Advancements in the Developmental Zebrafish Model for Predictive Human Toxicology

Abstract

The rapid assessment of chemical hazards to human health, with reduced reliance on mammalian testing, is essential in the 21st century. Early life stage zebrafish have emerged as a leading model in the field due to their amenability to high throughput developmental toxicity testing while retaining the benefits of using a whole vertebrate organism with high homology with humans. Zebrafish are particularly well suited for a variety of study areas that are more challenging in other vertebrate model systems including microbiome work, transgenerational studies, gene-environment interactions, molecular responses, and mechanisms of action. The high volume of data generated from zebrafish screening studies is highly valuable for QSAR modeling and dose modeling for use in predictive hazard assessment. Recent advancements and challenges in using early life stage zebrafish for predictive human toxicology are reviewed.

Introduction

Of >350,000 chemicals registered for production globally, the vast majority of these have no toxicity data, and even fewer have in vivo toxicological data¹. New approach methodologies (NAMs) that reduce reliance on mammalian species and increase throughput in toxicological assessments are necessary to meet the demand for human health risk assessment. The use of non-mammalian model organisms in toxicology based on the conservation of disease response pathways is widely accepted as a viable solution to this problem². The recent EU initiative, PrecisionTox, emphasizes the need for non-mammalian model organisms for predictive human health toxicology³. In the US, the Tox21 project is developing alternative testing systems to predict human toxicology, its third phase seeks to advance non-mammalian models like zebrafish and C. elegans for this purpose⁴. For an overview of the benefits and drawbacks of commonly used non-mammalian models for predictive toxicology see Kahbib et al⁵. Zebrafish have emerged as a leading model in this field. Here we will focus on the benefits of and the most recent advancements in early life stage zebrafish (0–120 hpf) as a model organism for predictive human health toxicology and identify research opportunities and challenges unique to this model system (Figure 1).

Figure 1:

Summary of key advantages and advancements in using the developmental zebrafish model for predictive human toxicology.

Translation to humans

Disease responsive pathways are relatively well conserved across all species, including lower vertebrates and invertebrates². Zebrafish specifically, express a high degree of similarity with humans, with orthologs for >82% of human disease-associated proteins^2,6. Despite being fish, they have high anatomical and physiological homology to humans^6,7. While rodents have been the gold standard for pre-clinical chemical testing, there are some cases in which zebrafish are better models for human disease and xenobiotic response, especially in modeling cardiovascular disease^8,9. A recent study found that abnormal zebrafish morphological endpoints were able to predict the human teratogenicity of 31 chemicals with a similar accuracy as rodent studies, and a lower false negative rate¹⁰. Zebrafish have a highly annotated genome and are conducive to the development of transgenic lines which support the characterization of conserved biological processes and humanization when necessary. Hala et al. leveraged pathway analysis to identify conserved and divergent reproductive regulatory networks¹¹. Another study by Kirchberger et al. utilized zebrafish single cell transcriptomics and transgenic reporter lines to describe neutrophil maturation and identify key characteristics of individual human patient’s neuroblastoma tumor cells¹². The availability of single-cell sequencing and ligand-receptor interactomes are expanding the possibilities for translational toxicology by uncovering similarities and differences between zebrafish and humans^13–15.

Screening and hazard identification

In contrast to mammalian model systems, zebrafish are amenable to high throughput developmental toxicity testing while retaining the benefits of using a whole vertebrate organism. Over the course of a typical early life stage high throughput screen the embryo develops externally from a fertilized egg to a fully developed, free-swimming larval fish in 5 days. This assay enables the screening of dozens of chemicals per day at various concentrations and numerous independent replicate samples per concentration. This makes it possible to assess hundreds of chemicals per month in a single laboratory¹⁶. In contrast, a typical rodent study achieves only a small fraction of this throughput and involves far fewer replicate samples, which diminishes confidence in the results. Zebrafish development is ideal for chemical hazard detection because essentially all potential biological targets of chemical insult are active and thus susceptible over that short 5-day period. And because some of these targets are also active in disease responsive pathways in adults, the developmental zebrafish is an effective surrogate for any life stage¹⁷. Unlike in vitro high-throughput screening, it is not necessary to have a priori knowledge of the zebrafish target organs or pathways. Zebrafish developmental toxicity testing provides an unbiased integrated systems approach with morphological and behavioral endpoints for comparative hazard assessment of individual chemicals, mixtures, and environmental samples^18–20. Machine learning and AI approaches further increase the throughput, decrease bias in assessment of morphological and behavioral phenotypes, and help detect subtle phenotypes^21–23.

Beyond typical high throughput screening, zebrafish are effective for studying the complex effects of xenobiotic exposures including transgenerational impacts, epigenetics, microbiome interactions, and gene-environment interactions. Their relatively short generation time makes zebrafish advantageous for transgenerational studies. Haimbaugh et al. utilized zebrafish to demonstrate variable transgenerational behavioral and transcriptomic effects of exposure to environmental polyfluoroalkyl substances (PFAS) levels²⁴. Hao et al. studied the impacts of bisphenol S (BPS) exposure on F0, F1, and F2 generations through methylation status, gonad morphology, steroid concentrations, and methyl transferase gene expression and were able to demonstrate substantial transgenerational reproductive effects from low level exposure to BPS²⁵. For a more thorough review of the use of zebrafish to understand epigenetic effects of xenobiotic exposures see Terrazas-Salgado et al²⁶. The recognition of the microbiome’s role in mediating xenobiotic effects is increasing and zebrafish have proven to be a reliable and versatile model for studying the microbiome, especially in the context of xenobiotic dynamics. Jia et al. provides a recent review of this topic²⁷. Genetic variability can play a major role in toxic outcomes and some zebrafish lines have high genetic variability, in stark contrast to laboratory rodents²⁸. This diversity can be leveraged to identify gene-environment interactions associated with exposure, exemplified by studies of PFAS and endocrine disrupting chemicals^29,30. Zebrafish offer comprehensive insights into developmental toxicity, transgenerational effects, epigenetics, gene-environment interactions and the role of the microbiome in xenobiotic exposures, at a relatively high throughput, making them valuable in chemical hazard identification.

Lastly it is important to note that environmental exposures occur in complex mixtures, which poses a challenge to traditional hazard assessment approaches. The throughput demands of studying the effects of chemical mixtures and their components can be prohibitive in many model systems. However, the high throughput nature of the early life stage zebrafish assay has made zebrafish core to advancements in our understanding of mixture toxicity. Their utility as applied to polycyclic aromatic hydrocarbons specifically is reviewed in Wilson et al¹⁹.

QSAR modeling

Quantitative structure-activity relationship (QSAR) models are essential in understanding the structural underpinnings of toxicity and, by leveraging structural similarity, they reduce reliance on animal models by becoming more and more predictive of hazard from new, untested but correlated structures. QSAR models have been in use by a variety of governmental agencies including the U.S. Food and Drug Administration (US FDA) and European Chemical Agency (ECHA) among others to support regulatory decisions for decades^31,32. However, most of the models in use are based on in vitro data like hepatotoxicity or receptor binding affinity. The high volume of data generated from zebrafish screening studies makes it possible to develop QSAR models for whole organism toxicity. A number of recent studies have developed QSAR models to predict either a binary toxicity outcome or half maximal response (AC50) values^33–35. So far these models can predict toxic outcomes of chemicals within limited domains of applicability reasonably well. The use of interpretable machine learning models for these predictions can identify the most important chemical features likely to be drivers of toxicity. The usefulness of these models in a regulatory context is still limited because of the small training sets relative to the massive untested chemical space and the challenges associated with predicting such a complex endpoint as developmental toxicity³⁶. Growing databases like the US EPA’s ToxCast are widening the chemical space available for developing these models and increasing their applicability³⁷. The zebrafish model is also a valuable tool for the in vivo verification of in silico models predicting specific endpoints like estrogenic potencies³⁸. A more thorough review of machine learning applications in morphological, behavioral, neurological and genotoxicity studies in zebrafish is provided in Wang et al³⁹.

Elucidating in vivo molecular responses

Understanding the molecular response to chemical exposure is a core theme of recent regulatory initiatives^3,4. Zebrafish are particularly well-suited for these studies due to their amenability to high-throughput testing, gene editing techniques, and transcriptomic approaches. Much of the work in this space through 2020 was reviewed in Tal et al⁷. This review covered studies which used unbiased and targeted transcriptomics to identify pathways associated with in vivo toxicity and leveraged high throughput behavioral responses to identify novel chemicals acting through previously understood mechanisms. Also discussed was the use of CRISPR gene editing technologies to verify in vivo the importance of gene products in the toxic mechanism of action, a technique frequently used in zebrafish studies⁷. This section will cover the most recent developments in this area and areas not covered in the previous review.

Advancements in RNA sequencing technologies, such as bulk RNA sequencing, single-cell RNA sequencing, and spatial transcriptomics, are broadening the potential for investigating molecular responses zebrafish and translation to higher vertebrates. Bulk RNA sequencing, the average transcriptomic profile of a whole embryo or pooled embryos, is a common approach used to investigate gene expression responses to exposure in zebrafish. The high-throughput nature of zebrafish exposures enables the generation of large transcriptomic datasets, rich for co-expression network analysis. This method visualizes gene networks and their expression patterns across chemical exposures, identifying transcriptomic network modules responsive to specific chemical classes or moieties within a class, and revealing the centrality of specific genes in xenobiotic responses^40–42. Singel-cell RNA sequencing, which provides sequencing of individual cells that make up the fish as opposed to whole animals, has been utilized to better characterize normal zebrafish developmental trajectories and define shared developmental cascades with humans and other model systems^14,15. This has direct advantages for toxicological work as exemplified in a study using single-cell RNA sequencing, coupled with phenotypic outcomes, to identify the cell types and functions most impacted by exposure to polyfluorooctanoic acid (PFOA)⁴³. Building on these approaches, spatial transcriptomics is an innovative technique poised to make significant contributions to zebrafish toxicology research. With this technology it is possible to section a whole zebrafish embryo and localize transcriptomic changes. To date this technology has been used to create a spatiotemporal map of gene expression during normal zebrafish development and localize transcriptomic responses in optic nerve regeneration^44,45. While its use in toxicology is still forthcoming, spatial transcriptomics promises a greater ability to uncover regional xenobiotic responses and precisely identify targeted tissues.

The development of transgenic reporter lines further compliments these approaches. Reporters enable the visualization of chemical effects on neurological and cardiovascular development or the localized expression of important genes in optically clear, early life stage zebrafish. This is exemplified in a study that used fluorescent transgenic lines marking blood vessel formation and heart development to detect chemical induced perturbations⁴⁶. Another study combined the high-throughput abilities of the zebrafish with a transgenic reporter line for cytochrome P4501A as a proxy for aryl hydrocarbon receptor (AHR) activation to bin the activity profiles and molecular responses of 123 polycyclic aromatic hydrocarbons⁴⁷.

Overall, the diverse array of tools available for investigating molecular responses in zebrafish facilitates the integration of multiple approaches to understanding the effects of chemical exposure. Integrative omics approaches are increasingly common in zebrafish studies and provide more comprehensive insights into molecular mechanisms than transcriptomics alone, demonstrating the value of leveraging multiple techniques. This is exemplified in a zebrafish study which used both metabolomics and gene expression to identify biological pathways perturbed by exposure to triclosan and methyl-triclosan with increased clarity⁴⁸. This revealed upregulation of the pentose phosphate pathway as the source of altered glucose 6-phosphate levels⁴⁸. Lee et al. used an integrated multi-omics approach (metabolomics, transcriptomics and proteomics) to more decisively identify pathways impacted by perfluorooctanesulfonic acid (PFOS) exposure than was possible with individual omics, like negative impacts on neuron structure and formation, neuroinflammation, and Ca²⁺ signaling pathways⁴⁹.

Dose/uptake modeling

One major challenge in using zebrafish, or any other model system, to predict human health toxicology is the translation of dose. Most high throughput zebrafish exposure studies use a polystyrene 96-well plate format. A small volume of medium is loaded into each well and a single zebrafish embryo is then added, usually between 1 and 6 hours post fertilization, and chemical is dosed immediately afterward¹⁶. Some protocols call for renewal while many others do not. Due to the high cost and low throughput of body burden measurements, most concentrations are only reported as the nominal target in the medium. However, in a 96-well plate format there is often high loss from medium to plate wall or through evaporation, making the bioavailable concentration and the internal dose much lower than the reported nominal concentration^50,51. For zebrafish screening data to be applied to human risk assessment, accurate dose predictions are essential. A brief overview of basic toxicokinetic considerations in early life stage zebrafish toxicity testing is offered in Bauer et al⁵². This has been an active area of research in recent years, similar to in vitro to in vivo extrapolation (IVIVE) which is key to expanding the applicability of in vitro assays⁵³. Achenbach et al. demonstrated that differences in chemical toxicokinetic profiles could partially explain differences in toxicity profiles⁵⁴. Understanding chemical uptake, metabolism, and distribution etc. is essential to translational and comparative toxicology

Recent work has focused on enhancing the ability to predict dose in zebrafish exposure studies. Some groups have continued to establish physiologically based pharmacokinetic (PBPK) models for small groups of similar compounds like bisphenol A (BPA) analogs⁵⁵. Other studies have integrated QSAR methods to predict the uptake of chemicals by zebrafish tissues, usually within a certain class like cephalosporins or valproic acid analogs^56,57. Zhang et al. measured the bioconcentration of eight ionizable organic compounds and found that the human apparent volume of distribution (a parameter describing the sorption capacity in tissue relative to blood) was the most reliable predictive property for uptake into zebrafish embryos, not the octanol-water coefficient commonly used for neutral chemicals⁵⁸. But humans are not exposed to single chemicals in the environment, making it essential to understand how mixture exposures impact chemical uptake and toxic outcomes. Still, there are a limited number of studies examining the effects of mixtures on PBPK models in zebrafish, most of which examined binary mixtures⁵⁹. However, recent zebrafish study used seven PFAS chemicals to model the impact of PFAS mixture kinetics and revealed that individual chemical kinetics varied substantially with the composition of the PFAS mixture, revealing the importance of studying kinetic in the context of complex mixtures⁶⁰.

Summary

Early life stage zebrafish are a valuable tool in predicting chemical hazard to human health in the chemical space of the 21^st century. Zebrafish are particularly well suited for a variety of study areas that are more challenging in other vertebrate model systems including microbiome work, transgenerational studies, mixture effects, gene-environment interactions, molecular responses and mechanisms of action. Despite the value of the zebrafish for predictive toxicology, better dose modeling and a much larger and more diverse QSAR dataset are needed to achieve full integration of the model into human risk assessment.

Abbreviations:

NAMs

New approach methodologies

PFAS

polyfluoroalkyl substances

BPS

bisphenol S

QSAR

Quantitative structure-activity relationship

US FDA

U.S. Food and Drug Administration

ECHA

European Chemical Agency

AC50

half maximal response

PFOA

polyfluorooctanoic acid

AHR

aryl hydrocarbon receptor

PFOS

perfluorooctanesulfonic acid

IVIVE

in vitro to in vivo extrapolation

PBPK

physiologically based pharmacokinetic

BPA

bisphenol A