Application of book-keeping principles to metabolic networks provides a powerful technique for understanding the properties of microorganisms and predicting the results of genetic modification.
Stoichiometric analysis has the same function in biochemistry as book-keeping has in business. Both deal with the inputs and outputs in flow systems, and studying the stoichiometric structure of a metabolic network might seem only slightly more exciting than a visit to an accountant's office. Times change, however, and progress in systems biology will involve paying closer attention to stoichiometry than seemed necessary in the past. The benefits should be that, with minimal knowledge of kinetic parameters, we will be able to predict how systems will respond to changes in conditions, and how they can be genetically engineered to produce desirable characteristics.
The growing knowledge of genomes of different organisms has brought new life to the study of metabolic networks, and a striking example appears on page 190 of this issue1. Stelling and collaborators discuss there the idea that, by breaking a network down into 'elementary flux modes', based on simple accounting for metabolic inputs and outputs, possible properties of the network can be predicted.
A typical biological network such as the central metabolism of the gut bacterium Escherichia coli consists of many processes that operate simultaneously and in parallel. For example, many different metabolic products are being synthesized at different rates at the same time as substrates such as glucose are being consumed to supply power for all the activity. Even when all the individual processes have been identified, it is no trivial matter to predict how the properties of the network as a whole will change if the activity of one enzyme is changed.
It used to be widely assumed, for example, that complete elimination of an enzyme activity would have obvious effects, but this expectation has been overthrown by observations of the effects of gene knockouts in various organisms: in E. coli, fewer than 300 out of 4,000 genes are 'essential' in the sense that deleting one of them prevents growth on a rich medium2. Many of the others can be deleted (one at a time) without producing any evident consequences, even for growth on a medium containing restricted nutrients; in other words, they are 'silent'. As well as discovering the functions of these silent genes, it is crucial to know why they are not essential, and elementary flux modes provide a tool for addressing this question.
Consider, for example, the simple branched network of six reactions shown in Fig. 1a. There are three subnetworks that account for everything the whole network can do. These are the elementary flux modes, labelled A, B and C in Fig. 1b, of which mode A consists of reactions 1, 2, 3 and 4; mode B of reactions 1, 2, 5 and 6; and mode C of reactions 3, 4, 5 and 6. Analysing the network in this way makes it immediately obvious that transformation of X into Y is a function of flux mode A and that deleting a reaction that is not part of it, for example reaction 5, will not prevent the system from achieving this transformation. Predicting whether a mutant will be able to grow involves identifying elementary flux modes that do not use the missing reaction. In this simple example, the analysis merely confirms what is evident from inspection, but in larger systems this is not the case.
Systems of moderate size have previously been decomposed into elementary flux modes3,4. But there has been no analysis of a system that encompasses most of an organism's metabolic activity, something that is necessary if unintuitive properties are to be deduced. The approach has now come of age with the analysis by Stelling and colleagues1 of a network of 110 reactions, linking 89 different metabolites, that encapsulates our knowledge of the central metabolism of E. coli. This system is much larger than any analysed previously, and its complexity is illustrated by the huge number — 43,279 — of elementary flux modes, far beyond the range of analysis possible by inspection.
The results of Stelling et al. allow us to understand some of the experimental properties of E. coli in terms of its stoichiometric structure. For example, glucose is involved in 27,099 elementary modes, more than twice as many as glycerol, in agreement with biochemical experience that glucose is the more important energy source. Moreover, around 7% of the elementary modes for glucose allow energy production and growth in the absence of oxygen. But none of the elementary modes for glycerol, succinate or acetate allows this, in agreement with the observation that E. coli can grow anaerobically on glucose but not on these other energy sources.
The agreement between theory and experiment provides a basis for believing that this method can produce valid conclusions even without knowing the answer in advance, and so can be used for organisms that are less well known than E. coli. Specifically, we can now predict which mutations, in which combinations, a bacterial or yeast culture can tolerate, and what genetic modifications (addition or suppression of catalytic activities) will permit new properties to be created. Moreover, as flux modes may differ in the amount of product they generate per unit of input, they allow insight into how yields may be improved. Organisms such as E. coli that can survive in widely varying conditions owe their metabolic robustness, or resistance to mutation, to a high degree of redundancy — in other words, they have a much larger number of elementary flux modes than are absolutely necessary for growth in any particular conditions. More specialized organisms will certainly have less redundancy in their design and fewer elementary flux modes.
Until now, biochemical progress has been driven mainly by new observations rather than theories — in contrast, say, to modern physics. However, knowledge of elementary flux modes allows predictions of what should happen after a particular genetic manipulation or change in culture conditions, and these predictions can, of course, be tested. Stoichiometric analysis will thus give a boost to hypothesis-driven experimental research in biology.
Stelling, J., Klamt, S., Bettenbrock, K., Schuster, S. & Gilles, E. D. Nature 420, 190–193 (2002).
Csete, M. E. & Doyle, J. C. Science 295, 1664–1669 (2002).
Schuster, S. et al. Bioinformatics 18, 351–361 (2002).
Van Dien, S. J. & Lidstrom, M. E. Biotechnol. Bioeng. 78, 296–312 (2002).
About this article
Ecological Complexity (2018)
Progress in Biophysics and Molecular Biology (2011)
The generalized flux-minimization method and its application to metabolic networks affected by enzyme deficiencies