Although bisphenol A (BPA) is commonly recognized as an endocrine disruptor, the metabolic consequences of its exposure are still poorly understood. In this study, we present a non-targeted LC-MS based metabolomic analysis in combination with a full-genome, high-throughput RNA sequencing (RNA-Seq) to reveal the metabolic effects and the subjacent regulatory pathways of exposing zebrafish embryos to BPA during the first 120 hours post-fertilization. We applied multivariate data analysis methods to extract biochemical information from the LC-MS and RNA-Seq complex datasets and to perform testable predictions of the phenotypic adverse effects. Metabolomic and transcriptomic data revealed a similar subset of altered pathways, despite the large difference in the number of identified biomarkers (around 50 metabolites and more than 1000 genes). These results suggest that even a moderate coverage of zebrafish metabolome may be representative of the global metabolic changes. These multi-omic responses indicate a specific metabolic disruption by BPA affecting different signaling pathways, such as retinoid and prostaglandin metabolism. The combination of transcriptomic and metabolomic data allowed a dynamic interpretation of the results that could not be drawn from either single dataset. These results illustrate the utility of -omic integrative analyses for characterizing the physiological effects of toxicants beyond the mere indication of the affected pathways.

In this work, two knowledge integration strategies based on multivariate curve resolution alternating least squares (MCR-ALS) were used for the simultaneous analysis of data from two metabolomic platforms. The benefits and the suitability of these integration strategies were demonstrated in a comparative study of the metabolite profiles from yeast (Saccharomyces cerevisiae) samples grown in non-fermentable (acetate) and fermentable (glucose) carbon source. Untargeted metabolomics data acquired by capillary electrophoresis-mass spectrometry (CE-MS) and liquid chromatography-mass spectrometry (LC-MS) were jointly analysed. On the one hand, features obtained by independent MCR-ALS analysis of each dataset were joined to obtain a biological interpretation based on the combined metabolic network visualization. On the other hand, taking advantage of the common spectral mode, a low-level data fusion strategy was proposed merging CE-MS and LC-MS data before the MCR-ALS analysis to extract the most relevant features for further biological interpretation. Then, results obtained by the two presented methods were compared. Overall, the study highlights the ability of MCR-ALS to be used in any of both knowledge integration strategies for untargeted metabolomics. Furthermore, enhanced metabolite identification and differential carbon source response detection were achieved when considering a combination of LC-MS and CE-MS based platforms.