MAGNETIC RESONANCE IN MEDICINE
25, 107-1 I9 ( 1992)
‘H Spectroscopic Imaging of Rat Brain at 7 Tesla E. J. FERNANDEZ,* A. A. MAUDSLEY, T. HIGUCHI,AND M. W. W E I N E R ~ Departments of Radiology and ?Medicine. University of California, Sun Francisco, and Veterans Administration Medical Center, 4150 Clement Street, 1 l M , San Francisco, California 94121 Received December 3, 1990; revised April 30, 1991
‘H magnetic resonance spectroscopic imaging has been used to obtain metabolite maps of the rat brain. The spin-echo-based technique has been evaluated with respect to water and lipid suppression and sensitivity. Metabolite maps were constructed for choline. creatine + phosphocreatine, amino acids, N-acetyl aspartate, and lactate. A spatial resolution of 3 X 3 mm (in plane) with 7-mm-thick slices was achieved routinely in 60-min ( 16 X 16 phase encoding) acquisitions. For higher intensity resonances, metabolite maps could be constructed in as little as 10 min. Results from phantoms and from rats under normal and focal ischemia conditions are presented. Q 1992 Academic press, Inc. INTRODUCTION
Heterogeneity of cerebral metabolites has been noted in normal brain ( 1) using a number of invasive techniques which have been developed for studying brain metabolism, such as microdialysis, histological stains, enzymatic assays, and radioactive markers. Substantial alterations of these metabolite distributions are observed with tissue injury. Unfortunately, because of the invasive nature of these analytical procedures, there often remain uncertainties about the stability of metabolite levels in such preparations. Moreover, to carry out a study of metabolite dynamics (for example, before, during, and following cerebral ischemia) many animals are required to produce a complete time course. Magnetic resonance (MR) methods can provide information about the metabolic state of the brain noninvasively, although with limited sensitivity and spatial resolution. Using surface coil observation or single-volume localized MR spectroscopic techniques, previous investigators have demonstrated spatial heterogeneities in metabolism in normal (2) and diseased (3, 4 ) brain. In particular, ‘H MR methods are now finding successful application to examination of both humans ( 5 , 6 ) and animal models ( 7). However, single-volume localization techniques restrict the shape of the region of tissue that can be examined, and collecting data from different regions (e.g., normal vs diseased) can sometimes be difficult. Poor localization can lead to spectral “contamination” by tissue of a different type or different metabolic state, confounding the interpretation of the data ( 8 ) . Spectroscopic imaging (SI) techniques, such as those based on phase encoding ( 9 , l o ) ,can circumvent many of these problems. To date, both ‘H and 31PMR SI have * To whom correspondence should be addressed at: Department of Chemical Engineering, Thornton Hall, University of Virginia, Charlottesville, VA 22903. 107
0740-3 194192 $3.00 Copyright Q 1992 by Academic Press. Inc. Ail rights of reproduction in any form reserved.
108
FERNANDEZ ET AL
Prepare
Gx
;
Excite
1 ;
+-
Refocus
Echo
I
FIG. I . RF and gradient sequence used for 'H spin-echo spectroscopic imaging. See text for description.
been applied to humans (22, 1 2 ) , but there has been less work in animal models. Although they result in a small loss of signal relative to localized MR spectroscopic methods, spatially sensitive data are collected from the entire sample ( 2 3 ) .After acquisition, the data can be reconstructed to produce spectra for arbitrary volumes as well as images showing the distributions of observed resonances. The development of software for the processing and analysis of spectroscopic imaging data is critical. The data sets involved are large (for example, 32 X 32 phase encodings in a two-dimensional SI experiment would result in 1024 spectra), and convenient methods for examining and analyzing such data are required. The construction of metabolite images provides a quick, convenient method for viewing a portion of the spectral domain of the entire dataset. Registration of SI data with conventional MR images is also important, since MRI can provide anatomical reference points for interpretation of SI data. A combination data processing/display software package has been developed in our laboratory for this purpose ( 2 4 ) . Automated curve fitting and statistical analysis of individual spectra, such as that developed by Nelson and Brown. also facilitates quantitation of spectroscopic imaging data ( 1 5 ) . The goal of this study was to develop a 'H spectroscopic imaging technique useful for studies of brain metabolism in the rat at 7 T. The technique was optimized and evaluated with respect to water suppression, lipid suppression, magnetic susceptibility artifacts, and sensitivity. MATERIALS A N D METHODS
Instrumentation Experiments were performed on a Nalorac (Martinez, CA) Quest 4400 Imaging Spectrometer system operating at 300 MHz with a useable bore of 1 1.5 cm. The region of the magnet that could quickly be shimmed to 0.1 ppm half-height linewidth was roughly a sphere 3.5 cm in diameter. The actively shielded gradient/shim coils had maximum strength of 2.2 G / c m axially and 1.1 G / c m in the transverse directions. A birdcage resonator 4.0 cm long and 4.0 cm in diameter was used.
'H SPECTROSCOPIC IMAGING OF RAT BRAIN
109
Pulse Sequence The spin-echo spectroscopic imaging pulse sequence is shown in Fig. 1. The preparation period consisted of selective excitation of the water resonance and homospoil before the excitation pulse ( 1 6 ) . For excitation, a five-lobed sinc pulse was used to select a 7-mm-thick axial slice. The calculated chemical-shift artifact between the regions from which water and lactate were detected was approximately 1 mm in the axial direction. Phase encoding gradients were applied in the two transverse directions immediately after the excitation pulse. Refocusing was accomplished with a 2662 binomial chemical-shift-selective pulse (17, 18) with the excitation null set at the water resonance and the maximum excitation set for NAA at 2.0 ppm. The refocusing pulse was bracketed by gradient pulses ofequal area (0.6 G/cm, 10 ms) which selected only the magnetization that was properly refocused. These gradient pulses improved water suppression significantly ( 19). Gradient pulses in the first half of the echo were applied immediately following the excitation pulse to minimize any residual eddy currents that might interfere with the refocusing pulse. Likewise, the rephasing gradient was applied immediately following the refocusing pulse to minimize any effects of eddy currents on the acquisition. Trapezoidal gradients with 400 ps ramp times were used throughout. Magnetic field homogeneity was first optimized over the entire sensitive region of the birdcage coil and then over the 7-mm slice used for spectroscopic imaging experiments. In vivo water linewidths were typically 35-40 Hz. The echo time was chosen as a multiple of 1 /2J to keep J-coupled multiplets in phase. An echo time of 272 ms water used to provide lipid and additional water suppression ( 2 0 ) , and water suppression factors of over 10,000 were regularly achieved. A 1-s TR was used, which is significantly less than 3 times the T I values reported for human brain at 4.7 T (2). Thus the tip angle of the excitation pulse was set to 135", such that together with the refocusing pulse the effective tip angle would be less than 90". Sixteen phase encoding steps were typically used in both of the transverse directions. Two to twelve signal averages were collected at each phase encoding, with add/subtract phase cycling was incorporated to reduce DC offset artifact, leading to an acquisition times from 8.5 to 60 min. The field of view (FOV) for the spectroscopic images was 5 cm, except where noted otherwise.
Data Processing Data were usually collected as symmetrical echoes with 512 points and a sweep width of 4000 Hz. For TE = 136 ms it was necessary to acquire asymmetric spin echoes, since the TE was too short to allow sampling of a symmetrical echo. In these cases, only 179 points were collected in the first half of the echo. Before Fourier transformation, asymmetrical echoes were zero filled in the spectral domain in such a way as to make the echoes symmetrical. No significant change in lineshape was observed from the symmetrical echo case. Exponential line broadening of 5.0 Hz was applied in the spectral domain. No resolution enhancement was performed. Typically, 16 phase encoding steps were used in both spatial dimensions, and data sets were zero filled to 32 points. Mild exponential apodization (symmetric with the weighting func-
110
FERNANDEZ ET AL.
tion equal to e-'.' at the edge of the data) was used in the spatial domain to reduce Gibbs ringing artifacts. The reconstructed data were examined with Spectroscopic Imaging Display (SID) software developed in our laboratory ( 14). This software incorporates several features useful for the analysis of spectroscopic imaging data, including construction of spectroscopic images based on integrals over user-defined regions of the spectrum, overlay of MRI images or image edges for anatomical registration, multiple spectral plot options, and arithmetic manipulation of multiple images (e.g., differences or ratios of two images, masking of one image from another, construction of Bo field maps). Pixel intensity in spectroscopic images of individual metabolites was proportional to the signal integrated over a user-defined chemical-shift range (e.g., 1.25-1.40 ppm for lactate).
Protocol for Rat Spectroscopic Imaging E-xperiments Fisher 344 rats (200-250 g) were used in all experiments. Anesthesia was induced with pentobarbital (60 mg/kg body wt, administered i.p.) and maintained with isoflurane ( I % ) in a mixture of 30% oxygen:70%nitrous oxide for studies of normal rats. During the experiments on normal rats, rectal temperature and blood pressure were monitored. Body temperature was maintained by warming the bore with warm air. The rats were positioned in the magnet such that the brain was centered in the birdcage coil, and the position was verified by MRI. Focal ischemia was accomplished by the method of Longa et al. (21) . The middle cerebral artery was occluded by insertion of a 4-0 suture in the right carotid artery up to the origin of the middle cerebral artery. Using this model, reperfusion of the affected region can be accomplished by careful removal of the suture. RESULTS A N D DISCUSSION
' H Spectroscopic Imaging of Phantoms Figure 2 shows the spectroscopic images of four tubes, each containing solutions of either acetate or alanine. The high SIN permitted 32 X 32 phase encodings to be acquired over the 30-mm FOV. Thus, the spatial resolution was approximately 1 X 1 X 7 mm. The pixel intensity is proportional to peak area in a particular region of the 'H MR spectrum. Figure 2A shows the spectroscopic image of signal intensity in both the alanine and acetate regions of the spectrum; thus, all four tubes are visible in the image. Figures 2B and 2C were constructed from the acetate and alanine regions of the spectrum, respectively, demonstrating images from a single metabolite. Neither image is contaminated by either water or the other metabolite. In addition, the image intensity is relatively uniform throughout each tube. Figure 2D shows a profile through the peak integral image as a function of position horizontally along a line through the two tubes containing alanine (tubes c and d). The profile shows that the signal integral does not have sharp edges at tube-solution interfaces. The gradual transitions between the alanine-containing and alanine-free regions represent both partial volume effects and the influence of the point spread function. These considerations are important in the interpretation of spectroscopic
‘H SPECTROSCOPIC IMAGING OF RAT BRAlN
111
FIG.2. Spectroscopic images of four-tube phantom. Solutes were dissolved in tap water. The contents of the four tubes were (a) 50 mM acetate, (b) I M acetate, (c) 50 m M alanine, and ( d ) 20 rnM alanine. Spectroscopic images were constructed based on the integral of signals due to ( A ) alanine and acetate, ( B ) only alanine, (C) only acetate. ( D ) Profile of signal integral through tubes ( c ) and (d). Over the 30 mm FOV, 32 X 32 phase encodings were acquired. Linebroadening of 5.0 Hz was applied. No zero filling was used in the spectral dimension. Spatial filtering was as described under Materials and Methods. The scaling on image ( B ) was increased to show detail in tube ( d ) .
imaging experiments, in which one goal may be to determine the spatial extent of a metabolite.
Evaluation of the Spectroscopic Imaging Method Variations in signal intensity observed on spectroscopic images may arise for a number of reasons. B1inhomogeneity effects are expected to be minor given the birdcage coil used. Loss of peak area was observed in voxels with broad peaks ( for example on the left side of tube “a” in Fig. 2), suggesting that Bo inhomogeneity might have been involved as observed in the distortion of MR images despite the fact that signal integration was performed over a large range ( 1 ppm) either side of the resonance. Magnetic field homogeneity critically affects S I N , spectral resolution, and water
-
I12
FERNANDEZ ET AL.
suppression in any high-resolution 'H NMR of biological samples. Field homogeneity is determined by the heterogeneity of the empty magnet and by spatial variations in the magnetic susceptibility of the sample. Changes in susceptibility can cause severe artifacts in conventional MR images (22). Moreover, the effects of magnetic susceptibility (such as at the skull-brain interface) are more severe at higher field (22). However, it should be noted that magnetic susceptibility effects in spectroscopic images may largely be overcome by producing images from line-fitted total peak areas after accounting for shifts in resonance position. ' H Spectroscopic Imaging of Cerebral Metabolites in Norma/ Rat
'H SI spectra from a normal rat brain are shown in Fig. 3. All spectra were obtained from approximately the same location in the basal ganglia. Figure 3a shows a spectrum obtained from a single voxel of a 4-h acquisition. Using NAA as a chemical-shift reference, resonances are observed for choline (Cho, 3.2 ppm), creatine phosphocreatine (Cr PCr, 3.0 ppm), four resonances in the amino acid region (2.68, 2.48,
+
+
7
5
4
3
2
1
PPM
FIG.3. Water-suppressed 'H SI spectra of normal rat brain. (a) Single-voxel ( 3 X 3 X 7 mm) spectrum from 4-h acquisition, TE 272 ms. ( b ) Single-voxel spectrum from 10-min acquisition, TE 272 ms. ( c )Singlevoxel spectrum from 60-min acquisition, TE 272 ms. ( d ) Single-voxel spectrum from 60-min acquisition, TE 136 ms. Resonances observed are assigned to ( I ) choline (3.2 ppm), (2) creatine/phosphocreatine (3.0 ppm), (3-6) amino acid resonances (2.65.2.47, 2.33.2.12 ppm), and ( 7 ) N-acetyl aspartate (2.0 ppm). All spectra were obtained from approximately the same location in the basal ganglia.
I13
‘H SPECTROSCOPIC IMAGING OF RAT BRAIN
2.32, and 2.12 ppm), N-acetyl aspartate (NAA, 2.0 ppm), and a methyl group at 1.3 ppm which could be resting level lactate but could also contain contaminating lipid from the scalp. This “spatial contamination” precludes the observation of resting level lactate, even at TE 272 ms. Unfortunately, none of the resonances in the amino acid region can be conclusively assigned to a single compound. Based on chemical shifts from the literature, the four amino acid resonances are believed to represent aspartate and N-acetyl aspartate (Asp + NAA, 2.68 pprn), glutamine and NAA (Gln NAA, 2.48 ppm), glutamate and GABA (Glu + GABA, 2.32 ppm), and glutamate and glutamine (Glu Gln, observed as a shoulder to the NAA resonance at 2. I ppm) (23, 2 4 ) . Signal-to-noise ratio in the 4-h TE 272 ms spectrum was sufficient to estimate peak areas from all metabolites except resting level lactate (Fig. 3a), while a 10-min acquisition was suitable for studies of NAA and elevated levels of lactate produced by ischemia (Fig. 3b). S / N can be improved by summing individual voxel spectra for a large region of the brain, although the S / N of such a spectrum would be lower than the spectrum obtained from that region by a single-volume localization method ( 2 5 ) . Two of the greatest concerns in performing ‘H MR spectroscopy in vivo are water and lipid suppression. Figures 3c and 3d show that the water suppression with the sequence of Fig. 1 is excellent at both 136 and 272 ms echo times and a total acquisition time of 60 min. The water resonance linewidths from the slice selected shimming were generally less than about 40 Hz, and under these conditions the water suppression was equivalent to that in Fig. 3. The NAA linewidth for a single voxel was variable, but was usually about 20 Hz. Where water linewidths were greater, the water suppression usually deteriorated to the point of obscuring the Cr PCr and Cho regions of the spectrum. The spectral quality obtained using SI methods compares favorably with that obtained using surface coil methods at 360 MHz ( 7). Spectral resolution is an important consideration for experiments involving frequency-selective pulses, such as magnetization transfer 3’Pmeasurements of enzyme reaction rates and ‘H homonuclear editing. The spectral resolution obtained here shows that large magnetic susceptibility variations are not normally observed at this high field in the rat head and thus will not normally preclude such experiments. However, areas that have been shown to exhibit large susceptibility shifts (e.g., the sinuses in the human head) may need to be avoided for homonuclear editing experiments. Magnetic field homogeneity might be improved by incorporating some degree of localization into the shimming procedure and SI data acquisition (e.g., such methods as STEAM, PRESS, or surface coil localization ( 19, 26, 27)). Shown in Fig. 4 are spectroscopic images of protonated metabolites in a normal rat taken with TE 136 ms. Images are shown for Cho, Cr PCr, amino acids, NAA, and total metabolites. A conventional MR image of the same FOV is also shown (Fig. 4F). Pixel intensity is proportional to integrated signal intensity within a user-defined region of the spectrum. All of the metabolites are found throughout the brain, though regional variations of NAA and resting level lactate are frequently observed (unpublished observations). In the case of Cho (Fig. 4A) and Cr PCr (Fig. 4B) images there is some signal intensity outside the brain which may arise from the muscle in other parts of the head. Strong signals were obtained from the lipid lactate region
+
+
+
+
+
+
114
FERNANDEZ ET AL.
FIG.4. Spectroscopic images of cerebral metabolites in normal rat acquired with TE = 136 ms. Pixel intensity is proportional to the integrated signal intensity. ( A ) choline, (B) creatine phosphocreatine, ( C ) amino acids, ( D ) glutamate + glutamine N-acetyl aspartate (combining the peak at 2.0 ppm and the shoulder at 2.1 ppm), ( E ) lactate and lipid, (F) MR image of the rat head.
+
+
115
'H SPECTROSCOPIC IMAGING OF RAT BRAIN
of the spectrum at TE I36 ms (Fig. 4E), while TE 272 ms images yielded little signal (data not shown),
Eflects of Echo Time To increase sensitivity, there is a great motivation to obtain 'H MR data at shorter spin-echo times. Therefore, the effects of a shorter echo time on spectral quality of 'H SI data were investigated. Figures 3c and 3d show single-voxel spectra obtained from 136-ms and 272-ms echo time SI data sets from the same region of the same normal rat brain. The S I N was increased in the 136-ms echo time spectrum by a factor depending on the T , of each metabolite. Because of the complex strong Jcouplings involved in the case of the amino acid resonances, the calculated values do not correspond to simple T2relaxation times. Individual metabolite peak areas have been measured using NMRl (New Methods Research, Syracuse, NY) and the metabolite ratios and T2values calculated from a two-point fit for NAA, Cr PCr, and Cho agree well with published values. Decreased echo times can lead to reduced water and lipid suppression since water and lipid have much shorter T , values than cerebral metabolites. While water suppression in TE = 136 ms experiments did not deteriorate significantly, the lipid signal was considerably higher. Figure 5 shows stackplots of individual voxel spectra along a horizontal line through the brain from two 'H SI data sets collected with TE = 272 and 136 ms from the same rat. The TE = 272 ms dataset shows little or no contamination of brain voxels by lipid. However, at TE = 136 ms a very large lipid signal was present, and while other parts of the spectrum are not affected, the lactate region in the rat brain was contaminated by noise ridges propagating from the lipid resonance. This contamination is more than expected from the point spread function, and we
+
FIG. 5. Stackplot of individual voxel spectra taken from a horizontal line of voxels across the center of the brain of a normal rat. The SI data set was collected with 16 phase encodings in both transverse directions. The data set was zero filled to 32 data points in the spatial dimensions. Acquisition time was approximately 1 h. ( A ) TE 272 ms. (B) TE 136 ms.
116
FERNANDEZ ET AL.
are investigating the possibility of instrumental instabilities as a contributing factor. Such contamination has been observed consistently at 136 ms echo times, precluding the use of such data sets for quantitative assessment of lactate concentrations. In such cases, it may be necessary to modify the experiment to include the 'H homonuclear editing technique to selectively observe lactate (28).Indeed, similar problems of lipid contamination have been noted in 'H MRSI studies of human brain even with volume localization applied in addition to phase encoding ( 12).
Spectroscopic Imaging of Focal Cerebral Ischemia Figure 6 demonstrates how spectroscopic imaging can determine altered spatial metabolite distributions. A study of focal cerebral ischemia was camed out 30 min after occlusion of the middle cerebral artery, performed as outlined in the methods section. Figure 6 shows a comparison of a conventional MR image, 'H MRSI images of NAA and lactate (60 min acquisition, TE 272 ms), and a photomicrograph of a 2,3,5-triphenyltetrazolium chloride (TTC) staining of the same brain section subjected to severe focal ischemia. TTC stains normal tissue deep red and leaves white regions of irreversible ischemic damage. Notably, the region of elevated lactate in the spectroscopic image shown in Fig. 6C corresponds well with the region of the brain not stained by TTC as shown in Fig. 6D. These results demonstrate that the metabolite maps provided by spectroscopic imaging accurately complement the anatomical detail provided by MR images and that the observed distribution of increased lactate corresponds well to the region of tissue damage as determined from tissue staining performed after the experiment. CONCLUSIONS
These results demonstrate the feasibility of studying cerebral metabolism in rat brain by IH SI at high field. Sensitivity was found to be sufficient to provide spectra with a homogeneous volume coil at 3 X 3 X 7-mm nominal spatial resolution in approximately 20 min. Magnetic susceptibility variations were not found to be a limiting factor in carrying out such experiments. Spectroscopic images were obtained under normal and focal ischemia conditions. There are important limitations of this technique. First, S I N limits the spatial resolution that can be achieved. Improvements in sensitivity will come with the use of small surface coils, but at the expense of uniform B I , and therefore more difficult quantitation of metabolites. Since the S / N in the SI experiment is nearly that of singlevolume localization techniques, there is little sensitivity advantage to be gained by simple localization for observation of spectra from small tissue volumes; however, if
FIG.6 . Comparison of conventional MR image. 'HMR SI, and TTC-stained section of brain from a rat subjected to 3.5 h of focal ischemia. (A) Conventional 128 X 128 MR image of the rat head during ischemia with an overlay indicating the smaller region which is shown in figures (B) and ( C ) . Spectroscopic images during ischemia of ( B ) NAA, and (C) lactate. ( D ) Photomicrograph of TTC-stained brain section to the same scale as ( B ) and (C). TTC stains normal tissue dark, while damaged tissue remains white.
'H SPECTROSCOPIC IMAGING OF RAT BRAIN
117
118
FERNANDEZ ET AL.
a spectrum from a large volume is desired, localization by conventional techniques such as ISIS ( 2 9 ) , rather than summing individual SI voxels, will give higher S I N ( 2 5 ) .Of course, improved S I N and/or time resolution can be obtained by collecting SI data with large voxels, for example, by phase encoding in only one dimension ( 3 0 ) . Finally, spectral contamination remains a significant issue, especially with regard to lipid at shorter echo times. Lipid suppression at short echo times can be improved by incorporating some degree of localization into the SI data acquisition (e.g., such methods as STEAM, PRESS, or surface coil localization (19, 26, 27)). These results demonstrate that ‘H SI can delineate variations in protonated metabolite concentrations in the rat brain. In cerebral ischemia there are substantial alterations of metabolite concentrations with regional differences which correlate with the degree and extent of tissue injury (31-33). MR SI techniques can monitor a number of these metabolites in a completely noninvasive manner, thereby allowing repeated measurements in a single animal. In particular, spectroscopic images of lactate, NAA, and amino acids such as glutamine, glutamate, aspartate, and GABA will provide insights into the metabolic role of these important compounds in stroke and other diseases. ACKNOWLEDGMENTS This work is supported by PHS Grants R01-CA48815, RO1-DK33293. and HL-07192-14, and the Veterans Administration Medical Research Service. The authors are very grateful to Dr. Hiroaki Shimizu and Dr. Philip R. Weinstein for animal preparation in the focal ischemia experiments.
REFERENCES 1. J. PEELING,D. WONG, AND G. R. SUTHERLAND, Stroke 20,633 ( 1989). 2. J. FRAHM,H. BRUHN,M. L. GYNGELL,K. D. MERBOLDT,W. HANICKE,AND R. SAUTER.Magn. Reson. Med. I I , 4 7 (1989). 3. M. W. WEINER,H. HETHERINGTON, B. HUBESCH,G. KARCZMAR,B. MASSIE,A. MAUDSLEY,D. J. MEYERHOFF. D. SAPPEY-MARINIER, S. SCHAEER, D. B. TWIEG,AND G. B. MATSON.NMR Biomed. 2,290 (1989). 4. B. Ross, P. T. NARASIMHAN, J. TROPP,AND K. S. DERBY,NMR Biorned. 2, 340 ( 1989). 5. C. C. HANSTOCK,D. L. ROTHMAN,T. JUE,AND R. G. SHULMAN,J. Magn. Reson. 77,583 ( 1988). 6. P. R. LUYTEN,C. SEGEBARTH, D. BAL~RIAUX, AND J. A. DEN HOLLANDER,Radiology 165(P), 66 (1987). 7. K. L. BEHAR,J. A. DEN HOLLANDER, M. E. STROMSKI, T. O ~ I N OR., G. SHULMAN, 0. A. C. PETROFF. AND J. W. PRICHARD, Proc. Natl. Acad. Sci. USA 80, 4945 ( 1983). 8. P. A. BOTTOMLEY, C. J. HARDY,P. B. ROEMER,AND R. G. WEISS,NMR Biomed. 2,284 ( 1989). 9. T. R. BROWN,B. M. KINCAID,AND K. UGURBIL,Proc. Natl. Acad. Sci. USA 79,3523 ( 1982). 10. A. A. MAUDSLEY,S. K. HILAL,W. H. PERMAN, AND H. E. SIMON,J. Magn. Reson. 51, 147 (1983). 11. A. A. MAUDSLEY, D. B. TWIEG,D. SAPPEY-MARINIER, B. HUBESCH,J. W. H u m , G. B. MATSON, AND M. W. WEINER,Magn. Reson. Med. 14,415 ( 1990). 12. P. R. LUYTEN,A. J. H. MARIEN,W. HEINDEL,P. H. J. VAN GERWEN,K. HERHOLZ,J. A. DEN HOLLANDER, G . FRIEDMA”,AND W.-D. HEISS,Radiology 176,791 ( 1990). 13. A. A. MAUDSLEY, J. Magn. Reson. 68,363 (1983). 14. A. A. MAUDSLEY, M. A. ELLIOT.AND M. W. WEINER,“Society of Magnetic Resonance in Medicine, Proceedings of 9th Annual Meeting, New York. NY, 1990,” p. 571. 15. S. J. NELSONAND T. R. BROWN,J. Magn. Reson. 84,95 (1989). 16. A. HAASE,J. FRAHM,W. HAENICKE,AND D. MATTHEI,Phys. Med. Biol. 30, 341 ( 1985). 17. P. J. HORE,J. Magn. Reson. 55, 283 (1983). 18. P. J. HORE. J. Magn. Reson. 54, 539 (1983).
‘H SPECTROSCOPIC IMAGING OF RAT BRAIN
119
19. A. A. DE GRAAF,W. M. M. J. BOVEE,N. E. P. DEUTZ,A N D R. A. F. M. CHAMULEAU. Mugn. Reson. Imaging 6,255 (1988). “Biomedical Magnetic Resonance” (T. L. James and A. R. 20. K. M. BRINDLEAND I. D. CAMPBELL, Margulis, Eds.), p. 243, Radiology Research and Education Foundation, San Francisco, 1984. 21. E. Z. LONGA,P. R. WEINSTEIN, S. CARLSON,AND R. CUMMINS, Stroke 20, 84 ( 1989). 22. K. M. LODEKE.P. RUSCHMANN, AND R. TISCHLER, Magn. Reson. Imaging 3,329 ( 1985). 23. C. ARCIS,YEN-CHANG, AND M. BARANY,Phys. Chem. Phys. Med. NMR 17,23 (1985). 24. A. A. DE GRAAF,J. E. VAN DUK,AND W. M. M. J. BOV~E, Magn. Reson. Med. 13,343 ( 1990 j. 25. W. A. EDELSTEIN, G. H. GLOVER,C. J. HARDY,AND R. W. REDINGTON. Mugn. Reson. Med. 3,604 ( 1986j. 26. P. R. LUYTEN,A. J. H. M A R I ~ N AND , J. A. DEN HOLLANDER, “Society of Magnetic Resonance in Medicine, Proceedings of 7th Annual Meeting, San Francisco, 1988,” p. 327. 27. P. R. LUYTEN,J. A. DEN HOLLANDER. C. SEGEBARTH, AND D. BAL~RIAUX, “Society of Magnetic Resonance in Medicine, Proceedings of 7th Annual Meeting, San Francisco, 1988,” p. 252. 28. D. L. ROTHMAN,K. L. BEHAR,H. P. HETHERINGTON, AND R. G. SHULMAN, Proc. Nail. Acud. Sci. USA 81,6330 (1984). 29. R. J. ORDIDGE,A. CONNELLY. AND J. A. B. LOHMAN, J. Mugn. Reson. 66,283 ( 1986). 30. J. C. HASELGROVE, V. H. SUBRAMANIAN, J. J. S. LEIGH,L. GYULAI,AND B. CHANCE,Science 220, 1 I70 (1983). 31. S. H. GRAHAM,K. SHIRAISHI, S. S. PANTER,R. P. SIMON,AND A. FADEN,Neurosci. Len. 110, 124 (1990).
32. V. R. S. KOLLURIAND G. Y. C. V. S. LAKSHMI, Neurochem. Res. 14, 621 ( 1989). 33. 0. U. SCREMIN AND D. J. JENDEN,Stroke 20, 1524 ( 1989).
25, 107-1 I9 ( 1992)
‘H Spectroscopic Imaging of Rat Brain at 7 Tesla E. J. FERNANDEZ,* A. A. MAUDSLEY, T. HIGUCHI,AND M. W. W E I N E R ~ Departments of Radiology and ?Medicine. University of California, Sun Francisco, and Veterans Administration Medical Center, 4150 Clement Street, 1 l M , San Francisco, California 94121 Received December 3, 1990; revised April 30, 1991
‘H magnetic resonance spectroscopic imaging has been used to obtain metabolite maps of the rat brain. The spin-echo-based technique has been evaluated with respect to water and lipid suppression and sensitivity. Metabolite maps were constructed for choline. creatine + phosphocreatine, amino acids, N-acetyl aspartate, and lactate. A spatial resolution of 3 X 3 mm (in plane) with 7-mm-thick slices was achieved routinely in 60-min ( 16 X 16 phase encoding) acquisitions. For higher intensity resonances, metabolite maps could be constructed in as little as 10 min. Results from phantoms and from rats under normal and focal ischemia conditions are presented. Q 1992 Academic press, Inc. INTRODUCTION
Heterogeneity of cerebral metabolites has been noted in normal brain ( 1) using a number of invasive techniques which have been developed for studying brain metabolism, such as microdialysis, histological stains, enzymatic assays, and radioactive markers. Substantial alterations of these metabolite distributions are observed with tissue injury. Unfortunately, because of the invasive nature of these analytical procedures, there often remain uncertainties about the stability of metabolite levels in such preparations. Moreover, to carry out a study of metabolite dynamics (for example, before, during, and following cerebral ischemia) many animals are required to produce a complete time course. Magnetic resonance (MR) methods can provide information about the metabolic state of the brain noninvasively, although with limited sensitivity and spatial resolution. Using surface coil observation or single-volume localized MR spectroscopic techniques, previous investigators have demonstrated spatial heterogeneities in metabolism in normal (2) and diseased (3, 4 ) brain. In particular, ‘H MR methods are now finding successful application to examination of both humans ( 5 , 6 ) and animal models ( 7). However, single-volume localization techniques restrict the shape of the region of tissue that can be examined, and collecting data from different regions (e.g., normal vs diseased) can sometimes be difficult. Poor localization can lead to spectral “contamination” by tissue of a different type or different metabolic state, confounding the interpretation of the data ( 8 ) . Spectroscopic imaging (SI) techniques, such as those based on phase encoding ( 9 , l o ) ,can circumvent many of these problems. To date, both ‘H and 31PMR SI have * To whom correspondence should be addressed at: Department of Chemical Engineering, Thornton Hall, University of Virginia, Charlottesville, VA 22903. 107
0740-3 194192 $3.00 Copyright Q 1992 by Academic Press. Inc. Ail rights of reproduction in any form reserved.
108
FERNANDEZ ET AL
Prepare
Gx
;
Excite
1 ;
+-
Refocus
Echo
I
FIG. I . RF and gradient sequence used for 'H spin-echo spectroscopic imaging. See text for description.
been applied to humans (22, 1 2 ) , but there has been less work in animal models. Although they result in a small loss of signal relative to localized MR spectroscopic methods, spatially sensitive data are collected from the entire sample ( 2 3 ) .After acquisition, the data can be reconstructed to produce spectra for arbitrary volumes as well as images showing the distributions of observed resonances. The development of software for the processing and analysis of spectroscopic imaging data is critical. The data sets involved are large (for example, 32 X 32 phase encodings in a two-dimensional SI experiment would result in 1024 spectra), and convenient methods for examining and analyzing such data are required. The construction of metabolite images provides a quick, convenient method for viewing a portion of the spectral domain of the entire dataset. Registration of SI data with conventional MR images is also important, since MRI can provide anatomical reference points for interpretation of SI data. A combination data processing/display software package has been developed in our laboratory for this purpose ( 2 4 ) . Automated curve fitting and statistical analysis of individual spectra, such as that developed by Nelson and Brown. also facilitates quantitation of spectroscopic imaging data ( 1 5 ) . The goal of this study was to develop a 'H spectroscopic imaging technique useful for studies of brain metabolism in the rat at 7 T. The technique was optimized and evaluated with respect to water suppression, lipid suppression, magnetic susceptibility artifacts, and sensitivity. MATERIALS A N D METHODS
Instrumentation Experiments were performed on a Nalorac (Martinez, CA) Quest 4400 Imaging Spectrometer system operating at 300 MHz with a useable bore of 1 1.5 cm. The region of the magnet that could quickly be shimmed to 0.1 ppm half-height linewidth was roughly a sphere 3.5 cm in diameter. The actively shielded gradient/shim coils had maximum strength of 2.2 G / c m axially and 1.1 G / c m in the transverse directions. A birdcage resonator 4.0 cm long and 4.0 cm in diameter was used.
'H SPECTROSCOPIC IMAGING OF RAT BRAIN
109
Pulse Sequence The spin-echo spectroscopic imaging pulse sequence is shown in Fig. 1. The preparation period consisted of selective excitation of the water resonance and homospoil before the excitation pulse ( 1 6 ) . For excitation, a five-lobed sinc pulse was used to select a 7-mm-thick axial slice. The calculated chemical-shift artifact between the regions from which water and lactate were detected was approximately 1 mm in the axial direction. Phase encoding gradients were applied in the two transverse directions immediately after the excitation pulse. Refocusing was accomplished with a 2662 binomial chemical-shift-selective pulse (17, 18) with the excitation null set at the water resonance and the maximum excitation set for NAA at 2.0 ppm. The refocusing pulse was bracketed by gradient pulses ofequal area (0.6 G/cm, 10 ms) which selected only the magnetization that was properly refocused. These gradient pulses improved water suppression significantly ( 19). Gradient pulses in the first half of the echo were applied immediately following the excitation pulse to minimize any residual eddy currents that might interfere with the refocusing pulse. Likewise, the rephasing gradient was applied immediately following the refocusing pulse to minimize any effects of eddy currents on the acquisition. Trapezoidal gradients with 400 ps ramp times were used throughout. Magnetic field homogeneity was first optimized over the entire sensitive region of the birdcage coil and then over the 7-mm slice used for spectroscopic imaging experiments. In vivo water linewidths were typically 35-40 Hz. The echo time was chosen as a multiple of 1 /2J to keep J-coupled multiplets in phase. An echo time of 272 ms water used to provide lipid and additional water suppression ( 2 0 ) , and water suppression factors of over 10,000 were regularly achieved. A 1-s TR was used, which is significantly less than 3 times the T I values reported for human brain at 4.7 T (2). Thus the tip angle of the excitation pulse was set to 135", such that together with the refocusing pulse the effective tip angle would be less than 90". Sixteen phase encoding steps were typically used in both of the transverse directions. Two to twelve signal averages were collected at each phase encoding, with add/subtract phase cycling was incorporated to reduce DC offset artifact, leading to an acquisition times from 8.5 to 60 min. The field of view (FOV) for the spectroscopic images was 5 cm, except where noted otherwise.
Data Processing Data were usually collected as symmetrical echoes with 512 points and a sweep width of 4000 Hz. For TE = 136 ms it was necessary to acquire asymmetric spin echoes, since the TE was too short to allow sampling of a symmetrical echo. In these cases, only 179 points were collected in the first half of the echo. Before Fourier transformation, asymmetrical echoes were zero filled in the spectral domain in such a way as to make the echoes symmetrical. No significant change in lineshape was observed from the symmetrical echo case. Exponential line broadening of 5.0 Hz was applied in the spectral domain. No resolution enhancement was performed. Typically, 16 phase encoding steps were used in both spatial dimensions, and data sets were zero filled to 32 points. Mild exponential apodization (symmetric with the weighting func-
110
FERNANDEZ ET AL.
tion equal to e-'.' at the edge of the data) was used in the spatial domain to reduce Gibbs ringing artifacts. The reconstructed data were examined with Spectroscopic Imaging Display (SID) software developed in our laboratory ( 14). This software incorporates several features useful for the analysis of spectroscopic imaging data, including construction of spectroscopic images based on integrals over user-defined regions of the spectrum, overlay of MRI images or image edges for anatomical registration, multiple spectral plot options, and arithmetic manipulation of multiple images (e.g., differences or ratios of two images, masking of one image from another, construction of Bo field maps). Pixel intensity in spectroscopic images of individual metabolites was proportional to the signal integrated over a user-defined chemical-shift range (e.g., 1.25-1.40 ppm for lactate).
Protocol for Rat Spectroscopic Imaging E-xperiments Fisher 344 rats (200-250 g) were used in all experiments. Anesthesia was induced with pentobarbital (60 mg/kg body wt, administered i.p.) and maintained with isoflurane ( I % ) in a mixture of 30% oxygen:70%nitrous oxide for studies of normal rats. During the experiments on normal rats, rectal temperature and blood pressure were monitored. Body temperature was maintained by warming the bore with warm air. The rats were positioned in the magnet such that the brain was centered in the birdcage coil, and the position was verified by MRI. Focal ischemia was accomplished by the method of Longa et al. (21) . The middle cerebral artery was occluded by insertion of a 4-0 suture in the right carotid artery up to the origin of the middle cerebral artery. Using this model, reperfusion of the affected region can be accomplished by careful removal of the suture. RESULTS A N D DISCUSSION
' H Spectroscopic Imaging of Phantoms Figure 2 shows the spectroscopic images of four tubes, each containing solutions of either acetate or alanine. The high SIN permitted 32 X 32 phase encodings to be acquired over the 30-mm FOV. Thus, the spatial resolution was approximately 1 X 1 X 7 mm. The pixel intensity is proportional to peak area in a particular region of the 'H MR spectrum. Figure 2A shows the spectroscopic image of signal intensity in both the alanine and acetate regions of the spectrum; thus, all four tubes are visible in the image. Figures 2B and 2C were constructed from the acetate and alanine regions of the spectrum, respectively, demonstrating images from a single metabolite. Neither image is contaminated by either water or the other metabolite. In addition, the image intensity is relatively uniform throughout each tube. Figure 2D shows a profile through the peak integral image as a function of position horizontally along a line through the two tubes containing alanine (tubes c and d). The profile shows that the signal integral does not have sharp edges at tube-solution interfaces. The gradual transitions between the alanine-containing and alanine-free regions represent both partial volume effects and the influence of the point spread function. These considerations are important in the interpretation of spectroscopic
‘H SPECTROSCOPIC IMAGING OF RAT BRAlN
111
FIG.2. Spectroscopic images of four-tube phantom. Solutes were dissolved in tap water. The contents of the four tubes were (a) 50 mM acetate, (b) I M acetate, (c) 50 m M alanine, and ( d ) 20 rnM alanine. Spectroscopic images were constructed based on the integral of signals due to ( A ) alanine and acetate, ( B ) only alanine, (C) only acetate. ( D ) Profile of signal integral through tubes ( c ) and (d). Over the 30 mm FOV, 32 X 32 phase encodings were acquired. Linebroadening of 5.0 Hz was applied. No zero filling was used in the spectral dimension. Spatial filtering was as described under Materials and Methods. The scaling on image ( B ) was increased to show detail in tube ( d ) .
imaging experiments, in which one goal may be to determine the spatial extent of a metabolite.
Evaluation of the Spectroscopic Imaging Method Variations in signal intensity observed on spectroscopic images may arise for a number of reasons. B1inhomogeneity effects are expected to be minor given the birdcage coil used. Loss of peak area was observed in voxels with broad peaks ( for example on the left side of tube “a” in Fig. 2), suggesting that Bo inhomogeneity might have been involved as observed in the distortion of MR images despite the fact that signal integration was performed over a large range ( 1 ppm) either side of the resonance. Magnetic field homogeneity critically affects S I N , spectral resolution, and water
-
I12
FERNANDEZ ET AL.
suppression in any high-resolution 'H NMR of biological samples. Field homogeneity is determined by the heterogeneity of the empty magnet and by spatial variations in the magnetic susceptibility of the sample. Changes in susceptibility can cause severe artifacts in conventional MR images (22). Moreover, the effects of magnetic susceptibility (such as at the skull-brain interface) are more severe at higher field (22). However, it should be noted that magnetic susceptibility effects in spectroscopic images may largely be overcome by producing images from line-fitted total peak areas after accounting for shifts in resonance position. ' H Spectroscopic Imaging of Cerebral Metabolites in Norma/ Rat
'H SI spectra from a normal rat brain are shown in Fig. 3. All spectra were obtained from approximately the same location in the basal ganglia. Figure 3a shows a spectrum obtained from a single voxel of a 4-h acquisition. Using NAA as a chemical-shift reference, resonances are observed for choline (Cho, 3.2 ppm), creatine phosphocreatine (Cr PCr, 3.0 ppm), four resonances in the amino acid region (2.68, 2.48,
+
+
7
5
4
3
2
1
PPM
FIG.3. Water-suppressed 'H SI spectra of normal rat brain. (a) Single-voxel ( 3 X 3 X 7 mm) spectrum from 4-h acquisition, TE 272 ms. ( b ) Single-voxel spectrum from 10-min acquisition, TE 272 ms. ( c )Singlevoxel spectrum from 60-min acquisition, TE 272 ms. ( d ) Single-voxel spectrum from 60-min acquisition, TE 136 ms. Resonances observed are assigned to ( I ) choline (3.2 ppm), (2) creatine/phosphocreatine (3.0 ppm), (3-6) amino acid resonances (2.65.2.47, 2.33.2.12 ppm), and ( 7 ) N-acetyl aspartate (2.0 ppm). All spectra were obtained from approximately the same location in the basal ganglia.
I13
‘H SPECTROSCOPIC IMAGING OF RAT BRAIN
2.32, and 2.12 ppm), N-acetyl aspartate (NAA, 2.0 ppm), and a methyl group at 1.3 ppm which could be resting level lactate but could also contain contaminating lipid from the scalp. This “spatial contamination” precludes the observation of resting level lactate, even at TE 272 ms. Unfortunately, none of the resonances in the amino acid region can be conclusively assigned to a single compound. Based on chemical shifts from the literature, the four amino acid resonances are believed to represent aspartate and N-acetyl aspartate (Asp + NAA, 2.68 pprn), glutamine and NAA (Gln NAA, 2.48 ppm), glutamate and GABA (Glu + GABA, 2.32 ppm), and glutamate and glutamine (Glu Gln, observed as a shoulder to the NAA resonance at 2. I ppm) (23, 2 4 ) . Signal-to-noise ratio in the 4-h TE 272 ms spectrum was sufficient to estimate peak areas from all metabolites except resting level lactate (Fig. 3a), while a 10-min acquisition was suitable for studies of NAA and elevated levels of lactate produced by ischemia (Fig. 3b). S / N can be improved by summing individual voxel spectra for a large region of the brain, although the S / N of such a spectrum would be lower than the spectrum obtained from that region by a single-volume localization method ( 2 5 ) . Two of the greatest concerns in performing ‘H MR spectroscopy in vivo are water and lipid suppression. Figures 3c and 3d show that the water suppression with the sequence of Fig. 1 is excellent at both 136 and 272 ms echo times and a total acquisition time of 60 min. The water resonance linewidths from the slice selected shimming were generally less than about 40 Hz, and under these conditions the water suppression was equivalent to that in Fig. 3. The NAA linewidth for a single voxel was variable, but was usually about 20 Hz. Where water linewidths were greater, the water suppression usually deteriorated to the point of obscuring the Cr PCr and Cho regions of the spectrum. The spectral quality obtained using SI methods compares favorably with that obtained using surface coil methods at 360 MHz ( 7). Spectral resolution is an important consideration for experiments involving frequency-selective pulses, such as magnetization transfer 3’Pmeasurements of enzyme reaction rates and ‘H homonuclear editing. The spectral resolution obtained here shows that large magnetic susceptibility variations are not normally observed at this high field in the rat head and thus will not normally preclude such experiments. However, areas that have been shown to exhibit large susceptibility shifts (e.g., the sinuses in the human head) may need to be avoided for homonuclear editing experiments. Magnetic field homogeneity might be improved by incorporating some degree of localization into the shimming procedure and SI data acquisition (e.g., such methods as STEAM, PRESS, or surface coil localization ( 19, 26, 27)). Shown in Fig. 4 are spectroscopic images of protonated metabolites in a normal rat taken with TE 136 ms. Images are shown for Cho, Cr PCr, amino acids, NAA, and total metabolites. A conventional MR image of the same FOV is also shown (Fig. 4F). Pixel intensity is proportional to integrated signal intensity within a user-defined region of the spectrum. All of the metabolites are found throughout the brain, though regional variations of NAA and resting level lactate are frequently observed (unpublished observations). In the case of Cho (Fig. 4A) and Cr PCr (Fig. 4B) images there is some signal intensity outside the brain which may arise from the muscle in other parts of the head. Strong signals were obtained from the lipid lactate region
+
+
+
+
+
+
114
FERNANDEZ ET AL.
FIG.4. Spectroscopic images of cerebral metabolites in normal rat acquired with TE = 136 ms. Pixel intensity is proportional to the integrated signal intensity. ( A ) choline, (B) creatine phosphocreatine, ( C ) amino acids, ( D ) glutamate + glutamine N-acetyl aspartate (combining the peak at 2.0 ppm and the shoulder at 2.1 ppm), ( E ) lactate and lipid, (F) MR image of the rat head.
+
+
115
'H SPECTROSCOPIC IMAGING OF RAT BRAIN
of the spectrum at TE I36 ms (Fig. 4E), while TE 272 ms images yielded little signal (data not shown),
Eflects of Echo Time To increase sensitivity, there is a great motivation to obtain 'H MR data at shorter spin-echo times. Therefore, the effects of a shorter echo time on spectral quality of 'H SI data were investigated. Figures 3c and 3d show single-voxel spectra obtained from 136-ms and 272-ms echo time SI data sets from the same region of the same normal rat brain. The S I N was increased in the 136-ms echo time spectrum by a factor depending on the T , of each metabolite. Because of the complex strong Jcouplings involved in the case of the amino acid resonances, the calculated values do not correspond to simple T2relaxation times. Individual metabolite peak areas have been measured using NMRl (New Methods Research, Syracuse, NY) and the metabolite ratios and T2values calculated from a two-point fit for NAA, Cr PCr, and Cho agree well with published values. Decreased echo times can lead to reduced water and lipid suppression since water and lipid have much shorter T , values than cerebral metabolites. While water suppression in TE = 136 ms experiments did not deteriorate significantly, the lipid signal was considerably higher. Figure 5 shows stackplots of individual voxel spectra along a horizontal line through the brain from two 'H SI data sets collected with TE = 272 and 136 ms from the same rat. The TE = 272 ms dataset shows little or no contamination of brain voxels by lipid. However, at TE = 136 ms a very large lipid signal was present, and while other parts of the spectrum are not affected, the lactate region in the rat brain was contaminated by noise ridges propagating from the lipid resonance. This contamination is more than expected from the point spread function, and we
+
FIG. 5. Stackplot of individual voxel spectra taken from a horizontal line of voxels across the center of the brain of a normal rat. The SI data set was collected with 16 phase encodings in both transverse directions. The data set was zero filled to 32 data points in the spatial dimensions. Acquisition time was approximately 1 h. ( A ) TE 272 ms. (B) TE 136 ms.
116
FERNANDEZ ET AL.
are investigating the possibility of instrumental instabilities as a contributing factor. Such contamination has been observed consistently at 136 ms echo times, precluding the use of such data sets for quantitative assessment of lactate concentrations. In such cases, it may be necessary to modify the experiment to include the 'H homonuclear editing technique to selectively observe lactate (28).Indeed, similar problems of lipid contamination have been noted in 'H MRSI studies of human brain even with volume localization applied in addition to phase encoding ( 12).
Spectroscopic Imaging of Focal Cerebral Ischemia Figure 6 demonstrates how spectroscopic imaging can determine altered spatial metabolite distributions. A study of focal cerebral ischemia was camed out 30 min after occlusion of the middle cerebral artery, performed as outlined in the methods section. Figure 6 shows a comparison of a conventional MR image, 'H MRSI images of NAA and lactate (60 min acquisition, TE 272 ms), and a photomicrograph of a 2,3,5-triphenyltetrazolium chloride (TTC) staining of the same brain section subjected to severe focal ischemia. TTC stains normal tissue deep red and leaves white regions of irreversible ischemic damage. Notably, the region of elevated lactate in the spectroscopic image shown in Fig. 6C corresponds well with the region of the brain not stained by TTC as shown in Fig. 6D. These results demonstrate that the metabolite maps provided by spectroscopic imaging accurately complement the anatomical detail provided by MR images and that the observed distribution of increased lactate corresponds well to the region of tissue damage as determined from tissue staining performed after the experiment. CONCLUSIONS
These results demonstrate the feasibility of studying cerebral metabolism in rat brain by IH SI at high field. Sensitivity was found to be sufficient to provide spectra with a homogeneous volume coil at 3 X 3 X 7-mm nominal spatial resolution in approximately 20 min. Magnetic susceptibility variations were not found to be a limiting factor in carrying out such experiments. Spectroscopic images were obtained under normal and focal ischemia conditions. There are important limitations of this technique. First, S I N limits the spatial resolution that can be achieved. Improvements in sensitivity will come with the use of small surface coils, but at the expense of uniform B I , and therefore more difficult quantitation of metabolites. Since the S / N in the SI experiment is nearly that of singlevolume localization techniques, there is little sensitivity advantage to be gained by simple localization for observation of spectra from small tissue volumes; however, if
FIG.6 . Comparison of conventional MR image. 'HMR SI, and TTC-stained section of brain from a rat subjected to 3.5 h of focal ischemia. (A) Conventional 128 X 128 MR image of the rat head during ischemia with an overlay indicating the smaller region which is shown in figures (B) and ( C ) . Spectroscopic images during ischemia of ( B ) NAA, and (C) lactate. ( D ) Photomicrograph of TTC-stained brain section to the same scale as ( B ) and (C). TTC stains normal tissue dark, while damaged tissue remains white.
'H SPECTROSCOPIC IMAGING OF RAT BRAIN
117
118
FERNANDEZ ET AL.
a spectrum from a large volume is desired, localization by conventional techniques such as ISIS ( 2 9 ) , rather than summing individual SI voxels, will give higher S I N ( 2 5 ) .Of course, improved S I N and/or time resolution can be obtained by collecting SI data with large voxels, for example, by phase encoding in only one dimension ( 3 0 ) . Finally, spectral contamination remains a significant issue, especially with regard to lipid at shorter echo times. Lipid suppression at short echo times can be improved by incorporating some degree of localization into the SI data acquisition (e.g., such methods as STEAM, PRESS, or surface coil localization (19, 26, 27)). These results demonstrate that ‘H SI can delineate variations in protonated metabolite concentrations in the rat brain. In cerebral ischemia there are substantial alterations of metabolite concentrations with regional differences which correlate with the degree and extent of tissue injury (31-33). MR SI techniques can monitor a number of these metabolites in a completely noninvasive manner, thereby allowing repeated measurements in a single animal. In particular, spectroscopic images of lactate, NAA, and amino acids such as glutamine, glutamate, aspartate, and GABA will provide insights into the metabolic role of these important compounds in stroke and other diseases. ACKNOWLEDGMENTS This work is supported by PHS Grants R01-CA48815, RO1-DK33293. and HL-07192-14, and the Veterans Administration Medical Research Service. The authors are very grateful to Dr. Hiroaki Shimizu and Dr. Philip R. Weinstein for animal preparation in the focal ischemia experiments.
REFERENCES 1. J. PEELING,D. WONG, AND G. R. SUTHERLAND, Stroke 20,633 ( 1989). 2. J. FRAHM,H. BRUHN,M. L. GYNGELL,K. D. MERBOLDT,W. HANICKE,AND R. SAUTER.Magn. Reson. Med. I I , 4 7 (1989). 3. M. W. WEINER,H. HETHERINGTON, B. HUBESCH,G. KARCZMAR,B. MASSIE,A. MAUDSLEY,D. J. MEYERHOFF. D. SAPPEY-MARINIER, S. SCHAEER, D. B. TWIEG,AND G. B. MATSON.NMR Biomed. 2,290 (1989). 4. B. Ross, P. T. NARASIMHAN, J. TROPP,AND K. S. DERBY,NMR Biorned. 2, 340 ( 1989). 5. C. C. HANSTOCK,D. L. ROTHMAN,T. JUE,AND R. G. SHULMAN,J. Magn. Reson. 77,583 ( 1988). 6. P. R. LUYTEN,C. SEGEBARTH, D. BAL~RIAUX, AND J. A. DEN HOLLANDER,Radiology 165(P), 66 (1987). 7. K. L. BEHAR,J. A. DEN HOLLANDER, M. E. STROMSKI, T. O ~ I N OR., G. SHULMAN, 0. A. C. PETROFF. AND J. W. PRICHARD, Proc. Natl. Acad. Sci. USA 80, 4945 ( 1983). 8. P. A. BOTTOMLEY, C. J. HARDY,P. B. ROEMER,AND R. G. WEISS,NMR Biomed. 2,284 ( 1989). 9. T. R. BROWN,B. M. KINCAID,AND K. UGURBIL,Proc. Natl. Acad. Sci. USA 79,3523 ( 1982). 10. A. A. MAUDSLEY,S. K. HILAL,W. H. PERMAN, AND H. E. SIMON,J. Magn. Reson. 51, 147 (1983). 11. A. A. MAUDSLEY, D. B. TWIEG,D. SAPPEY-MARINIER, B. HUBESCH,J. W. H u m , G. B. MATSON, AND M. W. WEINER,Magn. Reson. Med. 14,415 ( 1990). 12. P. R. LUYTEN,A. J. H. MARIEN,W. HEINDEL,P. H. J. VAN GERWEN,K. HERHOLZ,J. A. DEN HOLLANDER, G . FRIEDMA”,AND W.-D. HEISS,Radiology 176,791 ( 1990). 13. A. A. MAUDSLEY, J. Magn. Reson. 68,363 (1983). 14. A. A. MAUDSLEY, M. A. ELLIOT.AND M. W. WEINER,“Society of Magnetic Resonance in Medicine, Proceedings of 9th Annual Meeting, New York. NY, 1990,” p. 571. 15. S. J. NELSONAND T. R. BROWN,J. Magn. Reson. 84,95 (1989). 16. A. HAASE,J. FRAHM,W. HAENICKE,AND D. MATTHEI,Phys. Med. Biol. 30, 341 ( 1985). 17. P. J. HORE,J. Magn. Reson. 55, 283 (1983). 18. P. J. HORE. J. Magn. Reson. 54, 539 (1983).
‘H SPECTROSCOPIC IMAGING OF RAT BRAIN
119
19. A. A. DE GRAAF,W. M. M. J. BOVEE,N. E. P. DEUTZ,A N D R. A. F. M. CHAMULEAU. Mugn. Reson. Imaging 6,255 (1988). “Biomedical Magnetic Resonance” (T. L. James and A. R. 20. K. M. BRINDLEAND I. D. CAMPBELL, Margulis, Eds.), p. 243, Radiology Research and Education Foundation, San Francisco, 1984. 21. E. Z. LONGA,P. R. WEINSTEIN, S. CARLSON,AND R. CUMMINS, Stroke 20, 84 ( 1989). 22. K. M. LODEKE.P. RUSCHMANN, AND R. TISCHLER, Magn. Reson. Imaging 3,329 ( 1985). 23. C. ARCIS,YEN-CHANG, AND M. BARANY,Phys. Chem. Phys. Med. NMR 17,23 (1985). 24. A. A. DE GRAAF,J. E. VAN DUK,AND W. M. M. J. BOV~E, Magn. Reson. Med. 13,343 ( 1990 j. 25. W. A. EDELSTEIN, G. H. GLOVER,C. J. HARDY,AND R. W. REDINGTON. Mugn. Reson. Med. 3,604 ( 1986j. 26. P. R. LUYTEN,A. J. H. M A R I ~ N AND , J. A. DEN HOLLANDER, “Society of Magnetic Resonance in Medicine, Proceedings of 7th Annual Meeting, San Francisco, 1988,” p. 327. 27. P. R. LUYTEN,J. A. DEN HOLLANDER. C. SEGEBARTH, AND D. BAL~RIAUX, “Society of Magnetic Resonance in Medicine, Proceedings of 7th Annual Meeting, San Francisco, 1988,” p. 252. 28. D. L. ROTHMAN,K. L. BEHAR,H. P. HETHERINGTON, AND R. G. SHULMAN, Proc. Nail. Acud. Sci. USA 81,6330 (1984). 29. R. J. ORDIDGE,A. CONNELLY. AND J. A. B. LOHMAN, J. Mugn. Reson. 66,283 ( 1986). 30. J. C. HASELGROVE, V. H. SUBRAMANIAN, J. J. S. LEIGH,L. GYULAI,AND B. CHANCE,Science 220, 1 I70 (1983). 31. S. H. GRAHAM,K. SHIRAISHI, S. S. PANTER,R. P. SIMON,AND A. FADEN,Neurosci. Len. 110, 124 (1990).
32. V. R. S. KOLLURIAND G. Y. C. V. S. LAKSHMI, Neurochem. Res. 14, 621 ( 1989). 33. 0. U. SCREMIN AND D. J. JENDEN,Stroke 20, 1524 ( 1989).
