Impact of Genomics Platform and Statistical Filtering on Transcriptional Benchmark Doses (BMD) and Multiple Approaches for Selection of Chemical Point of Departure (PoD).

Leslie Recio,A. Francina Webster,Byron Kuo,Nikolai Chepelev,Carole L. Yauk,Rémi Gagné,Andrew Williams,Hikmet Budak

doi:10.1371/journal.pone.0136764

Leslie Recio, A. Francina Webster + Show 6 more

Open Access

https://doi.org/10.1371/journal.pone.0136764

Copy DOI

Abstract

Many regulatory agencies are exploring ways to integrate toxicogenomic data into their chemical risk assessments. The major challenge lies in determining how to distill the complex data produced by high-content, multi-dose gene expression studies into quantitative information. It has been proposed that benchmark dose (BMD) values derived from toxicogenomics data be used as point of departure (PoD) values in chemical risk assessments. However, there is limited information regarding which genomics platforms are most suitable and how to select appropriate PoD values. In this study, we compared BMD values modeled from RNA sequencing-, microarray-, and qPCR-derived gene expression data from a single study, and explored multiple approaches for selecting a single PoD from these data. The strategies evaluated include several that do not require prior mechanistic knowledge of the compound for selection of the PoD, thus providing approaches for assessing data-poor chemicals. We used RNA extracted from the livers of female mice exposed to non-carcinogenic (0, 2 mg/kg/day, mkd) and carcinogenic (4, 8 mkd) doses of furan for 21 days. We show that transcriptional BMD values were consistent across technologies and highly predictive of the two-year cancer bioassay-based PoD. We also demonstrate that filtering data based on statistically significant changes in gene expression prior to BMD modeling creates more conservative BMD values. Taken together, this case study on mice exposed to furan demonstrates that high-content toxicogenomics studies produce robust data for BMD modelling that are minimally affected by inter-technology variability and highly predictive of cancer-based PoD doses.

Highlights

Toxicogenomics is expected to become an asset to human health risk assessment because it provides mechanistic data in a more efficient and cost-effective manner, using fewer experimental animals than the majority of standard toxicity testing methods
If toxicogenomics data are to be included in chemical risk assessment, a consensus must be reached regarding which genomics platforms are appropriate, how the data should be modeled, and how to choose appropriate point of departure (PoD) values
Inter-platform consistency of DEG fold change, and DEG and pathway benchmark dose (BMD) values, 2) the effect of filtering data based on statistically significant changes in gene expression prior to modeling in BMDExpress, 3) the ability to produce transcriptional BMD values that are predictive of known cancer BMD values, and 4) different approaches for selecting a single PoD from a toxicogenomics dataset for use in risk assessment

Summary

Introduction

Toxicogenomics is expected to become an asset to human health risk assessment because it provides mechanistic data in a more efficient and cost-effective manner, using fewer experimental animals than the majority of standard toxicity testing methods. Toxicogenomics data can be used to determine the molecular mode of action (MoA) of chemical carcinogens and has been shown to be predictive of genotoxicity and cancer outcomes [1,2,3,4,5,6,7,8,9,10,11,12,13]. For these reasons, many regulatory agencies worldwide are exploring ways to incorporate toxicogenomic data into their chemical risk assessments. In order to consider RNA-seq as an alternative to microarrays, it will be important to understand how these two high-content technologies compare with respect to the mechanistic insight and quantitative outputs they produce

Objectives

Methods

Results

Conclusion