Deep autoencoder-based behavioral pattern recognition outperforms standard statistical methods in high-dimensional zebrafish studies.

Zebrafish have become an essential model organism in screening for developmental neurotoxic chemicals and their molecular targets. The success of zebrafish as a screening model is partially due to their physical characteristics including their relatively simple nervous system, rapid development, experimental tractability, and genetic diversity combined with technical advantages that allow for the generation of large amounts of high-dimensional behavioral data. These data are complex and require advanced machine learning and statistical techniques to comprehensively analyze and capture spatiotemporal responses. To accomplish this goal, we have trained semi-supervised deep autoencoders using behavior data from unexposed larval zebrafish to extract quintessential "normal" behavior. Following training, our network was evaluated using data from larvae shown to have significant changes in behavior (using a traditional statistical framework) following exposure to toxicants that include nanomaterials, aromatics, per- and polyfluoroalkyl substances (PFAS), and other environmental contaminants. Further, our model identified new chemicals (Perfluoro-n-octadecanoic acid, 8-Chloroperfluorooctylphosphonic acid, and Nonafluoropentanamide) as capable of inducing abnormal behavior at multiple chemical-concentrations pairs not captured using distance moved alone. Leveraging this deep learning model will allow for better characterization of the different exposure-induced behavioral phenotypes, facilitate improved genetic and neurobehavioral analysis in mechanistic determination studies and provide a robust framework for analyzing complex behaviors found in higher-order model systems.

Diverse PFAS produce unique transcriptomic changes linked to developmental toxicity in zebrafish.

Per- and polyfluoroalkyl substances (PFAS) are a widespread and persistent class of contaminants posing significant environmental and human health concerns. Comprehensive understanding of the modes of action underlying toxicity among structurally diverse PFAS is mostly lacking. To address this need, we recently reported on our application of developing zebrafish to evaluate a large library of PFAS for developmental toxicity. In the present study, we prioritized 15 bioactive PFAS that induced significant morphological effects and performed RNA-sequencing to characterize early transcriptional responses at a single timepoint (48 h post fertilization) after early developmental exposures (8 h post fertilization). Internal concentrations of 5 of the 15 PFAS were measured from pooled whole fish samples across multiple timepoints between 24-120 h post fertilization, and additional temporal transcriptomics at several timepoints (48-96 h post fertilization) were conducted for Nafion byproduct 2. A broad range of differentially expressed gene counts were identified across the PFAS exposures. Most PFAS that elicited robust transcriptomic changes affected biological processes of the brain and nervous system development. While PFAS disrupted unique processes, we also found that similarities in some functional head groups of PFAS were associated with the disruption in expression of similar gene sets. Body burdens after early developmental exposures to select sulfonic acid PFAS, including Nafion byproduct 2, increased from the 24-96 h post fertilization sampling timepoints and were greater than those of sulfonamide PFAS of similar chain lengths. In parallel, the Nafion byproduct 2-induced transcriptional responses increased between 48 and 96 h post fertilization. PFAS characteristics based on toxicity, transcriptomic effects, and modes of action will contribute to further prioritization of PFAS structures for testing and informed hazard assessment.

Deep autoencoder-based behavioral pattern recognition outperforms standard statistical methods in high-dimensional zebrafish studies.

Zebrafish have become an essential tool in screening for developmental neurotoxic chemicals and their molecular targets. The success of zebrafish as a screening model is partially due to their physical characteristics including their relatively simple nervous system, rapid development, experimental tractability, and genetic diversity combined with technical advantages that allow for the generation of large amounts of high-dimensional behavioral data. These data are complex and require advanced machine learning and statistical techniques to comprehensively analyze and capture spatiotemporal responses. To accomplish this goal, we have trained semi-supervised deep autoencoders using behavior data from unexposed larval zebrafish to extract quintessential "normal" behavior. Following training, our network was evaluated using data from larvae shown to have significant changes in behavior (using a traditional statistical framework) following exposure to toxicants that include nanomaterials, aromatics, per- and polyfluoroalkyl substances (PFAS), and other environmental contaminants. Further, our model identified new chemicals (Perfluoro-n-octadecanoic acid, 8-Chloroperfluorooctylphosphonic acid, and Nonafluoropentanamide) as capable of inducing abnormal behavior at multiple chemical-concentrations pairs not captured using distance moved alone. Leveraging this deep learning model will allow for better characterization of the different exposure-induced behavioral phenotypes, facilitate improved genetic and neurobehavioral analysis in mechanistic determination studies and provide a robust framework for analyzing complex behaviors found in higher-order model systems.

The Effects of Long-term, Low-dose ß-N-methylamino-L-alanine (BMAA) Exposures in Adult SODG93R Transgenic Zebrafish.

ß-N-Methylamino-L-alanine (BMAA) is a non-proteinogenic amino acid produced by cyanobacteria, which has been implicated in several neurodegenerative diseases, including amyotrophic lateral sclerosis (ALS). It is postulated that chronic exposure to BMAA can lead to formation of protein aggregates, oxidative stress, and/or excitotoxicity, which are mechanisms involved in the etiology of ALS. While specific genetic mutations are identified in some instances of ALS, it is likely that a combination of genetic and environmental factors, such as exposure to the neurotoxin BMAA, contributes to disease. We used a transgenic zebrafish with an ALS-associated mutation, compared with wild-type fish to explore the potential neurotoxic effects of BMAA through chronic long-term exposures. While our results revealed low concentrations of BMAA in the brains of exposed fish, we found no evidence of decreased swim performance or behavioral differences that might be reflective of neurodegenerative disease. Further research is needed to determine if chronic BMAA exposure in adult zebrafish is a suitable model to study neurodegenerative disease initiation and/or progression.

Leveraging a High-Throughput Screening Method to Identify Mechanisms of Individual Susceptibility Differences in a Genetically Diverse Zebrafish Model.

Understanding the mechanisms behind chemical susceptibility differences is key to protecting sensitive populations. However, elucidating gene-environment interactions (GxE) presents a daunting challenge. While mammalian models have proven useful, problems with scalability to an enormous chemical exposome and clinical translation faced by all models remain; therefore, alternatives are needed. Zebrafish (Danio rerio) have emerged as an excellent model for investigating GxE. This study used a combined bioinformatic and experimental approach to probe the mechanisms underlying chemical susceptibility differences in a genetically diverse zebrafish population. Starting from high-throughput screening (HTS) data, a genome-wide association study (GWAS) using embryonic fish exposed to 0.6 µM Abamectin revealed significantly different effects between individuals. A hypervariable region with two distinct alleles-one with G at the SNP locus (GG) and one with a T and the 16 bp deletion (TT)-associated with differential susceptibility was found. Sensitive fish had significantly lower sox7 expression. Due to their location and the observed expression differences, we hypothesized that these sequences differentially regulate sox7. A luciferase reporter gene assay was used to test if these sequences, alone, could lead to expression differences. The TT allele showed significantly lower expression than the GG allele in MCF-7 cells. To better understand the mechanism behind these expression differences, predicted transcription factor binding differences between individuals were compared in silico, and several putative binding differences were identified. EMSA was used to test for binding differences in whole embryo protein lysate to investigate these TF binding predictions. We confirmed that the GG sequence is bound to protein in zebrafish. Through a competition EMSA using an untagged oligo titration, we confirmed that the GG oligo had a higher binding affinity than the TT oligo, explaining the observed expression differences. This study identified differential susceptibility to chemical exposure in a genetically diverse population, then identified a plausible mechanism behind those differences from a genetic to molecular level. Thus, an HTS-compatible zebrafish model is valuable and adaptable in identifying GxE mechanisms behind susceptibility differences to chemical exposure.

Utilizing a Population-Genetic Framework to Test for Gene-Environment Interactions between Zebrafish Behavior and Chemical Exposure.

Individuals within genetically diverse populations display broad susceptibility differences upon chemical exposures. Understanding the role of gene-environment interactions (GxE) in differential susceptibility to an expanding exposome is key to protecting public health. However, a chemical's potential to elicit GxE is often not considered during risk assessment. Previously, we've leveraged high-throughput zebrafish (Danio rerio) morphology screening data to reveal patterns of potential GxE effects. Here, using a population genetics framework, we apportioned variation in larval behavior and gene expression in three different PFHxA environments via mixed-effect modeling to assess significance of GxE term. We estimated the intraclass correlation (ICC) between full siblings from different families using one-way random-effects model. We found a significant GxE effect upon PFHxA exposure in larval behavior, and the ICC of behavioral responses in the PFHxA exposed population at the lower concentration was 43.7%, while that of the control population was 14.6%. Considering global gene expression data, a total of 3746 genes showed statistically significant GxE. By showing evidence that heritable genetics are directly affecting gene expression and behavioral susceptibility of individuals to PFHxA exposure, we demonstrate how standing genetic variation in a heterogeneous population such as ours can be leveraged to test for potential GxE.

The Ahr2-Dependent wfikkn1 Gene Influences Zebrafish Transcriptome, Proteome, and Behavior.

The aryl hydrocarbon receptor (AHR) is required for vertebrate development and is also activated by exogenous chemicals, including polycyclic aromatic hydrocarbons (PAHs) and 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD). AHR activation is well-understood, but roles of downstream molecular signaling events are largely unknown. From previous transcriptomics in 48 h postfertilization (hpf) zebrafish exposed to several PAHs and TCDD, we found wfikkn1 was highly coexpressed with cyp1a (marker for AHR activation). Thus, we hypothesized wfikkn1's role in AHR signaling, and showed that wfikkn1 expression was Ahr2 (zebrafish ortholog of human AHR)-dependent in developing zebrafish exposed to TCDD. To functionally characterize wfikkn1, we made a CRISPR-Cas9 mutant line with a 16-bp deletion in wfikkn1's exon, and exposed wildtype and mutants to dimethyl sulfoxide or TCDD. 48-hpf mRNA sequencing revealed over 700 genes that were differentially expressed (p < .05, log2FC > 1) between each pair of treatment combinations, suggesting an important role for wfikkn1 in altering both the 48-hpf transcriptome and TCDD-induced expression changes. Mass spectrometry-based proteomics of 48-hpf wildtype and mutants revealed 325 significant differentially expressed proteins. Functional enrichment demonstrated wfikkn1 was involved in skeletal muscle development and played a role in neurological pathways after TCDD exposure. Mutant zebrafish appeared morphologically normal but had significant behavior deficiencies at all life stages, and absence of Wfikkn1 did not significantly alter TCDD-induced behavior effects at all life stages. In conclusion, wfikkn1 did not appear to be significantly involved in TCDD's overt toxicity but is likely a necessary functional member of the AHR signaling cascade.

Systematic developmental toxicity assessment of a structurally diverse library of PFAS in zebrafish.

Per- and polyfluoroalkyl substances (PFAS) are a class of widely used chemicals with limited human health effects data relative to the diversity of structures manufactured. To help fill this data gap, an extensive in vivo developmental toxicity screen was performed on 139 PFAS provided by the US EPA. Dechorionated embryonic zebrafish were exposed to 10 nominal water concentrations of PFAS (0.015-100 µM) from 6 to 120 h post-fertilization (hpf). The embryos were assayed for embryonic photomotor response (EPR), larval photomotor response (LPR), and 13 morphological endpoints. A total of 49 PFAS (35%) were bioactive in one or more assays (11 altered EPR, 25 altered LPR, and 31 altered morphology). Perfluorooctanesulfonamide (FOSA) was the only structure that was bioactive in all 3 assays, while Perfluorodecanoic acid (PFDA) was the most potent teratogen. Low PFAS volatility was associated with developmental toxicity (p < 0.01), but no association was detected between bioactivity and five other physicochemical parameters. The bioactive PFAS were enriched for 6 supergroup chemotypes. The results illustrate the power of a multi-dimensional in vivo platform to assess the developmental (neuro)toxicity of diverse PFAS and in the acceleration of PFAS safety research.

Concurrent Evaluation of Mortality and Behavioral Responses: A Fast and Efficient Testing Approach for High-Throughput Chemical Hazard Identification.

The continual introduction of new chemicals into the market necessitates fast, efficient testing strategies for evaluating their toxicity. Ideally, these high-throughput screening (HTS) methods should capture the entirety of biological complexity while minimizing reliance on expensive resources that are required to assess diverse phenotypic endpoints. In recent years, the zebrafish (Danio rerio) has become a preferred vertebrate model to conduct rapid in vivo toxicity tests. Previously, using HTS data on 1060 chemicals tested as part of the ToxCast program, we showed that early, 24 h post-fertilization (hpf), behavioral responses of zebrafish embryos are predictive of later, 120 h post-fertilization, adverse developmental endpoints-indicating that embryonic behavior is a useful endpoint related to observable morphological effects. Here, our goal was to assess the contributions (i.e., information gain) from multiple phenotypic data streams and propose a framework for efficient identification of chemical hazards. We systematically swept through analysis parameters for data on 24 hpf behavior, 120 hpf behavior, and 120 hpf morphology to optimize settings for each of these assays. We evaluated the concordance of data from behavioral assays with that from morphology. We found that combining information from behavioral and mortality assessments captures early signals of potential chemical hazards, obviating the need to evaluate a comprehensive suite of morphological endpoints in initial screens for toxicity. We have demonstrated that such a screening strategy is useful for detecting compounds that elicit adverse morphological responses, in addition to identifying hazardous compounds that do not disrupt the underlying morphology. The application of this design for rapid preliminary toxicity screening will accelerate chemical testing and aid in prioritizing chemicals for risk assessment.