Big-data and machine learning to revamp computational toxicology and its use in risk assessment.

The creation of large toxicological databases and advances in machine-learning techniques have empowered computational approaches in toxicology. Work with these large databases based on regulatory data has allowed reproducibility assessment of animal models, which highlight weaknesses in traditional in vivo methods. This should lower the bars for the introduction of new approaches and represents a benchmark that is achievable for any alternative method validated against these methods. Quantitative Structure Activity Relationships (QSAR) models for skin sensitization, eye irritation, and other human health hazards based on these big databases, however, also have made apparent some of the challenges facing computational modeling, including validation challenges, model interpretation issues, and model selection issues. A first implementation of machine learning-based predictions termed REACHacross achieved unprecedented sensitivities of >80% with specificities >70% in predicting the six most common acute and topical hazards covering about two thirds of the chemical universe. While this is awaiting formal validation, it demonstrates the new quality introduced by big data and modern data-mining technologies. The rapid increase in the diversity and number of computational models, as well as the data they are based on, create challenges and opportunities for the use of computational methods.

Machine Learning of Toxicological Big Data Enables Read-Across Structure Activity Relationships (RASAR) Outperforming Animal Test Reproducibility.

Earlier we created a chemical hazard database via natural language processing of dossiers submitted to the European Chemical Agency with approximately 10 000 chemicals. We identified repeat OECD guideline tests to establish reproducibility of acute oral and dermal toxicity, eye and skin irritation, mutagenicity and skin sensitization. Based on 350-700+ chemicals each, the probability that an OECD guideline animal test would output the same result in a repeat test was 78%-96% (sensitivity 50%-87%). An expanded database with more than 866 000 chemical properties/hazards was used as training data and to model health hazards and chemical properties. The constructed models automate and extend the read-across method of chemical classification. The novel models called RASARs (read-across structure activity relationship) use binary fingerprints and Jaccard distance to define chemical similarity. A large chemical similarity adjacency matrix is constructed from this similarity metric and is used to derive feature vectors for supervised learning. We show results on 9 health hazards from 2 kinds of RASARs-"Simple" and "Data Fusion". The "Simple" RASAR seeks to duplicate the traditional read-across method, predicting hazard from chemical analogs with known hazard data. The "Data Fusion" RASAR extends this concept by creating large feature vectors from all available property data rather than only the modeled hazard. Simple RASAR models tested in cross-validation achieve 70%-80% balanced accuracies with constraints on tested compounds. Cross validation of data fusion RASARs show balanced accuracies in the 80%-95% range across 9 health hazards with no constraints on tested compounds.

The human AHR: identification of single nucleotide polymorphisms from six ethnic populations.

BACKGROUND: The effects of 2,3,7,8-tetrachlorodibenzo-p-dioxin and related dioxin-like chemicals are mediated through binding-dependent activation of the cytosolic aryl hydrocarbon receptor (AHR). The human AHR is a low-affinity receptor relative to most rodents, but some reports suggest that there may be individuals with polymorphic high-affinity receptors, thereby possibly increasing the sensitivity to dioxins in such people. METHODS: Although no polymorphisms have been reported in the ligand binding region of the AHR in the over 100 reported sequences, we sequenced 108 additional human AHR genes in an effort to further identify single single nucleotide polymorphisms (SNPs) within the open reading frames of the AHR locus. The DNA was sequenced from six ethnic populations that included Japanese, Chinese, European/Caucasian, African-American, South East Asian, and Hispanic. RESULTS: Six exonic SNPs were identified; four had been described as previously reported and two seem to be novel. Four of the SNPs identified lead to amino acid changes in the AHR protein and two of the SNPs lead to synonymous substitutions. An additional four SNPs have been reported elsewhere that were not identified in the current analysis. With these new sequences, more than 200 human AHR gene sequences have been analyzed for SNPs. CONCLUSION: The results indicate a very limited presence of polymorphisms in the core ligand binding region of the human AHR. Other regions, such as the transactivation domain, seem to be slightly more polymorphic in the human population and the impact on functionality should be further examined.

Mortality rates among workers exposed to dioxins in the manufacture of pentachlorophenol.

OBJECTIVE: We sought to determine if workers exposed to dioxins in pentachlorophenol (PCP) manufacturing were at increased risk of death from specific causes. METHODS: We examined death rates among 773 workers exposed to chlorinated dioxins during PCP manufacturing from 1937 to 1980 using serum dioxin evaluations to estimate exposures to five dioxins. RESULTS: Deaths from all causes combined, all cancers combined, lung cancer, diabetes, and ischemic heart disease were near expected levels. There were eight deaths from non-Hodgkin lymphoma (standardized mortality ratios = 2.4, 95% CI = 1.0 to 4.8). We observed no trend of increasing risk for any cause of death with increasing dioxin exposure. However, the highest rates of non-Hodgkin lymphoma were found in the highest exposure group (standardized mortality ratios = 4.5, 95% CI = 1.2 to 11.5). CONCLUSIONS: Other than possibly an increased risk of non-Hodgkin lymphoma, we find no other cause of death related to the mixture of the dioxin contaminants found in PCP.

The aryl hydrocarbon receptor as a target for estrogen receptor-negative breast cancer chemotherapy.

2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) and the relatively non-toxic selective aryl hydrocarbon receptor (AhR) modulator 6-methyl-1,3,8-trichlorodibenzo-furan (MCDF) induced CYP1A1-dependent ethoxyresorufin O-deethylase activity and inhibited proliferation of seven estrogen receptor (ER) negative breast cancer cell lines. MCDF, TCDD and structurally related 2,3,7,8-tetrachlorodibenzofuran, 1,2,3,7,8-pentachlorodibenzo-p-dioxin, 2,3,4,7,8-pentachlorodibenzofuran, and 3,3',4,4',5-pentachlorobiphenyl induced CYP1A1 and inhibited proliferation of BT-474 and MDA-MB-468 cells. In BT474 and MDA-MB-468 cells transfected with a small inhibitory RNA for the AhR, the antiproliferative activity of the chlorinated aromatic compounds was reversed, whereas for MCDF, only partial reversal was observed, suggesting that this compound acts through both AhR-dependent and AhR-independent pathways in these two cell lines. MCDF also inhibited tumor growth in athymic nude mice in which MDA-MB-468 cells were injected directly into the mammary fat pad. These results suggest that the AhR is a potential drug target for treatment of ER-negative breast cancer.

Toxicokinetics of 2,3,7,8-TCDF and 2,3,4,7,8-PeCDF in mink (Mustela vison) at ecologically relevant exposures.

Wild mink (Mustela vison) living along the Tittabawassee River in central Michigan exhibit elevated hepatic and dietary polychlorinated dibenzofuran (PCDF) concentrations exceeding mink-specific, literature-reported toxicity reference values (TRVs) on a toxicity equivalents basis. However, no apparent effects on individuals or population are evident, suggesting that available TRVs may overpredict risk for the site-specific mix of congeners. To investigate this discrepancy, a 180-day spiked feed study was conducted to assess: (1) the dosages of key congeners necessary to achieve liver concentrations bracketing those observed in wild mink, (2) time to achieve steady-state concentrations, and (3) effect of coadministration of 2,3,7,8-tetrachlorodibenzofuran (TCDF) and 2,3,4,7,8-pentachlorodibenzofuran (4-PeCDF) on the toxicokinetics and distribution of each congener. Adipose and hepatic PCDF concentrations were measured at 0, 90, and 180 days. PCDF concentrations in mink scat were determined at several time points and indicated nearly complete absorption of both TCDF and 4-PeCDF from the diet. Elimination half-times of TCDF were < 15 h and were inversely proportional to dose, while those for 4-PeCDF were approximately 7-9 days with no clear dose dependency in the tested dose range. Coadministration of 4-PeCDF and TCDF accelerated clearance of TCDF compared to administration of TCDF alone. Clearance of 4-PeCDF was not affected by TCDF coadministration. Distribution of 4-PeCDF, but not TCDF, demonstrated increased hepatic sequestration with increasing dose. 4-PeCDF toxicokinetics were described using a previously published two-compartment model. Overall, the toxicokinetic information gathered here illustrates the impact of CYP1A1 induction on bioaccumulation and toxicity potential of TCDF and 4-PeCDF. This information may provide insight into why the current TRVs do not appear to correctly characterize the risk for these two congeners when they are the primary components of an environmental mixture.

Ligand-dependent interactions of the Ah receptor with coactivators in a mammalian two-hybrid assay.

2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) is a high affinity ligand for the aryl hydrocarbon receptor (AhR). In this study, we investigated structure-dependent differences in activation of the AhR by a series of halogenated aromatic hydrocarbons. TCDD, 1,2,3,7,8-pentachlorodibenzo-p-dioxin (PeCDD), 2,3,7,8-tetrachlorodibenzofuran (TCDF), 2,3,4,7,8-pentachlorodibenzofuran (PeCDF), and 3,3',4,4',5-pentachlorobiphenyl (PCB126) induced CYP1A1-dependent activities in HEK293 human embryonic kidney, Panc1 pancreatic cancer, and Hepa1c1c7 mouse hepatoma cell lines. There was a structure-dependent difference in the efficacy of TCDF and PCB126 in HEK293 and Panc1 cells since induced CYP1A1 mRNA levels were lower than observed for the other congeners. A mammalian two-hybrid assay in cells transfected with GAL4-coactivator and AhR-VP16 chimeras was used to investigate structure-dependent interactions of these chimeras in Panc1, HEK293, and Hepa1c1c7 cells. The reporter construct pGAL4-luc contains five tandem GAL4 response elements linked to the luciferase gene and the GAL4-coactivator chimeras express several coactivators including steroid receptor coactivator 1 (SRC-1), SRC-2 and SRC-3, the mediator coactivator TRAP220, coactivator associated arginine methyl transferase 1 (CARM-1), and peroxisome proliferator-activated receptor gamma coactivator 1 (PGC-1). Results of the mammalian two-hybrid studies clearly demonstrate that activation of pGAL4-luc in cells transfected with VP-AhR and GAL4-coactivator chimeras is dependent on the structure of the HAH congener, cell context, and coactivator, suggesting that the prototypical HAH congeners used in this study exhibit selective AhR modulator activity.