PBTK modeling of the pyrrolizidine alkaloid retrorsine to predict liver toxicity in mouse and rat

Retrorsine is a hepatotoxic pyrrolizidine alkaloid (PA) found in herbal supplements and medicines, food and livestock feed. Dose-response studies enabling the derivation of a point of departure including a benchmark dose for risk assessment of retrorsine in humans and animals are not available. Addressing this need, a physiologically based toxicokinetic (PBTK) model of retrorsine was developed for mouse and rat. Comprehensive characterization of retrorsine toxicokinetics revealed: both the fraction absorbed from the intestine (78%) and the fraction unbound in plasma (60%) are high, hepatic membrane permeation is dominated by active uptake and not by passive diffusion, liver metabolic clearance is 4-fold higher in rat compared to mouse and renal excretion contributes to 20% of the total clearance. The PBTK model was calibrated with kinetic data from available mouse and rat studies using maximum likelihood estimation. PBTK model evaluation showed convincing goodness-of-fit for hepatic retrorsine and retrorsine-derived DNA adducts. Furthermore, the developed model allowed to translate in vitro liver toxicity data of retrorsine to in vivo dose-response data. Resulting benchmark dose confidence intervals (mg/kg bodyweight) are 24.1–88.5 in mice and 79.9–104 in rats for acute liver toxicity after oral retrorsine intake. As the PBTK model was built to enable extrapolation to different species and other PA congeners, this integrative framework constitutes a flexible tool to address gaps in the risk assessment of PA. Supplementary Information The online version contains supplementary material available at 10.1007/s00204-023-03453-z.


PBTK Model Ordinary Differential Equations
The following ordinary differential equations describe the change of amount of substance (mol) of retrorsine or its metabolites over time t (h) in the tissues (tis) of the PBTK model indicated below.All model parameters are listed in Table S1.
Retrorsine (RET) in peritoneum (for intraperitoneal administration): Retrorsine in gut lumen (for oral administration): Retrorsine in venous blood (for intravenous administration): V ven tis = adi, bon, bra, hea, kid, mus, ski (S3) Retrorsine in arterial blood: Retrorsine in lungs: Retrorsine in adipose, bone, brain, heart, muscle, skin or spleen: bon, bra, hea, mus, ski, spl (S6) Retrorsine in kidneys: Retrorsine in urine: Retrorsine in in gut tissue: Retrorsine in liver vascular/interstitial (superscript vi; also termed extracellular) space: Retrorsine in liver cellular (superscript c) space: Retrorsine in bile: GSH conjugates (DHR:GSH) in liver cellular space: Protein adducts (DHR:PROT) in liver cellular space: DNA adducts (DHR:DNA) in liver cellular space: Pre-processing of in vivo kinetic data Kinetic data from in vivo studies were standardized into one common base unit, amount of substance (nmol): Retrorsine (RET) data were either reported as % of dose in urine and bile or as µg/mL in plasma and liver.The conversion of RET data into amount of substance (nmol) is shown in Eqs.S16-S19.Glutathione conjugate (DHR:GSH) and protein adduct (DHR:PROT) data were reported as A Analyte /A IS /mg protein expressing the ratio of peak areas of analyte and internal standard (IS) in LC-MS/MS per mg protein of supernatant from liver homogenates.It was assumed that the mass spectrometric response of analyte and internal standard are similar.However, due to lack of mass spectrometric standards for quantification of DHR:GSH and DHR:PROT, data transformed to amount of substance (nmol) have to be regarded as relative quantities (Eqs.S20-S21).In the experimental protocolYang et al. (2017,2018) reported that mouse livers were homogenized and the resulting supernatants were used for mass spectrometric analysis.To account for the difference of protein amount in supernatants versus total liver homogenates, DHR:GSH and DHR:PROT data were multiplied by the protein ratio R prot of supernatant versus homogenate (Eqs.S20-S21), which was provided by personal correspondance with X. Yang (Yang et al. 2017).
DNA adduct (DHR:DNA) data were transformed from adducts/10 8 nucleotides to adducts/liver in nmol using scaling factors (Eq.S22).SF genome describes the number of base pairs/ haploid cell that is identical to the number of nucleotide pairs/ haploid cell.By multiplication with SF genome DHR:DNA were tranformed to adducts/ cell.We accounted for individual nucleotides by multiplication with factor 2, as well as for a diploid cell also using factor 2. The hepatocellularity SF liv scales adducts/cell to adducts/g liver.The Avogadro constant N A relates the number of hepatic adduct molecules to amount of substance.

Chemicals and Biological Materials
Retrorsine was purchased from AppliChem (Darmstadt, Germany) and from Phytoplan (Heidelberg, Germany).William's Medium E and supplements were puchased from PAN-Biotech (Aidenbach, Germany).Methanol and water (LC-MS grade) were obtained from Merck KGA (Darmstadt, Germany).All other chemicals and co-factors for liver microsomal assays were purchased from Carl Roth (Karlsruhe, Germany) or Sigma-Aldrich (Steinheim, Germany).Liver microsomes of CD-1 mice (11 weeks old) and Sprague-Dawley rats (eight to 10 weeks old) were obtained by Corning (Woburn, MA, USA).Characteristics of microsomal preparations are summarized in Table S3.Male CD-1 mice (eight weeks old) for isolation of primary hepatocytes were purchased from Janvier Labs (Le Genest-Saint-Isle, France).Male Sprague-Dawley rats (weighing 180 to 200 g) for isolation of primary hepatocytes were purchased from Charles River Laboratories (Wilmignton, MA, USA).The human colon adenocarcinoma cell line Caco-2 was obtained from the European Collection of Cell Cultures (ECACC, Porton Down, UK).

Isolation and Culture of Primary Hepatocytes
Primary hepatocytes were isolated from male CD-1 mice and male Sprague-Dawley rats by the two-step collagenase perfusion method.A detailed protocol of isolation including materials is given in Godoy et al. (2013) according to the protocol 'Isolation of primary rat and mouse hepatocytes' in Appendix I (see pages 1471-1475).Isolated cells were plated as a monolayer on 12-well plates onto dried collagen I at a density of 4.5 • 10 5 cells/mL in pre-warmed culture medium.The culture medium consisted of William's E medium supplemented with 100 U/mL penicillin/0.1 mg/mL streptomycin, 50 µg/mL gentamycin, 100 nM dexamethasone in EtOH, 20 mM L-glutamine, 2 ng/mL insulin and 10% fetal calf serum.Hepatocytes were incubated at 37°C under an atmosphere of 5% CO 2 and left for 3 h to attach.Then, cells were washed three times with William's E medium before the start of the medium loss assay and the cytotoxicity assay.
A detailed protocol of the collagen monolayer preparation procedure including materials is given in Godoy et al. (2013) according to the protocol 'Collagen monolayer protocol' in Appendix II (see page 1477).

Medium Loss Assay
Three hours after seeding, cultured hepatocytes were used for the medium loss assay.At the start of the experiment, cells were treated with 0.7 µM retrorsine diluted in serum-free incubation medium either at 37 • C or at 4 • C. Time-dependent retrorsine depletion was measured and therefore medium samples were taken at 0, 2.5, 5, 10, 30 and 60 min.Samples were stored at -20°C until they were diluted in 5% methanol and analysed for retrorsine by LC-MS/MS as reported in Geburek et al. (2020).Primary hepatocytes of three (rat) or two (mouse) animals were used as biological replicates and for each animal three technical replicates were examined.Measured retrorsine concentrations were normalized to the inital retrorsine concentration RET 0 .Retrorsine depletion-time profiles were described by a monoexponential decay model, where the parameter λ T (1/h) represents the rate constant of retrorsine loss in medium at the respective temperature T: A two-compartment model (Fig. S1) was used to derive the passive influx diffusion flow rate into the cells PS diff,in in vitro and the active uptake clearance CL act,in in vitro (both L/h/10 6 cells).PS diff,in in vitro and CL act,in in vitro were approximated under the assumption that passive diffusion out of the cells is negligible during the first monoexponential phase (Eqs.S24-S25).In vitro-to-in vivo extrapolation of both parameters was performed by multiplication with hepatocellularity SF liv and fraction unbound in vitro fu in vitro yielding in vivo PS diff,in and CL act,in (both L/h/g liver) (Eqs.S26 -S27).Fig. S1 Two-compartment model characterizing the uptake and clearance processes involved in the medium loss assay based on Schweinoch (2014).The external compartment represents the incubation medium surrounding the hepatocytes, while the internal compartment represents the hepatocytes' intracellular space.It was assumed that in the medium loss assay most compound exporting proteins are not active in unpolarized monolayers of hepatocytes (absence of biliary clearance CL bile and active efflux CL act,ef ).At a physiological temperature of 37°C active influx CL act,in , metabolism CL met,liv and passive diffusion PS diff,in , PS diff,out were involved in compound clearance (a).At a non-physiological temperature of 4°C active influx and metabolism were regarded as negligible (b).Note: PS diff,in and CL act,in (Eqs.S24-S27) were approximated under the assumption that passive diffusion out of the cells PS diff,out is negligible during the first monoexponential decay phase

Liver Microsomal Assay
Kinetics of retrorsine hepatic metabolism were determined by measuring retrorsine depletion in the presence of mouse and rat liver microsomes.Incubation mixtures were prepared on ice.Microsomal preparations were diluted in 50 mM Tris-HCl buffer (pH 7.5) to yield a protein concentration of 1 mg protein/mL.For simulation of phase I metabolism 33 mM potassium chloride, 8 mM magnesium chloride, 1 mM nicotinamide adenine dinucleotide phosphate (NADPH), 5 mM glucose-6-phosphate and 0.5 U/mL glucose-6-phosphate dehydrogenase were added. 2 mM glutathione were included to fasciliate formation of phase II glutathione conjugates.Mixtures were incubated at 37°C and 400 rpm with 1, 15, 50 and 200 µM of retrorsine.Reactions were stopped at 8, 15, 20, 30, 40, 50 and 60 min by addition of ice-cold methanol containing 1% ammonium formate.All experiments were performed in duplicate.Samples were vortexted and stored at -80°C.Thawed samples were centrifuged at 14,000xg at 4°C to precipitate salts and proteins.Supernatants were diluted in 5% methanol and were analysed for retrorsinne by LC-MS/MS as previously reported in Geburek et al. (2020).Measured retrorsine concentrations were normalized to the initial retrorsine concentration RET 0 .The following end-product inhibition model was developed to describe time-dependent retrorsine depletion: In the end-product inhibition model, we assumed that RET (µM) is converted to a mixture of products P with reaction rate constant k (1/h) (Eqs.S28-S30).Highly reactive dehydroretrorsine (DHR), one of RET's reaction products, was considered to unspecifically bind to the active center of microsomal enzymes thereby inhibiting their activity over time.The irreversible inhibition of CYP3A4 by resulting from metabolic activation of RET has been demonstrated in vitro by Dai et al. (2010).As a consequence k is inhibited in dependency of the product concentration.Half-maximal inhibition of k is achieved at the product concentration IC 50 (µM) (Eq.S31).k 0 represents the initial reaction rate constant where inhibition is still absent (t = 0).Its dependency on RET 0 was described by a Michaelis-Menten-like relationship (Obach and Reed-Hagen 2002) with maximum reaction velocity V max,liv,in vitro (µM/min) and RET concentration at half-maximal reaction velocity K M (µM).In vitro-to-in vivo extrapolation of V max,liv,in vitro yielded V max,liv (nmol/min/g liver): V max,liv =V max,liv,in vitro • 1 where: ρ prot Protein concentration (mg protein/mL): 1.00 (Experimental) fu in vitro Fraction unbound in vitro 1.00 (Assumption) SF microsomes Scaling factor (mg protein/g liver): 47.0 (mouse,rat) (Ring et al. 2011) fu in vitro was assumed 1, which was supported by the predicted value of 0.998 (Eq.for bases in Turner et al. (2007)).The predicted low protein binding is in line with the low lipophilicity of retrorsine (log P=-1.26).

Caco-2 Permeability Assay
Incubation of Caco-2 cells and transport experiments were performed as previously described in (Hessel et al. 2014).Briefly, Caco-2 cells were seeded on Transwell TM inserts (1.12 cm 2 growth area, 0.4 µm pore size, polycarbonate membrane; Corning B.V. Life Sciences, Amsterdam, The Netherlands) at a density of 6 •10 5 cells per insert and cultivated for 21 days.Before and after transport experiments the integrity of the cell monolayer was routinely checked by measuring the transepithelial electrical resistance.At the start of the experiment, 1 µM of retrorsine was placed either on the apical (A) or on the basolateral (BL) side.Medium samples (50 µL) from both sides were collected at 4, 8 and 24 h.All experiments were performed in triplicate.Samples were stored at -20 • C until they were analyzed for retrorsine by LC-MS/MS using the workflow reported in (Hessel et al. 2014).
The efflux ratio (ER) of retrorsine was calculated based on apparent permeability (P app , cm/s) values of A→BL and BL→A, where C acceptor (ng/mL) is the concentration in the acceptor compartment over time t (s), V acceptor (mL) is the volume of the acceptor compartment, C 0,donor (ng/mL) is the initial concentration in the donor comparment and A (cm 2 ) is the Transwell TM growth area.Data were reported as mean ± standard deviation.
The in vivo intestinal absorption F a (fraction) was predicted making use of the correlation between Caco-2 permeability and human intestinal absorption, which was reported for a set of 93 diverse compounds by Skolnik et al. (2010) (Eq.S35).We assumed that human F a is equal to that of rat and mouse.This assumption was made based on an in vivo study reporting a similar extent of intestinal absorption in rats compared to humans (Zhao et al. 2003).

Cytotoxicity Assay
The cytotoxicity assay was performed using the CellTiter-Blue® (CTB) assay as described in (Gu et al. 2018) according to the SOP in Supplement 3A.Three hours after seeding, cultured hepatocytes were incubated with concentrations of retrorsine ranging between 0 and 250 µM diluted in serum-free incubation medium for a time span of 48 hours.Cells incubated with 10% DMSO were used as a positive control, while cells incubated with medium containing 1.25% acetonitrile were used as solvent control.After incubation, the medium was removed and cells were washed with phosphate-buffered saline.100 µL serum-free medium containing 20% CTB reagent were added to each well.Supernatants were transferred to black polystyrene 96-well plates and fluorescence was read out with the Tecan Infinite M200 Pro plate reader.Primary mouse and rat hepatocytes of three animals were used as three biological replicates and for each animal three technical replicates were assessed.
Read-outs were corrected for background fluorescence and were normalized animal-wise to fluorescence of the solvent control (representing 100% cell viability at 0 µM retrorsine).The concentration-response curve was described by the sigmoidal inhibition model (Eq.S36), where IC 50 (µM) is the retrorsine concentration (RET) at half-maximal cell viability.
Cell viability = 100 1 + 10 (RET−log IC 50 ) (S36) Fig. S2 Time-dependent passage of retrorsine (RET) through the Caco-2 monolayer in the Transwell TM system used as a model for the small intestinal epithelium.Cells were exposed to 1 µM of RET either on the apical (A) or on the basolateral (BL) side.RET levels in the respective acceptor compartment are given as % of the initial RET concentration in the donor compartment of n=3 independent experiments (circles) with mean (solid line) ± standard deviation (error bars) Fig. S3 Pairs plot of 1000 Monte Carlo samples of estimated parameter distributions of the PBTK model of retrorsine.Diagnonal: histograms of the marginal posterior distribution of the parameters.Lower triangle: pairwise correlation plots with Loess smooth (red line).Upper triangle: Pearson correlation coefficients