Adverse outcome pathway-driven analysis of liver steatosis in vitro : a case study with cyproconazole

Adverse outcome pathways (AOPs) describe causal relationships between molecular perturbation and adverse cellular effects, and are being increasingly adopted for linking in vitro mechanistic toxicology to in vivo data from regulatory toxicity studies. In this work, a case study was performed by developing a bioassay toolbox to assess key events in the recently proposed AOP for chemically induced liver steatosis. The toolbox is comprised of in vitro assays to measure nuclear receptor activation, gene and protein expression, lipid accumulation, mitochondrial respiration, and formation of fatty liver cells. Assay evaluation was performed in human HepaRG hepatocarcinoma cells exposed to the model compound cyproconazole, a fungicide inducing steatosis in rodents. Cyproconazole dose-dependently activated RARα and PXR, two molecular initiating events in the steatosis AOP. Moreover, cyproconazole provoked a disruption of mitochondrial functions and induced triglyceride accumulation and the formation of fatty liver cells as described in the AOP. Gene and protein expression analysis, however, showed expression changes different from those proposed in the AOP, thus suggesting that the current version of the AOP might not fully reflect the complex mechanisms linking nuclear receptor activation and liver steatosis. Our study shows that cyproconazole induces steatosis in human liver cells in vitro and demonstrates the utility of systems-based approaches in the mechanistic assessment of molecular and cellular key events in an AOP. AOP-driven in vitro testing as demonstrated can further improve existing AOPs, provide insight regarding molecular mechanisms of toxicity, and inform predictive risk assessment.


Table of Contents Graphic 1 Introduction
An important goal in the development of mechanism-based testing strategies for potential toxicants is to explore the utility of in silico and in vitro tools as parts of a refined risk assessment strategy. The concept of adverse outcome pathways (AOPs) has been introduced to link in vitro data to phenomenological effects observed in in vivo studies. AOPs are a conceptual framework to portray mechanistic toxicological knowledge about how chemicals lead to an adverse outcome. It directly links a molecular initiating event along a causal chain of molecular and cellular key events to an adverse outcome, which is relevant to risk assessment at a level of biological organization 1, 2 . The concept of AOPs has gained attention in regulatory toxicology to facilitate the inclusion of mechanistic toxicological evidence into risk assessment in a formalized manner 3 . Knowledge of the molecular effects underlying the adverse outcome observed in in vivo studies is essential for establishing AOPs.
However, measurement of these molecular events along an AOP is not included in current protocols for regulatory in vivo studies and moreover, for many of the events, in vivo or ex vivo analysis may be difficult or impossible. In contrast, in vitro analyses of individual molecular events appear as a viable strategy in order to determine changes along the AOP induced by a chemical of interest. This might be (0.01%) served as a positive control. One hour before the end of incubation 10 µl WTS-1 reagent were added to each well containing 100 µl medium and plates were returned to the incubator. At the end of the incubation period, absorbance was measured at 450 nm with a reference wavelength of 620 nm using the plate reader Infinite M200 Pro (Tecan group, Männedorf, Switzerland). Values of the reference wavelength were subtracted from absorbance values and data were corrected for background absorbance by subtracting the values from wells incubated without cells. Data were referred to solvent control which was set to 100%. Dose response curve fitting (four parameter logistic curve) was performed with SigmaPlot 13.0 (Systat Software, Erkrath, Germany) and values for inhibitory concentration of 10% (IC10) were determined. At least two independent, biological replicates with five to six technical replicates per condition were performed. Transcription of the reporter gene firefly luciferase is controlled by the GAL4-specific upstream activation sequence (UAS) on a second plasmid. Agonistic binding of a test compound to the LBD of the nuclear receptor of interest leads to the activation of the fusion protein, which binds to the UAS and initiates transcription of the firefly luciferase gene.

Reporter gene assays
Since the GAL4-dependent assay is only dependent on the components encoded by the plasmids and is independent of any endogenous factors provided by the host cell line, and due to an increased transfection efficiency of HEK-293 in comparison to HepG2 cells, the transactivation assays were performed in HEK-293 cells.
Reporter gene assays for AHR, GRE as well as CAR-CYP2B6, PXR-CYP2B6 and VDR-CYP2B6 are based on nuclear receptor-specific DNA response elements that directly regulate transcription of the reporter gene firefly luciferase. These assays were performed in HepG2 cells as they require the presence of certain endogenous factors provided by the host cell line and we observed no induction of reporter gene expression in HEK-293 cells (data not shown).
HepG2 and HEK-293 cells were seeded in 96-well plates and after 24 h, cells were transiently transfected for 4−6 h with the appropriate plasmids using TransIT-LT1 (Mirus Bio, Madison, USA) according to the manufacturer instructions. A plasmid constitutively expressing the reporter gene Renilla luciferase was co-transfected as an internal control for normalization. Following transfection, cells were exposed for 24 h to 15 µM, 30 µM or 60 µM cyproconazole in culture medium with 0.5% DMSO. For each reporter gene assay a specific positive control was included (supplement , Table S1).
Cells were lysed with 50 µl lysis buffer (100 mM potassium phosphate with 0.2% (v/v) Triton X-100, pH 7.8) on an orbital shaker, then centrifuged (5000 x g, 5 min). Subsequently, 5 µl of supernatant were analyzed for firefly and Renilla luciferase activities in a dual luciferase assay as previously described (Hampf and Gossen, 2006) using a Mithras LB940 (Berthold Technologies, Bad Wildbad, Germany) or an Infinite M200 Pro (Tecan group, Männedorf, Switzerland) luminometer. Three to four independent, biological replicates were performed with four technical replicates per condition. Firefly luciferase activity was normalized to Renilla luciferase activity and expressed as relative activity referred to the positive control. Statistical analysis was performed with SigmaPlot 13.0 (Systat Software, Erkrath, Germany) using the non-parametric Kruskal-Wallis test followed by Dunn's test. Statistical significance was assumed at p < 0.05. An overview of the specific conditions for each reporter gene assay (plasmid, plasmid amount, cell line, positive control) is provided as a supplement (Table S1).

Analysis of mRNA expression levels
HepaRG cells were differentiated in 12-well plates and treated with 25 µM, 50 µM, 100 µM or 200 µM cyproconazole or solvent control (0.5% DMSO) for 24 h as described above. Cells were washed twice with ice-cold PBS and lysed with 350 µl RLT buffer (RNeasy Mini Kit, Qiagen, Hilden Germany). Total RNA was extracted according to the manufacturer protocol. The RNA was quantitated spectrophotometrically at 260 nm (A260) by using a nanophotometer P330 (Implen, München, Germany). An A260/280 ratio of > 1.8 was considered an acceptable measure of RNA purity. RNA integrity was estimated by visual examination of two distinct rRNA bands (28S and 18S) on a denaturing 1% agarose gel stained with SYBR Gold. Only RNA samples with a clear and sharp 28S rRNA band about twice as intense as that of 18S rRNA were used. For first-strand cDNA synthesis, 2 µg of total RNA were reverse-transcribed into cDNA in a total volume of 20 µl, using the SuperScript First-Strand Synthesis System for RT-PCR (Thermo Fisher Scientific, Waltham, USA) according to the manufacturer instructions using oligo dT primers for the reaction. TaqMan Gene expression assays (Thermo Fisher Scientific, Waltham, USA) were used to measure the expression of 69 genes linked to liver steatosis, nuclear receptor activation and hepatotoxicity (supplement , Table S2). TaqMan Gene Expression Assays consisted of a pair of unlabeled PCR primers and a TaqMan probe with FAM dye label on the 5' end and minor groove binder (MGB) as well as a nonfluorescent quencher (NFQ) on the 3' end. All genes were amplified by real-time PCR in the Step One Plus detection system with StepOnePlus Software v2.3 (Thermo Fisher Scientific, Waltham, USA). Each amplification reaction was carried out in a total volume of 20 µl containing 10 µl TaqMan gene expression master mix (Thermo Fisher Scientific, Waltham, USA), 1 µl TaqMan gene expression assays and 20 ng cDNA. The reactions were cycled 40 times using the following parameters: 95 °C for 15 s and 60 °C for 1 min during which the fluorescence data were collected. A non-template control was run with every set of primers. Expression levels of the target genes were normalized to the reference genes ACTB (actin beta: Hs01060665_g1), GAPDH (glyceraldehyde -3-phosphate dehydrogenase: Hs02758991_g1) and B2M (beta-2-microglobulin: Hs00187842_m1) which were found to be stably expressed throughout treatments. RNA from three independent, biological replicates was used. Each cDNA was analyzed at least in duplicate by real-time PCR. Relative gene expression was calculated using the ∆∆CT method 37 . Statistical significance of differences in expression was assessed by the non-parametric Kruskal-Wallis test followed by Dunn's test, using GraphPad Prism v.7 (GraphPad Software, La Jolla, USA). A p value < 0.05 was assumed statistically significant. The statistical calculation was based on 2 -∆Ct values.

Protein extraction and parallel reaction monitoring (PRM) analysis
HepaRG cells were differentiated in 15 cm dishes and exposed to 25 µM, 50 µM, 100 µM or 200 µM cyproconazole or solvent control (0.5% DMSO) as described above. After 72 h cells were harvested by incubating with 2 mM EDTA for 15 minutes at 37 °C. Harvested cells were centrifuged at 300 x g for 3 min at 4 °C, supernatants were removed and cell pellets were snap-frozen in liquid nitrogen and stored at -80 °C until further analysis. Cells (3 independent, biological replicates per experimental condition) were lysed with 8 M urea in ammonium bicarbonate, sonicated twice in a VialTweeter (Hielscher, Wanaque, USA) at 100% amplitude, 0.8 cycle, for 10s and centrifuged at 20,000 x g for 20 min to remove any remaining debris. Protein concentration in the extracts was determined by the bicinchoninic acid assay (BCA Protein Assay Kit, Thermo Fisher Scientific, Waltham, USA) and 100 µg per sample were used for further protein extraction. RapiGest surfactant (0.1%) was added to each sample and proteins were reduced with 5 mM TCEP (Thermo Fisher Scientific, Waltham, USA) at 37 °C for 30 min and then alkylated with 10 mM iodoacetamide (Fluka, Buchs, Switzerland) in the dark at 37 °C for 30 min. Samples were diluted with 100 mM ammonium bicarbonate to 4 M urea and LysC (Wako, Neuss, Germany) was added at a 1:100 enzyme to protein ratio and incubated for 3 h at 37 °C.
Samples were then diluted to 1.5 M urea using 100 mM ammonium bicarbonate before adding trypsin (Sigma-Aldrich, St. Louis, USA) at a 1:100 trypsin to protein ratio. Tryptic digestion was carried out overnight at 37 °C. The peptide mixtures were acidified to pH 3 by the addition of formic acid to inactivate trypsin and the precipitated RapiGest was removed by centrifugation (13,000 x g for 10 min).
The peptide mixture was further subjected to C18 purification using Sep-Pak Vac 1cc (50 mg) tC18 Cartridges (Waters, Eschborn, Germany) according to the manufacturer instructions.  , Table S3). Raw peak areas were exported from Skyline and further analyzed. Peptide intensities were summarized at the protein level by averaging.
Quantification was performed by comparing to the untreated control samples. Abundance changes were reported as log2 fold changes ± SEM using Student's t-test under the assumptions of normal distribution and equal variance.

AdipoRed assay
Levels of triglycerides were measured using the AdipoRed assay essentially according to the Erkrath, Germany) using the non-parametric Kruskal-Wallis test followed by Dunn's test. Statistical significance was assumed at p < 0.05.

Mitochondrial respiration
Mitochondrial respiration as an indicator of cellular metabolism and fitness in response to the exposure of HepaRG cells to cyproconazole was determined by extracellular flux analysis using an Agilent Seahorse XF24 Analyzer (Agilent Seahorse Bioscience, Santa Clara, CA USA). Extracellular flux analysis comprises measurement of oxygen consumption rate (OCR) in treated and control cells.
Measurements of OCR can decipher key parameters of mitochondrial respiration when different aspects of electron transport chain are shut down by respiration modulators. For this purpose oligomycin (2 µM) was used to block ATP synthase, carbonyl-cyanide-4-(trifluoromethoxy) phenyhydrazone (FCCP, 1 µM) was used to make the inner mitochondrial membrane permeable for protons and allow maximum electron flux through the electron transport chain, and a mix of rotenone (0.5 µM) and antimycin A (0.5 µM) were used together to inhibit complexes I and III, respectively.
Through use of mitochondrial inhibitors, four key mitochondrial respiration parameters were measured: basal, ATP production-linked, maximal, and proton leak-linked OCR 42 . Based on these data, spare respiratory capacity and non-mitochondrial respiration also were calculated.
To meet stringent prerequisites of the extracellular flux analysis for highly optimized and controlled seeding density of the cells 43 a minor adaptation was made to the methods described under section 2.3 (HepaRG cell culture and exposure to test compounds). Prior to seeding for the treatment, On the day of analysis plates were taken from the incubator, and cells were visually checked for their morphology, seeding uniformity, and adherence. Each plate was washed three times with 100 ml of pre-warmed (37 °C) filter sterilized XF assay medium (non-buffered DMEM base medium supplemented with 10 mM glucose, 1 mM sodium pyruvate and 2 mM L-glutamine; adjusted to pH 7.4).
The washing medium was removed from each well and cells were rinsed two times with 1 ml of prewarmed (37 °C) assay medium. Finally, 525 µl of the assay medium was added to each well and cells were observed under the microscope to ensure that cells were not washed away. The plates were placed in a 37 °C non-CO 2 incubator for 1 h prior to the assay. During this time, solutions of mitochondrial modulators oligomycin, FCCP and rotenone/antimycin A were prepared with pre-made XF assay media (Seahorse Bioscience) warmed to 37 °C. These were loaded (75 µl) into designated injection ports of sensor cartridges. Subsequently, the loaded XF sensor cartridge with the XF utility plate was placed into the XF24 Analyzer and calibrated. After calibration, the XF utility plate containing calibration fluid was replaced with the plate containing the cells. The software of the XF24 instrument was programmed to first perform 3 basal OCR measurements, followed by the sequential addition of oligomycin, FCCP, and rotenone/antimycin A. Measurement cycles performed after each addition consisted of 3 min of mixing the cell suspension, 2 min of waiting, and 3 min of OCR measurement.
Results were presented as values of OCR corresponding to basal respiration, ATP production-linked respiration, maximal respiration, proton leak-linked respiration, spare respiratory capacity and nonmitochondrial respiration. The statistical significance of values reflecting the impact of cyproconazole on measured respiration parameters in three independent, biological replicates with three technical replicates was tested by the non-parametric Kruskal-Wallis test against negative control values (assay medium with 0.5% DMSO) with p < 0.05 indicating statistically significant differences.

Nile Red staining and neutral lipid droplet analysis by high content cell imaging
HepaRG Two hours before reading 50 µl of 0.3 µg/ml DAPI in PBS were added to each well (incubation at ambient temperature in the dark). The multi-well plates were scanned (9 images) with an Arrayscan XTI using a 20x NA 0.4 objective (Plan NeoFluar, Zeiss, Oberkochen, Germany). The Photometrics X1 CCD camera was set with a binning 2 (14 bits dynamic range, 4 x 106 pixels with a size of 4.54 µm).
Identification of the nuclei was done by tracking DAPI with an XF100-386-23 filter set and used to focus the instrument. Identification of neutral lipid spot was done by tracking Nile Red green emission with an XF 100_485_20 filter set. The Spot Detector V3 Bioapplication analysis algorithm (software V.6.5) identifies nuclei upon fluorescent intensity and size. The nuclear mask was dilated to define the cytoplasmic region. The following parameters were measured at the cell level. For the nuclei: area as well as total and average intensity for each cell; for the neutral lipid spot: spot number, spot area and spot intensity as well as total fluorescence of spot intensity within each cell.
To analyze the quantitative data for neutral lipid droplet accumulation obtained after the image analysis, a workflow was built in Statistica v13.2 (Tibco, Palo Alto, USA). Firstly, each independent plate was standardized in order to eliminate inter-experiment variation. The whole data set for neutral lipid spot total intensity within cells was submitted to a median MAD standardization in order to fix the values in the same order of magnitude (robust Z-score). Then the three independent experiments were grouped. Data were now normalized to the median of control cells (DMSO treated cells = 1). Results are presented as box plot median ± percentile 25% and 75% and non-outlier data. For statistical analysis the non-parametric Kruskal-Wallis test was performed. Statistical significance was assumed at p < 0.05.

BMC 50 calculation
Dose-response modeling and benchmark concentration (BMC) analysis were performed using the Rbased software package PROAST (version 64.14, RIVM, Bilthoven, Netherlands). Response data were provided as relative data normalized to the solvent control and submitted to PROAST as continuous, summary data containing mean, standard deviation and sample size in tab-delimited text files. Data were fitted to an exponential 3-parameter model (1) or to an exponential 4-parameter model (2): The model with the lowest Akaike information criterion (AIC) value was selected and the BMC and the corresponding two-sided 90% BMC confidence interval given by BMCL (lower bound of the BMC confidence interval) and BMCU (upper bound of the BMC confidence interval) were calculated for a benchmark response of 50% (BMR 50 ).

Results
The aim of this study was to compile an in vitro test battery covering the AOP for liver steatosis and to evaluate its capacity to detect the steatotic potential of a known steatotic chemical, namely the

Nuclear receptor activation
Activation of different nuclear receptors is considered to constitute the molecular initiating event of liver steatosis 4, 12 , as delineated in the AOP for liver steatosis (Figure 1). Activation of a large set of nuclear receptors, namely AHR, CAR, FXR, GR, LXRα, PPARα, PPARγ, PPARδ, PXR, RARα, RXRα, and VDR by cyproconazole was monitored using luciferase-based reporter assays either in human HepG2 or HEK-293 cells, depending on the respective reporter gene assay. As listed in Figure 3A, most of the receptors were not affected by cyproconazole. Dose-dependent statistically significant induction of reporter activities was observed, however, for PXR ( Figure 3B) activation of this receptor was consistently demonstrated in form of a GAL4 fusion construct-based assay aimed to monitor binding of a substance to the PXR ligand binding domain, as well as by a classic reporter assay utilizing the PXRresponsive human CYP2B6 promoter ( Figure 3B). In addition, cyproconazole significantly activated luciferase reporter activity driven by the retinoid receptor RARα ( Figure 3B). According to the AOP for liver steatosis, the antagonistic binding to PPARα also represents a molecular initiating event.
However, no repression of basal activity of the reporter gene assay plasmid after exposure to cyproconazole could be observed, indicating that cyproconazole does not exhibit antagonistic PPARα binding properties (data not shown). Baseline levels of the PPARα reporter assay were high enough to be able to monitor potential down-regulation of the signal by a test compound.

PCR-based gene expression analysis
According  Figure 4B). Regarding the genes listed in the AOP, only FASN and SCD were detected to be down or upregulated in the screening approach, respectively ( Figure 4A). However, no significant gene expression changes of both genes could be confirmed in samples exposed to increasing cyproconazole concentrations ( Figure 4A). Detailed data of the heat map in Figure 4B are provided in the supplement (Table S4).

Protein abundance changes in HepaRG cells upon cyproconazole exposure
Changes in the abundance of nuclear receptor targets and important players in steatosis were additionally monitored at the protein level using a mass spectrometric-based assay in HepaRG cell lysates following 72 h exposure to 25 µM, 50 µM, 100 µM or 200 µM cyproconazole. A set of parallel reaction monitoring (PRM) assays was designed for the identification and quantification of selected proteins corresponding to gene transcripts examined by gene array analysis (Figure 4). A total of 33 proteins were monitored using quantotypic peptides and their log2-fold changes relative to untreated cells were plotted as a heat map ( Figure 5). The quantitative PRM data used for constructing the heat map in Figure 5 can be found in the supplement (Table S5)

Liver triglyceride accumulation
In the steatosis AOP, liver cell triglyceride accumulation is a late key event that follows nuclear receptor activation and transcriptional changes ( Figure 1). This parameter was monitored in human HepaRG cells after 24 h and 72 h of cyproconazole exposure using the fluorescence-based AdipoRed assay. As shown in Figure 6A, cyproconazole treatment led to a statistically significant and dose-dependent increase in cellular triglycerides, which was more pronounced after 72 h of exposure. This was confirmed by GC-FID analyses ( Figure 6B, 6C), which demonstrated an even stronger increase in relative cellular triglyceride levels by cyproconazole, especially after 72 h for triglycerides with higher molecular weight (C54 and C56).

Mitochondrial disruption
Mitochondrial disruption is thought to be an organelle level effect causally linked to liver steatosis ( Figure 1). Mitochondrial respiration was thus monitored in cyproconazole-treated human HepaRG cells using a Seahorse analyzer. Results of these assays demonstrate a decrease in basal and maximal respiration as well as in proton leak, spare respiratory capacity and ATP production after 72 h of 200 µM cyproconazole exposure (Figure 7).

Assessment of triglyceride accumulation at the single-cell level
At the cellular level, the occurrence of fatty liver parenchyma cells is considered a hallmark of the development of steatosis. Following treatment of HepaRG cells with cyproconazole, a high content cell imaging approach was employed by staining triglycerides with the lipophilic dye nile red. Figure 8A visualizes fatty hepatocyte-like cells within the cultures of HepaRG cells consisting of a mixture of hepatocyte-and bile duct epithelium-like cells. The data presented in Figure 8B demonstrate that cyproconazole induced a statistically significant increase in lipid droplets in liver cells for both time points analyzed, i.e. 24 h and 72 h of incubation with the fungicide. These data are consistent with the observation of increased triglyceride levels in cyproconazole-treated HepaRG cells ( Figure 6).

Dose-response and temporal relationships between key events
A principal characteristic of AOPs is the existence of dose-response relationships and temporal concordance between the different key events 46, 47 . In order to compare the different key events analyzed in this study and the employed assays, we performed dose-response modeling for each data set and calculated the BMC 50 value (i.e. the concentration at which a 50% change in the response occurs, relative to the background) (Table 1). Due to the complexity of gene and protein expression data we focused on the following key events: nuclear receptor activation, liver triglyceride accumulation, mitochondrial disruption and fatty liver cells. According to AOP principles different key events have to causally concur in their dose-response relationships. It implies that upstream (i.e. preceding) key events occur at lower concentrations than subsequent downstream events.
Consequently, the incidence for an upstream event is higher than that of a downstream key event for the same concentration of a tested stressor. This concordance of dose-response relationships is reflected by the data in Table 1. At 24 h of exposure, nuclear receptor activation occurs at lower cyproconazole concentrations (reflected by a lower BMC 50 ) than liver triglyceride accumulation followed by mitochondrial disruption. The same is true for 72 h of exposure where liver triglyceride accumulation again occurs at lower cyproconazole concentrations than mitochondrial disruption. However, the key event fatty liver cells can also be detected at lower concentrations than mitochondrial disruption although it succeeds mitochondrial disruption according to the AOP. This does not necessarily reflect an inconsistency in the expected dose-response relationship but might be due to differences in sensitivity or variability of the employed in vitro assays. For instance data suggest that liver triyglyceride accumulation analysis via GC-FID seems to be more sensitive than AdipoRed assay as BMC 50 values for GC-FID are lower at both time points. Temporal concordance implying an occurrence of upstream key events at earlier time points than downstream key events is also reflected by the data. In general the BMC 50 values for GC-FID are lower at 72 h than at 24 h which indicates that liver triglyceride accumulation is much more pronounced after 72 h exposure than after 24 h exposure and thus is a key event that occurs later in time than earlier key events like the nuclear receptor activation. The same is true for mitochondrial disruption where the BMC 50 values are also lower after 72 h than after 24 h indicating a higher incidence at later time points.

Discussion
Using the example of the AOP on liver steatosis, the data presented in this study illustrate how a battery of in vitro tests can be used to quantitatively assess key events in an AOP and thereby to  24 . In a number of studies, it was shown that cyproconazole increases liver weight, induces hepatocyte hypertrophy and leads to lipid droplet accumulation in rodent hepatocytes in vivo [24][25][26] . The present data obtained with human HepaRG hepatoma cells convincingly show that the compound also can induce steatosis in human cells.
In the absence of evidence of CAR activation in non-rodent cell systems, it is tempting to speculate that PXR activation most likely is the main underlying mechanism by which cyproconazole induces steatosis in human cells. These data suggest that at the level of key events, the AOP in its present state does not fully reflect the biological processes that link nuclear receptor activation to increased cellular fatty acid contents via transcriptional changes at the mRNA level.
Activation of LXR may be capable of triggering the key event genes FASN, SCD and SREBF1 of the AOP in HepaRG cells. This situation might indicate that the present version of the AOP is centered around LXR activation, whereas signaling pathways through other nuclear receptors, which also induce steatotic processes, are not yet fully reflected. While a detailed time-resolved analysis of all molecular alterations caused by cyproconazole is outside the scope of the present study, further work, for example using specific PXR ligands, could help to close remaining gaps of the AOP network for hepatic steatosis and to include critical target genes of PXR and/or other nuclear receptors relevant for steatosis.
Another possibility for why the induction of target genes at the key event level was not observed in our analyses might be that the rate of transcriptional activation is not suitable for detecting alterations with mRNA isolation 24 hours after the start of incubation with cyproconazole. This explanation is unlikely, however, since induction of PXR target genes is generally evident in HepaRG cells 24 h after exposure to PXR agonists 19 . Moreover, known PXR target genes such as CYP3A4 were markedly induced in the present analysis. Finally, this assessment was consistent with the profile resulting from our proteomicsbased evaluation of cyproconazole exposure in HepaRG cells.
After 72 h, later key events like triglyceride accumulation and mitochondrial disruption are clearly induced, however no changes in the abundance of proteins encoded by the target genes postulated in the AOP were evident except for significant reductions in ACOX1 and FASN. ACOX1 is a rate-limiting enzyme in peroxisomal fatty acid oxidation and its deletion causes microvesicular steatosis, which can progress to steatohepatitis and hepatocellular carcinoma 54,55 . The decrease in protein abundance of ACOX1 is in agreement with the AOP, although no PPARα antagonistic binding, the underlying molecular initiating event, was found for cyproconazole. In contrast, FASN is an important enzyme for de novo fatty acid synthesis and it is upregulated in patients with hepatic steatosis in the absence of hepatic inflammation 56 . Thus, in the AOP, it is proposed to be activated by LXR, PPARγ, FXR or CAR and the reduction of FASN at the protein level observed in our study is not in line with the AOP.
In the future, in vitro determination of a limited number of molecular changes induced by a test compound might be used for predicting the steatotic potential of chemicals without the need to conduct costly and ethically controversial studies in vivo. This future approach, however, will require more complete AOPs and AOP networks containing molecular mechanisms not currently (or not sufficiently) represented. Given discrepancies between observed mRNA induction and the mRNA induction pattern postulated in the steatosis AOP, the in vitro test battery developed in this work was able to recapitulate most key events of the AOP, including the molecular initiating event of nuclear receptor activation, mitochondrial disruption, lipid accumulation, and the occurrence of fatty liver cells. Furthermore, concordance in dose-response relationship as well as temporal association could be demonstrated and is in line with central AOP principles. Thus, while existing AOPs such as that for steatosis are expected to require ongoing improvements whenever more detailed mechanistic information becomes available, they nonetheless provide a potentially useful framework for guiding toxicological studies in vitro.
Further research involving case studies such as the present work are anticipated to, on one hand, help confirm aspects of an AOP network, and on the other hand to critically evaluate and identify weaknesses. As elucidated in the present study, the current AOP on liver steatosis only sparsely reflects the mechanistic link between nuclear receptor activation and gene expression changes leading to hepatic steatosis. In principle, discrepancies between observed nuclear receptor activation and induction of gene expression might results from differences in cellular metabolism of the test characterizes those later key events as a consequence of mRNA and protein expression, regulated by the diverse nuclear receptors activated by steatotic compounds (see also Figure 1). Thus, even not directly measured, information on these apical key events can be extracted from our data.
Nonetheless, future integration of additional apical endpoints suggested by Angrish et al. 13 and Angrish et al. 14 might further increase the values of the system. Testing of multiple mixtures in vitro in a test battery such as ours might constitute a feasible approach for mixture testing of chemical compounds.

Funding Information
The research leading to these results received funding from the European Union´s Horizon 2020 research and innovation programme under Grant Agreement 633172 (European Test and Risk Assessment Strategies for Mixtures). This publication reflects only the author's views, and the Community is not liable for any use made of the information contained therein.

Acknowledgments
We would like to thank Regina Al-Hamwi and Beatrice Rosskopp for their excellent technical support.

Supporting information
The supporting information provides as supplement Figure S1 as well as Table S1 to Table S5: Figure S1: Cytotoxicity of cyproconazole in HepaRG, HepG2 and HEK-293 cells Table S1: Plasmids, cell lines and positive controls used for the different reporter gene assays Table S2: List of selected genes linked to liver steatosis, nuclear receptor activation and hepatotoxicity Table S3: Quantotypic peptides used to compare protein level changes in cyproconazole-exposed HepaRG cells relative to untreated controls Table S4: Results for gene expression analysis of genes linked to liver steatosis, nuclear receptor activation and hepatotoxicity Table S5: Results for protein abundance changes in cyproconazole-exposed HepaRG cells  a Dose-response modeling and benchmark dose (BMC) analysis were performed using the software package PROAST. The BMC and the respective two-sided 90% BMC confidence interval given by BMCL (lower bound of the BMC confidence interval) and BMCU (upper bound of the BMC confidence interval) were calculated for a benchmark response of 50% (BMR 50 ). Data were sorted for BMC 50 . For high content cell imaging data after 24 h exposure no BMC 50 could be calculated due to the maximal response lying below a 50% change in response relative to the background.