Factors affecting the robustness of metabolite fingerprinting using 1H NMR spectra

1H NMR spectroscopy is one of the techniques whose potential is currently being explored in the emerging field of metabolomics. It is a non-targeted method, producing signals for all proton-containing chemical species. For crude plant materials the spectra are always complex, with many signals overlapping. Hence a most suitable approach for analysing them is 'metabolite fingerprinting', which is aimed at highlighting compositional similarities and exploring the overall natural variability in a population of samples. The most commonly used method for this is principal component analysis (PCA), as it allows the whole spectral trace to be analysed and the vast quantity of information to be simplified. In this paper we investigate whether there are factors which may affect the NMR spectra in a way that subsequently decreases the robustness of the metabolite fingerprinting by PCA. Imperfections in the signal registration (i.e. inconsistency of the peak position) are generally detrimental to analysing whole traces by multivariate methods. The sources of such problems are illustrated through specially designed repeatability studies using potato and tomato samples, and the analysis of a tea dataset containing many samples. Careful sample preparation can help to limit peak shifts; for instance here by attempting to control the pH of the extracts. In addition, some compounds are susceptible to interactions affecting their chemical shifts and mathematical alignment of peaks may be necessary. Lastly factors such as resolution can also affect analyses and must be carefully adjusted. Our choice of examples aims to raise awareness of potential problems. We do not question the validity of the NMR approach, but point out those areas where special care may need to be taken.