https://orcid.org/0000-0002-8700-4834
Abstract
Assessment of potential impacts of chemicals on the environment traditionally involves regulatory standard data requirements for acute aquatic toxicity testing using algae, daphnids, and fish (e.g., Organisation for Economic Co-operation and Development [OECD] test guidelines 201, 202, and 203, respectively), representing different trophic levels. In line with the societal goal to replace or reduce vertebrate animal testing, alternative bioassays were developed to replace testing with fish: the fish cell line RTgill-W1 acute toxicity assay (OECD test guideline 249) and the zebrafish embryo acute toxicity test (zFET, OECD test guideline 236). However, previous studies revealed the lower sensitivity of the RTgill-W1 cell line assay and zFET for some neurotoxic chemicals and allyl alcohol, which is presumably biotransformed in fish to the more toxic acrolein (which is predicted well through the cell line assay). To provide an additional alternative to acute fish toxicity, in this study we analyzed historic ecotoxicity data for fish and daphnids from the EnviroTox Database. We found a considerable variability in acute fish median lethal concentration and acute daphnids median effect concentration values, particularly for neurotoxic chemicals. Comparing sensitivity of these taxonomic groups according to different neurotoxicity classification schemes indicates that fish rarely represent the most sensitive trophic level of the two. Exceptions here most prominently include a few cyclodiene compounds, which are no longer marketed, and a chemical group that could be identified through structural alerts. Moreover, daphnids are more sensitive than fish to acrolein. This analysis highlights the potential of the Daphnia acute toxicity test, which is usually a standard regulatory data requirement, in safeguarding the environmental protection level provided by the RTgill-W1 cell line assay and the zFET. This research, rooted in decades of efforts to replace the fish acute toxicity test, shifts the focus from predicting fish toxicity one-to-one to emphasizing the protectiveness of alternative methods, paving the way for further eliminating vertebrate tests in environmental toxicology.
Introduction
Traditional assessment of potential impacts of chemicals on the environment involves standard regulatory data requirements for acute aquatic toxicity testing using algae, daphnids, and fish, representing diverse trophic levels. Standardization of experimental setup and conditions in test guidelines, such as those by the Organisation for Economic Co-operation and Development (OECD), does not aim to capture the huge biological variability but to pragmatically enable toxicity comparisons under similar conditions. These include test guideline 203 for acute fish toxicity and test guideline 202 on Daphnia sp. acute immobilization (as a surrogate for mortality; OECD, 1992, 2004, 2019).
Replacing the acute fish toxicity test
Particularly for fish, the motivation for finding adequate replacement methods is strong for ethical, practical, and scientific reasons. Ethically, test guideline 203 is of considerable animal welfare concern because mortality (or moribundity) is the endpoint. Practically, the relatively high volume of water and chemicals required, the necessary infrastructure, and the 96 hours of exposure time are disadvantageous in terms of environmental burden and testing throughput. Scientifically, the basic study design of test guideline 203 causes uncertainty for experimental variability due to the low number of fish per concentration, the lack of tank replication, the absence of an internal positive control, and a high degree of flexibility in the test design. It allows for use of 11 different fish species, different endpoints for clinical signs in addition to lethality, and, until its revision in 2019, animal life stages only defined by specimen body length (OECD, 1992, 2019). Moreover, there is considerable uncertainty for the extrapolation of such laboratory data to environmental protection goals (Paparella et al., 2021).
Despite these concerns, the acute fish toxicity test is by far the most frequently conducted aquatic toxicity test using approximatley 50,000 fish per year in Europe, which represents approximately 50% of all aquatic vertebrates used for all types of regulatory ecotoxicity testing (Burden et al., 2020; European Commission, 2024). Furthermore, regulatory ecotoxicology faces an ever-growing chemical landscape that was estimated in a 2020 review article to include more than 350,000 chemicals and mixtures currently registered on the market worldwide (Wang et al., 2020). For example, the European Union legislation for the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH) requires the testing of chemicals for acute fish toxicity at a production/import volume above 10 tons per year, which currently covers approximately 12,000 chemicals (European Chemical Agency, 2024). Therefore, there have been considerable efforts to reduce and replace the use of fish in acute toxicity testing, including the development of a testing strategy (OECD Guidance Document 126 the threshold approach for acute fish toxicity; OECD, 2010) and the development of alternative methods.
The threshold approach for acute fish toxicity
The observation that fish are not always the most sensitive test taxon (Weyers et al., 2000) provided the foundation for developing the threshold approach for acute fish toxicity (OECD, 2010). The threshold approach addresses fish toxicity by initially using a single-concentration test (limit test), thus requiring fewer fish compared to the full acute fish toxicity study. The selection of a single concentration is based on the derivation of a threshold concentration from algae and acute invertebrate (e.g., daphnids) toxicity data. The concept initially described for pharmaceuticals (Hutchinson et al., 2003) was further developed for chemical substances at the European Commission’s Joint Research Centre (Jeram et al., 2005), taking into consideration the requirements of the limit test in OECD test guideline 203 (European Centre for the Validation of Alternative Methods, 2006; OECD, 2019). In addition, several publications confirm the potential of the threshold approach to reduce the number of fish used for acute toxicity testing (Hoekzema et al., 2006; Rawlings et al., 2019), also when applied to substances other than industrial chemicals, such as pesticides and pharmaceuticals. Furthermore, retrospective analyses of plant protection products indicate that the overall pain and distress experienced by the fish used in the threshold approach procedure is much less than that of a full acute fish toxicity study, where mortality and adverse symptoms are intended (Creton et al., 2014; OECD, 2002).
Alternative approaches for acute fish toxicity
Several bioassay-based alternative methods (new approach methodologies [NAMs]) have been developed or are currently under development to replace the acute fish toxicity test (OECD, 2019) and offer ethical, practical, and economic advantages as well as lower variability and scientific validation (Paparella et al., 2021). These, most notably, include the fish cytotoxicity test using the cell line RTgill-W1 (OECD test guideline 249; OECD, 2021) and the zebrafish embryo acute toxicity test (zFET; OECD test guideline 236; OECD, 2013). As an additional advantage, data from these alternative methods may support the improvement of computational approaches for predictive ecotoxicology due to lower data variability (Paparella et al., 2021).
However, some limitations pose challenges to the widespread adoption of alternative methods as replacements for the fish acute toxicity test (Westerink, 2013): (1) Studies have highlighted lower sensitivity of the RTgill-W1 cell line assay and zFET to certain modes of neurotoxicity (Glaberman et al., 2017; Klüver et al., 2015; Knöbel et al., 2012; Sobanska et al., 2018; Tanneberger et al., 2013), that is, causing specific adverse functional or structural changes in the nervous system (Legradi et al., 2018). For the RTgill-W1 cell line assay, this is hypothesized to be due to the absence of specific channels or receptors present in nerve but not gill cells (Fischer et al., 2019; Tanneberger et al., 2013). For the zFET, this appears at least to some degree to be related to the oxygen supply of the larvae relying on passive diffusion instead of being dependent on active gill ventilation, as is the case for adults. Active ventilation can be affected through neurotoxicants, rendering older fish life stages more susceptible to respiratory failure syndrome (Kämmer et al., 2024; Klüver et al., 2015). (2) Differences in biotransformation capacity may lead to the underestimation of acute fish toxicity via zFET and the RTgill-W1 cell line assay for compounds more toxic after biotransformation. As a prominent example, allyl alcohol is presumably biotransformed in fish to the more toxic chemical acrolein, whereas allyl alcohol is considered to be less toxic in the RTgill-W1 cell line assay and zFET than in the fish acute toxicity test (Glaberman et al., 2017; Klüver et al., 2015; Sobanska et al., 2018). Allyl alcohol is not biotransformed in zebrafish embryos due to the lower expression of the alcohol dehydrogenase isoenzyme adh8a (Braunbeck et al., 2020; Klüver et al., 2014; Knöbel et al., 2012). (3) Another limitation of the zFET is that some molecules may not pass the chorion due to their molecular size, charge, and/or structure (>3,000 Da for uncharged molecules; Busquet et al., 2014; Braunbeck et al., 2020; Pelka et al., 2017).
Currently, an OECD project advancing the threshold approach into a framework of more comprehensive integrated approaches to testing and assessment (IATA) for acute fish toxicity is ongoing (project 2.54 on the OECD Test Guidelines Programme [last accessed November 13, 2024]). This IATA will provide adequate information for classification and labeling according to the United Nations Globally Harmonized System of Classification (UN GHS) as well as for hazard and risk assessment. It integrates algae and daphnid testing, the zFET, the RTgill-W1 cell line assay, computational approaches such as quantitative structure activity relationship (QSAR) methods, physicochemical information, and additional relevant data, for example, from high-throughput screening and other taxa. However, concerns regarding the ability of the alternative methods to be protective for neurotoxic chemicals have hampered their use for regulatory purposes and have been a motivation for the investigation into whether invertebrate toxicity data are protective for neurotoxicants in fish presented in this study.
Can invertebrate toxicity data be protective for neurotoxicants in fish?
Recently, it was found that mainly pesticides acting through acetylcholinesterase inhibition and aromatic amines are much more toxic to Daphnia magna than to fish (Kienzler et al., 2016). That study analyzed databases available within the OECD QSAR Toolbox 3.2 (Dimitrov et al., 2016; OECD, 2023), exploring extrapolation approaches for avoiding chronic fish testing on the basis of existing data. For example, the neurotoxic insecticide fenitrothion had a medium toxicity to fish but appeared as one of the most chronically toxic chemicals to D. magna (Kienzler et al., 2016). In this context, the hypothesis of the data analysis in the present study was to elucidate whether invertebrate toxicity data can be protective for fish for a broad range of neurotoxic compounds. This alternative route to acute fish toxicity prediction could present an important key for the full replacement of acute fish toxicity testing through alternative methods.
In this study, we present a systematic analysis of quality-assured ecotoxicity data from the EnviroTox Database (Connors et al., 2019) with a focus on chemicals with neurotoxic modes of action. We investigate biological data variability and compare the sensitivity of fish and daphnids for chemicals with a neurotoxic mode of action (MoA), based on different classification schemes. The resulting insights are contextualized to assess the potential of daphnids to compensate for the described current weaknesses of biological alternative methods for neurotoxicity in the environment.
Materials and methods
Data sources and curation
The data used in this analysis originate from the EnviroTox database, a curated aquatic toxicity database (version downloaded January 23, 2024 from https://EnviroToxdatabase.org). The database was selected because it is compiled based on quality criteria in accordance with the Stepwise Information-Filtering Tool (Beasley et al., 2015) outlined in Connors et al. (2019) and therefore contains high-quality data. A number of filtering criteria were applied to reduce the data to entries that can be reasonably assumed to relate to experiments carried out in accordance with the OECD test guideline 203 for fish (OECD, 2019) and OECD test guideline 202 for daphnids (OECD, 2004).
In short, this includes only data points for acute toxicity testing with the 11 fish species suggested in the test guideline 203 at 96 hr and all daphnid species in the test guideline 202 at 48 hr. For fish, this is limited to outcomes of mortality and moribundity and represented by the lethal concentration 50% (LC50) or median effect concentration (EC50). Throughout the study, LC50 is used for fish but encompasses both LC50 and EC50 data. For daphnids, the outcome is intoxication/immobilization as a surrogate for mortality (EC50). Additionally, if terms within the study title or effect descriptions were related to egg, embryo, or larval data, these data points were excluded to ensure that only tests with juvenile or adult life stages were included. Finally, using the binary filtering option contained in the EnviroTox database, all entries where the LC50 or EC50 lay fivefold above the predicted chemical water solubility were excluded.
To label the tested chemicals as relevant to neurotoxicity, a number of different classification approaches were compared. The EnviroTox database already includes MoA classifications according to Verhaar, TEST, ASTER, and OASIS (Connors et al., 2019; Kienzler et al., 2017, 2019). The TEST and ASTER classifications were deemed to be most informative for the purpose of discerning neurotoxic and non-neurotoxic MoA, due to the inclusion of neurotoxicity-specific classes. These classifications were expanded on through the inclusion of pesticide-specific modes of action from the Insecticide Resistance Action Committee (IRAC) Mode of Action Database, which is built on experimental data for the purpose of pesticide resistance management (https://irac-online.org/mode-of-action). Additionally, two other sources of MoA categorization were included: (1) the candidate compound list by the European Food Safety Authority (EFSA) for the implementation of an in vitro test battery focused on developmental neurotoxicity, contributing labeling for 119 chemicals (Masjosthusmann et al., 2020) and (2) a list of chemicals that are regularly found in chemical monitoring of European rivers, amounting to more than 3,300 compounds with an assigned MoA class (Kramer et al., 2024). The EFSA data set was augmented by SMILES codes to check overlap with the EnviroTox data set based on CAS, SMILES, and chemical name. Except for the IRAC categorization, all approaches include both positive and negative compounds to allow for the contextualization of neurotoxically active compounds against chemicals that are not considered neurotoxic or that have not been assigned with a category. Figure 1 gives an overview of the 805 final chemicals included in the analysis (i.e., those for which data for both fish and daphnids in accordance with our filtering criteria are available), their subcategories across classification schemes, and the number of compounds in the different categories. However, the Sankey chart (see online supplementary material Figure S1) gives an overview of which subcategories chemicals are assigned to across the different categorization schemes and highlights some discrepancies and limitations of individual approaches, where certain chemicals are labeled neurotoxic in one classification but non-neurotoxic in another. Because the categorization schemes provide a broad range of MoA classifications related to different applications (e.g., pesticides vs. surface water monitoring), categories and chemicals will overlap or differ between these schemes. We see this as a redundancy mechanism strengthening the analysis.
Overview of the 805 chemicals included in the analysis for which both fish and daphnid data in accordance with our filtering criteria was available, their subcategories across classification schemes, and the number of compounds in the different categories of categorization schemes used to label compounds as potentially neurotoxically active and nonactive based on different data sources. Note. Op = organophosphate; AChE = acetylcholine esterase; Clc = chloride channel; nAChr = nicotinic acetylcholine receptor.
After merging data from the EnviroTox database with the supplementary MoA data, the effect values were summarized by calculating median values for each chemical per taxonomic group (fish and daphnids). Afterwards, each of these median values per chemical was matched with each median value for daphnids using the same chemical (based on CAS number). We considered different approaches of summarizing and matching the data, and this approach may be justified considering that the current OECD test guideline allows the use of multiple species and there is no robust knowledge on species-specific sensitivity to the different chemicals on the market. Further reasoning and comparison with alternative approaches for this, including calculations based on the geometric mean, is described in the online supplementary material.
To compare sensitivity between the two taxonomic groups, the ratio Rdf was calculated according to the equation in Figure 2.
Calculation of Rdf, the ratio between daphnid acute median effect concentration (EC50) and fish acute median lethal concentration (LC50). The colored bars to the right indicate different levels of concern: Rdf ≤1 and ≤10 are considered unproblematic, because they indicate that daphnids are either more sensitive or up to a factor of 10 less sensitive than fish to the respective compound and can, thus, be considered protective for fish. The Rdf values above 10 are critical, because they indicate that fish are more sensitive than daphnids by at least a factor of 10. The figure was created with BioRender.com.
This ratio gives an intuitive understanding of relative sensitivity between the two taxonomic groups. A ratio of 1 indicates that the two groups are equally sensitive to the respective chemical, whereas values below 1 relate to a higher sensitivity in daphnids. For the analysis, ratios more than 10 were considered to be potentially problematic, because they indicate that, for the respective chemical, fish are at least 10 times more sensitive than daphnids in the standardized test guideline tests.
Data analysis
Data transformation, analysis, and visualization were carried out using R (R Core Team, 2018) with RStudio (Ver. 2023.12.1 + 402; RStudio Team, 2024) and the tidyverse package (Wickham et al., 2019). Ridgeline plots were created using the ggridges package (Wilke, 2024). Figure 2 was created with BioRender.com. The Sankey chart (see online supplementary material Figure S1) was created with ggsankey (Sjoberg, 2024). Random subsampling of data points for Figure 3 was done using the slice_sample function from the dplyr package. The Kruskal-Wallis rank sum test was performed using the kruskal.test function of the stats package. The geometric mean for online supplementary material Figures S11–S14 was calculated in R with the geometric.mean function from the psych package. The SMILES codes for the supplementary data were retrieved using the Python package cirpy (https://cirpy.readthedocs.io). The layout of Figure 5 was adjusted using Inkscape 1.2.2 (732a01da63, 2022-12-09) for Windows. The R code used to produce the analyses and figures in this publication is available as two quarto documents on GitHub (https://github.com/schuerc/fish_daphnia_neurotox_public; last accessed November 25, 2024).
Variability of median lethal concentration (LC50) values across more than four species in fish for 20 randomly selected chemicals from the chemicals assigned (A) a neurotoxic mode of action (32 chemicals total) and (B) no categorization (90 chemicals total) through the Insecticide Resistance Action Committee. CAS numbers of test chemicals are included in brackets. Online supplementary material (Figure S5) displays significance levels of species variability across the whole chemical set.
Results and discussion
Overview of the data and variability across taxonomic groups
The analysis was limited to 805 chemicals for which acute toxicity data on both fish and daphnids (related to OECD test guideline 203 and 202, respectively) were contained in the EnviroTox database, and for which MoA classifications were available. Online supplementary material Tables S1 and S2 give an overview of effect types, test statistics, and different species included in the data set. As an initial step, we characterized this data set by visualizing the distribution of median EC50/LC50 values across the two taxonomic groups fish and daphnids (see online supplementary material Figure S2, A) as well as the interspecies variability of the data for the two taxonomic groups (see online supplementary material Figure S2, B and C; Figure S2, D contains a similar characterization of amphibian toxicity data). The range of median values is similar across fish and daphnids with slightly lower values for the latter, indicating a generally higher sensitivity. This is also indicated by the quartiles (vertical lines in the plots A–C in online supplementary material Figure S2) being lower as well (i.e., toward lower effective concentrations and, hence, higher sensitivity) compared to the quartiles in the fish distribution.
Interspecies variability for the selected compounds seems to be influenced by data availability, as can be seen for nonsummarized data points (i.e., raw data points, not median values) in online supplementary material Figure S2, B–D. The fish species with the most data available after prefiltering (indicated by the numbers given in the y-axis labels in online supplementary material Figure S2B–D) appear to be approximately normally distributed and largely overlap in their range, whereas species with fewer tested chemicals seem to have more unevenly distributed LC50 values (see online supplementary material Figure S2, B). This also holds true for the daphnid data (see online supplementary material Figure S2, C), albeit this subset is, for the most part, dominated by a single species (Daphnia magna). Daphnia carinata and Daphnia laevis generally appear to be more sensitive or have been tested on more toxic chemicals with their distributions centered on the left of the scale (lower EC50 values, higher toxicity). Contrary to that, data for Daphnia longispina and Daphnia obtusa are less evenly distributed, which likely is related to the low number of data points (5 and 6, respectively). Accordingly, observations based on such low numbers of data points should be taken cum grano salis. There is a poor correlation between the number of data points per chemical and the coefficient of variation (CV; R2 of 0.11 and 0.29 for daphnids and fish, respectively; see online supplementary material Figure S3). The CV is the relative standard deviation, that is, the standard deviation divided by the mean, expressed as percentage. Here, it supports the notion that the CV does not necessarily increase with more data points, but it becomes more robust (see online supplementary material Figure S3).
From the overlap of the two data sets, the ratios of median daphnid effect values divided by the median fish effect values were calculated as a marker of sensitivity differences between the two taxonomic groups (Figure 2; see online supplementary material Figure S2E). At this point, the analysis is agnostic about the associated MoA. The distribution is skewed toward the left of a ratio of one, that is, higher sensitivity for daphnids with only a small fraction of ratios indicating a higher sensitivity of fish by a factor of 10 or 100 compared to daphnids (orange and red area, respectively). A similar pattern emerges when corresponding median fish LC50 and daphnids EC50 values are plotted against each other to visualize the correlation differently (see online supplementary material Figure S4). Here, it can be observed that over all 805 chemicals, daphnids are the more sensitive taxonomic group and only for approximately 4% of the chemicals are fish more than tenfold more sensitive than daphnids.
Biological data variability
Because the analysis is based on ecotoxicological in vivo data, it is important to acknowledge the high biological variability associated with such information. Here, we restrict to chemicals that are covered by the IRAC categorization scheme; 32 and 90 chemicals had been tested on daphnids and at least on four fish species and had a neurotoxicity MoA or no categorization in the IRAC categorization scheme, respectively. We use IRAC as the reference categorization here because we expect that knowing the MoA of a pesticide constituent gives the highest reliability of a correct categorization as opposed to, for example, a purely predicted MoA. This assumption comes with the caveat that IRAC contains only 86 active compounds and is biased toward insecticides. To keep the two groups comparable, we randomly subsampled 20 chemicals that have been tested on at least four different fish species from each of those two groups. We visually oppose them in Figure 3, displaying the range of LC50. Even though data were selected for only one experimental duration and limited as closely as possible to OECD test guideline 203 conformity and excluded LC50 values that were more than fivefold above the chemicals water solubility limits, we still observed high variability of the experimental outcomes. In some cases, the LC50 values for a single species and chemical span several orders of magnitude. Visually, this is more distinct for the neurotoxic compounds in Figure 3A. Although this LC50 variability can be explained only partially by interspecies differences, the proportion of compounds for which this is the case is much higher for neurotoxic compounds versus nonclassified ones (see results from the Kruskal-Wallis rank sum test in online supplementary material Figure S5).
A figure similar to the boxplot in Figure 3 for daphnids (at least two tested species) is included in the online supplementary material as Figure S6. There, the data mostly come from studies using D. magna as the primary test species (see online supplementary material Table S2), which hinders the comparison of inter- and intraspecies variability.
Additionally, we visualize the CV, that is, the standard deviation normalized against the mean, across the broad categories of non-neurotoxic, no categorization, and neurotoxic across all chemicals and categorization schemes (Figure 4, A). This is done for fish data, but limited to the chemicals that have been tested on both fish and daphnids. Notably, because this is based on all categorization schemes, chemicals can be assigned to several categories at once (see online supplementary material Figure S1), leading to duplicate data points. The median of the CVs for neurotoxic chemicals is approximately 100%, which is approximately twice the median of the CVs of other chemicals. However, the lower quartile of the neurotoxic CVs is overlapping with the upper quartile of the other chemicals CVs. As a refinement, Figure 4B focuses only on IRAC-labeled chemicals. Here, the median of the CVs is also higher in the group of chemicals that have been assigned a neurotoxic MoA compared to those without an IRAC classification. In conclusion, toxicity outcomes in fish appear more variable for chemicals with an assigned neurotoxic MoA. Consequently, when only a few (pseudo-)replicates are available, this higher uncertainty about the “true” median LC50 could influence the Rdf values derived from such data, potentially leading to very high or very low Rdf values. An analysis with similar results is provided for daphnids (see online supplementary material Figure S6, A).
Coefficient of variation across fish for all the chemicals in the dataset that also have been tested on daphnids. (A) Assigned to the broad classes of non-neurotoxic, no categorization, and neurotoxic across all five categorization schemes and (B) only the chemicals with an assigned mode of action in the Insecticide Resistance Action Committee categorization scheme.
Neurotoxicity classification schemes
Because the main goal of this study was to analyze whether daphnid acute data could be protective of fish acute mortality for neurotoxic compounds, it was important to establish a ground truth for the MoA of the chemicals in the data set. In other terms, it was necessary to introduce knowledge on which chemicals are considered to cause neurotoxicity. For this, six classification schemes were selected that, for the most part, contained positive (i.e., potentially neurotoxic) as well as non-neurotoxic compounds (Figure 1). For the sake of consistency, subcategories for neurotoxicity were retained where available (some category names were adapted for clarity), but data for chemicals that either were labeled as something other than neurotoxic (“Non-Neurotoxic”) or were not labeled in the specific scheme at all (“No Categorization”) were summarized separately.
Within this analysis, we considered that chemical median ratios between daphnids EC50 and fish LC50 values of less than 10 (i.e., fish are not more than 10 times more sensitive than daphnids) are in agreement with the hypothesis that daphnid EC50 values are sufficiently protective in the absence of fish LC50 values. This consideration is based on several observations: (1) Chemical-specific LC50 and EC50 values for fish and daphnids generated according to OECD test guideline may range several orders of magnitude with a typical CV of 100% for chemicals with neurotoxic MoA, which is approximately twice as high as for non-neurotoxic chemicals (see section on Biological data variability). Irrespective of MoA, this observation is in line with the findings of Hrovat et al. (2009), who found variability of up to six orders of magnitude for fish LC50 derived from the ECOTOX database, albeit without selecting for specific experimental setups. Another analysis by Braunbeck et al. (2020) applied more stringent filtering to manually curated fish acute toxicity test data, which resulted in a smaller data set of 58 chemicals, with only two chemicals with neurotoxic MoAs in accordance with our classifications. They kindly made the raw data available to us, enabling a comparative analysis with consistent data handling and metric calculation. We found that only 18 of the chemicals in their data set are also part of the current analysis, albeit none of them are classified as neurotoxic. Furthermore, in that analysis, essentially for non-neurotoxic chemicals, CVs ranged up to 150% with a median of approximately 20% (see online supplementary material Figure S6B). Fischer et al. (2019; see Figure 5 therein) support the particularly high fish interspecies variability for the neurotoxic pesticide malathion (>4 orders of magnitude). Scholz et al. (2016) reported variability of fish LC50 values of more than 50-fold for chemicals that are toxic mainly after bioactivation. Paparella et al. (2021) further summarized the current knowledge on variability of data related to OECD test guideline 203. (2) Within current regulatory practice, safety factors of 1,000 are applied to acute aquatic toxicity data for the derivation of predicted no effect concentrations (PNEC). This approach aims to arrive at protective values for the environment based on acute laboratory experiments. However, this, by default, assumes high uncertainty for extrapolating from acute EC50 or LC50 values to safe environmental concentrations. In this context, a variability of experimental data by a factor of 10 may be considered insignificant relative to the overall uncertainty. (3) United Nations GHS classification for aquatic toxicity stratifies acute toxicity into tenfold ranges with the category boundaries <1 mg/L, 1–10 mg/L, and >100 mg/L, for categories 1, 2, and 3, respectively. Additionally, a tenfold potency stratification is used for the derivation of M-factors for the purpose of toxicity-based weighting of components for mixture assessment. Even though these ranges may be too granular to adequately account for the observed biological variability, they imply that deviations of experimental data by a factor of 10 may be considered acceptable for regulatory purposes.
![(A) Heatmap/table of median ratios (effect concentration [EC50] and lethal concentration [LC50]) across the classification schemes and subclasses. Median ratios are divided into the groups ≤1, between >1 and ≤10, between >10 and ≤100, and >100. Darker shading of cells correlates to higher percentages. (B) Violin plot of the ratio distributions across the different categorization schemes and subcategories. Gray dots are not relevant to neurotoxicity. Colorful dots indicate the range of Rdf ratios for neurotoxicity-related endpoints (green ≤10; orange 10–100). Blue triangles indicate chemicals that are part of the ASTER subcategory “Neurotoxicant: Cyclodiene-Type”). Note. Op = organophosphate; AChE = acetylcholine esterase; Clc = xhloride channel; nAChr = nicotinic acetylcholine receptor.](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/etc/PAP/10.1093_etojnl_vgaf014/4/m_vgaf014f5.jpeg?Expires=1747873057&Signature=HKJj4JeKurkmH70DBNdADXMY1PXP~4qV-4V7tH1OwwxFOc23-f7lSY9MwA0~0~Od0YcdckYc6Qi4T1ZqNgDLVykiOIZ0cCHEMZW6JLhjVr8YQKrd84cE0q2qXc25~JfBty1QXKIUM0E4BlDTUODiJmD5Kh9L~cQeNb3Lj0q~HIXfDRa9UrPPWewoIZFhsHIu08FlCT3ISyV3gK9b4NjGIu9VOtrd8HvuKPaefr1zN3ZI9N~1EY5sGvr~pEMqlNlpy21NkcYCd7crjk3o1~VBim0E-aZiyoyoRcOB9jFTX6svO-99fF0DjQmt1kMLRo~k5mR7sw7sKxWA2syD3UZOYw__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
(A) Heatmap/table of median ratios (effect concentration [EC50] and lethal concentration [LC50]) across the classification schemes and subclasses. Median ratios are divided into the groups ≤1, between >1 and ≤10, between >10 and ≤100, and >100. Darker shading of cells correlates to higher percentages. (B) Violin plot of the ratio distributions across the different categorization schemes and subcategories. Gray dots are not relevant to neurotoxicity. Colorful dots indicate the range of Rdf ratios for neurotoxicity-related endpoints (green ≤10; orange 10–100). Blue triangles indicate chemicals that are part of the ASTER subcategory “Neurotoxicant: Cyclodiene-Type”). Note. Op = organophosphate; AChE = acetylcholine esterase; Clc = xhloride channel; nAChr = nicotinic acetylcholine receptor.
Generally, the choice of a threshold for what can still be considered a safe Rdf is arbitrary. Accordingly, we arrived at, what we consider, a conservative value of 10 based on previous works and what is deemed acceptable within the regulatory context.
Median ratios across MoA categories
Generally, the emerging patterns indicate that for most neurotoxic compounds, daphnids are more sensitive than fish compared to non-neurotoxic compounds (Figure 5). Exceptions are the “Cyclodiene-type” neurotoxins in ASTER and the analogous “GABA-gated Clc Blockers Cyclodiene Organochlorines” in IRAC. The cyclodiene-type category in ASTER contains four chemicals with an Rdf above 10, namely, dieldrin, endosulfan, endosulfan sulfate, and endrin. These chemicals also appear within the EFSA, Kramer, and TEST classifications, which do not specify cyclodiene-type neurotoxicity as such (blue triangles visualize these chemicals across the categories in Figure 5B). In addition, two pyrethroids, that is, flucythrinate and resmethrin, have a ratio slightly above 10 (13.8 and 19.7, respectively) within the ASTER pyrethroid category as well as within the other classification systems.
Out of the 32 compounds with ratios ≥10, seven were categorized as neurotoxic in at least one of the classification schemes (five out of these seven chemicals were labeled as neurotoxic by at least three out of five classification schemes). The observation of most compounds being labeled as neurotoxic in several classification schemes strengthens confidence that the approach applied in this analysis is highly reliable. All five cyclodienes ratios with >10 (dieldrin, endosulfan, endosulfan sulfate, endrin, and 1,2,3,4,5,6-hexachlorocyclohexane [lindane]) are covered in Annex A of the Stockholm Convention. The two pyrethroids (flucythrinate and resmethrin) are excluded from being used as ingredients in plant protection products through the European Commission Regulation (EC) No. 2076/2002 (European Commission, 2002). Hence, all compounds labeled as neurotoxic that have a median ratio higher than 10 are no longer relevant to the market. This is true for the European market for the REACH-regulated chemicals and globally for those covered by the Stockholm Convention ratified by 186 countries (https://treaties.un.org/Pages/ViewDetails.aspx?src=IND&mtdsg_no=XXVII-15&chapter=27&clang=_en [last accessed November 13, 2024]; Table 1). Resmethrin products are no longer sold or distributed across the United States since 2015, and the use of any remaining product is highly limited (https://www.epa.gov/mosquitocontrol/permethrin-resmethrin-d-phenothrin-sumithrinr-synthetic-pyrethroids-mosquito [last accessed November 13, 2024]). It needs to be stated, however, that the CVs for fish LC50 and daphnids EC50 values are particularly high for these chemicals (Table 1), such that the daphnids/fish median ratio contains considerable uncertainty.
Overview of the seven chemicals labeled with a neurotoxic mode of action in at least one of the classification schemes and with a median ratio above 10. Five out of seven compounds are covered in the “Elimination” Annex of the Stockholm Convention, whereas the other two are prohibited from use in plant protection products through European Commission (EC) Regulation No. 2076/2002.
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | CV fish [n] | CV daphnids [n] | Covered by Stockholm Convention Annex A | Covered by Regulation (EC) No. 2076/2002 |
|---|---|---|---|---|---|---|
| Dieldrin | 60-57-1 | 23.7 | 206.1 [73] | 28.8 [8] | Yes | No |
| Endosulfan | 115-29-7 | 200 | 187.4 [115] | 51.5 [19] | Yes | No |
| Endosulfan sulfate | 1031-07-8 | 657.1 | NA [1] | NA [1] | Yes | No |
| Endrin | 72-20-8 | 53.3 | 368.7 [80] | 107.7 [12] | Yes | No |
| Flucythrinate | 70124-77-5 | 13.8 | 72.1 [15] | NA [1] | No | Yes |
| Lindane | 58-89-9 | 16 | 516.6 [87] | 108.6 [16] | Yes | No |
| Resmethrin | 10453-86-8 | 19.7 | 74.9 [51] | 112.0 [4] | No | Yes |
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | CV fish [n] | CV daphnids [n] | Covered by Stockholm Convention Annex A | Covered by Regulation (EC) No. 2076/2002 |
|---|---|---|---|---|---|---|
| Dieldrin | 60-57-1 | 23.7 | 206.1 [73] | 28.8 [8] | Yes | No |
| Endosulfan | 115-29-7 | 200 | 187.4 [115] | 51.5 [19] | Yes | No |
| Endosulfan sulfate | 1031-07-8 | 657.1 | NA [1] | NA [1] | Yes | No |
| Endrin | 72-20-8 | 53.3 | 368.7 [80] | 107.7 [12] | Yes | No |
| Flucythrinate | 70124-77-5 | 13.8 | 72.1 [15] | NA [1] | No | Yes |
| Lindane | 58-89-9 | 16 | 516.6 [87] | 108.6 [16] | Yes | No |
| Resmethrin | 10453-86-8 | 19.7 | 74.9 [51] | 112.0 [4] | No | Yes |
Note. CV = coefficient of variation.
Overview of the seven chemicals labeled with a neurotoxic mode of action in at least one of the classification schemes and with a median ratio above 10. Five out of seven compounds are covered in the “Elimination” Annex of the Stockholm Convention, whereas the other two are prohibited from use in plant protection products through European Commission (EC) Regulation No. 2076/2002.
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | CV fish [n] | CV daphnids [n] | Covered by Stockholm Convention Annex A | Covered by Regulation (EC) No. 2076/2002 |
|---|---|---|---|---|---|---|
| Dieldrin | 60-57-1 | 23.7 | 206.1 [73] | 28.8 [8] | Yes | No |
| Endosulfan | 115-29-7 | 200 | 187.4 [115] | 51.5 [19] | Yes | No |
| Endosulfan sulfate | 1031-07-8 | 657.1 | NA [1] | NA [1] | Yes | No |
| Endrin | 72-20-8 | 53.3 | 368.7 [80] | 107.7 [12] | Yes | No |
| Flucythrinate | 70124-77-5 | 13.8 | 72.1 [15] | NA [1] | No | Yes |
| Lindane | 58-89-9 | 16 | 516.6 [87] | 108.6 [16] | Yes | No |
| Resmethrin | 10453-86-8 | 19.7 | 74.9 [51] | 112.0 [4] | No | Yes |
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | CV fish [n] | CV daphnids [n] | Covered by Stockholm Convention Annex A | Covered by Regulation (EC) No. 2076/2002 |
|---|---|---|---|---|---|---|
| Dieldrin | 60-57-1 | 23.7 | 206.1 [73] | 28.8 [8] | Yes | No |
| Endosulfan | 115-29-7 | 200 | 187.4 [115] | 51.5 [19] | Yes | No |
| Endosulfan sulfate | 1031-07-8 | 657.1 | NA [1] | NA [1] | Yes | No |
| Endrin | 72-20-8 | 53.3 | 368.7 [80] | 107.7 [12] | Yes | No |
| Flucythrinate | 70124-77-5 | 13.8 | 72.1 [15] | NA [1] | No | Yes |
| Lindane | 58-89-9 | 16 | 516.6 [87] | 108.6 [16] | Yes | No |
| Resmethrin | 10453-86-8 | 19.7 | 74.9 [51] | 112.0 [4] | No | Yes |
Note. CV = coefficient of variation.
Moreover, beyond neurotoxicity, just 3.9% of the 805 chemicals for which both fish and daphnid data were available showed a median Rdf above 10 (see online supplementary material Table S3), which supports that within the current overall chemical universe, fish LC50 values rarely drive PNEC derivation and classification. In contrast to earlier published work, our analysis accounted for data variability via the use of taxonomic median values, considered an Rdf below 10 as sufficiently protective, engaged a high number of compounds (805), and applied clear and stringent data filtering criteria. A tabular overview on differences to earlier analyses is provided in the online supplementary material (see Table S5). Importantly, the 3%–4% of non-neurotoxic chemicals with ratios higher than 10 are of low concern if alternative methods to the fish test are used (test guideline 236 and test guideline 249) because, based on current knowledge, they are protective for chemicals with non-neurotoxic mechanisms.
In our presentation, the comparison of daphnids and fish toxicity data is based on the calculation of the chemicals’ median LC50 or EC50 values averaging all LC50 or EC50 values within the trophic level. It is acknowledged that different approaches also may be used, that is, grouping by chemical and species or no grouping and matching all fish LC50 values with all daphnids EC50 values. The arguments and results for these alternative approaches are provided in the online supplementary material under “Comparison of different data matching approaches.” Importantly, the overall pattern emerging from these other approaches is similar (see online supplementary material Figures S9 and S10, Table S5).
Moreover, the Rdf was calculated from the median of the pseudo-replicate EC50s for each compound in daphnids relative to fish. Alternatively, the Rdf could be calculated from the geometric mean (GM) of the pseudo-replicates. The GM would be more robust against large distances between the EC50s, whereas the median is more robust against extreme (outlier) EC50s. In the case of an “ideal” log-normal distribution of the EC50 values, the median would be identical to the GM. We performed the analysis comparatively with both the median and the GM and found that the differences between the outcomes are negligible. Figures using the GM analogous to those presented with the median in the main article text are given in the online supplementary material (see Figures S11–S14).
Chemicals underestimated in the fish cell acute toxicity assay and zFET
Tanneberger et al. (2013) analyzed the correlation between EC50 from the fish cell acute toxicity assay and fish acute LC50 values. Out of the 35 compounds investigated in that study, five compounds had fish LC50 values at least 10 times lower than the EC50 derived from the in vitro experiments: permethrin (190-fold), caffeine (18-fold), lindane (63-fold; all three neurotoxicants), allyl alcohol (2,700-fold), and 4-fluoroaniline (12-fold). Accordingly, the authors of that study, based on which the OECD test guideline 249 (OECD, 2021) was later developed, found these compounds to be underpredicted by the RTgill-W1 cell line assay. Those compounds were specifically sought out in our data set (where available) and are summarized in Table 2. Because data were not always available at the desired time point of 48 hr for daphnids, the filter criteria had to be adapted to be less stringent. Expanding toward daphnid experiments performed at 24 hr (in addition to 48 hr) and LC50/mortality data in addition to EC50 and immobilization allowed four of the chemicals of interest (permethrin, lindane, caffeine, and allyl alcohol) to be included. No daphnid data were available for 4-fluoroaniline.
Outlier chemicals that Tanneberger et al. (2013) found to be underpredicted by fish cell acute toxicity assays compared to fish and their median lethal/effect concentration (LC50/EC50) in fish and daphnids in mg/L, in our analysis as well as median ratios, coefficient of variation (CV), and number of data points [n].
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | Median fish | CV fish [n] | Median daphnids | CV daphnids [n] | Rdaph/RTgill W1 | EC50 RTgill W1 cell line [p05-p95] | CV RTgill W1 cell line |
|---|---|---|---|---|---|---|---|---|---|
| Allyl alcohol | 107-18-6 | 0.4 | 0.59 | 57.3 [9] | 0.25 | NA [1] | 0.0003 | 812 [622-1060] | ca. 30%a |
| Permethrin | 52645-53-1 | 0.2 | 0.01 | 87.7 [109] | 0.001 | 197.8 [42] | 0.0003 | 3.76 [2.98-4.76] | ca. 30%a |
| Caffeine | 58-08-2 | 2.8 | 151 | NA [1] | 422 | 87.8 [2] | 0.1591 | 2652 [2357-2984] | ca. 30%a |
| Lindane | 58-89-9 | 20.8 | 0.09 | 516.6 [87] | 1.79 | 116.6 [40] | 0.2594 | 6.9 [nd] | ca. 30%a |
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | Median fish | CV fish [n] | Median daphnids | CV daphnids [n] | Rdaph/RTgill W1 | EC50 RTgill W1 cell line [p05-p95] | CV RTgill W1 cell line |
|---|---|---|---|---|---|---|---|---|---|
| Allyl alcohol | 107-18-6 | 0.4 | 0.59 | 57.3 [9] | 0.25 | NA [1] | 0.0003 | 812 [622-1060] | ca. 30%a |
| Permethrin | 52645-53-1 | 0.2 | 0.01 | 87.7 [109] | 0.001 | 197.8 [42] | 0.0003 | 3.76 [2.98-4.76] | ca. 30%a |
| Caffeine | 58-08-2 | 2.8 | 151 | NA [1] | 422 | 87.8 [2] | 0.1591 | 2652 [2357-2984] | ca. 30%a |
| Lindane | 58-89-9 | 20.8 | 0.09 | 516.6 [87] | 1.79 | 116.6 [40] | 0.2594 | 6.9 [nd] | ca. 30%a |
Not compound specific but generic CV for interlaboratory replicate values for RTgill W1 cell line assay as provided in Organisation for Economic Co-operation and Development (OECD) 2021 (OECD Series of Testing and Assessment No. 334).
Note. NA = not applicable; nd = not determined.
Outlier chemicals that Tanneberger et al. (2013) found to be underpredicted by fish cell acute toxicity assays compared to fish and their median lethal/effect concentration (LC50/EC50) in fish and daphnids in mg/L, in our analysis as well as median ratios, coefficient of variation (CV), and number of data points [n].
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | Median fish | CV fish [n] | Median daphnids | CV daphnids [n] | Rdaph/RTgill W1 | EC50 RTgill W1 cell line [p05-p95] | CV RTgill W1 cell line |
|---|---|---|---|---|---|---|---|---|---|
| Allyl alcohol | 107-18-6 | 0.4 | 0.59 | 57.3 [9] | 0.25 | NA [1] | 0.0003 | 812 [622-1060] | ca. 30%a |
| Permethrin | 52645-53-1 | 0.2 | 0.01 | 87.7 [109] | 0.001 | 197.8 [42] | 0.0003 | 3.76 [2.98-4.76] | ca. 30%a |
| Caffeine | 58-08-2 | 2.8 | 151 | NA [1] | 422 | 87.8 [2] | 0.1591 | 2652 [2357-2984] | ca. 30%a |
| Lindane | 58-89-9 | 20.8 | 0.09 | 516.6 [87] | 1.79 | 116.6 [40] | 0.2594 | 6.9 [nd] | ca. 30%a |
| Chemical name | CAS number | Rdf Daphnid/fish median ratio | Median fish | CV fish [n] | Median daphnids | CV daphnids [n] | Rdaph/RTgill W1 | EC50 RTgill W1 cell line [p05-p95] | CV RTgill W1 cell line |
|---|---|---|---|---|---|---|---|---|---|
| Allyl alcohol | 107-18-6 | 0.4 | 0.59 | 57.3 [9] | 0.25 | NA [1] | 0.0003 | 812 [622-1060] | ca. 30%a |
| Permethrin | 52645-53-1 | 0.2 | 0.01 | 87.7 [109] | 0.001 | 197.8 [42] | 0.0003 | 3.76 [2.98-4.76] | ca. 30%a |
| Caffeine | 58-08-2 | 2.8 | 151 | NA [1] | 422 | 87.8 [2] | 0.1591 | 2652 [2357-2984] | ca. 30%a |
| Lindane | 58-89-9 | 20.8 | 0.09 | 516.6 [87] | 1.79 | 116.6 [40] | 0.2594 | 6.9 [nd] | ca. 30%a |
Not compound specific but generic CV for interlaboratory replicate values for RTgill W1 cell line assay as provided in Organisation for Economic Co-operation and Development (OECD) 2021 (OECD Series of Testing and Assessment No. 334).
Note. NA = not applicable; nd = not determined.
Within this analysis, the Rdf are in an unproblematic range (<3) for allyl alcohol, permethrin, and caffeine, whereas for lindane the ratio of approximately 21 is only slightly above the ratio of 10 and needs to be contextualized with its fish CV of 517%, rendering this ratio uncertain.
Importantly, this analysis puts allyl alcohol in a different light. Rather than being an example for a compound considered “difficult” for alternative testing schemes due to its presumed toxification via biotransformation in fish, it is here considered an example of outliers in the performance of alternative methods that may be well covered with the obligatory daphnid test.
In summary, the toxicity of the outliers not sufficiently covered by the RTgill-W1 cell line assay is covered by daphnid data, thus reducing an uncertainty in the toolbox of bioassay-based NAMs for fish acute toxicity.
In a similar analysis, Scholz et al. (2016) identified 17 compounds for which juvenile fish were more than tenfold more sensitive than zebrafish embryos. Yet, daphnids show a similar or higher sensitivity compared to juvenile fish for all these compounds except for cyazofamid, which appears to be 86-fold more toxic in fish than daphnids but does not have a clear neurotoxic mode of action (see online supplementary material Table S6).
Rawlings et al. (2019) arrive at a similar conclusion, supporting that the tenfold potency stratification as used for acute aquatic UN GHS classification is not affected. They can still be considered protective if, within the standard acute fish, daphnids, algae data requirements, the acute juvenile fish LC50 values were replaced by zFET EC50 values. However, due to limited algae data availability, that study included only 82 compounds (vs. 805 in our analysis) and very few neurotoxic chemicals (5 vs. 141 in our analysis). Moreover, our analysis provides Rdf values and a more comprehensive neurotoxic mode of action annotation.
Complementing fish acute toxicity alternative methods
Given the currently available data, test guideline 236 and 249 are considered to be of limited use for chemicals with a neurotoxic MoA. In the still only recently (in 2021) instated test guideline 249, this is acknowledged to concern chemicals specifically acting on ion channels or receptors typical for brain tissues (OECD, 2021). Given the other underpredicted chemical, allyl alcohol, we argue that the assay needs to be conducted on both the original chemical as well as relevant transformation products, should indications arise that those transformation products contribute to or drive toxicity. This is based on the observation that the more toxic transformation product acrolein was predicted well through the RTgill-W1 assay as well as the zFET assay.
Complementarily to that, our analysis demonstrates that daphnid acute toxicity data from studies conducted according to OECD test guideline 202 can be considered protective for fish for such compounds due to the higher sensitivity of daphnids. Exceptions, as we have shown, are seven out of 141 compounds with neurotoxic mechanisms, mainly belonging to a distinct group of chemicals. They were identified and predicted to be neurotoxic via the ASTER tool and the IRAC database, showcasing how computational tools and biological assays can further complement each other. The chemicals of concern are included either in the Annex A (“Elimination”) of the Stockholm Convention or the EC regulation No. 2076/2002 and, accordingly, not relevant for markets in the European Union and in the 186 countries that ratified the Stockholm Convention.
Limitations and outlook
Data were filtered for test guideline conformity in terms of test duration, endpoints, and species while excluding EC50 values more than fivefold above water solubility. We acknowledge that the EnviroTox database does not allow filtering for nominal versus measured chemical concentrations and differences in exposure designs (e.g., flow-through versus static or semi-static exposure). Although this may have affected data variability to some extent, it is unlikely to have affected the Rdf. Likewise, as we discussed earlier, different approaches to comparing the daphnids and fish data are possible (e.g., using the geometric mean instead of the median), but the general outcome of the analysis proved to be similar, substantiating our trust in the chosen approach (see online supplementary material Figures S11–S14).
The determination of the ground truth for neurotoxic MoAs (i.e., the way in which chemicals considered to cause neurotoxicity were identified) presents an inherent bias in this analysis (see the section on Neurotoxicity classification schemes). Therefore, a broad range of classification schemes was selected, spanning both very specific pesticide MoAs (e.g., IRAC) and a broad selection of chemicals relevant for freshwater monitoring (e.g., the data from Kramer et al. [2024]). The fact that the neurotoxic chemicals with a median ratio more than 10 were for the most part flagged by multiple classification schemes shows that this redundancy mechanism added robustness to the analysis and demonstrates that its findings are indeed not biased. However, it is possible and likely that new chemicals that are more toxic to fish than to daphnids may emerge in the future, probably belonging to completely new chemical classes. Therefore, this analysis serves as an example that the current regulatory framework can benefit from the modular adoption of alternative methods. For this, a concerted effort is required toward full integration of such methods into a new framework that accounts for the strengths and weaknesses of each alternative method as opposed to expecting a single method to serve as a full one-to-one replacement of a traditional animal test. Even though our analysis focuses on data derived from biological toxicity assays, we acknowledge the high potential of in silico methods, such as QSARs with and without the use of machine learning (Gasser et al., 2024; Kleinstreuer & Hartung, 2024; Muratov et al., 2020). They may serve as the uniting puzzle piece that enables the integration of different kinds of data (physicochemical, in vivo, in vitro, etc.) into models that ultimately can allow for predictions across a broad space of endpoints and biology, given the availability of sufficiently high-quality data.
However, this work already supports a fully NAM-based IATA for fish acute toxicity including the zFET, the RTgill-W1 cell line assay, and computational approaches, in addition to daphnids and algae testing, supporting a full replacement of the acute fish toxicity test.
Finally, considering the observed variability of acute fish and daphnids data generated according to OECD test guideline, it is scientifically warranted to discuss how the UN GHS approach could be improved. The current UN GHS approach foresees a deterministic classification of chemicals into three acute aquatic toxicity category or no classification, depending on the lowest LC/EC50 value from fish, daphnids, and algae being below 1 mg/L, between 1 and 100 mg/L, or above 100 mg/L. Moreover, tenfold ranges are used to derive compound specific M-factors for the purpose of mixture assessment. However, considering the confirmation within the present analysis that LC/EC50 values from fish and daphnids may span several orders of magnitude, an approach providing concentration ranges and categorization probabilities rather than point estimates may be more adequate.
Toward next-generation hazard and risk assessment
In the context of a chemical hazard assessment that is based on the 3Rs principle (replacement, reduction, and refinement of animal tests), this analysis showcases how the integration of alternative systems can benefit the replacement of vertebrate species. Because daphnid toxicity data is currently a standard regulatory data requirement, testing on daphnids is conducted irrespective of whether fish are tested as well. Therefore, focusing on a combination of alternative methods in conjunction with the anyway created daphnid data leads to a net reduction in testing on animals, because fish tests can be omitted. Interestingly, acute daphnid EC50 values appear also protective for acute amphibian EC50 values, irrespective of the underlying MoA and as far as available within the EnviroTox database (see online supplementary material Figure S8). A next-generation risk assessment (NGRA) framework that is fit for the challenge of an ever-growing chemical landscape and concerned with environmental health foremost needs to be considerate of the protection goals. Achieving high capacities of chemical testing while ensuring environmental protection will likely be feasible sooner through the combination of alternative methods with traditional daphnid testing. This way of considering strengths and shortcomings of individual methods creates a framework that covers a range of outcomes rather than aiming at replacing highly variable acute fish toxicity testing one-to-one.
Nonetheless, we consider the use of daphnids as safeguards to be an intermediate solution that does not negate the need for the further development and adoption of approaches that are completely free of animal use. Here, the already more advanced fully nonanimal-based NGRA concepts for human health may stimulate the evolution of environmental NGRA concepts (Langan et al., 2024). Referring to the approach taken here for analyzing mechanism of action, this work presents also an important stepping-stone toward such a broader, fully animal-free, IATA for environmental safety.
Supplementary material
Supplementary material is available online at Environmental Toxicology and Chemistry.
Data availability
Data, associated metadata, and calculation tools are available through the online supplementary files, and the R-code will be made available via GitHub: https://github.com/schuerc/fish_daphnia_neurotox_public.
Author contributions
Christoph Schür (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft), Martin Paparella (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft), Christopher Faßbender (Conceptualization, Writing – original draft), Gilly Stoddart (Conceptualization, Writing – review & editing), Marco Baity Jesi (Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing), and Kristin Schirmer (Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing)
Funding
This work was funded through the Swiss Data Science Center (SDSC) grant “Enhancing Toxicological Testing through Machine Learning” (Project No. C20-04) and partly carried out in the framework of the European Partnership for the Assessment of Risks from Chemicals (PARC) and has received funding from the European Union’s Horizon Europe research and innovation program under Grant Agreement No. 101057014. The work of M.P. at the Medical University is co-financed via PARC and the Austrian Federal Ministry for Climate Action, Environment, Energy, Mobility, Innovation and Technology, Department V/5—Chemicals Policy and Biocides.
Conflicts of interest
The authors declare no conflicts of interest.
Acknowledgments
We thank T. Braunbeck (University of Heidelberg, Germany) for providing us with the raw data from his analysis of fish median lethal concentration variability (Braunbeck et al., 2020), which allowed us to compare our analysis.
References
Author notes
C.S. and M.P. shared the majority of the work and are ranked according to their contributions.
