Development of algal interspecies correlation estimation models for chemical hazard assessment

Evaluation of interspecies correlation estimation models to increase taxonomic diversity while reducing reliance on animal testing for chemicals evaluated under the Toxic Substances Control Act

The U.S. Environmental Protection Agency is committed to the implementation of new approach methodologies (NAMs) to enhance the scientific basis for chemical hazard assessments. Chemical evaluations under the Toxic Substance Control Act (TSCA) are often conducted with limited test data and are well suited for NAMs applications. Interspecies correlation estimation (ICE) models are log-linear least squares regressions of the sensitivity between two species that estimate the acute toxicity of an untested species from the sensitivity of a surrogate. Interspecies correlation estimation models have been developed from and validated for diverse chemical modes of action, but their application in TSCA chemical assessments has not been previously evaluated. We use ICE models and a dataset of measured acute values for five chemicals, increasing the taxonomic diversity from which concentrations of concern (CoCs) are derived. Concentrations of concern were developed using approaches typically applied in TSCA risk evaluations, including application of assessment factors to the most sensitive species and the development of species sensitivity distributions where a minimum of eight species are represented by measured data. These CoCs were compared with those derived from datasets supplemented with ICE-predicted values, as well as comparing ICE predicted species mean acute values (SMAVs) to their respective measured values. Interspecies correlation estimation models predicted SMAVs within a factor of 5 and 10 for 87% and 92% of measured values, respectively. The CoCs developed from measured data only and data supplemented with ICE predicted toxicity were generally within five-fold, showing comparable protection. The taxonomic diversity in the ICE supplemented dataset was substantially higher than the measured data for species sensitivity distributions, providing a data-driven way of reducing uncertainty and potentially reducing the need for assessment factors. Interspecies correlation estimation models show promise as a NAM to improve the taxonomic representation included in chemical evaluations under TSCA.

Uncertainty analysis and updated user guidance for interspecies correlation estimation models and low toxicity compounds

Interspecies correlation estimation (ICE) models are log-linear relationships of acute sensitivity between two species that estimate the sensitivity of an untested species from the known sensitivity of a surrogate. As ICE model use increases globally, additional user guidance is required to ensure consistent use across chemicals and applications. The present study expands ICE uncertainty analyses and user guidance with a focus on low toxicity compounds whose acute values (i.e., reported as mg/L) can be greater than those used to develop a model. In these cases, surrogate values may be outside the ICE model domain and require additional extrapolations to predict acute toxicity. We use the extensive, standardized acute toxicity database underlying ICE models to broadly summarize inter-test variability of acute toxicity data as a measure by which model prediction accuracy can be evaluated. Using the data and models found on the USEPA Web-ICE (www3.epa.gov/webice), we created a set of "truncated" models from data corresponding to the lower 75th percentile of surrogate toxicity. We predicted toxicity for chemicals in the upper 25th percentile as both μg/L beyond the model domain and converted to mg/L (i.e., "scaled" value) and compared these predictions with those from cross-validation of whole ICE models and to the measured value. For ICE models with slopes in the range 0.66-1.33, prediction accuracy of scaled values did not differ from the accuracy of the models when data were entered as μg/L within or beyond the model domain. An uncertainty analysis of ICE confidence intervals was conducted and an interval range of two orders of magnitude was determined to minimize type I and II errors when accepting or rejecting ICE predictions. We updated the ICE user guidance based on these analyses to advance the state of the science for ICE model application and interpretation. Integr Environ Assess Manag 2024;20:1554-1565. Published 2023. This article is a U.S. Government work and is in the public domain in the USA.

A QSAR-ICE-SSD model prediction of the PNECs for alkylphenol substances and application in ecological risk assessment for rivers of a megacity

Alkylphenols (APs) are ubiquitous and generally present in higher residue levels in the environment. The present work focuses on the development of a set of in silico models to predict the aquatic toxicity of APs with incomplete/unknown toxicity data in aquatic environments. To achieve this, a QSAR-ICE-SSD model was constructed for aquatic organisms by combining quantitative structure-activity relationship (QSAR), interspecies correlation estimation (ICE), and species sensitivity distribution (SSD) models in order to obtain the hazardous concentrations (HCs) of selected APs. The research indicated that the keywords "alkylphenol" and "nonylphenol" were most commonly studied. The selected ICE models were robust (R2: 0.70-0.99; p-value < 0.01). All models had a high reliability cross- validation success rates (>75%), and the HC5 predicted with the QSAR-ICE-SSD model was 2-fold than that derived with measured experimental data. The HC5 values demonstrated nearly linear decreasing trend from 2-MP to 4-HTP, while the decreasing trend from 4-HTP to 4-DP became shallower, indicates that the toxicity of APs to aquatic organisms increases with the addition of alkyl carbon chain lengths. The ecological risks assessment (ERA) of APs revealed that aquatic organisms were at risk from exposure to 4-NP at most river stations (the highest risk quotient (RQ) = 1.51), with the highest relative risk associated with 2.9% of 4-NP detected in 82.9% of the sampling sites. The targeted APs posed potential ecological risks in the Yongding and Beiyun River according to the mixture ERA. The potential application of QSAR-ICE-SSD models could satisfy the immediate needs for HC5 derivations without the need for additional in vivo testing.

Ecological risk assessment for difenoconazole in aquatic ecosystems using a web-based interspecies correlation estimation (ICE)-species sensitivity distribution (SSD) model

Difenoconazole is a typical triazole fungicide that can inhibit demethylation during ergosterol synthesis. Due to its wide use, difenoconazole is frequently detected in surface water, paddy water, agricultural water, and other aquatic environments. Presently, an assessment of the ecological risk posed by difenoconazole in aquatic ecosystems is lacking. Here, a web-based interspecies correlation estimation (ICE)-species sensitivity distribution (SSD) model was first applied to assess the ecological risk of difenoconazole in aquatic environments. Meanwhile, maximum acceptable concentration (MAC), maximum risk-free concentration (MRFC), and risk quotient (RQ) values were used to evaluate the potential risk of difenoconazole to aquatic organisms. Our results showed that an aquatic MAC value of 0.31 μg/L was acceptable for difenoconazole in aquatic environments. Further, the detected concentration of difenoconazole was lower than the MRFC value of 0.09 μg/L indicating no risk to aquatic organisms. Assessment data suggested that difenoconazole exhibited potential risks to eight studied aquatic ecosystems (including surface water, paddy water, and agricultural water) in different countries (RQ > 1), indicating that difenoconazole overuse could cause adverse effects to aquatic organisms in these aquatic ecosystems. Thus, restricted use and rational use of difenoconazole are recommended.

Prediction of chemical reproductive toxicity to aquatic species using a machine learning model: An application in an ecological risk assessment of the Yangtze River, China

The endocrine disrupting chemicals (EDCs) have been at the forefront of environmental issues for over 20 years and are a principle factor considered in every ecological risk assessment, but this kind of risk assessment faces difficulties. The expense, time cost of in vivo tests, and lack of toxicity data are key limiting factors for the ability to conduct ecological risk assessments of EDCs to aquatic species. In this study, a machine learning model named the support vector machine (SVM) was used to predict the reproductive toxicity of EDCs, and the performance of the models was evaluated. The results showed that the SVM model provided more accurate toxicity prediction data compared with the interspecies correlation estimation (ICE) model developed by previous study to predict the reproductive toxicity. The application of the predicted toxicity data was an important supplement to the observed data for the ecological risk assessment of EDCs in the Yangtze River, where estrogens and phenolic compounds have been found at some sampling sites in the middle and lower reaches. The results showed that the ecological risk of estrone, 17β-estradiol, and ethinyl estradiol were significant. This study revealed the application potential of machine learning models for the prediction of reproductive toxicity effects of EDCs. This can provide reliable alternative toxicity data for the ecological risk assessments of EDCs.

Cross-species extrapolation of chemical sensitivity

Ecosystems are usually populated by many species. Each of these species carries the potential to show a different sensitivity towards all of the numerous chemical compounds that can be present in their environment. Since experimentally testing all possible species-chemical combinations is impossible, the ecological risk assessment of chemicals largely depends on cross-species extrapolation approaches. This review overviews currently existing cross-species extrapolation methodologies, and discusses i) how species sensitivity could be described, ii) which predictors might be useful for explaining differences in species sensitivity, and iii) which statistical considerations are important. We argue that risk assessment can benefit most from modelling approaches when sensitivity is described based on ecologically relevant and robust effects. Additionally, specific attention should be paid to heterogeneity of the training data (e.g. exposure duration, pH, temperature), since this strongly influences the reliability of the resulting models. Regarding which predictors are useful for explaining differences in species sensitivity, we review interspecies-correlation, relatedness-based, traits-based, and genomic-based extrapolation methods, describing the amount of mechanistic information the predictors contain, the amount of input data the models require, and the extent to which the different methods provide protection for ecological entities. We develop a conceptual framework, incorporating the strengths of each of the methods described. Finally, the discussion of statistical considerations reveals that regardless of the method used, statistically significant models can be found, although the usefulness, applicability, and understanding of these models varies considerably. We therefore recommend publication of scientific code along with scientific studies to simultaneously clarify modelling choices and enable elaboration on existing work. In general, this review specifies the data requirements of different cross-species extrapolation methods, aiming to make regulators and publishers more aware that access to raw- and meta-data needs to be improved to make future cross-species extrapolation efforts successful, enabling their integration into the regulatory environment.

Derivation of algal acute to chronic ratios for use in chemical toxicity extrapolations

Algal toxicity studies are required by regulatory agencies for a variety of purposes including classification and labeling and environmental risk assessment of chemicals. Algae are also frequently the most sensitive taxonomic group tested. Acute to chronic ratios (ACRs) have been challenging to derive for algal species because of the complexities of the underlying experimental data including: a lack of universally agreed upon algal inhibition endpoints; evolution of experimental designs over time and by different standardization authorities; and differing statistical approaches (e.g., regression versus hypothesis-based effect concentrations). Experimental data for developing globally accepted algal ACRs have been limited because of data availability, and in most regulatory frameworks an ACR of 10 is used regardless of species, chemical type or mode of action. Acute and chronic toxicity (inhibition) data on 17 algal species and 442 chemicals were compiled from the EnviroTox database (https://envirotoxdatabase.org/) and a proprietary database of algal toxicity records. Information was probed for growth rate, yield, and final cell density endpoints focusing primarily on studies of 72 and 96 h duration. Comparisons of acute and chronic data based on either single (e.g., growth rate) and multiple (e.g., growth rate, final cell density) endpoints were used to assess acute and chronic relationships. Linear regressions of various model permutations were used to compute ACRs for multiple combinations of taxa, chemicals, and endpoints, and showed that ACRs for algae were consistently around 4 (ranging from 2.43 to 5.62). An ACR of 4 for algal toxicity is proposed as an alternative to a default value of 10, and recommendations for consideration and additional research and development are provided.

Oryzias sinensis, a new model organism in the application of eco-toxicity and water quality criteria (WQC)

Fish play an important role as a primary eco-toxicity test organism in environmental hazard assessment. Toxicity data of native species are often sought for use in the derivation of water quality criteria (WQC). The Chinese medaka, Oryzias sinensis, is an endemic species of China. The acute toxicity of 6 chemicals on O. sinensis was tested in this work, and toxicity effect of 10 chemicals to O. sinensis was compared with 4 commonly used species globally. A total of 9 robust interspecies correlation estimation (ICE) models using O. sinensis as the surrogate species were constructed and used to derive predicted no effect concentration and hazardous concentrations of 5% species (HC₅) values based on species sensitivity distribution. Results showed that the 96 h median lethal concentration of Hg²⁺, Cr⁶⁺, linear alkylbenzene sulfonates, triclosan, 3,4-dchloroaniline, sodium chloride to O. sinensis were 0.29, 50, 6.0, 0.63, 9.2 and 14,400 mg/L, respectively. The sensitivity of O. sinensis and other 4 testing organisms were statistically indistinguishable (P > 0.05). No significant difference among HC₅-ICE, HC₅-measured and HC₅ from published literatures was identified. All results indicated the O. sinensis is a potential model organism in the application of eco-toxicity and WQC in China and other Asian countries.

Effective extrapolation models for ecotoxicity of benzene, toluene, ethylbenzene, and xylene (BTEX)

Benzene homologues have significant toxic effects to aquatic organisms. In this study, the acute toxicity data of benzene, toluene, ethylbenzene and xylene (BTEX) were collected and screened, and the toxicity extrapolation model of paired BTEX was established. The results showed that except the correlation between benzene and xylene was not strong due to insufficient data, the linear correlation of the other five paired BTEX was good (p < 0.01), and R² fitted by the four out of five paired BTEX was greater than 0.85. The cross validation showed that ethylbenzene-xylene model was optimal, and for most species (81.8%), the established five BTEX models had a prediction error of less than 10%. Also, these extrapolation models were validated by experimental results of Pseudorasbora parva. The difference between the predicted and measured values of the acute toxicity of BTEX was less than 1 fold, which indicated that the extrapolation model had high accuracy.

Development and use of interspecies correlation estimation models in China for potential application in water quality criteria

Establishment of numerical water quality criteria (WQC) has brought increasing interest in China. However, toxicity data to develop robust WQC values (number of toxicity data ≥8) of contaminants based solely on endemic and indigenous species are insufficient. In this study, interspecies correlation estimation (ICE) models were developed using a combination of North American ICE models supplemented with China-specific species to resolve this problem. A total of 207 significant surrogate-predicted models (p < 0.05, F-test) were derived: 119, 66 and 22 models for vertebrates, invertebrates and plant surrogate species, respectively. Model cross-validation success rate (≥80%), mean square error (MSE, ≤ 0.54), R² (≥0.78) and taxonomic distance (≤4, within the same class) were selected as guiding criteria to screen the resulted ICE models. The differences of 5th percentile hazard concentrations (HC₅s) for 6 chemicals (2,4-dichlorophenol, triclosan, tetrabromobisphenol A, nitrobenzene, perfluorooctane sulfonate and octabromodiphenyl ether) calculated from ICE-based and measured toxicity-based SSDs were within 3-fold among models. Although the number of derived ICE models was not comprehensive and continues to be improved, they can already be used in the development of WQC targeting protection of aquatic life and environmental risk assessments for chemicals lacking toxicity data.

Application of Interspecies Correlation Estimation (ICE) models and QSAR in estimating species sensitivity to pesticides

Ecological risk assessment is challenged by the need to assess hazard to the diverse communities of organisms inhabiting aquatic and terrestrial systems. Computational approaches, such as Quantitative Structure Activity Relationships (QSAR) and Interspecies Correlation Estimation (ICE) models, are useful tools that provide estimates of acute toxicity where data are lacking or limited for ecological risk assessments (ERA). This review describes the technical basis of ICE models for use in pesticide ERA that may be used in conjunction with QSAR model estimates or surrogate species toxicity data and demonstrates the potential for improving hazard assessment. Validation and uncertainty analysis of ICE model predictions are summarized and used as guidance for selecting ICE models and evaluating toxicity predictions. A user-friendly web-based ICE modelling platform (Web-ICE) is described and demonstrated through case studies. Case studies include the development of Species Sensitivity Distributions generated from QSAR and ICE estimates, comparative sensitivity for a pesticide and its degradate, and application of ICE-estimated toxicity values for listed species assessments.

Development of interspecies correlation estimation (ICE) models to predict the reproduction toxicity of EDCs to aquatic species

Endocrine disrupting chemicals (EDCs) threaten the reproductive fitness of aquatic organisms at concentrations lower than those associated with longevity and development. However, the small number of aquatic species assessed for reproductive toxicity has limited the ecological risk assessment of EDCs, making sensible decisions more difficult. In response to this, interspecies correlation estimation (ICE) models were established for EDCs to enable the estimation of reproduction toxicity values to a wider range of organisms. A total of 16 ICE models of EDCs for 6 surrogate species were statistically significant. Of the 16 models, 37.5% (6 models) had a cross-validation success rate > 60%, with a relatively small model squared error, indicating that the model fit is robust. These model results implied that the action of EDCs for each species pair might involve the same mechanisms, and taxonomic relationships did not influence the prediction precision. The cross-validation success rate corroborated the consistency between the projected and experimental values for the EDC ICE models. Sixty-seven percent of the projected values fell within a 10-fold difference of the experimental data. The results indicated that a proven ICE model can greatly increase the amount of EDCs chronic toxicity data for predicted species, without the need for extensive animal experiments, thus providing substitute chronic toxicity data for rapid assessment of EDCs ecological risks.

Aquatic plants: Test species sensitivity and minimum data requirement evaluations for chemical risk assessments and aquatic life criteria development for the USA

Phytotoxicity results from the publicly-available ECOTOX database were summarized for 20 chemicals and 188 aquatic plants to determine species sensitivities and the ability of a species-limited toxicity data set to serve as a surrogate for a larger data set. The lowest effect concentrations reducing the sublethal response parameter of interest by 50% relative to the controls (EC50) usually varied several orders of magnitude for the 119 freshwater and 69 saltwater plants exposed to the same test chemicals. Generally, algae were more sensitive than floating and benthic species but inter-specific differences for EC50 values were sometimes considerable within and between phyla and no consistently sensitive species was identified for the morphologically-diverse taxa. Consistent equivalencies of the phytotoxicity databases for freshwater-saltwater plants and floating-benthic macrophyte species were not demonstrated. Two species-sensitivity distribution plots (SSDs) were constructed for each of the 20 chemicals, one based on all available phytotoxicity information (range = 10-76 test species) and another based on information for only five species recommended for pesticide hazard evaluations. HC5 values (hazardous concentration to 5% of test species) estimated from the two SSDs usually differed four-fold or less for the same chemical. HC5 values for the five species were often conservative estimates of HC5 values for the more species-populated data sets. Consequently, the collective response of the five test species shows promise as an interim aquatic plant minimum data requirement for aquatic life criteria development. In contrast, the lowest EC50 values for the five species usually were greater than HC5 values for the same test chemicals, a finding important to criteria-supporting Final Plant Values. The conclusions may differ for comparisons based on other test chemicals, test species, response parameters and calculations.