Table of Contents

1. Definition

Tox21/ToxCast denotes a coordinated pair of United States federal programs that perform large-scale, high-throughput in vitro toxicology screening. Tox21 (Toxicology Testing in the 21st Century) is a research collaboration among four federal partners — the Environmental Protection Agency (EPA), the National Toxicology Program (NTP) of the National Institute of Environmental Health Sciences (NIEHS), the National Center for Advancing Translational Sciences (NCATS) of the National Institutes of Health, and the Food and Drug Administration (FDA) — initiated in 2008¹. ToxCast (Toxicity Forecaster) is the companion EPA program, launched in 2007, that pioneered the application of in vitro bioassays and computational toxicology to the prioritization of environmental chemicals².

The biological basis of both programs is the toxicity pathway concept: adverse outcomes in intact organisms are presumed to originate in the perturbation of conserved, quantifiable cell-signaling circuits. The assay panels therefore model molecular and cellular initiating events — nuclear receptor signaling, oxidative and other stress-response pathways (for example, p53, NF-κB, and ARE/Nrf2), mitochondrial function, DNA damage, and general cytotoxicity — rather than the apical endpoints of classical animal bioassays³. Structurally, screening is performed by quantitative high-throughput screening (qHTS), in which every chemical is tested as a titration series of concentrations, typically in 1536-well microplates, so that a full concentration-response curve is recorded and fitted to yield potency metrics such as the half-maximal activity concentration (AC50)⁴. Assay readouts rely principally on reporter-gene constructs (for example, pathway-responsive luciferase reporters), receptor-binding and enzyme measurements, and multiplexed cell viability measurements in human-derived cell systems^{3 4}.

2. Purpose

The primary value proposition is scalable, human-relevant hazard identification. Classical guideline animal testing is slow, resource-intensive, and of limited throughput, leaving tens of thousands of commercial chemicals with little or no toxicity data. The National Research Council’s 2007 report, Toxicity Testing in the 21st Century, called for a transition from observational apical animal studies toward mechanism-based, pathway-oriented in vitro testing anchored in human biology⁵. Tox21 and ToxCast implement that vision by screening thousands of chemicals at a fraction of the cost and duration of animal studies, producing concentration-resolved data that rank chemicals for targeted testing and mechanistic follow-up^{1 6}.

The programs improve upon legacy models in three principal respects. First, single-concentration, single-endpoint screening is replaced by titration-based concentration-response profiling, which distinguishes genuine biological activity from noise and ranks chemicals by potency⁴. Second, the assay repertoire is built on human molecular targets — human nuclear receptors, human cell lines, and human primary cells — so results are directly relevant to human pathway biology rather than extrapolated across species⁶. Third, in vitro potencies are routinely translated into administered-dose equivalents through in vitro-to-in vivo extrapolation (IVIVE), allowing comparison against realistic human exposure estimates and thereby supporting risk-based rather than purely hazard-based prioritization⁷.

3. Methodology

Quantitative high-throughput screening. The core technique is qHTS, in which each compound is assayed as a multi-point dilution series spanning several orders of magnitude, generating a pharmacological profile from which curve class, efficacy, and AC50 are computed⁴.

Screening platform and protocols. Tox21 screening is executed on a fully automated robotic platform operating 1536-well plates. Compounds are transferred by a 23-nanoliter pintool or by acoustic, non-contact dispensing, and each assay is qualified before production screening against acceptance criteria including a signal-to-background ratio greater than three, a coefficient of variation below ten percent, and a Z-prime factor above 0.5³. The platform can screen the entire approximately 10,000-compound (10K) library in triplicate in approximately one week³. Tox21 Phase II testing covers roughly 10,000 chemicals in triplicate at fifteen concentrations, yielding more than fifty million concentration-response data points⁸.

Chemical libraries and substrates. The Tox21 10K library was assembled through multi-agency curation of commercial chemicals, drugs, food additives, pesticides, and industrial and consumer-product chemicals, with standardized structural annotation to support read-across and predictive modeling⁹. The ToxCast Phase I library comprised 309 chemicals, mostly pesticide active ingredients, screened across 467 in vitro assays spanning cell-free biochemical platforms and cell-based platforms in human primary cells, human cell lines, and rat primary hepatocytes⁶. Phase II profiling covered 976 chemicals across 331 enzymatic and receptor-signaling assays¹⁰, and the inventory subsequently grew to approximately 1,800 chemicals in Phase II and more than 4,500 after Phase III was initiated in late 2014¹¹.

Data processing and tools. All concentration-response data are processed through the open-source tcpl (ToxCast pipeline) R package and its associated MySQL database, which perform normalization, quality control, curve fitting, and hit calling¹². Chemical identity, structures, and assay annotations are integrated through the public CompTox Chemistry Dashboard, which links ToxCast and Tox21 bioactivity to physicochemical and exposure data¹³.

4. Performance

Empirical performance.

• Throughput: the robotic platform profiles the full 10K library in triplicate in approximately one week³.
• Data scale: more than fifty million qHTS data points across approximately 10,000 chemicals and dozens of pathway assays⁸.
• Assay reproducibility: enforced through pre-qualification criteria (signal-to-background, coefficient of variation, Z-prime factor) and through triplicate, multi-concentration screening supported by positive and negative controls³.
• Predictive performance: cluster-based modeling of the Tox21 10K profiles produced predictive models for 72 in vivo toxicity endpoints and supported mechanism characterization⁸.
• Pathway-level performance: an integrated network model of 18 in vitro assays classified estrogen receptor pathway activity with strong concordance against curated reference data and supported prioritization decisions¹⁴.

Technical boundaries and limitations.

• Cytotoxicity burst: at concentrations approaching cell death, apparent activity arises simultaneously across many unrelated assays; such hits must be triaged as nonspecific rather than pathway-specific¹⁵.
• Dosimetry: potency values alone can mislead; IVIVE incorporating in vitro pharmacokinetic measurements is required to express bioactivity as administered-dose equivalents comparable to human exposure⁷.
• Biological coverage: many reporter assays lack full metabolic competence and capture short-term, largely transcriptional responses; apical and chronic outcomes therefore require integration with other new approach methodologies and targeted in vivo studies¹⁶.
• Assay coverage per chemical is uneven, so the absence of activity in ToxCast does not establish the absence of hazard¹⁷.

5. Applications

Research applications. The programs generate pathway-level bioactivity maps used for mechanism characterization, read-across, and the training of machine-learning toxicity models^{8 13}. The Tox21 Data Challenge 2014 crowdsourced predictive models for twelve nuclear-receptor and stress-response assays, demonstrating the modeling value of the public data¹⁸.

Regulatory and industrial applications. The data are used principally for chemical prioritization, ranking substances for targeted in vivo testing^{1 6}; for endocrine screening within the EPA Endocrine Disruptor Screening Program^{19 20}; for risk-based screening through IVIVE and exposure integration^{7 16}; and for prioritization of existing substances under Canada’s Bioactivity Exposure Ratio approach¹⁷.

Standardized contexts of use. EPA has proposed validated high-throughput assays and computational pathway models as alternatives within its Endocrine Disruptor Screening Program¹⁹, and the estrogen receptor model has been applied to more than 1,800 chemicals^{20 14}. Health Canada published a science approach document in March 2021 formalizing the use of ToxCast and Tox21 bioactivity, combined with exposure estimates, for priority setting under the Canadian Environmental Protection Act¹⁷.

Deployment stage and readiness. The programs are operationally mature for screening and prioritization and are embedded in the EPA’s computational toxicology blueprint as core new approach methodologies; they are not yet accepted as standalone replacements for definitive apical hazard characterization¹⁶.

6. Variations

• Tox21 Phase I — The proof-of-concept phase of the interagency program, which screened approximately 2,800 compounds in more than 75 biochemical and cell-based assays covering cytotoxicity, DNA damage, and signaling pathway readouts³.
• Tox21 Phase II — The scale-up phase, which expanded testing to the nearly 10,000-compound 10K library on the NCATS robotic platform, screened in triplicate at fifteen concentrations^{3 8 9}.
• ToxCast Phase I — The initial EPA phase, which screened 309 mostly pesticide chemicals across 467 in vitro assays on multiple cell-free and cell-based platforms⁶.
• ToxCast Phase II — The expanded EPA phase, which grew the chemical inventory to approximately 1,800 compounds, including an endocrine-focused subset, profiled across 331 enzymatic and receptor-signaling assays^{10 11}.
• Tox21 Data Challenge — The 2014 crowdsourced modeling competition built on twelve Tox21 qHTS assay datasets for nuclear-receptor and stress-response pathway prediction¹⁸.