What is Metabonomics?
Metabonomics is a general approach to analysing the regulation of metabolism within in an organism. By measuring hundreds or thousands of metabolites simultaneously a picture (which we sometimes refer to as a "fingerprint") of the current metabolic status of the organism is generated. It is then possible to compare this metabolic profile in the same organism at different times, or else in a groups of different organisms.
In principle, the metabonomic approach consists of two distinguishable parts: firstly, an experimental technique must be used to collect the input dataset - the concentration of multiple metabolites within the sample under study. Secondly, a data processing technique must be applied to the dataset in order to sift out of the millions of individual datapoints the patterns that are of interest. Our work in the metabonomics field has, so far, relied entirely on NMR spectroscopy as the experimental data collection technique. NMR offers a number of advantages, including very high reproducibility (the co-efficient of variation for replicate measures of the same sample are in the range 0.5-2% across the NMR spectrum, which compares with 3-10% for techniques such as ELISA). Proton NMR is also absolutely non-selective with regard to the metabolites which can be detected, since all biomolecules contain hydrogen atoms. Its principle disadvantage is the cost of the equipment required and the relative insensitivity when applied to low abundance molecular species. Other experimental data collection techniques include mass spectroscopy, typically following liquid chromatography (LC-MS).
Similarly, a number of different data processing methodologies can be adopted. We have primarily used the megavariate analysis methods first developed by Svantë Wold and his colleagues at the University of Umea. Examples of such techniques include OSC, PCA and PLS-DA. These techniques can be used both to identify patterns within large and complex datasets, but also to identify which of the many parameters under study were most important in defining those patterns. These approaches are, to some extent, limited by their assumption of linear relationships between the variables, and neural network techniques can be applied and are particularly useful when many unrelated causes can underlie membership of a particular sample class. Conversely, however, neural network approaches make it more difficult to determine which parameters were most important in the patterns that are identified.
Metabonomics has been extensively applied to experimental models of human disease, with particular emphasis on toxicology studies (reviewed in Nature Rev Drug Discovery 1:153). In response to toxin, the metabolic profile of an organism traverses a trajectory where it first moves away from the "normal" region in a direction which is characteristic of the particular injury caused by the toxin, then, depending on the severity of the insult, either stops in a new region characteristic of the permanent injury caused, or else returns to "normal" over a period of time. This approach is likely to be of particular value in screening novel chemical entities for toxic side-effects.
More recently, metabonomics has been applied for the diagnosis of disease states. Megavariate data processing techniques can be used to classify metabolic profiles on the basis of known disease status, and then this "mathematical model" tested by querying unknown samples against it. This classification approach has been used previously in experimental models of disease, but has only recently made the transition into human disease. Our work in this field is centred on this transition into "Clinical Metabonomics", which is the use of metabonomics in the clinic to diagnose human medical conditions.