Proton magnetic resonance spectroscopy (1H-MRS) of skeletal muscle provides complex spectra that offer a wealth of information (1–11). Metabolites that are detectable include carnosine (12, 13), taurine (14, 15), choline-containing compounds (13), creatine/phosphocreatine (16, 17), extra- (EMCL), and intramyocellular (IMCL) lipids (1, 2). Recently, new knowledge in the field brought the discovery of the effect of muscle fiber orientation (18–20). MRS of skeletal muscle showed a great potential for investigating physical and biochemical changes induced by disease processes (4, 6, 9, 11, 16) or physiological activity (7, 21, 22) and enhanced understanding of the underlying biophysics. The growing interest in this field raised the issue of absolute quantification of muscular metabolite concentrations to allow for the comparison of different populations and assessment of metabolic flux rates under in vivo conditions.
The unsuppressed water line is almost exclusively used as a concentration reference in the processing of muscle spectra. The acquisition of the water line has found widespread use in single-voxel MRS, particularly because it is a simple matter of acquiring a couple of scans from the measured volume of interest (VOI) either immediately before, or after, the water-suppressed acquisition. The most important advantage of water reference is the high spectral intensity. The large water signal provides excellent signal-to-noise ratio data that can be used to determine local magnetic field (B0) shifts, to correct for eddy-current distortions, and to assist in phasing of the spectra (23). In contrast, the acquisition of a water MR spectroscopic imaging (MRSI) data set in addition to the water-suppressed is not always done. If performed as an additional acquisition it can lengthen the study undesirably, especially for volumetric (three-dimensional) MRSI. In spectroscopy of the skeletal muscle, the necessity for relaxation correction of water reference is a fundamental disadvantage. This correction complicates the relative wide range of T2 values (25 < T2 < 35 ms, B0 = 1.5 T) (7–9, 17, 21, 24–26), which depend on the age, type, and distribution of muscle fibers.
Alternatively, the fat signal can be used as the reference in quantitation of muscle lipids. Yellow bone marrow (11, 22), adipose, or subcutaneous fat (27) can serve as the internal standard, because EMCL and IMCL have essentially the same composition of fatty acid triglycerides with very close relaxation times T1, T2 (1, 8, 9). The fat reference is extraordinarily suitable for the quantitation of muscular lipids because no relaxation corrections are needed.
The main purpose of the present work was to compare two methods for the determination of lipid content in human skeletal muscle: relaxation effects sensitive water referenced single-voxel 1H MRS and relaxation effects robust high-spatial-resolution MRSI with fat (yellow bone marrow) as the internal or vegetable oil as the external standard.
Five healthy male volunteers were measured with a mean age of 42.2 ± 11.0 years (range = 28–56 years), and mean body mass index (BMI) of 29.2 ± 5.4 kg/m2. BMI of the three subjects was close to the upper normal value of 25 kg/m2 (24.2, 25.8, and 26.0 kg/m2). Two subjects were overweight (33.0, 36.8 kg/m2). The subjects were informed about the experimental procedures and gave written consent. The study was conducted in accordance with the regulations of the local ethics committee.
All data were acquired on a 1.5 T Gyroscan Intera MR scanner (Philips Medical Systems) using a combination of whole body and knee coils for radiofrequency transmitting and signal receiving, respectively. The subject lay in a supine position with the most extended part of the calf in the center of the coil. The external fat reference was fixed to the calf. The reference consisted of plastic probe (Φ = 13 mm i.d., 110 mm long) filled with vegetable oil. The long axis of the probe was parallel to B0 (Fig. 1). Multislice T1-weighted spin-echo images were first acquired to guide the positioning of the volume of interests for single-voxel MRS and high-spatial-resolution MRSI.
The measurements started with the water-suppressed high-spatial-resolution MRSI sequence (27) in axial slice (Fig. 1). Vegetable oil probe center was in the middle of the measured slice. The slice thickness was 15 mm. Incrementing the echo time of the subsequent 48 images by ΔTE = 2.6 ms led to a spectral bandwidth of 6 ppm and spectral resolution 0.125 ppm. The first image was measured with TE1 = 8 ms. A read gradient of 5.44 mT/m was applied, flip angle 35°. Image matrix 256 × 256, FOV = 150 mm and 128 phase encoding steps led to resolution in the plane of 0.59 × 1.17 mm. The net measurement time was 15 min 22 s (1 acquisition, TR = 150 ms).
High-spatial-resolution MRSI was followed by the single-voxel spectroscopy (PRESS). The voxels were placed within the MRSI slice, that is, the axial PRESS box slice of thickness 15 mm was identical to the spectroscopic imaging slice. Large vascular structures were avoided. Figure 1a shows representative voxel positions. Single-voxel spectra were acquired using the following parameters: TR/TE = 3000/26 ms, 64 accumulations, 1024 points, spectral bandwidth 1000 Hz. The measurement sequence included 16 unsuppressed water reference acquisitions for quantitation. The voxel size was 10 × 10 × 15 mm3. In total, 17 spectra were measured.
Relaxation times T1 and T2 of vegetable oil were measured by single voxel PRESS sequence (TR = 3000 ms). For evaluation of the transverse relaxation time T2, the first echo time of localization sequence was set to 30 ms and increased nine times with the step of 15 ms. Longitudinal relaxation time T1 was evaluated using an inversion recovery technique. The inversion times TI were set to 6, 90, 186, 312, 456, 624, 798, 978, 1158, and 1332 ms.
T map in the measured slice was computed by single-exponential fit of the first 32 gradient echo images acquired in MRSI experiment. T1 and T2 relaxation times of the vegetable oil were obtained by single-exponential fit of the methylene spectral line intensities.
Estimation of total lipid content of single voxel spectra was accomplished by LCModel. The LCModel algorithm was customized by the manufacturer (version 6.1-4F). Concentrations of methylene resonances of extra- (EMCLCH2) and intramyocellular (IMCLCH2) lipids were computed as mM (equivalent to millimoles per liter of muscle tissue), and were corrected for T1, T2 relaxation effects of the water reference using LCModel's control parameter atth2o. This value was determined by the expression exp(−TE/T2)[1−exp(−TR/T1)] (28) assuming relaxation times T1 = 1300 ms, T2 = 28 ms (7). The concentration of lipid molecules (total lipid content) was computed by the summation of EMCLCH2 and IMCLCH2 concentrations and division by factor 31. Value 31 is based on the assumption of Boesch et al. (7) that the average number of methylene protons is 62 per triacylglycerol (TG) molecule (equivalent to 31 CH2 groups). The resulting concentration was then corrected for relaxation effects of methylene lines using the same expression as for water line and assuming T1 = 340 ms, T2 = 85 ms for the EMCLCH2 and IMCLCH2 lines (9). Division by the tissue density (1.05 kg/liter for normal muscle tissue) (24) was performed to convert mM to millimoles per kg wet weight.
High-spatial-resolution MRSI raw data were processed on a Linux PC. The processing algorithms were developed in house and have been described elsewhere (11, 27). The resulting spectra were computed by the summation of the hundreds of magnitude spectra. The VOI was defined by the mask matrix (11) that represents voxel size and position in axial slice (Fig. 1a). Mask matrices were placed manually. We avoided summation of real spectra because large quantities (102) of the high-spatial-resolution MRSI spectra cannot be reliably phase corrected before summation. The magnitude spectra were processed by advanced method for accurate, robust and efficient spectral fitting AMARES (29) in the Java-based magnetic resonance user interface (MRUI) (30). Prior knowledge was described elsewhere (11). The lipid content in volume % (LV) was computed using EMCLCH2 intensity of the voxels with a 100% fat content. The spectrum of yellow bone marrow (11) and spectrum of vegetable oil were used as the internal and external reference, respectively. To convert concentration based on volume % to mmol/kg wet weight, a specific density of muscle tissue 1.05 kg/liter, average density of lipids 0.918 kg/liter, and average molecular mass of TG molecule 0.858 kg/mol was assumed (7). It can be shown that:
where LC is the lipid content in mmol/kg wet weight, and LTV = 100 − LV represents lean tissue content in volume %. No relaxation corrections were performed for yellow bone marrow referenced quantitation because its relaxation times T1, T2 are almost identical to EMCL and IMCL (1, 8, 9). However, when vegetable oil was used as the external concentration reference, the relaxation time differences between bone marrow (T1BM,T) and vegetable oil (T1OIL,T) were taken into account. The correction factor was derived using the well-known equation for the steady state signal intensity (SI) of radiofrequency spoiled fast gradient echo sequence (T2*<TR<T1) (31, 32):
where C is the hardware related constant, ρ is the spin density, α is the flip angle and E1 = exp(−TR/T1). It can be shown that the spectral intensity of vegetable oil (EMCLCH2) has to be multiplied by the correction factor CF:
where E1BM = exp(−TR/T1BM), E1OIL = exp(−TR/T1OIL), and TE1 is the echo time of the first image record, that is, time of the first point of the k-FID (33). The assumption of equivalency between yellow bone marrow and vegetable oil spin density (ρBM=ρOIL) was also considered.
Standard correlation and linear regression was performed to evaluate the relationship between lipid concentrations estimated by different methods. P < 0.05 was considered statistically significant. Results are reported as mean ± SD (standard deviation).
Spin echo image of the right calf (TR/TE = 300/10 ms, slice thickness 15 mm) is shown in Figure 1a. White squares show representative voxel positions in different muscles. The arrow shows voxel in soleus muscle whose spectrum was used in Fig. 3 and 4. Figure 1b shows T map and mean T values in the regions of interest (ROI) depicted by circles. The extent of ROIs was either 73 or 27 mm2 and included 213 or 79 pixels, respectively. T value of vegetable oil (48 ms) was higher than T of bone marrow (40 ms). T values of the subcutaneous fat in selected ROI (small circles) did not significantly differ from the bone marrow. Note that T values of subcutaneous fat are influenced by bulk magnetic susceptibility effects on the air–soft tissue interfaces. T values of the water in the muscles (∼24 ms) are almost constant and close to T2 (28 ms; 7). We consider T values of the muscles reliable, although water suppression by factor ∼5. T1 and T2 value of vegetable oil measured by single-voxel (PRESS) technique was 225 and 55 ms, respectively.
High-spatial-resolution MRSI reference fat spectra of the vegetable oil and bone marrow are shown in Figure 2a and 2b, respectively. Spectrum of the vegetable oil and bone marrow was computed by summation of 144 and 564 voxel spectra corresponding to a VOI 0.7 and 2.9 cm3, respectively. VOIs indicate gray pixels in the probe with vegetable oil and tibial bone marrow (Fig. 1a). From Figure 2a,b it can be seen that tibial bone marrow contains only lipids.
Figure 3a shows the spectrum of the soleus muscle measured by single-voxel spectroscopy (SVS). VOI is shown by the arrow in Figure 1a. Correspondent high-spatial-resolution MRSI spectrum is shown in Figure 4a. Single-voxel spectrum was processed by LCModel (Fig. 3b–d), whereas MRSI spectrum was processed by AMARES (MRUI; Fig. 4b–d). Both EMCLCH2 at 1.5 ppm and IMCLCH2 at 1.3 ppm were evaluated. Figure 5a shows the correlation between lipid content measured by water referenced SVS and high-resolution MRSI with fat (bone marrow) as the internal standard. Correlation between lipid content estimated using internal (bone marrow) and external (vegetable oil) fat standard is shown in Figure 5b. The relationships between IMCL concentrations are summarized in Figure 6. Concentrations of 11 spectra were considered. IMCLs were undetectable in two voxels. In four voxels, the IMCL concentrations were unreliable as indicated by Cramér-Rao lower bounds (CRLB > 20 %).
The purpose of the study was to compare single-voxel magnetic resonance spectroscopy with high-resolution spectroscopic imaging for the assessment of lipid content in human skeletal muscle. The main task of the experiments was, therefore, measurement of the proton spectra that cover a broad range of lipid concentrations in skeletal muscle. Experiments with normal (BMI = 24.2 kg/m2), slight (26.0 kg/m2), and overweight (36.8 kg/m2) volunteers were, therefore, performed.
LCModel and MRUI software packages were used for the assessment of the total fat content in calf muscles. MRUI has the advantage of being free of charge to nonprofit organizations. This software tool analyzes spectra in the time domain using a priori information that can be introduced flexibly. Preparation of the prior knowledge is however laborious. MRUI offers only Lorentzian or Gaussian lineshapes for a given peak. Oscillations of the residue around EMCLCH2 and IMCLCH2 line positions (Figs. 2d, 4d) are a consequence of such restrictions. A more appropriate Voigt lineshape (34) has not yet been implemented. AMARES (MRUI) enables input of prior knowledge through a so-called singlet approach and the possibility of applying soft constrains (imposing lower and upper bounds) on the freely estimated parameters.
The LCModel fits real spectra in the frequency domain using a basis set of spectra of in vitro metabolite solutions. Exceptions are lipids that are fitted using a simulated basis set. The regularization procedure allows considerable flexibility in the estimation of baseline and lineshape. The “black box” approach inherent to the LCModel reduces user interaction and individual bias to a minimum.
Figure 5a shows the correlation of total lipid concentrations estimated by SVS and high-spatial-resolution MRSI. In total, 17 muscle spectra were evaluated. Linear regression (solid line) reveals a correlation coefficient of 0.979, with a highly significant slope (P < 0.0001) close to 1 and low intercept. Somewhat larger dispersion of the concentrations from identity line (dashed line), particularly in the higher concentration range were probably caused by errors in manual positioning of VOIs (±1 pixel) in processing of MRSI data. Correlation between concentrations determined by SVS and high-spatial-resolution MRSI is appealing taking into account different experimental approaches, spectrum processing and quantitation methods. Deviations can be explained by differences in prior knowledge, different baseline correction methods and by limited precision of the relaxation times T1 and T2. Quantitation complicates short relaxation time T2 (∼ 28 ms) of the reference water line and heterogeneity of the water distribution in the muscle tissue. A non-negligible problem is the relatively wide range of T2 values of the water (25 < T2 < 35 ms) that depend on the age, type, and distribution of muscle fibers (25, 26). These difficulties can be avoided by high-resolution MRSI which offers unique possibilities for VOI definition. VOI can be noncontinuous and irregularly shaped. This feature enables VOI selection with 100% fat content. Tibial yellow bone marrow was used in this study (11). The following arguments support this choice: (i) no trabecular bone is in the center part of tibialis; (ii) tibialis contains only yellow bone marrow with 100% of fat; (iii) relaxation times T1, T2 of the yellow bone marrow and muscular fat (EMCL, IMCL) are almost identical (1, 8, 9); and (iv) the bulk magnetic susceptibility effects between the bone and bone marrow are small because tibialis is parallel to B0 and resembles an “infinite” long cylinder (35). Because our MRSI sequence uses gradient echoes (27), differences in T between reference bone marrow and muscle lipids can influence quantitation. From the T map (Fig. 1b), it follows that T differences of water in the measured slice are insignificant because of the small voxel size (5.2 mm3) and good magnetic field homogeneity (< 0.3 ppm) in the slice. Differences in T values of muscle lipids should, therefore, be insignificant as well. Exceptions are relatively thin layers of subcutaneous fat influenced by bulk magnetic susceptibility effect on the air tissue interface. Muscle lipid concentrations obtained by MRSI with the bone marrow as the internal standard are most likely independent of relaxation effects and can be used as the “gold standard.”
Figure 6a shows the relationship between IMCL concentrations estimated by SVS and high-spatial-resolution MRSI. Correlation is good, however, somewhat inferior when compared with the total lipid content shown in Figure 5a. As in the case of total lipids the deviations could be explained by differences in prior knowledge, baseline corrections and by limited precision of the relaxation times T1, T2. In addition, the correlation was impaired by decreased accuracy in IMCL estimations because of lower concentrations and difficulties in separation of IMCLCH2 spectral line from the dominant EMCLCH2 peak. Figures 5b and 6b reveal very good correlations between lipid contents evaluated using bone marrow and vegetable oil reference. From these relationships, it follows that it is possible to use external fat standard (e.g., vegetable oil) in MRSI quantitation. However, differences in relaxation times between muscle lipids and external fat standard (Eq. ) have to be taken into account.
In this work, two approaches for quantitation of lipids in skeletal muscle have been compared: standard water referenced single-voxel MRS and high-spatial-resolution MRSI with fat (yellow bone marrow) as the internal standard. Spectrum processing was performed using LCModel and AMARES. From comparison of the water and fat referenced quantitations, it follows that the single-voxel MRS approach requires relaxation corrections of the reference water line and methylene fat line. It was demonstrated that the high-spatial-resolution MRSI approach with the internal fat reference circumvents the problem of relaxation corrections and enables using vegetable oil as the external fat standard. High-resolution MRSI is, therefore, a promising alternative in quantitation of lipids in skeletal muscles.
The authors thank S. Provencher for providing the customized LCModel and for assistance with the spectrum analysis. The MRUI software package was kindly provided by the participants of the EU founded network programmes “Human Capital and Mobility”, CHRX-CT94-0432 and “Training and Mobility of Researches”, ERB-FMRX-CT970160.