Demonstrating the reliability of in vivo metabolomics-based chemical grouping: Part 2 – Consistency of group-specific metabolic effects

Abstract Previously, the MetAbolomics ring-Trial for CHemical groupING (MATCHING) demonstrated high inter-laboratory reproducibility in the blinded application of rat plasma metabolomics to group eight substances (aniline, 2-chloroaniline, 4-chloro-3-nitroaniline, dichlorprop-p, fenofibrate, 17α-methyl-testosterone, trenbolone, WY-14643). Each ring-trial partner applied their preferred metabolomics workflows to acquire, process, quality assess and statistically analyse their data. All five partners whose datasets passed quality control correctly clustered the substances into three groups according to previously established toxicological effects, both for female and male rats. However, to ensure high confidence in each grouping hypothesis, metabolomics data should support a plausible toxicological interpretation by linking changes in annotated metabolite profiles to known toxicological effects associated with the substances. In this follow-up study, our first data-driven objective was to determine the consistency of group-specific discriminatory metabolites (including lipids) across partners. For all six cases (three groups in female and male rats), statistically significant consistency was observed, confirming that all partners discovered similar metabolic perturbations underpinning each group. Strikingly, this was achieved even though a range of statistical and computational approaches were used for both discovering and annotating discriminatory metabolites. Secondly, we evaluated whether pre-existing toxicological knowledge, organised using an Adverse Outcome Pathway-informed framework of metabolomic associative events (mAEs), could provide plausible toxicological interpretations of the partner’s discriminatory metabolite profiles, for each toxicological effect (androgen receptor activity (AR), peroxisome proliferation (PP), anaemia). Binomial distribution analysis revealed statistically significant detection of characteristic mAEs for all six cases, albeit with insufficient specificity across the three effects. Consequently, a third objective was developed, to refine the mAEs by combining the pre-existing toxicological knowledge with partner-to-partner detection reproducibility of metabolites and their directional change, thereby significantly enhancing mAE detection specificity. Overall, the discovery by the five partners of significantly consistent metabolic perturbations underpinning each group further demonstrates the reliability of metabolomics for chemical grouping. Moreover, an integrated knowledge and data-driven approach is presented as a basis for deriving metabolomic signatures that were used to support plausible toxicological interpretations of the metabolomics data. Considering the broader implications of this study, we propose the development of a collaborative framework to facilitate community-driven contributions to a shared knowledgebase of metabolomic signatures.