Transcriptional profiling efficacy to define biological activity similarity for cosmetic ingredients' safety assessment based on next generation read-across

The objective of this work was to use transcriptional profiling to assess the biological activity of structurally related chemicals to define their biological similarity and with that, substantiate the validity of a read-across approach usable in risk assessment. Two case studies are presented, one with 4 short alkyl chain parabens: methyl (MP), ethyl (EP), butyl (BP), and propyl paraben (PP), as well as their main metabolite, p-Hydroxybenzoic acid (pHBA) with the assumption that PP was the target chemical; and a second one with caffeine (CA) and its main metabolites theophylline (TP), theobromine (TB) and paraxanthine (PX), where caffeine was the target chemical. The comprehensive transcriptional response of MCF7, HepG2, A549 and ICell cardyomiocytes was evaluated (TempO-Seq) after exposure to vehicle-control, each paraben or pHBA, caffeine or its metabolites, at 3 non-cytotoxic concentrations, for 6h. Differentially expressed genes (FDR = 0.05, and fold change ±1.2=) were identified for each chemical, at each dose of exposure, and used to determine similarities between them. Each of the chemicals is able to elicit changes in the expression of a number of genes, as compared to controls, particularly at the highest dose tested. Importantly, the transcriptional profile elicited by each of the parabens shares a high degree of similarities across the category members. The highest number of genes commonly affected by the parabens was found between BP and PP. The transcriptional profile of the parabens is similar to the one elicited by some estrogen receptor agonist, among other actives. Pathway enrichment analysis (MSigDB v7.0) of the transcriptional profile for each paraben indicated a significant overlap in the up- and down-regulated pathways across the four parabens.The highest similarity in biological activity was found between BP and PP. This was indicative of their biological similarity, and thus the validity of the read across among the group, with BT being the closest structural and biological analogue for PP. In the caffeine case and its main metabolites, the transcriptional profile elicited of all four compounds had a high degree of similarity across the cell types, with CA and TP being the most active. The most robust response was obtained in the cardiomyocytes with the highest transcriptional profile similarity between CA and TP. The transcriptional profile of the methylxantines is similar to the one elicited by inhibitors of phosphatidylinositol 3-kinase as well as other actives. Pathway enrichment analysis (MSigDB v7.0) of the transcriptional profile for each compound indicated a significant overlap in the up- and down-regulated pathways across the four methylxanthines. The highest similarity in biological activity was found between CA and TP, supporting the conclusion that from the group of methylxantines evaluated TP is the most appropriate analogue to read-across for CA. Overall, our results support the approach of incorporating transcriptional profiling in well-designed in vitro tests to support biological similarity driven read-across procedures and strengthening the traditional structure-based approaches useful in risk assessment. Overall design: Four cell types were used for the transcriptomic experiments: MCF-7 (breast epithelial adenocarcinoma), A549 (lung epithelial carcinoma), HepG2 (hepatocellular carcinoma) and iCell cardiomyocytes (derived from induced pluripotent stem cells, FujiFilm Cellular Dynamics, Madison, WI). MCF-7, A549 and HepG2 cells were purchased from American Type Culture Collection (Manassas, VA) and grown in phenol red-free DMEM media containing 10% serum (Invitrogen, Carlsbad, CA). The iCell cardiomyocytes were grown in a proprietary maintenance medium (FujiFilm Cellular Dynamics, Inc.) for 168 h before chemical treatment. All cell cultures were performed in 96-well plates with 5 vehicle controls on each plate. Cells were seeded on 96 well plates and treated with 3 concentrations of each chemical or vehicle (DMSO) for 6 h. Treatments were randomized across the plate to minimize batch effect and dispensed using an Andrew System 1000G liquid handling robot (Andrew Alliance, Waltham, MA). Following 6 h treatment, cell lysates were harvested according to a protocol provided by Biospyder (TempO-Seq Workflow — BioSpyder) and then stored at -80 °C until analysis. Each experiment was performed in 3 biological replicas. Cell lysate plates were shipped frozen to BioSpyder (BioSpyder Technologies, Inc., Carlsbad, CA) and Tempo-Seq human whole transcriptome (version 2.0) assay was performed at BioSpyder as previously described by Biospyder, Inc. (Yeakley et al., 2017). ***RAW data not provided for A549 and iCAR cell lines***