Distribution of intramyocellular lipids in human calf muscles as determined by MR spectroscopic imaging

Authors


Abstract

In this study the distribution of intramyocellular lipids (IMCL) in human calf muscles was determined by 1H-MR spectroscopic imaging (MRSI) measurements. An obstacle for MRSI measurements in the calf, including different muscles, is the inevitable inclusion of regions with high concentrations of extramyocellular lipids (EMCL). This can lead to signal bleeding and consequently to unpredictable overlaps of IMCL resonances with EMCL in voxels of interest. The results of this study show that signal bleeding from EMCL can be substantially reduced in voxels from calf muscles by the application of a lipid extrapolation (LE) procedure (Haupt et al., Magn Reson Med 1996;35:678). The spectra of all voxels located within muscle tissue were fitted, and the metabolite values were assigned to one of 10 different muscles based on image segmentation. Significant IMCL differences between some muscles were obtained, with high values in m. soleus and two to three times lower values in the tibialis anterior, tibialis posterior, and gastrocnemius muscles. In addition to gross differences between muscles, significant intersubject differences were observed in both IMCL content and distribution over different muscles. A significant correlation between fiber orientation (obtained from orientation-dependent dipolar coupling of creatine and taurine resonances) and IMCL content was found, indicating that IMCL content is directly correlated to biomechanical properties. Magn Reson Med 51:253–262, 2004. © 2004 Wiley-Liss, Inc.

Recently, the measurement of intramyocellular lipids (IMCL) by 1H MR spectroscopy (MRS) has become of interest in the study of human skeletal muscle tissue (1–7), since it has been demonstrated that IMCL is negatively correlated with insulin sensitivity in sedentary and diabetic volunteers (3, 4), and also serves as an important energy supply (Refs. 7 and8, and references therein). However, although several studies have investigated IMCL using 1H MRS, all of these studies used either single-voxel spectroscopy or analyzed a few selected voxels from MR spectroscopic imaging (MRSI) data (5, 6). The lack of comprehensive MRSI studies is mainly due to strong lipid signals from surrounding subcutaneous or interstitial fat and bone marrow (i.e., extramyocellular lipids (EMCL)), which lead to contamination of the spectra of interest due to signal bleeding at the low MRSI resolution. This primarily affects the detection of IMCL. The bulk magnetic susceptibility (BMS) shift leads to a separation of IMCL and EMCL resonances. The frequency shift amounts to 0.25 ppm when the lipid layer is parallel to B0. However, because of the anisotropic structure of EMCL, the EMCL resonances shift and broaden depending on the orientation of the lipid layer relative to B0 (2). Therefore, signal bleeding originating from different EMCL layers outside the voxel can lead to unpredictable overlaps with IMCL resonances in the voxel of interest. Strong spatial apodization can reduce the problem of signal bleeding, at the expense of much greater voxel sizes. However, this in turn may lead to increased EMCL contributions from inside the voxel and thus complicate IMCL estimation. In MRSI studies of the brain, it has been shown that contamination caused by subcutaneous lipids can be significantly reduced by employing data extrapolation (9). It may therefore also be possible to reduce EMCL contributions from signal bleeding in muscle spectra, without applying strong spatial apodization.

IMCL evaluation is hampered also by the spectral signal overlap with EMCL from inside the voxel. It has been shown in single-voxel studies, and for selected voxels from MRSI studies, that sophisticated fitting strategies are necessary to reliably estimate IMCL beside overlapping EMCL signal contributions (2, 5, 6). However, no automated fitting strategies have been presented for full MRSI data sets, which in addition take into account residual EMCL contamination due to signal bleeding.

Strong IMCL differences between different muscles have been detected in studies employing single-voxel MRS at different locations, and in one study analyzing selected voxels from different locations of an MRSI data set (2, 5, 10). Furthermore, it has been shown that endurance exercise can reduce IMCL content (2, 6, 11). It is possible that IMCL usage is specific to certain muscle groups depending on the type of exercise involved.

Previously, we showed that 1H-MRSI methods can be used in human calf muscles to measure fiber orientation (12), using the orientation-dependent dipolar splittings of creatine/phosphocreatine (Cr) and taurine (Tau) resonances (13). In addition, the distribution of Cr and trimethyl-ammonium groups (TMA) over different muscles was determined. However, IMCL content and distribution were not determined in that study.

The overall goal of this study was to determine the spatial distribution of IMCL in different muscles by proton MRSI. Our specific goals were to 1) address the problem of signal bleeding by employing a lipid extrapolation (LE) procedure; 2) employ sophisticated fitting procedures that allow the incorporation of prior knowledge; and 3) segment the muscle and analyze the content of IMCL and other metabolites, as well as the fiber orientation in different muscles.

MATERIALS AND METHODS

MRSI

Thirty measurements on the calves of 10 healthy subjects (eight men and two women; mean age = 29 ± 7 years, range = 20–42) were carried out. The subjects' degree of fitness ranged from untrained to medium well trained; no completely sedentary nor highly trained subjects were included in this study. The subjects were measured at different stages of diet and exercise. The measurements were performed on a 1.5 T system (SIGNA; General Electric, Milwaukee, WI) using a standard linearly polarized transmit/receive coil (20 cm diameter) for extremities. Included in these measurements was a subset of nine measurements with previously published results on muscle fiber orientation by MRSI (12). The subjects were in a supine position and the long axis of the leg was placed parallel to the magnetic field, with the center of the coil approximately 10 cm inferior to the proximal end of the tibia. The coil was fixed to the patient table always at the same position, to ensure similar parallel positioning of all subjects. 2D MR images in sagittal and transverse orientations were acquired to localize an MRSI plane at similar inferior–superior positions. Imaging parameters were chosen to visualize fasciae separating different muscles (gradient-echo sequence with flip angle = 30°, TR = 51 ms, and TE = 5 ms).

For spectroscopic imaging, a 2D MRSI sequence with point-resolved spectroscopy (PRESS) volume preselection was used. The slice (15 mm thick) was selected in the transverse orientation. The position and size of the PRESS box was adjusted to partly exclude subcutaneous fat without excluding muscle tissue (Fig. 1). Two slightly different measurement parameters were used: The spectra were acquired with TR = 1600 or 1200 ms, TE = 35 ms, water presaturation, and outer volume suppression (OVS). A matrix of 36 × 36 was used over an FOV of 22 or 20 cm, resulting in a scan time of 35 or 26 min, respectively. In addition to OVS, oblique saturation bands were used for 17 measurements. They were adjusted to saturate signals from bone (tibia and fibula) and subcutaneous fat (Fig. 1).

Figure 1.

MR image of the calf of a volunteer. A sketch of a typical PRESS box location with OVS is overlaid. The typical position of oblique saturation bands that were applied additionally in a subset of measurements (17 of 30) is also indicated.

The study was approved by the local ethics committee, and informed consent was obtained from all of the subjects before the MRS studies were conducted.

Data Processing

Postprocessing included spatial zero-filling to 64 × 64, and moderate Gaussian spatial apodization (weight equal to 0.37 at the diameter of a circle inscribed in the k-space square) after LE. The LE procedure was performed as described in Ref. 9. The procedure makes use of the fact that the main EMCL signals are localized to narrow, predefinable regions. These strongly band-limited EMCL areas were extrapolated to higher k-space values. The actually measured data were kept unchanged. For the LE procedure, a lipid mask defining the regions of EMCL from bulk lipids was created, using either an additionally acquired and coregistered fluid-attenuated inversion recovery (FLAIR) MR image (TR/TE/TI = 3500/69/730 ms) or a lipid image produced directly from the zero-filled MRSI data by integrating the spectra over the lipid region. However, since the two procedures performed equally well, only the results employing the latter method are presented. A few MRSI data sets were also processed without LE and with different spatial Gaussian apodizations (weight equal to 0.37 and 0.05 at the diameter of the k-space data—denoted as GW 0.37 and GW 0.05, respectively) for comparison purposes and in order to determine the performance of the LE procedure. In the time domain, a fast Fourier transform (FFT) was performed without additional postprocessing (i.e., no zero-filling, apodization, or residual H2O elimination was applied). Phantom measurements were performed to map the excitation profile of the PRESS box. Within the box the intensity profile was relatively uniform and dropped only close to the edges. The signal intensities were >90% of maximum intensity in approximately two-thirds of the voxels within the prescribed area. Therefore, no correction was applied. However, we analyzed only spectra from voxels that were (based on this criterion) sufficiently distant from the edges of the PRESS box.

The spectra were fitted using an iterative nonlinear least-squares fitting algorithm (time domain fitting of frequency domain spectra (TDFDFIT) (14)), which allows a mixed lineshape fit and incorporation of prior knowledge. Prior knowledge was used for EMCL and IMCL. The lineshape model for IMCL and EMCL consisted of a pattern of six lines each, plus a single line for the methylene resonance (Fig. 2a). The lineshape model was obtained in a first step by fitting a series of spectra that displayed only EMCL contributions (spectra from regions containing predominantly subcutaneous or interstitial fat) by using Voigt lines without prior knowledge. The same total lipid lineshape model was used for both IMCL and EMCL. The frequency, phase, and linewidth of the IMCL lineshape (pattern and methylene resonance) were calculated via Cr, which was fitted prior to lipid fitting (see below). IMCL phase and linewidth were then fixed, and constraints were used for the frequency (–3 Hz and +4 Hz). The areas of the IMCL pattern and methylene resonances were fitted independently and without constraints. For EMCL, the phase, linewidth, and frequency of the methylene resonance and the pattern were fixed to each other. The total EMCL lineshape (i.e., methylene resonance + pattern) was fitted with wide constraints; only the frequency was constrained to not overlap with IMCL. To account in addition for residual EMCL contributions from signal bleeding, and for asymmetric EMCL lines (15), EMCL signal contributions were fitted with three total lineshapes, including asymmetric lines (Fig. 2b and c). The asymmetric lineshapes were created empirically by convolving the lineshape pattern with a Gaussian distribution of EMCL orientations. Cr, TMA, and Tau were fitted prior to lipid fitting, excluding the lipid region. Cr and Tau are subject to dipolar coupling (13), which results in a dependence of the spectra on the orientation of muscle fibers with respect to B0. The dipolar coupling was taken into account for fitting of these metabolites, as described previously for a subset of the measurements (12). The fit yielded the degree of splitting due to dipolar coupling, which can be directly converted into fiber orientation. In addition to the metabolite intensities, the fitting of the dipolar splittings allowed us to determine muscle fiber orientation in different muscles.

Figure 2.

Lineshape models used to fit IMCL (a) and EMCL (a–c). Each lineshape consisted of a pattern of six lines each, plus a single line for the methylene resonance. Image a shows the pattern and the methylene resonance separated (top spectrum, with the pattern enlarged), and the sum of both (bottom spectra). IMCL was fitted with the symmetrical lineshape shown in a, fitting the pattern and methylene area independently. EMCL was also fitted with the lineshape shown in a, and the two asymmetrical lineshape models shown in b and c (see text for more details).

All of the spectra were fitted with identical starting values. Spectra from outside the calf and from regions including bone or subcutaneous fat were not fitted. On average, 425 voxels were fitted for each subject. The fitting of the MRSI data was performed by a completely automated procedure. Spectra with fit residues greater than twice the average residue of all spectra from this data set (χ2 > 2 × mean(χ2)) were excluded from further analysis. In addition, spectra were automatically excluded if the Cr linewidth was >8 Hz.

Muscle Segmentation

The calf was segmented manually into 10 muscles using a coregistered gradient-echo MRI (for more details see Ref. 12), to obtain IMCL, Cr, TMA, and fiber orientation in different muscles. The metabolites of each voxel of the zero-filled MRSI data were assigned to its originating muscle based on the segmented images. The evaluated muscles included the m. tibialis anterior (TA), m. tibialis posterior (TP), m. soleus (which was divided into a medial (SM) and a lateral (SL) part), m. gastrocnemius lateralis (GL) and medialis (GM), m. extensor digitorum (ED), m. peroneus brevis (PB), flexor digitorum longus (FDL), and m. flexor hallucis longus (FHL) (Fig. 3). To ensure that spectra were assigned completely to an individual muscle, a gap of at least one MRSI voxel was left between different muscles. On average, approximately 280 of 425 fitted voxels were segmented for each subject. Muscles that were segmented with less than five MRSI voxels were excluded from the analysis.

Figure 3.

Example showing manual segmentation of the calf of a volunteer overlaid on a coregistered MR image. The segmented muscles include the m. tibialis anterior (TA), m. tibialis posterior (TP), m. soleus medialis (SM) and lateralis (SL), m. gastrocnemius medialis (GM) and lateralis (GL), m. extensor digitorum (ED), m. peroneus brevis (PB), flexor digitorum longus (FDL), and m. flexor hallucis longus (FHL).

Metabolite Analysis

The metabolite contents and fiber orientation values that were assigned to a muscle were averaged in each subject, yielding for each muscle the average and standard error of the mean (SE).

Before intersubject averaging and comparisons can be performed, the metabolite values must be corrected for measurement differences or referenced to another signal experiencing the same variations. Different referencing and correction methods were applied as follows: 1) The values were referenced to the mean Cr in m. soleus medialis (CrSM). Total Cr has been shown to vary to only a minor degree between subjects for identical muscles—at least within relatively homogeneous subject groups, as in the current study (16, 17). However, the reference to CrSM may have added some variances to the findings. 2) To validate the use of CrSM as an internal reference, the reciprocity principle was employed, and the metabolite values were corrected for receiver gain and transmitter value (18). This enabled us to compare absolute metabolite contents between subjects (in arbitrary units). The results, whether obtained via the reciprocity principle or by referencing to CrSM, were almost identical and therefore only the latter are presented in detail. 3) IMCL was referenced to IMCL in m. soleus medialis (IMCLSM) to determine whether IMCL distribution over muscles was similar among subjects despite different total IMCL contents. 4) IMCL was referenced to Cr in each voxel to exclude voxel profile or partial volume effects as reasons for the findings.

The values were corrected for longitudinal relaxation times (T1) using literature values (Cr: T1 = 1100 ms, IMCL: T1 = 300 ms (1)). No correction for transverse relaxation times was performed, because the same TE was used for all measurements.

Statistics

For statistical comparisons of metabolite contents and distribution, and fiber orientation between subjects, one-way analyses of variance (ANOVAs) were performed. For comparisons between different muscles, two-tailed paired t-tests were performed followed by Bonferroni corrections for multiple comparisons. Linear regression analysis was used with partial correlations to control for covariates to compare IMCL content with fiber orientation. A P-value < 0.05 was assumed to be statistically significant.

RESULTS

LE

In order to determine optimal postprocessing parameters for IMCL estimation beside overlapping EMCL signal contributions, several datasets were processed with different Gaussian k-space weighting functions, and with and without LE. The theoretical dispersion of a single voxel with homogeneous lipid content for different Gaussian k-space apodization values is shown in Fig. 4a and b. The contour plot of signal intensity after moderate apodization (GW 0.37, Fig. 4a) shows that the signal drops to a 0.02 level within approximately one nominal voxel. However, there are signal contributions at the 0.02 level distant from the voxel. Since EMCL signals from adipose tissue are about 100 times more intense than IMCL intensities, this leads to substantial contamination. The LE procedure is used to eliminate or reduce these contributions. After stronger spatial apodization (GW 0.05, Fig. 4b), the contour plot does not show signal contributions at the 0.02 level distant from the voxel; however, the voxel size is clearly increased, with the 0.02 contour level spanning approximately four nominal voxels. Accordingly, lipid images of actually acquired MRSI data that were created by spectral integration over the lipid region (0.5–2.5 ppm; Fig. 4c–e) showed substantial ringing after GW 0.37 without LE (Fig. 4c), while it is clearly reduced after LE (using the same Gaussian apodization, Fig. 4d). The lipid image after GW 0.05 and without LE shows no ringing; however, regions with high EMCL are smeared out (Fig. 4e).

Figure 4.

Demonstration of the signal bleeding and blurring of MRSI data that occurs with different spatial apodization values. a and b: Calculated contour plots of the signal intensity dispersion of a single voxel using moderate (weight equal to 0.37 at the diameter of the k-space data, GW 0.37) and strong (GW 0.05) k-space apodization, respectively. The axes denote nominal voxel numbers. c–e: Lipid images of actually acquired MRSI data. In c, GW 0.37 was used; in d, the same apodization was used as in c, but with lipid extrapolation; in d, GW 0.05 was applied. The images are scaled identically.

The effect of LE is also demonstrated in Fig. 5, which shows sample spectra from six different locations obtained with LE (bottom spectra of each comparison) and without LE (top and center spectra). For the top spectra, GW 0.05 was employed, whereas the center and bottom spectra were obtained with GW 0.37. Spectra obtained with GW 0.37 and without LE (center spectra) showed out-of-phase EMCL signal contributions and/or duplicated EMCL signals (i.e., relatively strong contaminations from outside the voxel). In contrast, spectra from the same locations obtained with GW 0.05 (top spectra) showed only a single EMCL signal pattern, and no out-of-phase signal. However, compared to spectra with LE and GW 0.37 (bottom spectra), the EMCL signal contribution is much higher, and thus it is more difficult to distinguish between IMCL and EMCL—especially since close to adipose tissue, bone marrow, and lipid-containing fasciae, the greater voxel size due to stronger apodization results in increased EMCL signals. The spectra with LE and GW 0.37 were much less contaminated with EMCL from outside the voxel than spectra without LE, and EMCL signal contributions were relatively low. However, LE did not completely suppress EMCL signal contributions due to signal bleeding (Fig. 5, spectrum 5). Nevertheless, as the best compromise, we chose moderate apodization (GW 0.37) plus LE. Spectra distant from areas with high EMCL appeared to be relatively unaffected by the chosen processing scheme (Fig. 5, spectrum 6).

Figure 5.

Comparison of spectra after different spectral postprocessing procedures were performed from six positions of the calf, as indicated on the MR image. The top spectrum of each comparison was obtained with strong spatial apodization (GW 0.05). The center and bottom spectra were obtained with moderate apodization (GW 0.37) and with and without lipid extrapolation (with and without LE, respectively). The spectra for each comparison are scaled approximately for identical Cr area.

Spectral Fitting

Fitting of all of the spectra within muscle tissue yielded only small residues for the vast majority of the spectra. Three spectra with corresponding fits and residues are shown in Fig. 6. The figure demonstrates that the fitting also performed well (i.e., yielded only small residues) for spectra with strongly overlapping IMCL and EMCL resonances (Fig. 6b), and for spectra with residual EMCL contamination from signal bleeding (Fig. 6c).

Figure 6.

Experimental (top), fitted (center), and residual (bottom) spectra from three different voxels. The spectra of the three chosen voxels demonstrate (a) good and (b) poor separation between IMCL and EMCL, and (c) residual EMCL contamination from outside the voxel.

IMCL and Fiber Orientation Images

The IMCL distribution over different muscles can be obtained qualitatively by creating images from the fitted IMCL signal area of all voxels. An example of an IMCL image obtained from the fitting of all spectra is shown in Fig. 7a. A contour plot of the coregistered MRI is overlaid for anatomical guidance. The image shows distinct areas of high and low IMCL, with high signal contributions in m. soleus and much lower intensities in m. tibialis anterior and m. gastrocnemius. Areas of high and low intensities are confined by fasciae separating different muscles. Similarly, fiber orientation images were created after the fitted dipolar splittings of the Cr and Tau resonances were converted into angles (Fig. 7b).

Figure 7.

a: Fitted IMCL image from one subject. A contour plot of the coregistered MR image is overlaid for anatomical guidance. b: Fiber orientation image. Large angles between B0 and fiber orientation are displayed by convention in bright shades.

Metabolite Distribution and Fiber Orientation

To determine the consistency of IMCL determination in different muscles, intrasubject SEs were calculated for each muscle from all allocated IMCL values. The average SEs for all muscles are given in Table 1. The average SE was below 10% for the main muscles of the calf (TA, TP, SL, SM, GL, and GM), except for GL.

Table 1. MRSI Results for IMCL and Fiber Orientation in Different Muscles
 TATPSLSMGLGMEDPBFDLFHL
  1. IMCL SE, averaged intrasubject standard error of the mean (in %) of all IMCL values assigned to a certain muscle. IMCL/CrSM, mean values (±1 SD) of IMCL content (relative to Cr in m. soleus medialis, SM); IMCL/IMCLSM, mean values (±1 SD) of IMCL distribution (relative to IMCL in SM); IMCL/Cr, mean values (±1 SD) of IMCL referenced to Cr in each voxel; Fiber orientation, (obtained from the orientation dependent dipolar splitting of Cr and Tau resonances) in different muscles of the calf. See Fig. 3 for muscle abbreviations.

Mean individual IMCL SE (%)7.39.86.04.011.88.014.516.022.713.6
IMCL/CrSM2.2 ± 0.92.6 ± 0.85.5 ± 1.77.2 ± 2.72.1 ± 1.12.4 ± 1.12.9 ± 1.23.1 ± 1.62.9 ± 0.94.3 ± 1.3
IMCL/IMCLSM0.31 ± 0.110.38 ± 0.90.79 ± 0.191.00 (Ref)0.29 ± 0.110.34 ± 0.140.40 ± 0.140.43 ± 0.170.45 ± 0.90.62 ± 0.13
IMCL/Cr4.3 ± 1.63.9 ± 1.26.8 ± 1.77.3 ± 2.73.3 ± 1.63.6 ± 1.55.1 ± 2.24.9 ± 2.65.3 ± 2.15.7 ± 0.9
Fiber orientation (°)9 ± 427 ± 538 ± 545 ± 322 ± 424 ± 515 ± 723 ± 423 ± 432 ± 6

The average IMCL contents relative to CrSM (IMCL/CrSM) in different muscles are listed in Table 1. The highest IMCL was found in SM; it was lower in SL, and it was lower by about a factor of 3 compared to SM in TA, TP, or gastrocnemius. A relatively high intersubject standard deviation (SD) for IMCL contents (40% on average) was found for all muscles. Statistical analysis by ANOVA (using the IMCL contents of all voxels assigned to a specific muscle as the dependent variable) showed that these intersubject IMCL differences were significant for all muscles except FDL.

To determine whether the IMCL distribution over different muscles was similar, despite different total IMCL contents among subjects, the IMCL of each muscle was normalized to IMCLSM (IMCL/IMCLSM). The individual and average (±SD) relative IMCL levels in 10 muscles are shown in Fig. 8 and Table 1. While the mean values duplicate the results on IMCL contents, demonstrating similar IMCL distribution over different muscles for all subjects, the SD here denotes intersubject differences in IMCL distribution rather than absolute content. However, although the average intersubject SD over all muscles of relative IMCL was smaller than for IMCL/CrSM (28%), it was still considerably high. Similar to IMCL/CrSM, intersubject differences in IMCL/IMCLSM were significant for all muscles except for FDL. Table 2 shows the results of statistical comparisons of IMCL content between different muscles. Despite the relatively high intersubject SD and the use of conservative Bonferroni corrections for multiple comparisons, significant IMCL differences were detected between a number of muscles, especially between SL and SM within m. soleus, and between m. soleus (SL and SM) and most other muscles.

Figure 8.

Relative IMCL content (relative to IMCL in SM) in different muscles of the calf from 30 measurements. The individual relative IMCL content is shown, as well as the average ± 1 SD. For some muscles, less than 30 values were obtained, because 1) values for a muscle were excluded from the analysis if less than five voxels contributed to the individual mean (affecting mainly smaller muscles, such as FDL and FHL), and 2) some muscles were saturated in a subset of measurements, which used oblique saturation bands (affecting TP, FDL, and FHL).

Table 2. Statistical Comparison of IMCL Content (Relative to CrSM, Upper Right Triangle) and Fiber Orientation (Lower Left Triangle) Between Different Muscles of the Calf
MuscleTATPSLSMGLGMEDPBFDLFHL 
  1. Differences between muscles were determined with paired t-tests with Bonferroni corrections for multiple comparisons. Comparisons between muscles with P-values less than 0.05 are denoted with “x”. See Fig. 3 for muscle abbreviations.

TA xxxIMCL
TPx xx
SLxx xxxxx
SMxxx xxxxx
GLxxx x
GMxxx x
EDxxxxx 
PBxxxx 
FDLxx 
FHLxx 
Fiber orientation

To exclude artifacts (e.g., voxel profiles or partial volume effects) as the reason for these findings, the analysis was repeated after IMCL was normalized to Cr for each voxel, and an average IMCL/Cr ratio was calculated for each segmented muscle (Table 1). The results showed essentially the same IMCL distribution, although the differences between muscles were smaller.

Fiber orientation was determined by fitting the dipolar coupling of Cr and Tau resonances, as described previously (12) (Table 1). For TA, a fiber orientation almost parallel to the external field was found, whereas in m. soleus the fibers were angulated by 38° ± 5°, and by 45° ± 3° for SL and SM, respectively, with respect to the external field. A statistical analysis showed that the fiber orientations were significantly different in the majority of muscles (Table 2).

A highly significant linear relationship between fiber orientation and relative IMCL content in different muscles was found by linear regression analysis, controlling for muscle as the covariate (Fig. 9). Furthermore, a linear regression between IMCL and fiber orientation for individual muscles revealed a positive slope in nine of 10 muscles, although the correlation was significant only for SL (P < 0.0001).

Figure 9.

Correlation between relative IMCL content and fiber orientation. Different symbols denote different muscles.

DISCUSSION

The present work demonstrates that 1H MRSI in human muscle is feasible and yields reliable results in different muscles for IMCL levels besides other metabolites. The application of an LE procedure substantially reduced EMCL contamination from outside the voxel. IMCL was highest in the soleus muscle, while it was low in the tibialis anterior, posterior, and gastrocnemius muscles, confirming prior studies on selected voxels in a few muscles (2, 5, 10). In addition to these gross differences between some muscles, which were similar for all subjects, significant intersubject differences were observed for both IMCL content and distribution.

IMCL Estimation Procedure

The acquisition of in vivo MRSI spectra from human muscles, including different muscle types (i.e., covering large areas or a complete cross section) is hampered by the almost inevitable inclusion of areas with high concentrations of EMCL from subcutaneous or interstitial fat and bone marrow. EMCL concentrations from these areas are about two orders of magnitude greater than the concentration of IMCL. Thus, signal bleeding from EMCL can lead to contamination of the spectra of interest. A further complication arises in the BMS shift of EMCL, in addition to potential spatial B0 shifts (due to B0 field distortions), leading to unpredictable frequency shifts of the signal contribution from outside the voxel. While it still may be possible to manually select and analyze spectra without or with only moderate (visible) contamination, this signal bleeding complicates reliable fitting of IMCL signals of all spectra in muscle tissue, or may even render it impossible. As in previous studies in the brain (9), this study shows that the use of an LE procedure in muscle, with only moderate spatial apodization, significantly reduces contamination from outside the voxel while maintaining relatively small voxel sizes. In all data sets, a clear improvement in comparison with results from simple spatial apodization were obtained. The alternative approach of applying strong Gaussian apodization without LE results in increased voxel sizes and hence stronger EMCL signals in voxels relatively close to adipose tissue, bone marrow, or lipid-containing fasciae. Despite the improvements, the LE procedure did not remove EMCL contamination completely, and residual EMCL contamination was present in parts of the dataset. This is to some extent simply because not all lipid layers (especially thin layers) could be included in the lipid mask. Furthermore, bleeding from regions that were included in the lipid mask was not completely removed by the LE procedure. It is therefore difficult to determine an ideal processing scheme for a full MRSI data set. The optimum strategy depends on the amount of fat tissue present, and possibly geometrical structures. As the best compromise, we chose moderate apodization plus LE for all data sets. To further suppress strong lipid signals from bone (tibia and fibula) and subcutaneous fat, additional saturation bands were used for a subset of measurements in addition to retrospective LE (Fig. 1). However, in all data sets after LE, an overall spectral quality was obtained that allowed reliable spectral fitting of IMCL in the majority of voxels (see below). Alternative EMCL suppression procedures (e.g., lipid nulling by inversion recovery techniques) are not suitable because EMCL and IMCL have similar relaxation properties.

For spectral fitting, EMCL was fitted with three lineshape models to account for residual EMCL contamination. The primary objective was a reliable fitting of IMCL. EMCL contributions were not further analyzed because of residual contamination from outside the voxel, and because EMCL contributions vary strongly depending on exact voxel position (2). The fitting revealed areas of relatively uniform IMCL contents: in general, strong changes in IMCL contributions from one voxel to the next were observed only at fasciae. Since signal bleeding does not depend on muscle boundaries, and its contributions are expected to change more rapidly, this confirms that the fitting revealed real IMCL differences and not artifactual differences due to EMCL signal bleeding. EMCL originating from inside the voxel (and which is therefore not removed by LE) overlaps strongly with IMCL, depending on the angle between the lipid layer and B0 (for example, the soleus muscle generally shows a stronger overlap between EMCL and IMCL than TA). To minimize the possibility of incorrectly attributing signal contributions from EMCL to IMCL depending on the BMS shift, tight-fitting constraints were employed and some parameters were fixed (e.g., the IMCL linewidth was fixed to the previously-fitted Cr linewidth).

In this study, no absolute metabolite values in millimoles were calculated. In principle, the use of an internal standard (H2O or lipid signal from bone marrow) acquired in an additional scan, or assumptions about Cr as the internal standard (i.e., concentration, T1, and T2), would allow IMCL concentration to be estimated in millimoles (19). However, the main goal of this methodological study was to determine whether IMCL distribution over different muscles could be assessed by MRSI. Therefore, only relative IMCL contents were reported.

IMCL Distribution

The highest level of IMCL was found in m. soleus, with slightly lower values in SL than in SM. Other muscles (e.g. TA, TP, GM, and GL) showed about two to three times lower IMCL contents. The detection of higher IMCL signals in m. soleus compared to other muscles is in agreement with previous studies (2, 5, 10). However, in our study, IMCL levels in m. gastrocnemius were lower (compared to m. soleus) than previously observed (10). IMCL in TA was three times lower than in SM. This is in excellent agreement with a recent study at 4T on selected voxels (5).

Instead of analyzing manually selected voxels, we analyzed all voxels located completely within a muscle (∼280 of 425 fitted voxels per subject); therefore, voxels with remaining EMCL contamination and voxels with poor EMCL–IMCL differentiation were included. In view of this, the intraindividual SEs in muscles were relatively small, especially for the main muscles of the calf (TA, TP, SL, SM, GL, and GM; Table 1). This is particularly important for clinical studies, to prevent any potential bias due to manual interference (e.g., manual voxel selection). The low intraindividual SEs also suggest that relevant differences in longitudinal studies, such as before and after exercise or diet, can be detected by 1H-MRSI.

In addition to gross differences in IMCL content between some muscles, significant intersubject differences were detected for almost all muscles, demonstrating a large range of individual IMCL contents, as reported in previous 1H-MRS studies (2, 5). This can be explained by differences in physical activity, diet, or insulin sensitivity among subjects. Furthermore, in addition to interindividual differences in IMCL contents, the IMCL distribution over muscles was found to be significantly different among subjects. This is in line with a previous MRS study (20) in which differences in IMCL distribution in swimmers and orientiees between m. rectus and TA were observed.

Muscle biopsy studies have shown that IMCL concentration is correlated with muscle fiber type composition. Type I fibers (slow twitch) contain three times as much IMCL as type II fibers (fast twitch) (21, 22). M. soleus contains a high percentage (about 88%) of type I fibers (23). In contrast, GL, GM, and ED have lower percentages (about 50%) of type I fibers (23). TA falls in between, being composed of approximately 70% type I fibers. Therefore, qualitatively, the results of our study are also in agreement with these morphometry studies. However, quantitatively, the results are not in agreement, and other factors besides fiber-type composition may play a role for IMCL differences between muscles. The difference in the percentage of fiber type I between m. soleus and “faster” muscles (e.g., tibialis anterior and gastrocnemius) is only 20–40%. Thus, even when size differences between fiber types (24) are accounted for, threefold (or even more) higher IMCL levels in type I fibers can only partially account for the differences between muscles found in this and other MRS studies.

Fiber Orientation and IMCL Distribution

The fiber orientations in different muscles obtained in this study from 30 measurements are very close to previously published values from a subset of nine measurements, which were included in the present study (12).

Although there is some controversy as to whether structural parameters, such as the pennation angle, have functional effects (e.g., maximal shortening velocity and maximal force) (25), it has been suggested that biochemical effects, which determine muscle shortening velocity, are modulated by muscle architecture (26, 27), and that muscles that are recruited during low-velocity contractions have large pennation angles (28). Furthermore, several studies have demonstrated that the pennation angle increases in response to resistance training and muscle hypertrophy (29, 30), and decreases with disuse muscle atrophy (31). Smaller pennation angles were observed in different muscles of the leg in sprinters compared to distance runners (32). The architectural parameters appear not to be directly related to fiber composition (25, 33).

Thus, the finding of this study that fiber orientation and IMCL distribution are significantly correlated suggests that IMCL concentrations are high in muscles with lower shortening velocity. IMCL levels thus appear to be directly correlated with biomechanical properties (although this relationship does not necessarily indicate a causative relationship) and not only to fiber-type composition. This is in line with the observation that IMCL concentration is much higher in endurance-trained subjects than in sedentary subjects (Ref. 11, and references therein). This also cannot be explained by different fiber-type composition alone, and indicates that muscles recruited during low-velocity contractions have high IMCL levels. Alternatively, it is possible that the detected correlation between IMCL distribution and fiber orientation is coincidental or artificial, and is the result of IMCL overestimation in angulated muscles due to stronger EMCL and IMCL signal overlap. However, the selected fitting strategy tried to minimize this effect. Artificial group effects due to repeated measures of muscles with different IMCL levels and fiber orientation were controlled for (with muscle as the covariate) and thus can be excluded. In conclusion, the difference in IMCL levels observed among different muscles can not be explained by different fiber-type distribution alone, but appears to also be correlated with different muscle tasks that require different levels of force and velocity.

Acknowledgements

The authors thank H. Hoppeler for very helpful discussions, and A.A. Maudsley for providing the software for LE.

Ancillary