Biochemical characterization of pediatric brain tumors by using in vivo and ex vivo magnetic resonance spectroscopy

J Neurosurg 96:1023 1031, 2002 Biochemical characterization of pediatric brain tumors by using in vivo and ex vivo magnetic resonance spectroscopy A. ARIA TZIKA, PH.D., LEO LING CHENG, PH.D., LILIANA GOUMNEROVA, M.D., JOSEPH R. MADSEN, M.D., DAVID ZURAKOWSKI, PH.D., LOUKAS G. ASTRAKAS, PH.D., MARIA K. ZARIFI, M.D., R. MICHAEL SCOTT, M.D., DOUGLAS C. ANTHONY, M.D., PH.D., R. GILBERTO GONZALEZ, M.D., PH.D., AND PETER MCL. BLACK, M.D., PH.D. Departments of Radiology, Neurosurgery, Pathology, and Biostatistics, Children s Hospital; Departments of Radiology and Pathology, Massachusetts General Hospital; and Harvard Medical School, Boston, Massachusetts Object. Magnetic resonance (MR) spectroscopy provides biochemical information about tumors. The authors sought to determine the relationship between in vivo and ex vivo biochemical characterization of pediatric brain tumors by using MR spectroscopy. Their hypothesis was that ex vivo MR spectroscopy provides a link between in vivo MR spectroscopy and neuropathological analysis. Methods. In vivo proton MR spectroscopy was performed before surgery in 11 patients with neuroepithelial tumors. During resection, a total of 40 tumor biopsy samples were obtained from within the volume of interest identified on in vivo MR spectroscopy and were frozen immediately in liquid nitrogen. High-Resolution Magic Angle Spinning (HRMAS) was used to perform ex vivo MR spectroscopy in these 40 tumor biopsy samples. Neuropathological analysis was performed using the same biopsy samples, and the tumors were classified as ependymoma, choroid plexus carcinoma, pineoblastoma (one each), and pilocytic astrocytoma, medullobastoma, low-grade glioma, and glioblastoma multiforme (two each). Ex vivo HRMAS MR spectroscopy improved line widths and line shapes in the spectra, compared with in vivo MR spectroscopy. Choline (Cho) detected in vivo corresponded to three different peaks ex vivo (glycerophosphocholine, phosphocholine [PCho], and Cho). Metabolite ratios from in vivo spectra correlated with ratios from ex vivo spectra (Pearson correlation coefficient range r = 0.72 0.91; p 0.01). Metabolite ratios from ex vivo spectra, such as PCho/ total creatine (tcr) and lipid/tcr, correlated with the percentage of cancerous tissue and percentage of tumor necrosis, respectively (r = 0.84; p 0.001). Conclusions. Agreement between in vivo and ex vivo MR spectroscopy indicates that ex vivo HRMAS MR spectroscopy can improve resolution of this modality and provide a link between in vivo MR spectroscopy and neuropathological analysis. KEY WORDS brain neoplasm biopsy magnetic resonance spectroscopy children P EDIATRIC brain neoplasms, like all tumors, exhibit aberrant morphology at the macroscopic and microscopic levels. Although MR imaging can be used to detect this aberrant morphology with great sensitivity, 3 histopathological analysis of biopsy samples is still needed for a definitive diagnosis. In vivo brain proton MR spectroscopy has become a clinical diagnostic tool and promises to contribute to presurgical evaluation based on the ability Abbreviations used in this paper: Cho = choline; GBM = glioblastoma multiforme; Gd = gadolinium; Glx = glutamate and/or glutamine; GPCho = glycerophosphocholine; HRMAS = High-Resolution Magic Angle Spinning; MR = magnetic resonance; NAA = N-acetylaspartate; PCho = phosphocholine; tcr = total creatine; VOI = volume of interest. it provides to discriminate between the biochemical characteristics of healthy and neoplastic tissue. 18,19 Proton MR spectra of tumors reveal a characteristic pattern that is different from that found in normal tissue and that may be different among various tumor types. 19,23,24 Nevertheless, overlapping of peaks prevents detailed biochemical analysis of tumors in vivo. A clear understanding of the spectral pattern of a tumor on MR spectroscopy requires better knowledge of the biochemical compounds contributing to the peaks of the in vivo spectra. To gain this knowledge, acquisition of increased-resolution MR spectra from brain tumors is needed. Because this resolution is currently missing from the spectra acquired in vivo, in vitro spectra have been used, 9,15,21 even though they are limited. These limitations include er- 1023

A. A. Tzika, et al. rors resulting from extraction procedures and the fact that the biopsy sample is used up during these procedures, thus preventing neuropathological analysis of the original, intact biopsy sample. 5 Ideally, MR spectroscopy of intact biopsy samples is preferred. Unfortunately, proton MR spectra of intact biopsy samples are of suboptimal quality because of increased line widths. The method of HRMAS, which offers decreased line widths, has been applied to acquire spectra with optimal quality. 6 By spinning the biopsy sample at the magic angle, residual chemical shift anisotropies and dipole dipole interactions are averaged, thus increasing resolution and sensitivity. The purpose of this investigation was to determine whether there was agreement in the biochemical characterization between in vivo and ex vivo MR spectroscopy of pediatric brain tumors. Patient Population Clinical Material and Methods Eleven children (10 boys and one girl) 1 to 14 years of age were studied. Their tumors consisted of the following: ependymoma, choroid plexus carcinoma, and pineoblastoma were found in one each, and pilocytic astrocytoma, medulloblastoma, low-grade glioma, and GBM were found in two each. These children were eligible for inclusion in this study because they had brain tumors that were neuroepithelial in origin. Approval of the study was granted by the Institutional Review Board. In Vivo MR Imaging/MR Spectroscopy The MR images and in vivo MR spectra were acquired using a 1.5-tesla clinical imager. The VOIs for MR spectroscopy were selected using T 2 -weighted MR images and T 1 -weighted MR images obtained after contrast administration. Proton MR spectra were acquired using stimulated echo acquisition mode and/or point-resolved spectroscopy with TEs 20 msec and 144 msec, respectively (a 1500-msec TR was used). The MR spectral processing was performed using commercially available spectroscopic analysis software (SAGE; General Electric Medical Systems, Milwaukee, WI), as well as software developed in house. Brain Tumor Tissue Specimens During surgery, we collected samples from within the VOI shown on the MR images, which were available in the operating room. Tissue collection was either performed using a navigation system or with intraoperative MR imaging, and when it was performed free hand, the investigators examined the preoperative MR images carefully in three dimensions until they felt comfortable that the biopsy sample to be examined ex vivo was obtained from within the volume element examined with in vivo MR spectroscopy. Forty intact tumor biopsy specimens were collected and immediately frozen in liquid nitrogen. These specimens were kept at 80 C until ex vivo HRMAS MR spectroscopy was performed. Each frozen specimen was transferred into a specially designed probe that was precooled to 5 C. The HRMAS MR spectroscopy study was started as soon as the tissue thawed. Subsequently, the specimens were refrozen in liquid nitrogen and sent for histopathological evaluation. Ex Vivo Proton HRMAS MR Spectroscopy Procedures for ex vivo proton HRMAS MR spectroscopy have been described elsewhere. 6 Typically, the studies were performed at 5 C on a model MSL400 MR spectrometer (proton frequency 400.13 Hz) coupled to a BD-MAS probe (Bruker Instruments, Inc., Billerica, MA). Temperature was controlled by a VT-1000 unit, in combination with an MAS-DB pneumatic unit (Bruker Instruments, Inc.). The specimen spinning rate was stabilized at either 2.25 or 2.3 khz for individual samples. A rotor-synchronized Carr Purcell Meibom Gill pulse sequence (90-[ -180- ] n -acquisition) functioned as a T 2 filter. The interpulse delay ( = 2 / r ) was synchronized with the rotor rotation ( indicates the specimen spinning speed in time units and r /2 represents the MAS speed in kilohertz). The value of n for each sample was adjusted according to sample spinning rates to create a T 2 filter of 2n 50 msec. The 90 pulse length, which varied from 10.5 to 12.1 sec, was also adjusted individually for each sample. The number of transients was 256, with an acquisition time of 10 msec. A TR of 3 seconds and a spectral width of 8 khz (20 ppm) was used. All free induction decays were subjected to 1 Hz of apodization before Fourier transformation and phase adjustment. Tetramethylsilane at 0 ppm was used as an external chemical shift reference, from which the internal reference of the lactate doublet was found to be 1.32 and 1.34 ppm. 5 Histopathological Studies Each specimen was thawed, fixed in 10% formalin, embedded in paraffin, cut into 8- m sections, and stained with hematoxylin and eosin. The features evaluated in each specimen included the following: 1) areas of highly malignant glioma with high cell density and mitoses; 2) areas of moderate or low cellularity but clearly diagnostic of glioma; 3) areas of necrosis; 4) areas of nearly normal brain tissue that contained normal cells and some reactive astrocytes but in which the number of infiltrating neoplastic cells was too low to establish the diagnosis of glioma; and 5) vascular structures and areas of blood clot. All biopsy specimens were evaluated by the same neuropathologist (D.C.A.), who was blinded to the MR spectroscopy results. The entire area composed of five types of tissue was measured to the nearest 5%, and the proportions of each of the five pathological features were independently calculated as a percentage of the total area of the specimen. Statistical Analysis Each in vivo and ex vivo metabolite variable was assessed for normality by using the Kolmogorov Smirnov test, 26 and no significant departures were found. Therefore, the Pearson product-moment correlation coefficient (r) was used to measure the linear association between selected metabolite variables. Least-squares regression analysis 8 was used to determine the best-fitting regression lines. Data analysis was completed using commercially available software (Version 10.0; SPSS Inc., Chicago, IL) for the person- 1024

In vivo and ex vivo MR spectroscopy in pediatric brain tumors FIG. 1. In vivo and ex vivo HRMAS proton MR spectroscopy readings obtained in a GBM. Conventional T 1 -weighted MR image with Gd administration (T1Gd), in vivo proton MR spectra (1 and 2) obtained with single-voxel MR spectroscopy, and ex vivo MR spectrum (3) in a 15-year-old boy with GBM. The ex vivo proton MR spectrum was obtained using the HRMAS method in an intact biopsy sample from within the VOI on in vivo proton MR spectroscopy. The lesion enhances on the T 1 -weighted image. The MR spectrum 1 demonstrates prominent Cho, tcr, and NAA peaks, whereas spectrum 2 exhibits additionally high lipid (L) peaks. The ex vivo MR spectrum 3 exhibits a similar pattern to spectrum 2, with Glx peaks additionally detected. ms = milliseconds. al computer. A two-tailed alpha level of 0.05 was used for all statistical tests. Results Ex vivo HRMAS MR spectroscopy yielded improved line widths and line shapes compared with in vivo MR spectroscopy (Figs. 1 5). It also allowed correlation of in vivo MR spectroscopy and neuropathological analysis. Qualitative analysis of the ex vivo spectra showed that inositol and/or glycine peaks were very prominent in highgrade tumors (Figs. 3 and 4). The tumor-specific Cho peak that was detected in vivo exhibited three different peaks, which corresponded to GPCho, PCho, and Cho in ex vivo MR spectroscopy. In certain cases, all three peaks were discernable (Figs. 3 5). Increased relative detection of Cho was observed in medulloblastomas (Fig. 3). Taurine was detected prominently in the pineoblastoma (Fig. 5). In addition, Glx resonances were detected in all highly malignant tumors. The GBM (Fig. 1), choroid plexus carcinoma (Fig. 2), and ependymoma (Fig. 4) proton MR spectra were dominated by lipid and macromolecule peaks. These were the tumors that had extensive necrosis on histopathological analysis. Linear regression analysis showed that the selected metabolite ratios derived from in vivo spectra correlated highly with the ratios from ex vivo spectra (Pearson coefficient range, r = 0.72 0.91; p 0.01, Fig. 6). Both PCho/tCr and the sum of PCho Cho/tCr ratios that were determined ex vivo correlated with the in vivo ratio of Cho/tCr. Also, PCho/NAA and the sum of PCho Cho/NAA ratios that were determined ex vivo correlated with the in vivo ratio of Cho/NAA. Based on histopathological analysis, choroid plexus carcinoma, medulloblastomas, and pineoblastoma were characterized primarily by extensive areas of high cellularity. The GBM specimens revealed both areas of high cellularity and areas of necrosis. Both pilocytic and the low-grade astrocytomas were characterized by areas of low cellularity that were nevertheless clearly diagnostic of glioma. The relationship between selected metabolites in the biopsy specimens detected on ex vivo MR spectroscopy and histopathological studies is presented also (Fig. 7). In the seven patients with high-grade tumors (one each: ependymo- 1025

A. A. Tzika, et al. FIG. 2. In vivo and ex vivo HRMAS proton MR spectroscopy readings obtained in a choroid plexus carcinoma. Conventional T 2 -weighted (T2W) and Gd-enhanced MR images, in vivo proton MR spectra (1 and 2) obtained by singlevoxel MR spectroscopy, and ex vivo MR spectrum (3) in a 1-year-old girl with newly diagnosed choroid plexus carcinoma. The ex vivo proton MR spectrum was obtained using the HRMAS method in an intact biopsy sample from within the VOI on in vivo proton MR spectroscopy. The lesion enhances on T 1 -weighted image. The MR spectrum 1 demonstrates prominent Cho, NAA, Glx, and lipid peaks, whereas spectrum 2 exhibits only high Cho. The figure illustrates that spectrum 3 exhibits a similar spectral pattern to the short-echo, in vivo, spectrum 1. ma, choroid plexus carcinoma, pineoblastoma; two each: medulloblastoma, and GBM) the linear relationship of the PCho/tCr ratio and the amount of malignant glioma was significant (r = 0.8392; p 0.001). When all 11 patients in the study, together with the patients with low-grade tumors, were included in the linear regression analysis, the relationship between the PCho/tCr ratio and the amount of malignant glioma was also significant, but with a smaller correlation coefficient (r = 0.37; p 0.001). Finally, a significant relationship was observed between the lipid/tcr ratio and the percentage of tumor necrosis (r = 0.8487; p 0.001). Discussion Significant correlations between in vivo and ex vivo MR spectroscopy indicate that the latter can be used to provide a link between clinical MR spectroscopy and histopathological analysis of biopsy samples. The HRMAS method can be used to improve resolution of MR spectra, and it offers certain advantages over other high-resolution studies. Procedures involving chemical extraction or tissue homogenization destroy the specimen and do not allow histopathological analysis of the same material. Because the HRMAS method does not disrupt the tissue, higher levels of Cho have been found with this modality compared with others requiring chemical extraction. 6 Despite these concerns, histopathological evaluation of tissues before and after HRMAS MR spectroscopy has shown that both macroscopic and microscopic integrity of the tissue is for the most part maintained. 4 We have used multiple ex vivo samples in our correlations of in vivo and ex vivo MR spectroscopy, thus to some extent accounting for tumor heterogeneity, which, especially in ependymomas or GBMs, may become a serious source of error because of extensive areas of necrosis intermingling with areas of low and high cellularity. Our sampling of heterogeneous tumors has presented us with the opportunity to evaluate correlations of quantitative ex vivo MR spectroscopy and quantitative neuropathological studies. In this report, although the number of patients is small, the use of multiple ex vivo samples resulted in good correlation between the in vivo and ex vivo findings (Pearson coefficient range, r = 0.72 0.91; p 0.01, Fig. 6). 1026

In vivo and ex vivo MR spectroscopy in pediatric brain tumors FIG. 3. In vivo and ex vivo HRMAS proton MR spectroscopy readings obtained in medulloblastoma. Conventional T 2 -weighted MR image, in vivo proton MR spectrum obtained by single-voxel MR spectroscopy, and ex vivo proton MR spectrum were performed in a 10-year-old girl with medulloblastoma. Ex vivo proton MR spectroscopy was performed using the HRMAS method in an intact biopsy sample from within the VOI on in vivo proton MR spectroscopy. The in vivo and ex vivo MR spectroscopy spectral patterns are similar. The ex vivo HRMAS spectrum has higher resolution, however, and thus the Cho-containing compounds (from left to right, GPCho, PCho, and free Cho) that give rise to the Cho peak in vivo are distinguishable from one another. A prominent tcr peak was detected; NAA was not detected. Lipids were not detected and the prominent peak at 1.33 ppm is lactate, which is usually detected in biopsy samples. Inl = inositol. In our study, ex vivo HRMAS MR spectroscopy provided improved line widths and line shapes compared with in vivo MR spectroscopy (Figs. 1 5), and indicated a good correlation between in vivo MR spectroscopy and neuropathological analysis. Meanwhile, the general pattern of the ex vivo spectra compares well with the in vivo MR spectra. The tumor-specific Cho peak that was detected in vivo exhibited at least three different peaks, which corresponded to GPCho, PCho, and Cho ex vivo; in certain cases taurine was also detected (Fig. 5). The PCho, levels of which have been shown to correlate with the number of cells in the S phase, 20 and depletion of which corresponds to growth arrest, 11 appears to be the primary Cho-containing metabolite detected ex vivo that gives rise to the Cho peak detected in vivo (Figs. 3 5). Our linear regression analysis showed that Cho/tCr and Cho/NAA ratios derived from in vivo spectra correlated highly with the ex vivo PCho/tCr and PCho/NAA ratios (r = 0.87 and r = 0.91, respectively; p 0.001). Furthermore, both PCho/tCr and the sum of PCho Cho/tCr ratios determined ex vivo correlated with the in vivo ratio of Cho/tCr (Fig. 6 upper) and thus indicat- ed that the in vivo Cho peak in our tumor samples reflected the PCho peak detected ex vivo. This substantiates the results of animal studies. 11,20 An increase in the Cho peak exhibited by the MR spectra in our study supports the hypothesis that elevated mobile precursors participate in cell membrane turnover during rapid cell growth and neoplasia. 16 The peak is found to consist of water-soluble Cho-containing compounds such as PCho and GPCho, and free Cho. 2 These chemically distinct compounds, which are elevated in tumors and rapidly proliferating tissues, 11 and differentially accumulated in the early stages of growth arrest or apoptosis, 20 were detected on ex vivo MR spectroscopy, especially in certain cases in our study (Figs. 3 5). Furthermore, the significant relationship between the PCho/tCr ratio and the amount of cancerous tissue (percent malignant glioma, which includes both moderately and highly cellular areas) indicated a strong correlation between an ex vivo MR spectroscopy tumor-specific biochemical index and histopathological findings (Fig. 7 left). Because Cho/tCr detected in vivo correlated highly with the ex vivo PCho/tCr, it follows that the in vivo MR 1027

A. A. Tzika, et al. FIG. 4. In vivo and ex vivo HRMAS proton MR spectroscopy readings obtained in ependymoma. Conventional T 2 - weighted and Gd-enhanced MR images, in vivo proton MR spectrum (upper) obtained by single-voxel MR spectroscopy, and ex vivo MR spectrum (lower) in a 1-year-old boy with newly diagnosed ependymoma. The ex vivo proton MR spectrum was obtained using the HRMAS method in an intact biopsy sample from within the VOI on in vivo proton MR spectroscopy. The lesion enhances on the T 1 -weighted image. The figure illustrates that the ex vivo MR spectrum (lower) exhibits a similar spectral pattern to the short-echo, in vivo MR spectrum (upper). In addition, the Cho-containing compounds peak, which was detected as a single peak on in vivo MR spectroscopy, is seen as more than one peak on the ex vivo MR spectrum. Inls = inositols. spectroscopy tumor-specific marker Cho/tCr also correlates highly with histopathological findings. We have therefore shown in this study that ex vivo MR spectroscopy may be used as a link between in vivo MR spectroscopy and histopathological findings. In vivo and ex vivo spectra of several tumors in our study exhibited substantial levels of lipids, which may be due to apoptosis and/or necrosis. For instance, MR spectra from ependymomas and GBMs were dominated by lipid and macromolecule peaks (Figs. 1 and 4). In general, cancer cells are typically apoptotic and they die of apoptosis, and also after conventional chemotherapy, radiation, 22 and as a result of experimental approaches such as antiangiogenic 14 and ganciclovir treatments. 10,25 Recently, gene therapy induced apoptosis of experimental gliomas has been shown to be associated with substantial accumulation of polyunsaturated fatty acids, as detected on proton MR spectroscopy in vivo. 12 Moreover, in our study, when we tested the correlation between lipids and the percentage of necrosis in our specimens, the result was significant, indicating that the histological feature of necrosis and/or apoptosis is reflected by the presence of lipid signals in our ex vivo MR spectra. The differences in visibility of compounds detected on proton MR spectroscopy in vivo compared with ex vivo have been minimized by performing an expedited ex vivo analysis in cold temperatures. These conditions have minimized tissue degradation and prevented lactate accumulation. Further investigation is needed to clarify the effects that tissue freezing has on membrane damage and intra- and intermolecular mobility, both of which affect visibility of metabolites on MR spectra. Metabolite breakdown has been shown to be increased by time and freezing. 17 Therefore, we recognize that tissue freezing in our study may have influenced the visibility of compounds detected on ex vivo proton MR spectroscopy, and may also have attenuated the correlations between the metabolite ratios. Nevertheless, this is not a serious concern in our study because this effect would only have made our correlations weaker. We have not yet found an alternative to performing HRMAS MR spectroscopy immediately after surgical removal of the specimen in a cooled probe. Indeed, this will be the preferred procedure if HRMAS MR spectroscopy becomes a clinical tool. Whether it becomes a clinical tool will depend on the po- 1028

In vivo and ex vivo MR spectroscopy in pediatric brain tumors FIG. 5. In vivo and ex vivo HRMAS proton MR spectroscopy readings obtained in pineoblastoma. Conventional T 2 - weighted MR image, in vivo proton MR spectra obtained by single-voxel MR spectroscopy (1 and 2) and ex vivo proton MR spectrum (3) in a 4-year-old boy with pineoblastoma. Ex vivo proton MR spectroscopy was performed using the HRMAS method in an intact biopsy sample from within the VOI on in vivo proton MR spectroscopy. The short-echo, in vivo and ex vivo MR spectral patterns are similar (1 and 3). The ex vivo spectrum has higher resolution, however, and thus the Cho-containing compounds (from left to right, GPCho, PCho, and free Cho) that give rise to the Cho peak in vivo are distinguishable from one another. A prominent taurine (Tau) peak was also detected. The peaks detected to the left of putative N-acetyl compounds are probably due to Glx. Lipids are not as prominently detected, and the prominent peak at 1.33 ppm is lactate (Lac), which is usually detected in biopsy samples. tential for HRMAS MR spectroscopy to provide accurate tumor diagnosis information beyond what is found on in vivo MR spectroscopy. By providing a more detailed biochemical characterization of tumors in addition to morphological findings, as assessed on histopathological studies, HRMAS MR spectroscopy may assist in tumor diagnosis intraoperatively. Postoperatively and during treatment, when the distinction between tumor recurrence and necrosis is clinically important, HRMAS MR spectroscopy of stereotactic biopsy material might complement histopathological studies. Meanwhile, high-resolution in vivo MR spectroscopy (with excellent data from voxels of approximately 0.2 cm 3 ) combined with regional cerebral blood volume images, correlates well with histological findings, and this correlation has led to the suggestion that in vivo MR spectroscopy may be used to distinguish high- from lowgrade tumors. 7,13 This notion should not detract, however, from the importance of our line of research, which identifies ex vivo MR spectroscopy as a possible link between in vivo MR spectroscopy and neuropathological findings. In fact, it might provide additional supporting evidence that in vivo MR spectroscopy is a diagnostic and prognostic tool of clinical importance. After all, prognosis can only be attained using an in vivo tool, although diagnosis may rely solely on typing of tissue excised during surgery, inasmuch as all operable tumors will undergo resection. In this regard, ex vivo MR spectroscopy combined with histopathological studies might indeed be clinically useful; however, its clinical utility warrants further proof. In this report, the limited number of pediatric tumors in each histological category did not allow us to determine the potential of ex vivo HRMAS MR spectroscopy in tumor diagnosis; however, this should not detract from the importance of those findings. This potential has been demonstrated previously in adults. 5 Nevertheless, what is true in adult tumors does not necessarily hold in pediatric lesions. For instance, tumors in children are more aggressive than in adults, but often long-term disease control is possible. 1 We believe that use of ex vivo HRMAS MR spectra and pattern recognition techniques, in addition to estimation of levels 1029

A. A. Tzika, et al. Conclusions In this study we have demonstrated the feasibility of performing in vivo and ex vivo proton MR spectroscopy, as well as neuropathological analysis of the same samples, which were obtained in pediatric brain tumors. Using this approach, we have tested the hypothesis that ex vivo HRMAS MR spectroscopy can improve resolution of this modality and provide a link between in vivo MR spectroscopy and neuropathological analysis. A direct future implication of this study that warrants further investigation is that higher-grade tumors may be distinguished from lowergrade ones. Therefore, we suggest that this tool will allow surgeons to make an important distinction among inoperable pediatric brain tumors. FIG. 6. Graphs showing relationships between selected metabolites detected by in vivo proton MR spectroscopy and ex vivo, HRMAS, proton MR spectroscopy. The fitted regression line, Pearson correlation coefficient (r), and probability values are shown. of metabolites, may enhance the potential of HRMAS MR spectroscopy for tumor diagnosis. References 1. Albright AL: Pediatric brain tumors. CA Cancer J Clin 43: 272 288, 1993 2. Barker PB, Breiter SN, Soher BJ, et al: Quantitative proton spectroscopy of canine brain: in vivo and in vitro correlations. Magn Reson Med 32:157 163, 1994 3. Barkovich AJ: Neuroimaging of pediatric brain tumors. Neurosurg Clin N Am 3:739 769, 1992 4. Cheng LL, Anthony DC, Comite AR, et al: Quantification of microheterogeneity in glioblastoma multiforme with ex vivo highresolution magic-angle spinning (HRMAS) proton magnetic resonance spectroscopy. Neurooncol 2:87 95, 2000 5. Cheng LL, Chang IW, Louis DN, et al: Correlation of high-resolution magic angle spinning proton magnetic resonance spectroscopy with histopathology of intact human brain tumor specimens. Cancer Res 58:1825 1832, 1998 6. Cheng LL, Ma MJ, Becerra L, et al: Quantitative neuropathology by high resolution magic angle spinning proton magnetic resonance spectroscopy. Proc Natl Acad Sci USA 94:6408 6413, 1997 7. Dowling C, Bollen AW, Noworolski SM, et al: Preoperative proton MR spectroscopic imaging of brain tumors: correlation with histopathologic analysis of resection specimens. AJNR 22: 604 612, 2001 8. Draper NR, Smith H: Applied Regression Analysis, ed 3. New York: John Wiley & Sons, 1998 9. Florian CL, Preece NE, Bhakoo KK, et al: Characteristic metabolic profiles revealed by 1H NMR spectroscopy for three types of human brain and nervous system tumors. NMR Biomed 8: 253 264, 1995 10. Freeman SM, Abboud CN, Whartenby KA, et al: The bystander effect : tumor regression when a fraction of the tumor mass is genetically modified. Cancer Res 53:5274 5283, 1993 11. Gillies RJ, Barry JA, Ross BD: In vitro and in vivo 13C and 31P NMR analyses of phosphocho metabolism in rat glioma cells. Magn Reson Med 32:310 318, 1994 FIG. 7. Graphs showing relationships between selected metabolites detected by ex vivo proton MR spectroscopy and histopathological features. The Pcho/tCr ratio compared with the percentage of highly cellular malignant glioma (left) and the Lipids/tCr ratio compared with the percentage of necrosis (right) are plotted. The fitted regression line, Pearson correlation coefficient (r), and probability values are shown. 1030

In vivo and ex vivo MR spectroscopy in pediatric brain tumors 12. Hakumaki JM, Poptani H, Sandmair AM, et al: 1H MRS detects polyunsaturated fatty acid accumulation during gene therapy of glioma: implications for the in vivo detection of apoptosis. Nat Med 5:1323 1327, 1999 13. Henry RG, Vigneron DB, Fischbein NJ, et al: Comparison of relative cerebral blood volume and proton spectroscopy in patients with treated gliomas. AJNR 21:357 366, 2000 14. Holmgren L, O Reilly MS, Folkman J: Dormancy of micrometastases: balanced proliferation and apoptosis in the presence of angiogenesis suppression. Nat Med 1:149 153, 1995 15. Kinoshita Y, Kajiwara H, Yokota A, et al: Proton magnetic resonance spectroscopy of astrocytic tumors: an in vitro study. Neurol Med Chir 33:350 359, 1993 16. Kugel H, Heindel W, Ernestus RI, et al: Human brain tumors: spectral patterns detected with localized H-1 MR spectroscopy. Radiology 183:701 709, 1992 17. Middleton DA, Bradley DP, Conner SC, et al: The effect of sample freezing on proton magic-angle spinning NMR spectra of biological tissue. Magn Reson Med 40:166 169, 1998 18. Nelson SJ, Vigneron DB, Star-Lack J, et al: High spatial resolution and speed in MRSI. NMR Biomed 10:411 422, 1997 19. Preul MC, Caramanos Z, Collins DL, et al: Accurate, noninvasive diagnosis of human brain tumors by using proton magnetic resonance spectroscopy. Nat Med 2:323 325, 1996 20. Smith TA, Eccles S, Ormerod MG, et al: The phosphocholine and glycerophosphocholine content of an oestrogen-sensitive rat mammary tumor correlates strongly with growth rate. Br J Cancer 64:821 826, 1991 21. Sutton LN, Wehrli SL, Gennarelli L, et al: High-resolution 1Hmagnetic resonance spectroscopy of pediatric posterior fossa tumors in vitro. J Neurosurg 81:443 448, 1994 22. Thompson CB: Apoptosis in the pathogenesis and treatment of disease. Science 267:1456 1462, 1995 23. Tzika AA, Vajapeyam S, Barnes PD: Multivoxel proton MR spectroscopy and hemodynamic MR imaging of childhood brain tumors: preliminary observations. AJNR 18:203 218, 1997 24. Wald LL, Nelson SJ, Day MR, et al: Serial proton magnetic resonance spectroscopy imaging of glioblastoma multiforme after brachytherapy. J Neurosurg 87:525 534, 1997 25. Wei SJ, Chao Y, Hung YM, et al: S- and G2-phase cell cycle arrests and apoptosis induced by ganciclovir in murine melanoma cells transduced with herpes simplex virus thymidine kinase. Exp Cell Res 241:66 75, 1998 26. Zar JH: Biostatistical Analysis, ed 3. Englewood Cliffs, NJ: Prentice-Hall, 1996 Manuscript received April 23, 2001. Accepted in final form December 28, 2001. Support was provided by American Cancer Society Grant No. RPG-98-056-01-CCE, and in part by a Public Health Service/National Institutes of Health Grant No. CA77727. Address reprint requests to: A. Aria Tzika, Ph.D., Nuclear Magnetic Resonance Surgical Laboratory, Massachussetts General Hospital, Harvard Medical School, and Shriners Burns Institute, 51 Blossom Street, Room 261, Boston, Massachusetts 02114. email: atzika @partners.org. 1031