Evaluation of robustness in untargeted metabolomics: Application of multivariate analysis, linear regression and hierarchical modeling

Abstract

Gas chromatography coupled with mass spectrometry (GC-MS) is considered as a reliable and effective technique for metabolomics analysis of human tissues and body fluids. However, sample preparation procedure for GC-MS is multi-stage, long-lasting and error-prone. That is why Design of Experiments approach is often used to assess the performance of commonly applied sample preparation procedures. In this work, Plackett-Burman design with 5 center points was used to evaluate the impact of 10 variables on the extraction and derivatization of metabolites from human urine samples. The obtained data was subjected to statistical analysis. In the first approach, the principal component analysis was performed to simplify the data matrix. The first two components (explaining the highest percent of the variability) were selected as response variables. They were regressed with parameters to assess the effects of the applied changes. The obtained PCA-based models suggest that alterations applied in input parameters settings may contribute to significant changes in the obtained responses, e.g. altered overnight incubation time influences the obtained metabolic profiles. In the second approach, the whole dataset was simplified by calculating the average signal intensity across metabolites. Those values, subjected to linear regression also pointed the overnight incubation time as the factor affecting sample preparation process the most. Finally, hierarchical linear modeling was applied in order to provide detailed evaluation of the applied changes. Such an approach enables the selection of variables influencing sample's metabolic profiles and each metabolite individually, providing also robustness to outliers. Ultimately, 12 experiments enabled to identify critical stages of the applied sample preparation procedure in terms of metabolic profile (abundances of specific metabolites). Furthermore, the obtained results might also suggest the direction for optimization of GC-MS method for targeted analysis of urine metabolites of different physico-chemical properties and allow to propose some minor modifications that might be beneficial for quality and repeatability of non-targeted experiments.

Keywords

• Design of experiments
• Quality by design
• Metabolomics
• Sample preparation
• Robustness
• Gas chromatography