Moving toward the Real World: Zebrafish Transcript Map Predicts Mixture Effects Using Single-Compound Data.

Human exposure to mixtures of chemicals is ubiquitous, but regulation has long focused on single compounds.1 Today investigators are interested in predicting the effects of chemical mixtures based on known toxicities from single compounds.2 A recent study in Environmental Health Perspectives3 found toxicogenomic predictions to be accurate even when mixture components belonged to different chemical classes or differed in their pharmacological mode of action (MOA).

“High-throughput technology allows us to measure the molecular effects of many chemicals at once, but appropriate experimental designs are key for predicting mixture effects,” says Wibke Busch, a toxicologist at the Helmholtz Centre for Environmental Research in Leipzig, Germany, and the paper’s senior author. “The goal of our proof-of-concept study was to test how well we can predict genome-wide mixture effects from time-resolved dose–response curves for single chemicals [developed in an earlier study4].”

Zebrafish embryo at 27 hours postfertilization. Image: © Ho-Wen Chen/https://www.flickr.com/photos/chenhowen/ (CC BY-NC-ND 2.0; cropped).

In earlier research,4 the authors developed whole-transcriptome time-resolved dose–response curves for three compounds found in many aquatic environments5: two drugs with the same molecular target and MOA in fish (diclofenac, naproxen) and one herbicide with an unknown and presumably dissimilar MOA in fish (diuron). In the new study, the team exposed zebrafish embryos to mixtures of the three chemicals and analyzed whole-transcriptome responses—in other words, the entire range of gene expression.

To reduce the complexity of analyzing the responses of more than 20,000 gene transcripts, the researchers used a grid onto which they had mapped all zebrafish transcripts—a “toxicogenomic universe”—in their earlier study.4 The grid’s 3,600 nodes depict groups of genes in the same anatomical site or with similar functions. Thus, the map is a visual display of whole-transcriptome gene expression “fingerprints” of certain conditions or chemicals to which zebrafish embryos are exposed. The researchers had used a concentration- and time-dependent response (CTR) model to describe the change in transcriptome fingerprints when zebrafish embryos were exposed to the test compounds or mixtures.

“Mixture compounds often have different uptake rates,” explains Busch. “If one chemical reaches the embryo before another, it triggers pathways or signaling mechanisms earlier. To account for this, our CTR model extends the classical dose–response curve by adding time as a third dimension.”

The researchers combined different mixture concepts with the CTR model to predict exposure effects. For example, the concentration addition (CA) concept6 calculates the mixture effect as the sum of the component concentrations, scaled by their individual effect. The researchers chose mixture proportions that would maximize the number of nodes expected to respond to the mixture.

They found that the CTR model accurately predicted most observed mixture effects for both the similar and dissimilar MOAs despite the different natures of the three compounds. Some effects involved known pathways with several contributors; others were linked to general stress response pathways. Among the mixture concepts, CA achieved the best balance between false-positive and false-negative predictions, according to the authors. That concept also identified nodes that responded only to the mixture and not the individual exposures.7

The study may be helpful for risk assessment because it found that mixture effects on the molecular scale—at least for these three components—were additive and could be predicted from single-compound knowledge. Model-based predictions are also useful as a reference point for assessing whether observed effects are additive, synergistic, or antagonistic.

For Emma Schymanski, an associate professor of environmental cheminformatics at the University of Luxembourg, the study is a significant advance. “Its strengths include realistic concentrations, an environmentally relevant mixture, connections to pathways, and a bioassay that is not considered an animal experiment,” she says. (In Europe, experiments with zebrafish embryos up to 120 hours postfertilization are considered alternatives to animal experiments.8)

“Since the largest open chemical database includes 110 million chemicals,9 the next challenge is to test if the method can be scaled up to mixtures of many more chemicals,” adds Schymanski, who was not involved in the study. She also notes that a carefully defined static mixture differs from environmental mixtures that vary seasonally and even daily in their components and respective concentrations.

Gerald LeBlanc, a professor of toxicology at North Carolina State University, who also was not involved in the study, praised the CTR model and the reduction in dimensionality from more than 20,000 single-gene expression levels to 3,600 nodes. “This is a rapid screen that provides a wealth of information for developing and validating predictive models for mixture effects in species that mimic human toxicity better than zebrafish,” he says. “An important next step is to relate gene expression changes to phenotypic outcomes.” That, says Busch, is one of her group’s goals. For this step, she plans to use the concept of adverse outcome pathways,10 or linkages between molecular events caused by a toxicant and resulting organism- or population-level health effects.