Method selectivity evaluation using the co-feature ratio in LC/MS metabolomics: Comparison of HILIC stationary phase performance for the analysis of plasma, urine and cell extracts

Evaluation of the chromatographic separation in metabolomics studies has primarily been done using preselected sets of standards or by counting the number of detected features. An alternative approach is to calculate each feature's co-feature ratio, which is a combined selectivity measurement for the separation (i.e. extent of co-elution) and the MS-signal (i.e. adduct formation and in-source fragmentation). The aim of this study was to demonstrate how the selectivity of different HILIC stationary phases can be evaluated using the co-feature ratio approach. The study was based on three sample types; plasma, urine and cell extracts. Samples were analyzed on an UHPLC-ESI-Q-ToF system using an amide, a bare silica and a sulfobetaine stationary phase. For each feature, a co-feature ratio was calculated and used for multivariate analysis of the selectivity differences between the three stationary phases. Unsupervised PCA models indicated that the co-feature ratios were highly dependent on type of stationary phase. For several metabolites a 15-30 fold difference in the co-feature ratio were observed between the stationary phases. Observed selectivity differences related primarily to the retention patterns of unwanted matrix components such as inorganic salts (detected as salt clusters), glycerophospholipids, and polyethylene glycols.These matrix components affected the signal intensity of co-eluting metabolites by interfering with the ionization efficiency and/or their adduct formation. Furthermore, the retention pattern of these matrix components had huge influence on the number of detected features. The co-feature ratio approach has successfully been applied for evaluation of the selectivity performance of three HILIC stationary phases. The co-feature ratio could therefore be used in metabolomics for developing selective methods fit for their purpose, thereby avoiding generic analytical approaches, which are often biased, as type and amount of interfering matrix components are metabolome dependent. (C) 2018 Elsevier B.V. All rights reserved.