Between-person comparison of metabolite fitting for NMR-based quantitative metabolomics

Nuclear magnetic resonance (NMR) spectroscopy is widely used as an analytical platform for metabolomics. Many studies make use of 1D spectra, which have the advantages of relative simplicity and rapid acquisition times. The spectral data can then be analyzed either with a chemometric workflow or by an initial deconvolution or fitting step to generate a list of identified metabolites and associated sample concentrations. Various software tools exist to simplify the fitting process, but at least for 1D spectra, this still requires a degree of skilled operator input. It is of critical importance that we know how much person-to-person variability affects the results, in order to be able to judge between different studies. Here we tested a commercially available software package (Chenomx' NMR Suite) for fitting metabolites to a set of NMR spectra of yeast extracts and compared the output of five different people for both metabolite identification and quantitation. An initial comparison showed good agreement for a restricted set of common metabolites with characteristic well-resolved resonances but wide divergence in the overall identities and number of compounds fitted; refitting according to an agreed set of metabolites and spectral processing approach increased the total number of metabolites fitted but did not dramatically increase the quality of the metabolites that could be fitted without prior knowledge about peak identity. Hence, robust peak assignments are required in advance of manual deconvolution, when the widest range of metabolites is desired. However, very low concentration metabolites still had high coefficients of variation even with shared information on peak assignment. Overall, the effect of the person was less than the experimental group (in this case, sampling method) for almost all of the metabolites.