Cyproconazole, epoxiconazole, and prochloraz were obtained from BASF or Syngenta, respectively. The batches used identical to what has been used in a previous study by Seeger et al. (2019[66]). All other pesticidal active compounds were purchased from Sigma (Taufkirchen, Germany) in the highest available purity, dissolved in DMSO (purity > 99 %; Carl Roth, Karlsruhe, Germany), and stored at -20 °C until further use. In vitro testing was performed up to the highest possible concentrations, limited by the compound-specific solubility in cell culture medium. All other chemicals were obtained from Sigma or Merck (Darmstadt, Germany) in the highest available purity.

Human HepaRG hepatocarcinoma cells were purchased from Biopredic International (Saint Grégoire, France) and seeded according to the manufacturer's protocol at densities of 9,000, 100,000, or 200,000 cells/well in 96-well, 12-well, or 6-well plates, respectively. For cultivation William's Medium E with 2 mM glutamine (PAN-Biotech, Aidenbach, Germany), 10 % (v/v) fetal bovine serum (FBS; FBS Good Forte EU approved; PAN-Biotech), 100 U/ml penicillin and 100 µg/ml streptomycin (Capricorn Scientific, Ebsdorfergrund, Germany), 0.05 % human insulin (PAA Laboratories GmbH, Pasching, Austria) and 50 µM hydrocortisone hemisuccinate (Sigma) were used. Cells were cultured under standard cell culture conditions (37 °C in a humidified atmosphere with 5 % CO₂). Cells were differentiated for 14 days, followed by 14 days in medium additionally containing 1.7 % DMSO. Differentiated HepaRG cells were pre-adapted to treatment medium (culture medium containing only 2 % FBS and 0.5 % DMSO) for 48 h prior to exposure in treatment medium for 24 h, with a final DMSO concentration of 0.5 %. Cell identity and differentiation status were checked microscopically. Differentiated HepaRG cells exhibit characteristic, unique morphological features, as provided by the manufacturer´s specifications as well as described in several publications (Cerec et al., 2007[11]; Guillouzo et al., 2007[24]; Parent et al., 2004[56]). Additionally, cells were routinely checked for the absence of mycoplasma contaminations.

Viability of cells treated with pesticidal active compounds was analyzed in HepaRG cells using the WST-1 cell assay (Sigma-Aldrich, St. Louis, USA) as described by Luckert et al. (2018[45]). In brief, cells were seeded in 96-well plates and incubated with the respective test compound. Triton X-100 (0.01 %) served as a positive control for cytotoxicity. One hour before the end of incubation, 10 µl WST-1 reagent was added to each well containing 100 µl medium and incubated again for one hour. Afterwards, absorbance was measured at 450 nm and corrected by values measured at the reference wavelength of 620 nm, using an Infinite M200 Pro plate reader (Tecan, Männedorf, Switzerland). At least three independent biological replicates, each with six technical replicates per condition, were run.

Cells were seeded in 12-well plates and incubated for 24 h with the respective test compounds or solvent control (0.5 % DMSO). Afterwards, cells were washed twice with ice-cold phosphate-buffered saline (PBS) and lysed using 350 µl RLT buffer per well (RNeasy Mini Kit; Qiagen, Hilden, Germany) containing 3.5 µl β-mercaptoethanol. RNA was isolated according to the manufacturer's protocol. Additionally, potential DNA contaminations were removed by DNase digestion (RNase-Free DNA Set; Qiagen). For higher yields, RNA was eluted twice in 30 µl H₂O. RNA was quantified at 260 nm using a Tecan spectrometer (Tecan, Männedorf, Switzerland). For purity evaluation the 260 nm/280 nm absorbance ratio was determined and ratios > 1.8 were considered an acceptable RNA purity. RNA integrity was checked by micro-gel electrophoresis using the Agilent RNA 6000 Nano Kit in a Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Only RNAs with calculated RIN (RNA Integrity Number) values > 9 were used (quality ranging from 10, i.e. highly intact RNA to 1, i.e. completely degraded RNA). RNA samples were stored at -80 °C. Primer design for quantitative RT-PCR experiments (Supplementary Table 1) was performed by using qPrimerDepot (Cui et al., 2006[12]) and RTPrimerDB (Lefever et al., 2008[41]) and was validated by melting curve analysis. Relative expression from treatment procedures against DMSO-treated Reference RNA of all 47 genes was analyzed using 30 ng total RNA and the SensiFAST™ SYBR® No-ROX Kit (Bioline, Luckenwalde, Deutschland). All measurements were conducted in triplicates from biological replicates on a LightCycler® 96 System (Roche, Basel, Schweiz) applying sample maximization as experimental plate design. Efficiency of gene amplification was calculated for each run with LinRegPCR (Ramakers et al., 2003[61]; Ruijter et al., 2009[62]). For calculation of relative expression - including normalization by three reference genes - the software REST384 (Pfaffl, 2001[57]; Pfaffl et al., 2002[58]) was applied.

Cells were seeded in 12-well plates and incubated for 24 h with the respective test compounds or solvent control (0.5 % DMSO). Afterwards, cells were washed twice with ice-cold PBS and lysed using 400 µl/well lysis buffer containing 1 % NP-40, 0.01 % sodium dodecyl sulfate (SDS); 0.15 M sodium chloride; 0.01 M sodium phosphate; 2 mM ethylenediaminetetraacetic acid (EDTA); and 2.5 U/ml benzonase at pH 7.2. After 1 h of shaking at 4 °C, lysates were collected and stored at -80 °C. After lysis, the protein concentration was determined using the bicinchoninic acid assay (Thermo Fisher Scientific, Waltham, USA). The manufacturer's protocol was used for this purpose. The samples were diluted 1:5 before analysis and the concentrations determined with the microplate reader BioTek ELx808 (BioTek, Winooski, USA) at a wavelength of 562 nm. This was followed by tryptic proteolysis. Triethanolamine (TEA) with a final concentration of 50 mM and lysis buffer was added to the samples, which were then denatured at 99 °C for five minutes. After cooling to room temperature, tris(2-carboxyethyl) phosphine (TCEP; final concentration 5 mM) was added. After shaking for 30 sec, iodoacetamide (IAA) with a final concentration of 10 mM was added and samples were shaken again for 30 min at room temperature in the dark. Proteolysis was initiated with the addition of trypsin (trypsin:protein ratio 1:40) and the samples were digested at 37 °C for either 2 or 16 h. To stop digestion, phenylmethanesulfonyl fluoride (PMSF) with a final concentration of 1 mM was added. The samples were centrifuged for 10 min at 13,000 x g to remove cell debris. Before the samples were analyzed by liquid chromatography-mass spectrometry (LC-MS), immunoprecipitation was performed. The sample was mixed with PBS containing CHAPS (3-[(3-cholamidopropyl) dimethylammonio] -1-propanesulfonate), internal isotopically labeled standard peptide and the respective antibodies to enrich the target analytes. Special antibodies (called Triple X Proteomics; TXP) were used. These TXP antibodies specifically recognize the last four amino acids of the C-terminus. Thus, an antibody can enrich not only one peptide but whole peptide groups at once (Poetz et al., 2009[59]). The complete immunoprecipitation was performed as described in Weiß et al. (2018[80]).

An Acclaim PepMap RSLC C18 (75 μm I.D. x 150 mm, 2 μm, Thermo Fisher Scientific, Waltham, USA) analytical column and a trapping column Acclaim PepMap 100 C18 μ precolumn (0.3 mm I.D. x 5 mm, 5 μm, Thermo Fisher Scientific, Waltham, USA) were used for LC-MS analysis. The samples were measured in parallel reaction monitoring (PRM). The method duration was either 20 min (for the CYP 17-plex with a flow rate of 0.3 µl/min and an oven temperature of 40 °C) or 10 min (for all other multiplex assays with a flow rate of 1 µl/min and an oven temperature of 55 °C). Eluent A (aqueous phase) consisted of LC-MS grade water, with 0.1 % formic acid (FA) added. Eluent B (organic phase) of 80 % acetonitrile (ACN) and 20 % LC-MS grade water with 0.1 % FA. As loading buffer, 2 % ACN with LC-MS grade water and 0.05 % trifluoroacetic acid (TFA) was used. The evaluation of the data was carried out via Skyline 4.2.0. 19072. The ratio of the obtained endogenous signal and the internal isotopically labeled standard peptide was calculated. For each analyte, only the most intense fragment ion (quantifier ion) was used. For each test substance, cells were treated three times. Each sample was measured and mean values were calculated from the biological replicates. Fold changes were obtained by referencing the results of the treatments to the solvent control.

As recommended by EFSA (2011[14]; EFSA Scientific Committee et al., 2017[16]; Kortenkamp et al., 2009[38]) and described previously (Kienhuis et al., 2015[33]; Staal et al., 2018[69]), a benchmark dose (BMD) approach was used which is currently considered the most appropriate approach for deriving reference points. One kind of reference point obtained by this method is the relative potency of a compound. By scaling the concentration-response of one compound relatively against the concentration-response of another, i.e. the reference compound, the relative potency factor (RPF) was obtained as a scaling factor. Usually, RPFs are computed for a single parameter of interest, e.g. cytotoxicity or the induction of expression of a specific target gene. In the context of omics data, this would lead to a plethora of different RPFs and constitute a time-consuming and inefficient approach. To overcome this drawback we developed a strategy which considers all analytes obtained during transcriptomic or proteomics analysis applying confidence interval statistics. For this purpose, the width of the confidence interval of transcriptomic alterations as an equivalent for the overall degree of transcriptomic alterations by one compound was used. Thus, half the width of confidence interval - mathematically defined as the absolute error e and measure of the accuracy of the estimation of a parameter - is defined as the potency factor (equation 1).

There, e denotes the absolute error or potency factor, z denotes the (1-α/2)-quantile of standard normal distribution and α = 0.05, i.e. including 95 % of the values. The parameter σ reflects the standard deviation and n is the number of data points per experiment. Due to different tested concentrations a correction for compound concentration by which effect was achieved is additionally required. Thus, e was furthermore divided into the administered concentration per compound. The relative potencies were computed for all compounds against each other. The software tool Genesis 1.8.1 (Sturn et al., 2002[71]) was applied for Cluster analysis.

A panel of candidate hepatotoxicity markers was assembled based on data from the literature and own previous studies (Grinberg et al., 2018[22]; Seeger et al., 2019[66]). The panel included various nuclear receptor target genes (e.g. from the cytochrome P450 (CYP) superfamily), key players of fatty acid and bile acid metabolism-related pathways as compiled in the AOPs for steatosis and cholestasis (e.g. (Mellor et al., 2016[49]; Vinken, 2015[74]; Vinken et al., 2013[76]), as well as genes recently identified as biomarkers for drug-induced liver injury (e.g. Albrecht et al., 2019[2]; Grinberg et al., 2018[22]). Following an initial screening approach with several hepatotoxic pesticides, some functionally redundant entries with very similar regulation (e.g. closely related CYP genes known to be affected by identical transcription factors), as well as some genes not expressed in HepaRG cells or not influenced by any of the test compounds were removed from the panel, resulting in a final number of 51 hepatotoxicity markers used for subsequent analyses. The selected markers are listed in Table 1(Tab. 1).

Pesticidal active compounds were selected as test chemicals with well-known toxicological profiles. In order to cover a broad spectrum of exposure-relevant pesticides, 30 different substances were chosen for in vitro testing (Table 2(Tab. 2)). These belong to various chemical classes of pesticides, namely: anilinopyrimidines, benzimidazoles, carboxamides / dicarboximides, dithiocarbamates, imidazoles, morpholines, neonicotinoids, organophosphates, phthalimides, phenylpyrazoles / pyrazoles, strobilurins, triazoles, and quaternary ammonium compounds (Table 2(Tab. 2)), with liver as their main target organ in vivo.

Using the WST-1 assay, all compounds were screened in HepaRG cells for 24 h regarding their cytotoxic potential, for the evaluation of suitable test concentrations for transcriptomic and proteomic analysis (data not shown). The highest non-cytotoxic concentrations from the WST-1 assay were selected for subsequent transcriptomic and proteomic analysis and are listed in Table 2(Tab. 2).

Using the selected highest non-toxic concentration of each compound, mRNA expression of the 44 hepatotoxicity marker genes was determined in HepaRG cells. The complete dataset is contained in Supplementary Table 2. In total, a heterogeneous response was recorded at the mRNA level, involving up- and downregulation as well as pronounced and more subtle responses. Data are summarized in Figure 1(Fig. 1). Individual examples of regulation of selected genes are provided in Figure 2(Fig. 2), showing the results for the preferentially upregulated genes e.g. S100P, the preferentially downregulated genes CYP7A1, and for e.g. NEAT1 showing only weak alterations upon pesticide treatment.

In order to obtain correlating data at the protein level, a set of multiplexed targeted MS-based assays was used, consisting of existing and newly developed assays. In total, quantitative determination of 24 proteins was performed, with 17 proteins corresponding to important marker genes already assayed at the mRNA level (Table 1(Tab. 1)). HepaRG cells were incubated with the same concentrations of pesticidal active compounds as used for the mRNA analyses. Total results can be found in Supplementary Table 3 and are visualized in Figure 3(Fig. 3). Selected protein level data are depicted in Figure 4(Fig. 4).

Subsequently, correlation of the data obtained individually at the mRNA and protein levels was assessed. For each of the 17 hepatotoxicity markers for which mRNA and protein data were available, correlations were analyzed. This is exemplarily depicted for S100P, NQO1, CYP2C9, and HSD11B2 in Figure 5(Fig. 5). The full datasets are available in Supplementary Tables 2 and 3. Similar to S100P, CYP1A1, and TNFRSF12A also predominantly showed a consistent upregulation at the mRNA and protein levels by most chemicals (Figure 5(Fig. 5); and data not shown). Predominant downregulation at both, the mRNA and protein levels was observed for ADH1B, ARG1, CYP2C9, CYP2E1, SULT1B1, and UGT2B7 (Figure 5(Fig. 5); and data not shown). A tendency for that behavior was also seen for CYP7A1; however, the fact that protein levels of CYP7A1 were below the LOQ in several samples impeded comprehensive comparative analysis of this marker gene/protein (data not shown). A positive correlation between mRNA and protein data was also visible for ALDH3A1, FASN, and POR, even though a considerable number of deviations (8, 9, and 12 out of 30 compounds, respectively) from consistent up- or downregulation in the two datasets were recorded (data not shown). A comparable weak regulation of NQO1, without clear preference for up- or downregulation by a majority of compounds, was seen at the mRNA as well as protein levels (Figure 5(Fig. 5)). Inverse correlation of the direction of regulation was only observed for 2 of the 17 mRNAs/proteins, namely LMNA and PRKDC (data not shown). For HSD11B2, downregulation of mRNA expression but no clear-cut regulation at the protein was visible, possibly indicating a delayed response at the protein level (Figure 5(Fig. 5)). Overall, these findings point towards an overall good accordance of the responses obtained at the transcriptional and proteomic levels.

In addition to the qualitative correlation of the direction of responses at the level of individual genes or proteins, we were also interested in comparing the overall relative potencies of the test compounds to alter RNA and protein levels on a broader basis. For this purpose, we decided to use relative potency factors derived by a benchmark dose approach. Such potency factors constitute a useful compound-characterizing parameter that is usually related to one specific biological endpoint, e.g. CYP3A4 expression. To adapt this approach to multi-endpoint data as resulting from omics analysis, the development of a strategy that considers the whole dataset needs to be developed. For this purpose, we developed a method that uses the width of the confidence interval of transcriptomic or proteomic alterations as an equivalent for the overall degree of transcriptomic alterations by one compound, independent of the nature of the specific genes or proteins that were subject to regulation. The mathematical background of the total omics RPF approach is described in detail in the 'Materials and Methods' section, paragraph 'Determination of compound potency'. This way, we were able to deduce RPFs for all compounds, based on their ability to deregulate gene or protein expression in HepaRG cells. Heatmap visualization of the RPFs is presented in Figure 6A(Fig. 6). For a full table of RPFs, please refer to Supplementary Table 4. When comparing the RPFs derived from transcriptomic and proteomic analysis, it became evident that a high degree of concordance was present in the two datasets, i.e. that compounds which strongly altered the transcriptomic pattern of HepaRG cells were also potent at the protein level. Correlation analysis resulted in an R² value of 0.9515 (Figure 6B(Fig. 6)) with difenoconazole (DIF), fenpyroximate (FPX), and pyraclostrobin (PCL) being identified as the most potent compounds.