Analytical precision, biological variation, and mathematical normalization in high data density metabolomics

Metabolic serotypes sensitive to caloric intake may enable sera metabolomic profiles to validate epidemiological parameters and predict disease risk in humans. This long-range goal is complicated by the lack of known state markers and the requirement for simultaneous monitoring of multiple small changes. Therefore, analytical precision for appropriate high data density studies using HPLC separations coupled with coulometric array detectors was evaluated over a two month period in pooled rat sera samples (previously collected and stored at −80 °C), and in authentic biochemical standards. In sera, mean coefficients of variation (CV) of retention time and ratio accuracy within the established metabolic serotype varied within ±1% and ±3%, respectively. In sets of purified standards, the same parameters fluctuated, correspondently, in ranges of ±0.1% and ±1%. Median CV of the metabolite concentrations were ~13% in standards and ~11–19% in sera, and varied non-monotonically with the analytical system status and experimental design. These parameters were shown to be sufficiently controlled so as not to dominate intra-group biological variability in serum metabolomics studies. Continuation of experimental runs across an analytical breakpoint (column replacement) was associated with disproportionate changes in metabolite concentrations, independent of maintained analytical precision. These changes were sufficient to shift overall profile localization in megavariate projection analyses. We developed a mathematical approach to normalize this break and use partial least squares projection to latent structure discriminant analysis to confirm validity of this normalization approach. This generally applicable mathematical correction helps enable longer term high data density studies by removing a critical source of systemic variation.