High-Resolution Magic Angle Spinning - American Chemical Society

University of Kuopio, Finland, and School of Sport and Exercise Sciences, University of Birmingham, Birmingham, U.K.. The functional genomic approache...
1 downloads 0 Views 121KB Size
Anal. Chem. 2006, 78, 1546-1552

High-Resolution Magic Angle Spinning 1H NMR Spectroscopy and Reverse Transcription-PCR Analysis of Apoptosis in a Rat Glioma Julian L. Griffin,*,† Cherie Blenkiron,‡ Piia K. Valonen,§ Carlos Caldas,‡ and Risto A Kauppinen|

Department of Biochemistry and the Hutchison/MRC Research Centre, Department of Oncology, University of Cambridge, Cambridge, U.K., Biomedical NMR and National Bio-NMR Facility, A.I. Virtanen Institute for Molecular Sciences, University of Kuopio, Finland, and School of Sport and Exercise Sciences, University of Birmingham, Birmingham, U.K.

The functional genomic approaches of transcriptomics, proteomics and metabolomics aim to measure the mRNA, protein or metabolite complement of a cell, tissue or organism. In this study we have investigated the compatibility of transcriptional analysis, using Reverse Transcription (RT)-PCR, and metabolite analysis, by high-resolution magic angle spinning (HRMAS) 1H NMR spectroscopy, in BT4C rat glioma following the induction of programmed cell death. The metabolite and transcriptional changes that accompanied apoptosis were examined at 0, 4 and 8 days of ganciclovir/thymidine kinase gene therapy. Despite the high spinning speeds employed during HRMAS 1H NMR spectroscopy of one-half of the tumor samples, RT-PCR analysis of the pro-apoptotic transcripts Bcl-2, BAK-1, caspase-9 and FAS was possible, producing similar results to those detected in the unspun half of the tumors. Furthermore, the expression of FAS was inversely correlated with some of the key metabolic changes across the time period examined including the increases CHdCH and CHdCHCH2 lipid resonances which accompany apoptosis. This study demonstrates how combined transcriptomic and metabolomic studies of tumors can be used to understand the molecular events that accompany well documented metabolic perturbations during cell death processes.

analytical approaches, such as mass spectrometry, often only capable of detecting 10-20 metabolites within a tumor, these metabolic fingerprints are highly diagnostic for tumor type4-8 and response to drug treatment.9-10 One suggested reason for this is that many of the metabolites measured are major nodes in the metabolic network of a tissue and, hence, are highly sensitive to metabolic perturbations.11 However, while this makes for a sensitive tool for detecting genetic modifications or toxicological lesions, it is difficult to conclusively trace these metabolic changes back to the primary cause or the perturbed pathway. Thus, to further define the biological processes that accompany a 1H NMR spectroscopic-derived metabolic fingerprint requires the acquisition of either more global metabolic data or cross-correlation of these changes with transcriptional and proteomic changes. For metabolomics of tissues, metabolites are usually extracted into solution prior to analysis. However, extraction procedures make it difficult to directly compare lipophilic and water-soluble metabolites, and require tissue destruction, potentially limiting what analyses can be performed subsequently. High-resolution magic angle spinning (HRMAS) 1H NMR spectroscopy produces spectra from intact tissue with resolution comparable to that obtained using solution-state spectroscopy by spinning samples at the magic angle (54.7°) to the magnetic field. This process reduces the effects of dipolar coupling, chemical shift anisotropy, and magnetic susceptibility differences.12-13 In conjunction with

Applications of NMR spectroscopy, including in vivo magnetic resonance spectroscopy (MRS) and high-resolution solution-state analysis of tissue extracts, have been widely used to distinguish between different cell lines and tumor types (see ref 1 for a review of this subject). In addition, high-resolution 1H NMR spectroscopy is one of the analytical techniques currently being used to produce metabolic fingerprints of biological tissues and fluids as part of a metabolomic approach to certain biological questions.2-3 Despite the relative insensitivity of the approach compared with other

(2) Nicholson, J. K.; Lindon, J. C.; Holmes, E. Xenobiotica 1999, 29 (11), 11811189. (3) Raamsdonk, L. M.; et al. Nat. Biotechnol. 2001, 19 (1), 45-50. (4) Cheng, L. L.; Chang, I. W.; Smith, B. L.; Gonzalez, R. G. J. Magn. Reson. 1998, 135 (1), 194-202. (5) Howells, S. L.; Maxwell, R. J.; Peet, A. C.; Griffiths, J. R. Magn. Reson. Med. 1992, 28 (2), 214-236. (6) Usenius, J. P.; Tuohimetsa, S.; Vainio, P.; Ala-Korpela, M.; Hiltunen, Y.; Kauppinen, R. A. NeuroReport 1996, 7 (10), 1597-1600. (7) Tate, A. R.; Watson, D.; Eglen, S.; Arvanitis, T. N.; Thomas, E. L.; Bell, J. D. Magn. Reson. Med. 1996, 35 (6), 834-840. (8) Preul, M. C.; Caramanos, Z.; Leblanc, R.; Villemure, J. G.; Arnold, D. L. NMR Biomed. 1998, 11 (4-5), 192-200. (9) Griffiths, J. R.; McSheehy, P. M. J.; Robinson, S. P.; et al. Cancer Res. 2002, 62, 688-695. (10) Griffiths, J. R.; Stubbs, M. Adv Enzyme Regul. 2003, 43, 67-76. (11) Brindle, K. M. Biochemist 2003, 25, 15-17. (12) Cheng, L. L.; Lean, C. L.; Bogdanova, A.; Wright, S. C.; Ackerman, J. L.; Brady, T. J.; Garrido L. Magn. Reson. Med. 1996, 36, 653-658. (13) Millis, K.; Weybright, P.; Cambell, N.; Fletcher, J. A.; Fletcher, C. D.; Cory, D. G.; Singer, S. Magn Reson. Med. 1999, 41, 257-267.

* To whom correspondence should be sent. Tel.: 01223 333626. Fax: 01223 766002. E-mail: [email protected]. † Department of Biochemistry, University of Cambridge. ‡ Hutchison/MRC Research Centre, Department of Oncology, University of Cambridge. § University of Kuopio. | University of Birmingham. (1) Griffin, J. L.; Shockcor, J. P. Nat. Rev. Cancer 2004, 4 (7), 551-561.

1546 Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

10.1021/ac051418o CCC: $33.50

© 2006 American Chemical Society Published on Web 01/25/2006

pattern recognition techniques, HRMAS 1H NMR spectroscopy has been used to metabolically profile a number of diseases through direct analysis of tissues including a wide range of tumors.4,13-15 One disadvantage is that metabolites that are coresonant with lipid peaks can be obscured. Programmed cell death (PCD) can readily be induced in BT4C glioma using ganciclovir (GCV)/thymidine kinase gene therapy. We have previously shown that there is a marked increase in lipids and, in particular, polyunsaturated fatty acids (PUFAs) during PCD in these tumors.15-16 These PUFAs are relatively mobile, as determined by NMR relaxation and diffusion experiments, suggesting they are contained within cytosolic droplets.15 However, the cause of this lipid accumulation and how it interrelates with PCD are still unknown, with lipid accumulation suggested to result from either cell membrane damage and repartitioning or increased lipid synthesis.16-18 In this study, we have examined the metabolic profile of rat glioma during apoptosis, and compared metabolite perturbations with the transcription of key apoptotic genes using Reverse Transcription Polymerase Chain Reaction (RT-PCR). We demonstrate how NMR-derived metabolomics can be performed in conjunction with transcriptomics as part of a systems biology approach to understanding apoptosis. METHODS Animal Handling and Tissue Preparation. All animal experiments were performed according to the guidelines approved by the Ethical Committee of the National Laboratory Animal Centre (Kuopio, Finland). BT4C gliomas carrying the herpes simplex virus (HSV)-tk gene were induced by implanting 104 cells in 5 µL of Optimem to a depth of 2.5 mm into the corpus callosum of female BDIX rats weighing 200-270 g as described previously.19 When the tumors reached a diameter of 4-5 mm as measured using magnetic resonance imaging, GCV treatment was introduced for the duration of the study. At key time points (days 0, 4, and 8; n ) 3 for each time point), the brain of terminal pentobarbitone-anaesthetized rats were funnel frozen using liquid nitrogen, and the brain was rapidly removed. Tumors were dissected frozen on dry ice for sampling for either HRMAS 1H NMR spectroscopy followed by RT-PCR or RT-PCR alone. This approach minimized sample degradation during tissue dissection. HRMAS Spectroscopy. Tumor tissue samples were examined using a HRMAS 1H NMR probe interfaced with a 9.4-T superconducting magnet operating at a frequency of 400.1-MHz 1H NMR frequency (Bruker BioSpin GmBH, Rheinstetten, Germany). Tissue samples weighing 5-10 mg were placed into a zirconium oxide HRMAS rotor with 10 µL of D2O (deuterium lock reference) containing 10 mM 3-(trimethylsilyl)propionic acid (TSP, as chemi(14) Tomlins, A.; Foxall, P. J. D.; Lindon, J. C.; Lynch, M. J.; Spraul, M.; Everett, J.; Nicholson, J. K. Anal. Commun. 1998, 35, 113-115. (15) Griffin, J. L.; Lehtimaki, K. K.; Valonen, P. K.; Grohn, O. H.; Kettunen, M. I.; Yla-Herttuala, S.; Pitkanen, A.; Nicholson, J. K.; Kauppinen, R. A. Cancer Res. 2003, 63 (12), 3195-201. (16) Hakumaki, J. M.; Poptani, H.; Puumalainen, A.-M.; Loimas, S.; Paljarvi, L. A.; Yla-Herttuala, S.; Kauppinen R. A. Cancer Res. 1998, 58, 3791-3799. (17) Blankenberg, F. G.; Storrs, R. W.; Naumovski, L.; Goralski, T.; Spielman, D. Blood 1996, 87 (5), 1951-1956. (18) Williams, S. N.; Anthony, M. L.; Brindle, K. M. Magn. Reson. Med. 1998, 40, 411-420. (19) Poptani, H.; Puumalainen, A.-M.; Gro ¨hn, O. H. J.; et al. Cancer Gene Ther. 1998, 5, 101-109.

cal shift reference with δ ) 0.0 ppm). Spectra were acquired at 4 °C using a three-pulse NOESY presaturation sequence (Bruker Gmbh) for solvent suppression with a fixed evolution time of 6 µs and a mixing time of 150 ms. The following parameters were used: 128 transients, an acquisition time 4.09 s, a total pulse recycle delay of 6.24 s, sweep width of 10 kHz, 32K complex data points, and a spinning rate of 4 kHz. All 1H NMR spectra were manually corrected for phase and baseline distortions and referenced to TSP (δ 0.0) within ACD NMR Software manager (Advanced Chemistry Development Inc,, Toronto, Canada). For multivariate analysis, the spectra were automatically integrated using a macro program within the ACD software. This program integrated spectra across integral regions of 0.04 ppm between the chemical shift ranges of δ 0.2 and 10.0. The region δ 4.5-5.0 was set to zero integral to remove the effects of variations in the presaturation of the water resonance in all NMR spectra. Each integrated spectral region was divided by the sum of all the integrals in order to standardize samples and partially removed concentration differences between samples. RT-PCR. RNA was isolated from gliomas using TriReagent (Sigma) on homgenised tissue. From this, first strand cDNA was synthesized with Superscript II RT enzyme as per manufacturer’s instructions (Invitrogen). RT-PCR was performed using the system described by Applied Biosciences “Assays On Demand” gene expression system and used to analyze transcription levels of the proapoptotic rat transcripts BCL-2 (Rn00591516), BAK-1 (Rn00587491), caspase-9 (Rn00581212), and FAS (Rn00594913) as well as rat hexokinase 1 (Rn00562436) used as a positive control and internal standard between samples and human 18S (Hs99999901) used as a negative control. The four proapoptotic transcripts were chosen as changes in their transcriptional expression have all been correlated with the process of apoptosis, albeit in different branches of the apoptosis pathway. All transcriptional levels were normalized to hexokinase expression and then represented as a ratio to the amount detected in tumors prior to treatment (day 0). Histology. For histology, a satellite group of rats (n ) 3 for each time point) were anaesthetized by CO2 and transcardially perfused with phosphate-buffered saline for 10 min (25 mL/min) followed by 4% paraformaldehyde in 0.1 M phosphate buffer, pH 7.4, for 15 min (25 mL/min). The fixed brains were removed from the skull, rinsed in phosphate-buffered saline, and embedded in OCT compound (Bayer Corp., Emeryville, CA) for cryosectioning. Nissl staining was used to reveal the cell damage in the sections (20-µm slices; 1 section from each group of 45 was analyzed) as well as for quantitative cell count as described previously.20 Four sections (300 µm apart) from the middle of the tumor were stained for terminal deoxyribonucleotidyl transferase-mediated dUTP nick end labeling (TUNEL) (ApopTag Plus, Oncor, Gaithersburg, MD) to reveal apoptotic nuclei, using a methyl green counterstain. Apoptotic nuclei were counted in tumor tissue from arbitrarily chosen high-power microscopy fields (× 20; AX-70 microscope; Olympus, Tokyo, Japan). Statistical Analysis and Pattern Recognition. Results from univariate statistical analyses are presented as mean values ( standard error of mean, and values were compared using an (20) Gro ¨hn, O. H.; Lukkarinen, J. A.; Silvennoinen, M. J.; Pitka¨nen, A.; van Zijl, P. C.; Kauppinen, R. A. Magn. Reson. Med. 1999, 42, 268-276.

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

1547

ANOVA test within the package Matlab (Mathworks, Natick, MA). For the comparison between transcripts in spun and unspun samples, a paired repeated measures analysis of variance was performed. However, this test did not detect any significant differences between transcript expression in spun and unspun samples. To apply multivariate analysis of the NMR spectra and identify which metabolites were correlated with the transcriptional changes measured, the combined data set consisting of NMR spectra and transcription results were analyzed using partial least squares (PLS) within the SIMCA package (version 9.0, Umetrics, Umea, Sweden). This approach is a mathematical extension of principal components analysis, where a projection model is developed to predict a Y matrix (in this example, transcript data) from the X matrix (NMR data) via PLS scores of X. This approach is a generalized multiple regression method, which copes with multiple collinear X and Y variables. To ensure a predictive pattern recognition model was formed, the amount of variation in the data explained by the model (R2) was measured as a percentage for both the X and Y matrixes. In addition, Q2, a measure of model robustness, was calculated by iteratively predicting the PLS scores of individual samples. Q2 is the fraction of the total variation of the Y variables that can be predicted by a component according to cross-validation and is calculated according to

Q2 ) (1.0 - PRESS/SS)

where PRESS is the prediction error sum of squares and is the squared differences between observed Y and predicted values when observations were left out during cross-validation; SS is the residual sum of squares of the previous component. This was reported also as a percentage, where a Q2 > 0 represents a model with a predictive capability greater than chance, with a cumulative score of over 50% representing a highly robust model (SIMCA html help file). The Q2 algorithm was also used to determine how many components were fitted in a PLS model. Only components that had a positive Q2 were considered statistically robust and described by the model. In the models presented, all data are Pareto scaled, where each variable is divided by the square root of the variance for that variable. This approach increases the contribution that lowconcentration metabolites/transcripts make to the pattern recognition model compared with models produced where no scaling is used (e.g., so-called mean centering), without significantly increasing the contribution that noise makes to the model as may occur in models where univariate scaling is used (where the variance is set to unity for each variable). To further examine these correlations produced by multivariate analysis of the calculated metabolite concentrations with transcriptional changes and time, a Pearson correlation analysis was performed using Matlab (Mathworks, Natick, MA; Matlab help file on bootstrap). Because of the relatively small number of pairs of data used to calculate these correlation coefficients, the results were bootstrapped across a 1000 iterations of the data set. In this process, the distributions of the correlation coefficients are simulated by generated data sets where some pairs of data are missed out and a correlation coefficient is calculated for the remaining pairs of data. Each bootstrap sample contains n rows, 1548 Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

Figure 1. Three typical HRMAS 1H NMR spectra obtained from tumors undergoing PCD. By day 8, large increases in a number of lipid resonances were clearly detected, and in particular those associated with unsaturated resonances. Boldface type signifies proton positions associated with a given resonance. Spectra are scaled to constant noise across the series.

where n is the number of data pairs examined, chosen randomly (with replacement) from the corresponding input data set. This is repeated in an iterative manner to generate a distribution of the correlation coefficient. This procedure is used to generate distributions of the correlation coefficients giving an uncertainty distribution about the measured values. For two variables to be positively correlated, >95% of the population should give a positive score in the distribution. Similarly, for a negative correlation, >95% of the population should be less than 0. RESULTS AND DISCUSSION Figure 1 shows the metabolic changes that accompany GCV/ thymidine kinase-induced PCD in glioma across the time period investigated. Marked increases in a number of lipid resonances were detected across the time period. This was particularly pronounced for those resonances associated with unsaturated lipids including CHdCH, CHdCHCH2CHdCH, and CHdCHCH2CH2 (where the boldface type signifies the resonance observed). Contrary to this, there were relatively few changes apparent visually for choline-containing metabolites and the metabolites that produced sharp resonances in the spectra. Despite the relatively high spinning speeds used during HRMAS, RNA degradation was minor in terms of the reproducibility of the RT-PCR results between halves of tumors. The relative stability of mRNA may have also been assisted by the very rapid tissue freezing that is possible during funnel freezing. While subsequent handling required tissue thawing, even during this process the tissue was maintained at ∼4 °C. Table 1 demonstrates that there were no significant differences in the measured mRNA ratios between tumor tissue analyzed after spinning and tissue that had not been spun, indicting that HRMAS 1H NMR spectroscopy could be used as part of a combined transcriptomic and metabolomic analysis of biological tissues. While variation across the three replicates for a given transcript/time point combination was present, this probably resulted from the act of splitting tumors, which may not have provided homogeneous halves, with a similar level of variation being present in both spun and unspun samples. Because of the similarity between the transcriptional changes measured in spun and unspun tissue, data were combined to form an average transcription level for each time point regardless of

Table 1. RT-PCR Results Comparing Mean Transcription Levels ((SE) between Halves of Tumors Either Spun at 4 kHz and 4 °C for ∼15 min or Not (n ) 3) Using a Paired Analysis of Variance Testa BCL-2

BAK-1

caspase-9

FAS

day

spun

non

spun

non

spun

non

spun

non

0 4 8

1.00 ( 0.28 0.87 ( 0.08 1.08 ( 0.30

1.00 ( 0.14 1.01 ( 0.17 1.12 ( 0.58

1.00 ( 0.25 0.88 ( 0.28 0.85 ( 0.02

1.00 ( 0.06 1.07 ( 0.20 1.34 ( 0.45

1.00 ( 0.10 0.93 ( 0.07 1.17 ( 0.31

1.00 ( 0.02 1.14 ( 0.12 1.41 ( 0.24

1 ( 0.28 0.46 ( 0.22 0.38 ( 0.02

1.00 ( 0.20 0.85 ( 0.24 0.65 ( 0.20

a Expression levels were initially normalized to the housekeeping gene hexokinase 1, whose expression level is assumed to remain constant across the time course, and then represented as a ratio to the mean values found in the untreated tumors. No significant change was detected between the spun and unspun samples.

Figure 2. Summary of RT-PCR results from Table 1 including mean transcriptional levels for both halves of tumors either spun at 4 kHz and 4 °C for ∼15 min or not. Expression levels were initially normalized to the house keeping gene hexokinase 1 to correct for differences in cell numbers and then reported with respect to the mean values found in the untreated tumors (day 0). Each mean represents the combined average of expression in tumor tissue following HRMAS or in unspun tissue (n ) 6, standard errors shown). Of the four transcripts quantified across the time course, only FAS changed significantly with time (p ) 0.013 according to an ANOVA test).

tissue handling. Across the time period, only FAS expression decreased significantly across the three time points according to an ANOVA test (p ) 0.013; Figure 2). Cross-correlating metabolite profiles against the selected mRNA changes and time of treatment using PLS multivariate regression analysis demonstrated that the PUFA resonances at 5.3 (from vinyl protons) and 2.75 ppm (from bis-allylic protons) were highly correlated with the progression of PCD as measured by time of treatment (Figure 3), as well as inversely correlated with the expression of FAS. The most robust regression model was built for Pareto scaled data when only time and FAS expression was correlated with metabolite changes (R2(X) ) 57%, R2(Y) ) 80.0%, Q2 ) 52.1%; two PLS components; results from the first PLS component are shown). Indeed no multivariate model could be built between the NMR data and the expression of bcl-2, BAK-1, and caspase-9 either considered together or considered alone. Thus, the other transcripts were considered to be poorly correlated with metabolite changes across the data set (as expected given that no significant temporal correlation was calculated for any of the other transcripts). The loadings score (Figure 3b) was used to calculate which metabolite regions contributed most to the PLS regression model and, hence, were most correlated with either time or FAS expression. The most correlated metabolite regions, as determined by an arbitrary cutoff of half the maximum loading score for the first PLS component, were in ppm: 5.32 (CHdCH lipid), 3.24-3.28 (phosphatidylcho-

line), 1.28 (CH2CH2CH2 lipid), 0.88 (CH3 lipid), 2.8 (CHdCHCH2CHdCH lipid), 3.44 (taurine and myo-inositol). In terms of FAS expression, the concentrations of taurine and phosphatidylcholine were positively correlated, and CHdCH, CHdCHCH2CHdCH, CH2CH2CH2, and CH3 lipid resonances were negatively correlated. The time of progression was also correlated with cell density in the satellite animals where histology was performed (Table 2). The PLS model of NMR spectra regressed against FAS expression was alone used to predict FAS expression on a per sample basis. This model predicted FAS expression to be 0.74 ( 0.08 at day 0, 0.69 ( 0.04 at day 4, and 0.39 ( 0.04 at day 8. Using ANOVA to compare predicted and measured FAS expression, no significant difference was found. Because of the relatively small number of samples examined, we employed a bootstrapping routine to generate a distribution for the correlation coefficients between the metabolite regions identified using PLS as being correlated with PCD progression in terms of the day of treatment and FAS expression. Again the PUFA resonances (CHdCH and CHdCHCH2CHdCH resonances) were positively correlated with time of treatment and negatively correlated with FAS expression (Figure 4). In addition, significant correlations were also calculated for phosphatidylcholine (3.28 ppm), phosphocholine (3.24 ppm), and myo-inositol (3.56 and 3.88 ppm) against time of treatment (negative correlation) and FAS expression (positive correlation). Time of treatment was positively correlated with the saturated lipid resonance at 1.28 ppm, but no correlation was found with FAS expression with this resonance. In addition, time and FAS expression were found to be negatively correlated. However, no correlation was found between any of the metabolites and the transcripts bcl-2, BAK-1, and caspase-9 (i.e., for a correlation with an overall positive value, distributions had >5% of the distribution with a value of 0 or less for that correlation). From the calculated correlation coefficients and the PLS models, in terms of predicting the progression of PCD, the lipid resonances were more highly discriminatory than the transcripts that were chosen to monitor changes in apoptosis. A number of previous studies of tumor metabolism have focused on lipid metabolism, particularly during apoptosis and necrosis.16-18,21 We have previously shown in this system that using both MRS and HRMAS 1H NMR spectroscopy the concentration of PUFAs increases 3-fold during apoptosis, as determined by increased CHdCH and CHdCHCH2CHdCH resonances.15 Furthermore, by (21) Mountford, C. E.; Wright, L. C. Trends Biochem. Sci. 1988, 13, 172-177.

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

1549

Figure 3. Multivariate partial least-squares analysis of the combined data from HRMAS 1H NMR and RT-PCR analysis of tumors undergoing PCD. (A) shows a scores plot for the regression analysis, which demonstrates a linear trend between the NMR spectral profile (x axis) and the transcriptional data/time of treatment. (B) shows the corresponding loadings plot for the first PLS component. Only the most discriminatory variables are labeled on the figure. An arbitrary cutoff of half the maximum loading score for the PLS was chosen.

Table 2. Cell Density, Number of Apoptotic Cells, and Relative Tumor Volumes of BT4C Gliomas (( SEM)a treatment

day

cell density (×103 cells/mm3) (n ) 3 for each time point)

TUNEL (cells/high-power field) (n ) 3 for each time point)

tumor volume (relative) (day 0 n ) 9; day 4 n ) 6; day 8 n ) 3)

0 4 8

178.4 ( 10.4 154.7 ( 13.9 71.3 ( 1.1b

4.0 ( 0.1 11.3 ( 0.5b 14.2 ( 1.9b

1 1.36 ( 0.05 0.79 ( 0.09c

a Day 0 volume was taken to be one in each animal, and the subsequent values were referenced to this value. b Significance level of p < 0.01 for a change relative to the untreated tumor using a Student unpaired t-test. c Significance level of p < 0.01 for a change relative to the day 4 value using a Student unpaired t-test.

examining the line widths of the resonances of these lipids under different physical conditions, such as temperature and spinning rate, alongside NMR measurements of the diffusion rates of these 1550 Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

lipids, the evidence indicates that the PUFA lipids were most likely to be the constituents of cytoplasmic lipid droplets rather than being membrane associated. The mechanism of apoptosis induced by ganciclovir in HSV-1 thymidine kinase-expressing cells has recently been defined in Chinese hamster ovary cells stably transfected with HSV-1 thymidine kinase.22-23 Ganciclovir induced double-stranded breaks in DNA followed by apoptosis, although this process required one or two passages through the cell cycle, producing a time lag between exposure and apoptosis of at least 24 h in cell culture. This process was found to be largely FAS independent and involved caspase-9 activation and BCL-2 decline at the protein level, indicating the mitochondrial damage pathway was responsible for the subsequent apoptosis. In the present study, a decrease in FAS expression was found during gene therapy, indicating that this pathway played a minor role in any apoptosis induced in the (22) Tomicic, M. T.; Bey, E.; Wutzler, P.; Thust, R.; Kaina, B. Mutat. Res. 2002, 505, 1-11. (23) Tomicic, M. T.; Thust, R.; Kaina, B. Oncogene 2002, 21, 2141-2153.

Figure 4. To confirm correlations identified using PLS between NMR spectra and FAS expression and time of treatment, Pearson correlation coefficients were calculated using a bootstrapping procedure. Each distribution represents 1000 iterations. The distributions of the correlation coefficients are shown for key pairs. In addition the mean correlation coefficient (R) for each pair is shown.

glioma. Furthermore, even in FAS-dependent apoptosis induced by chemotherapy-induced oxidative stress in glioma, there is only a small increase in FAS expression,24 suggesting that this pathway is controlled by FAS receptor activation rather than being under transcriptional control. The decrease in FAS expression cannot be explained by a global decrease in mRNA content as expression values were standardized to rat hexokinase 1 expression, and the expression of the other transcripts did not change across the time course. Instead, it appears FAS mRNA was selectively targeted for degradation, which may be a response particular to glioma, as these tumors appear to be resistant to apoptosis induced by FAS ligands in general.24 There were no changes in bcl-2 or caspase-9 expression across the time course, confirming the activation of this pathway is at the proteomic level rather than under transcriptional control. Caspase-9 cleaves bcl-2 to bcl-2 p23 fragment, which induces cytochrome c release. This release stimulates caspase-9 activity producing an amplification cycle for apoptosis induction via the mitochondrial damage pathway. These observations indicate that metabolic profiling through HRMAS can be conducted alongside transcriptional analysis-based

techniques. Weckwerth and colleagues25 have previously suggested an extraction procedure suitable for transcriptomics, proteomics, and gas chromatography-mass spectrometry-based metabolomics, but our approach produces a metabolic fingerprint of the tissue without any significant sample destruction or degradation. Indeed, Waters and colleagues26 have used HRMAS 1H NMR spectroscopy prior to histology, with the latter detecting no significant damage in liver tissue. The potential for tissue damage can be further reduced by the use of pulse sequences designed for slow spinning rates during HRMAS spectroscopy.27-29 This suggests that HRMAS 1H NMR could be used as a screening tool to identify tumors with distinctive metabolic phenotypes for further analysis by approaches that are more costly on a per (24) Xia, S.; Rosen, E. M.; Laterra, J. Cancer Res. 2005, 65, 5248-5255. (25) Weckwerth, W.; Wenzel, K.; Fiehn, O. Proteomics 2004, 4 (1), 78-83. (26) Waters, N. J.; Garrod, S.; Farrant, R. D.; Haselden, J. N.; Connor, S. C.; Connelly, J.; Lindon, J. C.; Holmes, E.; Nicholson, J. K. Anal. Biochem. 2000, 282 (1), 16-23. (27) Hu, Z. J.; Wind, R. A. J. Magn. Reson. 2002, 159 (1), 92-100. (28) Wind, R. A.; Hu, J. Z.; Rommereim, D. N. Magn. Reson. Med. 2003, 50 (6), 1113-1119. (29) Burns, M. A.; Taylor, J. L.; Wu, C. L.; Zepeda, A. G.; Bielecki, A.; Cory, D.; Cheng, L. L. Magn. Reson. Med. 2005, 54 (1), 34-42.

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

1551

sample basis (e.g., using DNA microarrays) or more skilled-labor intensive (e.g., histology). In conclusion, this study demonstrates the potential for conducting NMR-based metabolomics alongside transcriptional analysis using HRMAS 1H NMR spectroscopy. Such approaches can be used to place key metabolic changes in the context of the molecular changes that accompany a process, and this may be

1552

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

particularly useful in understanding the biological processes that produce a profound increase in PUFA lipids in some tumors during apoptosis. Received for review August 8, 2005. Accepted December 12, 2005. AC051418O