR in healthy tis-
sues. In cancer studies, the increased glucose uptake, lack
of oxygen supply, and accumulation of lactate are the
main features of the tumor cells. Lactate content in the
tumor tissue can predict metastases and overall survival
of cancer patients (4–7). Lactate also reveals the recur-
rence of tumor cells after cancer treatment and the resist-
ance to both radiotherapy and chemotherapy (6,8,9). The
abnormal lactate elevation has an important role in both
diagnosis and therapeutic management of cancer.Mobile lipids are the MR visible neutral lipids accu-
mulated in the form of small droplets under long-term
stress. A number of in vivo and in vitro studies have
demonstrated that accumulated mobile lipids are closely
related to the cell death in response to cancer therapy
(10). The concentration of mobile lipids indicates the
extent of necrosis and apoptosis (11,12). Therapeutic
outcomes such as radiation necrosis can be evaluated by
the mobile lipid concentration, while the lactate concen-
tration can serve as an indicator of tumor growth and
recurrence. The accurate measurement of lactate and
lipid concentrations is considered to be valuable for
prognosis, discovering tumor recurrence, and evaluating
the efficacy of therapeutic interventions (5).Lactate is also related to hypoxia and hypoperfusion.
In stroke and muscle ischemia, lactate content is an indi-
cator of the time that tissue has been subjected to
hypoxia and is closely related with pH value (13–15). As
lactate has been reported to accumulate quickly in the
acute phase in stroke, lactate concentration can be an
important marker for grading or staging the disease or
injury (13,14,16). Recent studies have also reported that
mobile lipids are related to apoptosis signaling in stroke
patients (17,18). The concentration of mobile lipids indi-
cates the quantity of necrotic cell death in the injured
tissue (14,17). In summary, monitoring lactate and lipid
1Laboratory of Biomedical Imaging and Signal Processing, The Universityof Hong Kong, Pokfulam, Hong Kong SAR, China.2Department of Electrical and Electronic Engineering, The University ofHong Kong, Pokfulam, Hong Kong SAR, China.3Department of Surgery, The University of Hong Kong, Pokfulam, HongKong SAR, China.4School of Biomedical Sciences, The University of Hong Kong, Pokfulam,Hong Kong SAR, China.5State Key Laboratory of Pharmaceutical Biotechnology, The University ofHong Kong, Pokfulam, Hong Kong SAR, China.
Grant sponsor: Hong Kong Research Grant Council; Grant number: GRFHKU17124314.
*Correspondence to: Ed X. Wu, PhD, Department of Electrical and ElectronicEngineering, The University of Hong Kong, Hong Kong SAR, China. Tele-phone: Tel: 852-2859-7096; fax: 852-2559-8738; E-mail: [email protected].
Received 15 October 2015; revised 16 December 2015; accepted 7January 2016
DOI 10.1002/mrm.26144Published online 00 Month 2016 in Wiley Online Library (wileyonlinelibrary.com).
Magnetic Resonance in Medicine 00:00–00 (2016)
VC 2016 Wiley Periodicals, Inc. 1
is essential for diagnosis and prognosis of a number ofdiseases.
Spectroscopy can provide direct observation and quan-tification of important biomarkers. However, in 1H-MRS,mobile lipids emit a strong signal from the methylene(�CH2�) protons at 1.2–1.4 ppm, and it overlaps withthe lactate methyl (�CH3) protons at 1.3 ppm. These twosignals can be separated using multiple echo times (TEs),because the lactate doublets at 1.3 ppm reverse in polar-ity with every TE increase of 144 ms (19). This methodis simple and straightforward, but the resulting signal-to-noise ratio (SNR) is very low, and it is often compro-mised by the chemical shift displacement artifact(20,21). In the past two decades, lactate editing byJ-difference has been developed and applied for the sep-aration of lactate and lipid in tumor tissue (22,23). Spec-tral editing methods such as Meshcher-Garwood point-resolved spectroscopy and band selective inversion withgradient dephasing have demonstrated high sensitivityfor detecting lactate. However, these methods are timeconsuming, require long TE, and usually take more than30 min to acquire the difference spectra with reasonablelactate SNR (22–25). Zero-quantum coherence spectralediting combined with the stimulated echo (STE)sequence for spatial selection can also separate lactatefrom lipid (26). Furthermore, the implementation ofselective multiple quantum coherence filtering with thechemical shift imaging sequence allows the sensitvedetection of lactate without lipid contamination (27).
In the NMR community, there are various methods forresolving the overlapped resonances in certain dimen-sions based on different mechanisms (28,29). Separatingthe spectrally overlapped molecules or/and macromole-cules by diffusion has been previously proposed anddemonstrated in ex vivo samples as two-dimensional dif-fusion-ordered spectroscopy (2D-DOSY) (30,31). By add-ing diffusion weighting to the overlapped resonances,DOSY is ideal for resolving different molecules in thedimension of diffusivity and characterizing their sizes(32–34).
Similar approaches can be applied in vivo to separatemolecules and/or macromolecules with different diffu-sivities. Diffusion MR is a powerful tool for probingmolecular diffusivity and tissue microstructure noninva-sively. Most DW-MRS studies so far have focused on theresonances arising from relatively small metabolites(35,36). Recently, DW-MRS has been employed to inves-tigate the diffusion behavior of macromolecules such asintramyocellular lipids (IMCLs) in skeletal muscles(37–39) and proteoglycans (PGs) in intervertebral discs(40). The diffusivities of macromolecules such as lipid,PG, and collagen have been shown to be much lowerthan those of the solutes in cytosol, such as NAA, chol-ine, creatine, glucose, and lactate.
Lactate is a small metabolite (molecular weight ofC3H6O3: 90 Da). It is expected to exhibit a relatively highdiffusivity compared with lipid (molecular weight of anunsaturated triglyceride C55H98O6: 854 Da), especiallygiven that the diffusion of mobile lipid in vivo is oftenrestricted within the submicron-sized lipid droplets. Alarge difference in diffusivity between lactate and lipidmay allow their robust separation and quantification
from diffusion-weighted spectra. In this study, DW-MRSwas applied as a tool for separating the overlapped lac-tate and lipid signals at 1.3 ppm. The feasibility ofresolving their spectral overlapping by diffusion weight-ing was demonstrated in phantoms and rat brain tumors.
METHODS
MR Instrument and DW-MRS Sequence
All MR experiments were performed on a 7T MRI scan-ner (70/16 Pharmascan, Bruker BioSpin, Rheinstetten,Germany) equipped with a 370 mT/m gradient systemalong each axis. A receive-only quadrature surface coiland a circular polarized transmit-only coil were used.The DW-MRS sequence was implemented by modifyingthe stimulated echo acquisition mode (STEAM) basedsingle-voxel MRS sequence with a pair of diffusion gra-dients applied along the shortest axis of the MRS voxel.Oil phantom experiments were first performed to evalu-ate and optimize the DW-MRS protocol by minimizingeddy current and maintaining adequate SNR whileachieving adequate diffusion decay (38,40).
Phantom Experiments
All phantom experiments were conducted at room tem-perature (20
�C). One double-layer-cylinder phantom
with 400 mM lactate solution in the outer layer and oliveoil in the inner tube was made for demonstrating diffu-sion suppression of the lactate signal. The phantom wasplaced in parallel with the main magnetic field so thatthe lipid resonances in the inner tube would be shiftedto the higher frequency as a result of the bulk suscepti-bility effect (41,42).
Six phantoms were made to demonstrate the lactateand lipid separation and quantification. The lipid drop-let emulsion (�0.5-mm diameter, Intralipid 20%, Frese-nius Kabi, Lake Zurich, Illinois) was mixed with lactate(L7022, Sigma-Aldrich, St. Louis, Missouri) in agarosegel (A9414, Sigma-Aldrich, St. Louis, Missouri). Theywere made with constant lactate concentration (50 mM)and varied lipid concentrations (lipid volume fractionfrom 0.1 to 0.6% in steps of 0.1%). These gel phantomsprovided better imitation than the olive oil and lactatesolution phantom of the live tissue where the mobile lip-ids are accumulated with lactate. DW-MRS experimentswere performed on the gel phantoms with 11 diffusionweighting factors or b-values (0, 1.5, 2.5, 4, 6.5, 9, 30, 40,and 21� 103 s/mm2). Other parameters were TE¼ 32 ms,repetition time (TR)¼ 1300 ms, mixing time (TM)¼ 64ms, diffusion time (D)¼ 80 ms, diffusion gradient dura-tion (d)¼ 9 ms, and number of excitations (NEX)¼ 32.
Rat Brain C6 Glioma Experiments
The brains of six Sprague-Dawley (SD) rats (body massapproximately 260 g) were injected with C6 glioma cells(43–45). In brief, the rats were anesthetized with 80 mg/kg ketamine and 10 mg/kg xylazine and placed in a ste-reotactic device. Their heads were shaved and an inci-sion made in the anteroposterior direction to expose thebregma of the skull. A 1-mm hole was drilled throughthe skull at 1 mm anterior and 2.5 mm to the right of the
2 Wang et al
Bregma. 10mL of cell culture medium (Dulbecco’s modi-
fied Eagle’s medium) containing approximately 106 C6
glioma cells were then injected into the cortex at 5 mm
depth at 2 mL/min with a 26-G needle Hamilton syringe
(Reno, Nevada). Afterward, the hole was filled with bone
wax and the scalp was sutured.DW-MRS experiments with nine b-values (0, 2, 4, 6,
10, 20, 30, 40, and 50� 103 s/mm2) were conducted on
the 18th day after C6 glioma cell injection. The MRS
voxel was placed at the tumor core and covered most of
the tumor tissue based on the T2-weighted images. MRS
voxel sizes ranged from 151 to 215mL, depending on the
actual tumor sizes. The scan parameters were TE/TM/
TR¼ 52/34/1300 ms, D/d¼ 60/18 ms, and NEX¼128.Oil red O staining was performed on the tumor tissue
samples to confirm lipid accumulation in the tumor
and to identify the lipid droplets (46–48). Rats exam-
ined by DW-MRS were sacrificed with transcardial per-
fusion using 4% paraformaldehyde in 0.1 M PBS
(pH¼7.4) after an overdose of intraperitoneal pentobar-
bital. Brain samples postfixed with 4% paraformalde-
hyde were further processed by 30% sucrose overnight
at 4 �C for cryoprotection. They were embedded in opti-
mum cutting temperature compound (OCT; Thermo Sci-
entific, Waltham, Massachusetts) and 15-m cryosections
were prepared and stored frozen until staining. Frozen
sections from OCT-embedded brain tissues were air-
dried at room temperature, washed in 60% isopropanol
and stained with 0.5% Oil red O solution (O1391,
Sigma-Aldrich, St. Louis, Missouri) at room temperature
for 1 h. Sections were washed through 60% isopropanol
and mounted. Lipid droplets were visualized under the
microscope and images were acquired.
Data Analysis
For all DW-MRS data, each individual free-induction
decay (FID) was stored and phase-corrected before aver-
aging to minimize the effect of zero-order phase varia-
tions caused by motion and gradient eddy current
(38,40). Spectral analysis was performed using JMRUI
software (http://www.mrui.uab.es/mrui/). Overlapped
signals from lipid methylene (� CH2-) and lactate
methyl (� CH3) protons at 1.3 ppm were identified and
integrated using AMARES algorithm. Fitting errors of
the overlapping background signals were assessed by
Cram�er-Rao lower bounds (CRLBs) (49–51). Quantifica-
tion was considered relevant only when the correspond-
ing CRLB was below 25%. The DW signals of lactate
and lipid were normalized by the spectral baseline
noise. Their relative concentrations in arbitrary unit
were calculated by normalizing the lactate and lipid sig-
nals with the MRS voxel volume.Two calculation approaches were applied to quantify
the lactate and lipid signals. For the biexponential fitting
approach, the lactate and lipid signals and their apparent
diffusion coefficients (ADCs) were computed by fitting
the b-value-dependent DW signals to a biexponential
model using least-squares regression according to the fol-
lowing equation:
S ¼ Slactate � e�b � ADClactate þ Slipid � e�b � ADClipid [1]
The spectral subtraction approach was also applied toestimate lactate and lipid signals from two or three DWspectra, which can offer the advantage of shorter acquisi-tion time in practice. This way of estimation is justifiedbecause lipid diffusion is very slow and the lipid signal
loss resulting from diffusion can be negligible at certainlow b-value (6500 and 20000 s/mm2 for gel phantomsand C6 glioma, respectively) where the lactate signal isalready mostly suppressed by diffusion. Two DW spectra
acquired at such b-value and b¼ 0 allow the direct esti-mation of lipid spectrum (ie, the DW spectrum acquiredat such nonzero b-value), and indirect estimation of lac-tate spectrum by subtracting two spectra in the complexspectral domain.
Three DW spectra can further improve the lactate and
lipid signal estimations and measure the lipid diffusivityusing b¼ 0, and two nonzero b-values at which lipid dif-fusivity can be measured while the lactate signal ismostly suppressed. Such approach can yield more accu-rate lipid spectrum estimation by correcting for the lipid
diffusion decay, and better lactate spectrum estimation.In this study, the two nonzero b-values were 6500 and21000 s/mm2 for gel phantoms, and 20000 and 50000 s/mm2 for C6 glioma.
The accuracy of the lactate and lipid quantification in
the gel phantoms was evaluated by the quality of the lin-ear correlation between the relative lipid concentrationsmeasured by DW-MRS and the known lipid volume frac-tions in the phantoms. Subtraction approaches using twoand three b-values were also compared with the biexpo-
nential fitting approach for their quantification quality.
RESULTS
Figure 1 illustrates the robust separation of lactate andlipid by diffusion weighting. An axial T2-weighted image
of the double layer phantom is shown in the upper-rightcorner. As shown in the DW spectra, the 1.3-ppm signalfrom lipid methylene (� CH2-) protons of the olive oil inthe inner tube was shifted to 1.5 ppm and separated fromthe 1.3-ppm signal from lactate methyl (� CH3) protons
of the lactate solution contained in the outer layer by thebulk susceptibility effect. The effective suppression oflactate signal was achieved at a b-value of 6000 s/mm2,whereas the lipid signal remained relatively unchanged.
Figure 2 demonstrates the separation and quantifica-tion of lactate and lipid signals in six gel phantoms. The
lactate concentration was fixed at 50 mM, whereas thelipid volume fraction was varied. The location of theMRS voxel in the gel phantom is shown in the T2-weighted image in Figure 2a. The representative DW
spectra are also shown in Figure 2a. In these spectra, lac-tate doublets are clearly seen at the low b-value range(0–1500 s/mm2), and they overlapped with the lipid res-onance at 1.3 ppm. At high b-values, lactate doubletsdecayed to an undetectable level, but lipid resonance at
1.3 ppm only showed a very slow decay. The corre-sponding biexponential fitting of the integrated DW sig-nals at 1.3 ppm is shown in the upper-right corner ofFigure 2a. Figure 2b shows the relative lactate and lipid
Separation of Lactate and Lipid by Diffusion-Weighted MRS 3
concentrations in gel phantoms calculated from the biex-ponential fitting. Although the measured lactate concen-
trations maintained constant as expected, the measuredlipid concentrations were highly correlated with theirknown lipid volume fractions. Figure 2c shows that lac-
tate ADCs were relatively constant at (8.07 6 0.05)�10�4
mm2/s, and that lipid ADCs exhibited a relatively largefluctuation ((5.05 6 0.38)�10�7 mm2/s). This fluctuation
in lipid ADC likely resulted from the limited b-valuerange used in the experiments (0–21000 s/mm2) that did
not provide strong lipid signal decay for accurate lipiddiffusivity estimation. The variation of the lipid dropletsize in the gel phantoms might also contribute to the
lipid ADC fluctuation.Figure 3 shows the DW-MRS quantification of lactate
and lipid in C6 glioma in six rat brains. In the upper-leftcorner of Figure 3a, the typical localization of the MRS
voxel is shown as the rectangle superimposed on thetumor in a T2-weighted image. At lower b-value range(0–10000 s/mm2), the overlapped resonances at 1.3 ppm
showed a relatively faster decay; in contrast, the lipidresonance at 0.9 ppm exhibited little decay. The corre-
sponding biexponential fitting of the DW signals at1.3 ppm is presented in the upper-right corner of Figure3a. Figures 3b and 3c summarize the quantification of
lactate and lipid concentrations and ADCs. The lipidmethylene (� CH2-) proton signals were generally muchstronger than the lactate methyl (� CH3) signals, indicat-
ing a great extent of mobile lipid accumulation in C6 gli-oma. The presence of mobile lipids in C6 glioma was
verified by the Oil red O staining as shown in Figure 4.The mobile lipid accumulation was observed in thetumor tissue but not in the normal tissue. The typicaldiameter of lipid droplets was approximately 0.5 mm.
Figure 5 demonstrates the use of spectral subtractionapproach to calculate lactate and lipid contents in thegel phantoms. As shown in Figure 5a, clean separationof lactate and lipid spectra was achieved by directly sub-tracting the two spectra acquired at b¼0 and 6500 s/mm2, without and with the spectrum at b¼6500 s/mm2
corrected for diffusion loss. The lactate doublets can beclearly observed after subtraction. Two lipid resonancesfrom methylene (� CH2-) and methyl (� CH3) protonswere well removed from the lactate doublets The lactateand lipid signals calculated using such spectral subtrac-tion approaches and 11 b-value biexponential fittingapproach are compared in Figures 5b and 5c. A spectralsubtraction approach using either two b-values (0 and6500 s/mm2) or three b-values (0, 6500, and 21000 s/mm2) produced results that were highly correlated withthose by the biexponential fitting. Similarly, Figure 6ashows that lactate and lipid in the C6 glioma can also beseparated by subtracting the spectra acquired at b¼ 0and 20000 s/mm2. Figures 6b and 6c show that the accu-racy of spectral subtraction approach using three b-values (0, 20000, and 50000 s/mm2) was better than thatof spectral subtraction using two b-values (0 and 20000s/mm2).
DISCUSSION
Separation and Quantification of Lactate and Lipid byDW-MRS
Lipids are a group of naturally abundant macromoleculesin various tissue types in the human body (37,52). Themolecular weight of lipid is much larger than that ofwater-soluble small molecules, including lactate. There-fore, the diffusivity, characterized by the ADC, differsgreatly between lipid and lactate (53). As shown in Fig-ure 2c, the lactate ADC in gel phantoms was(8.07 6 0.05)�10�4 mm2/s. It was three orders of magni-tude higher than the lipid ADC ((5.05 6 0.38)�10�7
mm2/s). Figure 3c shows that lactate ADC in C6 gliomawas (2.83 6 0.83)�10�4 mm2/s, which was two orders ofmagnitude higher than the lipid ADC ((2.68 6 1.29)�10�6
mm2/s). These large diffusivity differences benefitted theeffective separation and quantification of lactate andlipid by diffusing weighting.
As shown in Figures 5 and 6, the complex spectralsubtraction approach allows the relatively simple estima-tion of the lactate spectrum by the difference spectrumof two or three DW spectra, thus incurring less acquisi-tion time. For accurate lactate and lipid quantification,lipid diffusion decay needs to be considered. For exam-ple, based on the lactate and lipid ADC measurements inC6 glioma in the present study, approximately 94% ofthe lactate signal can be suppressed by diffusion weight-ing at b¼ 10000 s/mm2 when lipid signal loss is approxi-mately 3%, and 99% of the lactate signal can besuppressed at b¼ 20000 s/mm2, whereas the lipid signalloss is approximately 5%. These estimations suggest thatlipid signal loss resulting from diffusion should not be
FIG. 1. Demonstration of the effective lactate signal suppressionby diffusion weighting. The DW spectra were acquired from the
MRS voxel (the box in the T2-weighted image in the upper-rightcorner). The lactate signal was suppressed to an undetectablelevel at 6000 s/mm2 (purple spectra), while the lipid signals
showed little decay. Note that, by positioning the cylinder phan-tom in parallel to the main magnetic field, the 1.3-ppm lipid signalshifted to 1.5 ppm by the bulk susceptibility effect to facilitate
visualization.
4 Wang et al
neglected if lipid concentration is more than an order of
magnitude higher than lactate concentration. In this sce-
nario, the subtraction approach using two b-values
would be problematic, and the three-b-value subtraction
approach would become more desirable. As shown in
Figures 6b and 6c, use of three b-values to include the
lipid diffusivity estimation was necessary for accurate
lactate and lipid quantification in C6 glioma. It should
be further noted that the lactate phase evolution result-
ing from scalar coupling is neglected in Eq. 1 because of
the relatively small TEs used in the present study.
Future studies can account for such phase evolution in
Eq. 1 for better quantification accuracy if the biexponen-
tial fitting approach is used.The presence of lactate and lipid is not limited to
tumor tissues. In fact, we also performed a pilot study to
test the diffusion separation of lactate and lipid in a rat
model of intracerebral hemorrhage (ICH) (Supporting
Figure S1). In brief, collagenase solution was infused
into the right striatum of six SD rats to induce the blood
vessel dissolution and focal bleeding (54,55). The blood
vessel disruption resulted in regional shortage of oxygen
supply and gave rise to the lactate elevation. The cell
death caused by hypoxia in the hematoma led to the
increased mobile lipids. Biexponential decay of the DW
signals at 1.3 ppm was observed as a result of the lactate
and mobile lipids in the hematoma 20 h after collagenase
injection. Compared with the rat brain C6 glioma, lipid
concentration in hematoma was much lower, whereas
the lactate concentration was similar.
Biological Insights from the ADCs of Lipid and Lactate
In this study, the lipid ADC in rat brain C6 glioma meas-
ured at a diffusion time of 60 ms (2.68�10�6 mm2/s) is
comparable with those reported in previous glioma stud-
ies (56,57). The lipid ADC varies between different tissue
types and physiological conditions. For example, the
intramyocellular lipid ADC measured in the rat hindlimb
skeletal muscle is 1.01� 10�6 mm2/s at a diffusion time
of 80 ms and 0.25� 10�6 mm2/s at 220 ms, revealing the
strongly restricted diffusion behavior of lipid macromole-
cules within the micron-sized intramyocellular lipid
droplets that are largely immobilized inside the muscle
cells (37,40). The IMCL ADC is much lower than the
mobile lipid ADCs measured in this study, both in C6
glioma and ICH hematoma. Note that the free and unre-
stricted lipid ADC was measured as 6.57� 10�6 mm2/s
in pure olive oil (40). This value was much higher than
the lipid ADC measured in the gel phantoms in the pres-
ent study (5.05�10�7 mm2/s) and lower than the lipid
ADC measured from the lipid droplet emulsion diluted
FIG. 2. DW-MRS measurements in the six agarose gel phantoms containing a constant lactate concentration but varied lipid dropletconcentrations at 20
�C. DW spectra acquired from one phantom (lipid volume fraction 0.2%) are presented (a). The MRS voxel localiza-
tion is shown as the box in the T2-weighted image. The lactate doublets can clearly be seen at b-values lower than 4500 s/mm2. The
corresponding biexponential fitting of the 1.3-ppm DW signals is shown in the upper-right figure. Relative lactate and lipid concentra-tions calculated by the biexponential fitting (b). Lactate and lipid ADCs (c).
Separation of Lactate and Lipid by Diffusion-Weighted MRS 5
in saline (1.86�10�5 mm2/s, data not shown). All
together, these large lipid ADC variations among differ-
ent forms of lipids suggest that mobile lipid ADCs were
determined by both size and mobility of the lipid drop-
lets. Tissue type, cell size and shape, cytosol viscosity,
and intracellular skeleton may all affect the mobility of
lipid droplets in vivo, which can lead to the broad
mobile lipid ADC variations in different tissues.In this study, the lipid and lactate ADCs measured in
C6 glioma showed a similar interanimal variation (Fig.
3c). This might arise from the inhomogeneous tumor
microstructure among the six animals. The ADC differen-
ces could be also associated with tissue microstructure,
such as heterogeneity in cell size, extent of edema, and
blood coagulation in the lesion (1,2,58). The diffusion
environment is complex in both tumor and ischemic tis-
sues. Characterization of the lactate and mobile lipid dif-
fusion behaviors may provide extra information about
the tissue microstructural changes in diseased tissues.
Separation of Different Molecules and Macromolecules byDiffusion Weighting
In DOSY and DW-MRS experiments, diffusivity is often
dominated by molecular size. Our recent study has
shown that water can be suppressed by diffusion weight-
FIG. 3. In vivo DW-MRS results from the six rat brains with C6 glioma. The experiments were performed on the 18th day after the injec-tion of glioma cells. DW spectra acquired from one rat are presented (a). The MRS voxel is indicated as the rectangle box in the T2-
weighted image. In the DW spectra, lipid and lactate signals at 1.3 ppm were overlapped. The corresponding biexponential fitting of the1.3-ppm DW signals is shown in the upper-right corner. Relative lactate and lipid concentrations calculated by the biexponential fitting(b). Lactate and lipid ADCs (c).
FIG. 4. Oil red O staining of a C6 glioma sample at the border ofnormal tissue and C6 glioma. Red dots show the lipid presence in
the form of small droplets within C6 glioma. The enlarged viewshows that the typical diameter of these lipid droplets was
approximately 5 mm.
6 Wang et al
ing and be separated from PGs at 4.2–4.5 ppm in interver-tebral discs. PGs are the important macromolecules in theextracellular matrix. This water and PG separation by dif-fusion weighting was based partly on the large differencein their molecular sizes (40). In the present study, DW-MRS was proved to be robust when applied to separatethe overlapped resonances from the small lactate metabo-lite and large lipid macromolecules. Note that moleculardiffusion is also sensitive to microstructural characteris-tics. Diffusion weighting has been applied to separatespectrally overlapped components that differed greatly intheir microstructural characteristics. For example, IMCLsand extramyocellular lipids (EMCLs) at 1.2–1.5 ppm inskeletal muscles can be separated by diffusion weighting,because the difference in their microstructural distribu-tion has dominated their diffusion properties (37,38,40).
In fact, diffusion weighing of water molecules has been
often applied to alter the contrast among vascular water,
extracellular water, and intracellular water and separate
these different components (59–61).
Technical Issues in DW-MRS
DW-MRS experiments are technically challenging. Non-
water molecules are of very low concentrations in vivo.
They exhibit large molecular weights and slow diffusion,
and strong diffusion weighting is required to probe and
utilize their diffusion properties. Consequently, DW-
MRS experiments suffer from low SNR, require strong
gradients, and are vulnerable to gradient eddy currents
and motions. At present, the relatively low gradient
strength in clinical MRI systems is one of the major
FIG. 5. Separation and quantification of the lipid and lactate signals in the gel phantoms using spectral subtraction approaches. Lactate
spectrum generated by directly subtracting the DW spectra acquired at b¼0 and 6500 s/mm2 from one gel phantom in the spectraldomain, without and with the spectrum at b¼6500 s/mm2 corrected for lipid diffusion loss (a). Comparison of the lactate and lipid sig-
nals calculated by two-b-value (b) and three-b-value (c) subtraction approaches with those by the 11-b-value biexponential fittingapproach.
Separation of Lactate and Lipid by Diffusion-Weighted MRS 7
obstacles to robust DW-MRS implementation. Neverthe-less, short TE and TM are generally desired to preserveSNR for accurate signal measurement in both preclinicaland clinical DW-MRS studies. It is imperative to opti-mize the choice of TR and TE values, diffusion time, anddiffusion gradient duration according to the T1 and T2values of the molecule of interest and the b-valuerequirement. Compared to PRESS, STEAM-based DW-MRS loses half of the transverse magnetization, resultingin 50% SNR reduction. However, STEAM-based DW-MRS has an advantage over a PRESS-based one becausethe former offers the flexibility to increase diffusion timewithout increasing TE and significantly compromisingSNR. In the present study, given the b-value requirementof 50000 s/mm2 and the short lipid T2 value, a STEAM-based DW-MRS sequence was employed.
Strong diffusion gradient can induce substantial eddycurrent and contaminate spectral quantification in DW-MRS study. In DW-STEAM, adding two gradientsbetween the second and third 90 � RF pulses with thesame amplitudes can reduce the eddy current effect (62).Furthermore, a short diffusion gradient duration isalways preferred for the cancellation of eddy currentsfrom the rising and falling edges of the gradient pulse.Using a prolonged TE will also reduce the short-termeddy current effect at the expense of SNR in practice.
DW signals at high b-values are highly sensitive tophysiologic motions. In our study, the respiration wascontrolled by the mechanical ventilation system. In brainDW-MRS, the cerebrospinal fluid (CSF) pulsation andrespiratory motion can easily lead to the ADC overesti-mation. For applying DW-MRS to study human brain
FIG. 6. Separation and quantification of the lipid and lactate signals in C6 glioma using spectral subtraction approaches. Lactate spec-trum generated by directly subtracting the DW spectra acquired at b¼0 and 20000 s/mm2 from one typical C6 glioma, without and
with the spectrum at b¼20000 s/mm2 corrected for lipid diffusion loss (a). Comparison of the lactate and lipid signals calculated by thetwo-b-value (b) and three-b-value (c) subtraction approaches with those by the nine-b-value biexponential fitting approach.
8 Wang et al
tumors, cardiac and respiratory gated acquisition withappropriate trigger delay can help alleviate this problem.In a previous human study, peripheral pulse gating(PPG) has been used to reduce the motion artifactscaused by CSF pulsation in the brain to improve diffu-sion measurements (63). In addition, we also found thatacquisition with interleaved b-values will help improvethe accuracy of ADC measurement in the presence ofphysiologic motions such as respiration.
Scan-to-scan phase realignment was adopted in thisstudy to alleviate the motion and gradient eddy current–related ADC overestimation. Our recent in vivo DW-MRSstudy of IMCL showed that in individually stored FIDs,scan-to-scan phase corrections can significantly reducethe macromolecule ADC overestimation caused by physi-ologic motion (40). For animal study, the use of animalfixation, paralyzation, and synchronized ventilationwould also be helpful in reducing the detrimentalmotion effect. In summary, careful sequence adjustmentis essential for DW-MRS experiments. For in vivo exami-nation of both animal models and human subjects, thecontrol of physiologic motion is of great necessity.
CONCLUSION
Lactate and lipid elevations are the hallmark of cancerousand ischemic tissues. Using diffusion weighting, robust sep-aration of lactate and lipid signals at 1.3 ppm was demon-strated in the lactate/lipid phantoms and C6 glioma in vivo.In practice, it can be implemented by the simple spectralsubtraction approach using two or three diffusion-weightedspectra acquired with carefully chosen b-values. Such DW-MRS presents a new approach to separate and quantify spec-trally overlapped signals, such as lactate and lipid, by usingthe diffusivity difference associated with their differentmolecular sizes or mobility within tissue microstructure.
ACKNOWLEDGMENTS
This work was in part supported by a grant from HongKong Research Grant Council (GRF HKU17124314) andHKU matching fund for the State Key Laboratory of Pharma-ceutical Biotechnology. The authors thank Dr. Gerda Grim-nisdottir for assistance during the manuscript preparation.
REFERENCES
1. Nicolay K, Braun KP, Graaf RA, Dijkhuizen RM, Kruiskamp MJ. Dif-
fusion NMR spectroscopy. NMR Biomed 2001;14(2):94–111.
2. Nicolay K, van der Toorn A, Dijkhuizen RM. In vivo diffusion spec-
troscopy: an overview. NMR Biomed 1995;8(7-8):365–374.
3. Van Zijl PC, Moonen CT, Faustino P, Pekar J, Kaplan O, Cohen JS.
Complete separation of intracellular and extracellular information in
NMR spectra of perfused cells by diffusion-weighted spectroscopy.
Proc Natl Acad Sci U S A 1991;88(8):3228–3232.
4. Kim JH, Chang KH, Na DG, Song IC, Kwon BJ, Han MH, Kim K. 3T
1H-MR spectroscopy in grading of cerebral gliomas: comparison of
short and intermediate echo time sequences. AJNR Am J Neuroradiol
2006;27(7):1412–1418.
5. Walenta S, Wetterling M, Lehrke M, Schwickert G, Sundfor K,
Rofstad EK, Mueller-Klieser W. High lactate levels predict likelihood
of metastases, tumor recurrence, and restricted patient survival in
human cervical cancers. Cancer Res 2000;60(4):916–921.
6. Hirschhaeuser F, Sattler UG, Mueller-Klieser W. Lactate: a metabolic
key player in cancer. Cancer Res 2011;71(22):6921–6925.
7. Saraswathy S, Crawford FW, Lamborn KR, Pirzkall A, Chang S, Cha
S, Nelson SJ. Evaluation of MR markers that predict survival in
patients with newly diagnosed GBM prior to adjuvant therapy.
J Neurooncol 2009;91(1):69–81.
8. Sattler UG, Meyer SS, Quennet V, et al. Glycolytic metabolism and
tumour response to fractionated irradiation. Radiother Oncol 2010;
94(1):102–109.
9. Groussard C, Morel I, Chevanne M, Monnier M, Cillard J, Delamarche
A. Free radical scavenging and antioxidant effects of lactate ion: an
in vitro study. J Appl Physiol (1985) 2000;89(1):169–175.
10. Delikatny EJ, Chawla S, Leung DJ, Poptani H. MR-visible lipids and
the tumor microenvironment. NMR Biomed 2011;24(6):592–611.
11. Cheng LL, Anthony DC, Comite AR, Black PM, Tzika AA, Gonzalez
RG. Quantification of microheterogeneity in glioblastoma multiforme
with ex vivo high-resolution magic-angle spinning (HRMAS) proton
magnetic resonance spectroscopy. Neuro Oncol 2000;2(2):87–95.
12. Rosi A, Ricci-Vitiani L, Biffoni M, et al. (1) H NMR spectroscopy of
glioblastoma stem-like cells identifies alpha-aminoadipate as a
marker of tumor aggressiveness. NMR Biomed 2015;28(3):317–326.
13. Wagner KR, Xi G, Hua Y, Kleinholz M, de Courten-Myers GM, Myers
RE. Early metabolic alterations in edematous perihematomal brain
regions following experimental intracerebral hemorrhage. J Neurosurg
1998;88(6):1058–1065.
14. Graham GD, Hwang JH, Rothman DL, Prichard JW. Spectroscopic
assessment of alterations in macromolecule and small-molecule metab-
olites in human brain after stroke. Stroke 2001;32(12):2797–2802.
15. Aronowski J, Zhao X. Molecular pathophysiology of cerebral hemor-
rhage: secondary brain injury. Stroke 2011;42(6):1781–1786.
16. Schubert GA, Poli S, Schilling L, Heiland S, Thome C. Hypothermia
reduces cytotoxic edema and metabolic alterations during the acute
phase of massive SAH: a diffusion-weighted imaging and spectros-
copy study in rats. J Neurotrauma 2008;25(7):841–852.
17. Ahn SY, Yoo HS, Lee JH, et al. Quantitative in vivo detection of brain
cell death after hypoxia ischemia using the lipid peak at 1.3 ppm of
proton magnetic resonance spectroscopy in neonatal rats. J Korean
Med Sci 2013;28(7):1071–1076.
18. Gasparovic C, Rosenberg GA, Wallace JA, Estrada EY, Roberts K,
Pastuszyn A, Ahmed W, Graham GD. Magnetic resonance lipid sig-
nals in rat brain after experimental stroke correlate with neutral lipid
accumulation. Neurosci Lett 2001;301(2):87–90.
19. Yamasaki F, Takaba J, Ohtaki M, et al. Detection and differentiation
of lactate and lipids by single-voxel proton MR spectroscopy. Neuro-
surg Rev 2005;28(4):267–277.
20. Yablonskiy DA, Neil JJ, Raichle ME, Ackerman JJ. Homonuclear J
coupling effects in volume localized NMR spectroscopy: pitfalls and
solutions. Magn Reson Med 1998;39(2):169–178.
21. Lange T, Dydak U, Roberts TPL, Rowley HA, Bieljac M, Boesiger P.
Pitfalls in lactate measurements at 3T. AJNR Am J Neuroradiol 2006;
27(4):895–901.
22. Rothman DL, Behar KL, Hetherington HP, Shulman RG. Homonuclear
1H double-resonance difference spectroscopy of the rat brain in vivo.
Proc Natl Acad Sci U S A 1984;81(20):6330–6334.
23. Smith MA, Koutcher JA, Zakian KL. J-difference lactate editing at 3.0
Tesla in the presence of strong lipids. J Magn Reson Imaging 2008;
28(6):1492–1498.
24. Star-Lack J, Spielman D, Adalsteinsson E, Kurhanewicz J, Terris DJ,
Vigneron DB. In vivo lactate editing with simultaneous detection of
choline, creatine, NAA, and lipid singlets at 1.5 T using PRESS exci-
tation with applications to the study of brain and head and neck
tumors. J Magn Reson 1998;133(2):243–254.
25. Kelley DA, Wald LL, Star-Lack JM. Lactate detection at 3T: compen-
sating J coupling effects with BASING. J Magn Reson Imaging 1999;
9(5):732–737.
26. Sotak CH. A volume-localized, two-dimensional NMR method for the
determination of lactate using zero-quantum coherence created in a
stimulated echo pulse sequence. Magn Reson Med 1988;7(3):364–370.
27. Thakur SB, Yaligar J, Koutcher JA. In vivo lactate signal enhancement
using binomial spectral-selective pulses in selective MQ coherence
(SS-SelMQC) spectroscopy. Magn Reson Med 2009;62(3):591–598.
28. Levitt MH. Spin dynamics: basics of nuclear magnetic resonance.
John Wiley & Sons: Chichester; 2001.
29. Cavanagh J, Fairbrother WJ, Palmer III AG, Skelton NJ. Protein NMR
spectroscopy: principles and practice. Elsevier Academic Press: Burling-
ton; 1995.
30. Wu D, Chen A, Johnson CS. An improved diffusion-ordered spectros-
copy experiment incorporating bipolar-gradient pulses. J Magn Reson
1995;Series A 115(2):260–264.
Separation of Lactate and Lipid by Diffusion-Weighted MRS 9
31. Morris KF, Johnson Jr CS. Diffusion-ordered two-dimensional nuclear
magnetic resonance spectroscopy. J Am Chem Soc 1992;114(8):3139–
3141.
32. Barjat H, Morris GA, Smart S, Swanson AG, Williams SC. High-reso-
lution diffusion-ordered 2D spectroscopy (HR-DOSY)—a new tool for
the analysis of complex mixtures. J Magn Reson 1995;Series B 108(2):
170–172.
33. �Smejkalov�a D, Piccolo A. Aggregation and disaggregation of humic
supramolecular assemblies by NMR diffusion ordered spectroscopy
(DOSY-NMR). Environ Sci Technol 2007;42(3):699–706.
34. Keresztes I, Williard PG. Diffusion-ordered NMR spectroscopy
(DOSY) of THF solvated n-butyllithium aggregates. J Am Chem Soc
2000;122(41):10228–10229.
35. Liu Z, Zheng D, Wang X, Zhang J, Xie S, Xiao J, Jiang X. Apparent
diffusion coefficients of metabolites in patients with MELAS using
diffusion-weighted MR spectroscopy. AJNR Am J Neuroradiol 2011;
32(5):898–902.
36. Zheng DD, Liu ZH, Fang J, Wang XY, Zhang J. The effect of age and
cerebral ischemia on diffusion-weighted proton MR spectroscopy of
the human brain. AJNR Am J Neuroradiol 2012;33(3):563–568.
37. Brandejsky V, Kreis R, Boesch C. Restricted or severely hindered dif-
fusion of intramyocellular lipids in human skeletal muscle shown by
in vivo proton MR spectroscopy. Magn Reson Med 2012;67(2):310–
316.
38. Xiao L, Wu EX. Diffusion-weighted magnetic resonance spectroscopy:
a novel approach to investigate intramyocellular lipids. Magn Reson
Med 2011;66(4):937–944.
39. Cao P, Fan SJ, Wang AM, Xie VB, Qiao Z, Brittenham GM, Wu EX.
Diffusion magnetic resonance monitors intramyocellular lipid droplet
size in vivo. Magn Reson Med 2015;73(1):59–69.
40. Wang AM, Cao P, Yee A, Chan D, Wu EX. Detection of extracellular
matrix degradation in intervertebral disc degeneration by diffusion
magnetic resonance spectroscopy. Magn Reson Med 2015;73(5):1703–
1712.
41. Szczepaniak LS, Dobbins RL, Stein DT, McGarry JD. Bulk magnetic
susceptibility effects on the assessment of intra- and extramyocellular
lipids in vivo. Magn Reson Med 2002;47(3):607–610.
42. Schick F, Eismann B, Jung WI, Bongers H, Bunse M, Lutz O. Compar-
ison of localized proton NMR signals of skeletal muscle and fat tissue
in vivo: two lipid compartments in muscle tissue. Magn Reson Med
1993;29(2):158–167.
43. Auer R, Del Maestro R, Anderson R. A simple and reproducible
experimental in vivo glioma model. Can J Neurol Sci 1981;8(4):325–
331.
44. Wassenaar W, Tator CH, So WS. The development of an intracerebral
glioma model for brain tumor chemotherapy. J Neurosurg 1973;39(3):
293–301.
45. Miura FK, Alves MJF, Rocha MC, da Silva R, Oba-Shinjo SM, Marie
SKN. Xenograft transplantation of human malignant astrocytoma cells
into immunodeficient rats: an experimental model of glioblastoma.
Clinics 2010;65(3):305–309.
46. Fowler SD, Greenspan P. Application of Nile red, a fluorescent
hydrophobic probe, for the detection of neutral lipid deposits in tis-
sue sections: comparison with Oil red O. J Histochem Cytochem
1985;33(8):833–836.
47. Ramirez-Zacarias J, Castro-Munozledo F, Kuri-Harcuch W. Quantita-
tion of adipose conversion and triglycerides by staining intracytoplas-
mic lipids with Oil red O. Histochemistry 1992;97(6):493–497.
48. Wang H, Wang H, Xiong W, Chen Y, Ma Q, Ma J, Ge Y, Han D. Eval-
uation on the phagocytosis of apoptotic spermatogenic cells by Ser-
toli cells in vitro through detecting lipid droplet formation by Oil
Red O staining. Reproduction 2006;132(3):485–492.
49. Cavassila S, Deval S, Huegen C, Van >Ormondt D, Graveron-Demilly
D. Cramer-Rao bound expressions for parametric estimation of over-
lapping peaks: influence of prior knowledge. J Magn Reson 2000;
143(2):311–320.
50. Chan KC, Khong PL, Cheung MM, Wang S, Cai KX, Wu EX. MRI of
late microstructural and metabolic alterations in radiation-induced
brain injuries. J Magn Reson Imaging 2009;29(5):1013–1020.
51. Zhou IY, Ding AY, Li Q, McAlonan GM, Wu EX. Magnetic resonance
spectroscopy reveals N-acetylaspartate reduction in hippocampus
and cingulate cortex after fear conditioning. Psychiatry Res 2012;
204(2):178–183.
52. Hamilton G, Smith DLJ, Bydder M, Nayak KS, Hu HH. MR properties
of brown and white adipose tissues. J Magn Reson Imaging 2011;
34(2):468–473.
53. Ribeiro ACF, Lobo VMM, Leaist DG, Natividade JJS, Ver�ıssimo LP,
Barros MCF, Cabral AMTDPV. Binary diffusion coefficients for aque-
ous solutions of lactic acid. J Solution Chem 2005;34(9):1009–1016.
54. Tang J, Liu J, Zhou C, Alexander JS, Nanda A, Granger DN, Zhang
JH. Mmp-9 deficiency enhances collagenase-induced intracerebral
hemorrhage and brain injury in mutant mice. J Cereb Blood Flow
Metab 2004;24(10):1133–1145.
55. Rosenberg GA, Mun-Bryce S, Wesley M, Kornfeld M. Collagenase-
induced intracerebral hemorrhage in rats. Stroke 1990;21(5):801–807.
56. Perez Y, Lahrech H, Cabanas ME, Barnadas R, Sabes M, Remy C,
Arus C. Measurement by nuclear magnetic resonance diffusion of the
dimensions of the mobile lipid compartment in C6 cells. Cancer Res
2002;62(20):5672–5677.
57. Lahrech H, Zoula S, Farion R, Remy C, Decorps M. In vivo measure-
ment of the size of lipid droplets in an intracerebral glioma in the
rat. Magn Reson Med 2001;45(3):409–414.
58. Balls GT, Frank LR. A simulation environment for diffusion weighted
MR experiments in complex media. Magn Reson Med 2009;62(3):
771–778.
59. Pfeuffer J, Tk�ac I, Gruetter R. Extracellular-intracellular distribution
of glucose and lactate in the rat brain assessed noninvasively by
diffusion-weighted 1H nuclear magnetic resonance spectroscopy in
vivo. J Cereb Blood Flow Metab 2000;20(4):736–746.
60. Pfeuffer J, Provencher SW, Gruetter R. Water diffusion in rat brain in
vivo as detected at very largeb values is multicompartmental. Mag-
netic Resonance Materials in Physics, Biology and Medicine 1999;
8(2):98–108.
61. Wu EX, Cheung MM. MR diffusion kurtosis imaging for neural tissue
characterization. NMR Biomed 2010;23(7):836–848.
62. Steidle G, Schick F. Echoplanar diffusion tensor imaging of the lower
leg musculature using eddy current nulled stimulated echo prepara-
tion. Magn Reson Med 2006;55(3):541–548.
63. Ercan AE, Techawiboonwong A, Versluis MJ, Webb AG, Ronen I. Dif-
fusion-weighted chemical shift imaging of human brain metabolites
at 7T. Magn Reson Med 2015;73(6):2053–2061.
SUPPORTING INFORMATION
Additional supporting information may be found in the online version of thisarticle
Supporting FIG. S1. Diffusion separation of lipid and lactate in the hema-toma region in the intracerebral hemorrhage (ICH) rat model. The DW-MRSexperiments were performed on the collagenase-induced ICH rat model at20 h post–collagenase infusion. The typical DW spectra acquired from thehematoma region are presented (a). The MRS voxel is indicated as the rec-tangle box in the T2-weighted image. The corresponding biexponential fit-ting of DW signals is shown in the upper-right corner. Lactate and lipidconcentrations calculated by the biexponential fitting in all six ICH rats (b).
10 Wang et al