Endocrinology and Metabolism


We validate the use of 1H magnetic resonance spectroscopy (MRS) to quantitatively differentiate between adipocyte and intracellular triglyceride (TG) stores by monitoring the TG methylene proton signals at 1.6 and 1.4 ppm, respectively. In two animal models of intracellular TG accumulation, intrahepatic and intramyocellular TG accumulation was confirmed histologically. Consistent with the histological changes, the methylene signal intensity at 1.4 ppm increased in both liver and muscle, whereas the signal at 1.6 ppm was unchanged. In response to induced fat accumulation, the TG concentration in liver derived from 1H MRS increased from 0 to 44.9 ± 13.2 μmol/g, and this was matched by increases measured biochemically (2.1 ± 1.1 to 46.1 ± 10.9 μmol/g). Supportive evidence that the methylene signal at 1.6 ppm in muscle is derived from investing interfascial adipose tissue was the finding that, in four subjects with generalized lipodystrophy, a disease characterized by absence of interfacial fat, no signal was detected at 1.6 ppm; however, a strong signal was seen at 1.4 ppm. An identical methylene chemical shift at 1.4 ppm was obtained in human subjects with fatty liver where the fat is located exclusively within hepatocytes. In experimental animals, there was a close correlation between hepatic TG content measured in vivo by 1H MRS and chemically by liver biopsy [R = 0.934;P < .0001; slope 0.98, confidence interval (CI) 0.70–1.17; y-intercept 0.26, CI −0.28 to 0.70]. When applied to human calf muscle, the coefficient of variation of the technique in measuring intramyocellular TG content was 11.8% in nonobese subjects and 7.9% in obese subjects and of extramyocellular (adipocyte) fat was 22.6 and 52.5%, respectively. This study demonstrates for the first time that noninvasive in vivo 1H MRS measurement of intracellular TG, including that within myocytes, is feasible at 1.5-T field strengths and is comparable in accuracy to biochemical measurement. In addition, in mixed tissue such as muscle, the method is clearly advantageous in differentiating between TG from contaminating adipose tissue compared with intramyocellular lipids.

  • fatty acid analysis
  • quantification
  • dog
  • rabbit
  • human
  • congenital generalized lipodystrophy
  • proton
  • skeletal muscle
  • liver

great strides have been made over the last few years in the use of magnetic resonance spectroscopy (MRS) for the noninvasive measurement of metabolic products and intermediates. 31P MRS has been applied to the study of energy metabolism, whereas 13C has been used to monitor carbohydrate metabolism (29, 52). Proton (1H) MRS has successfully been applied to the evaluation of brain disorders (29). In addition, it has proven to be a sensitive and precise tool for the quantification of gross tissue fat content (5, 49, 59, 64) and has been used for studying derangements in lipid metabolism in vivo (4, 6, 7,48, 54, 57, 60, 61, 62, 64).

The resonances from methylene and methyl protons of triglyceride (TG) acyl chains appear between 1.0 and 1.6 ppm. The exact significance of the multiple peaks arising in this region has recently been the subject of detailed investigation. Several studies of intrahepatic lipid accumulation (where adipocytes are absent) consistently report a single methylene signal although the chemical shifts were not reported (37,48). On the other hand, reports of studies of muscle have inconsistently assigned the methylene signal to one, two, or more peaks resonating from 1.26 to 1.6 parts/million (ppm) (3, 6, 13, 49, 54). Barany et al. (6) observed a single peak at 1.6 ppm in healthy human muscle in vivo and also reported a second peak at 1.4 ppm, which they considered to be uniquely characteristic of muscle undergoing myopathic degeneration. Recently, it was reported that in vivo localized proton spectroscopy can identify two sets of proton resonances from fatty acyl chains within muscle shifted in frequency from each other by 0.2 ppm (12, 57). Boesch et al. (12) and Schick et al. (57) suggested that these signals originate from two distinct compartments, an adipocyte and intramyocellular lipid pool. Magnetic susceptibility differences between compartments and the geometric arrangement of the tissue in musculature might cause the observed frequency shift. If the assignment of the spectral signal is correct, then MRS could be used to monitor the TG pool within adipose cells or in metabolically active cells such as myocytes. The two pool types are kinetically distinct in that the former is thought to turn over very slowly and serves as a long-term storage depot (7, 8), whereas the latter is thought to be in dynamic and rapid equilibrium with substrate utilization and supply (22, 49).

1H MRS has been used with some success to monitor TG content in studies of hepatic steatosis, and the results have generally correlated with histological and biochemical analysis; however, these relationships have not been assessed quantitatively (37, 48, 64). Although we and others have demonstrated a tight correlation between1H MRS and biochemical measurement of bulk TG content in skeletal (59) or cardiac muscle (49) ex vivo, to our knowledge a quantitative relationship has yet to be demonstrated in vivo. Previous attempts at quantifying skeletal muscle TG content by invasive biopsy have likely been confounded by tissue inhomogeneity, particularly due to the presence of contaminating adipose cell lipid (26, 57, 69). This problem is exemplified by large coefficients of variation (20–50%) for TG measurement in muscle biopsies (69), with the range for soleus muscle differing up to 20-fold (reviewed in Ref. 26). With regard to the use of 1H MRS for measurement of tissue TG levels, another factor that has not previously been taken into account is the influence of variable proton densities of different fatty acid species on the signals obtained.

The goal of the present study, therefore, was twofold: 1) to determine if the previous methylene and methyl proton resonance assignments are consistent with interventions known to modify TG concentrations and 2) to determine if 1H MRS can be used to quantify intracellular TG, particularly within muscle tissue. We have extended the previous observations of Schick et al. (57) and tested whether the low-frequency1 methylene and methyl resonances (1.40 and 1.0 ppm) are derived from a metabolically active intracellular lipid pool while the higher-frequency resonances stem from a storage pool within adipocytes. We conducted studies in both canine and rabbit models of induced fatty infiltration where changes in the level and/or location of TG pools as deduced from MR proton spectroscopy could be compared directly with biochemical and histological analysis of tissue biopsies. The influence of constituent fatty acid species on 1H MRS-calculated TG content was also directly assessed. Finally, we have examined the chemical shift of the methylene signal within adipose tissue, muscle, and liver in a variety of human subjects: healthy controls, hyperlipidemic individuals with fatty liver, and patients with congenital generalized lipodystrophy (CGL). The results support prior assignments and suggest that under defined conditions the intracellular TG store can be quantified by1H MRS in vivo.



Mixed-breed dogs (7–20 kg) were obtained through the University of Texas Animal Resources Center from outside suppliers. Retired female breeder New Zealand White rabbits were obtained from Myrtle’s (Thompson Station, TN). Animals were allowed to acclimate with ad libitum access to chow before experimentation. Animals were variably fasted from 4 to 12 h on the day of experimentation. All procedures were approved by the University of Texas Southwestern Institutional Review Board for Animal Research.

Norepinephrine Infusion

Prolonged exposure to intravenous catecholamines has been shown to accelerate adipocyte lipolysis and cause generalized organ fat infiltration (70). Norepinephrine bitartrate (A9512, Sigma, St. Louis, MO) was dissolved before use in sterile buffer containing 114 μM ascorbic acid, 54 μM sodium EDTA, and 50 IU/ml heparin. Dogs were anesthetized (10–25 mg/kg iv thiopental induction) followed by isoflurane/oxygen via inhalation. The level of anesthesia was deepened during liver studies to minimize respiratory motion. The left leg was subjected to 1H MRS before biopsy. Tissue was blotted to remove excess blood and was either immediately frozen in liquid nitrogen or in supercooled isobutane for biochemical or histological analysis, respectively. Animals were then fitted with subcutaneous miniosmotic pumps (Alza model 2ML1) delivering norepinephrine at 0.5 μg ⋅ kg−1 ⋅ min−1intravenously in the right jugular vein. After recovery, animals were given ad libitum access to 10% glucose drinking water. Twenty-four hours after placement of the norepinephrine pump, animals were reanesthetized, and 1H spectra and biopsy material were obtained from the opposite leg and liver. Additional animals were studied only one time to obtain baseline 1H spectra and liver biopsies.

Inhibition of Hepatic Fat Oxidation with CPI-975

CPI-975 is an inhibitor of carnitine palmitoyltransferase I, the rate-limiting step in mitochondrial long-chain fatty acid oxidation (2), and was obtained from Novartis Pharmaceuticals (Summit, NJ). Overnight-fasted rabbits were given intravenous boluses of CPI-975 dissolved in 45% Molecusol HPB (CTD, Gainesville, FL) at a dose of 2 mg/kg hourly and were studied 2–6 h after the first injection. They were anesthetized with isoflurane/oxygen, and hepatic triglyceride content was measured by 1H MRS. Animals were then killed, and tissue samples were processed as described inTissue TG.

Biochemical Analyses

Tissue TG.

Small amounts (∼0.5 g) of tissue were powered under liquid nitrogen, and aliquots were taken in triplicate for TG measurement. Total lipids were extracted with 30 vol (vol/wt) of acidified chloroform-methanol (2:1). After removal of the upper phase and conversion of the infranatant to a uniform phase with methanol, total TG content was determined enzymatically with a commercially available assay (no. 337; Sigma) using triolein as the calibration standard. Tissue TG fatty acids were analyzed by gas chromatography after hydrolysis and conversion to individual methyl esters (71). Tissue (300–500 mg) was powdered under liquid nitrogen and extracted with 50 vol of Folch reagent (2:1 chloroform-methanol). This was passed through filter paper and reconstituted to 100 ml with fresh Folch. The chloroform phase (4 ml) was dried under nitrogen gas in 13 × 100 glass tubes. Two milliliters of freshly prepared methanol-benzene 4:1 were added followed by 200 μl acetyl chloride with gentle vortexing. Tubes were closed with Teflon-coated caps and heated at 100°C for 1 h. Samples were then placed in an ice bath, and 5 ml of 6% K2CO3 followed by 1.5 ml of benzene were added with vortexing in between. Samples were then centrifuged at 2,000 rpm for 10 min. The top (benzene) layer was removed to 12 × 75 glass tubes and dried under nitrogen gas at room temperature. Methyl ester samples were reconstituted in 0.5 ml hexane and transferred to gas chromatography (GC) vials (Hewlett Packard, Palo Alto, CA). These were injected in a model 5890A HP gas chromatograph containing a 50-m long, 0.25-mm ID fused silica capillary column (SP CP Sil88) using nitrogen as carrier gas at 58 psi. Samples were injected at 110°C and run at 175–220°C with total run time of 32 min. Individual fatty acids were identified based on retention time compared with a standard free fatty acid (FFA) mixture, and abundances were calculated from area under the curve measurement. TG contents are expressed as micromoles per gram wet weight.

Plasma FFA were measured using a commercially available assay kit (Boehringer-Mannheim, Indianapolis, IN). Plasma TG were measured with the same kit as described above (Sigma 337). Plasma osmolarity was measured by freezing-point depression on an Advanced Osmometer model 3D3 (Norwood, MA), and serum sodium was measured by a commercial autoanalyzer. All chemicals were obtained from Sigma/Aldrich and were of the highest grade possible unless otherwise specified.


Muscle for histological analysis was mounted in a cross-sectional orientation and snap-frozen in isopentane cooled in a liquid nitrogen bath. Frozen tissue was cut into 6- to 10-μm sections, which were then air-dried and stained with oil red O for neutral lipid. Muscle fiber type was determined using established enzyme histochemical stains for NADH tetrazolium reductase and myofibrillar ATPase activities (56).

NMR Spectroscopy

Image-guided, 1H localized, MRS and high-resolution T1 weighted imaging were performed on a 1.5 T Gyroscan NT whole body system (Philips Medical Systems) using a combination of whole body and knee coils for radiofrequency transmitting and signal receiving or body coil alone for studies of skeletal muscle and liver, respectively. Imaging parameters were chosen for suitable separation of muscle, fascia, and fat [repetition time (TR) = 400 ms, echo time (TE) = 30 ms]. Volumes of interest within muscle were centered over the gastrocnemius/lateral digital flexor and tibialis anterior (12 mm3) in dogs, and over midsoleus in humans (12–15 mm3) with minimal contribution from gastrocnemius. Voxel positions were placed to avoid vascular structures and gross adipose tissue deposits and to ensure consistent orientation of muscle fibers along the magnetic field. Similarly, volumes of interest in liver were located away from major vascular structures, typically within the right lobe, with size varying from 30 to 50 mm3. Automatic routines were employed for course and fine shimming, with typical line widths of the water signal of ∼10 Hz being obtained in muscle. Localized proton spectra within muscle were collected using a PRESS sequence with the following parameters: TR = 5 s, TE = 33 ms, 1,024 data points over 1,000 kHz spectral width. The long interpulse interval was chosen to ensure fully relaxed water signal (99.2%), since it served as an internal standard for quantitation. In liver, spectra were collected using the same sequence and parameters, except for TE = 40 ms. Similar procedures were followed for dogs, rabbits, and human subjects.

Spectral Data Evaluation

The high-resolution spectrum of a TG mixture with chemical shift assignments is shown in Fig. 1 (63). Signals in spectra of skeletal muscle, fat, and liver in vivo were assigned and processed as follows. Signal in the time domain was multiplied by a second-order Gaussian function before Fourier transformation and manual phase correction. Employing prior knowledge (23) of signals from the polyolefinic/monolefinic methylenes α-CH2, (CH2)n−2, and CH3 (57), water and fat resonances were line fit using a mixed Lorentzian/Gaussian function, and methylene and methyl peak areas were quantified using a standard analysis package (NUTS; ACORNNMR, Fremont, CA; Fig. 2). Chemical shifts were measured relative to water at 4.80 ppm. The water signal was integrated over 4.4–5.2 ppm. Methylene and methyl signals were integrated over 0.8–2.0 ppm. The (CH2)nsignal was used for calculation of intracellular TG content due to its higher signal intensity and narrower line width compared with the CH3 resonance. Intracellular TG content (expressed as methylene-to-water peak area ratio × 100) was corrected for spin-spin relaxation time (T2) relaxation of water and fat (Table 1). T2 values were calculated according to the T2 relaxation equation (29, 52) and found to be comparable to those described previously (48, 57, 64). The (CH2)n−2 at 1.40 ppm was used to represent the intramyocellular TG (IMCL) pool due to its high intensity. The IMCL-β-CH2 is obscured by the signals from extramyocellular (EMCL) (CH2)n−2arising from adipose TG at 1.60 ppm (compare Fig. 1 vs. Fig. 2). In liver, the β-CH2 was not resolved from the (CH2)n−2, so both were assumed to contribute to the proton signal for quantification purposes. This is possible since there is no adipocyte-associated TG within liver parenchyma.

Fig. 1.

High-resolution proton spectrum of soybean oil acquired at 300 MHz using a Varian-Inova spectrometer and a single-pulse sequence. Resonances are assigned to protons of fatty acid chains and glycerol as indicated. Assignments are relative to CH3 at 1.0 parts/million (ppm).

Fig. 2.

Proton spectrum of human soleus muscle. A: raw spectrum;B: deconvolution with fitted peaks for signals identified by prior knowledge; C: difference spectrum after subtracting fitted curves from B from raw data in A. EMCL, extramyocellular; IMCL, intramyocellular. IMCL CH3(1) and CH2 (3) at 1.0 and 1.4 ppm; EMCL CH3 (2) and CH2 (4) at 1.2 and 1.6 ppm; IMCL + EMCL α-CH2 (5, 6) and polyolefinic/monolefinic (7, 8) CH2 overlap and were approximated by four peaks.

View this table:
Table 1.

T2 relaxation times of various resonances in muscle and liver


For conversion to an absolute scale, methylene proton peak areas were corrected for molar proton densities of methylenes from TG containing the most common acyl chains (14:0, 16:0, 16:1, 18:0, 18:1, 18:2, and 18:3). Further corrections were made for the density of component TG at 37°C (34), nonfat mass relative to water mass (36, 43, 50, 58), and tissue density (30). The proton density of water at 37°C was taken as 111.1 mmol/ml. Tissue water content was assumed to be constant since there was no change in plasma sodium or osmolality, which provides evidence of dehydration when elevated (data not shown; see Ref. 68). The above procedure was applied similarly to rabbits, dogs, and humans. Calculations are presented in the .

Human Subjects

The characteristics of the study subjects are given in Table2. All were studied on an ad libitum diet in the postprandial state. Nine were normolipidemic subjects without evidence of metabolic disease, five were obese with fatty liver, and four had CGL. All subjects gave informed consent, and all procedures were approved by the University of Texas Southwestern Medical Center Institutional Review Board.

View this table:
Table 2.

Subject characteristics

Cadaveric diaphragm muscle tissue was obtained from three individuals. One individual died suddenly of presumed cardiac arrest, and two succumbed to accidental trauma. All had been described as being in good health before death. Pleural and peritoneal membranes were dissected free, and pure muscle was extracted for TG fatty acid analysis as described in previous paragraphs. The diaphragm is a striated muscle and has little to no associated interfacial fat. It was therefore considered to be representative of other striated muscles for purposes of intracellular lipid analysis.


Regression analyses were performed using an automated curve fitting program (Sigmaplot; Jandel Scientific, San Rafael, CA). Comparison of the biochemically vs. spectroscopically measured TG concentrations was performed with the Pearson product moment statistic. Group means were compared utilizing a paired Student’s t-test or the Student-Newman-Keuls test when appropriate, with P < 0.05 considered significant.


Norepinephrine Infusion: A Model of Intracellular Fat Accumulation

Norepinephrine infusion into dogs caused a modest elevation in plasma FFA and TG concentrations (data not shown). At baseline in liver,1H lipid signals were undetectable, and this was matched by minimal amounts of TG measured biochemically (0 ± 0 and 2.1 ± 1.1 μmol/g). In muscle, strong methylene signals were observed at both 1.4 and 1.6 ppm (Table 3and Fig. 3). After 24 h of norepinephrine infusion, there was substantial accumulation of hepatic lipid with a single (CH2)n−2methylene peak identified at 1.4 ppm. Intrahepatic TG concentrations increased similarly when measured by spectroscopic and biochemical means (44.9 ± 13.2 and 46.1 ± 10.9 μmol/g, respectively; Fig.3 A). In muscle, there was a trend toward a decreased EMCL at 1.6 ppm, whereas the total muscle lipid (TML) signal in the posterior or anterior leg muscles was unchanged (Fig. 3, B andC). In contrast, the IMCL at 1.4 ppm increased significantly with calculated concentrations, changing from 3.5 ± 2.1 to 11.4 ± 3.9 μmol/g within the gastrocnemius/lateral digital flexor complex and from 6.2 ± 2.3 to 10.5 ± 2.2 μmol/g in the tibialis anterior (P < 0.05 for both). Biopsy measurement of muscle TG was greater than IMCL but less than TML calculated from the spectroscopy data, consistent with residual adipocyte contamination despite attempts to dissect away all visible contaminating fat.

View this table:
Table 3.

Chemical shifts of triglyceride methylene (CH2)n−2 protons

Fig. 3.

Changes in liver (A) and muscle (B and C) triglycerides (TG) in dogs before and after norepinephrine (NE) infusion monitored by 1H magnetic resonance spectroscopy (MRS) and biochemical measurement. B: gastrocnemius complex;C: tibialis anterior. 1H MRS TG, proton MRS measurement, IMCL, intramyocellular (1.4 ppm); EMCL, extramyocellular (1.6 ppm); TML, total muscle lipid; BX-TG, biopsy with biochemically measured TG content. * P < 0.05 post- vs. pre-NE.

The accumulation of intrahepatic and intramyocyte TG was further confirmed histologically (Fig. 4). Review of the sections by a pathologist blinded to the treatment status of the animals was consistent with greater intensity of neutral lipid staining by oil red O postnorepinephrine compared with that at baseline in both liver and muscle. Staining for muscle fiber type confirmed the preferential accumulation of lipid within type I (oxidative) fibers compared with type IIa or IIb (glycolytic) fibers.

Fig. 4.

Histological analysis of dog liver (A and B, magnification ×62.5) and soleus muscle (C and D, magnification ×31.2) before and after NE, respectively. Tissue sections were processed as described in methods and were stained with oil red O, which specifically stains neutral lipids. Note the increase in lipid staining after NE (A and C, before NE; B and D, after NE).

Validation of Biochemical vs. Spectroscopic Measurement of TG

Using data from the dogs subjected to norepinephrine, from the rabbits treated with SDZ CPI-975, and normal fasted animals, biochemically measured hepatic TG content vs. spectroscopic TG abundance was compared (Fig. 5). In normal healthy liver, TG levels were minimal as has been previously reported (19, 24, 28, 45, 46, 70). Both norepinephrine and SDZ CPI-975 acutely increased intrahepatic TG stores from baseline, as expected. Norepinephrine markedly increases adipose tissue lipolysis, resulting in generalized organ TG accumulation (70). CPI-975 blocks mitochondrial long-chain fatty acid oxidation, causing increased cytoplasmic stores of TG (2). In the 13 animals studied, there was a very good correlation between methods with r = 0.934 (P < 0.0001). The slope of the line was 0.98 [confidence interval (CI) 0.70–1.17] with a y-intercept of 0.26 (CI −0.28 to 0.70), suggesting that the technique is precise for quantification of TG in vivo.

Fig. 5.

Correlation between measurement of hepatic TG concentration by1H MRS and biopsy. Dogs and rabbits were studied before and after NE or the long-chain fatty acid oxidation inhibitor (CPI-975). Group sizes are indicated in parentheses.

The intensity of the methylene proton signal (which makes up the bulk of the lipid signal in liver) is dependent not only on the absolute concentration of tissue TG but also on the relative length and saturation of the constituent fatty acids. The fatty acid profiles determined by GC from liver tissue TG of the rabbits and dogs are given in Table 4. For individual fatty acids this was significant for the 14:0, 16:0, 18:1, and the 18:3 moieties, but the calculated proton densities used for determining tissue TG differed by <1% (70.1 vs. 70.6 mol [CH2]n−2 + β-CH2protons/ml).

View this table:
Table 4.

Analysis of liver and muscle triglyceride fatty acids

Studies in Human Subjects

The chemical shift of the methylene proton signals was measured in liver, subcutaneous juxtahepatic adipose fat, and muscle in normal controls, subjects with fatty liver, and patients with CGL and compared with data from dogs and rabbits (Table 3). Methylene resonances have a chemical shift of either ∼1.4 or 1.6 ppm, or both, depending on the tissue studied. A single methylene signal located at 1.6 ppm is observed in fat and at 1.4 ppm in liver (Table 3 and Fig. 6, Aand B). In contrast, two signals are observed in normal muscle (Figs. 2 and7), consistent with prior reports (12, 57) suggesting the existence of an intracellular TG pool and an EMCL (adipose) tissue TG pool resonating at ∼1.4 and at ∼1.6 ppm, respectively. Two peaks with these same chemical offsets were observed when the voxel was placed simultaneously over liver and juxtahepatic fat (Fig. 6 C).

Fig. 6.

1H MR spectra of the abdomen of a human subject with steatosis. Voxel was sequentially placed over subcutaneous juxtahepatic fat (A), liver (B), or both (C). Chemical shifts were measured relative to tissue water at 4.80 ppm. Note that a single methylene peak originates from adipose fat and is located at 1.6 ppm, whereas the intracellular fat within liver appears at 1.4 ppm. IHL, intrahepatic lipid; EHL, extrahepatic (adipocyte) lipid.

Fig. 7.

1H MR spectra from the soleus muscle of a normal control (A) and a subject with congenital generalized lipodystrophy (CGL, B). Note in the normal control there are two methylene peaks at 1.4 and 1.6 ppm, whereas in the CGL subject there is only a single peak at 1.4 ppm. This finding is consistent with a known lack of interfacial adipose tissue in CGL subjects (see also Fig. 8).1H MRS only detects IMCL in these subjects. Car, cartinine; Cr, creatine.

Subjects with CGL

Patients with CGL are known to have almost no detectable adipose tissue, especially in metabolically active areas investing muscle groups (1, 18, 31, 61). This was confirmed with the high-resolution T1 weighted images (Fig. 8). When studied with 1H MRS, no signal was detectable from muscle at 1.6 ppm, whereas there was a strong signal located at 1.4 ppm (Table 3 and Fig. 7 B). Identical results were obtained in all subjects with lipodystrophy in whom the IMCL content was two times as high as in normal controls (19.8 ± 4.6 vs. 10.7 ± 1.4 μmol/g, P < 0.05) and no EMCL was detectable (0 ± 0 vs. 27.2 ± 2.0 μmol/g,P < 0.05). In a fifth subject with CGL who had never undergone spectroscopy but had died and undergone autopsy evaluation, skeletal muscle tissue was available for analysis (18). On careful review of multiple sections, only rare adipocytes were identified in the subdermal layers and connective tissue elements (data not shown).

Fig. 8.

T1 weighted transaxial image of the calf in a normal individual (A) and in a subject with CGL (B). Imaging parameters are given in methods. Images are displayed with the same window and level. Small amount of adipose tissue is detected subdermally but not within muscle itself in CGL patients. Magnetic resonance imaging shows an absence of fat in the “gross” intermuscular fasciae compared with normal subjects.

Reproducibility of 1H MRS Measurement of Intracellular Lipids

Reproducibility of intracellular TG measurement was estimated by studying five lean individuals three times and five obese subjects two or three times during the same spectroscopy session. Subjects were removed from the magnet between measurements. A similar positioning of the voxel was attempted for each measurement. The mean coefficient of variation for the calculated intracellular skeletal muscle TG level was 11.8 and 7.9% for lean and obese and 22.6 and 52.5% for the extracellular (adipose) TG, respectively.


The possibility of using 1H MRS to detect different cellular compartments for stored lipids in vivo has only recently become appreciated (57). Nevertheless, the proper assignment for TG methylenes within muscle and other organs has remained in doubt due to conflicting findings (3, 6, 12, 13, 49, 53). In addition, the ability of 1H MRS to quantify absolute concentrations of intracellular TG in vivo has not been previously validated, particularly taking into account the influence of different fatty acid species on calculated TG abundance. In this study, we demonstrate that methylene signals deriving from TG acyl chains that are not isotropically restricted resonate at either 1.6 or 1.4 ppm, depending on their anatomic location, i.e., either in bulk fat deposits or from within nonadipose cells such as myocytes and hepatocytes. This is entirely consistent with previous reports showing two fat signals originating from muscle (12, 13, 54, 57) and the suggestion that two lipid compartments are present within the muscle bed (57). Various kinds of data now exist in support of the notion that the methylene signal identified at 1.4 ppm derives from intracellular lipid stores and that at 1.6 ppm is due to TG bulk storage depots in adipose tissue. We have directly measured tissue TG fatty acid composition and, utilizing the proper methylene assignments, show for the first time that 1H MRS can quantitatively measure intracellular TG in vivo at 1.5 T.

All in vivo studies to date that have carefully examined the proton frequency of methylene signals emanating from tissues without associated adipose cells find that they have a single resonance at 1.3–1.4 ppm. This was true for isolated cells in tissue culture (39), whole liver (this study), and cardiac muscle (4). By contrast, in adipose tissue a single fat resonance is also seen, but it is located at 1.5–1.6 ppm (12, 57, and this paper). As with the “mixing” studies originally reported by Schick et al. (57) in which a voxel centered over muscle and fat gave rise to two peaks separated by 0.2 ppm, we found that the same is true when the VOI is centered over liver and adipose fat. By its nature, muscle has two compartments, myocytes themselves and investing facial layers that run along the myofibril bundles and are difficult to separate from each other macroscopically. Included in our study group were four subjects with CGL in whom there is essentially no adipose tissue within muscle planes when studied either by high-resolution magnetic resonance imaging or with histological methods (1, 18, 31, this report). Consistent with the expectation that the methylene signal deriving from adipose tissue is located at 1.6 ppm, this peak was absent in the CGL patients, whereas a strong peak was identified at 1.4 ppm, indicating the presence of substantial quantities of TG in an IMCL compartment. Thus localized 1H MRS seems to be highly sensitive in revealing intracellular fat (compare Fig. 7 with Fig. 8) and is therefore superior to imaging methods in studies of intracellular lipids.

Several lines of physiological data also support the notion that the methylene peak at 1.4 ppm (and the methyl peak at 1.0 ppm) obtained from whole muscle 1H MRS represents the IMCL storage pool of TG while the resonance at 1.6 ppm (and methyl at 1.2 ppm) represents EMCL (adipocyte-based) lipid. The IMCL peak has been shown to scale with markers of muscle mass, such as creatine, carnitine, and water, whereas the EMCL peak at 1.6 ppm does not (12). This same group also reported that the IMCL peak decreased in one subject who underwent prolonged aerobic exercise and then recovered to baseline after rest and eating. In another animal model, Balschi et al. (4) reported increases in intramyocardial TG in postischemic dog heart preinfused with a lipid emulsion. Localized 1H MRS in vivo of the anterior ventricular wall demonstrated increases in the methylene resonance at 1.3 ppm that generally correlated with biochemical and histological indexes of TG accumulation.

The above studies clearly suggest that an intracellular TG pool exists and that it can be monitored via changes in the methylene and methyl peaks at 1.4 and 1.0 ppm. The increase in intracellular TG monitored by1H spectroscopy was independently confirmed in our studies by direct biochemical and histological measurement. As seen in Figs. 3and 5, the measurement is linear over the range of intracellular TG encountered in normal tissues (0–30 μmol/g; see Refs. 19, 25,45, 46, 58, 64) and over that seen in pathological conditions (30–60+ μmol/g; see Refs. 24, 25, 45, 64, 70). The spectroscopic measurement of TG is not sensitive to contamination by plasma TG contained within the hepatic blood pool. Only signal from stationary tissue and fluids is detected under the present experimental conditions (29). It can be calculated that residual plasma TG can account for 0.1–0.3 μmol/g under normal conditions and more in the setting of severe hypertriglyceridemia.

It should be noted that biochemically measured fasting hepatic TG content is somewhat species and protocol dependent. Data recalculated from the literature using standard error measurements indicate that levels have been reported as low as 1–2 μmol/g in dogs (14, 15,25, 70), 0.1–6.0 μmol/g in rabbits (33, 38, 41), 3–6 μmol/g in rats (16, 17, 51, 65), and 3–17 μmol/g in humans (19, 28, 45, 46, 58, 64). We suspect that the fact our basal biochemical measurements were on the low side of those previously reported was due to the exposure of animals to deep and prolonged inhalational anesthesia. This setting was likely a strong stimulus to intrahepatic lipolysis; therefore, unstressed basal levels would have been higher. Nonetheless, the levels of intrahepatic TG were consistent when measured by either technique, this being the major focus of this report. From our experiments, it appears that the 1H MRS sensitivity threshold in liver under the present conditions lies between 2 and 13 μmol/g. This may be more precisely defined with further improvements in the technique and refinements in MR technology. Fortunately, this should not prevent the application of this technique to hepatic studies in humans and conscious unstressed animals where levels are anticipated to be greater than ∼13 μmol/g.

One caution that should be kept in mind is that, when hepatic TG stores are minimal, spectroscopy will likely only detect the storage pool of mobile lipids (see below). In contrast, biochemical analysis may in addition detect that small amount that is entrained in the biosynthetic pathway of hepatic very low density lipoprotein (VLDL) production. Because these TG are tightly associated with the endoplasmic reticulum and lipophilic assembly proteins such as apolipoprotein B100 and microsomal transport protein (35), they may not be visible by nuclear magnetic resonance (NMR; see Refs. 10, 40). This limitation, however, should not apply to muscle, since no VLDL TG are synthesized, and all TG are stored as lipid droplets (67).

The use of the methylene peak is quantitative for measurement of TG by NMR (Fig. 5), given knowledge of the fatty acid composition. Theoretically, some error could occur if the composition of the fatty acids within TG differs substantially from that expected. Despite dramatic species differences in dietary preferences, which were reflected in the fatty acid composition of liver lipids between dogs and rabbits, the calculated proton density of tissue TG was hardly affected (Table 4). This is largely due to the predominance of three fatty acids (18:0, 18:1, and 16:0) accounting for over two-thirds of all fatty acids. Although extremes of diet do occur and can be reflected in the fatty acid composition of adipose TG, the latter rarely changes by more than a few percent (11, 42, 66). In fact, the proton density of the adipose fat from Pacific Islanders eating a Western diet vs. the indigenous subsistence diet differs in proton density of adipose TG only by 3% (11). The only spectroscopic measure of TG that is independent of fatty acid composition would be that derived from methyl proton abundances at 1.0 ppm (see Fig. 1). Unfortunately, these peaks are inadequately resolved at 1.5 T measured at a TE = 33 ms and therefore do not lend themselves to accurate measurement at routinely available field strengths.

There are several other potential sources of error in our spectroscopic assessment of intracellular TG concentration. Mobile lipids such as long-chain acylcarnitine and acyl-CoA esters could also contribute to the methylene signal. However, the molar content of acyl esters is <1% of tissue TG when measured biochemically (20). Another source of signal contamination is from the methyl group of lactate resonating at ∼1.3 ppm. Under in vivo conditions and in the absence of tissue hypoxia, tissue lactate concentrations are 0.5–1.0 μmol/g (3,44). At an echo time of 33 ms, lactate methyl signal is dephased by 50% due to J-modulation such that the residual signal is <1–10% of the total methylene signal. The properties of cellular phospholipids and cholesterol esters should also be considered. The absolute concentrations of these moieties is low, on the order of 5–10 (27) and 0.5–1 (24, 45, 55, 72) μmol/g, respectively. More important, they are known to be anisotropic due to their association with plasma membranes, and the resulting extremely broad lines (>10 kHz) cause them to be lost in the background under the present experimental conditions (10, 40). Last, line broadening due to respiratory motion is problematic in spectroscopy of liver without gating and could have affected the accuracy of peak deconvolution and quantification; however, this concern would not apply to static measurements in muscle.

In muscle, excessive EMCL signal created difficulties with deconvolution. This could be avoided by careful placement of the VOI to avoid fat deposits visible on the high-resolution locator images and, when necessary, decreasing voxel volumes, e.g., 1 cm3. Interestingly, obese subjects who generally had higher IMCL concentrations had a lower coefficient of variation than lean subjects. In contrast, EMCL measurements were highly variable, consistent with a greater degree of inhomogeneous tissue marbling. Thus we conclude that this technique is applicable to all individuals irrespective of their degree of obesity. Based on the 1H MRS data, we estimate a current practical lower limit of detection of 0.5% CH2-to-H2O ratio (equivalent to ∼7 μmol/g) in liver. In liver, respiratory motion-induced magnetic field inhomogeneities decrease sensitivity and precision due to accentuated line broadening. In muscle, the technique easily resolved the lowest IMCL contents observed to date (3–4 μmol/g); however, the true lower limit was not further delineated.

The physical basis for the observed 0.2-ppm shift in lipid resonances between the IMCL and EMCL storage pools is the distribution of magnetic susceptibility in tissue (21). The factors that control the distribution of magnetic susceptibility within tissue compartments are incompletely understood and may be due to ultrastructural or microanatomic differences. IMCL represents a metabolically highly active pool that is optimized for rapid turnover and supply of lipid substrates for cellular oxidation. As such, lipid droplets are very small (<200 Å radius) and are associated with cytoplasmic components and many associated enzymes involved in fatty acid esterification, hydrolysis, and transport into the mitochondrion (67). EMCL turns over slowly, and thus TG exist in bulk with relatively little associated enzymatic components or cytoplasmic matrix. It is therefore possible that the differences in bulk magnetic susceptibility of TG arise, at least in part, from the more highly polar environment for IMCL compared with that for EMCL. Another recently recognized ultrastructural difference between the lipid pools is the unique presence of the phosphoprotein perilipin ringing the intracellular lipid droplets within adipocytes but not within other cell types such as hepatocytes (9).

The difference in bulk magnetic susceptibility accounting for the observed chemical shift may be an intrinsic property of lipid compartmentation (21). Using this theoretical framework, Boesch et al. (12) have attempted to explain the differential chemical shift of EMCL and IMCL resonances. According to the theory of Chu et al. (21), the resonance from IMCL, which is contained within spherical droplets, remains independent of orientation relative to the magnetic field (12,21). In contrast, the EMCL has a small magnetic moment due to the triglycerides being located in an annular compartment oriented along muscle fibers and connective tissue. The EMCL resonance is affected by its orientation relative to the magnetic field. When the muscle fibers are oriented parallel to the magnetic field, it can be calculated using the above approach, that EMCL will resonate 0.2 ppm higher in frequency (12). In our experiments, which focused attention on the soleus muscle, the fibers were nearly parallel to the Z-axis of the magnetic field thus maximizing the observed shift difference between IMCL and EMCL resonances. Misalignment of muscle fibers relative to the magnetic field can be predicted to broaden and shift the EMCL line, and for some orientations overlap of IMCL and EMCL resonances may occur. This could create errors in quantification of IMCL and thus should be avoided.

In summary, we provide several novel lines of evidence that an intracellular pool of lipid in nonfat cells, predominantly TG, is spectroscopically distinct from fat stored within adjacent adipocytes and that it can be detected at 1.5 T. The presence of a methylene peak at 1.40 ppm and methyl peak at 1.0 ppm (relative to water at 4.80 ppm) appears to be a general property of the TG storage pool within cells, regardless of tissue type. We confirm earlier reports that the measurement of intracellular TG by 1H MRS correlates highly with biochemical analysis and show, for the first time, that given some reasonable assumptions on the general abundance of fatty acid species, this is quantitative. In light of the increasing interest in the noninvasive and accurate measurement of intracellular lipids in vivo, this method should find great applicability in studies of normal and disturbed fat metabolism, including exercise, starvation, obesity, diabetes, and various states of insulin resistance.


We thank Dr. Craig Malloy for technical suggestions and for critical reading of the manuscript, Murphy Daniels for technical assistance, Ben Alexander for help with animal studies, Laura Lee for tissue processing for histochemistry, and Steve Turley and Brian Jimmerson for triglyceride fatty acid analysis. We are grateful to Drs. Ron Peshock and Paul Wetherall at the Roger’s Magnetic Resonance Center for general technical support. Last, we thank the volunteers who gave their time to participate in this study.


  • 1  Lower frequency is classically defined as a parts per million shift that decrease in absolute value.

  • This study was supported in part by grants from Philips Medical Systems, Hoechst-Marrion Merrill Dow, National Institutes of Health (NIH) Grants P41-RR-02584 (Facility Grant) and RO1 DK-53358-01 (D. T. Stein), the Lattner Foundation, and Juvenile Diabetes Foundation/NIH no. 995003.

  • Address for corrspondence and reprint requests: D. T. Stein, Department of Medicine, Rm. G47, Albert Einstein College of Medicine, 1300 Morris Park Ave., New York, NY 10461 (E-mail:dstein{at}aecom.yu.edu).


Calculated peak areas of methylene and water signals were corrected for T2 relaxation according to the spin-spin relaxation equation (52) using the relaxation times from Table 1 and spin echo times as in methods.


tissue mass


fat mass


lean tissue mass


tissue water


1H spectroscopically derived fat-to-water ratio


weighted proton density of TG relative to water (mol/mol in a volume)


tissue density (g/ml)


tissue TG density relative to triolein standard

Let TG content in units of micromoles per gram tissue was calculated as follows. LetM=FM+LTM=1 Equation 1

The proportion of body water to ash and protein is relatively invariant except under conditions of acute changes in hydration (47). Therefore, within a volume of tissue,W=TW/LTM Equation 2This is a constant value for a given tissue and is available from standard tables (36, 43, 50, 58).

Rearranging Eq. 2, one getsLTM=TW/W Equation 3From the proton spectra one hasZ=T2corrected methylene­to­water ratio from the proton spectrum fromastandard1.0­ml voxel Equation 4 =crude fat signal/tissue water signal=FM/TW LetP=weighted relative proton density(mol/mol)of tissue TGvs.water Equation 5

The methylene (CH2)n−2 and [(CH2)n−2 + β-CH2] proton densities for the various fatty acid species in TG are given in Table 5. Using the tissue fatty acid analysis (e.g., Table 4) a weighted proton density is calculated relative to water.Proton density of water=111.1mmol proton/ml water e.g.Proton density of hepaticTG(CH2)n2+β­CH2 =70.35mmol proton/mlTG =70.35/111.1=0.6332 Proton density of skeletal muscleTG(CH2)n2 =65.2mmol proton/mlTG =65.2/111.1=0.5869 Rearranging terms from Eq. 4 and 5 to calculate absolute quantities we obtainTW=FM×P/Z Equation 6Solving for fat mass, rearrange Eq. 1 FM=1LTM Equation 7Substituting Eq. 3 for LTMFM=1TW/W Substituting Eq. 6 for TW one getsFM=1FM×PZ×W Equation 8Rearranging terms givesFM=ZWZW+P Equation 9

View this table:
Table 5.

Physical properties of common triglycerides

The weighted molar density of tissue TG is converted from proton molar density per unit volume to proton molar density per unit weight by correcting for tissue density. The calculated fat mass is further corrected for the weighted density of the TG fatty acids relative to the triolein standard and lastly converted to molar mass by dividing by the triolein standard molecular weight (885.4)FM(μmol/g)=ZW×106885.4DT(ZW+P) Equation 10

The weighted densities calculated from Tables 4 and 5 are as follows: dog and rabbit liver = 1.025; dog skeletal muscle = 1.021; and human skeletal muscle = 1.024.

Relative tissue water content to total weight (kg/kg) was as follows: dog and rabbit liver = 0.725 (36, 43, 50); dog and rabbit skeletal muscle = 0.688 (43, 50); and human skeletal muscle = 0.810 (58).

Tissue density (g/ml) was as follows: liver = 1.03 and skeletal muscle = 1.05. Data were taken from Ref. 30.


View Abstract