Metabolomics, the large-scale study of the metabolic complement of the cell [1][2][3] , is a mature science that has been practiced for over 20 years 4 . Indeed, it is now a commonly used experimental systems biology tool with demonstrated utility in both fundamental and applied aspects of plant, microbial and mammalian research [5][6][7][8][9][10][11][12][13][14][15] . Among the many thousands of studies published in this area over the last 20 years, notable highlights [5][6][7][8]10,11,16 are briefly described in Supplementary Note 1.Despite the insight afforded by such studies, the nature of metabolites, particularly their diversity (in both chemical structure and dynamic range of abundance 9,12 ), remains a major challenge with regard to the ability to provide adequate coverage of the metabolome that can complement that achieved for the genome, transcriptome and proteome. Despite these comparative limitations, enormous advances have been made with regard to the number of analytes about which accurate quantitative information can be acquired, and a vast number of studies have yielded important biological information and biologically active metabolites across the kingdoms of life 14 . We have previously estimated that upwards of 1 million different metabolites occur across the tree of life, with between 1,000 and 40,000 estimated to occur in a single species 4 .

Regulation of enzyme expression is one key mechanism by which cells control their metabolic programs. In this work, a quantitative analysis of metabolism in a model bacterium under different conditions shows that expression alone cannot explain the majority of the observed metabolic changes.

Recent data suggest that the majority of proteins bind specific metabolites and that such interactions are relevant to metabolic and gene regulation. However, there are no methods to systematically identify functional allosteric protein-metabolite interactions. Here we present an experimental and computational approach for using dynamic metabolite data to discover allosteric regulation that is relevant in vivo. By switching the culture conditions of Escherichia coli every 30 s between medium containing either pyruvate or (13)C-labeled fructose or glucose, we measured the reversal of flux through glycolysis pathways and observed rapid changes in metabolite concentration. We fit these data to a kinetic model of glycolysis and systematically tested the consequences of 126 putative allosteric interactions on metabolite dynamics. We identified allosteric interactions that govern the reversible switch between gluconeogenesis and glycolysis, including one by which pyruvate activates fructose-1,6-bisphosphatase. Thus, from large sets of putative allosteric interactions, our approach can identify the most likely ones and provide hypotheses about their function.

Metabolic systems are often the first networks to respond to environmental changes, and the ability to monitor metabolite dynamics is key for understanding these cellular responses. Because monitoring metabolome changes is experimentally tedious and demanding, dynamic data on time scales from seconds to hours are scarce. Here we describe real-time metabolome profiling by direct injection of living bacteria, yeast or mammalian cells into a high-resolution mass spectrometer, which enables automated monitoring of about 300 compounds in 15-30-s cycles over several hours. We observed accumulation of energetically costly biomass metabolites in Escherichia coli in carbon starvation-induced stationary phase, as well as the rapid use of these metabolites upon growth resumption. By combining real-time metabolome profiling with modeling and inhibitor experiments, we obtained evidence for switch-like feedback inhibition in amino acid biosynthesis and for control of substrate availability through the preferential use of the metabolically cheaper one-step salvaging pathway over costly ten-step de novo purine biosynthesis during growth resumption.