Predictive models of prenatal developmental toxicity from ToxCast high-throughput screening data

Area of Science:

• Environmental toxicology
• Computational toxicology
• Chemical safety assessment

Background:

• The Environmental Protection Agency's ToxCast project analyzes chemical in vitro bioactivity.
• Predicting in vivo toxicity from in vitro data is crucial for chemical safety.
• Existing databases like ToxRefDB contain valuable in vivo developmental toxicity data from animal studies.

Purpose of the Study:

• To determine if in vitro high-throughput screening (HTS) data correlates with in vivo developmental toxicity.
• To develop predictive models for identifying chemicals with developmental toxicity potential.
• To uncover mechanistic relationships between in vitro assays and in vivo outcomes.

Main Methods:

• Statistical associations were built between HTS data and ToxRefDB developmental toxicity data.
• Univariate analysis filtered HTS assays based on correlation with specific in vivo endpoints.
• Linear discriminant analysis with cross-validation was used to construct predictive models.

Main Results:

• 423 significant associations were found between HTS assays and developmental toxicity endpoints.
• Species-specific patterns emerged: rat models showed correlations with TGF-beta, RAR, and GPCR signaling.
• Rabbit models correlated with inflammatory signals like interleukins (IL1a, IL8) and chemokines (CCL2).
• Models achieved high balanced accuracy (>70%) in predicting developmental toxicity.
• Common Gene Ontology processes linked species-specific toxicity endpoints, e.g., cleft palate to urogenital defects via embryonic development.

Conclusions:

• HTS assays are valuable tools for developing pathway-level models predictive of developmental toxicity.
• This approach can enhance the identification of chemicals posing developmental risks.
• Mechanistic insights were gained into species-specific responses to chemical exposures.