Extended steroid profiling in H295R cells provides deeper insight into chemical-induced disturbances of steroidogenesis: Exemplified by prochloraz and anabolic steroids

Human adrenocortical H295R cells have been validated by the OECD Test Guideline 456 to detect chemicals disrupting testosterone and 17 β -estradiol (estradiol) biosynthesis. This study evaluated a novel approach to detect disturbances of steroidogenesis in H295R cells, exemplified by prochloraz and five anabolic steroids. Steroid profiles were assessed by an untargeted LC-MS-based method, providing a relative quantification of 57 steroids annotated according to their accurate masses and retention times. Such a panel of steroids included several mineralocorticoids, glucocorticoids, progestins and adrenal androgens. The coverage of a high number of metabolites in this extended steroid profiling facilitated grouping of chemicals with similar effects and detecting subtler differences between chemicals. It allowed, for example, distinguishing between the effects of turinabol and oxymetholone, supposed to act similarly in a previous characterization including only nine adrenal steroids. Furthermore, the results revealed that product/substrate ratios can provide superior information on altered enzyme activities compared to individual metabolite levels. For example, the 17 α -hydroxypregnenolone/preg- nenolone ratio was found to be a more sensitive marker for detecting 17 α -hydroxylase inhibition by prochloraz than the corresponding individual steroids. These results illustrate that chemical grouping and calculation of product/substrate ratios can provide valuable information on mode-of-action and help prioritizing further experimental work.

Cell-based steroidogenesis assays are useful tools for studying chemical effects on steroidogenesis, including direct enzyme inhibition, disturbances of gene expression and changes due to post-translational modifications. The Organization of Economic Cooperation and Development (OECD) and the United States Environmental Protection Agency (USEPA) Endocrine Disruptor Screening Program (EDSP) have developed test guidelines for an in vitro steroidogenesis assay based on H295R cells (EPA, 2009;Hecker et al., 2011;OECD, 2011). The human H295R adrenocortical carcinoma cell line expresses the main enzymes involved in steroidogenesis (Gazdar et al., 1990;Gracia et al., 2006) (see Fig. 1 for an overview of steroid biosynthesis), and has been validated in guideline 456 to test chemicals potentially disturbing testosterone and estradiol production (EPA, 2009;Hecker et al., 2011;OECD, 2011). This guideline did not define the analytical quantification method and accepted antibody-based steroid quantification, which often leads to overestimation due to cross-reactivity of the antibody used (Handelsman andWartofsky, 2013,Olesti et al., 2021).
Several groups worked on improved and extended protocols of the H295R-based steroidogenesis assay (Mangelis et al., 2016;Nakano et al., 2016;Rijk et al., 2012;Schloms et al., 2012;Strajhar et al., 2017). The use of targeted mass spectrometry-based methods has extended the initial measurement of testosterone and estradiol to include mineralocorticoids, glucocorticoids, progestins and adrenal androgens (Odermatt et al., 2016). Further modifications of the steroidogenesis assay include the use of stimulated cells, for example using the potent CYP11B1/2 inducer torcetrapib to facilitate the identification of compounds diminishing cortisol and/or aldosterone production (Akram et al., 2019), or the more general steroidogenesis inducer forskolin to detect a broader range of inhibitors (Karmaus et al., 2016;Patt et al., 2020;Schloms et al., 2012;Schloms et al., 2014;Swart et al., 2019;Swart and Smith, 2016;von Krogh et al., 2010). Targeted analysis of the main adrenal steroids combined with gene expression analyses, enzyme activity assays and molecular docking calculations provide a valuable toolbox to obtain information on the mechanisms underlying chemical-induced steroidogenesis disturbances. However, an individual evaluation of effects on all key steroidogenic enzymes is time-and cost-intensive. Thus, a comprehensive analysis of the chemical-induced alterations in the steroid profile could greatly help prioritizing further experimental investigations.
The present study aimed to explore whether an extended steroid profiling by untargeted steroidomics analysis combined with extended steroid annotation using exact mass and retention time (Gonzalez-Ruiz et al., 2018) can reveal a fast and comprehensive picture of alterations in a panel of progestins, mineralocorticoids, glucocorticoids and androgens. To assess the added value of an untargeted strategy compared to a targeted steroid profiling method, five anabolic steroids (also known as anabolic-androgenic steroids, AAS) and the reference compound prochloraz were selected from a preceding study using targeted ultra-high performance liquid chromatography tandem mass spectrometry (UHPLC-MS/MS) for steroid quantification (Patt et al., 2020): prochloraz, known to inhibit CYP17A1 and used as a positive control in the OECD test guideline 456; mesterolone, proposed to mainly inhibit CYP17A1; stanozolol, assumed to inhibit CYP17A1 with a prominent effect on 17,20-lyase activity; turinabol and oxymetholone, suggested to act very similarly by inhibiting CYP21A2 and CYP11B1, whereby turinabol seemed to be more potent; and danazol, which showed only moderate effects (see Suppl. Fig. 1 for chemical structures). In this earlier study, nine adrenal steroids were quantified in supernatants of H295R cells to assess potential effects on CYP17A1, CYP21A2, CYP11B1, CYP11B2 and 3β-HSD2 (Patt et al., 2020).
To detect chemical-mediated effects on steroidogenic enzyme activities, the present study aimed to compare the informative value of product/substrate ratios with that of individual steroid levels. Inhibition of an enzyme is expected to lead to an increase of substrate and decrease of product concentration. However, in an endogenous system the flux of an entire pathway and possible interferences due to inhibition of downstream and upstream enzymes need to be taken into account. Development and application of specific targeted LC-MS/MS methods are labor-intensive and are challenged by the limited availability of reference standards and restricted in the number of steroids that can be detected. Thus, a preselection for potentially affected pathways and involved metabolites is desirable. The extended steroid profiling approach used in this work allows a broad coverage of steroid metabolites, including several substrates and products for a given enzyme. In Fig. 1. Schematic representation of steroidogenesis pathways. Enzymes and steroids of steroidogenesis pathways in H295R cells are presented. Steroidogenesis pathways are colored as blue for progestins, yellow for mineralocorticoids, red for glucocorticoids, green for androgens and purple for estrogens. ACTH: adrenocorticotropic hormone; CYP: cytochrome P450; HSD: hydroxysteroid dehydrogenase; StAR: steroidogenic acute regulatory protein; SULT: sulfotransferase; SRD: steroid 5α-reductase. order to select the most suitable read-outs for detecting altered activities of steroidogenic enzymes, several product/substrate ratios were considered per enzyme.
2.4. Lentiviral construction of a COS-1 cell line stably expressing CYP17A1 and cytochrome P450 oxidoreductase (POR) pMDLg/pRRE and pRSV-rev (encoding the packaging signal for the virus) and pVSV-G (encoding the envelope signal) were a kind gift from Dr. Christoph Dehio and HEK293T from Dr. Michael Hall, University of Basel. HEK293T cells (1 × 10 6 ) were seeded in a T75-flask and transfected with 750 ng of pVSV-G, 1500 ng pMDLg/pRRE, 500 ng of pRSVrev and 500 ng of the lentiviral plasmid encoding POR and CYP17A1 using FuGene. The transfection medium was replaced by fresh medium after 5 h. After 24 h, the supernatant was collected, passed through a 0.45 μm filter and stored at − 20 • C until further use. COS-1 cells (30,000) were seeded in a T25-flask and incubated for 24 h before replacing the medium by 5 mL of transduction medium composed of 2.5 mL of the prepared viral supernatant from the HEK239T cells and 2.5 mL culture medium. Successfully transduced cells were selected applying 2 μg/mL puromycin for 14 d. Afterwards, 0.5 μg/mL puromycin was constantly added to the culture medium.

H295R steroidogenesis assay
The H295R steroidogenesis assay was accomplished according to the OECD test guideline No. 456 (OECD, 2011), except for certain modifications described previously (Patt et al., 2020). H295R cells between passages 5-10 were grown in 24-well plates (200,000 cells/mL) for 24 h, followed by a change to fresh phenol red-free medium containing the test compound (1 μM). A single concentration of 1 μM was chosen to allow a comparison of effects on the extended steroid profile by the different compounds, and because such a concentration can be reached in situations of AAS abuse. Culture supernatants were collected after 48 h of incubation and stored at − 20 • C until steroid hormones were quantified. Experiments were conducted three times independently, each performed in duplicates.
After another 2 h, the absorbance was monitored at 450 and 650 nm (reference wavelength). Experiments were performed three times independently with technical triplicates. None of the test compounds at concentrations up to 10 μM showed any effects on cell morphology or decreased cellular metabolic activity (Patt et al., 2020).

Determination of CYP11B1 enzyme activity
A cell-based CYP11B1 activity assay was performed as described previously with modifications . Briefly, V79-4 cells stably expressing CYP11B1 were seeded in 24-well plates (100,000 cells/well). After 18 h, the medium was changed to a serum-free medium containing either 0.03% DMSO (vehicle control), 3 μM of the respective AAS or prochloraz at final concentrations ranging from 1 nM to 3 μM. Cells were preincubated for 1 h. Afterwards, the medium was replaced by 500 μL serum-free medium containing the respective test compound and the CYP11B1 substrate 11-DOC at a final concentration of 0.1 μM. The supernatant was collected after 20 min and stored at − 20 • C, followed by preparation of samples for 11-DOC and corticosterone quantification by targeted UHPLC-MS/MS (see Suppl. Table 2 for steroid standards and calibrators used for targeted analyses).

Determination of CYP21A2 activity
COS-1 cells were seeded in 24-well plates (100,000 cells/well). After overnight incubation, cells were transfected with CYP21A2 and POR using FuGene (Promega, Madison, USA). Cells were incubated for another 18 h prior to the treatment as described above but using 17αhydroxyprogesterone as substrate. IC 50 curves were determined for prochloraz (10 nM-1 μM) and stanozolol (187.5 nM-3 μM). After 20 min of incubation, supernatants were frozen, followed by quantification of 17α-hydroxyprogesterone and 11-deoxycortisol using targeted UHPLC-MS/MS.

Determination of CYP17A1 hydroxylase activity
COS-1 cells stably expressing CYP17A1 and POR were seeded in 96well plates (50,000 cells/well). The medium was changed to serum-free medium containing the respective test compound, as described in paragraph 2.7, in a total volume of 25 μL. For IC 50 determination, 10 nM to 3 μM danazol, 1 nM to 10 μM stanozolol, and 1 nM to 3 μM prochloraz were applied. The reaction was started by adding 100 nM progesterone including 3 nM of a radiolabeled tracer. The reaction was stopped after 45 min by adding methanol containing an excess of unlabeled progesterone and 17α-hydroxyprogesterone at a final concentration of 1 mM. Steroids were separated by thin layer chromatography (TLC) using ethyl acetate and chloroform (1:4) (Acros Organics). The unlabeled progesterone and 17α-hydroxyprogesterone allowed visualization under the UV-light. Spots were excised and the radiolabeled substrate and product were analyzed by scintillation counting. Radioactive counts per minute (CPM) were normalized to the solvent control. A substrate conversion of 20-30% was reached after 45 min of incubation with the negative control containing DMSO.

Determination of CYP17A1-lyase activity
COS-1 cells (3 × 10 6 ) were transfected with 4000 ng CYP17A1 and 4000 ng POR using polyethylenimine (PEI, 25 kDa linear, Polyscience, Warringron, USA) and a PEI/DNA ratio of 3:1 (m/m). After 4 h, the medium was replaced by fresh medium and cells were incubated overnight. Transfected COS-1 cells were seeded in 96-well plates (50,000 cells/well). After 18 h, cells were incubated as described above in a total volume of 50 μL containing 100 nM 17α-hydroxypregnenolone. For IC 50 determination, 1 nM to 3 μM prochloraz were used. The supernatant was collected after 60 min and stored at − 20 • C until quantification of dehydroepiandrosterone (DHEA) by ELISA according to the manufacturer ((DEH3344, Demeditec, Kiel, Germany).

Targeted steroid quantification for CYP11B1 and CYP21A2 assay
Samples were spiked with deuterium-labeled reference compounds to a final concentration of 0.125 ng/mL each (for CYP11B1 assay: corticosterone-D8, for CYP21A2 assay: 17α-hydroxyprogesterone-D8 and 11-deoxycortisol-D5) and volumes were adjusted to 1 mL with water prior to solid phase extraction using Oasis HLB 3 cc cartridges (60 mg, 30 μm particle size, Waters, Massachusetts, USA). Columns were preconditioned with methanol and water, 3 mL each, prior to applying 950 μL of sample. Columns were washed with 3 mL each of water and methanol:water (10%) and with 1 mL methanol:water (40%) and dried for 5 min. Steroids were eluted twice with 0.75 mL methanol, followed by evaporation to dryness. Samples were reconstituted with 0.5 mL of methanol and diluted 1:4 (v:v; aqueous sample:methanol) prior to injecting 5 μL into the instrument. Steroids were analyzed by UHPLC-MS/MS using an Agilent 1290 Infinity II UHPLC coupled to an Agilent 6495 triple quadrupole mass spectrometer equipped with a jet-stream electrospray ionization interface (Agilent Technologies, Santa Clara, CA, USA). Analyte separation was achieved using a reverse-phase column (1.7 μm, 2.1 mm × 150 mm; Acquity UHPLC BEH C18; Waters) with mobile phases consisting of ACN, water and FA (95/5/0.1 and 5/ 95/0.1; v/v). Data acquisition and quantitative analysis was performed by MassHunter (Version B.10.0. Build 10.0.27, Agilent Technologies).

Untargeted steroid profiling
For steroid determination using an untargeted acquisition mode, samples were prepared as described previously (Patt et al., 2020), without relying on internal reference standards. After preconditioning with 1 mL ethyl acetate and 1 mL Milli-Q water, Oasis HLB 1 cc SPE cartridges (30 mg, 30 μm particle size, Waters), were applied to extract the samples. The columns were washed three times with H 2 O and a mixture of MeOH/H 2 O (10/90, v/v), followed by eluting the steroids twice with 0.5 mL of ethyl acetate. After evaporating eluates to dryness, the samples were reconstituted with 50 μL of MeOH:H 2 O 6:5 v/v. 9 μL of each sample were taken and pooled to produce quality control (QC) samples. QC samples were injected at the beginning of the analytical sequence in order to condition the system, and every six samples to monitor the analytical performance of the LC-MS (Pezzatti et al., 2020).
Chromatography was performed on a Waters H-Class Acquity UHPLC system (Waters) coupled to a maXis 3G Q-TOF high-resolution mass spectrometer from Bruker (Bruker Daltonik GmbH, Bremen, Germany) through an electrospray interface (ESI) operated in positive ionization mode. Data between 50 and 1000 m/z were acquired in profile mode at a rate of 2 Hz. ESI and MS parameters were optimized using a mix of standards fed by a syringe pump and mixed with the LC eluent (mid-gradient conditions) through a tee-junction. MS and UHPLC control and data acquisition were performed through the HyStar v3.2 SR2 software (Bruker Daltonik) running the Waters Acquity UHPLC v.1.5 plug-in. Formate adducts in the 90-1247 m/z range were employed for automatic m/z calibration on each run (see Suppl. Data file for information on the 57 identified steroids).

Data analysis
Run alignment and peak piking were performed on Progenesis QI v2.3 (Nonlinear Dynamics, Waters, Newcastle upon Tyne, UK). Abundances were calculated as peak areas, and median-normalized using the values of the control samples of each biological batch. Analytical drift correction (LOESS regression using QC samples) and data filtering (features exhibiting relative standard deviation above 30% in QC samples) were performed using SUPreMe, an in house developed software for metabolomics data pre-processing.
Steroids were identified by their accurate masses and retention times (Rt) from different data sources. First, Rt and accurate masses were acquired in house for a set of 92 steroids commercially available, using the same LC-MS setup and conditions as the samples. Then, and in addition to the authentic standards, annotation was extended by using DynaSti (Codesido et al., 2019), a dynamic retention time database for steroids freely available at https://dynasti.vital-it.ch/. DynaSti can calculate Rt in a reverse-phase setup for any gradient profile by applying the linear solvent strength model and using steroids' molecular descriptors (Randazzo et al., 2016). Such descriptors have been experimentally measured for 92 steroids and calculated in silico for 106 more. DynaSti uses the same 92 steroid standards, which were used to experimentally measure Rt, so only the 106 in silico predictions were taken from DynaSti in the present work. The following tolerances were used for steroid annotation: 2 ppm mass error, 1% Rt error for steroids with experimentally measured Rt and masses, and 2% Rt error for steroids with only in silico calculated properties available. In total, 53 steroids were annotated by direct match to their experimentally measured accurate masses and retention times, and 4 were annotated thanks to in silico predicted parameters.

Statistics and multivariate analysis
Multivariate analysis was performed using SIMCA 15.0.2 (Sartorius Stedim Data Analytics AB, Umeå, Sweden). Data was centered and UVscaled, and Ward's method was used for hierarchical cluster analysis (HCA). Heat maps were plotted using Python 3.8 and Pandas 1.1.0 (The pandas development team, 2020).
Univariate statistical evaluation was performed in GraphPad Prism version 8.0.2. For analysis of data obtained from three independent experiments (k = 3) each performed in duplicate (n = 2), Shapiro-Wilk normality test was applied to verify the normality of data (N = k × n). Two-tailed Student's t-test for independent samples was applied to calculate statistical significance of differences between chemical treatments and the solvent control, followed by a Bonferroni correction to control the false positive rate. Values represent mean ± SD and were considered significant at a p-value <0.05.

Results and discussion
The effects of various AAS on adrenal steroidogenesis has been investigated recently in H295R cells using targeted quantification (UHPLC-MS/MS) of nine steroid metabolites (Patt et al., 2020). To be Fig. 3. Effects of test compounds on the extended steroid profile in H295R cells. H295R cells were treated for 48 h with solvent as control (SC; 0.01% DMSO) or the corresponding test compound at a concentration of 1 μM. Qualitative analysis of alterations in steroid hormone levels was performed by UHPLC-MS. Data are expressed as fold changes relative to SC and are presented as mean ± SD from three independent experiments, each performed in duplicate. Steroid metabolites significantly downregulated compared to SC are depicted in green and those significantly upregulated in red. Differences with p < 0.05 were considered significant. *Values obtained from previous targeted measurements (Patt et al., 2020). n.d.: not detected. able to detect subtler differences in the effects of the compounds on adrenal steroidogenesis, an untargeted UHPLC-MS acquisition with an extended annotation and identification strategy was applied. This approach allows a much broader coverage of the steroid profile. This should provide a better perspective of the chemical-induced disturbances on steroidogenesis, granting access to the detection of inhibitory effects on a larger number of enzymes. This approach also has the potential to group chemicals according to their ability of altering certain enzymatic activities. Therefore, this newly introduced strategy offers an exploratory first unbiased assessment of substances to prioritize further investigations.

Multivariate analysis of steroid metabolites in H295R cells
A multivariate data analysis (MVA) was performed to explore the complexity of the acquired data (57 assigned steroids, 7 treatments including DMSO control, 6 replicates each) by grouping AAS according to their induced disturbances and investigating potentially similar modes of actions. Hierarchical cluster analysis (HCA) allows assessing and visualizing natural partitions of the dataset using a tree structure by simultaneously maximizing differences between clusters (separability) and homogeneity within a given subset (compactness). The degree of similarity between the different AAS and prochloraz are presented in a dendrogram, revealing two major clusters ( Fig. 2A): The first included samples of turinabol and oxymetholone ( Fig. 2A), both showing diminished mineralocorticoid and glucocorticoid levels but elevated adrenal androgen levels (Fig. 3), while the second encompassed mesterolone, stanozolol and danazol ( Fig. 2A), showing increased mineralocorticoid and decreased adrenal androgen levels (Fig. 3). The effects of danazol remained undefined in the earlier study using targeted steroid quantification (Patt et al., 2020). This second cluster also included prochloraz, although mineralocorticoids were not decreased and it is a bit separated from the three AAS. The extended steroid profiling of the current study allowed further clustering of the different AAS with respect to their distinct mode-of-action. The first cluster revealed grouping of oxymetholone with DMSO, suggesting that oxymetholone exerts weaker effects on steroidogenesis than turinabol. The second cluster isolated prochloraz, most likely due to its potent inhibition of several steroidogenic CYP enzymes (discussed below) and differentiated danazol from stanozolol and mesterolone, thus suggesting that different effects are involved. Moreover, danazol did not cluster closely with DMSO, in contrast to the weak effects found in the previous study using targeted steroid quantification (Patt et al., 2020).
Principal component analysis (PCA) provided a complementary overview of the various treatments and highlighted steroid metabolites associated with the observed trends. Having 57 dimensions/variables (steroids) and 7 treatments, PCA was used to reduce the dimensionality of the datasets while retaining the highest possible amount of variability using a linear combination of the initial variables. The PCA score plot summarized more than 50% of the total variance confirming the HCA results (Fig. 2B): mesterolone, stanozolol and danazol clustered closely but separately from other samples, DMSO and oxymetholone were grouped, and prochloraz and turinabol were separated from all other treatments. The loadings plot added further information by showing the main responsible dimension for the AAS location on the PCA plot (Fig. 2C). Mineralocorticoid accumulation (upper right corner) in samples treated with mesterolone and stanozolol mainly accounted for their distinct behavior. Accumulation of mineralocorticoids by mesterolone and stanozolol might be attributed to CYP17A1 inhibition as already observed in the targeted study (Patt et al., 2020). An increased production of progestins (lower right corner) in samples of prochloraz suggests that this compound inhibited CYP17A1 and additional enzymes such as CYP21A2. Turinabol samples exhibited a different pattern, characterized by reduced mineralocorticoid production and increased androgen levels (lower left corner) ( Fig. 2B and C). Turinabol samples correlated closely with oxymetholone samples, although the effects evoked by oxymetholone seemed less pronounced due to its proximity to the DMSO solvent control (Fig. 2B). These observations were found to be consistent with the results obtained in the targeted study (Patt et al., 2020). To fully exploit the advantage of MVA and gain a global overview of potential similarities between the various steroidomic profiles, a heat map comprising a HCA dendrogram of the 57 steroids was computed, highlighting subgroups of metabolites underlying the steroid hormone patterns of test chemicals (Suppl. Fig. 2).

Investigation of effects by prochloraz on CYP17A1, CYP21A2 and CYP11B1 to evaluate the presented approach
Changes of hormone levels in the supernatants of H295R cells exposed to the test compounds were analyzed relative to the DMSO solvent control and presented as fold changes (Fig. 3). Instead of focusing on the behavior of individual metabolites, the present approach aimed at making use of the extended steroid profile for assessing product/substrate ratios of steroidogenic enzymes expressed in H295R cells to detect chemical-induced disturbances (Fig. 4). Product/substrate ratios are applied in clinical diagnosis of impaired functions of steroid metabolizing enzymes, including CYP21A2, CYP17A1, CYP11B1 and 11β-HSD2 (Hicks et al., 2014;Medeiros et al., 2013;Nomura et al., 1996;Thompson et al., 2017). It needs to be noted that such product/substrate ratios are affected by any other enzyme metabolizing either the substrate or the product and that the flux of steroidogenic pathways needs to be kept in mind. The use of more complex ratios, including downstream metabolites, may be considered. For reasons of simplicity, only single product/substrate ratios were used in this study. Despite their limitations, such ratios can provide initial mechanistic information on disrupted enzyme activities.
The extended steroid profile after prochloraz treatment, which was shown earlier to inhibit CYP17A1 activity and at higher concentrations also CYP21A2 and CYP19A1 but not CYP11B1 (Blystone et al., 2007;Higley et al., 2010;Ohlsson et al., 2009;Sanderson et al., 2002), confirmed changes observed in the previous targeted UHPLC-MS/MS-based study (Patt et al., 2020). Except for non-17α-hydroxylated progestins that were highly elevated, most steroid concentrations were decreased (Fig. 3). Using untargeted steroidomics, information on additional steroids could be obtained, providing further evidence on the functionality of particular enzymes, which would have otherwise remained inaccessible by using conventional targeted approaches. Pregnenolone and 17α-hydroxypregnenolone levels provided valuable information on CYP17A1 activity, both being more predictive than progesterone and 17α-hydroxyprogesterone, since they are a primary substrate-product pair and less influenced by 3β-HSD2 and CYP21A2 activities. Thus, 17α-hydroxypregnenolone and pregnenolone should be integrated into methods for targeted measurements in follow-on studies. Upon inhibition of an enzyme, its product and substrate concentrations are expected to be differentially altered and the corresponding ratio may provide more reliable information on the enzyme activity. The product/substrate ratios 17α-hydroxypregnenolone/pregnenolone and 17α-hydroxyprogesterone/progesterone may serve as indicators for an inhibition of CYP17A1 hydroxylase. These ratios were significantly reduced after prochloraz treatment, indicating a potent inhibition of 17α-hydroxylase in agreement with the PCA results ( Fig. 2B and C; Fig. 4). This hypothesis was then tested by performing a 17α-hydroxylase activity assay using overexpressed human enzyme, confirming the potent inhibition by prochloraz with an IC 50 value of 48 nM (Fig. 5, Table 1), in line with the literature (Blystone et al., 2007;Ohlsson et al., 2009).
The 17α-hydroxylase inhibition affects the substrate of the 17,20lyase reaction, making an assessment of the 17,20-lyase activity difficult. The reduced androstenedione/17α-hydroxyprogesterone ratio (Fig. 4) suggests an impaired 17,20-lyase activity, although this ratio should be interpreted cautiously because the human 17,20-lyase does not efficiently catalyze the formation of androstenedione (Auchus et al., 1998). A more reliable marker for 17,20-lyase activity would be the DHEA/17α-hydroxypregnenolone ratio. Since the sensitivity of the untargeted approach was not sufficient to detect DHEA, and this metabolite was also below LLOQ upon treatment with prochloraz in the targeted assay (Patt et al., 2020), the DHEA/17α-hydroxypregnenolone ratio could not be calculated. The CYP17A1 17,20-lyase activity assay based on overexpressed human enzyme confirmed the potent inhibition by prochloraz, with an IC 50 value of 301 nM (Fig. 5, Table 1), in line with the results from the present H295R assay and data in the literature (Blystone et al., 2007;Ohlsson et al., 2009).
The observed increase in 11-DOC along with the decrease in corticosterone may be explained by the inhibition of CYP21A2; however, this could partly be due to CYP11B1 inhibition. In contrast to the targeted study including only two substrates and products of CYP11B1 (Patt et al., 2020), the extended steroid profiling approach enabled the calculation of four different steroid ratios: the direct product/substrate ratios cortisol/11-deoxycortisol and corticosterone/11-DOC, and the alternative ratios 11β-hydroxyandrostenedione/androstenedione and 11β-hydroxyetiocholanolone/etiocholanolone. The first two ratios depend on an intact CYP21A2 activity and the 11β-hydroxyandrostenedione/androstenedione ratio on an intact CYP17A1 17,20-lyase activity. Whilst androstenedione is synthesized de novo by the H295R cells, etiocholanolone, like testosterone , is mainly contributed by the addition of Nu-serum, and the 11β-hydroxyetiocholanolone/etiocholanolone ratio may thus serve as an indicator of CYP11B1 activity. Since this ratio tended to be decreased (Fig. 4), CYP11B1 enzyme activity was measured in stably transfected V79 cells, revealing an IC 50 value of 171 nM (Fig. 5, Table 1). This finding contradicts an earlier study in H295R cells, reporting an effect of prochloraz on the expression of CYP11B1 but not directly on its enzyme activity (Ohlsson et al., 2009).
CYP11B2 converts 11-DOC to aldosterone, with corticosterone and 18-hydroxycorticosterone being intermediates that are generally thought not to be released from the catalytic site of CYP11B2 (Strushkevich et al., 2013). The aldosterone/11-DOC ratio was significantly reduced (Fig. 4), implying a block of CYP11B2. This mirrors CYP11B1 inhibition, explained by the fact that CYP11B1 and CYP11B2 share 93% homology and inhibitors are rarely selective (Lenzini et al., 2021;Mornet et al., 1989). This observation opposes the already mentioned study by Ohlsson et al., reporting no direct inhibition of CYP11B1 and CYP11B2 activities based on their data from H295R cells (Ohlsson et al., 2009) .   Fig. 4. Impact of test compounds on key product to substrate ratios. H295R cells were incubated for 48 h with the solvent as control (SC; 0.01% DMSO) or the respective test compound at a concentration of 1 μM. Steroid hormone levels were qualitatively analyzed using UHPLC-MS and product to substrate ratios of secreted metabolites were formed relative SC. Data are expressed as fold changes relative to SC and are denoted as mean ± SD obtained from three independent experiments, each performed in duplicate. Product to substrate ratios significantly downregulated compared to SC are depicted in green and ratios significantly upregulated in red. Differences with p < 0.05 were considered significant. n.a.: not analyzed. Preg: pregnenolone. Prog: progesterone. DHEA: dehydroepiandrosterone. AD: androstenedione. 11-DOC: 11-deoxycorticosterone. F: cortisol. B: corticosterone. Et: etiocholanolone. Aldo: aldosterone. E1: estrone. E2: estradiol. Testo: testosterone. Andro: androsterone. DHP: dihydroprogesterone. DHDOC: dihydro-11-deoxycorticosterone. DHB: dihydrocorticosterone.

Exploitation of the androgen profile to study effects of prochloraz on testosterone in H295R cells
The H295R assay was initially designed to detect chemical-induced alterations in testosterone and estradiol biosynthesis (Hecker et al., 2011), two steroids not produced in a healthy adrenal and emphasizing the fact that this cell line was derived from an adrenal tumor. The final step of testosterone biosynthesis involves 17β-HSD3 in the testis and AKR1C3 in the prostate; however, only AKR1C3 was found to be expressed in H295R cells Xing et al., 2011). A previous study indicated that AKR1C3 is responsible for the low amounts of de novo testosterone production in H295R cells . AKR1C3 also catalyzes the oxoreduction of progesterone at position 20 of the steroid backbone. The ratios testosterone/androstenedione and 17α,20α-dihydroxyprogesterone/17α-hydroxyproge sterone are influenced by CYP17A1 inhibition and are therefore not suitable to study prochloraz-mediated effects on AKR1C3. A possibly useful ratio is estradiol/estrone because both estrogens are mainly contributed to the system by the addition of Nu-serum and de novo estrogen formation in H295R cells is very low (Strajhar et al., 2017).  5. Inhibition of CYP17A1, CYP21A2 and CYP11B1 activities by prochloraz and AAS. The conversion of progesterone to 17αhydroxyprogesterone (A and B), 17αhydroxypregnenolone to DHEA (C and D), 17α-hydroxyprogesterone to 11-deoxycortisol (E and F) and 11-DOC to corticosterone (G and H) were measured in intact cells overexpressing human recombinant enzyme. Product formation was quantified as follows: corticosterone and 11-deoxycortisol concentrations for CYP11B1 and CYP21A2 activity, respectively, were measured using a targeted UHPLC-MS/MS method, 17αhydroxyprogesterone formation from radiolabeled progesterone (CYP17A1-hydroxylase) was determined using TLC separation and scintillation counting, and DHEA production (CYP17A1-lyase) was measured using an ELISA kit. IC 50 curves (B, D, F and H) were prepared when the conversion rate was inhibited by more than 60% in the presence of 3 μM of prochloraz or AAS (A, C, E and G). Inhibition curves were analyzed by nonlinear regression. Data were normalized to the control and represent mean ± SD of three independent experiments.
Another ratio is 20α-hydroxyprogesterone/progesterone, which was moderately reduced (Fig. 4), suggesting AKR1C3 inhibition. Nevertheless, the low AKR1C3 activity in H295R cells  should be taken into account and interpretations drawn carefully.
Testosterone may be metabolized by aromatase (CYP19A1), besides other enzymes, and the two ratios estradiol/testosterone and estrone/ androstenedione (Fig. 4) represent markers to assess its activity. While the estradiol/testosterone ratio was significantly decreased, the estrone/ androstenedione ratio was not (Fig. 4). Because an earlier study suggested low CYP19A1 expression in H295R cells and found no effect of the specific CYP19A1 inhibitor letrozole on the levels of androstenedione and testosterone , these ratios need to be interpreted cautiously and to conclude on CYP19A1 inhibition, enzyme activity assays should be performed. Nevertheless, evidence for an inhibition of CYP19A1 by prochloraz has been provided (Higley et al., 2010;Sanderson et al., 2002).
16α-Hydroxylase (CYP3A7) is a key enzyme in the fetal production of 16α-hydroxydehydroepiandrosterone, the precursor of estriol (Caron-Beaudoin et al., 2017). By inhibiting CYP17A1, prochloraz treatment abolished the production of DHEA (Fig. 3). Thus, the ratios 16α-hydroxydehydroepiandrosterone/DHEA and 16α-hydroxyandrostenedione/androstenedione are not suitable for evaluating CYP3A7 activity in this situation. Also the 16α-hydroxyprogesterone/progesterone ratio is of limited use, because 16α-hydroxyprogesterone can be generated by CYP17A1, which is inhibited by prochloraz. A useful marker of CYP3A7 inhibition represents the 16α-hydroxyandrosterone/androsterone ratio, which was found to be reduced (Fig. 4), suggesting inhibition of the enzyme.

Effects of selected AAS on H295R steroid profiles
Steroid profiles of H295R cells treated with AAS were analyzed in the same manner as described for prochloraz, with a focus on product/ substrate ratios to detect altered enzyme activities. Conclusions drawn on CYP21A2, CYP11B1 and CYP17A1 activities were further evaluated by performing enzyme activity assays. Intracellular access of the AAS was verified by androgen receptor (AR) activation experiments. All compounds successfully activated the AR and were thus able to enter the cells (data not shown).
Similar to prochloraz, mesterolone was found to significantly lower the 17α-hydroxypregnenolone/pregnenolone ratio (Fig. 4), supporting the previous assumption of CYP17A1 hydroxylase inhibition (Patt et al., 2020) and providing an explanation for its clustering with the compound group containing prochloraz as shown in Fig. 2A. However, the mechanism underlying mesterolone-mediated interference with steroidogenesis remains unclear because no inhibition could be detected in the CYP17A1 hydroxylase assay (Fig. 5) and a previous study found no altered CYP17A1 mRNA expression upon exposure of H295R cells to mesterolone (Patt et al., 2020). Inhibition of cytochrome P450 reductase POR seems also unlikely as this would be expected to affect also CYP21A2 activity (Auchus, 2017). The 11-DOC/progesterone ratio did not indicate CYP21A2 inhibition, which was confirmed by the enzyme activity assay (Figs. 4 and 5). Also 3β-HSD2 activity seemed not to be affected, considering the unchanged progesterone/pregnenolone ratio (Fig. 4). Furthermore, the activities of CYP11B1, CYP11B2 and CYP19A1 were not affected by mesterolone treatment. Assessment of AKR1C3 activity, when accompanied by CYP17A1 inhibition, is complicated, because most ratios reporting AKR1C3 activity rely on substrates or products of CYP17A1. However, the estradiol/estrone ratio seems to be appropriate. In addition, the 20α-hydroxyprogesterone/progesterone ratio might be used. Nevertheless, both ratios were not significantly altered, suggesting that AKR1C3 activity was unaffected. Moreover, the decrease of the 5α-dihydroprogesterone/progesterone and 5α-dihydrocorticosterone/corticosterone ratios may indicate a reduced SRD5A1 activity (Fig. 4). Importantly, hypotheses drawn on potential interferences with CYP19A1, AKR1C3, CYP3A7 and SRD5A1 need to be validated in follow-on experiments.
Taken together, the extended steroid profiling and calculation of enzyme product/substrate ratios confirmed the earlier findings that mesterolone inhibits CYP17A1 (Patt et al., 2020), and provided information on additional enzymes likely affected by this AAS such as CYP3A7 and SRD5A1. Whether the mesterolone-induced effects are due to direct inhibition or altered expression of the respective enzymes remains to be investigated.
Markedly enhanced levels of mineralocorticoids (aldosterone, corticosterone and 11-DOC) and progestins (progesterone and 17αhydroxyprogesterone) along with diminished production of 11-deoxycortisol and androstenedione were similarly associated with stanozolol (Patt et al., 2020), which closely clustered with mesterolone according to HCA (Fig. 2A). Further information on the impact of stanozolol was gained by analyzing the 17α-hydroxypregnenolone/pregnenolone and DHEA/17α-hydroxypregnenolone ratios, reflecting CYP17A1 hydroxylase and 17,20-lyase activity, respectively. Both ratios were slightly reduced, suggesting an inhibition of CYP17A1 with a more pronounced effect on the lyase activity (Fig. 4). Previous results from cell-free activity assays based on human and pig CYP17A1 are in line with potent lyase and less pronounced hydroxylase inhibition of stanozolol (Nakajin et al., 1989;Patt et al., 2020). In contrast, more pronounced inhibition of the 17α-hydroxylase than the 17,20-lyase activity was observed in activity measurements performed in intact cells overexpressing CYP17A1 (Fig. 5, Table 1). Product/substrate ratios provided additional information on stanozolol-mediated steroidogenesis disturbances. Both ratios, 11-DOC/progesterone and 11-deoxycortisol/17α-hydroxyprogest erone (depending on CYP17A1 influenced steroids that might be subject to changes) were significantly reduced (Fig. 4), implying inhibition of CYP21A2 activity, which could be confirmed by enzyme activity measurements (Fig. 4, Table 1). This first indication of CYP21A2 inhibition by stanozolol was overlooked in the targeted study, emphasizing that the ratios obtained from the untargeted analyses are more informative than a restricted number of individual metabolites.
Danazol grouped with mesterolone and stanozolol on the PCA score plot ( Fig. 2A and B) and was shown previously to moderately affect the steroid profile of H295R cells, with weak inhibitory effects on progestins, glucocorticoids and adrenal androgens, not allowing Table 1 IC 50 values of selected AAS and prochloraz for CYP17A1, CYP21A2 and CYP11B1. The conversion of progesterone to 17α-hydroxyprogesterone, 17αhydroxypregnenolone to DHEA, 17α-hydroxyprogesterone to 11-deoxycortisol, and 11-DOC to corticosterone were measured in the presence of different concentrations of prochloraz or AAS in intact cells transfected with recombinant enzymes using targeted analytical readouts. Inhibition curves were analyzed by nonlinear regression and data were normalized to the control. Values represent mean ± SD of three independent experiments.
conclusions on possibly involved enzymes (Patt et al., 2020). Nevertheless, the significantly decreased 17α-hydroxypregnenolone/pregnenolone and progesterone/pregnenolone ratios suggest a combined inhibition of 3β-HSD2 and CYP17A1 hydroxylase activity (Fig. 4). In line with this, activity measurements confirmed inhibition of 17α-hydroxylase (IC 50 of 3.25 μM) and also but less potently of 17,20-lyase activity by danazol (Fig. 5, Table 1). CYP17A1 and 3β-HSD2 inhibition by danazol were also reported earlier (Arakawa et al., 1989;Barbieri et al., 1977;Betz et al., 1981). A different steroid pattern was obtained after the incubation of cells with turinabol and oxymetholone, which represented a separate cluster ( Fig. 2A and B). In our previous study, turinabol and oxymetholone were found to act very similarly, whereby turinabol exhibited more pronounced effects on the H295R steroid profile (Patt et al., 2020). Due to reduced levels of mineralocorticoids and glucocorticoids along with enhanced DHEA and a trend increase in 17α-hydroxyprogesterone, an inhibition of CYP21A2 activity combined with a possible block of CYP11B1/2 was proposed. Whilst the results obtained from the targeted measurements could be confirmed, the extended analysis provided further insights into the mode-of-action of turinabol and oxymetholone. For both AAS, the ratios progesterone/pregnenolone and 17α-hydroxyprogesterone/17α-hydroxypregnenolone were significantly decreased (Fig. 4), indicating an inhibition of the upstream enzyme 3β-HSD2 rather than CYP21A2. In line with this, activity measurements showed no CYP21A2 inhibition (Fig. 5, Table 1). In this regard, the levels of pregnenolone and 17α-hydroxypregnenolone revealed important information on the potential inhibition of 3β-HSD2 by turinabol and oxymetholone that was missed in the targeted study (Patt et al., 2020). This demonstrates the importance of including these metabolites in future targeted approaches. Another benefit of the extended analysis is the high number of metabolites providing a more complete picture of the effects on steroidogenesis of test compounds. While treatments with turinabol and oxymetholone led to comparable effects on the nine metabolites measured by targeted quantification (Patt et al., 2020), distinct effects of these two AAS were identified by assessing 57 different steroids using an untargeted analytical approach (Fig. 3). Turinabol, but not oxymetholone, led to significantly diminished levels of the glucocorticoids 11-deoxycortisol and cortisol and of 11β-hydroxyandrostenedione along with highly augmented levels of androsterone and 11β-hydroxyepiandrosterone. Furthermore, metabolite ratios indicate that turinabol inhibits CYP19A1 and has a more pronounced impact on CYP3A7 than oxymetholone (Fig. 4). The effects of the two AAS are difficult to classify, as they affect several enzymes and also due to the steroid flux; however, the extended steroid profiling approach allowed distinguishing their effects. Further investigations are required to unravel their exact mode-of-action.

Strengths, limitations and outlook
The current study was designed to explore whether an extended panel of progestins, mineralocorticoids, glucocorticoids, androgens and estrogens measured by an untargeted steroidomics approach can provide more information on chemical-induced disturbances of steroidogenesis than earlier methods using testosterone and estradiol (Hecker et al., 2011) or a few steroid metabolites as read-out (Akram et al., 2019;Karmaus et al., 2016;Maglich et al., 2014;Mangelis et al., 2016;Nakano et al., 2016;Patt et al., 2020;Rijk et al., 2012;Schloms et al., 2012;Strajhar et al., 2017). Our extended steroid profiling provided useful information on a large number of individual metabolites and on product/substrate ratios to detect altered enzyme activities after incubation with prochloraz or the selected AAS. The proposed use of product/substrate ratios to obtain initial information on altered enzyme activity is derived from clinical diagnosis where such ratios are used to detect impaired steroidogenic enzyme activity. Although providing initial insights into altered enzyme activities and facilitating the prioritization of further experimental work, such ratios are limited by the fact that products can be further metabolized and the flux of steroidogenic pathways is not considered. More complex ratios taking into account additional upstream precursors and downstream products should be evaluated in follow-on studies. Nevertheless, hypotheses drawn from changes of individual metabolites or ratios thereof should be evaluated by performing enzyme activity assays, as exemplified in this study for CYP17A1, CYP21A2 and CYP11B1. In addition, gene and protein expression analyses will help to identify effects caused by indirect modulation of the involved enzymes.
The analysis of large datasets such as the measurements of numerous steroid metabolites is challenging and MVA methods seemed to be the most adequate option for this purpose. In contrast to the evaluation of individual steroids, these tools facilitate compact data presentation and allow grouping of chemicals according to their phenotypic response in H295R cells by using metabolite levels as a readout. This highlights detecting differences between chemicals producing similar steroid patterns and provides diagnostic tools for the detection of biomarkers (Boccard and Rudaz, 2013). In this project, this approach allowed distinguishing the effects of turinabol and oxymetholone that showed a similar pattern in the targeted measurements comprising only nine steroids.
The current proof-of-concept study included only a single concentration to compare effects of different chemicals. Follow-on research should evaluate concentration-dependent effects because chemicals causing unwanted effects are rarely specific and exhibit concentrationdependent effects. To clearly understand the resulting steroid pattern in H295R cells, the application of highly specific reference inhibitors or siRNA would help to define steroid patterns specific for the disruption of each enzyme. This would facilitate grouping of chemicals and prioritizing further experimental work. However, such reference inhibitors are not available and the use of siRNA is complicated by the time needed to achieve sufficient knock-down and further incubation required for the evaluation of cellular effects.
H295R cell-specific limitations that should be considered include the addition of Nu-serum, which contributes steroids to the system that influence the steroid pattern, the challenging handling of H295R cells including the short range of passage numbers, and the application of appropriate controls (vehicle, complete medium comparison at time zero) .
The present study provided evidence that an extended, untargeted steroid profiling in combination with comprehensive data analysis and the use of substrate/product ratios can help to identify new potential informative biomarkers of chemical-induced alterations of adrenal steroidogenesis. This can lead to a broader insight into the biochemical effects of a test compound, which together with the grouping of chemicals according to their mode-of-action is highly valuable for the prioritization of follow-on experiments.

Declaration of competing interest
The authors declare no conflict of interest.

Data availability
Data will be made available on request.