Comment on Coumarin Hazard Identification

Recently, Hsieh et al ¹ reviewed and evaluated the scientific data on carcinogenic potential of coumarin. The authors examined, among other end points, the active ToxCast high-throughput screening assays for coumarin and metabolite 3,4-dihydrocoumarin. The authors correlate the reported activity with cancer potential using the Comparative Toxicogenomics Database. Although we support the use of new approach methodologies in a weight of evidence assessment, there are several weaknesses in the authors’ use and interpretation of the ToxCast data.

Review of Dose–Response Curves and Assay Flags

A careful review of the dose–response curves for the active coumarin and 3,4-dihydrocoumarin assays from the United States Environmental Protection Agency (EPA) Interactive Chemical Safety for Sustainability Dashboard output version 2, identified curve fitting flags for 8 of the 13 assays for coumarin, and 1 of 1 assays for 3,4-dihydrocoumarin. These flags, designed by EPA through the tcpl data processing package to minimize false negatives, are intended to highlight possible issues with the curve fits and active hit calls and should be reviewed manually. ² In other words, these assays have the potential to generate false-positive responses and should be carefully reviewed in the context of use. Assay flags associated with the coumarin and 3,4-dihydrocoumarin include borderline active, only 1 concentration above baseline, close to the assay cutoff, noisy data and hit call potentially confounded by overfitting. These flags call into question the bioactivity of 9 of the assays for coumarin and 3,4-dihydrocoumarin with false-positive responses based on lack of clear dose–response curves, barely active responses above the background of the assay, and variable in responses where the assays were run multiple times.

Review of Biologically Consistent Responses Across Orthogonal Assays and/or Related Targets

Because of the high-throughput nature of ToxCast assays, one can’t expect all assays covering a specific biological target to be active when exposed to a particular chemical. ³ However, one would have greater confidence in the results if there are several robust assays to validate a specific target. ^4,5 For the 5 assays considered active for coumarin, after removing the 9 chemical/assay pairs based on false-positive responses as discussed above, a closer investigation of the biological consistency of the responses across orthogonal assays and/or related targets shows was performed.

NVS_ENZ_rMAOBC and NVS_ENZ_rMAOBP: The monoamine oxidase (MAO) cell-free assays are active at high coumarin concentrations only with the AC50 in the 15 to 20 µM range. Although 2 assays within the MAO family are active following coumarin exposure, 4 MAO assays are inactive.

TOX21_Era_LUC_BGI_Agonist: This is 1 of 18 estrogen receptor alpha agonist assays within ToxCast that covers multiple key events leading to estrogen receptor activation. ⁶ The prediction model for estrogen active chemical comprises of 18 assays and a lack of response of 17 assays within this model suggests that bioactivity in TOX21_Era_LUC_BGI_Agonist following coumarin exposure is likely a nonspecific response.

ATG_RXRb_TRANS_dn: Within the ToxCast high-throughput screening (HTS) assays, there are 5 assays that measure nuclear receptor RXRa expression, another 4 assays for RXRb, and RXRg. However, following exposure to coumarin, only one assay within the RXR family, specifically ATG_RXRb_TRANS_dn, was active. The limited response in RXR nuclear receptor assays to coumarin exposure suggests that this is likely a nonspecific response.

APR_HepG2_MitoMembPot_72h_dn: The Apredica (APR) technology uses 2 cell types and 3 time points to evaluate the impact of chemicals on mitochondrial membrane potential. ³ Coumarin is only positive for one of these assays, APR_HepG2_MitoMembPot_72h_dn: The lack of response in the other related assays combined with lack of target engagement data suggests that this is likely a nonspecific response.

From this analysis, 3 of the active assays, namely TOX21_Era_LUC_BGI_Agonist, ATG_RXRb_TRANS_dn, and APR_HepG2_MitoMembPot_72h_dn, show a lack of biological concordance as related assays did not show activity following coumarin exposure. Only NVS_ENZ_rMAOBC and NVS_ENZ_rMAOBP show consistent activity at high concentrations in the cell-free system following exposure to coumarin. This conclusion, that monoamine oxidase activity, is the only credible activity associated with coumarin from the ToxCast assays, was also shown in a recent publication where a next generation risk assessment approach was employed. ⁷

In addition, Wetmore et al analyzed bioactivity of commonly consumed fruits and vegetables, which are generally recognized as safe, and compared to the activity of agrichemicals tested in ToxCast. ⁸ The fruit and vegetable extracts were bioactive in several in vitro assays supporting the conclusion that bioactivity does not equate to adverse effects. This illustrates the complexity of drawing conclusions from the bioactivity of a compound in ToxCast to an apical, histological end point in toxicity tests.

Overall, we believe it’s critical to call attention to the need for a closer investigation into the ToxCast HTS data that are used in a weight of evidence assessment to associate toxicological effects with a chemical.