A preliminary evaluation of the applicability of the in vitro to in vivo extrapolation approach to quantitative risk assessment

Abstract

The applicability of the in vitro to in vivo extrapolation (IVIVE) approach for the quantitative risk assessment of chemicals was evaluated using the results of mouse uterotrophic bioassays conducted by the Japanese Ministry of Health, Labour and Welfare. Five chemicals were selected. In the uterotrophic bioassay, three chemicals exhibited positive estrogenic activity, whereas two exhibited negative activity. These chemicals were active in 16 in vitro assays that investigated the key events from estrogen receptor (ER) binding (initiating event) to ER-induced proliferation, enabling us to derive IVIVE conversion factors using a physiologically based kinetic model. The oral equivalent doses (OEDs) that were extrapolated from the activity concentrations at the half-maximal response (AC₅₀) and the cutoff point (ACC) were compared with those of other key events to determine the critical key event that plays the most important role in the occurrence of uterotrophic responses. For the three chemicals that exhibited positive estrogenic activity, the OEDs from the in vitro AC₅₀ values for the determined critical events were within a factor of 2 of the lowest observed effect levels in the uterotrophic bioassay. In addition, the OEDs from the ACC values for the critical key events of the two chemicals that exhibited negative activity were higher than the highest dose tested in the bioassay. Based on these findings, the IVIVE approach was largely valid. However, the critical key events that significantly affect in vivo responses need to be appropriately determined to apply the IVIVE approach for the quantitative risk assessment of chemicals.

INTRODUCTION

In current risk assessments of chemicals, doses determined from in vivo bioassays using experimental animals, including the no observed adverse effect level (NOAEL), are commonly used as the point of departure (PoD). In vivo bioassays, though widely used, require animal sacrifice and are time-consuming and costly. By contrast, in vitro assays provide convenient and efficient testing methods; however, they cannot provide reliable dose–response relationships for deriving a PoD, as they fail to account for absorption, distribution, metabolism, and excretion processes occurring in vivo. Thus, they have not been adopted for quantitative risk assessments. However, in recent years, the in vitro to in vivo extrapolation (IVIVE) approach, which uses a physiologically based kinetic (PBK) model to extrapolate the concentration–response relationship obtained from in vitro assays to the dose–response relationship in vivo, has attracted increasing attention as a rapid and relatively inexpensive alternative to in vivo bioassays. In this approach, the active concentration in the in vitro assay is used as the blood concentration that could produce in vivo effects. Moreover, the oral equivalent dose (OED) required to reach this blood concentration is calculated and can be used to derive the PoD.

The validity of the IVIVE approach has already been investigated in several chemicals. Several studies have compared the BMDLs (lower confidence limit of benchmark dose (BMD)) of OED–response curves extrapolated from in vitro assays using the IVIVE approach with NOAELs or BMDLs from in vivo studies. For example, Strikwold et al. (2013, 2017) compared the BMDL₀₅ with the NOAELs and BMDL₀₅ in rat developmental toxicity studies of phenols. Li et al. (2017) compared the BMDL₁₀ with the NOAELs in rat developmental toxicity studies of the fungicide tebuconazole. Moreover, Noorlander et al. (2022) compared the medium effective dose (ED₅₀) from the acute neurotoxicity of tetrodotoxin in rats and mice with that extrapolated from in vitro assays. In these studies, the BMDL or ED₅₀ values extrapolated from in vitro assays were within the same order of magnitude as the corresponding BMDL or ED₅₀ values in vivo.

In a recent study, the OEDs were extrapolated from the concentrations required to cause the lowest significant activity (activity concentrations at cutoff or ACCs) of 18 and 11 in vitro assays targeting agonistic/antagonistic estrogenic and androgenic activity for 233 chemicals tested in the US EPA Toxicology Prediction (ToxCast) program, and the ratios to the estimated human oral exposure (biological activity–exposure ratio) were calculated to screen chemicals that require further risk assessments (Pradeep et al., 2020). Taking advantage of the flexibility of in vitro assays, Wegner et al. (2020) calculated the combined exposure-biological activity index by considering the estimated average exposure in addition to the OEDs extrapolated from the activity concentrations at half maximum response (AC₅₀s), the activity concentrations at 10% response (AC₁₀s), and the ACCs to assess the health risk from combined exposure for 22 chemicals. The results of these studies suggest that the IVIVE approach could be used as an alternative to short-term animal toxicity testing. However, only a small number of chemicals were examined. Thus, the effectiveness of this approach remains to be validated.

For the uterotrophic bioassay, Zhang et al. (2018) compared the BMDL₁₀ with that in rat uterotrophic bioassays of 17β-estradiol and bisphenol A (BPA). Besides Zhang et al.’s (2018) study, other studies also compared the OEDs and BMD₅₀s of chemicals: one study compared the OEDs of three chemicals (ethinylestradiol, BPA, and genistein) extrapolated from the lowest observed effect concentrations of the in vitro assays with the lowest observed effect levels (LOELs) from the rat in vivo assays (Fabian et al., 2019) and another study compared the BMD₅₀s of five chemicals (diethylstilbestrol, ethinylestradiol, genistein, coumestrol, and methoxychlor) derived from the extrapolated OEDs with those from rat in vivo assays (Zhang et al., 2020). However, the number of chemicals that were examined was still small.

The objective of this study was to determine whether the OED derived using the IVIVE approach can serve as an alternative to the PoD for assessing short-term chemical effects. To this end, we compared OEDs obtained through IVIVE with LOELs and BMDs from mouse uterotrophic bioassays conducted by the Japanese Ministry of Health, Labour and Welfare (MHLW) to screen potential endocrine-disrupting chemicals, and evaluated the applicability of the IVIVE approach in quantitative chemical risk assessment.

MATERIALS AND METHODS

Chemicals

The chemicals to be examined were selected from the MHLW Priority List of Chemicals (unpublished) that were determined to require detailed testing to confirm their endocrine-disrupting effects (https://www.nihs.go.jp/edc/english/actions/scheme.htm). We considered chemicals that met the following criteria to be suitable for IVIVE:

- The chemical is active in any of the 16 high-throughput agonist estrogen assays available from the National Institutes of Health (NIH) Integrated Chemical Environment (ICE) tool (https://ice.ntp.niehs.nih.gov/).

- All chemical-specific parameters required for the PBK mouse model must be available.

The 16 agonist estrogen assays are used to evaluate uterine hypertrophy. The uterotrophic response may result from a series of consecutive events covered by the assays. The chemical-specific parameters included molecular weight, topological polar surface area (TPSA, Ų), intrinsic hepatic clearance (CLint, μL/min/10⁶ hepatocytes), and octanol/water partition coefficient (log Kow). The TPSA was obtained from the US NIH PubChem database (https://pubchem.ncbi.nlm.nih.gov/), whereas the molecular weight, CLint, and log Kow were obtained from the CompTox Chemicals Dashboard database of the US Environmental Protection Agency (https://comptox.epa.gov/dashboard/).

In vitro assays

Several mechanisms have been proposed for endocrine disruption (OECD, 2021). Hypothetically, receptor dimerization, dimer–DNA binding, RNA transcription, protein production, and ER-induced proliferation were the key events mediating the pathway from the initiation event (estrogen receptor [ER] binding) to the phenotypic outcome (i.e., proliferation) (Judson et al., 2015). Herein, we used the 16 in vitro assays conducted in the USA as part of the Tox21 and ToxCast programs. These are high-throughput assays designed to measure agonist activity at key events in the ER pathway. The AC₅₀ and ACC values were obtained from the following 16 assays: ER binding assays (NVS_NR_bER, NVS_NR_hER, and NVS_NR_mERa), receptor dimerization assays (OT_ER_ERaERa_0480, OT_ER_ERaERa_1440, OT_ER_ERaERb_0480, OT_ER_ERaERb_1440, OT_ER_ERbERb_0480, and OT_ER_ERbERb_1440), dimer–DNA binding assays (OT_ERa_EREGFP_0120 and OT_ERa_EREGFP_0480), RNA transcription assays (ATG_ERa_TRANS_up and ATG_ERE_CIS_up), protein production assays (TOX21_ERa_BLA_Agonist_ratio and TOX21_ERa_LUC_VM7_Agonist), and ER-induced proliferation (ACEA_ER_80hr). These values were used for the IVIVE of the selected chemicals. The obtained AC₅₀ and ACC values were metadata values displayed in the Results View of ICE’s Curve Surfer tool.

In vivo uterotrophic bioassay in mice

The uterotrophic bioassays, which were commissioned by the MHLW to contract testing laboratories, were performed according to or similar to the OECD test guideline 440 (OECD, 2007). The study results (unpublished) were obtained via oral gavage and/or subcutaneous injection. Herein, we only considered oral gavage studies to assess the relevance of the human exposure route. The test chemicals were administered to ovariectomized female mice (six mice/treatment group) for up to 7 consecutive days using corn oil as a vehicle. The dosages were 0 (vehicle only), 30, 100, 300, and 1,000 mg/kg/day or 0, 20, 60, 200, and 600 mg/kg/day. Moreover, ethinyl estradiol at 6 μg/kg/day was administered to the positive control group. After treatment, the mice were euthanized and dissected, and their uteruses were collected and weighed. The LOEL was determined when a significant weight gain was observed. The BMDs were calculated from the mean and standard deviation of uterine weights for each treatment group using the PROAST web application version 70.1 (https://proastweb.rivm.nl/) of the Dutch National Institute for Public Health and the Environment. In PROAST, six models can be applied to fit continuous data such as uterine weights: the Full model, Null model, Exponential model 3, Exponential model 5, Hill model 3, and Hill model 5. The BMD calculated for each model is reported together with the Akaike Information Criterion (AIC) value. In this study, the model with the lowest AIC was identified as optimal, and its corresponding BMD value was adopted.

The LOEL, BMD, and no observed effect level (NOEL) were compared with the OEDs extrapolated using the IVIVE approach. As described above, previous studies have compared OEDs derived from in vitro assays with the BMDL, BMD, and/or LOEL values obtained from in vivo bioassays. The ACC and AC₅₀ values are the expected values of the concentration calculated from in vitro concentration–activity curves, not the 95% lower confidence limit of the concentration. Moreover, the ACC value approximately corresponds to the concentration of activity at 20% of maximum activity. Therefore, the OED_{_crit} values calculated from the ACC and AC₅₀ values of the critical key events were compared with the BMD₂₀ and BMD₅₀ values calculated from the dose–response relationship of the uterotrophic bioassay. Furthermore, the ACC- and AC₅₀-derived OED_{_crit} values were compared with the NOEL and LOEL, respectively, assuming that the NOEL and LOEL could be approximated values of BMD₂₀ and BMD₅₀, respectively. A part of these assumptions is supported by Zhang et al. (2020). As a prerequisite for using the IVIVE approach, the present study assumes that the effects observed in the in vivo bioassays are because of the parent chemicals, not their metabolites.

PBK mouse modeling

The PBK mouse model for IVIVE comprises six compartments for blood, fat, rapidly perfused tissues, slowly perfused tissues, kidney, and liver (Fig. 1). This model assumes that the chemicals are absorbed from the gastrointestinal tract, transported through blood flow via the liver, and distributed to each compartment. Moreover, it assumes that the chemicals are metabolized in the liver and excreted by the kidneys.

Fig. 1

Schematic diagram of the PBK mouse model.

The mouse physiological parameters (body weight, tissue weight, blood flow, etc.) were taken from the values described in a previous study (Noorlander et al., 2022) (Table 1). Each compartment’s volume was calculated by multiplying the body weight of mice by the fraction of tissue volume. Moreover, the blood flow rate through each compartment was calculated by multiplying the cardiac output by the fraction of blood flow to the tissue.

Table 1. Values of physiological parameters that were used in the PBK mouse model.

The absorption rate constant from the gastrointestinal tract of a chemical via the oral route (ka) was estimated from the TPSA. First, the apparent permeability coefficient of Caco-2 cells (Papp, cm/s) was calculated from the TPSA using the following equation (Hou et al., 2004):

Next, the effective permeability coefficient (Peff, cm/s) was calculated from the Papp using the following equation (Sun et al., 2002):

The Peff calculated for humans was corrected to that of rats (Peff, rat) by dividing it by 3.6 (Fagerholm et al., 1996). Finally, the rat ka (k_{a, rat}) was calculated using the following equation (Yu and Amidon, 1999):

In this study, the k_{a, rat} was used in the PBK mouse model.

The tissue/blood partition coefficients of the chemicals in rats were estimated from the Kow using the following equation (DeJongh et al., 1997) and applied to the PBK mouse model:

Fat/blood partition coefficient (PF):

Rapidly perfused tissue/blood partition coefficient (PR):

Slowly perfused tissue/blood partition coefficient (PS):

Kidney/blood partition coefficient (PK):

Liver/blood partition coefficient (PL):

The hepatic clearance (CLH) of chemicals was calculated by multiplying the CLint from the CompTox Chemicals Dashboard database by the mouse scaling factor (135 × 10⁶ hepatocytes/g liver; Sohlenius-Sternbeck, 2006) and the liver volume. The CLint values in the database were measured in metabolic stability studies using primary human hepatocytes that were metabolically active in phases I and II (Wetmore et al., 2012).

The renal clearance (CLR) of chemicals was calculated using the equation of Jongeneelen and ten Berge (2011):

where GlomFiltr is the glomerular filtrate as a fraction of the renal arterial flow (0.16 for mice). The FWS was the fraction of a chemical dissolved in the blood and calculated as follows:

Freabsorp, which refers to the fraction of tubular resorption, was calculated as follows:

where x=2(log Kow + 0.5).

The mass balance of the chemicals in each compartment was described according to the model of Zhang et al. (2018):

Blood compartment:

Fat compartment:

Rapidly perfused tissue compartment:

Slowly perfused tissue compartment:

Kidney compartment:

Liver compartment:

Unabsorbed dose in the gastrointestinal tract:

Here, V and Q represent the volume and blood flow for each compartment, respectively. C refers to the concentration of a chemical in each compartment. BL, F, R, S, K, and L denote blood, fat tissue, rapidly perfused tissues, slowly perfused tissues, kidneys, and liver, respectively. QT is the cardiac output, and kfe and AING are the first-order rate constant for transfer to feces and the unabsorbed dose in the gastrointestinal tract, respectively. The kfe was calculated to be 0.3 h⁻¹ from the experimental data of Nagata and Sugimoto (1973).

The plasma-to-blood ratio (PBP), which was used to convert blood concentrations to plasma concentrations, was calculated using the following equation (Kamiya et al., 2022):

To calculate the plasma protein unbound concentration, the plasma concentration was multiplied by the fraction unbound to protein in the plasma (Fup). The Fup for rats was estimated using the Drug Metabolism and pharmacokinetics Analysis Platform ver.1.5 (Kawashima et al., 2023) and was also used for mice.

The PBK model was coded in R version 4.1.3 (R Core Team, 2023), and the differential equations that describe the mass balance were solved using the deSolve package version 1.27.1 (Soetaert et al., 2010).

Verification of the PBK mouse model

The mouse-specific values for many parameters of the developed PBK model were difficult to set, except for physiological parameters, and the values for rats were used without correction for ka, tissue/blood partition coefficients, and Fup. Therefore, we verified the developed model’s predictive performance. This verification requires the concentration–time profile of the chemical in mouse blood and the model parameter values. Although blood concentration–time profiles have been reported for several chemicals in mice, in this study we performed verification using previously published data for BPA, which is included on the MHLW Priority List of Chemicals. We used the measured concentration–time profiles of unconjugated BPA in mouse serum following oral administration of 0.4 and 100 mg/kg (Taylor et al., 2011). The concentrations were obtained by using WebPlotDigitizer (https://automeris.io/) to read the values in the figures of Taylor et al.’s (2011) study. The values shown in Table 2 were used for the BPA-specific model parameters. We used CLint and Fup values that were measured in rats (Fabian et al., 2019). As the CLint value was based on microsomal protein content, we derived the CLH by multiplying CLint by a scaling factor for rats (50 mg microsomal protein/g liver; Fabian et al., 2019) and the mouse liver volume.

Table 2. Values of BPA-specific parameters that were used for validation of the PBK mouse model.

To evaluate the model’s predictive accuracy, the fold error (FE), average fold error (AFE), and absolute average fold error (AAFE) were calculated as follows:

where Predicted(i) and Measured(i) are the predicted and measured concentrations at time point i, respectively, and n is the number of time points at which the concentrations were measured. FE indicates the predictive accuracy for each data point, whereas AFE indicates whether the predicted profile under- or overestimates the measured values. In addition, the AAFE quantifies the absolute error from the measured values. If the FE for all data points was within 0.33–3 (a threefold error) and the AFE and AAFE were both less than 2, then the model predictions were considered valid.

IVIVE

The PBK mouse model was used to calculate the maximum molar concentration (Cmax, µM) of the unbound chemical in plasma when the mice were orally administered at 1 mg/kg/day for up to 7 days. The IVIVE conversion factor was calculated as follows:

To calculate the OED that gave unbound plasma concentrations equivalent to the activity concentrations, the AC₅₀ and ACC values of the 16 in vitro assays were multiplied by the IVIVE conversion factor.

As mentioned above, the 16 in vitro assays using the IVIVE approach herein are designed to measure activity at a series of key events in the ER pathway initiated by ER binding and leads to proliferation. Therefore, when the activity concentrations are at the same level in each in vitro assay targeting two consecutive key events, if the activity is exhibited in the assay for the former key event, it will likely be exhibited in the assay for the latter key event at the same concentration. Moreover, the OED of the former key event plays a crucial role in the occurrence of uterotrophic responses. Conversely, if the activity concentration of the assay for the latter event is higher than that the former, the pathway is unlikely to proceed well unless this activity concentration is reached. Therefore, the OED of the latter key event will play a more important role in the occurrence of the uterotrophic response than the former. Considering the abovementioned hypothesis, the critical key event was defined as the key event that plays the most important role in inducing the uterotrophic response, and the mean of the OEDs extrapolated from the activity concentrations (AC₅₀ and ACC) of the in vitro assays targeting the critical key event was calculated. Subsequently, this mean was defined as the OED_{_crit_AC50} or OED_{_crit_ACC}, which can be compared with the NOEL, LOEL, and BMD values of the in vivo bioassay.

Because multiple in vitro assays target each key event, we followed the steps below to identify the critical key events and calculate the OED_{_crit_AC50} and OED_{_crit_ACC} values for each chemical:

1) Define a discrete uniform distribution comprising multiple OED_{_AC50} values extrapolated from the AC₅₀ values of the in vitro assays for each key event. Similarly, define a discrete uniform distribution comprising OED_{_ACC} values extrapolated from the ACC values.

2) Randomly generate one OED from each distribution of OEDs for the initiating key event (ER binding) (OED_{_AC50_er}) and the subsequent key event (receptor dimerization) (OED_{_AC50_di}). Then, compare the two generated OEDs to determine which is larger. Iterate this process 1,000 times, and calculate the probability that the distribution of OED_{_AC50_di} is greater than that of OED_{_AC50_er} (Pr(OED_{_AC50_di} > OED_{_AC50_er})).

3) Consider that the second key event (receptor dimerization) plays a more important role in inducing a uterotrophic response when Pr(OED_{_AC50_di} > OED_{_AC50_er}) exceeds 50%. Conversely, consider that the initial key event (ER binding) plays a more important role when the probability is below 50%. Next, make the key event that plays a more important role a candidate for the critical key event.

4) Perform the abovementioned analysis chronologically up to the final key event (proliferation). As there is only one assay for ER-induced proliferation, calculate the probability that the OED extrapolated from the AC₅₀ value of this assay is greater than the distribution of OEDs of the critical key event candidate determined in 3) to identify the final candidates. Next, adopt the finally selected candidate as the critical key event that plays the most important role in inducing the uterotrophic response. Finally, determine the OED_{_crit_AC50} by averaging the OED_{_AC50}s extrapolated from the AC₅₀ values of the in vitro assays targeting the decided critical key event.

5) For OED_{_ACC}, perform steps 2)–4) to decide the critical key event and calculate the OED_{_crit_ACC}.

Furthermore, because no prior knowledge exists regarding the probability used here as a criterion for comparing the relative magnitudes of two discrete uniform distributions, we provisionally set this value at 50%, the midpoint between the minimum (0%) and maximum (100%). To assess the validity of this provisional probability, critical key events for THBP, DZ, and 4CP, together with their OED_{_crit_AC50} and OED_{_crit_ACC} values, were derived using the same procedure with criterion probabilities of 25% and 75%. These were then compared with the BMD₅₀ and BMD₂₀ values.

RESULTS

Chemicals

Among the chemicals in the MHLW Priority List, eight chemicals had ACC and AC₅₀ values in one or more assays for each of the key events of ER binding, receptor dimerization, DNA binding of the dimer, RNA transcription, protein production, and ER-induced proliferation. Furthermore, 20 chemicals exhibited significant estrogenic activity in the oral uterotrophic bioassay. Among these chemicals, none were inactive in all 16 in vitro assays. Conversely, the measured or estimated values of all chemical-specific parameters were available for 65 chemicals in the MHLW Priority List, making it possible to calculate the IVIVE conversion factors using the PBK model. Based on these results, the IVIVE approach was viable with the five chemicals shown in Table 3. Among them, three chemicals exhibited positive estrogenic activity in mouse uterotrophic bioassays, and the LOELs were identified. The remaining two chemicals had negative activity, with no significant effects observed even at the highest doses tested. In heptyl 4-hydroxybenzoate (HHB), one of the assays targeting ER binding as the key event was flagged as “Flag-Omit.” However, based on the data from the remaining two assays targeting the same key event, we determined that analysis was still feasible and therefore included it in the analysis described above.

Table 3. Five chemicals that were selected.

In vitro assays

The AC₅₀ and ACC values of the in vitro assays for the five selected chemicals are shown in Table S1. As shown in Fig. 2, the assays targeting ER binding (three assays), receptor dimerization (six assays), and protein production (two assays) showed greater variability in AC₅₀ values than those targeting DNA binding of the dimer (two assays) and RNA transcription (two assays). Furthermore, no consistent trend was observed in the distributions of AC₅₀ values for each key event of the five chemicals, and the same was true for the ACC values (figure not shown).

Fig. 2

Distribution of AC₅₀ values by key events in the 16 in vitro assays. The lower and upper ends of the box correspond to the 25th and 75th percentiles, respectively. The horizontal line in the box corresponds to the median.

In vivo uterotrophic bioassay in mice

The results of the in vivo uterotrophic bioassays and the calculated BMDs of the five selected chemicals are shown in Table S2. Three chemicals (i.e., 2,2′,4,4′-tetrahydroxybenzophenone (THBP), daidzein (DZ), and 4-cumylphenol (4CP)) showed positive estrogenic activity, and the LOELs could be determined. Moreover, since the BMDs could be calculated, we adopted the BMD values of the fitted models with the lowest AIC value (Table S2). Not much difference was observed between the LOEL and BMD₅₀ values or between the NOEL and BMD₂₀ values for these chemicals as shown in Table S2.

Verification of the PBK mouse model

The serum unconjugated BPA concentrations measured after oral administration of 0.4 and 100 mg/kg compared with those estimated using the PBK mouse model are shown in Fig. 3. All FE values were within a factor of 3, and the AFE and AAFE values were 0.78 and 1.99, respectively. Although these results tended to be slightly underestimated, they showed a nearly adequate fit. Furthermore, the chemical-specific parameters required for the PBK model (TPSA, log Kow, and Fup), excluding CLint, were not considerably different between the five selected chemicals and BPA. Therefore, no considerable differences in absorption, distribution, and renal excretion were expected between the five selected chemicals and BPA. Thus, we determined that the PBK mouse model can be used for IVIVE.

Fig. 3

Unbound BPA concentrations in mouse serum upon oral administration. The circles represent the unbound BPA concentrations in mouse serum following oral administration of 0.4 and 100 mg/kg measured by Taylor et al. (2011). The lines indicate the unbound BPA concentrations calculated by the PBK model. The orange areas are ranges from one-third to three times the calculated values.

IVIVE

The parameter values of the PBK mouse model and the calculated IVIVE conversion factors for the five chemicals, along with the OEDs that were extrapolated from the AC₅₀ and ACC values of each assay, are shown in Tables S3 and S4, respectively. The critical key events of the five chemicals and their OED_{_crit_AC50} and OED_{_crit_ACC} values were determined based on a comparison between distributions of the OEDs (Table S5). Table 4 shows the OED_{_crit_AC50} values that were calculated from the AC₅₀ values of the in vitro assays targeting the critical key events for the three chemicals for which LOELs were determined, together with the LOEL and BMD₅₀ values. More detailed and relevant data are provided in Table S3. The OED_{_crit_AC50} to LOEL ratios of THBP, DZ, and 4CP were 0.82, 1.83, and 0.76, respectively. Meanwhile, the OED_{_crit_AC50} to BMD₅₀ ratios of THBP, DZ, and 4CP were 1.71, 2.00, and 0.76, respectively. In addition, Table 5 shows the OED_{_crit_ACC} values that were derived from the ACC values of the in vitro assays targeting the critical key event, along with the NOEL and BMD₂₀ values. More detailed and relevant data are provided in Table S4. The OED_{_crit_ACC} to NOEL ratios of THBP, DZ, 4CP, HHB, and 2-ethylhexyl 4-hydroxybenzoate (2EHB) were 1.70, 3.40, 0.53, 1.70, and 5.10, respectively. Meanwhile, the OED_{_crit_ACC} to BMD₂₀ ratios of THBP, DZ, and 4CP were 2.04, 3.78, and 0.27, respectively. With 2EHB, large doses were predicted to be required to detect significant responses in the uterotrophic bioassays.

Table 4. Comparison of OED_{_crit_AC50} values to the observed LOEL and BMD₅₀ values.

Table 5. Comparison of OED_{_crit_ACC} values to the observed NOEL and BMD₂₀ values.

As shown in Table S6, the critical key events in THBP and 4CP shifted due to changes in the criterion probability used to identify them, which in turn altered the corresponding OED_{_crit_AC50} and OED_{_crit_ACC} values. The AAFE of the OED_{_crit_AC50}/BMD₅₀ ratio, calculated from the OED_{_crit_AC50} and BMD₅₀ values of these three chemicals, was lowest when the criterion probability was 50%. Likewise, the AAFE of the OED_{_crit_ACC}/BMD₂₀ ratio for the same chemicals was also minimal at a criterion probability of 50%. These findings suggest that, within the scope of this paper, using a 50% criterion probability for analysis is appropriate. However, when performing IVIVE on chemicals outside those examined here, the probability of the criterion should be considered separately.

DISCUSSION

Although the number of chemicals was small and the endpoint was only the acute uterotrophic response, the present study demonstrated that the activity concentrations (AC₅₀ and ACC) determined in the in vitro assays could be converted by the IVIVE approach to the corresponding OEDs for the in vivo bioassay (LOEL, NOEL, and BMDs). This finding suggests that an OED that can be used as a PoD for human health risk assessment can be derived. However, to derive these OEDs, the appropriate in vitro assays for applying the IVIVE approach should be selected based on the knowledge of key events in the pathways relevant to the in vivo toxicological outcomes of interest. Then, the critical key event playing the most important role in the occurrence of the in vivo outcome should be determined based on the activity concentrations of the selected in vitro assays. Herein, the mean OEDs for each of the six key events ranged from 20.6–821 mg/kg/day (THBP), 19.9–1,100 mg/kg/day (DZ), 28.8–761 mg/kg/day (4CP), 123–2,240 mg/kg/day (HHB), and 120–4,400 mg/kg/day (2EHB), with approximately 20- to 60-fold differences between the minimum and maximum values of these OEDs for each chemical. Although the full range of OEDs that were derived from all in vitro assays may be used directly in the screening stage to prioritize chemicals of concern for health risks, these differences indicate that the OED should be derived from in vitro assays relevant to critical key events for quantitative health risk assessment.

As described in the “IVIVE” subsection of the “Results” section, the calculated OED_{_crit_AC50} values for THBP, DZ, and 4CP showed no substantial difference from the LOEL and BMD₅₀ values of the uterotrophic bioassay. Similarly, the OED_{_crit_ACC} values for the five chemicals were not considerably different from the NOEL and BMD₂₀ values. As shown in Table 4, the ranges of OED_{_AC50} values used for averaging were 0.3–1.7 (THBP), 1.0–1.0 (DZ), 0.4–1.6 (4CP), 0.7–1.3 (HHB), and 0.2–1.8 (2EHB) times of the mean value (OED_{_crit_AC50}). Meanwhile, as shown in Table 5, the ranges of OED_{_ACC} values were 0.1–1.9 (THBP), 1.0–1.0 (DZ), 0.6–2.01.9 (4CP), 0.2–1.8 (HHB), and 0.02–2.0 (2EHB) times of the mean value (OED_{_crit_ACC}). Compared with OED_{_AC50}, these ranges appeared slightly broader. The AC₅₀ value represents the concentration producing 50% of maximum activity, whereas the ACC value does not correspond to a fixed activity level because the cutoff point varies by assay. This variability may contribute to the wider range of ratios. It should be noted that when the minimum and maximum OED_{_ACC} values of in vitro assays targeting a critical key event differ greatly, as in the case of 2EHB, the calculated OED_{_crit_ACC} will carry considerable uncertainty. To reduce such uncertainty and obtain reliable estimates of the OED that can serve as alternatives to NOEL and BMD, it is important to analyze concentration–activity data from each in vitro assay fitting models such as the Hill model or gain–loss model. The activity concentration corresponding to a given level (e.g., AC₀₅, AC₁₀) derived from this analysis should then be used to identify critical key events and derive the associated doses, such as OED_{_crit_AC05} and OED_{_crit_AC10}.

Two chemicals (HHB and 2EHB) were positive for estrogenic activity in the in vitro studies. However, they showed no response in mouse uterotrophic bioassays at oral doses up to 1,000 mg/kg/day (the maximum dose specified in OECD TG 440). These results suggest that HHB and 2EHB are negative for estrogenic activity. Although specific information on these two chemicals is lacking, it has been reported that esters of 4-hydroxybenzoic acid (parabens) are generally hydrolyzed in vivo by nonspecific esterases following oral administration (Ye et al., 2006). Thus, in addition to hepatic metabolism, hydrolysis by nonspecific esterases—widely distributed throughout the body—is also thought to reduce the peak plasma concentrations of the parent chemical. This reduction increases the IVIVE conversion factor, leading to higher OED values calculated from in vitro activity concentrations. Unfortunately, no information could be found on the model parameters for the hydrolysis of these two parabens. Therefore, while direct calculations cannot be performed, it is reasonable to expect that the OED_{_crit_AC50} and OED_{_crit_ACC} values for HHB and 2EHB would exceed the values of 1,700 and 5,100 mg/kg/day reported in Table 5. Once such information becomes available, the PBK mouse model coded in R can readily be adapted by modifying the mass balance equations to perform these calculations.

Although the PBK model plays an important role in IVIVE, the present study had limited mouse-specific information for the PKB model. Thus, the rat values were directly substituted for ka, tissue/blood partition coefficients, and Fup. This substitution of rat values is often performed in PBK mouse modeling, which is used to estimate the pharmacokinetics of volatile organic compounds following inhalation exposure. Furthermore, the aforementioned IVIVE study on the acute oral neurotoxicity of tetrodotoxin conducted by Noorlander et al. (2022) used the same parameter values in rats and mice. This suggests that applying rat values is the second-best option in the absence of mouse-specific data. However, future IVIVE studies will need to target chemicals with diverse toxicokinetic properties. Therefore, the availability of pharmacokinetic study data for model validation is limited. Continued efforts should be made to validate the PBK model with a variety of chemicals with different properties to reduce the uncertainty in its predictions and improve its reliability.

In light of the above, we concluded that although the technology for IVIVE itself is now largely established, further studies are required on a large number of chemicals to determine how to appropriately identify the critical key event in the occurrence of in vivo outcomes from the many in vitro assays available, and how to derive reliable OED value as the PoD for quantitative human health risk assessment from the activity concentrations of multiple in vitro assays relevant to the critical key event.

Supplementary data

Supplementary data will be made available in the online version of the paper.

Funding

This study was supported by a Health and Labour Sciences Research Grant (21KD2005 and 24KD2004) from the Ministry of Health, Labour and Welfare, Japan.

Conflict of interest

The authors declare that there is no conflict of interest.

Data availability

The data in this study are included in the article/supplementary materials. Contact the corresponding author(s) directly to request the underlying data.

Author contributions

Conceptualization: Kikuo Yoshida, Mariko Matsumoto

Funding acquisition: Mariko Matsumoto

Investigation: Kikuo Yoshida, Takaaki Umano, Mariko Matsumoto

Supervision: Mariko Matsumoto

Visualization: Kikuo Yoshida

Writing – original draft: Kikuo Yoshida

Writing – review & editing: Kikuo Yoshida, Takaaki Umano, Mariko Matsumoto, Takashi Yamada

Ethical approval and consent to participate

Not applicable.

Patient consent for publication

Not applicable.

REFERENCES

•  DeJongh,  J.,  Verhaar,  H.J. and  Hermens,  J.L. (1997): A quantitative property-property relationship (QPPR) approach to estimate in vitro tissue-blood partition coefficients of organic chemicals in rats and humans. Arch. Toxicol., 72, 17-25.
•  Fabian,  E.,  Gomes,  C.,  Birk,  B.,  Williford,  T.,  Hernandez,  T.R.,  Haase,  C.,  Zbranek,  R.,  van Ravenzwaay,  B. and  Landsiedel,  R. (2019): In vitro-to-in vivo extrapolation (IVIVE) by PBTK modeling for animal-free risk assessment approaches of potential endocrine-disrupting compounds. Arch. Toxicol., 93, 401-416.
•  Fagerholm,  U.,  Johansson,  M. and  Lennernäs,  H. (1996): Comparison between permeability coefficients in rat and human jejunum. Pharm. Res., 13, 1336-1342.
•  Hou,  T.J.,  Zhang,  W.,  Xia,  K.,  Qiao,  X.B. and  Xu,  X.J. (2004): ADME evaluation in drug discovery. 5. Correlation of Caco-2 permeation with simple molecular properties. J. Chem. Inf. Comput. Sci., 44, 1585-1600.
• Jongeneelen, F.J. and ten Berge, W. (2011). IndusChemFate: A multi-chemical PBTK-model in MS-Excel applicable for workers, consumers and experimental animals User manual, version 2.00. IndusTox Consult, Nijmegen, NL.
•  Judson,  R.S.,  Magpantay,  F.M.,  Chickarmane,  V.,  Haskell,  C.,  Tania,  N.,  Taylor,  J.,  Xia,  M.,  Huang,  R.,  Rotroff,  D.M.,  Filer,  D.L.,  Houck,  K.A.,  Martin,  M.T.,  Sipes,  N.,  Richard,  A.M.,  Mansouri,  K.,  Setzer,  R.W.,  Knudsen,  T.B.,  Crofton,  K.M. and  Thomas,  R.S. (2015): Integrated model of chemical perturbations of a biological pathway using 18 in vitro high-throughput screening assays for the estrogen receptor. Toxicol. Sci., 148, 137-154.
•  Kamiya,  Y.,  Handa,  K.,  Miura,  T.,  Ohori,  J.,  Kato,  A.,  Shimizu,  M.,  Kitajima,  M. and  Yamazaki,  H. (2022): Machine learning prediction of the three main input parameters of a simplified physiologically based pharmacokinetic model subsequently used to generate time-dependent plasma concentration data in humans after oral doses of 212 disparate chemicals. Biol. Pharm. Bull., 45, 124-128.
•  Kawashima,  H.,  Watanabe,  R.,  Esaki,  T.,  Kuroda,  M.,  Nagao,  C.,  Natsume-Kitatani,  Y.,  Ohashi,  R.,  Komura,  H. and  Mizuguchi,  K. (2023): DruMAP: A novel drug metabolism and pharmacokinetics analysis platform. J. Med. Chem., 66, 9697-9709.
•  Li,  H.,  Zhang,  M.,  Vervoort,  J.,  Rietjens,  I.M.,  van Ravenzwaay,  B. and  Louisse,  J. (2017): Use of physiologically based kinetic modeling-facilitated reverse dosimetry of in vitro toxicity data for prediction of in vivo developmental toxicity of tebuconazole in rats. Toxicol. Lett., 266, 85-93.
•  Nagata,  M. and  Sugimoto,  J. (1973): The influences of drugs on the experimental constipation of mice treated with atropine and papaverine. J. Kansai Med. Univ., 25, 300-321.
•  Noorlander,  A.,  Zhang,  M.,  van Ravenzwaay,  B. and  Rietjens,  I.M. (2022): Use of physiologically based kinetic modeling-facilitated reverse dosimetry to predict in vivo acute toxicity of tetrodotoxin in rodents. Toxicol. Sci., 187, 127-138.
• OECD. (2007): Test No. 440: Uterotrophic bioassay in rodents: A short-term screening test for oestrogenic properties. In: OECD Guidelines for the Testing of Chemicals, Sect, 4. OECD Publishing, Paris.
• OECD. (2021): Test No. 455: Performance-based test guideline for stably transfected transactivation in vitro assays to detect estrogen receptor agonists and antagonists. In: OECD Guidelines for the Testing of Chemicals, Sect, 4. OECD Publishing, Paris.
•  Pradeep,  P.,  Patlewicz,  G.,  Pearce,  R.,  Wambaugh,  J.,  Wetmore,  B. and  Judson,  R. (2020): Using chemical structure information to develop predictive models for in vitro toxicokinetic parameters to inform high-throughput risk-assessment. Comput. Toxicol., 16, 100136.
• R Core Team. (2023). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Available online at https://www.R-project.org/.
•  Soetaert,  K.,  Petzoldt,  T. and  Setzer,  R. (2010): Solving differential equations in R: package deSolve. J. Stat. Softw., 33, 1-25.
•  Sohlenius-Sternbeck,  A.K. (2006): Determination of the hepatocellularity number for human, dog, rabbit, rat and mouse livers from protein concentration measurements. Toxicol. In Vitro, 20, 1582-1586.
•  Strikwold,  M.,  Spenkelink,  B.,  Woutersen,  R.A.,  Rietjens,  I.M. and  Punt,  A. (2013): Combining in vitro embryotoxicity data with physiologically based kinetic (PBK) modelling to define in vivo dose-response curves for developmental toxicity of phenol in rat and human. Arch. Toxicol., 87, 1709-1723.
•  Strikwold,  M.,  Spenkelink,  B.,  de Haan,  L.H.,  Woutersen,  R.A.,  Punt,  A. and  Rietjens,  I.M. (2017): Integrating in vitro data and physiologically based kinetic (PBK) modelling to assess the in vivo potential developmental toxicity of a series of phenols. Arch. Toxicol., 91, 2119-2133.
•  Sun,  D.,  Lennernas,  H.,  Welage,  L.S.,  Barnett,  J.L.,  Landowski,  C.P.,  Foster,  D.,  Fleisher,  D.,  Lee,  K.-D. and  Amidon,  G.L. (2002): Comparison of human duodenum and Caco-2 gene expression profiles for 12,000 gene sequences tags and correlation with permeability of 26 drugs. Pharm. Res., 19, 1400-1416.
•  Taylor,  J.A.,  Vom Saal,  F.S.,  Welshons,  W.V.,  Drury,  B.,  Rottinghaus,  G.,  Hunt,  P.A.,  Toutain,  P.L.,  Laffont,  C.M. and  VandeVoort,  C.A. (2011): Similarity of bisphenol A pharmacokinetics in rhesus monkeys and mice: relevance for human exposure. Environ. Health Perspect., 119, 422-430.
•  Wegner,  S.H.,  Pinto,  C.L.,  Ring,  C.L. and  Wambaugh,  J.F. (2020): High-throughput screening tools facilitate calculation of a combined exposure-bioactivity index for chemicals with endocrine activity. Environ. Int., 137, 105470.
•  Wetmore,  B.A.,  Wambaugh,  J.F.,  Ferguson,  S.S.,  Sochaski,  M.A.,  Rotroff,  D.M.,  Freeman,  K.,  Clewell,  H.J. 3rd,  Dix,  D.J.,  Andersen,  M.E.,  Houck,  K.A.,  Allen,  B.,  Judson,  R.S.,  Singh,  R.,  Kavlock,  R.J.,  Richard,  A.M. and  Thomas,  R.S. (2012): Integration of dosimetry, exposure, and high-throughput screening data in chemical toxicity assessment. Toxicol. Sci., 125, 157-174.
•  Ye,  X.,  Bishop,  A.M.,  Reidy,  J.A.,  Needham,  L.L. and  Calafat,  A.M. (2006): Parabens as urinary biomarkers of exposure in humans. Environ. Health Perspect., 114, 1843-1846.
•  Yu,  L.X. and  Amidon,  G.L. (1999): A compartmental absorption and transit model for estimating oral drug absorption. Int. J. Pharm., 186, 119-125.
•  Zhang,  M.,  van Ravenzwaay,  B.,  Fabian,  E.,  Rietjens,  I.M. and  Louisse,  J. (2018): Towards a generic physiologically based kinetic model to predict in vivo uterotrophic responses in rats by reverse dosimetry of in vitro estrogenicity data. Arch. Toxicol., 92, 1075-1088.
•  Zhang,  M.,  van Ravenzwaay,  B. and  Rietjens,  I.M. (2020): Development of a generic physiologically based kinetic model to predict in vivo uterotrophic responses induced by estrogenic chemicals in rats based on in vitro bioassays. Toxicol. Sci., 173, 19-31.