LC-HRMS-Driven Computational Toolbox to Assess Extraction Protocols Dedicated to Untargeted Analysis: How to Ease...

Metabolomics is a powerful approach that allows for high throughput analysis and the acquisition of large biochemical data. Nonetheless, it still faces several challenging requirements, such as the development of optimal extraction and analytical methods able to respond to its high qualitative and quantitative requisites. Hence, the objective of the present article is to suggest a LC-HRMS-based untargeted profiling approach aiming to provide performant tools that help assess the performance and the quality of extraction methods. It is applied in a herbicide-contaminated soil metabolomics context. The trifactorial experimental design consists of 150 samples issued from five different extraction protocols, two types of soils, and three contamination conditions (contaminated soils with two different formulated herbicides against uncontaminated soils). Four performance and quality criteria are investigated using adapted LC-HRMS-driven computational tools. First, 861 metabolic features are annotated, and then the width of metabolome coverage and quantitative performance of the five different extraction protocols are assessed in all samples using various optimized configurations of heatmaps as well as van Krevelen diagrams. Then, the reproducibility of LC-HRMS profiles issued from the five extractions is studied by two different approaches: Euclidean distances and relative standard deviations. The two methods are examined and compared. Their advantages and limitations are thus discussed. After, the capacity of the different extractions to discriminate between contaminated and uncontaminated soils will be evaluated using orthogonal projections to latent structures-discriminant analysis. Different data scaling parameters are tested, and the results are explored and discussed. All of the suggested computational and visualization tools are performed using public-access platforms or open-source software. They can be readapted by metabolomics developers and users according to their study contexts and fields of application.