The development and application of physiologically based kinetic (PBK) models have gained momentum in recent years in biomedical research as well as in chemical risk assessment providing insight into the dosimetry related to observed health risks in humans (Paini et al., 2019). The current paradigm shift in chemical risk assessment from animal testing to next-generation risk assessment (NGRA) aims at integrating novel, human-relevant, toxicity pathway-based alternative new approach methodologies (NAMs) including in silico methods, among which PBK modelling (Zhang et al., 2018). Though promising, the data generated from in vitro toxicity approach by using alternative cell or organoid-based of human origin systems could only be made useful with the application of quantitative in vitro-to-in vivo extrapolation (QIVIVE) (Zhang et al., 2018). PBK modelling provides a framework to help the translation of effect concentrations eliciting biological responses in a human-relevant in vitro system to in vivo exposure levels (Yoon et al., 2012). A combined approach of QIVIVE application utilizing PBK and benchmark dose (BMD) modelling could be used to identify a concentration value (then considered as point of departure-PoD) corresponding to exposure levels in human tissue associated with perturbations of toxicity pathways in vitro (Louisse et al., 2017; Zhang et al., 2018). Central to a successful QIVIVE is the selection of a human-relevant endpoint in an in vitro assay conducted with a robust study design (Rogiers et al., 2020). This will help define a reliable reference point from in vitro concentration-response data before extrapolating it to an in vivo apical endpoint alteration (Rogiers et al., 2020). In cases in which the toxicological profile of a chemical is not known, selecting an adequate endpoint in an in vitro test system can, however, become quite challenging. PBK modelling is also a valuable tool to evaluate internal exposures of chemicals in situations where information on the kinetics is sparse or lacking particularly in special populations, e.g., pregnant women, where kinetic changes affect drug/chemical disposition determining both maternal and fetal exposure (Coppola et al., 2021).

Chlorpyrifos (CPF), an organophosphorus pesticide, can readily cross the placenta and affect fetal growth and neurodevelopment (Abdel-Rahman et al., 2002; Saulsbury et al., 2008) with effects such as altered cognition, motor control, and behavior in rats and mice (Burke et al., 2017; EFSA, 2019). As evidenced by epidemiological studies, the in utero exposure to CPF, has been linked to developmental neurotoxicity (DNT) with effects appearing in children later in life such as learning disabilities, attention deficit/hyperactivity disorders, decrease in intelligent quotient and working memory (Rauh et al., 2006; Rauh et al., 2011; Burke et al., 2017). A pregnancy-adapted PBK model including the fetus can help predict CPF levels reaching the developing fetal brain and its possible accumulation over long-term exposure.

In this work, we aimed to examine the feasibility and predictivity of in vitro toxicity testing data applying a QIVIVE approach. We set to predict maternal doses, leading to the in vitro effect concentration in the target tissues, i.e., fetal brain, by performing PBK modelling-based reverse dosimetry. Applying a combined QIVIVE-dose-response approach for some selected in vitro toxicodynamic non-cholinergic endpoints described in (Di Consiglio et al., 2020) (i.e., neurite outgrowth, synaptogenesis, BDNF levels, percentages of neurons and astrocytes), we obtained the corresponding maternal doses. They were compared to doses modelled by using the maternal CPF blood concentrations associated with neurodevelopmental adverse outcomes in children from epidemiological studies. The result of the comparison allowed us to assess the contribution of the in vitro toxicodynamic effects other than the well-established mechanism of acetylcholinesterase (AChE) inhibition to the CPF mode of action (MoA) for DNT. By this, we could examine the potential of using QIVIVE and validate the predictive value of in vitro models for CPF-induced DNT.

In vitro and in silico approaches which facilitate the understanding of toxicodynamic processes are increasingly applied to move away from traditional animal testing (Coppola et al., 2021). The use of in silico methods such as PBK modelling serves as a predictive tool for any substance systemic exposure and could be valuable to support risk assessment when data are sparse, such as in pregnant women (Coppola et al., 2021). In this work, we performed QIVIVE to evaluate the quantitative predictive value of an vitro test system for DNT effects which was assumed to be promising in this respect (Pistollato et al., 2020; Pistollato et al., 2021; Nunes et al., 2022), using non-acetyl cholinesterase-dependent markers in hiPSC-derived neuronal/glial cell model.

The PBK model was successful in predicting the in vivo concentration-time profile for CPF for non-pregnant women, validating the kinetic model and the substance specific and physiological parameters (in non-pregnant women). Parameterizing a PBK model can become challenging when in vivo kinetic data are limited or totally lacking to calibrate and validate model predictions as was the case with CPF in pregnant women.

The human test system applied in the in vitro study (Di Consiglio et al., 2020) mimics key neurodevelopmental processes including neural cell proliferation, neuronal/glial differentiation, neurite outgrowth and early synaptogenesis, which can be observed in the first trimester in humans (Tau and Peterson, 2010). Due to the early stage of pregnancy in which the model was set (at the end of the first trimester), pregnancy-related anatomical and physiological changes, e.g., metabolizing enzymes, were not yet fully expressed and less likely to impact CPF kinetics. This was evidenced by the fact that the predicted concentration-time profile of CPF in pregnant women was only slightly different from that in non-pregnant women and the slight differences were in accordance with the expected differences due to pregnant-related changes in physiology at the indicated gestational age.

Using the maternal-fetal PBK population model, we modeled the in vivo doses starting from the in vitro actual measured concentrations provided by an in vitro study (Di Consiglio et al., 2020). Contrary to nominal concentrations, the actual concentrations reflect the true exposure of the cells in an in vitro model which is important especially after repeated exposure (Coecke et al., 2013; Kramer et al., 2015). We chose the average CPF concentration in cultured neuronal and glial cells derived from NSCs as the dose metric, since the metabolizing capacity of cultured cells at this stage of differentiation was limited and CPF intracellular concentrations did not decrease significantly with time (Di Consiglio et al., 2020).

The quantitative prediction of the in vivo maternal doses leading to concentrations which correspond to the in vitro concentration data did not match with CPF exposures measured in pregnant women with adverse developmental neurological outcomes in their children in epidemiological studies. We used data from epidemiological studies associating in utero CPF exposure to adverse neurodevelopmental outcomes such as learning/behavioral dysfunctions in children of exposed mothers. Since the in vitro non-cholinergic toxicodynamic effects were ascribed to the parent compound, CPF, we chose only studies with measured CPF blood concentrations (cord blood/mother’s blood). Studies providing a direct measurement of CPF in blood rather than the urinary excreted metabolite, TCPy could be more accurate to characterize the exposure and the dose that reaches the target tissue (Wessels et al., 2003). In this respect, few studies have measured CPF in maternal or cord blood. Silver et al. (Silver et al., 2017) described CPF blood concentrations in exposed pregnant women ranging between 0.56 and 7.33 ng/mL. In the infants, motor function was assessed at 6-week and 9-month and the scores were significantly lower for exposed versus unexposed infants. Chiu et al. (Chiu et al., 2021), reported an average CPF blood concentrations of 0.13 ng/mL in pregnant women and a concentration-dependent statistically significant poorer performance in the cognitive and language domains in their children at the age of 2 years in Taiwan. In the study by Rauh et al. (Rauh et al., 2006), attention deficit/hyperactivity disorder problems were observed in children and differences comparing high (above 0.011 ng/mL) versus low exposed children were statistically significant. Since no maternal doses (exposure levels in mg/kg bw) were described in these epidemiological studies, we applied reverse dosimetry to reconstruct the external maternal dose from the CPF blood concentrations and used it to inform the maternal-fetal model to simulate the resulting fetal brain concentrations. Accordingly, neurodevelopmental adverse effects were reported at predicted doses above 0.11 μg/kg/day of chronic exposure to CPF in those studies (considering a range of 0.00011–0.14 mg/kg bw/day oral dose). Based on BMD modelling, higher daily intake doses (i.e., 3.21–271 mg/kg bw/day) were required to induce developmental neurotoxic effects, which were captured by the in vitro analyzed test battery. Therefore, we conclude that the test battery used is not quantitatively predictive for developmental neurotoxic effects of CPF. The calculated dose of 20.5 mg/kg bw/day corresponding to the nominal concentration causing 5% decrease in cell viability of 21 μM, which could be considered an early event of toxicity induction was identified as the LOAEC for the analysis of synaptogenesis in the in vitro study (Di Consiglio et al., 2020) and is within this range. This is also higher (10–900 fold), than the LOAEL defined by the regulatory authorities, e.g., European Food Safety Authority (EFSA) based on animal studies: 0.3 mg/kg/day (EFSA, 2019). Similarly, much lower concentrations were simulated in the fetal brain (0.85–85.97 ng/mL), with an exceptional high maximum level of 1147.6 ng/mL reported in one study by Silver et al. (2017), than the IC₅ concentration showing effects in cultured neuronal/glial cells, i.e., 21 µM (corresponding to 7942 ng/mL). It is worth noting that CPF is highly lipophilic and concentrations of CPF in blood are much lower (by orders of magnitude) than the concentration in fat or the brain tissue. In our study, the simulated brain concentrations were approximately 100 times higher than plasma concentrations based on the algorithm of Schmidt (Schmitt, 2008a; Schmitt, 2008b) implemented in the PBK modelling software.

It is also important to mention that large differences exist between in vitro dosing patterns and in vivo exposure profiles. In our study, we estimated the oral doses after multiple exposure, which is a limitation in our approach, compared to the real-world where exposure could take place simultaneously through additional routes such as inhalation and dermal routes. However, in view of the very large differences between the predicted doses and exposure levels from epidemiological studies, it is unrealistic that exposure via dermal and inhalation routes, from which the bioavailability of CPF is lower (e.g., dermal absorption is less than 2% of the applied dose in humans) compared to the oral route (Nolan et al., 1984; Eaton et al., 2008), could account for the high predicted doses.

For a meaningful quantitative prediction of the in vivo dose eliciting adverse effects in vivo, the in vitro system should adequately represent the target tissue in vivo as well as react with the same sensitivity. Here, the cellular model used was a culture of human iPSC-derived NSCs undergoing differentiation towards a mixed culture and neurons and glia, which is suitable to model some key processes of the developing brain (i.e., NSC proliferation, neuronal/glial cell differentiation, neurite outgrowth, synaptogenesis and neuronal network formation and function) (Bal-Price et al., 2018; Pistollato et al., 2020). Fundamental to the predictive value of an in vitro system is the relevance of the measured in vitro effects to the mode of action (MoA) of a specific chemical in vivo. Concerning the MoA of CPF toxicity, although AChE inhibition has been an established mechanism for toxicity, other mechanisms have been shown to play a role in neurodevelopmental adverse effects. Experimental data provide evidence that CPF can interact with and change the activity of non-AChE targets and that DNT may occur in the absence of significant AChE inhibition and even at lower concentrations (Eaton et al., 2008). For example, non-cholinesterase-depending mechanisms induced by CPF altering synaptogenesis, neuronal network formation, and BDNF signaling in differentiating PC12 cells was shown in vitro (at 30 μM nominal concentration) as well as in young rats in vivo (Betancourt et al., 2007; Slotkin et al., 2008). Indeed, CPF (at 28.5 μM) was shown to inhibit neurite outgrowth in the PC12 cells and in primary cultures of embryonic rat sympathetic neurons (Das and Barone, 1999). These data suggest that the observed effects might be considered as a reliable functional readout, specific for neuronal cell differentiation, hence, justifying its use in the in vitro test battery established for neurodevelopmental neurotoxicity evaluation as described in Di Consiglio et al. (Di Consiglio et al., 2020). Nominal in vitro CPF concentrations of 10–20 μM may lead to alterations in expression of marker genes of differentiation as shown in the study by (Estevan et al., 2013). These concentrations are in the same range as in the in vitro study by (Di Consiglio et al., 2020).

Our results, however, show that the predictive value of the non-cholinesterase biomarkers is low. Although they may be qualitatively related to the toxicodynamic mechanisms of CPF, they could not be utilized quantitatively to provide a meaningful PoD for risk assessment. Our work was not designed to corroborate or disprove the presumed association of CPF with neurodevelopmental impairment suggested by epidemiological studies; however, we noticed the large discrepancy between the maternal doses needed to reach the effect concentration in the fetal brain and the reported regulatory LOAEC of 0.3 mg/kg bw/day. Our fundamental aim was to evaluate the quantitative predictivity of the used in vitro test battery for CPF in this respect.

The potential of the in vitro observed effects to successfully predict in vivo effects relies on whether the described in vitro effect can be a part of the toxicity pathway. The latter is defined as biochemical pathways or molecular circuits in the cells which would lead to adverse health outcomes when sufficiently perturbed by external factors (NRC, 2007). Effects causing perturbations of the signaling pathways leading to apical endpoint alterations both qualitatively and quantitatively can be considered as key molecular mechanisms in the adverse outcome pathway (AOP) (Zhang et al., 2018).

CPF and its oxon metabolite, CPFO, are both inhibitors of brain cholinesterase activity, with CPFO being more potent (almost 1,000 times) in vitro. In a system of cultured sensory neurons derived from embryonic rat dorsal root ganglia, CPF was found to exert an inhibitory effect of brain AChE in concentrations of 0.1 μM (i.e., 35 ng/mL) (Yang et al., 2008). While keeping into account species-specific differences in neural/neuronal cell sensitivity to toxicants (i.e., human vs. rodent) (Baumann et al., 2016), this concentration is much lower than the concentration suggested by the non-choline esterase-toxicodynamic endpoints applied in the in vitro study. Nevertheless, they are in concordance with the intracellular brain concentrations in the fetus predicted from biomonitoring data where concentration as low as 0.85 ng/mL were associated with DNT. Consequently, the selected toxicodynamic markers might be less sensitive than the well-established AChE inhibition for assessment of DNT.

Another factor to consider is the contribution of CPFO, an irreversible brain AChE inhibitor, to the neurologic toxicity induced by the parent compound in vivo. Given the limited metabolizing capacity of brain cells in vitro, the role of the bioactive oxon metabolite was neglected. This could have led to overestimate the effect concentration, consequently, the doses of the parent molecule needed to induce the specific responses, which deviates significantly from the in vivo situation. Although the expression of some CYPs, among which CYP2B6 and CYP2C19 have been found in this in vitro test system (Di Consiglio et al., 2020), the content of CYPs in the brain is low compared to that in the liver (approximately 0.5%–2%) (Hedlund et al., 2001). This implies that brain CYPs have a minor overall impact on CPF kinetics in vivo, however, in vitro, this limited extent of metabolism could still be relevant for metabolites toxic to neuronal cells.

Although other CYP isoforms shown to contribute to the biotransformation of CPF into the CPFO and TCPy metabolites in vitro, including CYP2A6, -2C9, -2D6 and -3A4, in addition to CYP1A2 (Buratti et al., 2002; Buratti et al., 2003), they were not accounted for in our PBK model, which may be regarded as a limitation. However, the majority of isoforms (CYP2A6, -2C9, -2D6, and -3A4) were shown to be more active at high substrate concentration (Buratti et al., 2002). Therefore, they play a less significant role in metabolism at lower substrate concentrations which are characteristic of subchronic exposure rather than the acute intoxication. For instance, the oral administration of 2 mg/kg bw CPF resulted in maximum blood concentrations of about 4 ng/mL, i.e., 0.011 nM, whereas, in vitro, the activity of the CYP isoforms were measured at relatively higher CPF concentration ranging from 25 to 100 μM (Buratti et al., 2002) or 0.02–10 μM (Buratti et al., 2003). The underrepresentation of these isoforms might have influenced the model predictions and lead to dose underestimation in the high dose ranges. The latter, however, are not relevant in the context of repeated low dose exposure and would not reflect the in vivo situation.

Notably, the data yielded by epidemiological studies are subject to several limitations, which contribute to the uncertainty inherent to these studies. These limitations include the single time point analysis of the exposure biomarker, the timing of the sample collection which occurs in late-stage pregnancy/delivery rather than throughout the pregnancy, and the influence of external risk factors such as co-exposure and the social environment on the measured neurological outcomes (Reiss et al., 2015).

A further limitation to be acknowledged is the time of exposure to CPF in vitro. We targeted the period around 15th week of gestation to primarily mirror effects of CPF on important cellular processes in brain organogenesis taking place within this time frame captured by the in vitro model. This period is perceived as the window of susceptibility regarding fetal brain formation (Moore et al., 2015). It might be that prolonged in vitro exposure could result in lower effect concentrations than those the described in 14-day study due to cumulative effects which translates to lower in vivo doses. We cannot exclude CPF-induced DNT at lower doses to occur with continuous exposure. Nevertheless, given the large difference between the modelled maternal exposure doses and the resulting fetal brain concentration, on one hand and those predicted from the in vitro system, on the other, the impact of such limitations on our findings is less than likely.

Finally, although a human cell-relevant in vitro assay merits the elimination of interspecies differences, nevertheless, the differentiating cultured neuronal and glial cells derived from human NSCs might still not react with the same sensitivity to the toxic insult as the native cells in vivo.

In conclusion, our developed CPF PBK maternal-fetal model was capable of predicting the in vivo kinetics. However, the non-cholinesterase- dependent markers showed to be less sensitive than the well-known AChE inhibition in human to predict DNT in a quantitative way. Effective daily intake doses predicted from the in vitro model were several orders of magnitude higher than exposures estimated from epidemiological studies to induce developmental non-cholinergic neurotoxic responses analyzed in the in vitro test battery (Di Consiglio et al., 2020) an also higher than reported regulatory LOAEC for acetyl-cholinesterase inhibition of 0.3 mg/kg bw/day. The quantitative predictive value of the investigated non-choline esterase-dependent toxicodynamic effects, although possibly relevant for other chemicals, may not adequately represent key events in the MoA/AOPs for CPF-associated DNT.