|Home | About | Journals | Submit | Contact Us | Français|
To quantify and examine the distribution of brain metabolites in normal young adults using single voxel MR spectroscopy at 3 Tesla (T).
Short-echo time single-voxel PRESS technique was used to measure the apparent concentration of five metabolites at nine locations in the brains of young adults. Concentrations were estimated by means of an automated fitting method (LCModel) with reference to an unsuppressed water signal and were corrected for T1 relaxation, T2 relaxation, and cerebrospinal fluid partial volume. Analysis of variance with Tukey post hoc test was used to evaluate regional variations.
Statistically significant differences in regional concentrations were detected for each of the metabolites. The number of significant differences was greatest for total choline, whereas myo-inositol and the sum of glutamine and glutamate had the fewest. Magnitude of variation was greatest for total choline and least for the sum of N-acetyl aspartate and N-acetylaspartylglutamate.
In agreement with previous studies at other field strengths, we found heterogeneous distribution of the major spectroscopically measurable brain metabolites. Although the most distinct differences are between tissue types, there is appreciable variation within a tissue type at different locations. The spectra and metabolite concentrations presented should provide a useful reference for both clinical and research MR spectroscopy studies performed at 3T.
Proton MR spectroscopy (MRS) of the brain is a useful technique for evaluating several neurological and psychiatric diseases (1). For clinicians interpreting spectra from individual patients, it is important to have knowledge of the normal range of spectral patterns from different brain regions, including how the spectra may depend on the patient’s age, and on the spectroscopic technique used. While normal regional and age-related spectral variations have been reported previously at field strengths of 1T (2–5), 2T (6–8), and 4T (9,10), to the best of our knowledge, there have been no reports using 3T scanners. Because use of 3T scanners is increasing for neuroimaging and spectroscopy in clinical practice, there is, therefore, a need for normative 3T data for comparison with patient studies. Also, even when results from all field strengths are taken together, there have been relatively few quantitative reports of normal values; therefore, expansion of coverage of different brain regions and confirmation of previously reported values is desirable.
The purpose of this study was to establish normative spectroscopic data at 3T, from a variety of regions often involved in brain pathology, using commonly available methodology (single voxel PRESS localization at short echo time, with analysis using the LCModel software (11)). Because MRS-measurable metabolic asymmetry in the human brain is minimal (12), unilateral measurements from multiple different anatomical regions were measured, rather than bilateral measurements in fewer structures. Quantitative analysis techniques are particularly important when evaluating patients with subtle metabolic abnormalities, or when no comparison spectra from presumed normal brain regions (in the patient under study) are available.
Thirty-five normal young adults (20 men, 15 women; age range, 20–41 years; mean, 31.4 years; 33 right-handed) were studied. Because metabolite concentrations have been shown to vary with age in children, and to a lesser extent in adolescents (13) and adults (10,14–18), we recruited subjects from a limited age range, over which metabolite concentrations were not expected to vary due to age. Handedness was not a criterion for subject selection, as metabolite levels have generally been observed to be symmetrical between the left and right hemispheres, independent of handedness (12). The protocol was approved by the local institutional review board, and informed written consent was obtained from each subject before the examination.
Scanning was performed on a 3T Philips Intera scanner using a quadrature birdcage transmit–receive head coil. Sagittal and axial T1-weighted localizer images were acquired before MRS. Single voxel MRS was performed on voxels graphically prescribed from the T1-weighted images (PRESS localization; “WET” water suppression using three suppression pulses (19); TE = 35 ms; TR = 2000 ms; 128 averages). An unsuppressed water spectrum (TE = 35 ms; TR = 5,000 ms; 4 averages) was also acquired for each voxel. Before data collection for each voxel, the field homogeneity was optimized using second order shim corrections and the FASTMAP technique (20). Four outer-volume suppression pulses were used to improve spatial localization and eliminate spurious signal from outside the desired region of interest. A total of 92 voxels were collected from nine anatomical locations (Fig. 1), with 9–11 voxels per location, and no repetitions of the same location in the same subject. From one to four locations were studied per session; some subjects returned for multiple sessions, for a total of one to eight locations per subject. All locations were on the left side or in the midline. Most voxels were approximately 20 × 20 × 20 mm3 in size (range, 5.7–8.8 cm3; mean, 7.9 cm3). Dimensions and orientation of each voxel were adjusted to match the size and shape of the targeted anatomical area. To correct for cerebrospinal fluid (CSF) included within the voxels, a heavily T2-weighted image with location, orientation, and slice thickness corresponding to the location, orientation, and z-dimension of each spectroscopic voxel was acquired (FSE; ETL = 8; effective TE = 500 ms; TR = 3,000 ms; 1.0 × 1.0 mm2 in-plane voxel dimensions). A phantom containing distilled water was placed beside the head and included in the field of view of this image to provide a calibration signal corresponding to 100% water.
Metabolite concentrations were estimated using an automated fitting routine, LCModel (11). This software automatically adjusts the phase and ppm shift of the spectra, estimates the baseline, and performs eddy current correction. Relative metabolite concentrations and their uncertainties are estimated by fitting the spectrum to a basis set of spectra acquired from individual metabolites in solution. The 16 metabolites included in our LCModel basis set were alanine, aspartate, creatine (Cr), γ-aminobutyric acid, glucose, glutamine (Gln), glutamate (Glu), glycerophosphocholine (GPC), phosphocholine (PC), lactate, myo-inositol (mI), N-acetyl aspartate (NAA), N-acetylaspartylglutamate (NAAG), scyllo-inositol, taurine, and guanine. Levels of many of these metabolites are not reliably detected above the noise level in normal brains at 3T. On this basis, five reliably measured metabolites and metabolite combinations were selected for further analysis: total NAA (tNAA, the sum of NAA and NAAG), total choline (tCho, predominantly GPC and PC), total Cr (tCr, the sum of Cr and phosphocreatine [PCr]), mI, and the sum (Glx) of Glu plus Gln.
LCModel quantifies metabolite levels by referencing to the unsuppressed water peak. In brief, this method divides the signal from a metabolite peak in a water-suppressed spectrum by the signal from the water peak in an unsuppressed spectrum acquired from the same voxel, applying corrections for factors such as different T1 and T2 relaxation times between the metabolite and water, and the number of 1H nuclei contributing to the metabolite and water peaks. This results in a molal concentration of the metabolite (moles of metabolite per mass of water). Optionally, LCModel can be configured to divide by the concentration of water in tissue to give a result in moles of metabolite per volume of tissue (molar concentration).
The output from LCModel gives the concentration of a metabolite relative to all of the MR-visible water within the voxel, regardless of whether the water arises from tissue, CSF, or blood vessels. However, molar or molal concentrations in the tissue alone are generally desired. The metabolite signal is assumed to arise entirely from brain tissue (which may not be true for some metabolites, but a valid assumption for all the metabolites considered in this study). Because the T2 of water in brain tissue is shorter than the T2 of water in CSF (by a factor of approximately 10) (21), at long echo times, nearly all of the MR-visible water can be ascribed to CSF. This can be used to estimate the partial volume of CSF in the voxel (22). The apparent metabolite concentrations, [M], output from LCModel, were corrected for CSF partial volume as follows:
where %CSF is given by
using MRI signal intensity (SI) on the heavily T2-weighted images. Signal in the phantom was summed over a 10 × 10 mm2 region of interest located centrally within the phantom to reduce the influence of susceptibility effects at the margin of the phantom. The method is sensitive to variations in B1 field strength, as well as water T1 and T2 relaxation time differences between the phantom and CSF; however, these effects are expected to be relatively small under the conditions used here. No correction was applied for the small volume of water contained within blood vessels.
The LCModel software uses basis spectra of pure compounds that are acquired by scanning phantoms composed of an aqueous solution of the metabolite of interest plus marker compounds (to calibrate the chemical shift scale [ppm]). The TR used to acquire the LCModel basis set was different from the TR used for the research subjects, and, furthermore, T1 relaxation times in vitro differ from relaxation times in vivo, so a T1 relaxation correction was applied. The T1 correction factor, Fcorr(T1), for each metabolite was calculated from:
The basis set was acquired with TR = 10,000 ms, so the numerator in equation  can be approximated to 1, and the correction factor becomes
Estimates of effective in vivo T1 relaxation times for each metabolite were taken from published values (23) for hemispheric white matter (mixed locations) at 3T as follows: tNAA = 1860 ms, tCho = 1320 ms, tCr = 1740 ms, mI = 1030 ms, and Glx = 1230 ms. Although these metabolites have multiple resonances (24), resulting in complicated decay patterns for several of the metabolites, the correction was based on a single-exponential model applied to the main resonance peak.
Although the LCModel basis spectra and the spectra for the research subjects were acquired using the same TE, the T2 of the metabolites in vivo differs from the T2 of the metabolites in vitro, necessitating a T2 correction as well. The T2 correction factor, Fcorr(T2), for each metabolite was calculated from:
Estimates of in vivo and in vitro T2 relaxation times for tCho, tCr, and tNAA in the supratentorial brain were taken from published values (25) for the main peaks. There is slight regional variation in the T2 relaxation times, so for supratentorial locations without a reported T2 relaxation time, we used the value from the available location with the greatest histological similarity. Published 3T T2 relaxation times for the pons and inferior vermis are not available, so we scanned an additional seven volunteers using echo times of 38, 140, and 280 ms to estimate the T2 in the pons (Cho = 220 ms, Cr = 141 ms, and NAA = 257 ms) and inferior vermis (Cho = 259 ms, Cr = 170 ms, and NAA = 226 ms). To our knowledge, published estimates of effective in vivo T2 relaxation times for mI and Glx are not available, and we report concentrations of these metabolites without correction for T2 effects. As a rough estimate of the error incurred by this omission, forgoing the correction for the shortest relaxation time in the reference cited (Cr in the cingulate gray matter) would result in a 16% underestimation of the concentration.
One-way analysis of variance (ANOVA) with Tukey post hoc test was used to compare the concentrations at each of the nine locations for each metabolite, and also to evaluate concentration differences between gray matter, white matter, and subcortical structures. To examine the effect of age on regional apparent metabolite concentration and average gray and white matter metabolite concentrations, linear regression was used, with an F-test applied to test for significance. In the linear regression analyses, P ≤ 0.01 was considered to be significant. All statistical analyses were performed using the SPSS software package (SPSS, Chicago, IL).
The output from the LCModel includes the original spectrum (phased and frequency-adjusted, based on the NAA peak), a fitted spectrum, an estimate of the underlying baseline (which may include resonances from macromolecular compounds, although no correction for macromolecules is explicitly applied), and the residuum from the fit (presumed to be mostly background noise). For each voxel, the baseline plot was subtracted from the phased and frequency-corrected spectrum. The results of this subtraction were normalized so that the area of the NAA peak reflects the concentration of NAA, and then were averaged together for each anatomical location (Fig. 2).
Estimated molal concentrations of mI, tCho, tCr, tNAA, and Glx (after correction for CSF partial volume, T1 effects, and T2 effects for tCho, tCr, and tNAA) are presented in Table 1. The results are graphically presented in Figure 3. Concentration of mI varied from 4.4 mmolal (centrum semiovale) to 6.7 mmolal (pons). The lowest tCho was 1.3 mmolal in the midline occipital gray matter, and the highest was 3.7 mmolal in the pons. Concentration of tCr varied from 9.4 mmolal (pons) to 16.7 mmolal (inferior vermis). Glx concentration varied from 10.0 mmolal (parietal-occipital white matter) to 18.0 mmolal (inferior vermis). The lowest tNAA was 15.0 mmolal in the midline frontal gray matter, and the highest was 18.9 mmolal in the thalamus.
ANOVA showed significant regional differences for each metabolite (P < 0.0015 for all metabolites). Table 2 shows P values corresponding to regional differences in apparent metabolite concentrations, while Table 3 presents the differences when all gray matter regions and all white matter regions are considered together. Qualitative observations on apparent metabolite concentrations in gray matter and white matter progressing from anterior to posterior are summarized in Table 4.
For purposes of comparison to other published studies covering similar locations, our results from Table 1 have been converted from molal (metabolite per mass of water) into molar (metabolite per volume of tissue) units and presented in Table 6. The conversion requires an estimate of the concentration of water in tissue, which is not well characterized at most locations in the brain. The concentration of water in tissue is assumed to be 35.88 moles per liter in white matter and 43.30 moles per liter in gray matter (21); we used the average of these concentrations for the other three locations.
The lowest mI was 4.4 mmolal in the centrum semiovale, and the highest was 6.7 mmolal in the pons. These two locations were involved in all four of the comparisons that reached statistical significance. Trends in the mI concentration are not readily apparent, although in gray matter it was intermediate or high, while in white matter it was intermediate or low. Levels of mI showed very little variation in the gray matter. However, the mI peak is complex and has modest signal-to-noise ratio (SNR), which may obscure the finer details of its regional variation.
The lowest concentration of tCho was 1.3 mmolal in the midline occipital gray matter, and the highest was 3.7 mmolal in the pons. This metabolite showed the greatest regional variation, with 20 of 36 comparisons reaching statistical significance. Most gray matter locations were different from most white matter locations, and as a group, there was a significant difference between gray matter and white matter (P < 0.001), with lower tCho levels in gray matter. Most gray matter locations were also different from most subcortical locations. The only exceptions involving gray matter involved the midline frontal gray matter; this location did not differ significantly from centrum semiovale or from parietal-occipital white matter, but did differ significantly from midline occipital gray matter. It also did not differ from the thalamus or inferior vermis. The thalamus more resembled the white matter than did the other subcortical locations. The concentration of tCho decreased from anterior to posterior in both white matter and gray matter. This pattern in the gray matter has been reported in the medial cortical gray matter (8), and in mesial temporal gray matter (26). As a group, gray matter differed from white matter and from all subcortical locations (P < 0.001), while white matter differed from the pons (P < 0.001).
The tCr concentration varied from 9.4 mmolal (pons) to 16.7 mmolal (inferior vermis). Of 36 comparisons, 14 reached statistical significance, 8 of them involving the inferior vermis, and all but 1 involving subcortical locations. The inferior vermis differed from all other locations; the thalamus differed from centrum semiovale, parietal– occipital white matter, midline frontal gray matter, and pons; and midline occipital gray matter differed from pons and parietal– occipital white matter. The tCr decreased in white matter from anterior to posterior, and increased in gray matter from anterior to posterior, as reported in other studies (8). Previous studies have also reported significantly higher tCr in the cerebellum compared with the cerebrum (8).
Although the LCModel provides estimates of the individual levels of Glu and Gln, because of substantial overlap (even at 3T) the sum of the concentrations of these two compounds (Glx) is more reliably determined than the individual components. Therefore, only the regional variations in Glx were considered in this study. The peak is dominated by Glu, which is approximately 3–6 times as prevalent as glutamine. Glx varied from 10.0 mmolal (parietal– occipital white matter) to 17.9 mmolal (inferior vermis). Of 36 comparisons, 7 reached statistical significance. These were inferior vermis versus all white matter locations, and midline frontal gray matter and midline parietal gray matter versus centrum semiovale and parietal-occipital white matter. There were no significant differences among white matter locations or among gray matter locations. Taken as a group, levels in gray matter were higher than levels in white matter (P < 0.001), as would be expected, because this neurotransmitter and neurotransmitter precursor should be found in close proximity to the synapses. Glx in white matter decreased from anterior to posterior. As with mI, the Glx peak is complex and has modest SNR, which may obscure the finer details of its regional variation.
As with Glx, the LCModel returns estimates of both NAA and NAAG; however, the sum (tNAA) is more reliably estimated because separation of these two closely overlapping resonances is difficult, even at 3T. The lowest tNAA was 15.0 mmolal in the midline frontal gray matter, and the highest was 18.9 mmolal in the thalamus. Despite the relatively small range in concentrations, the high SNR of this resonance and relatively small variations between individuals resulted in 9 of 36 comparisons reaching statistical significance. These were midline frontal gray matter and midline parietal gray matter versus centrum semiovale, thalamus, and pons; and inferior vermis versus centrum semiovale, thalamus, and pons. Taken as a group, levels in white matter were higher than in gray matter (P < 0.001), as has been previously reported (8). tNAA in gray matter increased from anterior to posterior, in agreement with a prior study (7).
Several qualitative observations about the spectra can be made by inspection of Figure 2. In gray matter, the Cho peak is distinctly lower than the Cr peak, whereas in white matter, the peaks are approximately equal height; and in gray matter, the Glx complex is higher than in white matter. The thalamus is intermediate between gray matter and white matter in both of these respects. The pons and inferior vermis are distinct from both gray and white matter, and distinct from each other. In the pons, the Cho peak is higher than the Cr peak, while in the inferior vermis, Cr is higher than NAA and higher than Cho.
The ratio of NAA peak height to Cr peak height is higher in white matter than in gray matter. In gray matter, Cho decreases from anterior to posterior, while NAA increases from anterior to posterior. Cho in white matter is relatively constant, while NAA is highest in the centrum semiovale. Linewidth and linewidth variability are least in the parietal–occipital white matter and greatest in the pons. Linewidth and variability are intermediate in the inferior vermis, midline frontal gray matter, and frontal white matter, and low in the remaining areas.
When taken region-by-region, or in gray matter or white matter as a whole, the only metabolite for which any statistically significant correlation with age was found was Glx in the inferior vermis (P < 0.001; Fig. 4). With this one exception, this confirms our underlying assumption that metabolite levels would be reasonably constant over the age range of these volunteers.
Relatively few quantitative normative studies performed with single voxel technique have been published, particularly studies that cover a large number of anatomical areas. We were unable to locate any other normative studies performed at 3T with single voxel PRESS technique. The results presented here generally agree with those reported by others (6,8,10,18,27) working at other field strengths, and using the single voxel STEAM or PRESS technique (Table 6). Compared with the study with anatomical locations most similar to those presented here (8), several findings were the same, including the location of the lowest tCho in the midline occipital gray matter, highest tCr in the cerebellum (superior cerebellar vermis in their study, inferior cerebellar vermis in our study), lowest tCr in the parietal– occipital white matter, lowest mI in the centrum semiovale, and lowest Glx in the parietal– occipital white matter. The anterior-to-posterior increase in tNAA in cortical gray matter agrees with a prior study (7), which suggested that most of this change was accounted for by changes in NAAG. We also detected anterior-to-posterior decrease in gray matter Cho concentration, similar to previous findings in medial cortical gray matter (8) and in the mesial temporal gray matter (26). In general, the various white matter regions differed more from gray matter regions than from each other, with higher NAA and Cho levels in the white matter. Higher tCho in the white matter compared with gray matter is consistent with the other reports in Table 6 (6,8,10,18,27). The reports in Table 6 are not concordant with regard to relative levels of NAA in white matter and gray matter, with one reporting NAA higher in white matter (8), one roughly equal (6), and two higher in gray matter (10,18). There are multiple factors potentially contributing to the discrepancies between the studies, including the use of a variety of different experimental and data processing techniques, and differences in locations examined. Of the other studies in Table 6, only two of five corrected for T2 effects. None of the other studies corrected for T1 effects, but four of five were STEAM studies using TR = 6000 ms, so for these the T1 correction would have been approximately 3%. Two of the studies corrected for CSF partial volume. Three studies do not say how they handled the concentration of water in tissue, one assumed the same concentration in gray matter and white matter, and the other assumed different concentrations in gray matter and white matter.
Metabolite levels in the pons and inferior vermis differed from both gray and white matter regions. Variance between subjects was high in pons, inferior vermis, and midline frontal gray matter; these are the three regions nearest to bone/air/tissue interfaces and are the most difficult to shim. These were also the three regions that accounted for nearly all of the failed attempts to obtain useable spectra (data not shown). Variance was also high for Glx, a complex peak that normally has low SNR. These observations have implications for prospective design of clinical MRS studies, in that given a choice of locations that could be followed in a study, it would be better to select the location or locations with the best prospects of obtaining reliable data.
Apparent metabolite concentrations reported here are for the most part similar to those reported in the prior studies (6,8,10,18,27). Differences are in general smaller than the reported standard deviations. For tNAA and tCr, our concentrations tend to be slightly higher than average, and some of our concentrations of mI are lower than the other reports. Multiple factors might contribute to these tendencies, particularly the application of correction factors that were not applied in some of the other studies. Correction for inclusion of CSF was done in the current study; CSF corrections for individual voxels ranged from 0.8% to 36%, average 5.6%, which uniformly increases all metabolite concentrations compared with uncorrected values. Correction for T1 effects was done in the current study and not in the comparison studies. As four of the five comparison studies were performed with long TR, the systematic underestimation due to this omission is small; the difference would be greatest for NAA and Cr, the metabolites with longest T1, and least for mI. T1 correction factors ranged from 16 to 50%. Correction for T2 effects was done for three of the metabolites in the current study, but T2 corrections were applied to these metabolites in only two of the five comparison studies. Omission of T2 correction would have had little effect for Cho at most locations, due to the similarity of the in vivo and in vitro relaxation times at 3T. T2 correction factors ranged from −4 to +16%. T2 relaxation effects would be less in the four of five comparison studies that were acquired with shorter TE than ours, so the concentrations of the metabolites that we did not correct for T2 effects, mI and Glx, would be relatively more underestimated than theirs.
Other than correction for inclusion of CSF, neither we nor the comparison studies corrected for inclusion of tissues other than the target tissue within the spectroscopic voxel. The inferior vermis and gray matter voxels were the most susceptible to this issue. The midline occipital gray matter voxel is particularly affected, as the gray matter at this location is thinner than at most other locations in the cortex, and the folding pattern of the cortex is the most complex; thus, this voxel contains a fair proportion of white matter. The inferior vermis voxel is also particularly affected, because not only are the gray matter and white matter layers finely interleaved at this location, but the voxel also contains portions of other medial structures of the cerebellum, particularly portions of the cerebellar tonsils. In clinical practice, partial volume inclusion of other tissues should not be a problem if patient voxels are consistently placed in a similar manner to the reference data. However, the results obtained by single voxel spectroscopy may differ quantitatively from MR spectroscopic imaging (MRSI), because the higher spatial resolution of MRSI has the potential to decrease the amount of partial volume inclusion. The technique used for quantification (internal reference, external standard, or phantom replacement) can also affect the apparent concentration of the metabolite (28).
Although the current study was not designed to determine the advantages or disadvantages of 3T relative to other field strengths, a few issues related to field strength should be noted. It has been reported that 3T provides better spectral resolution and SNR than 1.5T, although the improvements may be smaller than those expected from simple theoretical arguments (29,30). Increased sensitivity and spectral dispersion from the use of higher fields is partially offset by decreased T2 and T2* relaxation times (29). These appear to be primarily due to both macroscopic and microscopic susceptibility effects, which increase linearly with field strength. Nevertheless, spectral resolution (particularly for coupled spin systems such as Glx and mI) and SNR do appear to be higher at 3T than 1.5T, and improve reliability of detection of these compounds and others. It would appear that field strengths higher than 3T are needed for complete separation of Glu from Gln, or NAAG from NAA, based on chemical shift dispersion alone (i.e., without the use of homonuclear decoupling or spectral editing techniques) (31).
In conclusion, the establishment of normative values will aid diagnosis in clinical situations where no appropriate reference voxel can be obtained from the patient being studied, such as when the abnormality is in a midline structure, or when the disease process is widespread or global. This study has shown that the major spectroscopically measurable metabolites are heterogeneously distributed in the brain, and provides spectra recorded at 3T from the most commonly studied brain locations in neuropathology. While the most distinct differences are between tissue types, there is appreciable variation within a tissue type at different locations. This finding is perhaps not surprising, given the great variety of function and cytoarchitecture at different locations in the brain.
The authors thank Terri Brawner, RT, and Kathie Kahl, RT, for their assistance in acquisition of the data. We also thank Andrew J. Dwyer, MD, for statistical consultation.
Contract grant sponsor: NIH; Contract grant number: P41RR15241; Contract grant number: R01NS042851.