henhaobio
Sign in

Environmental Science & TechnologyPharmacology & ToxicologyPeer-reviewed journal article

Leveraging zebrafish embryo phenotypic observations to advance data-driven analyses in toxicology

Paul MichaelisiD, Nils KlüveriD, Silke Aulhorn, Hannes BohringiD, Jan BumbergeriD, Kristina Haase, Tobias KuhnertiD, Eberhard KüsteriD, Janet Krüger, Till LuckenbachiD, Riccardo MasseiiD, Lukas Nerlich, Sven PetruschkeiD, Thomas SchnickeiD, Anton SchnurpeliD, Stefan ScholziD, Nicole SchweigeriD, Daniel SielaffiD, Wibke BuschiD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Zebrafish have emerged as a central model organism in toxicological research. Zebrafish embryos are exempt from certain animal testing regulations which facilitates their use in toxicological testing. Next to the zebrafish embryo acute toxicity test (ZFET) according to the OECD TG 236, fish embryos are used in mechanistic investigations, chemical screenings, in ecotoxicology, and drug development. However, inconsistencies in the applied test protocols and the monitored endpoints in addition to a lack of standardized data formats, impede comprehensive meta-analyses and cross-study comparisons. To address these challenges, we developed the Integrated Effect Database for Toxicological Observations (INTOB), a comprehensive data management tool that standardizes collection of metadata and phenotypic observations using a controlled vocabulary. By incorporating data from more than 600 experiments into the database and subsequent comprehensive data analyses, we demonstrate its utility in improving the comparability and interoperability of toxicity data. Our results show that the ZFET can detect toxicity spanning seven orders of magnitude at the scale of effect concentrations. We also highlight the potential of read-across analyses based on morphological fingerprints and their connection to chemical modes of action, provide information on control variability of the ZFET, and highlight the importance of time for mechanistic understanding in chemical exposure-effect assessments. We provide the full FAIR dataset as well as the analysis workflow and demonstrate how professional data management, as enabled with INTOB, marks a significant advancement by offering a comprehensive framework for the systematic use of zebrafish embryo toxicity data, thus paving the way for more reliable, data-driven chemical risk assessment. SynopsisThis article shows how a novel data management tool for zebrafish embryo toxicity data advances comparative chemical risk assessment by enhancing data and metadata standardization, comparability, and cross-study analyses.

In-depth summary

Claude Sonnet 4.6 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (15 Oct 2024) · CC BY 4.0

A structured zebrafish embryo toxicity database (INTOB, 638 experiments) enabled control-quality tests, dose-response modeling and phenotype-fingerprint clustering of 47 substances (133 experiments); literature data proved unsuitable for meta-analysis.

Zebrafish embryo toxicity data are often not FAIR (Findable, Accessible, Interoperable, Reusable): they appear as analyzed results, with varying nomenclature and missing exposure metadata, which blocks joint analysis. INTOB stores metadata and per-embryo observations with a defined effect vocabulary in a machine-readable format.

Key results

  1. ShownA binomial-test control-quality filter that accounts for varying sample sizes removed a small share of experiments before analysis.

    Across 11080 control embryos, 93.6% were normal at the end of the experiment and 95.9% survived. Of 609 experiments, 62 had insufficient control quality, leaving 547. Experiments with fewer than 9 control embryos were also excluded. · Data processing and quality control; Control variability in ZFETs

  2. ShownDose-response modeling across 71 substances showed mostly small sensitivity ratios (SR = LC50/EC50) and effect concentrations spanning about 9 orders of magnitude (S16; the Discussion states 7). Exposure start (0 vs 24 hpf, hours post-fertilization) changed toxicity for some substances.

    Of 180 experiments, 166 allowed cumulative models. Only six substances had SR above 5 (benzovindiflupyr 22.4, valproic acid 16.3, ferbam 9.8, diclofenac sodium 6.2 and 8.4, acetaminophen 7.2, propoxur 6.7). Cyprodinil SR was 1.1 at 0 hpf and 3.1 at 24 hpf. Carbamazepine and imidacloprid were more toxic with 24 hpf exposure. · Dose response modeling; Sensitivity ratio analysis; Dynamic range of effect concentrations in the…; Discussion

  3. ShownHierarchical clustering of phenotype fingerprints at 96 hpf split 47 substances into six clusters. Some clusters were defined by one dominant effect (coagulation; low pigmentation; behavioral effects).

    Fingerprints came from 133 experiments and 6225 embryos; clusters had 5 to 12 substances. Cluster 3 (isoproturon, ferbam, carbendazim) was defined by coagulation with SR close to 1. Cluster 4 (6-PTU, 6-PPDQ, lenacil, acetaminophen, fipronil) was defined by low pigmentation. In a PCA, coagulation was second in importance only to no hatching. · Phenotypic fingerprints; Clusters of chemicals with similar morphological…

  4. ShownEffects at 48 hpf generally persist at 96 hpf, but decreased heartbeat progresses to more severe heart effects. Fingerprints for some substances change with observation time, e.g. triclosan.

    In 778 embryos from well-plate experiments, decreased heartbeat correlated only weakly with itself across time (Pearson r = 0.14). Normal at 48 hpf vs coagulated at 96 hpf gave r = -0.61, the strongest negative pair. For triclosan the SR approached 1 at 96 hpf. · Effect propagation; Occurrence and co-occurrence of morphological…

  5. ShownPhenotype data collected from the literature could not support meta-analysis of zebrafish embryo toxicity tests. Clusters formed by study rather than by substance.

    The review covered 162 publications (2014 to 2023) and 522 chemicals, samples or mixtures. Substances from separate studies rarely clustered together. Overlap with the INTOB data was small. The effect patterns of valproic acid and its analogues in two Brotzmann studies were hardly comparable, and the comparison with INTOB valproic acid data showed some overlaps but no clear picture. · Literature review on ZFE toxicological…; Comparison of chemical effects in ZFET based on…

Methods, data and reproducibility

Zebrafish embryo toxicity observations and metadata were stored in INTOB (relational database with web UI, GraphQL and REST API) and analyzed in R. Analyses covered binomial control QC, drc dose-response fits, fingerprint clustering (Ward, PCA, UMAP) and Pearson correlations of effects across time.

  • INTOB zebrafish embryo acute toxicity test experiments from one institute (last 20 years plus current direct entry); 609 after initial filtering, 547 after control QC · zebrafish · n = 638 experiments; 136 substances; 11080 control embryos · 10.5281/zenodo.11030299
  • Systematic literature review of zebrafish embryo morphological phenotypes after chemical exposure (2014 to 2023) · zebrafish · n = 162 publications; 522 chemicals, samples or mixtures · 10.5281/zenodo.11030299
Methods & tools
INTOB, R v.4.3.0, drc v.3.0-1, ghql v.0.1.0, jsonlite v.1.8.8, GraphQL, CompTox database (mirror), OECD TG 236, Zotero, PubMed, Science Direct, UMAP
Software & environment
R v.4.3.0; drc v.3.0-1; ghql v.0.1.0; jsonlite v.1.8.8. Data used were those recorded in INTOB before 7 March 2024.
Reproducibility gaps
The INTOB software itself is commercially available on request (https://www.ufz.de/intob/index.php?en=51326), not openly released. The text does not say how the Zenodo record is licensed.
Limitations (4)
  • Different trained observers judge phenotypes differently; manual microscopic scoring can add experimenter bias, which the authors suggest automated imaging could reduce. · stated by the authors
  • Less specific effects can obscure mode-of-action patterns and hinder clustering. Benzovindiflupyr, with low pigmentation and smaller head and eye size, grouped with other chemicals because it showed many other effects. · stated by the authors
  • Literature data were too heterogeneous (effects, time points, exposure windows, concentrations; mostly yes/no reporting) to compare with the INTOB data, so the overlap analysis was limited. · stated by the authors
  • All experiments come from one institute with varying designs (replicates, vessels, exposure start, time points), and cluster-to-mechanism links such as thyroid disruption are discussed from literature, not tested here. · noted in this summary

Read this if you work on zebrafish embryo toxicity data, need ideas for control-quality filtering and phenotype fingerprint clustering, or want an open dataset of 136 tested chemicals.

Citations
21
FWCI
9.45
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISAhmad '20Pfefferle '25Kagoo '22Torres-Ruíz '26Mundy '21Leuthold '25
65% semantic matchPharmacology & Toxicology
F. Ahmad, S. Ali, M. K. Richardson·bioRxiv·Oct 5, 2020