Elucidating the metabolic changes exhibited by cancer cells is important not only for diagnostic purposes, but also to more deeply understand the molecular basis of carcinogenesis, which could lead to novel therapeutic approaches. By regulating the expression of oncogenes or modulating various signal transduction systems, it is widely accepted that certain metabolic processes play fundamental roles in cancer progression. The significance of other metabolic phenotypes observed in cancer is more controversial, e.g. the shift in energy production from oxidative phosphorylation (respiration) to aerobic glycolysis known as the Warburg effect . The mainstream view has been that the Warburg effect is a consequence of the cancer process (secondary events due to hypoxic tumor conditions) rather than a mechanistic determinant, as originally hypothesized. Recently, however, a different picture of the role of metabolic changes in tumorigenesis has emerged; for example, the dichloroacetate-induced reversion from a cytoplasm-based glycolysis to a mitochondria-located glucose oxidation inhibits cancer growth, supporting the idea that the glycolytic shift is a fundamental requirement for cancer progression  and opening up the possibility of targeting metabolic pathways for cancer treatment .
Changes in intracellular concentrations of certain metabolites can influence the rate of cancer cell growth. A metabolite can exert this effect by acting as a signaling molecule, a role that does not preclude other important cellular functions. For instance, diacylglycerol, a lipid that confers specific structural and dynamic properties to biological membranes and serves as a building block for more complex lipids is also an essential second messenger in mammalian cells whose dysregulation contributes to cancer progression . Similarly, structural components of cell membranes such as ceramides and sphingosine are also second messengers with antagonizing roles in cell proliferation and apoptosis . Pyridine nucleotides constitute yet another example, having well characterized functions as electron carriers in metabolic redox reactions and roles in signaling pathways . In particular, NAD+ modulates the activity of sirtuins, a recently discovered family of histone deacetylases  that may contribute to breast cancer tumorigenesis . Arginine is yet another metabolite involved in numerous biosynthetic pathways that also has a fundamental role in tumor development, apoptosis and angiogenesis . Considering that the signaling role of many of these biomolecules was not even suspected a decade ago, it is likely that the role of other metabolites as second messengers will be discovered.
It is becoming increasingly clear that cellular metabolites can also be involved in the control of cell proliferation by directly regulating gene expression. Signaling pathway-independent modulation of gene expression by metabolites can occur on three levels : First, metabolites can bind to regulatory regions of certain mRNAs (riboswitches), inducing allosteric changes that regulate the transcription or translation of the RNA transcript; however, this type of direct metabolite-RNA interaction has not yet been detected in humans . Second, transcription factors can be activated upon metabolite binding, e.g. binding of steroid hormones to the estrogen receptor transcription factor induces gene expression events leading to breast cancer progression . Third, metabolites can be involved in epigenetic processes such as post-translational modification of histones that regulate gene expression by changing chromatin structure . The modulation of the rate of histone acetylation by nuclear levels of acetyl-CoA is an example of metabolic control over chromatin structure that involves epigenetic changes linked to cell proliferation and carcinogenesis .
The fact that metabolites can affect the cancer process on so many levels suggests that the manipulation of specific metabolic pathways may offer a reasonable therapeutic approach. In fact, this is the basis of several anticancer therapies that: i) have been proposed based on experimental evidence, ii) are currently the object of validation in clinical trials, or iii) are presently in clinical use. The inactivation of the metabolic enzymes KIAA1363  and indoleamine 2,3-dioxygenase  constitutes a good example of i). As for ii), several anticancer treatments that exploit the antiproliferative action of ceramide are examples of therapies based on the pharmacological manipulation of a metabolic pathway that are currently in clinical trials . Turning to iii), a metabolite-based therapy for acute lymphoblastic leukemia used since 1970  consists of depleting circulating asparagine by administration of the bacterial enzyme L-asparaginase.
The analysis of metabolic features associated with neuroendocrine cancers by a combination of experimental techniques (magic angle spinning NMR spectroscopy and microanalytical biochemical assays) and in silico methods (reconstruction of metabolic pathways from microarray gene expression data and predictions of possible biotransformations based on the chemical groups present in a given metabolite) have resulted in a promising metabolome-directed therapy . The goal is the detection of unusual pathways in the reconstructed metabolism of the cancer cell whose components can be targeted by already available drugs. In general, preventive and therapeutic anticancer approaches based on the pharmacological manipulation of metabolism aim to increase or decrease the intracellular levels of certain metabolites by administration of either the metabolites themselves, inhibitors/activators of relevant enzymes, or inhibitors/activators of specific transporters.
In this study, we hypothesize that the change in concentration of some metabolites that occurs in cancer cells could have an active role in the progress of the disease rather than merely being an inconsequential side effect. We explore whether the reversion to a metabolic phenotype more similar to the normal state might be of possible therapeutic value. Increasing the levels of certain compounds that are lowered in cancer cells could be straightforwardly achieved by directly administering the deficient metabolite. On the other hand, for metabolites whose levels are increased in cancer cells, reversion would involve activation or inhibition of key enzymes, an approach that is more difficult to implement. For that reason, here we decided to focus on the former case. Ideally, we would like to compare the actual intracellular levels of every human metabolite in normal and diseased states to identify those that are lowered in cancer cells. However, direct large-scale biochemical assays are currently unfeasible. Metabolite profiling based on NMR  or mass spectrometry techniques , although very powerful, require costly instruments and are not free of problems and limitations. In silico methods based on linking enzymes to upregulated microarray-detected transcripts and mapping to metabolic pathways have been applied to the qualitative reconstruction of the metabolome of cancer cells and some predictions have been successfully validated by biochemical experiments . Here, we describe CoMet, a fully automated and general Computational Metabolomics method that uses a Systems Biology approach to predict the human metabolites whose intracellular levels are more likely to be altered in cancer cells. We then prioritize the metabolites predicted to be lowered in cancer compared to normal cells as potential anticancer agents. We applied our methodology to a leukemia cell line and discovered several human metabolites that either alone or in combination, exhibit various degrees of antiproliferative activity.