In Vitro Exposure Assessment Improves Hazard Ranking of Bisphenol A Alternatives

Bisphenol A (BPA) was used for decades as a plastic chemical, but due to its endocrine disrupting properties it is now banned for food-contact materials (1) and thermal paper (2) in the European Union. Numerous BPA replacement chemicals have therefore been developed, but for most of them effects to human and environmental health have not been thoroughly assessed. (3, 4) Cell-based in vitro bioassays allow a fast screening for adverse effects and can help to identify which alternatives show similar toxicity compared to BPA itself and which ones can be considered more benign. Numerous in vitro toxicity studies have been published, but the majority has either focused on only a few of the BPA alternatives or few specific end points like estrogenicity and androgenicity (5, 6) or adipogenesis. (7, 8) The most comprehensive in vitro toxicity datasets for BPA and its alternatives can be found in the recent publication by Srebny et al. (9) that reports data from six assays for BPA and 26 alternatives and in the Tox21 database, which contains data from various assays for BPA and 15 alternatives as reviewed by Adamovsky et al. (4)

To date, comparative hazard assessment using in vitro toxicity data for BPA and its alternatives was exclusively done based on nominal effect concentrations. (9−12) However, nominal concentrations do not necessarily correlate with the bioactive concentration, because the test concentration of the chemicals added at the beginning of the bioassay can be reduced by plastic sorption, volatilization, cellular metabolism, abiotic degradation and binding to cell culture medium constituents (Figure 1). (13) Some of these binding and loss processes are only relevant for certain compound classes or assay types. For example, plastic sorption was shown to be negligible for assays that use a high serum content, i.e., 10% fetal bovine serum (FBS) in 384-well plates. (14) Furthermore, many cell lines applied in high-throughput screening assays (e.g., HEK293 cells) have limited metabolic capacity, which makes this loss process also less relevant. (15) Volatility can be evaluated using the medium-air partition constant (K_medium/air) of the test chemicals. Previous studies have shown that chemicals with log K_medium/air <4 are lost from the assay plates or even cross-contaminate neighboring wells and cannot be tested in open well plate formats. (16, 17)

Chem. Res. Toxicol. (2026): Figure 1. Binding and loss processes in cell-based in vitro bioassays and their relevance for BPA and its alternatives. Created in BioRender. Henneberger, L. (2026)

The SPME method was successfully optimized for BPA and 15 of the alternatives. Binding to the pin coating as well as equilibration times were found to increase with increasing hydrophobicity (Figure S4). Chemical recoveries (sum of amount in PBS and pin coating) for the optimization experiments in PBS were high (76-111%). Several chemicals showed nonlinear sorption to the pin coating, which had to be considered in the C_free calculation (Figure S3). Despite the good performance of the SPME method, the testable concentration range was limited for some BPA alternatives due to high LOQs (e.g., 1 μg/mL for BPP and 0.5 μg/mL for BPZ).

Chem. Res. Toxicol. (2026): Figure 3. Comparison of measured (EXP) and modelled (MBM) (A) bovine serum albumin-water distribution ratios (DBSA/w) and (B) membrane or liposome-water distribution ratios (Dm/w or Dlip/w), (C) dependence of measured (EXP) and modelled (MBM) free fractions (ffree) in assay medium (OptiMEM base medium +2% cs-FBS) of the test chemicals on the membrane-water distribution ratio (Dm/w), (D) comparison of average measured (EXP) and modelled (MBM) ffree, (E) comparison of measured freely dissolved concentrations (Cfree) with dosed nominal concentration (Cnom), (F) comparison of measured (EXP) and modelled (MBM) Cfree. Solid lines in A, B, D, E and F indicate the 1:1 line, the dotted lines indicate the 1:10 and 10:1 lines. AAFE (absolute average fold error).

Binding to BSA was measured for BPA and 15 of the alternatives (Figure S5, abbreviations in Table 1). For BPF, BPAF, BP-MIBK, BPA, BPS-MAE and D8 binding isotherms were linear, while for BPT, BPZ, BzBP, Bz, BPAP, BPB, BPE and BPS-MPE binding to BSA was concentration-dependent (Table S6). Table 2 reports the mean D_BSA/w. BPS and 24BPS showed weak affinities for BSA. For BPS D_BSA/w could only be determined at one concentration level and no isotherm was fitted. For 24BPS no binding could be measured at any of the tested concentrations indicating f_bound < 20% at the tested BSA concentration of 1.56 g/L. Figure 3A shows a comparison of the measured D_BSA/w with the values predicted from log K_ow by Srebny et al. (9) Because regression models based on log K_ow do not consider concentration dependence, the average D_BSA/w was calculated and used for the comparison in Fig. 3A even for the chemicals with nonlinear binding isotherms, resulting in generally larger error bars (see also Table 2). On average, the measured D_BSA/w agreed with the modelled values within a factor of 4.7 (AAFE). For most chemicals the measured and predicted D_BSA/w agreed within a factor of 7. For benzylparabene (BzBP), BPS-MAE and D8 the measured D_BSA/w were a factor of 17, 24, and 35 higher compared to the value predicted from log K_ow. For BPA the measured log D_BSA/w of 3.57 agrees well with literature data (29) (D_BSA/w = 3.65). This literature value was also used for the comparison in Figure 3A.

The same concept as applied in the present study can also be used for other classes of chemicals. Abiotic stability should always be measured first. For unstable chemicals hazard assessment will need to include the formed transformation products as well and a concentration metric that considers temporal changes (e.g., area under the curve). The same applies to stable chemicals if cell lines with high metabolic capacity are used for in vitro testing (e.g., HepG2 or HepaRG). For stable and not quickly metabolized chemicals EC_free should be used for hazard ranking, if the chemicals show significant binding to medium proteins and lipids or high chemical diversity like the BPA alternatives investigated in the present study. Simple prediction models can provide a fast alternative to experimental determination of f_free, but they rely on the quality of the used chemical input parameters and are less accurate for organic acids that show nonlinear protein binding.

Finally, for IVIVE, metabolism of the BPA alternatives needs to be considered as well. Metabolization does not only differ between BPA and its alternatives, but also between different cell lines and between in vitro and in vivo. For mechanistic understanding of metabolism and its influence on toxicity it might be advisible to use a defined external metabolic system as used by Srebny et al. (9) with non-metabolically competent cell lines before testing directly in metabolically competent cells. Metabolization generally decreases estrogenicity and cytotoxicity of BPA and its alternatives, (9) but data for other end points are largely missing.