The Role of Kinetics as Key Determinant in Toxicity of Pyrrolizidine Alkaloids and Their N-Oxides.

Pyrrolizidine alkaloids (PAs) are a large group of plant constituents of which especially the 1,2- unsaturated PAs raise a concern because of their liver toxicity and potential genotoxic carcinogenicity. This toxicity of PAs depends on their kinetics. Differences in absorption, distribution, metabolism, and excretion (ADME) characteristics of PAs may substantially alter the relative toxicity of PAs. As a result, kinetics will also affect relative potency (REP) values. The present review summarizes the current state-of-the art on PA kinetics and resulting consequences for toxicity and illustrates how physiologically-based kinetic (PBK) modelling can be applied to take kinetics into account when defining the relative differences in toxicity between PAs in the in vivo situation. We conclude that toxicokinetics play an important role in the overall toxicity of pyrrolizidine alkaloids. and that kinetics should therefore be considered when defining REP values for combined risk assessment. New approach methodologies (NAMs) can be of use to quantify these kinetic differences between PAs and their N-oxides, thus contributing to the 3Rs (Replacement, Reduction and Refinement) in animal studies. The Author(s). This is an open access article published by Thieme under the terms of the Creative Commons Attribution-NonDerivative-NonCommercial License, permitting copying and reproduction so long as the original work is given appropriate credit. Contents may not be used for commecial purposes, or adapted, remixed, transformed or built upon. (https://creativecommons.org/licenses/by-nc-nd/4.0/).

Introduction

Pyrrolizidine alkaloids (PAs) are a large group of plant constituents of which especially the 1,2-unsaturated PAs raise a concern because of their liver toxicity and potential genotoxic carcinogenicity. This toxicity of PAs is influenced by kinetics, and differences in absorption, distribution, metabolism, and excretion (ADME) characteristics of PAs may substantially alter the relative toxicity of PAs. As a result, differences in kinetics will also influence relative potency (REP) values. Such REP values are required to take differences in potency between PAs into account in combined risk assessment 1 .

Kinetics are of importance for the toxicity of 1,2-unsaturated PAs because these PAs require bioactivation to dehydroPAs to become toxic. Although the metabolic pathway for bioactivation of PAs is similar, there appear to be marked differences in kinetics between PAs resulting in substantial variation in metabolic clearance and bioavailability. This implies that metabolism and thus kinetics are key determinants in PA toxicity. It also implies that the definition of REP values for in vivo risk assessment should take toxicokinetics into account. The present review summarizes the current state-of-the art on kinetics of PAs and resulting consequences for their toxicity and illustrates how physiologically-based kinetic (PBK) modelling can be applied to take kinetics into account when defining the relative potencies for toxicity of PAs in the in vivo situation.

Literature Search Strategy

A literature search was performed using Google Scholar and PubMed using the Advanced Search feature, and the publication period set from 1950 to 2021. Table 1S in the supplementary materials presents an overview of the keywords used. It is noteworthy to highlight that each pyrrolizidine alkaloid has a distinct name; for that reason, the search was not limited to keywords found in the title but also included keywords anywhere in the article. For each search, the title of articles found in the first 10 web pages (100 most recent hits) were scanned and added to Endnote when considered relevant. Later, duplicates were removed, and the abstracts of the articles thus obtained were scanned for relevant information as part of the inclusion strategy. The snowballing effect was used to find additional articles to ascertain that relevant previous articles were also included, and after reviewing, 79 articles were selected.

Differences in ADME Characteristics

PAs may differ in their ADME characteristics; such differences may already become evident during absorption in the gastrointestinal tract. Some studies pointed at differences in the absorption of the PAs across the intestinal barrier. The intestinal uptake of PA-N-oxides, for example, was reported to be less efficient than that of the corresponding PAs 2 . Using Caco-2 monolayer cells, apparent permeability coefficient (Papp) values of tested PA-N-oxides were demonstrated to be low to moderate compared to those of tested PAs ( Fig. 1 ).

Fig. 1

 Apparent permeability coefficients (Papp) of PAs and their respective PA-N-oxides in Caco-2 monolayer cells (50 µM). Data were extracted from Yang et al. (2020).

Furthermore, although the absorption of PAs may proceed via passive diffusion, there may also be a role for active transport depending on the type of PA. For example, the organic cation transporter 1 (OCT1) was reported to play a role in active transport of some PAs into the liver 3 . The substrates of OCT1 are known to be mainly organic cations, while some weak bases, non-charged compounds and anions are also transported 4 ; especially retronecine-type PAs including monocrotaline and retrorsine were shown to be high affinity substrates of OCT1 3 ,  5 . At low pH, where monocrotaline is protonated to its corresponding cation, transport by OCT1 even appeared to be dominant and passive diffusion was relatively negligible 3 . At higher pH values in the intestinal compartment 6 ,  7 , a substantial part of monocrotaline would be neutral and transported via passive diffusion. The same study also revealed that in rat hepatocytes at pH 7.4 the OCT1 mediated transport and passive diffusion may contribute equally to the intestinal absorption of monocrotaline 3 .

Detailed studies on potential differences in characteristics for distribution of PAs have not been described so far. Yet, differences in, for example, lipophilicity, might cause variations in tissue distribution and protein binding, thereby influencing the unbound concentration of a PA available in the target tissue to induce toxicity. For example, PAs are known to be more lipophilic with higher log P octanol/water coefficients than their corresponding PA-N-oxides 2 , a characteristic potentially resulting in variations in distribution. To illustrate this, Table 1 lists the degree of protein binding, reflected by the fraction unbound (fub) and tissue partition coefficient values of various PAs and their N-oxides in adipose and liver tissue. These values were calculated using the log P and molecular weight of the compounds, organ permeability and elimination rate constant of the tissue into plasma as previously described 8 ,  9 . Compared to PAs, PA-N-oxides have lower log P and partition coefficient values in adipose but higher partition coefficient values in liver due to the higher fat content of adipose tissue than of liver tissue.

Table 1  Molecular weight (MW), lipophilicity (Log P), degree of protein binding reflected by the fraction unbound (fub) and partition coefficient of different PAs and corresponding PA N-oxides.

	Mw	Log P a	fub b	Partition coefficient value c
				Adipose	Liver
a Computed by XLogP3 3.0 (PubChem release 2019.06.18); b State of ionisation is considered neutral, and molecular weight and log P were inserted in the utilized tool for different PAs (available at www.qivivetools.wur.nl developed by Wageningen Food Safety Research); c Organ permeability and elimination rate constant of the tissue into plasma are already incorporated in the utilized tool (available at www.qivivetools.wur.nl developed by Wageningen Food Safety Research); d PA-N-oxides contain a quaternary nitrogen atom
Riddelliine	349.4	0.2	0.710	0.15	0.67
Riddelliine N-oxide d	365.4	− 0.4	0.994	0.14	0.76
Lycopsamine	299.36	− 0.4	0.820	0.13	0.69
Lycopsamine N-oxide d	315.36	− 1	0.997	0.13	0.76
Retrorsine	351.4	0.6	0.618	0.22	0.66
Retrorsine N-oxide d	367.4	0	0.992	0.17	0.78
Senecionine	335.4	1.1	0.491	0.42	0.70
Senecionine N-oxide d	351.4	0.5	0.987	0.27	0.82

Lipophilicity could also play a role in the level of protein binding and, as a result, affect the excretion of PAs via glomerular filtration. The fub of PAs and PA-N-oxides in rat or human plasma can be determined based on their log P and molecular weight, using the QIVIVE Tools ( www.qivivetools.wur.nl ) designed by Wageningen Food Safety Research 9 ,  10 . Although log P values presented in Table 1 are below 5 and considered low 11 , PAs generally have higher log P values and consequently lower fub than their corresponding PA-N-oxides. The fub is part of the PBK model equation describing glomerular filtration to describe urinary excretion 12 . Hence, at similar total blood concentrations, relatively less PAs will be excreted via glomerular filtration compared to the corresponding PA N-oxides.

In addition to absorption, distribution and excretion, the characteristics of metabolism could also play an important role in defining the toxicity of PAs via, for example, differences in bioactivation and/or metabolic clearance. Bioactivation of PAs to reactive pyrrolic metabolites may occur in the liver, while for PA-N-oxides their reduction to the corresponding PA by the intestinal microbiota and in the liver is an important initial step in the bioactivation pathway 13 . When scaled to a whole body, the contribution from gut microbiota to this PA-N-oxide reduction may be higher than that of the liver 13 ,  14 . Furthermore, the efficiency of the reduction by the gut microbiota may vary with the PA-N-oxide of interest 14 ,  15 ,  16 ,  17 ,  18 ,  19 .

For the parent PAs, in vitro studies using either microsomal incubations or rat hepatocyte sandwich cultures have revealed substantial variation in metabolic clearance between the examined PAs 20 ,  21 ,  22 . Based on kinetic data from in vitro liver microsomal incubations, clearance of lasiocarpine was shown to be about 10-fold more effective than that of riddelliine and 43-fold more effective than that of monocrotaline ( Table 2 ) 22 . These results were in line with data from Lester et al., (2019) 21 , who studied the clearance of PAs in rat hepatocyte sandwich cultures and reported the in vitro clearance for these three PAs to also decrease in the order: lasiocarpine > riddelliine > monocrotaline 21 . It is of interest to note that in this latter study the actual relative differences between the clearance rates of the PAs appeared to change with the PA concentration at which the experiments were performed, reflecting differences in the apparent K m for this clearance. The observations together raise the question of how such differences in kinetics for clearance should be accounted for when defining relative potency factors for PAs based on in vitro studies.

Table 2  Kinetic constants for metabolic clearance of lasiocarpine, riddelliine and monocrotaline in in vitro rat liver and intestinal microsomal incubations and scaled to the in vivo situation. Data were extracted from Suparmi et al. (2020) and Chen et al. (2018).

Organ/Compound	V max (nmol/min/mg microsomal protein)	K m (µM)	Catalytic efficiency (mL/min/mg microsomal protein)	Scaled V max (nmol/min/g tissue) a	Scaled catalytic efficiency (mL/min/g tissue) a	Scaled catalytic efficiency (mL/min/tissue) b
aIn vitro V max and catalytic efficiency were employed to calculate Scaled V max and scaled catalytic efficiency based on 35 mg microsomal protein/(g liver) or 20.6 mg microsomal protein/(g small intestine); b In vivo scaled catalytic efficiency was calculated based on the in vivo scaled catalytic efficiency employing 8.5 g (liver weight) and 3.5 g (small intestine weight)
Liver
Lasiocarpine	5.3	19.5	0.27	186	9.5	80.9
Riddelliine	2.1	75.7	0.03	73.5	0.97	8.2
Monocrotaline	0.06	9.2	0.01	2.1	0.2	1.9
Intestine
Lasiocarpine	1.7	23.4	0.07	35.0	1.50	5.2
Riddelliine	0.1	221	0.0005	2.06	0.009	0.03
Monocrotaline	0.02	13.4	0.001	0.4	0.03	0.1

One could consider using scaling of the in vitro toxicity of the PAs in a specific bioassay using the area under the concentration versus time curve (AUC) obtained in the same experimental model to quantify differences in metabolic clearance. This is what was done by Lester et al. (2019) 21 ; these authors divided the level of DNA adduct formation in the hepatocytes by the AUC of the respective PAs in the sandwich rat hepatocyte cultures. The AUC of the parent PA in the culture medium was considered a proxy for PA exposure including clearance, while the ratio of DNA adducts/AUC was considered a way to characterize relative potency, taking potential differences in clearance into account.

However, the actual clearance rate and the corresponding AUC appeared to vary not only with the PA of interest but also with the PA concentration at which the clearance and AUC were quantified. For example, at 1.0 µM PA the difference in the AUC for riddelliine or lasiocarpine appeared to be substantially larger than the values reported at 100 µM PA, an observation that can be ascribed to the relevant K m values ( Fig. 2 ). This is in line with the relatively less efficient clearance of riddelliine as compared to lasiocarpine observed in incubations with rat liver microsomes with the K m for riddelliine clearance being fourfold higher than that of lasiocarpine 20 . Thus, the question emerges as to what AUC to use when scaling the in vitro data to take kinetics characteristics into account. An alternative way to account for such differences in kinetics, including variations in Km values for clearance among several tested PAs, is using physiologically-based kinetic (PBK) modelling.

Fig. 2

 Clearance of PAs in rat hepatocyte sandwich cultures. Data were extracted from Lester et al. (2019).

Physiologically-based Kinetic (PBK) Modelling

In physiologically-based kinetic (PBK) modelling, the kinetics of a compound of interest within the body are described by the PBK model. Such PBK models consist of a set of mathematical equations that together describe the ADME characteristics of a compound within an organism 23 . With a PBK model, parameters for different kinetic processes can be integrated in order to predict the physiologically relevant concentrations of a compound in plasma or, when relevant, in the tissue of interest, all for a given dose, time point and route of administration of interest.

In the field of PA toxicity, PBK models have been used for reverse dosimetry based quantitative in vitro to in vivo extrapolation (QIVIVE), translating in vitro data on toxicity or on genotoxicity in primary hepatocytes to predict in vivo acute liver toxicity or in vivo genotoxicity of PAs (for references see Table 3 ). In this approach, the models are used to convert in vitro concentrations to relevant in vivo exposure levels enabling QIVIVE that takes kinetics into account 24 ,  25 . In this reverse dosimetry, in vitro concentrations of the concentration-response curve are set equal to plasma or tissue levels of the respective compound in the PBK model. After correcting for potential differences in protein binding, the PBK model can calculate the corresponding in vivo dose level for any given route of administration. Subsequent benchmark dose (BMD) modeling can be applied on the predicted in vivo dose-response data, enabling definition of a point of departure (PoD) for risk assessment, such as a BMDL x (the lower confidence limit of the benchmark dose causing an x% effect above background level) and BMDUx (the upper confidence limit of the benchmark dose causing an x% effect above background level).

Table 3  Overview of papers reporting PBK models for PAs and predicted points of departure (PoDs) for various PAs by PBK model-based reverse dosimetry of in vitro toxicity data.

PAs	Species	Ethnic group	Adverse effect	Predicted PoD (mg/kg bw/day)		Reference
				BMDL 5	BMD 10
n. a.: not available; a Calculated as 90th percentile for the population taking interindividual differences in kinetics into account; b Calculated as 99th percentile for the population taking interindividual differences in kinetics into account; c Combined Caucasian and Chinese population
Lasiocarpine	Rat	n. a.	Genotoxicity	n. a.	8.82	26
			Acute liver toxicity	23.0	32.5	20
	Human	Caucasian	Acute liver toxicity	7.4	n. a.	27
				1.4 a	n. a.	28
				0.4 b	n. a.	28
		Chinese	Acute liver toxicity	14.7	n. a.	27
				4.4 a	n. a.	28
				1.8 b	n. a.	28
		Combined c	Acute liver toxicity	2.7 a	n. a.	28
				0.8 b	n. a.	28
Riddelliine	Rat	n. a.	Genotoxicity	n. a.	3.41	26
			Acute liver toxicity	4.9	2.6	20
	Human	Caucasian	Acute liver toxicity	0.2	n. a.	27
		Chinese	Acute liver toxicity	1.0	n. a.	27
Monocrotaline	Rat	n. a.	Acute liver toxicity	n. a.	2.8	29

PBK modelling has recently been applied to describe the kinetics of three PAs, including riddelliine, lasiocarpine and monocrotaline 20 ,  26 ,  27 ,  28 ,  29 . Fig. 3 a and b reveal that at a similar oral dose of 10 mg/kg bw the PBK models predicted substantial differences in both the time-dependent blood concentration and area under the concentration time curve (AUC), respectively, for these three model PAs. Table 3 presents an overview of studies where the respective PBK models were used for reverse dosimetry to translate in vitro concentration response curves to in vivo dose response curves for either acute liver toxicity or in vivo genotoxicity, from which points of departure (PoDs) were derived.

Fig. 3

 PBK model predicted ( a ) Cmax and ( b ) AUC of monocrotaline (dashed dotted lines), riddelliine (dotted lines) and lasiocarpine (dashed line in insert) in rats upon an oral dose of 10 mg/kg bw PA. Data were extracted from the PBK models used previously described by Chen et al. (2018) and Suparmi et al. (2020).

PBK modelling-based studies on PA toxicity

Predicting acute liver toxicity

In these PBK studies, in vitro data on the toxicity of the PAs toward primary rat or human hepatocytes were translated to in vivo dose response curves for acute liver toxicity. The predictions for lasiocarpine toxicity in rats could be validated against available in vivo toxicity data. The PBK model based predicted dose response curve for lasiocarpine induced acute liver toxicity resulted in a BMDL 5 value of 23.0 mg/kg bw/day, which compared well to the in vivo data reported for acute toxicity of lasiocarpine upon a single oral dose amounting to a No Observed Adverse Effect Level (NOAEL) and a Lowest Observed Adverse Effect Level (LOAEL) of 8 – 80 and 12 – 120 mg/kg bw/day, respectively 20 ,  30 .

Given that in vivo kinetic data for lasiocarpine are not available, this part of the model evaluation could only be done for riddelliine for which in vivo kinetic data exist 20 ,  31 . However, the PBK model and its predictions for lasiocarpine could be evaluated based on comparison of the predicted dose response curve for liver toxicity to available in vivo data for liver toxicity of lasiocarpine 20 ,  30 ,  32 ,  33 , whereas this second validation step was not possible for riddelliine for which in vivo liver toxicity data were absent.

Fig. 4 summarizes the results of the PBK model-based reverse dosimetry as obtained for lasiocarpine as compared to riddelliine 20 . The results demonstrate the importance of taking kinetics into account when considering relative toxic potencies of PAs; the in vitro toxicity of the two PAs in primary rat hepatocytes appeared to indicate that the toxicity of riddelliine and lasiocarpine is comparable. However, after taking the differences in kinetics into account by PBK model-based translation of the in vitro toxicity data to the in vivo situation, lasiocarpine is predicted to be substantially less toxic than riddelliine. This can be ascribed to the fact that lasiocarpine has faster metabolic clearance; at a similar dose level, lasiocarpine leaves less of the parent PA to be bioactivated and to exert toxicity. The dose level at which riddelliine was predicted to start inducing liver toxicity in rats, reflected by the predicted BMDL 10 –BMDU 10 , amounted to 1.3 – 3.7 mg/kg bw/day for riddelliine, a value that matched well with the estimated toxic oral dose range of 1 – 3 mg PA/kg bw/day 34 .

Fig. 4

 Results of PBK model based reverse dosimetry for lasiocarpine and riddelliine. Data were extracted from Chen et al. (2018) with the continuous line representing the data for riddelliine and the dotted line representing the data for lasiocarpine.

A similar study predicted acute liver toxicity in human based on cytotoxicity towards human hepatocytes and reverse dosimetry using human PBK models for lasiocarpine and riddelliine 27 . The prediction revealed a similar influence of toxicokinetics, resulting in translation of comparable in vitro concentration response curves to in vivo dose response curves that display differences in toxicity. The predicted in vivo liver toxicity of riddelliine in human was predicted to be 10- to 28- fold higher than that of lasiocarpine, again due to the more efficient metabolic clearance of the latter 27 . Although in vitro incubations with human hepatocytes show that riddelliine was 2.1-fold less toxic than lasiocarpine 27 , the predicted in vivo difference was the other way around. The predicted in vivo acute liver toxicity of riddelliine and lasiocarpine in human appeared to be even more pronounced than that in rat, with riddelliine being predicted to be 4- to 5- fold more toxic than lasiocarpine for rat and 7- to 10- fold more toxic for human 20 ,  27 ,  35 . This result illustrates that differences in kinetics influence relative in vivo potencies of PAs, indicating that in vivo REP values are likely to vary from what would be obtained based on in vitro data alone, and that in vivo REP values may be species dependent.

The data in Table 3 also illustrate that PBK modelling can be used to study the role of kinetics in the toxicity of PAs for various ethnic groups. Chinese, for example, are predicted to be 2.0-fold less sensitive towards acute liver toxicity of lasiocarpine and 5.0-fold less sensitive towards the acute liver toxicity of riddelliine than Caucasians, due to differences in kinetics. The PBK models revealed that this could be ascribed mainly to less efficient bioactivation of the parent compounds and occurred in spite of less efficient clearance of the PAs in Chinese as compared to Caucasians.

It is also of interest to note that the PBK model-based predictions for in vivo toxicity of PAs reported until now did not refer to toxicity upon repeated dose exposure. In a 28-day study using daily repeated oral exposure to lasiocarpine, a NOAEL for treatment-related decrease in body weight gain in male rats of 0.6 mg/kg bw per day was reported 32 . This is 24- to 43- fold lower than the PoD presented above for acute liver toxicity. Available experimental data and PBK model-based predicted plasma profiles reveal that clearance of the PAs upon an oral dose of 10 mg/kg bw is expected to be complete within 10 hours 31 ,  36 . This suggests that the PA itself would not accumulate upon repeated dosing. However, repeated dosing may result in accumulation of liver damage that may not be repaired within the 24 hours between subsequent doses. This would explain the lower NOAEL values for repeated dose exposure as compared to single exposure regimens. To study this repeated dose toxicity and the potential accumulation of liver damage in in vitro studies, in vitro models other than primary rat hepatocytes will be required to define the toxicity. Various in vitro models with longer incubation times are available, allowing studies on longer exposure duration from 48 hours up to 14 days 37 ,  38 ,  39 .

Predicting genotoxicity

An important adverse effect of 1,2-unsaturated PAs is their metabolic activation to DNA reactive pyrrole metabolites resulting in a variety of genotoxic effects such as formation of DNA adducts, DNA and protein cross-links, chromosomal aberrations, micronuclei, and DNA double-strand breaks 40 ,  41 ,  42 ,  43 ,  44 ,  45 ,  46 ,  47 ,  48 ,  49 ,  50 . These genotoxic effects are considered to result in gene mutations, leading to tumour formation. Recent efforts have studied the relative potency of a series of PAs and PA-N-oxides in different in vitro assays for DNA damage and/or genotoxicity and defined relative potency (REP) values ( Table 4 ).

Table 4  REP values of PAs and their N-oxides derived from different models.

PAs and PA-N-oxides	Merz & Schrenk (2016)	Louisse et al., (2019)	Allemang et al., (2018)	Lester et al., (2019)
	Combined Cytotoxicity ( in vitro ), Genotoxicity in Drosophila and acute toxicity from rodent LD 50	γ H2AX in-cell Western (ICW) assay, HepaRG cells	Micronucleus assay, HepaRG cells	DNA adducts/AUC, rat hepatocytes
n. a.: not available
Riddelliine	1	1.00	1	1
Lasiocarpine	1	1.08	5.2 – 6.3	2.5
Monocrotaline	1	0.06	0.023 – 0.03	0.225
Riddelliine N-oxide	1	n. a.	0.014 – 0.015	n. a.
Lasiocarpine N-oxide	1	≤ 0.01	0.008 – 0.009	n. a.
Monocrotaline N-oxide	n. a.	n. a.	0.002 – 0.003	n. a.

Allemang et al, (2018) reported the activity of 15 PAs and some PA-N-oxides in the micronucleus assay performed using HepaRG cells, and ranked these compounds in terms of potency based on the BMD confidence intervals ( Table 4 ) 40 . Louisse et al. (2019) reported the in vitro concentration-response curves for a series of PAs in human liver HepaRG cells using the phosphorylated histone H2AX in-cell Western ( γ H2AX ICW) assay 51 . The γ H2AX ICW assay quantifies the amount of phosphorylated histone H2AX ( γ H2AX), which is known to represent an unspecific marker of DNA damage 52 ,  53 and has been used as a surrogate endpoint for in vitro genotoxicity 54 ,  55 ,  56 ,  57 . They reported PA-N-oxides to be the least potent and two to three orders or magnitude less potent than the corresponding PA, but also recognized that further biokinetics needed to be considered before such data could define robust REP values. A REP value for PA-N-oxides of less than 0.01, indicated by the results obtained, would not be in line with available in vivo data on DNA adduct formation of PA-N-oxides as compared to their corresponding PAs 45 ,  47 ,  58 . This corroborates the need to take in vivo kinetics into account when defining REP values, and that results from in vitro models that do not include possibilities for reduction of PA-N-oxides to their corresponding PA by intestinal microbiota cannot be used to characterize the relative in vivo potency of PA-N-oxides.

Chen et al. (2019) used PBK models for lasiocarpine and riddelliine in rats to convert the in vitro concentration response curves for the activity of these PAs in the γ H2AX assay in primary rat hepatocytes to in vivo dose-response curves 26 . These curves were used to derive PoDs, which were comparable to available in vivo genotoxicity data. This study also revealed that due to the faster metabolic clearance of lasiocarpine than that of riddelliine, the relative potency of lasiocarpine compared to riddelliine is lower in vivo than in vitro 26 . Primary rat hepatocytes were shown to be more sensitive than HepaRG cells in this bioassay. In primary rat hepatocytes, γ H2AX induction was already observed at one to two orders of magnitude lower concentrations, supporting the conclusion that primary hepatocytes provide the preferred model for studies on PA toxicity 26 .

Predicting relative potency of PA-N-oxides

PA-N-oxides constitute an important class of PAs. In botanical preparations, PAs often are encountered in their N-oxide form. Moreover, 8 out of the 17 PAs marked by EFSA as being relevant for monitoring in food and feed were PA-N-oxides 59 . As already outlined above, these PA-N-oxides must first be reduced to their parent PAs to exert toxicity.

Merz and Schrenk (2016) assigned equal interim relative potency (iREP) values for PA-N-oxides to that of the corresponding PAs ( Table 4 ). This equal treatment of PA-N-oxides and PAs is a worst-case approach. However, results of some in vivo studies reporting DNA adduct formation upon dosing either the PA-N-oxide or a similar dose of the corresponding PA indicate that the REP of PA-N-oxides may be lower than that of their PA analogues 45 ,  47 ,  58 . These results indicate that the REP values of the PA-N-oxides are not equal to those of their parent PAs but are not as low as indicated by results of the various in vitro genotoxicity studies described above.

PBK modelling might provide a better way to define the REP values of PA-N-oxides as compared to their respective PAs. Such models can integrate the kinetics of PA-N-oxide reduction to the parent PA based on results obtained in suitable in vitro models, and can subsequently be used to derive AUC values for systemic exposure to the parent PA upon exposure to the PA itself or to an equimolar dose of the corresponding PA-N-oxide. Comparison of the respective AUC values would provide a novel way to define a REP value for the PA-N-oxides relative to their corresponding PAs.

Consequences of differences in kinetics for toxicity and relative potency factors

Table 5 presents an overview of literature data on in vivo kinetic parameters of PAs. Fig. 5 presents a comparison of the dose-normalized C max and AUC for different PAs in rats as calculated based on the data presented in Table 5 . The normalized C max for adonifoline appears to be 5.5-fold higher than that for riddelliine which is often used as the reference PA, and 6.8-fold higher than that of senecionine N-oxide. The latter may be in part related to the differences in kinetics between PAs and PA N-oxides including the reduction of N-oxides to the parent PAs. When making the comparison on the basis of the dose-normalized AUC, there are also marked distinctions between the examined PAs. In addition, the time dependent concentration curves for the tested PAs show different shapes, illustrating that a high C max does not automatically imply a high AUC. The ADME characteristics causing these differences remain to be elucidated. Development of PBK models for PAs and PA N-oxides would provide a way to elucidate the underlying reasons causing such kinetic variations.

Table 5   In vivo kinetic parameters of PAs as obtained from literature.

PAs	Species	Dose (mg/kg)	Administration	Kinetic parameters				References
				C max (µg/mL)	AUC(0–t) (µg min/mL)	AUC(0–∞) (µg min/mL)	Tmax (min)
n. a.: not available
Indicine N-oxide	Human	12.16	Intravenous	25	n. a.	n. a.	n. a.	74
		27.02		60	n. a.	n. a.	n. a.
		40.54		100	n. a.	n. a.	n. a.
		81.08		400	n. a.	n. a.	n. a.
	Rabbit	25	Intravenous	40	n. a.	n. a.	n. a.	16
		56		100	n. a.	n. a.	n. a.
		111		200	n. a.	n. a.	n. a.
Riddelliine	Rat Male Female	10	Oral	0.25   0.4	n. a. n. a.	30.96   76.02	n. a. n. a.	31
	Mice Male Female	10	Oral	1.3   1.25	n. a. n. a.	78.42   63.84	n. a. n. a.
Riddelliine N-oxide	Rat	20	Oral	0.5481	1278.9	n. a.	120	14
Senecionine	Rat	1.5	Intravenous	9.83	n. a.	72.12	2.00	75 ,  76
		5.7	Oral	0.48	29.74	30.13	22.22	76
		11.5		0.49	32.97	33.14	15.56
		18.445		0.20579	17.24	n. a.	10.00	36
		22.9		1.27	76.30	76.72	21.11	76
Senecionine N-oxide	Rat	19.33	Oral	0.392	118.82	n. a.	213.33	36
Adonifoline	Rat	4.0	Intravenous	40.3	320.2	334.03	2.00	75 ,  76
		16	Oral	1.5	364	373.7	69.3	76
		32		4.4	804.2	814.9	52.8
		64		9.8	2296.6	2367.6	62.2
Seneciphylline	Rat	2	Intravenous	n. a.	101.41	106.32	n. a.	77
		10	Intragastric	0.82	57.45	58.24	18.6
		20		1.37	79.61	81.83	13.8
		40		1.75	109.04	112.52	19.2
Monocrotaline	Rat	60	Intravenous	36.765	n. a.	n. a.	n. a.	29 ,  78
	Mice	3	Intravenous	2.5538	179.51	193.55	n. a.	79
		15	Oral	9.8865	792.9	795.58	30

Fig. 5

 Normalized Cmax/dose (black bars) and normalized AUC/dose (grey bars) for PAs orally administered to rat as derived form available in vivo data on PA kinetics as presented in Table 5 .

Several studies already defined REP values for PAs based on in vitro testing strategies. Merz and Schrenk (2016), for example, defined interim relative potency factors (iREP) based on available literature data on in vitro cytotoxicity, in combination with genotoxicity data obtained in Drosophila , and acute toxicity data from rodent studies (LD 50 ) 60 . Allemang et al. (2018) defined REP values based on the potency of a series of 15 PAs, including some PA-N-oxides, in the micronucleus (MN) assay performed in HepaRG cells 40 . Louisse et al. (2019) also used HepaRG cells to define REP values using γ H2AX induction as the read out 51 .

Table 4 provides an overview of the REP values defined in previous studies for the model compounds used in our PBK-based studies until now. As already discussed, using these in vitro assays to define REP values for PA-N-oxides is hampered by the fact that the in vitro models do not contain the system for metabolic reduction of the PA-N-oxides to the corresponding PAs. The REP values of 1.00 listed by Merz and Schrenk (2016) were based on the worst-case assumption that all PA-N-oxides would be efficiently converted to the corresponding PA 60 .

The REP values derived for riddelliine, lasiocarpine and monocrotaline based on data from in vitro studies may also be affected by the fact that the in vitro models do not take differences in kinetics into account. Lester et al. (2019) 21 acknowledged this aspect by defining the REP values based on DNA adduct levels formed in exposed rat hepatocytes divided by the AUC of the respective PA in the hepatocyte incubation medium to account for differences in clearance efficiency. The study by Chen et al. (2019) reveals that use of the in vitro -PBK modelling-based approach for predicting in vivo genotoxicity results in a REP value of lasiocarpine relative to riddelliine amounting to 2.6, being the ratio between the BMD 10 values derived from the predicted in vivo dose response curves 26 . This value is comparable to the value of 2.5 derived by Lester et al. (2019) 21 , who also included a correction for differences in clearance, while both these values are lower than the value of 5.2 – 6.3 derived from the MN assay without considering the characteristics of clearance 21 . A lower REP value for lasiocarpine relative to riddelliine, upon taking kinetics into account, may be related to more efficient clearance of lasiocarpine than that of riddelliine. The about 2 – 2.5-fold higher in vivo genotoxic potency of lasiocarpine than of riddelliine is also in line with the difference in their BMDL 10 values derived from data on the incidence of liver hemangiosarcoma obtained in 2-year carcinogenicity studies in rats upon chronic oral exposure to lasiocarpine and riddelliine 61 ,  62 . EFSA derived a BMD 10 of 0.131 mg/kg bw/day for lasiocarpine and of 0.292 mg/kg bw/day for riddelliine 1 . These values also indicate that the relative potencies for in vivo acute liver toxicity (with riddelliine being most potent), vary from the values derived from in vivo genotoxicity and carcinogenicity (with lasiocarpine being most potent).

Discussion

The overview presented in the present paper illustrates that kinetics is a key determinant in PA or PA-N-oxide toxicity, and that definition of REP values should take differences in in vivo kinetics into account. Including kinetics when defining REP values can be done by using in vivo data to define the REP values. However, only for a limited number of the 1,2-unsaturated PAs and their N-oxides in vivo toxicity data are available. This data limitation combined with ethical constraints and the fact that data obtained in experimental animals may not reflect the human situation, indicate a need for new approach methodologies (NAMs) that apply alternative testing strategies.

When considering the use of in vitro -based NAMs to define in vivo relative potencies it appears important to take differences in kinetics into account. The use of clearance data for this consideration, quantifying the disappearance of the parent PA in the respective in vitro incubations, may not give the most adequate parameter to reflect the variations in kinetics. Clearance will include both bioactivation and detoxification. An example that illustrates this can be found in the comparison of the acute liver toxicity predicted for the Caucasian and Chinese populations. The Chinese are predicted to be less sensitive because of differences in kinetics, despite higher C max values for the PAs due to slower clearance. This is because at the same time the Chinese show relatively lower catalytic efficiency for bioactivation of the PAs 27 . Using PBK model-based reverse dosimetry of data obtained in an in vitro model can reflect the consequences of such differences for the in vivo toxicity since it converts the in vitro toxic concentration of the PA to a corresponding in vivo dose level taking all kinetics into account. For such PBK models the kinetics for both bioactivation and detoxification could be characterized in vitro by using glutathione as a scavenger for the reactive pyrrole intermediate. The rate of clearance minus the rate of bioactivation allows quantification of the rate of detoxification.

It is also relevant to point out that the type of endpoint may be another reason causing the variations in the REP values. The induction of micronuclei, for example, reflects chromosome damaging potential in cells that have undergone cell division 63 . The phosphorylation of histone H2AX can be caused by DNA double strand breaks 64 ,  65 , while γ H2AX can also be induced by other cellular processes and may not always reflect direct DNA damage and genotoxicity 52 ,  56 ,  66 . γ H2AX can be a consequence of apoptosis rather than reflecting true genotoxicity 67 ,  68 .

Given that kinetics and bioactivation play an essential role in the ultimate toxicity, the in vitro cell model used to study toxicity or genotoxicity of PAs may influence the results obtained. Using lasiocarpine as the model PA, this is illustrated by the results presented in Fig. 6 , showing the concentration dependent effect of lasiocarpine on the viability of pooled primary human hepatocytes, HepaRG cells and HepG2 cells. HepaRG cells are considered to have a similar biotransformation capacity as primary hepatocytes, also having high expression of CYP3A4, the key enzyme for PA metabolism in human 69 ,  70 ,  71 . The results presented in Fig. 6 a reveal that lasiocarpine appeared to be not toxic toward HepG2 cells at the concentrations tested, likely due to the lack of sufficient CYP activity. The results also reveal that lasiocarpine was more toxic in pooled primary human hepatocytes than in HepaRG cells. However, when for the HepaRG cells exposure duration was extended from 24 to 72 hours the effect of lasiocarpine on the viability of the cells appeared comparable to what was observed for the pooled human primary hepatocytes upon 24 h exposure ( Fig. 6 b ). When considering the use of an in vitro cell model it is also of importance to note that freshly isolated primary human hepatocytes may show substantial variation in terms of CYP3A4 activity pointing at the need to use pooled batches of human primary hepatocytes to eliminate interindividual variability in kinetics as much as possible. Furthermore, it has also been reported that HepaRG 3D spheroids have greater CYP 3A4 activity, than the HepaRG cultured in 2D monolayers 72 .

Fig. 6

 Concentration-dependent effect of lasiocarpine on the viability of ( a ) pooled primary human hepatocytes (5 donors) (circles), HepaRG cells (squares) and HepG2 cells (triangles) upon 24 hours exposure and ( b ) primary human hepatocytes (triangles) upon 24 hours exposure, and HepaRG cells upon 24 (squares) and 72 hours (circles) exposure. Data were extracted from Ning et al. (2019).

Previous studies pointed at a lower sensitivity of HepaRG cells than of primary hepatocytes towards the toxicity of PAs. The IC 50 value of lasiocarpine for effects on cell viability in the primary rat hepatocytes was 2- and 20-fold lower than the values for the primary human hepatocytes and HepaRG cells, respectively; while the IC 50 value of riddelliine in the primary rat hepatocytes was 7-fold and 22-fold lower than the values for the primary human hepatocytes and HepaRG cells, respectively 20 ,  27 . Quantifying γ H2AX formation as the endpoint for genotoxicity, the primary rat hepatocytes appeared to be about one to two orders of magnitude more sensitive to lasiocarpine and riddelliine induced genotoxicity than the HepaRG cells, with in these respective models lasiocarpine being 1.1 fold and 3.5 fold more potent than riddelliine 26 .

It is worthwhile to consider re-investigating the derived relative potency values from such assays using other cell models. Comparing the BMD confidence interval values with lower and upper confidence limits (CEDL and CEDU respectively) reported by Allemang et al. (2018) for lasiocarpine and riddelliine in HepaRG cells using the micronucleus assay (0.8 – 1.1 µM and 4.3 – 6.9 µM respectively) 40 , and the BMC 50 values for lasiocarpine and riddelliine derived from the γ H2AX concentration-response curves (5.3 µM and 5.8 µM respectively) 51 , and the BMC 10 values obtained when the γ H2AX assay is performed in rat hepatocytes (2.09 µM and 2.17 µM) 26 reveals that in the same cell model, the micronucleus endpoint is 6.6- and 1.3-fold more sensitive for lasiocarpine and riddelliine, respectively, than the γ H2AX endpoint. When comparing the same γ H2AX endpoint in different cell models, it appears that rat hepatocytes are 105- and 31- fold more sensitive toward lasiocarpine and riddelliine, respectively, than the HepaRG cells 26 . This may be due to a relatively higher level of bioactivation in primary hepatocytes than in HepaRG cells as observed for human hepatocytes ( Fig. 6 a ) and/or to species dependent differences. Altogether these examples indicate that the REP values may vary with the endpoint and in vitro cell model used, with the latter at least in part being likely due to variations in the metabolic characteristics of the employed cell models.

Ideally, an in vitro model to quantify PA liver toxicity should most closely resemble the human situation. Use of human primary hepatocytes would then be of value but they also come with limitations, such as donor-to-donor variability. Thus, at the current state-of-the art it seems best to use pooled human hepatocytes. Alternatively, human hepatocytes derived from iPSCs generated from multiple individuals with different polymorphic characteristics may provide a supply of hepatocytes for high-throughput screening with minor batch-to-batch variability to improve reproducibility 73 . Whatever in vivo model and endpoint applied, results obtained in an in vitro model need further translation to the in vivo situation, for example, by using PBK model based reverse dosimetry to also account for differences in in vivo kinetics between the examined PAs.

Taking it all together, it is concluded that the use of NAMs is important to fill the data gaps that currently exist for PAs relevant in food and feed. Whatever the alternative testing strategy applied, it is important to consider that the relative toxicity of PAs will depend on both toxicokinetics and toxicodynamics. While in vitro toxicity tests may reveal differences in relative potencies and thus in toxicodynamics, the role of toxicokinetics in PA induced toxicity should not be ignored.

Contributorsʼ Statement

Conception and design of the work: I. M. C. M. Rietjens; drafting the manuscript: I. M. C. M. Rietjens, F. Widjaja, Y. Alhejji; critical revision of the manuscript: I. M. C. M. Rietjens, F. Widjaja, Y. Alhejji; data collection: F. Widjaja, Y. Alhejji; analysis and interpretation of the data: I. M. C. M. Rietjens, F. Widjaja, Y. Alhejji.