Uncertainty quantification in ToxCast high throughput screening

MLinvitroTox reloaded for high-throughput hazard-based prioritization of high-resolution mass spectrometry data

MLinvitroTox is an automated Python pipeline developed for high-throughput hazard-driven prioritization of toxicologically relevant signals detected in complex environmental samples through high-resolution tandem mass spectrometry (HRMS/MS). MLinvitroTox is a machine learning (ML) framework comprising 490 independent XGBoost classifiers trained on molecular fingerprints from chemical structures and target-specific endpoints from the ToxCast/Tox21 invitroDBv4.1 database. For each analyzed HRMS feature, MLinvitroTox generates a 490-bit bioactivity fingerprint used as a basis for prioritization, focusing the time-consuming molecular identification efforts on features most likely to cause adverse effects. The practical advantages of MLinvitroTox are demonstrated for groundwater HRMS data. Among the 874 features for which molecular fingerprints were derived from spectra, including 630 nontargets, 185 spectral matches, and 59 targets, around 4% of the feature/endpoint relationship pairs were predicted to be active. Cross-checking the predictions for targets and spectral matches with invitroDB data confirmed the bioactivity of 120 active and 6791 nonactive pairs while mislabeling 88 active and 56 non-active relationships. By filtering according to bioactivity probability, endpoint scores, and similarity to the training data, the number of potentially toxic features was reduced by at least one order of magnitude. This refinement makes the analytical confirmation of the toxicologically most relevant features feasible, offering significant benefits for cost-efficient chemical risk assessment.Scientific Contribution:In contrast to the classical ML-based approaches for toxicity prediction, MLinvitroTox predicts bioactivity for HRMS features (i.e., distinct m/z signals) based on MS2 fragmentation spectra rather than the chemical structures from the identified features. While the original proof of concept study was accompanied by the release of a MLinvitroTox v1 KNIME workflow, in this study, we release a Python MLinvitroTox v2 package, which, in addition to automation, expands functionality to include predicting toxicity from structures, cleaning up and generating chemical fingerprints, customizing models, and retraining on custom data. Furthermore, as a result of improvements in bioactivity data processing, realized in the concurrently released pytcpl Python package for the custom processing of invitroDBv4.1 input data used for training MLinvitroTox, the current release introduces enhancements in model accuracy, coverage of biological mechanistic targets, and overall interpretability.

Chemistry domain of applicability evaluation against existing estrogen receptor high-throughput assay-based activity models

Introduction

The U. S. Environmental Protection Agency's Endocrine Disruptor Screening Program (EDSP) Tier 1 assays are used to screen for potential endocrine system-disrupting chemicals. A model integrating data from 16 high-throughput screening assays to predict estrogen receptor (ER) agonism has been proposed as an alternative to some low-throughput Tier 1 assays. Later work demonstrated that as few as four assays could replicate the ER agonism predictions from the full model with 98% sensitivity and 92% specificity. The current study utilized chemical clustering to illustrate the coverage of the EDSP Universe of Chemicals (UoC) tested in the existing ER pathway models and to investigate the utility of chemical clustering to evaluate the screening approach using an existing 4-assay model as a test case. Although the full original assay battery is no longer available, the demonstrated contribution of chemical clustering is broadly applicable to assay sets, chemical inventories, and models, and the data analysis used can also be applied to future evaluation of minimal assay models for consideration in screening.

Methods

Chemical structures were collected for 6,947 substances via the CompTox Chemicals Dashboard from the over 10,000 UoC and grouped based on structural similarity, generating 826 chemical clusters. Of the 1,812 substances run in the original ER model, 1,730 substances had a single, clearly defined structure. The ER model chemicals with a clearly defined structure that were not present in the EDSP UoC were assigned to chemical clusters using a k-nearest neighbors approach, resulting in 557 EDSP UoC clusters containing at least one ER model chemical.

Results and Discussion

Performance of an existing 4-assay model in comparison with the existing full ER agonist model was analyzed as related to chemical clustering. This was a case study, and a similar analysis can be performed with any subset model in which the same chemicals (or subset of chemicals) are screened. Of the 365 clusters containing >1 ER model chemical, 321 did not have any chemicals predicted to be agonists by the full ER agonist model. The best 4-assay subset ER agonist model disagreed with the full ER agonist model by predicting agonist activity for 122 chemicals from 91 of the 321 clusters. There were 44 clusters with at least two chemicals and at least one agonist based upon the full ER agonist model, which allowed accuracy predictions on a per-cluster basis. The accuracy of the best 4-assay subset ER agonist model ranged from 50% to 100% across these 44 clusters, with 32 clusters having accuracy ≥90%. Overall, the best 4-assay subset ER agonist model resulted in 122 false-positive and only 2 false-negative predictions compared with the full ER agonist model. Most false positives (89) were active in only two of the four assays, whereas all but 11 true positive chemicals were active in at least three assays. False positive chemicals also tended to have lower area under the curve (AUC) values, with 110 out of 122 false positives having an AUC value below 0.214, which is lower than 75% of the positives as predicted by the full ER agonist model. Many false positives demonstrated borderline activity. The median AUC value for the 122 false positives from the best 4-assay subset ER agonist model was 0.138, whereas the threshold for an active prediction is 0.1.

Conclusion

Our results show that the existing 4-assay model performs well across a range of structurally diverse chemicals. Although this is a descriptive analysis of previous results, several concepts can be applied to any screening model used in the future. First, the clustering of the chemicals provides a means of ensuring that future screening evaluations consider the broad chemical space represented by the EDSP UoC. The clusters can also assist in prioritizing future chemicals for screening in specific clusters based on the activity of known chemicals in those clusters. The clustering approach can be useful in providing a framework to evaluate which portions of the EDSP UoC chemical space are reliably covered by in silico and in vitro approaches and where predictions from either method alone or both methods combined are most reliable. The lessons learned from this case study can be easily applied to future evaluations of model applicability and screening to evaluate future datasets.

A Comparison of In Vitro Points of Departure with Human Blood Levels for Per- and Polyfluoroalkyl Substances (PFAS)

Per- and polyfluoroalkyl substances (PFAS) are widely used, and their fluorinated state contributes to unique uses and stability but also long half-lives in the environment and humans. PFAS have been shown to be toxic, leading to immunosuppression, cancer, and other adverse health outcomes. Only a small fraction of the PFAS in commerce have been evaluated for toxicity using in vivo tests, which leads to a need to prioritize which compounds to examine further. Here, we demonstrate a prioritization approach that combines human biomonitoring data (blood concentrations) with bioactivity data (concentrations at which bioactivity is observed in vitro) for 31 PFAS. The in vitro data are taken from a battery of cell-based assays, mostly run on human cells. The result is a Bioactive Concentration to Blood Concentration Ratio (BCBCR), similar to a margin of exposure (MoE). Chemicals with low BCBCR values could then be prioritized for further risk assessment. Using this method, two of the PFAS, PFOA (Perfluorooctanoic Acid) and PFOS (Perfluorooctane Sulfonic Acid), have BCBCR values < 1 for some populations. An additional 9 PFAS have BCBCR values < 100 for some populations. This study shows a promising approach to screening level risk assessments of compounds such as PFAS that are long-lived in humans and other species.

Investigating open access new approach methods (NAM) to assess biological points of departure: A case study with 4 neurotoxic pesticides

Open access new approach methods (NAM) in the US EPA ToxCast program and NTP Integrated Chemical Environment (ICE) were used to investigate activities of four neurotoxic pesticides: endosulfan, fipronil, propyzamide and carbaryl. Concordance of in vivo regulatory points of departure (POD) adjusted for interspecies extrapolation (AdjPOD) to modelled human Administered Equivalent Dose (AEDHuman) was assessed using 3-compartment or Adult/Fetal PBTK in vitro to in vivo extrapolation. Model inputs were from Tier 1 (High throughput transcriptomics: HTTr, high throughput phenotypic profiling: HTPP) and Tier 2 (single target: ToxCast) assays. HTTr identified gene expression signatures associated with potential neurotoxicity for endosulfan, propyzamide and carbaryl in non-neuronal MCF-7 and HepaRG cells. The HTPP assay in U-2 OS cells detected potent effects on DNA endpoints for endosulfan and carbaryl, and mitochondria with fipronil (propyzamide was inactive). The most potent ToxCast assays were concordant with specific components of each chemical mode of action (MOA). Predictive adult IVIVE models produced fold differences (FD) < 10 between the AEDHuman and the measured in vivo AdjPOD. The 3-compartment model was concordant (i.e., smallest FD) for endosulfan, fipronil and carbaryl, and PBTK was concordant for propyzamide. The most potent AEDHuman predictions for each chemical showed HTTr, HTPP and ToxCast were mainly concordant with in vivo AdjPODs but assays were less concordant with MOAs. This was likely due to the cell types used for testing and/or lack of metabolic capabilities and pathways available in vivo. The Fetal PBTK model had larger FDs than adult models and was less predictive overall.

The ToxCast pipeline: updates to curve-fitting approaches and database structure

Introduction: The US Environmental Protection Agency Toxicity Forecaster (ToxCast) program makes in vitro medium- and high-throughput screening assay data publicly available for prioritization and hazard characterization of thousands of chemicals. The assays employ a variety of technologies to evaluate the effects of chemical exposure on diverse biological targets, from distinct proteins to more complex cellular processes like mitochondrial toxicity, nuclear receptor signaling, immune responses, and developmental toxicity. The ToxCast data pipeline (tcpl) is an open-source R package that stores, manages, curve-fits, and visualizes ToxCast data and populates the linked MySQL Database, invitrodb. Methods: Herein we describe major updates to tcpl and invitrodb to accommodate a new curve-fitting approach. The original tcpl curve-fitting models (constant, Hill, and gain-loss models) have been expanded to include Polynomial 1 (Linear), Polynomial 2 (Quadratic), Power, Exponential 2, Exponential 3, Exponential 4, and Exponential 5 based on BMDExpress and encoded by the R package dependency, tcplfit2. Inclusion of these models impacted invitrodb (beta version v4.0) and tcpl v3 in several ways: (1) long-format storage of generic modeling parameters to permit additional curve-fitting models; (2) updated logic for winning model selection; (3) continuous hit calling logic; and (4) removal of redundant endpoints as a result of bidirectional fitting. Results and discussion: Overall, the hit call and potency estimates were largely consistent between invitrodb v3.5 and 4.0. Tcpl and invitrodb provide a standard for consistent and reproducible curve-fitting and data management for diverse, targeted in vitro assay data with readily available documentation, thus enabling sharing and use of these data in myriad toxicology applications. The software and database updates described herein promote comparability across multiple tiers of data within the US Environmental Protection Agency CompTox Blueprint.

Use of computational toxicology models to predict toxicological points of departure: A case study with triazine herbicides

BACKGROUND

Atrazine simazine and propazine, widely used triazine herbicides on food crops and in residential areas, disrupt the neuroendocrine system raising human health concerns. USEPA developed a PBPK model based on triazine common Mode of Action (MOA)-suppression of luteinizing hormone surge in female rats-to generate human regulatory points of departure (POD: mg/kg/day). We compared triazine Human Administered Equivalent Dose (AED_Human mg/kg/day) predictions from open access computational tools to the PBPK PODs to assess concordance.

METHODS

Computational tools were the following: ToxCast/Tox21 in vitro assays; Toxicogenomic databases to assess concordance with ToxCast/Tox21 targets; integrated chemical environment (ICE) models with ToxCast/Tox21 inputs to predict AED_Human PODs and population-based age-refined high throughput toxicokinetics (HTTK-Pop) to compare to age-related PBPK PODs.

RESULTS

ToxCast/Tox21 assays identified critical targets in the triazine common MOA and gene databases; ICE AED_Human predictions were mainly concordant with the USEPA PBPK PODs quantitatively. Low fold-differences between PBPK POD and ICE AED_Human predictions indicated that the ICE models are health-protective. HTTK-Pop age-refinements were within 10-fold of the USEPA PBPK PODs.

CONCLUSIONS

CompTox tools were used to identify assay targets in the MOA and identify potential molecular initiating targets in the adverse outcome pathway for potential use in risk assessment.

A Pragmatic Framework for the Application of New Approach Methodologies in One Health Toxicological Risk Assessment

Globally, industries and regulatory authorities are faced with an urgent need to assess the potential adverse effects of chemicals more efficiently by embracing new approach methodologies (NAMs). NAMs include cell and tissue methods (in vitro), structure-based/toxicokinetic models (in silico), methods that assess toxicant interactions with biological macromolecules (in chemico), and alternative models. Increasing knowledge on chemical toxicokinetics (what the body does with chemicals) and toxicodynamics (what the chemicals do with the body) obtained from in silico and in vitro systems continues to provide opportunities for modernizing chemical risk assessments. However, directly leveraging in vitro and in silico data for derivation of human health-based reference values has not received regulatory acceptance due to uncertainties in extrapolating NAM results to human populations, including metabolism, complex biological pathways, multiple exposures, interindividual susceptibility and vulnerable populations. The objective of this article is to provide a standardized pragmatic framework that applies integrated approaches with a focus on quantitative in vitro to in vivo extrapolation (QIVIVE) to extrapolate in vitro cellular exposures to human equivalent doses from which human reference values can be derived. The proposed framework intends to systematically account for the complexities in extrapolation and data interpretation to support sound human health safety decisions in diverse industrial sectors (food systems, cosmetics, industrial chemicals, pharmaceuticals etc.). Case studies of chemical entities, using new and existing data, are presented to demonstrate the utility of the proposed framework while highlighting potential sources of human population bias and uncertainty, and the importance of Good Method and Reporting Practices.

Characterizing Variability and Uncertainty Associated with Transcriptomic Dose-Response Modeling

Transcriptomics dose-response analysis (TDRA) has emerged as a promising approach for integrating toxicogenomics data into a risk assessment context; however, variability and uncertainty associated with experimental design are not well understood. Here, we evaluated n = 55 RNA-seq profiles derived from Japanese quail liver tissue following exposure to chlorpyrifos (0, 0.04, 0.1, 0.2, 0.4, 1, 2, 4, 10, 20, and 40 μg/g; n = 5 replicates per group) via egg injection. The full dataset was subsampled 637 times to generate smaller datasets with different dose ranges and spacing (designs A-E) and number of replicates (n = 2-5). TDRA of the 637 datasets revealed substantial variability in the gene and pathway benchmark doses, but relative stability in overall transcriptomic point-of-departure (tPOD) values when tPODs were calculated with the "pathway" and "mode" methods. Further, we found that tPOD values were more dependent on the dose range and spacing than on the number of replicates, suggesting that optimal experimental designs should use fewer replicates (n = 2 or 3) and more dose groups to reduce uncertainty in the results. Finally, tPOD values ranged by over ten times for all surveyed experimental designs and tPOD types, suggesting that tPODs should be interpreted as order-of-magnitude estimates.

High-throughput screening paradigms in ecotoxicity testing: Emerging prospects and ongoing challenges

The rapidly increasing number of new production chemicals coupled with stringent implementation of global chemical management programs necessities a paradigm shift towards boarder uses of low-cost and high-throughput ecotoxicity testing strategies as well as deeper understanding of cellular and sub-cellular mechanisms of ecotoxicity that can be used in effective risk assessment. The latter will require automated acquisition of biological data, new capabilities for big data analysis as well as computational simulations capable of translating new data into in vivo relevance. However, very few efforts have been so far devoted into the development of automated bioanalytical systems in ecotoxicology. This is in stark contrast to standardized and high-throughput chemical screening and prioritization routines found in modern drug discovery pipelines. As a result, the high-throughput and high-content data acquisition in ecotoxicology is still in its infancy with limited examples focused on cell-free and cell-based assays. In this work we outline recent developments and emerging prospects of high-throughput bioanalytical approaches in ecotoxicology that reach beyond in vitro biotests. We discuss future importance of automated quantitative data acquisition for cell-free, cell-based as well as developments in phytotoxicity and in vivo biotests utilizing small aquatic model organisms. We also discuss recent innovations such as organs-on-a-chip technologies and existing challenges for emerging high-throughput ecotoxicity testing strategies. Lastly, we provide seminal examples of the small number of successful high-throughput implementations that have been employed in prioritization of chemicals and accelerated environmental risk assessment.

Evaluating structure-based activity in a high-throughput assay for steroid biosynthesis

Data from a high-throughput human adrenocortical carcinoma assay (HT-H295R) for steroid hormone biosynthesis are available for >2000 chemicals in single concentration and 654 chemicals in multi-concentration (mc). Previously, a metric describing the effect size of a chemical on the biosynthesis of 11 hormones was derived using mc data referred to as the maximum mean Mahalanobis distance (maxmMd). However, mc HT-H295R assay data remain unavailable for many chemicals. This work leverages existing HT-H295R assay data by constructing structure-activity relationships to make predictions for data-poor chemicals, including: (1) identification of individual structural descriptors, known as ToxPrint chemotypes, associated with increased odds of affecting estrogen or androgen synthesis; (2) a random forest (RF) classifier using physicochemical property descriptors to predict HT-H295R maxmMd binary (positive or negative) outcomes; and, (3) a local approach to predict maxmMd binary outcomes using nearest neighbors (NNs) based on two types of chemical fingerprints (chemotype or Morgan). Individual chemotypes demonstrated high specificity (85-98%) for modulators of estrogen and androgen synthesis but with low sensitivity. The best RF model for maxmMd classification included 13 predicted physicochemical descriptors, yielding a balanced accuracy (BA) of 71% with only modest improvement when hundreds of structural features were added. The best two NN models for binary maxmMd prediction demonstrated BAs of 85 and 81% using chemotype and Morgan fingerprints, respectively. Using an external test set of 6302 chemicals (lacking HT-H295R data), 1241 were identified as putative estrogen and androgen modulators. Combined results across the three classification models (global RF model and two local NN models) predict that 1033 of the 6302 chemicals would be more likely to affect HT-H295R bioactivity. Together, these in silico approaches can efficiently prioritize thousands of untested chemicals for screening to further evaluate their effects on steroid biosynthesis.

Impact of High-Throughput Model Parameterization and Data Uncertainty on Thyroid-Based Toxicological Estimates for Pesticide Chemicals

Chemical-induced alteration of maternal thyroid hormone levels may increase the risk of adverse neurodevelopmental outcomes in offspring. US federal risk assessments rely almost exclusively on apical endpoints in animal models for deriving points of departure (PODs). New approach methodologies (NAMs) such as high-throughput screening (HTS) and mechanistically informative in vitro human cell-based systems, combined with in vitro to in vivo extrapolation (IVIVE), supplement in vivo studies and provide an alternative approach to calculate/determine PODs. We examine how parameterization of IVIVE models impacts the comparison between IVIVE-derived equivalent administered doses (EADs) from thyroid-relevant in vitro assays and the POD values that serve as the basis for risk assessments. Pesticide chemicals with thyroid-based in vitro bioactivity data from the US Tox21 HTS program were included (n = 45). Depending on the model structure used for IVIVE analysis, up to 35 chemicals produced EAD values lower than the POD. A total of 10 chemicals produced EAD values higher than the POD regardless of the model structure. The relationship between IVIVE-derived EAD values and the in vivo-derived POD values is highly dependent on model parameterization. Here, we derive a range of potentially thyroid-relevant doses that incorporate uncertainty in modeling choices and in vitro assay data.

Food, Beverage, and Feedstock Processing Facility Wastewater: a Unique and Underappreciated Source of Contaminants to U.S. Streams

Process wastewaters from food, beverage, and feedstock facilities, although regulated, are an under-investigated environmental contaminant source. Food process wastewaters (FPWWs) from 23 facilities in 17 U.S. states were sampled and documented for a plethora of chemical and microbial contaminants. Of the 576 analyzed organics, 184 (32%) were detected at least once, with concentrations as large as 143 μg L^-1 (6:2 fluorotelomer sulfonic acid), and as many as 47 were detected in a single FPWW sample. Cumulative per/polyfluoroalkyl substance concentrations up to 185 μg L^-1 and large pesticide transformation product concentrations (e.g., methomyl oxime, 40 μg L^-1; clothianidin TMG, 2.02 μg L^-1) were observed. Despite 48% of FPWW undergoing disinfection treatment prior to discharge, bacteria resistant to third-generation antibiotics were found in each facility type, and multiple bacterial groups were detected in all samples, including total coliforms. The exposure-activity ratios and toxicity quotients exceeded 1.0 in 13 and 22% of samples, respectively, indicating potential biological effects and toxicity to vertebrates and invertebrates associated with the discharge of FPWW. Organic contaminant profiles of FPWW differed from previously reported contaminant profiles of municipal effluents and urban storm water, indicating that FPWW is another important source of chemical and microbial contaminant mixtures discharged into receiving surface waters.

The effect of noise on the predictive limit of QSAR models

A key challenge in the field of Quantitative Structure Activity Relationships (QSAR) is how to effectively treat experimental error in the training and evaluation of computational models. It is often assumed in the field of QSAR that models cannot produce predictions which are more accurate than their training data. Additionally, it is implicitly assumed, by necessity, that data points in test sets or validation sets do not contain error, and that each data point is a population mean. This work proposes the hypothesis that QSAR models can make predictions which are more accurate than their training data and that the error-free test set assumption leads to a significant misevaluation of model performance. This work used 8 datasets with six different common QSAR endpoints, because different endpoints should have different amounts of experimental error associated with varying complexity of the measurements. Up to 15 levels of simulated Gaussian distributed random error was added to the datasets, and models were built on the error laden datasets using five different algorithms. The models were trained on the error laden data, evaluated on error-laden test sets, and evaluated on error-free test sets. The results show that for each level of added error, the RMSE for evaluation on the error free test sets was always better. The results support the hypothesis that, at least under the conditions of Gaussian distributed random error, QSAR models can make predictions which are more accurate than their training data, and that the evaluation of models on error laden test and validation sets may give a flawed measure of model performance. These results have implications for how QSAR models are evaluated, especially for disciplines where experimental error is very large, such as in computational toxicology.

Cardiotoxicity Hazard and Risk Characterization of ToxCast Chemicals Using Human Induced Pluripotent Stem Cell-Derived Cardiomyocytes from Multiple Donors

Heart disease remains a significant human health burden worldwide with a significant fraction of morbidity attributable to environmental exposures. However, the extent to which the thousands of chemicals in commerce and the environment may contribute to heart disease morbidity is largely unknown, because in contrast to pharmaceuticals, environmental chemicals are seldom tested for potential cardiotoxicity. Human induced pluripotent stem cell (iPSC)-derived cardiomyocytes have become an informative in vitro model for cardiotoxicity testing of drugs with the availability of cells from multiple individuals allowing in vitro testing of population variability. In this study, we hypothesized that a panel of iPSC-derived cardiomyocytes from healthy human donors can be used to screen for the potential cardiotoxicity hazard and risk of environmental chemicals. We conducted concentration-response testing of 1029 chemicals (drugs, pesticides, flame retardants, polycyclic aromatic hydrocarbons (PAHs), plasticizers, industrial chemicals, food/flavor/fragrance agents, etc.) in iPSC-derived cardiomyocytes from 5 donors. We used kinetic calcium flux and high-content imaging to derive quantitative measures as inputs into Bayesian population concentration-response modeling of the effects of each chemical. We found that many environmental chemicals pose a hazard to human cardiomyocytes in vitro with more than half of all chemicals eliciting positive or negative chronotropic or arrhythmogenic effects. However, most of the tested environmental chemicals for which human exposure and high-throughput toxicokinetics data were available had wide margins of exposure and, thus, do not appear to pose a significant human health risk in a general population. Still, relatively narrow margins of exposure (<100) were estimated for some perfuoroalkyl substances and phthalates, raising concerns that cumulative exposures may pose a cardiotoxicity risk. Collectively, this study demonstrated the value of using a population-based human in vitro model for rapid, high-throughput hazard and risk characterization of chemicals for which little to no cardiotoxicity data are available from guideline studies in animals.

The Alginate Immobilization of Metabolic Enzymes Platform Retrofits an Estrogen Receptor Transactivation Assay With Metabolic Competence

The U.S. EPA Endocrine Disruptor Screening Program utilizes data across the ToxCast/Tox21 high-throughput screening (HTS) programs to evaluate the biological effects of potential endocrine active substances. A potential limitation to the use of in vitro assay data in regulatory decision-making is the lack of coverage for xenobiotic metabolic processes. Both hepatic- and peripheral-tissue metabolism can yield metabolites that exhibit greater activity than the parent compound (bioactivation) or are inactive (bioinactivation) for a given biological target. Interpretation of biological effect data for both putative endocrine active substances, as well as other chemicals, screened in HTS assays may benefit from the addition of xenobiotic metabolic capabilities to decrease the uncertainty in predicting potential hazards to human health. The objective of this study was to develop an approach to retrofit existing HTS assays with hepatic metabolism. The Alginate Immobilization of Metabolic Enzymes (AIME) platform encapsulates hepatic S9 fractions in alginate microspheres attached to 96-well peg lids. Functional characterization across a panel of reference substrates for phase I cytochrome P450 enzymes revealed substrate depletion with expected metabolite accumulation. Performance of the AIME method in the VM7Luc estrogen receptor transactivation assay was evaluated across 15 reference chemicals and 48 test chemicals that yield metabolites previously identified as estrogen receptor active or inactive. The results demonstrate the utility of applying the AIME method for identification of false-positive and false-negative target assay effects, reprioritization of hazard based on metabolism-dependent bioactivity, and enhanced in vivo concordance with the rodent uterotrophic bioassay. Integration of the AIME metabolism method may prove useful for future biochemical and cell-based HTS applications.

Combined Naïve Bayesian, Chemical Fingerprints and Molecular Docking Classifiers to Model and Predict Androgen Receptor Binding Data for Environmentally- and Health-Sensitive Substances

Many chemicals that enter the environment, food chain, and the human body can disrupt androgen-dependent pathways and mimic hormones and therefore, may be responsible for multiple diseases from reproductive to tumor. Thus, modeling and predicting androgen receptor activity is an important area of research. The aim of the current study was to find a method or combination of methods to predict compounds that can bind to and/or disrupt the androgen receptor, and thereby guide decision making and further analysis. A stepwise procedure proceeded from analysis of protein structures from human, chimp, and rat, followed by docking and subsequent ligand, and statistics based techniques that improved classification gradually. The best methods used multivariate logistic regression of combinations of chimpanzee protein structural docking scores, extended connectivity fingerprints, and naïve Bayesians of known binders and non-binders. Combination or consensus methods included data from a variety of procedures to improve the final model accuracy.

Utility of In Vitro Bioactivity as a Lower Bound Estimate of In Vivo Adverse Effect Levels and in Risk-Based Prioritization

Use of high-throughput, in vitro bioactivity data in setting a point-of-departure (POD) has the potential to accelerate the pace of human health safety evaluation by informing screening-level assessments. The primary objective of this work was to compare PODs based on high-throughput predictions of bioactivity, exposure predictions, and traditional hazard information for 448 chemicals. PODs derived from new approach methodologies (NAMs) were obtained for this comparison using the 50th (PODNAM, 50) and the 95th (PODNAM, 95) percentile credible interval estimates for the steady-state plasma concentration used in in vitro to in vivo extrapolation of administered equivalent doses. Of the 448 substances, 89% had a PODNAM, 95 that was less than the traditional POD (PODtraditional) value. For the 48 substances for which PODtraditional < PODNAM, 95, the PODNAM and PODtraditional were typically within a factor of 10 of each other, and there was an enrichment of chemical structural features associated with organophosphate and carbamate insecticides. When PODtraditional < PODNAM, 95, it did not appear to result from an enrichment of PODtraditional based on a particular study type (eg, developmental, reproductive, and chronic studies). Bioactivity:exposure ratios, useful for identification of substances with potential priority, demonstrated that high-throughput exposure predictions were greater than the PODNAM, 95 for 11 substances. When compared with threshold of toxicological concern (TTC) values, the PODNAM, 95 was greater than the corresponding TTC value 90% of the time. This work demonstrates the feasibility, and continuing challenges, of using in vitro bioactivity as a protective estimate of POD in screening-level assessments via a case study.

Integrating endocrine-related health effects into comparative human toxicity characterization

Endocrine-disrupting chemicals have the ability to interfere with and alter functions of the hormone system, leading to adverse effects on reproduction, growth and development. Despite growing concerns over their now ubiquitous presence in the environment, endocrine-related human health effects remain largely outside of comparative human toxicity characterization frameworks as applied for example in life cycle impact assessments. In this paper, we propose a new methodological framework to consistently integrate endocrine-related health effects into comparative human toxicity characterization. We present two quantitative and operational approaches for extrapolating towards a common point of departure from both in vivo and dosimetry-adjusted in vitro endocrine-related effect data and deriving effect factors as well as corresponding characterization factors for endocrine-active/endocrine-disrupting chemicals. Following the proposed approaches, we calculated effect factors for 323 chemicals, reflecting their endocrine potency, and related characterization factors for 157 chemicals, expressing their relative endocrine-related human toxicity potential. Developed effect and characterization factors are ready for use in the context of chemical prioritization and substitution as well as life cycle impact assessment and other comparative assessment frameworks. Endocrine-related effect factors were found comparable to existing effect factors for cancer and non-cancer effects, indicating that (1) the chemicals' endocrine potency is not necessarily higher or lower than other effect potencies and (2) using dosimetry-adjusted effect data to derive effect factors does not consistently overestimate the effect of potential endocrine disruptors. Calculated characterization factors span over 8-11 orders of magnitude for different substances and emission compartments and are dominated by the range in endocrine potencies.

Comparison of Cellular Morphological Descriptors and Molecular Fingerprints for the Prediction of Cytotoxicity- and Proliferation-Related Assays

Cell morphology features, such as those from the Cell Painting assay, can be generated at relatively low costs and represent versatile biological descriptors of a system and thereby compound response. In this study, we explored cell morphology descriptors and molecular fingerprints, separately and in combination, for the prediction of cytotoxicity- and proliferation-related in vitro assay endpoints. We selected 135 compounds from the MoleculeNet ToxCast benchmark data set which were annotated with Cell Painting readouts, where the relatively small size of the data set is due to the overlap of required annotations. We trained Random Forest classification models using nested cross-validation and Cell Painting descriptors, Morgan and ErG fingerprints, and their combinations. While using leave-one-cluster-out cross-validation (with clusters based on physicochemical descriptors), models using Cell Painting descriptors achieved higher average performance over all assays (Balanced Accuracy of 0.65, Matthews Correlation Coefficient of 0.28, and AUC-ROC of 0.71) compared to models using ErG fingerprints (BA 0.55, MCC 0.09, and AUC-ROC 0.60) and Morgan fingerprints alone (BA 0.54, MCC 0.06, and AUC-ROC 0.56). While using random shuffle splits, the combination of Cell Painting descriptors with ErG and Morgan fingerprints further improved balanced accuracy on average by 8.9% (in 9 out of 12 assays) and 23.4% (in 8 out of 12 assays) compared to using only ErG and Morgan fingerprints, respectively. Regarding feature importance, Cell Painting descriptors related to nuclei texture, granularity of cells, and cytoplasm as well as cell neighbors and radial distributions were identified to be most contributing, which is plausible given the endpoint considered. We conclude that cell morphological descriptors contain complementary information to molecular fingerprints which can be used to improve the performance of predictive cytotoxicity models, in particular in areas of novel structural space.

Exploring the Potential of ToxCast Data in Supporting Read-Across for Evaluation of Food Chemical Safety

The intention of this study was to determine the utility of high-throughput screening (HTS) data, as exemplified by ToxCast and Tox21, for application in toxicological read-across in food-relevant chemicals. Key questions were addressed on the extent to which the HTS data could provide information enabling (1) the elucidation of underlying bioactivities associated with apical toxicological outcomes, (2) the closing of existing toxicological data gaps, and (3) the definition of the boundaries of chemical space across which bioactivity could reliably be extrapolated. Results revealed that many biological targets apparently activated within the chemical groupings lack, at this time, validated toxicity pathway associations. Therefore, as means of providing proof-of-principle, a comparatively well-characterized end point-estrogenicity-was selected for evaluation. This was facilitated through the preparation of two exploratory case studies, focusing upon groupings of paraben-gallates and pyranone-type compounds (notably flavonoids). Within both, the HTS data were seen to reflect estrogenic potencies in a manner which broadly corresponded to established structure-activity group relationships, with parabens and flavonoids displaying greater estrogen receptor affinity than benzoate esters and alternative pyranone-containing molecules, respectively. As such, utility in the identification of out-of-domain compounds was demonstrated, indicating potential for application in addressing point (3) as detailed above.

Selecting a minimal set of androgen receptor assays for screening chemicals

Screening certain environmental chemicals for their ability to interact with endocrine targets, including the androgen receptor (AR), is an important global concern. We previously developed a model using a battery of eleven in vitro AR assays to predict in vivo AR activity. Here we describe a revised mathematical modeling approach that also incorporates data from newly available assays and demonstrate that subsets of assays can provide close to the same level of predictivity. These subset models are evaluated against the full model using 1820 chemicals, as well as in vitro and in vivo reference chemicals from the literature. Agonist batteries of as few as six assays and antagonist batteries of as few as five assays can yield balanced accuracies of 95% or better relative to the full model. Balanced accuracy for predicting reference chemicals is 100%. An approach is outlined for researchers to develop their own subset batteries to accurately detect AR activity using assays that map to the pathway of key molecular and cellular events involved in chemical-mediated AR activation and transcriptional activity. This work indicates in vitro bioactivity and in silico predictions that map to the AR pathway could be used in an integrated approach to testing and assessment for identifying chemicals that interact directly with the mammalian AR.

Cytotoxicity Burst? Differentiating Specific from Nonspecific Effects in Tox21 in Vitro Reporter Gene Assays

BACKGROUND

High-throughput screening of chemicals with in vitro reporter gene assays in Tox21 has produced a large database on cytotoxicity and specific modes of action. However, the validity of some of the reported activities is questionable due to the "cytotoxicity burst," which refers to the supposition that many stress responses are activated in a nonspecific way at concentrations close to cell death.

OBJECTIVES

We propose a pragmatic method to identify whether reporter gene activation is specific or cytotoxicity-triggered by comparing the measured effects with baseline toxicity.

METHODS

Baseline toxicity, also termed narcosis, is the minimal toxicity any chemical causes. Quantitative structure-activity relationships (QSARs) developed for baseline toxicity in mammalian reporter gene cell lines served as anchors to define the chemical-specific threshold for the cytotoxicity burst and to evaluate the degree of specificity of the reporter gene activation. Measured 10% effect concentrations were related to measured or QSAR-predicted 10% cytotoxicity concentrations yielding specificity ratios (SR). We applied this approach to our own experimental data and to ∼ 8,000 chemicals that were tested in six of the high-throughput Tox21 reporter gene assays.

RESULTS

Confirmed baseline toxicants activated reporter gene activity around cytotoxic concentrations triggered by the cytotoxicity burst. In six Tox21 assays, 37%-87% of the active hits were presumably caused by the cytotoxicity burst (SR < 1 ) and only 2%-14% were specific with SR ≥ 10 against experimental cytotoxicity but 75%-97% were specific against baseline toxicity. This difference was caused by a large fraction of chemicals showing excess cytotoxicity.

CONCLUSIONS

The specificity analysis for measured in vitro effects identified whether a cytotoxicity burst had likely occurred. The SR-analysis not only prevented false positives, but it may also serve as measure for relative effect potency and can be used for quantitative in vitro-in vivo extrapolation and risk assessment of chemicals. https://doi.org/10.1289/EHP6664.

Identifying and Characterizing Stress Pathways of Concern for Consumer Safety in Next-Generation Risk Assessment

Many substances for which consumer safety risk assessments need to be conducted are not associated with specific toxicity modes of action, but rather exhibit nonspecific toxicity leading to cell stress. In this work, a cellular stress panel is described, consisting of 36 biomarkers representing mitochondrial toxicity, cell stress, and cell health, measured predominantly using high content imaging. To evaluate the panel, data were generated for 13 substances at exposures consistent with typical use-case scenarios. These included some that have been shown to cause adverse effects in a proportion of exposed humans and have a toxicological mode-of-action associated with cellular stress (eg, doxorubicin, troglitazone, and diclofenac), and some that are not associated with adverse effects due to cellular stress at human-relevant exposures (eg, caffeine, niacinamide, and phenoxyethanol). For each substance, concentration response data were generated for each biomarker at 3 timepoints. A Bayesian model was then developed to quantify the evidence for a biological response, and if present, a credibility range for the estimated point of departure (PoD) was determined. PoDs were compared with the plasma Cmax associated with the typical substance exposures, and indicated a clear differentiation between "low" risk and "high" risk chemical exposure scenarios. Developing robust methods to characterize the in vitro bioactivity of xenobiotics is an important part of non-animal safety assessment. The results presented in this work show that the cellular stress panel can be used, together with other new approach methodologies, to identify chemical exposures that are protective of consumer health.

Potential of ToxCast Data in the Safety Assessment of Food Chemicals

Tox21 and ToxCast are high-throughput in vitro screening programs coordinated by the U.S. National Toxicology Program and the U.S. Environmental Protection Agency, respectively, with the goal of forecasting biological effects in vivo based on bioactivity profiling. The present study investigated whether mechanistic insights in the biological targets of food-relevant chemicals can be obtained from ToxCast results when the chemicals are grouped according to structural similarity. Starting from the 556 direct additives that have been identified in the ToxCast database by Karmaus et al. [Karmaus, A. L., Trautman, T. D., Krishan, M., Filer, D. L., and Fix, L. A. (2017). Curation of food-relevant chemicals in ToxCast. Food Chem. Toxicol. 103, 174-182.], the results showed that, despite the limited number of assays in which the chemical groups have been tested, sufficient results are available within so-called "DNA binding" and "nuclear receptor" target families to profile the biological activities of the defined chemical groups for these targets. The most obvious activity identified was the estrogen receptor-mediated actions of the chemical group containing parabens and structurally related gallates, as well the chemical group containing genistein and daidzein (the latter 2 being particularly active toward estrogen receptor β as a potential health benefit). These group effects, as well as the biological activities of other chemical groups, were evaluated in a series of case studies. Overall, the results of the present study suggest that high-throughput screening data could add to the evidence considered for regulatory risk assessment of food chemicals and to the evaluation of desirable effects of nutrients and phytonutrients. The data will be particularly useful for providing mechanistic information and to fill data gaps with read-across.

Probabilistic human health risk assessment of perfluorooctane sulfonate (PFOS) by integrating in vitro, in vivo toxicity, and human epidemiological studies using a Bayesian-based dose-response assessment coupled with physiologically based pharmacokinetic (PBPK) modeling approach

BACKGROUND

Environmental exposure to perfluorooctane sulfonate (PFOS) is associated with various adverse outcomes in humans. However, risk assessment for PFOS with the traditional risk estimation method is faced with multiple challenges because there are high variabilities and uncertainties in its toxicokinetics and toxicity between species and among different types of studies.

OBJECTIVES

This study aimed to develop a robust probabilistic risk assessment framework accounting for interspecies and inter-experiment variabilities and uncertainties to derive the human equivalent dose (HED) and reference dose for PFOS.

METHODS

A Bayesian dose-response model was developed to analyze selected 34 critical studies, including human epidemiological, animal in vivo, and ToxCast in vitro toxicity datasets. The dose-response results were incorporated into a multi-species physiologically based pharmacokinetic (PBPK) model to reduce the toxicokinetic/toxicodynamic variabilities. In addition, a population-based probabilistic risk assessment of PFOS was performed for Asian, Australian, European, and North American populations, respectively, based on reported environmental exposure levels.

RESULTS

The 5th percentile of HEDs derived from selected studies was estimated to be 21.5 (95% CI: 10.6-36.3) ng/kg/day. After exposure to environmental levels of PFOS, around 50% of the population in all studied populations would likely have >20% of increase in serum cholesterol, but the effects on other endpoints were estimated to be minimal (<10% changes). There was a small population (~10% of the population) that was highly sensitive to endocrine disruption and cellular response by environmental PFOS exposure.

CONCLUSION

Our results provide insights into a complete risk characterization of PFOS and may help regulatory agencies in the reevaluation of PFOS risk. Our new probabilistic approach can conduct dose-response analysis of different types of toxicity studies simultaneously and this method could be used to improve risk assessment for other perfluoroalkyl substances (PFAS).

Assessing Toxicokinetic Uncertainty and Variability in Risk Prioritization

High(er) throughput toxicokinetics (HTTK) encompasses in vitro measures of key determinants of chemical toxicokinetics and reverse dosimetry approaches for in vitro-in vivo extrapolation (IVIVE). With HTTK, the bioactivity identified by any in vitro assay can be converted to human equivalent doses and compared with chemical intake estimates. Biological variability in HTTK has been previously considered, but the relative impact of measurement uncertainty has not. Bayesian methods were developed to provide chemical-specific uncertainty estimates for 2 in vitro toxicokinetic parameters: unbound fraction in plasma (fup) and intrinsic hepatic clearance (Clint). New experimental measurements of fup and Clint are reported for 418 and 467 chemicals, respectively. These data raise the HTTK chemical coverage of the ToxCast Phase I and II libraries to 57%. Although the standard protocol for Clint was followed, a revised protocol for fup measured unbound chemical at 10%, 30%, and 100% of physiologic plasma protein concentrations, allowing estimation of protein binding affinity. This protocol reduced the occurrence of chemicals with fup too low to measure from 44% to 9.1%. Uncertainty in fup was also reduced, with the median coefficient of variation dropping from 0.4 to 0.1. Monte Carlo simulation was used to propagate both measurement uncertainty and biological variability into IVIVE. The uncertainty propagation techniques used here also allow incorporation of other sources of uncertainty such as in silico predictors of HTTK parameters. These methods have the potential to inform risk-based prioritization based on the relationship between in vitro bioactivities and exposures.

The Next Generation Blueprint of Computational Toxicology at the U.S. Environmental Protection Agency

The U.S. Environmental Protection Agency (EPA) is faced with the challenge of efficiently and credibly evaluating chemical safety often with limited or no available toxicity data. The expanding number of chemicals found in commerce and the environment, coupled with time and resource requirements for traditional toxicity testing and exposure characterization, continue to underscore the need for new approaches. In 2005, EPA charted a new course to address this challenge by embracing computational toxicology (CompTox) and investing in the technologies and capabilities to push the field forward. The return on this investment has been demonstrated through results and applications across a range of human and environmental health problems, as well as initial application to regulatory decision-making within programs such as the EPA's Endocrine Disruptor Screening Program. The CompTox initiative at EPA is more than a decade old. This manuscript presents a blueprint to guide the strategic and operational direction over the next 5 years. The primary goal is to obtain broader acceptance of the CompTox approaches for application to higher tier regulatory decisions, such as chemical assessments. To achieve this goal, the blueprint expands and refines the use of high-throughput and computational modeling approaches to transform the components in chemical risk assessment, while systematically addressing key challenges that have hindered progress. In addition, the blueprint outlines additional investments in cross-cutting efforts to characterize uncertainty and variability, develop software and information technology tools, provide outreach and training, and establish scientific confidence for application to different public health and environmental regulatory decisions.

Development, validation and integration of in silico models to identify androgen active chemicals

Humans are exposed to large numbers of environmental chemicals, some of which potentially interfere with the endocrine system. The identification of potential endocrine disrupting chemicals (EDCs) has gained increasing priority in the assessment of environmental hazards. The U.S. Environmental Protection Agency (U.S. EPA) has developed the Endocrine Disruptor Screening Program (EDSP) which aims to prioritize and screen potential EDCs. The Toxicity Forecaster (ToxCast) program has generated data using in vitro high-throughput screening (HTS) assays measuring activity of chemicals at multiple points along the androgen receptor (AR) activity pathway. In the present study, using a large and diverse data set of 1667 chemicals provided by the U.S. EPA from the combined ToxCast AR assays in the framework of the Collaborative Modeling Project for Androgen Receptor Activity (CoMPARA). Two models were built using ADMET Predictor™; one is based on Artificial Neural Networks (ANNs) technology and the other uses a Support Vector Machine (SVM) algorithm; one model is a Decision Tree (DT) developed in R; and two models make use of differently combined sets of structural alerts (SAs) automatically extracted by SARpy. We used two strategies to integrate predictions from single models; one is based on a majority vote approach and the other on prediction convergence. These strategies led to enhanced statistical performance in most cases. Moreover, the majority vote approach improved prediction coverage when one or more single models were not able to provide any estimations. This study integrates multiple in silico approaches as a virtual screening tool for use in risk assessment of endocrine disrupting chemicals.

A case study on the application of an expert-driven read-across approach in support of quantitative risk assessment of p,p'-dichlorodiphenyldichloroethane

Deriving human health risk estimates for environmental chemicals has traditionally relied on in vivo toxicity databases to characterize potential adverse health effects and associated dose-response relationships. In the absence of in vivo toxicity information, new approach methods (NAMs) such as read-across have the potential to fill the required data gaps. This case study applied an expert-driven read-across approach to identify and evaluate analogues to fill non-cancer oral toxicity data gaps for p,p'-dichlorodiphenyldichloroethane (p,p'-DDD), an organochlorine contaminant known to occur at contaminated sites in the U.S. The source analogue p,p'-dichlorodiphenyltrichloroethane (DDT) and its no-observed-adverse-effect level of 0.05 mg/kg-day were proposed for the derivation of screening-level health reference values for the target chemical, p,p'-DDD. Among the primary similarity contexts (structure, toxicokinetics, and toxicodynamics), toxicokinetic considerations were instrumental in separating p,p'-DDT as the best source analogue from other potential candidates (p,p'-DDE and methoxychlor). In vitro high-throughput screening (HTS) assays from ToxCast were used to evaluate similarity in bioactivity profiles and make inferences toward plausible mechanisms of toxicity to build confidence in the read-across approach. This work demonstrated the value of NAMs such as read-across and in vitro HTS in human health risk assessment of environmental contaminants with the potential to inform regulatory decision-making.

High-throughput screening and environmental risk assessment: State of the science and emerging applications

In 2007 the United States National Research Council (NRC) published a vision for toxicity testing in the 21st century that emphasized the use of in vitro high-throughput screening (HTS) methods and predictive models as an alternative to in vivo animal testing. In the present study we examine the state of the science of HTS and the progress that has been made in implementing and expanding on the NRC vision, as well as challenges to implementation that remain. Overall, significant progress has been made with regard to the availability of HTS data, aggregation of chemical property and toxicity information into online databases, and the development of various models and frameworks to support extrapolation of HTS data. However, HTS data and associated predictive models have not yet been widely applied in risk assessment. Major barriers include the disconnect between the endpoints measured in HTS assays and the assessment endpoints considered in risk assessments as well as the rapid pace at which new tools and models are evolving in contrast with the slow pace at which regulatory structures change. Nonetheless, there are opportunities for environmental scientists and policymakers alike to take an impactful role in the ongoing development and implementation of the NRC vision. Six specific areas for scientific coordination and/or policy engagement are identified. Environ Toxicol Chem 2019;38:12-26. Published 2018 Wiley Periodicals Inc. on behalf of SETAC. This article is a US government work and, as such, is in the public domain in the United States of America.