- ABSTRACT –
Discovery of multi-omics biomarkers for toxicity using meta-analysis
Drug toxicity is a leading cause for drug-candidate attrition at all stages in the drug development pipeline. Therefore, the large number of animals, long time, and tremendous compound synthesis are required for drug toxicity assessment. However, globally used traditional toxicity indicators are lack of specific and sensitive biomarkers for predicting of drug toxicity and lately detect the drug toxicity. These properties delay drug development and increase drug price. Due to the lack of precise prediction of chemical compound toxicity using traditional toxicity indicators, identifying novel biomarkers for predicting toxicity is needed to reduce the cost and time of drug development such as preclinical safety testing. Furthermore, it is hard to understand complex toxicological process by using parallel omics level analysis. Therefore, multi-omics integrated analysis is needed to improve understanding of toxicological mechanism.
In the context of the overall situation, to improve the precise prediction of organ toxicity with biomarkers, we identified genomic and metabolic biomarkers and built toxicity prediction models with the identified genomic and metabolic biomarkers, respectively. To improve understanding of toxicological mechanism, we identified gene-metabolic network signatures using multi-omics integrated analysis.
At first, to identify genomics biomarkers for major organs (liver, heart and kidney) with three toxicity levels (normal and early and injured stages), totally nine classes, we have performed a step-wise feature selection method with multiple meta-analysis approaches. We used in-depth manually curated massive gene expression profiles from multiple resources, comprising >6,500 in vivo rat samples with >450 distinct compounds. As a result of step-wise feature selection, we discovered 21 genes as a signature for multi-organ toxicity (SMOT), which is involved in the common toxic responses. We built the predictor for multi-organ toxicity (PMOT) with the SMOT genes has a higher predictive performance for the nine classes than a model with integrated genes of previously known genomic biomarkers. Furthermore, some of the SMOT genes including NREP, TPM4, and TRPM4 are well experimentally validated using human cell lines.
Next, to identify metabolic biomarkers for nephrotoxicity with three toxicity levels (normal, low and high toxicities), we performed univariate and bivariate meta-analysis with multiple comparisons using manually curated metabolite expression profiles from multiple studies. After meta-analysis of metabolomics studies, we identified 18 metabolic biomarkers as a signature for kidney toxicity (SKT). Prediction model with the SKT metabolites has a higher predictive performance for the three toxicity classes than a model with integrated metabolites of previously known metabolic biomarkers for kidney toxicity. These metabolic biomarkers are associated to the general disturbed metabolic pathways by drug-induced kidney injury. Most of the SKT metabolites are well experimentally validated using in vivo rat.
Finally, to identify gene-metabolic network signatures for multi-organ toxicity, we performed an integrated analysis using incorporation of the analysis results of two omics into prior knowledge such as interactome and metabolic pathway. After integrated analysis of multi-omics, we identified 9 gene-metabolite network signatures which are relevant to toxicological response. Through using information and connectivity between components of network signature, these gene-metabolic network signatures with multiple molecular-level would be helpful for understanding of toxicological mechanism and reduce the cost and time in drug toxicity and safety assessment.
Although we applied and analyzed three major organs consist of kidneys, liver, and heart, this approach can be directly applied to other organs. These efforts may increase the efficiency to evaluate organ toxicity during drug development and reduce the cost and time of drug discovery and development. In addition, this approach could be applied to identify biomarkers of complex diseases such as cancers and neuron diseases.