The significant failure rate of compounds entering clinical trials, due in large part to safety and efficacy concerns, has led the pharmaceutical industry to invest heavily in new technologies and to integrate safety assessment into the discovery and development pipeline as early as possible.
Computational chemistry and gene-expression analysis have been widely adopted for safety assessment in recent years, and as the fields of proteomics and metabolomics mature, they too are being investigated with a view to combining the different technologies to give a more holistic understanding of the effects of compounds on biological systems.
The integration and analysis of this data presents a significant bioinformatics challenge. To address this need, GeneGo created MetaDrug™, a systems pharmacology and toxicology data-analysis platform.
MetaDrug incorporates chemical structural analysis, metabolite prediction, and ADMET quantitative structure activity relationship (QSAR) models with a database of molecular interactions, gene function, disease and toxicity ontologies and networks, as well as pathway- and network-analysis software. The user can upload, combine, analyze, and visualize multiple data types (genomic, proteomic, and metabolomic) individually or in concert to reveal on- and off-target pharmacological effects, mechanisms of toxicity, and biomarkers of safety and efficacy.
MetaDrug includes a comprehensive chemical and pharmacological database that comprises chemical structures of over 680,000 drugs, metabolites, and xenobiotics along with data on their biological effects from over 1,000,000 in vitro pharmacological-binding assays, 1,500,000 in vitro functional assays, and 600,000 in vivo functional assays.
Additional data covers a wide range of ADMET and physicochemical properties. This resource, coupled with chemical structure similarity and substructure searching; QSAR modeling, including the ability for the user to derive and incorporate custom QSAR models from their own data; and an extensive knowledge-base of drugs and their targets, facilitates comparison of discovery and development compounds to competitors on the market and in clinical trials.
MetaDrug further couples this pharmacological data to GeneGo’s network- and pathway-analysis tools. This provides the ability to query potential biological and toxicological concerns based on chemical structure, which enables triage and prioritization of compounds early in the discovery process.
As compounds progress and empirical data is generated, the data can be visualized on networks and pathways identified through structural analysis. Data-driven analysis can be performed independently or in concert with the structural data, the predicted activities investigated further, and conclusions refined.
Identifying Compound Liabilities
To demonstrate the power of this approach, propiconazole, an agricultural fungicide (Figure 1), was subjected to analysis in MetaDrug using the chemical structure of the compound as an entry point.
Propiconazole is widely used on fruit, vegetables, cereals, and seeds, and in construction as a wood preservative. The U.S. EPA classifies propiconazole as a developmental toxicant and a possible carcinogen. Propiconazole is hepatotoxic in rats in chronic studies, causes liver tumors in mice, and is a tumor promoter in rats. Propiconazole is also a reproductive toxicant in male rats.
MetaDrug identified several possible metabolites, the most likely of which were predicted to occur via aliphatic hydroxylation. The major site of enzymatic attack in rats given a single, oral dose of propiconazole was shown to be oxidation of the propyl side chain, though numerous other metabolites were identified, which confirmed both the diversity and the relative ranking of metabolites predicted by MetaDrug.
MetaDrug incorporates a variety of quantitative structure-activity models for evaluating ADMET properties. Using these models, propiconazole was predicted to be a substrate for and inhibitor of CYP2D6 and CYP3A4, a substrate of human organic anion transporter polypeptide C (OATP-C), as well as an inhibitor of human P-glycoprotein transporter. Pregnane-X receptor (PXR) agonist activity was also predicted (Table).
Using a 2-D Tanimoto similarity search, a total of 133 unique compounds with >50% similarity to propiconazole or one of its metabolites were identified. The pharmacological database identified 43 potential targets for these compounds—of which 16 were cytochromes P450, including CYP3A4 and CYP2D6, previously identified independently as possible targets by QSAR models.
PXR was also identified as a potential target in this search, again supporting the results of the QSAR prediction. Other potential targets included solute carrier organic anion transporter family member 1A2 (OATP-A), multidrug resistance protein 1 (MDR1), and solute carrier family 22 (organic anion transporter) member 8 (SLC22A8), suggesting that propiconazole has the potential to strongly impact drug and xenobiotic metabolism and clearance.
Enrichment analysis of the target list across three different ontological categories in MetaDrug; Gene Ontology (GO) biological processes, GeneGo canonical pathway maps and GeneGo toxicity networks; confirmed a strong impact on drug metabolism, with xenobiotic metabolic process and response to xenobiotic stimulus being the two most highly impacted GO processes and Metabolism_CYPs and Fanconi anemia-group proteins and Metabolism_Xenobiotic metabolism the two highest impacted GeneGo toxicity networks.
The most strongly impacted GeneGo canonical pathway maps were related to the metabolism of endogenous estrogenic and androgenic steroid hormones. The map Estradiol metabolism had the highest enrichment (Figure 2).
The disruption of steroid hormone homeostasis suggested by the enrichment in these pathways is a plausible mechanism for the developmental defects associated with exposure to the fungicide. Indeed, disruption of steroid hormone homeostasis as a result of altered metabolism following activation of nuclear receptors such as PXR has recently been hypothesized in the scientific literature in the scientific literature as a likely mechanism of the observed effects.
Further analysis of similar compounds and their targets, using the network building and analysis capabilities of MetaDrug, identified a highly enriched network (p=1.59x10-50) featuring PXR as a major hub (Figure 3).
Finally, to investigate the accuracy of the predicted biological effects, liver gene expression data from rats given the maximum-tolerated, oral dose of propiconazole for 1, 3, and 5 days was uploaded to MetaDrug and displayed on the pathways and networks.
A significant upregulation of CYP3A4 and 3A5, as well as downregulation of sulfotransferases and catechol-O-methyltransferase (Figure 2), is evident on the map for Estradiol metabolism. Agonism of PXR, with CYP3A4 and CYP3A5 significantly perturbed at all days of exposure, is also evident (Figure 3). PXR activation in rodents is known to lead to high levels of expression of DMEs, which leads to proliferation of the endoplasmic reticulum, hepatocellular hypertrophy, and hepatomegaly. All of these effects have been observed in the livers of rats treated with propiconazole. Persistent overexpression of cytochromes P450 may lead to oxidative stress and hepatocellular damage.
By using the in silico metabolite prediction, QSAR, and structural similarity searching tools in MetaDrug, alongside a database of molecular interactions, compounds, gene function, disease and toxicity ontologies and networks, and pathway- and network-analysis software, we were able to infer modes of toxic action and identify a major mechanism of toxicity of propiconazole.
Further development of MetaDrug with species-specific content, annotation of compound and gene associations with toxicity, and novel tools for mechanistic and predictive toxicity assessment is under way under the direction of MetaTox, a joint industry-FDA consortium.