Pacific Symposium on Biocomputing 12:181-192(2007) COMPARATIVE METABOLOMICS OF BREAST CANCER* CHEN YANG, ADAM D. RICHARDSON, JEFFREY W. SMITH, and ANDREI OSTERMAN The Burnham Institute for Medical Research, La Jolla, California 92037, USA Comparative metabolic profiling of cancerous and normal cells improves our understanding of the fundamental mechanisms of tumorigenesis and opens new opportunities in target and drug discovery. Here we report a novel methodology of comparative metabolome analysis integrating the information about both metabolite pools and fluxes associated with a large number of key metabolic pathways in model cancer and normal cell lines. The data were acquired using [U-13C]glucose labeling followed by two-dimensional NMR and GC-MS techniques and analyzed using isotopomer modeling approach. Significant differences revealed between breast cancer and normal human mammary epithelial cell lines are consistent with previously reported phenomena such as upregulation of fatty acid synthesis. Additional changes established for the first time in this study expand a remarkable picture of global metabolic rewiring associated with tumorigenesis and point to new potential diagnostic and therapeutic targets. 1. Introduction Since the completion the human genome, the main thrust of functional genomics has been to establish the link between gene/protein expression profiles and cellular phenotype in normal and disease states, most notably in cancer. Remarkable advancements in transcriptomics and proteomics technologies (1,2) led to the identification of novel therapeutic targets as well as tumor subtypes and biomarkers (3-6). Nevertheless, these technologies, taken alone, fall short of reflecting the entire picture of cellular networks and pathways. A direct assessment of a large number of intermediary metabolites and metabolic activities (metabolomics) is emerging as a powerful complementary approach for identifying pathways that are perturbed in a given pathology. Metabolic profiling is of special importance in cancer biology due to profound changes in central metabolism associated with many tumors as established by early biochemical studies (7) and recently confirmed by functional genomics techniques. Nevertheless, our current knowledge of the molecular processes associated with these metabolic changes is quite incomplete. * This work was partially supported by a California Breast Cancer Research Program Fellowship to CY and by the Burnham Institute Cancer Center Support Grant (2 P30 CA30199-24, PI R. Abraham). Both authors contributed equally to this work. Pacific Symposium on Biocomputing 12:181-192(2007) Despite the recent progress in metabolomics technology its applications in the field of human biology are still limited, mostly due to many technical challenges. Among the most obvious are: a limited availability of biological material, insufficient sensitivity and resolution of existing protocols, incomplete reference data (eg for NMR peak assignment) and the lack of established computational modeling framework. In this study we addressed some of these problems by combining [U-13C]-glucose labeling with two-dimensional NMR and GC-MS techniques to assess simultaneously the metabolite pools and fluxes associated with several interrelated metabolic pathways in human cells. We were able to assign two-dimensional NMR signals for 24 intermediary metabolites representing a substantial fraction of central metabolism. The acquired data were analyzed using an isotopomer model derived from reconstruction of an extensive metabolic network. We applied this approach for the comparative analysis of breast cancer and normal human mammary epithelial cell lines. An isotopomer model was developed for a metabolic network including the reactions of central carbon, fatty acid, and amino acid metabolism. We chose this metabolic network because it is the central backbone of metabolism providing energy, cofactor regeneration, and building blocks for cell synthesis. Moreover, cancer cells have been reported to display different activities of some of these pathways. We determined the active pathways and the flux distribution in this metabolic network. The observed pattern of metabolic changes is consistent with earlier observations of metabolic shifts in tumors (7), validating the developed methodology. A number of newly established changes in metabolic fluxes and pools provided us with new insights to potential diagnostic markers and therapeutic targets. 2. Materials and Methods 2.1 Experimental techniques Experimental procedures are only briefly introduced in this subsection. The details are provided in the Supplementary on-line Materials (SOM) 1. Cell lines and cultivation. Human cell lines used in this study were: MCF10A (ATCC), derived from normal mammary epithelial cells, and MDA-MB435 (NCI), a highly metastatic mammary epithelial cancer cell line. Cultivation, metabolic labeling with 20% [U-13C]glucose and harvesting was performed as described in SOM. 1 available at: http://www.burnham.org/labs/osterman/ Pacific Symposium on Biocomputing 12:181-192(2007) Gas Chromatography ­ Mass Spectrometry (GC-MS) was used to analyze fatty acids. Samples from ~ 5×107 cells were prepared as described (8) and analyzed on Trace GC/Trace MS Plus system (see SOM for details). Mass isotopomer distribution corrected for natural abundance (9), was used to assess de novo fatty acid synthesis as described (8). Nuclear Magnetic Resonance (NMR) was used to analyze a mixture of methanol- and water-soluble metabolites extracted from ~ 2.5×108 labeled cells as described in SOM. Two-dimensional [13C,1H] HSQC spectra were acquired using a Bruker Avance 500 NMR spectrometer. The 13C-13C scalar coupling fine structures were extracted from the cross sections taken along the 13C axis in a HSQC spectrum by using the Bruker XWINNMR software. The concentrations of metabolites were determined by integrating the cross peaks in the HSQC spectra using the NMRPipe (10) and Sparky (http://www.cgl.ucsf.edu/home/sparky/) software packages, comparing with the integral of resonances peaks of the L-methionine that was treated as an internal standard, and normalizing to the amount of total cellular protein. C- Isotopomer Model A mathematical model describing the 13C isotopomer distribution of metabolites in human cells fed with 13C-glucose was developed. It was used for the determination of metabolic fluxes using 13C-multiplet patterns of metabolites from HSQC spectra. The considered metabolic network included glycolysis, pentose phosphate pathway (PPP), tricarboxylic acid (TCA) cycle, anaplerotic reaction, and biosynthetic pathways of fatty acids and non-essential amino acids. The fraction of de novo synthesis of fatty acids was determined based on the mass isotopomer distribution measured by GC-MS as described (8). The fluxes through other pathways are derived as follows. Contribution of PPP to pyruvate/alanine formation. The assessment of PPP activity relies on the analysis of 13C multiplets of alanine C2. The observed relative multiplet intensities were transformed to the relative abundances of intact carbon fragments (11). According to the carbon rearrangements in PPP, three pentose molecules (C1-C2-C3-C4-C5 backbone) yield five pyruvate molecules. Three of them retain an intact C1-C2-C3 fragment, while two molecules carry only a C2-C3 fragment of the original backbone. The latter fraction denoted as f(2)(Ala-C2), is the main contributor to the isotopomer population of [2,3-13C2]alanine as assessed by the relative intensity of the doublet (1JCC = 35 Hz) of alanine C2. The total fraction of pyruvate derived from PPP can be estimated as 5/2 * f(2)(Ala-C2). TCA cycle and anaplerotic flux. [U-13C]pyruvate enters the TCA cycle either by pyruvate dehydrogenase oxidation or by the anaplerotic reaction of 2.2. 13 Pacific Symposium on Biocomputing 12:181-192(2007) pyruvate carboxylase. The first process generates [4,5-13C2]-ketoglutarate via [1,2-13C2]acetyl-CoA. Since intracellular -ketoglutarate concentration is too low to be detected by NMR, its labeling state was assessed via glutamate, an abundant metabolite in rapid exchange with -ketoglutarate. The isotopomer population of [4,5-13C2]glutamate reflects the flux through pyruvate dehydrogenase, which equals the TCA cycle (citrate synthase) flux, provided the acetyl-CoA synthetase flux is zero. The second process is expected to yield a distinct labeling pattern represented by [1,2,3-13C3] and [2,3-13C2] glutamate. This pattern reflects the formation of [1,2,3-13C3] and [2,3,4-13C3] oxaloacetate due to the pyruvate carboxylase reaction followed by the reversible interconversion between the asymmetric oxaloacetate and symmetric succinate (or fumarate). The relative activity of pyruvate carboxylase versus pyruvate dehydrogenase (vPC / vPDH) was calculated from the 13C multiplet components of glutamate at C3 and C4 using Eq. 1: v PC d (Glu - C3) = v PDH d * (Glu - C4) + q (Glu - C4) (1) where d(Glu-C3) is the contribution of doublet (1JCC = 35 Hz) to the multiplets of glutamate C3, while d*(Glu-C4) and q(Glu-C4) are the relative contributions of the doublet with 1JCC coupling constant of 55 Hz and the quartet to the glutamate C4 multiplets, respectively. Non-essential amino acid biosynthesis. We investigated the activities of biosynthetic pathways of cysteine, glutamate, glutamine, glycine, and proline. The equation used for glycine biosynthesis is derived as follows, similarly for the other amino acids. Glycine can be synthesized via serine from the glycolytic intermediate, 3-phosphoglycerate, or obtained directly from media components. 3-Phosphoglycerate has the same labeling pattern as pyruvate. Thus we obtain the isotopomer balance equation (2). s PGly - C2 = X syn PAla - C2 d Gly - C2 s + d + (1 - X syn ) Pn d * + q Ala - C2 1 - Pn Pn (2) where Xsyn is the fraction of glycine derived from glycolysis; s, d, d*, and q correspond to the relative intensities of singlet, doublet, doublet with a larger coupling constant and quartet, respectively. Pn is the natural 13C abundance (Pn = 0.011), and PGly-C2 and PAla-C2 are the specific enrichments of glycine C2 and alanine C2. PGly-C2 can be calculated from Xsyn, Pn and PAla-C2 using the relation PGly-C2 = Xsyn · PAla-C2 + (1 - Xsyn) · Pn. Therefore Xsyn can be derived from the analysis of 13C multiplets of alanine C2 and glycine C2 using Eq. 3: Pacific Symposium on Biocomputing 12:181-192(2007) X syn = PAla - C2 1 - Pn s Pn ( - ) d Gly - C2 Pn 1 - Pn s s s + d - ) + Pn ( - ( ) d Gly - C2 Pn d * + q Ala - C2 d Gly - C2 (3) 3. Results 3.1. NMR Spectral Assignment Fig. 1 shows a typical two-dimensional [13C,1H] HSQC spectrum of metabolites extracted from the human breast cancer cells. The assignment of 13 C-1H cross peaks for various metabolites was made by comparing the carbon and proton chemical shifts with literature values (12-17), with spectra of pure compounds and by spiking the samples. Overall, 24 metabolites could be unambiguously assigned. The details of peak assignments and the reference summary Table S1 of characteristic chemical shifts are provided in SOM. Fig. 1. A typical two-dimensional [13C, 1H] HSQC spectrum of the metabolites extracted from breast cancer cells. Abbreviations for the assigned peaks are as in Table S1. 3.2. Metabolic Fluxes A comparison of relative intensities of 13C-13C scalar coupling multiplet components of various metabolites extracted from [U-13C]glucose labeled MCF- Pacific Symposium on Biocomputing 12:181-192(2007) 10A and MDA-MB-435 cells are shown in Table 1. These data were used in the 13 C isotopomer model to determine the metabolic fluxes or flux ratios through individual pathways including glycolysis, PPP, TCA cycle and anaplerotic reaction, fatty acid and amino acid biosynthetic pathways (Fig. 2). Table 1. Relative intensities of 13C multiplet components of metabolites extracted from MCF-10A and MDA-MB-435 cells grown on [U-13C]glucose a Carbon position Alanine-C2 Isotopomer populations 2-13C 2,3-13C2 1,2-13C2 1,2,3-13C3 3-13C 2,3-13C2 3-13C 2,3-13C2 2-13C Multiplet components s d d* q s d s d s MCF-10A 0.27 0.01 0.01 0.71 0.28 0.72 0.16 0.84 0.29 MDA-MB435 0.16 0.11 0.01 0.72 0.17 0.83 0.20 0.80 0.14 Alanine-C3 Lactate-C3 Acetyl-CoA (GlcNAc/GalNAc)-C2 Glutamine-C4 d 0.71 0.86 1,2-13C2 s 0.50 -b 4-13C 3,4-13C2 d 0.01 -b 4,5-13C2 d* 0.48 -b 3,4,5-13C3 q 0.01 -b Glutamate-C3 3-13C s 0.73 0.72 2,3-13C2 / 3,4-13C2 d 0.27 0.27 13 t 0 0.01 2,3,4- C3 s 0.50 0.30 Glutamate-C4 4-13C d 0.01 0.01 3,4-13C2 d* 0.48 0.66 4,5-13C2 q 0.01 0.03 3,4,5-13C3 s 0.67 0.71 Glu (GSH)-C3 3-13C d 0.32 0.28 2,3-13C2 / 3,4-13C2 13 t 0.01 0.01 2,3,4- C3 s 0.24 0.13 Glu (GSH)-C4 4-13C 3,4-13C2 d 0.02 0.02 d* 0.70 0.73 4,5-13C2 q 0.04 0.12 3,4,5-13C3 s 0.88 0.27 Gly (GSH)-C2 2-13C d 0.12 0.73 1,2-13C2 s 0.86 0.27 Glycine-C2 2-13C 1,2-13C2 d 0.14 0.73 s 1.00 0.25 Proline-C4 4-13C d 0.00 0.71 4,5-13C2 t 0.00 0.04 3,4,5-13C3 s 1.00 0.25 Proline-C5 5-13C 13 d 0.00 0.75 4,5- C2 a s, singlet; d, doublet (1Jcc, ~35 Hz); d*, doublet split by a large coupling constant (1Jcc, ~60 Hz); t, triplet; q, quartet. b Resonance of glutamine C4 is below the detection level in the MDA-MB-435 cells. The relative activity of PPP versus glycolysis was determined based on the analysis of 13C multiplets of alanine C2 as described above. The contribution of Pacific Symposium on Biocomputing 12:181-192(2007) the signature doublet (1JCC = 35 Hz) to the multiplets of alanine C2 is very small in MCF-10A but significant in MDA-MB-435 cells (Table 1), suggesting that a relative contribution of PPP to production of pyruvate is substantially higher in malignant cells (28%) compared to nonmalignant cells (~2%), where the bulk of pyruvate stems from glycolysis (Fig. 2). The increased use of PPP enables the MDA-MB-435 cells not only to supply more ribose for nucleic acid synthesis, but to recruit more of the NADPH reducing power for fatty acid synthesis. Indeed, the GC/MS analysis performed in this study revealed that 47% of palmitate is newly synthesized from glucose in MDA-MB-435 cells (Fig. 2) in correlation with the observed increase in PPP flux. The de novo synthesized palmitoleate, stearate, and oleate is 37%, 35%, and 18%, respectively. This is in marked contrast with almost no de novo fatty acid synthesis in MCF-10A cells as evidenced by the lack of 13C tracer accumulation in palmitate, palmitoleate, stearate or oleate. 100 MCF-10A MDA-MB-435 80 Per cent of total (%) 60 40 20 0 Py r from PP Fatt y acid Contribut ion Gly Pro pathway s y nt hes iz ed of s y nt hes iz ed s y nt hes iz ed from glucose anapleros is from glucose from glucose to TCA cycle Fig. 2. Metabolic fluxes in MCF-10A and MDA-MB-435 cells (mean + s.d.; n=4). The relative fluxes through pyruvate carboxylase and pyruvate dehydrogenase were estimated from the analysis of glutamate labeling. The major isotopomer populations of 4,5-13C2 of glutamate and -glutamyl of glutathione indicated that these carbon atoms are derived from [1,2-13C2]acetylCoA (Table 1). The isotopomer ratio of acetyl-CoA C2, 1,2-13C2 / 2-13C1, which can be assessed via the acetyl moiety of GlcNAc or GalNAc, is 2.5 for MCF10A and 6.1 for MDA-MB-435. Whereas these ratios are similar to the isotopomer ratios of 4,5-13C2 + 3,4,5-13C3 / 4-13C1 + 3,4-13C2 of glutathione C4 Pacific Symposium on Biocomputing 12:181-192(2007) (2.8 for MCF-10A and 5.7 for MDA-MB-435), they are markedly different from the glutamate C4 ratios (0.96 for MCF-10A and 2.2 for MDA-MB-435). This indicates that the C4 and C5 in the -glutamyl moiety of glutathione are solely derived from acetyl-CoA, whereas glutamate is a likely subject of the isotopic dilution originating from a non-enriched carbon source (e.g. glutamine). Therefore, the isotopomer distribution of -glutamyl of glutathione was used to determine the relative activity of the anaplerotic reaction versus TCA cycle. The observed flux ratio of pyruvate carboxylase reaction over TCA cycle is slightly decreased in MDA-MB-435 compared to MCF-10A cells (Fig.2). Analysis of the 13C labeling pattern of the nonessential amino acids allowed us to determine the activity of the respective biosynthetic pathways. Using the 13 C isotopomer model, we found that cysteine is obtained directly from media components, and the activity of glutamate and glutamine biosynthesis is not changed significantly in MCF-10A and MDA-MB-435 cells (data not shown). Interestingly, MCF-10A cells do not utilize glucose for synthesis of glycine and proline, whereas these amino acids are actively synthesized from glucose in MDA-MB-435 cells (Fig. 2). 3.3. Metabolite Pools We used the 2D NMR data from the same labeling experiments to determine and compare the concentrations of unambiguously assigned metabolites (Table 2). Quantitation of metabolites with natural isotope abundance yields directly the total metabolite concentrations. At the same time, the differences observed for biosynthetically labeled metabolites may originate from changes in pool sizes as well as due to the 13C enrichment. In many cases these effects can be decoupled as illustrated below. Comparison of MCF-10A and MDA-MB-435 cell lines revealed significant changes in the pool sizes of many metabolites. For example, malignant cells exhibited significantly increased glutathione, m-inositol, and creatine concentrations and decreased isoleucine, leucine, valine, and taurine concentrations. Phosphocholine level is higher, whereas free choline and glycerophosphocholine were below the detection level in MDA-MB-435. The observed 12-fold increase in C2 and C3 peaks of succinate may not be explained solely by the 13C enrichment, which could account only for ~12% of the overall increase. The latter estimate is based on the labeling pattern of ketoglutarate deduced from the observed ~1.3-fold 13C enrichment at the C3 and C4 of -glutamyl moiety of glutathione. Therefore, the total pool size of succinate was significantly increased in MDA-MB-435 cells. A similar approach allowed us to establish a substantial increase in the total pool size of GlcNAc or GalNAc and a decrease in those of alanine, glutamine, and glycine (Fig.3). Pacific Symposium on Biocomputing 12:181-192(2007) Table 2. Comparison of metabolite concentrations in MCF-10A and MDA-MB-435 cells a Metabolites b Arginine GSH Isoleucine Leucine Lysine Valine m-Inositol Free choline Phosphocholine Glycerophosphocholine Total choline Phosphocholine / glycerophosphocholine Creatine Taurine Ratio MDA-MB-435 / MCF-10A 0.98 ± 0.15 1.59 ± 0.08 0.27 ± 0.04 0.48 ± 0.05 0.74 ± 0.16 0.26 ± 0.03 1.75 ± 0.10 < 0.25 1.72 ± 0.09 < 0.10 1.39 ± 0.17 > 17.2 1.74 ± 0.08 0.25 ± 0.03 Metabolites labeled at specific positions c Alanine C2 Alanine C3 Glutamine C4 Glutamate C3 Glutamate C4 Glu (GSH) C3 Glu (GSH) C4 Gly (GSH) C2 Glycine C2 Proline C4 Proline C5 Lactate C2 Ratio MDA-MB-435 / MCF-10A 1.02 ± 0.05 1.00 ± 0.05 < 0.05 1.56 ± 0.22 2.09 ± 0.31 2.17 ± 0.16 1.94 ± 0.15 6.54 ± 0.98 1.15 ± 0.06 8.57 ± 2.53 10.9 ± 3.3 0.67 ± 0.17 Succinate C2/C3 > 12.3 GlcNAc / GalNAc C2 > 14.7 UDP-GlcNAc / UDP2.56 ± 0.64 GalNAc C2 UTP / UDP C1 3.38 ± 0.53 a Relative amount of the various compounds were obtained by normalizing peaks to the internal reference standard, and further normalized per 1 mg of total protein (mean + s.d.; n=4) b Quantitation of metabolites with natural isotope abundance (a direct measure of metabolite concentrations). c Differences observed for biosynthetically labeled metabolites may reflect both, a 13C enrichment and a change in a total pool size. 4. Discussion The key aspects of the metabolomics methodology used in this study were: 1. A comparative approach was applied to assess metabolic changes in a model system of the highly metastatic cell line MDA-MB-435 versus the immortalized nontumorigenic cell line MCF-10A. 2. [U-13C]glucose labeling followed by the high-resolution 2D NMR spectroscopy allowed us to monitor twenty-four intracellular metabolites (Tables 1 and 2) in addition to fatty acids analyzed by GC-MS. 3. An extensive 13C isotopomer model was developed to determine and compare fluxes through the key central metabolic pathways including glycolysis, PPP, TCA cycle and anaplerotic reactions, and biosynthetic pathways of fatty acids and non-essential amino acids (Fig.2). 4. A combination of fluxes with individual metabolite pools within the single metabolic reconstruction framework expanded our ability to interpret underlying metabolic transitions (Fig.3). Although most of the individual components of this approach have been previously described, to our knowledge this is the first study when a combination of these techniques was systematically applied for metabolomics of Pacific Symposium on Biocomputing 12:181-192(2007) cancer. Although comprehensive isotopomer models are widely used in microbial systems (18,19), only a few models have been described for human cells (20-29). Most of these models were restricted by relatively narrow metabolic subnetworks (20-25) or based on the labeling data for one (i.e. glutamate (25,26)) or a few individual metabolites (27-29). Due to the higher sensitivity of HSQC method compared to regular 13C-NMR we were able to decrease the amount of cells required for the analysis. The increased signal dispersion in 2D spectra allowed us to analyze a wide range of metabolites without prior separation. Glucose Creatine Taurine G6P GAP 3PG R5P Ile Gly Gly Ala Leu Val Arg Lactate Pyruvate Fatty acids m-Inositol Acetyl-CoA GlcNAc Lys Phospho- Oxaloacetate choline Citrate -Ketoglutarate Glu* GSH Gln Glu Pro Glycerophosphocholine Succinate Fig. 3. Metabolic profile changes in breast tumors compared with normal human mammary epithelial cells. The arrows represent the fluxes. Fluxes are normalized to glucose uptake rate. The boldface arrows indicate the fluxes that are significantly upregulated. The pool sizes of boxed metabolites are directly assessed by [13C,1H] HSQC. Metabolites are colored if their concentrations are increased (black), decreased (white), or not changed (gray). G6P, glucose-6-phosphate; R5P, ribose-5-phosphate; GAP, glyceraldehydes-3-phosphate; 3-PG, 3-phosphoglycerate. See other abbreviations in Table S1 given in SOM. An integration of fluxes and pool sizes acquired within a single experiment Pacific Symposium on Biocomputing 12:181-192(2007) gives a more detailed fingerprint of the phenotype compared to conventional approaches based on one parameter. Although fluxes provide a direct measure of metabolic activities pointing to potential targets, they can be usually obtained only for a subset of central metabolic pathways. Metabolite pools can be readily assessed for both central and peripheral metabolites. While providing only an indirect evidence of metabolic activities, they can be used as biomarkers. We observed a sharp increase in metabolic activity of several pathways in cancer cells (Fig.2 and 3). Some of these observations such as upregulation of PPP and fatty acid synthesis are consistent with previous reports (30,31) providing us with a validation of the approach. An increase in other fluxes, eg the synthesis of glycine and proline, are reported here for the first time. Possible implications of these changes in establishing and maintaining a breast cancer phenotype are yet to be explored. Some of the observed changes in metabolite pools can be readily interpreted in the context of respective fluxes. For example the pools of all monitored amino acids decreased or remained largely unchanged in cancer cells, despite the established upregulation of some of the respective biosynthetic pathways (Fig.3). This is consistent with accelerated consumption of amino acids for protein synthesis. At the same time, the pool of glutathione (GSH in Fig.3), which is not consumed at the same level increased in keeping with the increased synthetic flux. Overproduction of GSH in tumors may reflect the increased resistance towards oxidative stress (32). We observed significant alterations in pools of several peripheral metabolites (eg creatine and taurine), whose metabolism may not be easily assessed via flux measurements. Therefore, the results obtained in this study, in addition to the validation of the approach, provide new information about metabolic aspects of tumorigenesis, and can aid the identification of new diagnostic and therapeutic targets. The presented approach constitutes a promising analytical tool to screen different metabolic phenotypes in a variety of cell types and pathological conditions. 1. 2. 3. 4. 5. REFERENCES Chu, S., DeRisi, J., Eisen, M., Mulholland, J., Botstein, D., Brown, P. O., and Herskowitz, I. (1998) Science 282, 699-705 Klose, J., Nock, C., Herrmann, M., Stuhler, K., Marcus, K., Bluggel, M., Krause, E., Schalkwyk, L. C., Rastan, S., Brown, S. D., Bussow, K., Himmelbauer, H., and Lehrach, H. (2002) Nat Genet 30, 385-393 Voss, T., Ahorn, H., Haberl, P., Dohner, H., and Wilgenbus, K. (2001) Int J Cancer 91, 180-186 Moch, H., Schraml, P., Bubendorf, L., Mirlacher, M., Kononen, J., Gasser, T., Mihatsch, M. J., Kallioniemi, O. P., and Sauter, G. (1999) Am J Pathol 154, 981-986 Celis, J. E., Celis, P., Ostergaard, M., Basse, B., Lauridsen, J. B., Ratz, G., Rasmussen, H. H., Orntoft, T. F., Hein, B., Wolf, H., and Celis, A. (1999) Cancer Res 59, 3003-3009 Pacific Symposium on Biocomputing 12:181-192(2007) 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. Golub, T. R., Slonim, D. K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J. P., Coller, H., Loh, M. L., Downing, J. R., Caligiuri, M. A., Bloomfield, C. D., and Lander, E. S. (1999) Science 286, 531-537 Dang, C. V., Lewis, B. C., Dolde, C., Dang, G., and Shim, H. (1997) J Bioenerg Biomembr 29, 345-354 Lee, W. N., Bassilian, S., Guo, Z., Schoeller, D., Edmond, J., Bergner, E. A., and Byerley, L. O. (1994) Am J Physiol 266, E372-383 Wittmann, C., and Heinzle, E. (1999) Biotechnol Bioeng 62, 739-750 Delaglio, F., Grzesiek, S., Vuister, G. W., Zhu, G., Pfeifer, J., and Bax, A. (1995) J Biomol NMR 6, 277-293 Szyperski, T. (1995) Eur J Biochem 232, 433-448 Gribbestad, I. S., Petersen, S. B., Fjosne, H. E., Kvinnsland, S., and Krane, J. (1994) NMR Biomed 7, 181-194 Gribbestad, I. S., Sitter, B., Lundgren, S., Krane, J., and Axelson, D. (1999) Anticancer Res 19, 1737-1746 Pal, K., Sharma, U., Gupta, D. K., Pratap, A., and Jagannathan, N. R. (2005) Spine 30, E68-72 Patel, A. B., Srivastava, S., Phadke, R. S., and Govil, G. (1999) Anal Biochem 266, 205-215 Sharma, U., Atri, S., Sharma, M. C., Sarkar, C., and Jagannathan, N. R. (2003) NMR Biomed 16, 213-223 Sharma, U., Mehta, A., Seenu, V., and Jagannathan, N. R. (2004) Magn Reson Imaging 22, 697-706 Dauner, M., Bailey, J. E., and Sauer, U. (2001) Biotechnol Bioeng 76, 144-156 Schmidt, K., Nielsen, J., and Villadsen, J. (1999) J Biotechnol 71, 175-189 Fernandez, C. A., and Des Rosiers, C. (1995) J Biol Chem 270, 10037-10042 Lapidot, A., and Gopher, A. (1994) J Biol Chem 269, 27198-27208 Jeffrey, F. M., Storey, C. J., Sherry, A. D., and Malloy, C. R. (1996) Am J Physiol 271, E788-799 Malloy, C. R., Sherry, A. D., and Jeffrey, F. M. (1988) J Biol Chem 263, 69646971 Vercoutere, B., Durozard, D., Baverel, G., and Martin, G. (2004) Biochem J 378, 485-495 Lu, D., Mulder, H., Zhao, P., Burgess, S. C., Jensen, M. V., Kamzolova, S., Newgard, C. B., and Sherry, A. D. (2002) Proc Natl Acad Sci U S A 99, 27082713 Cline, G. W., Lepine, R. L., Papas, K. K., Kibbey, R. G., and Shulman, G. I. (2004) J Biol Chem 279, 44370-44375 Boren, J., Cascante, M., Marin, S., Comin-Anduix, B., Centelles, J. J., Lim, S., Bassilian, S., Ahmed, S., Lee, W. N., and Boros, L. G. (2001) J Biol Chem 276, 37747-37753 Boren, J., Lee, W. N., Bassilian, S., Centelles, J. J., Lim, S., Ahmed, S., Boros, L. G., and Cascante, M. (2003) J Biol Chem 278, 28395-28402 Portais, J. C., Schuster, R., Merle, M., and Canioni, P. (1993) Eur J Biochem 217, 457-468 Boros, L. G., Cascante, M., and Lee, W. N. (2002) Drug Discov Today 7, 364372 Baron, A., Migita, T., Tang, D., and Loda, M. (2004) J Cell Biochem 91, 47-53 Meister, A. (1991) Pharmacol Ther 51, 155-194