Consensus QSAR models estimating acute toxicity to aquatic organisms from different trophic levels: algae,Daphnia and fish

ABSTRACT

We report new consensus models estimating acute toxicity for algae, Daphnia and fish endpoints. We assembled a large collection of 3680 public unique compounds annotated by, at least, one experimental value for the given endpoint. Support Vector Machine models were internally and externally validated following the OECD principles. Reasonable predictive performances were achieved (RMSE_ext = 0.56–0.78) which are in line with those of state-of-the-art models. The known structural alerts are compared with analysis of the atomic contributions to these models obtained using the ISIDA/ColorAtom utility. A benchmarking against existing tools has been carried out on a set of compounds considered more representative and relevant for the chemical space of the current chemical industry. Our model scored one of the best accuracy and data coverage.

Nevertheless, industrial data performances were noticeably lower than those on public data, indicating that existing models fail to meet the industrial needs. Thus, final models were updated with the inclusion of new industrial compounds, extending the applicability domain and relevance for application in an industrial context. Generated models and collected public data are made freely available.     

Assessment of in silico models for acute aquatic toxicity towards fish under REACH regulation

We evaluated the performance of eight QSAR in silico modelling packages (ACD/ToxSuite?, ADMET Predictor?, DEMETRA, ECOSAR, TerraQSAR?, Toxicity Estimation Software Tool, TOPKAT? and VEGA) for acute aquatic toxicity towards two species of fish: Fathead Minnow and Rainbow Trout. For the Fathead Minnow, we compared model predictions for 567 substances with the corresponding experimental values for 96-h median lethal concentrations (LC50). Some models gave good results, with r² up to 0.85. We also classified the predictions of all the models into four toxicity classes defined by CLP. This permitted us to assess other parameters, such as the percentage of correct predictions for each class. Then we used a set of 351 substances with toxicity data towards Rainbow Trout (96-h LC50). In this case the predictability was unacceptable for all the in silico models. The calculated r² gave poor correlations (≤0.53). Another analysis was performed according to chemical classes and for mode of action. In the first case, all the classes show a high percentage of correct predictions, in the second case only narcotics and polar narcotics were predicted with good confidence. The results indicate the possibility of using in silico methods to estimate aquatic toxicity within REACH regulation, after careful evaluation.     

Multivariate Analysis of the First 10 MEIC Chemicals

Abstract

Nonlinear mapping coupled to powerful graphical tools was used to compare the texicological responses of 32 in vivo and in vitro test systems to the first 10 MEIC chemicals. The obtained results clearly underline the usefulness of our methodological approach for the comparison of the different endpoints and the selection of a battery of in vitro toxicity tests allowing to estimate the possible harmful effects of chemicals in vivo.     

Predictions for existing chemicals-a multilateral QSAR project

Abstract Following a previous collaborative EU/EPA project focussed on QSAR predictions for a selection of new chemicals which had been notified in the EU, a similar exercise was started in 1993 on existing chemicals. In a first phase, the project addresses the High Production Volume (HPV) chemicals which are produced or imported at levels above a 1000t/year in the EU and 454t/year in the US. The relevant EU (Annex 1 of Existing Chemicals Regulation No. 793/93) and US-EPA lists contain 1036 and 2881 organic substances respectively of which HPV 749 chemicals are in common. The joint project aims at an estimation through validated QSAR models of the physical-chemical, ecotoxicity and toxicity endpoints which are included in the regulation and where experimental data will become available in IUCLID (International Unified Chemicals Information Database). Next to EC-JRC (ECB) and US-EPA, various laboratories in the EU are contributing to the project and recently, two institutes in Japan have joined in this project.     

Ecotoxicological modelling of cosmetics for aquatic organisms: A QSTR approach

In this study, externally validated quantitative structure–toxicity relationship (QSTR) models were developed for toxicity of cosmetic ingredients on three different ecotoxicologically relevant organisms, namely Pseudokirchneriella subcapitata, Daphnia magna and Pimephales promelas following the OECD guidelines. The final models were developed by partial least squares (PLS) regression technique, which is more robust than multiple linear regression. The obtained model for P. subcapitata shows that molecular size and complexity have significant impacts on the toxicity of cosmetics. In case of P. promelas and D. magna, we found that the largest contribution to the toxicity was shown by hydrophobicity and van der Waals surface area, respectively. All models were validated using both internal and test compounds employing multiple strategies. For each QSTR model, applicability domain studies were also performed using the “Distance to Model in X-space” method. A comparison was made with the ECOSAR predictions in order to prove the good predictive performances of our developed models. Finally, individual models were applied to predict toxicity for an external set of 596 personal care products having no experimental data for at least one of the endpoints, and the compounds were ranked based on a decreasing order of toxicity using a scaling approach.     

Toxicity Measurement of Imidazolium Ionic Liquids Using Acute Toxicity Test

Ionic liquids (ILs) can be considered as environmentally friendly solvent, but they have the ability to dissolve in water and accumulate in the environment. Therefore, the toxicity of ILs should be assessed in order to prevent their harm to human and environment. This study was carried out to investigate the toxicity of ILs towards marine and freshwater fish. Three ILs have been tested, which are 1-Butyl-3-methylimidazolium hydrogen sulfate and 1-Butyl-3-methylimidazolium bis (trifluoromethylsulfonyl) imide toward marine fish and 1-Hexyl-3-methylimidazolium bis (trifluoromethylsulfonyl) imide toward freshwater fish. Two different marine fish were employed, which are: Cephalopholis cruentata (grouper) and Lates calcarifer (barramundi). For freshwater fish, male Poecilia reticulate (guppy) was employed. The toxicity tests were conducted according to OECD (Organisation for Economic Cooperation and Development) guideline 203. For 1-Butyl-3-methylimidazolium hydrogen sulphate [BMIM][HSO4], the median lethal concentration (LC50) estimated toward Cephalopholis cruentatato be 199.98 mg.L^-1. For 1-Butyl-3-methylimidazolium bis (trifluoromethylsulfonyl) imide [BMIM][TFSI], LC50 estimated toward LatesCalcariferto be 374.11 mg.L^-1. While, for 1-Hexyl-3-methylimidazolium bis (trifluoromethylsulfonyl) imide, [HMIM][NTf2], LC50 estimated toward Poecilia Reticulate to be 207.49 mg.L^-1. All the LC50 values obtained can be identified as practically nontoxic liquids based on Acute Toxicity Rating Scale by Fish and Wildlife Service (FWS). As to our knowledge, there is no previous reported toxicity studies of [BMIM][HSO4] and [BMIM][TFSI] on marine fish and [HMIM] [NTf2] on freshwater fish. Thus, this paper can be used as a benchmark for researchers who are dealing with these three ILs.     

External validation of EPIWIN biodegradation models

The BIOWIN biodegradation models were evaluated for their suitability for regulatory purposes. BIOWIN includes the linear and non-linear BIODEG and MITI models for estimating the probability of rapid aerobic biodegradation and an expert survey model for primary and ultimate biodegradation estimation. Experimental biodegradation data for 110 newly notified substances were compared with the estimations of the different models. The models were applied separately and in combinations to determine which model(s) showed the best performance. The results of this study were compared with the results of other validation studies and other biodegradation models. The BIOWIN models predict not-readily biodegradable substances with high accuracy in contrast to ready biodegradability. In view of the high environmental concern of persistent chemicals and in view of the large number of not-readily biodegradable chemicals compared to the readily ones, a model is preferred that gives a minimum of false positives without a corresponding high percentage false negatives. A combination of the BIOWIN models (BIOWIN2 or BIOWIN6) showed the highest predictive value for not-readily biodegradability. However, the highest score for overall predictivity with lowest percentage false predictions was achieved by applying BIOWIN3 (pass level 2.75) and BIOWIN6.     

New Models to Predict the Acute and Chronic Toxicities of Representative Species of the Main Trophic Levels of Aquatic Environments

To assess the impact of chemicals on an aquatic environment, toxicological data for three trophic levels are needed to address the chronic and acute toxicities. The use of non-testing methods, such as predictive computational models, was proposed to avoid or reduce the need for animal models and speed up the process when there are many substances to be tested. We developed predictive models for Raphidocelis subcapitata, Daphnia magna, and fish for acute and chronic toxicities. The random forest machine learning approach gave the best results. The models gave good statistical quality for all endpoints. These models are freely available for use as individual models in the VEGA platform and for prioritization in JANUS software.     

Application of Hydrophobicity Parameters to Prediction of the Acute Aquatic Toxicity of Commercial Surfactant Mixtures

Abstract

Ethoxylated alcohols are the most extensively used nonionic surfactants in detergent products. The application of QSAR to their aquatic toxicity is complicated by the fact that they are multicomponent mixtures, the parent alcohols being often mixtures of isomers and homologues, each one being ethoxylated to varying degrees. A spreadsheet method for calculation of aquatic toxicity of such nonionic surfactant mixtures is presented. The method is based on a combination of the Könemann narcosis QSAR and mixture toxicity equations based on the principle of concentration addition. Log P values used in the spreadsheet calculations are themselves calculated by spreadsheet formulae based on the Leo and Hansch method modified by incorporation of the position dependent branching factor originally applied to linear alkylbenzene sulphonates. Close agreement between calculated and experimental EC50 values (48 hr Daphnia tests) is obtained for a range of ethoxylated alcohols having a diversity of branching patterns, carbon numbers and degrees of ethoxylation. The effects of increasing carbon number (decreasing EC50), branching (increasing EC50) and increasing degree of ethoxylation (increasing EC50) are all quantified.     

Development of classification models for predicting chronic toxicity of chemicals to Daphnia magna and Pseudokirchneriella subcapitata

Both the acute toxicity and chronic toxicity data on aquatic organisms are indispensable parameters in the ecological risk assessment priority chemical screening process (e.g. persistent, bioaccumulative and toxic chemicals). However, most of the present modelling actions are focused on developing predictive models for the acute toxicity of chemicals to aquatic organisms. As regards chronic aquatic toxicity, considerable work is needed. The major objective of the present study was to construct in silico models for predicting chronic toxicity data for Daphnia magna and Pseudokirchneriella subcapitata. In the modelling, a set of chronic toxicity data was collected for D. magna (21 days no observed effect concentration (NOEC)) and P. subcapitata (72 h NOEC), respectively. Then, binary classification models were developed for D. magna and P. subcapitata by employing the k-nearest neighbour method (k-NN). The model assessment results indicated that the obtained optimum models had high accuracy, sensitivity and specificity. The model application domain was characterized by the Euclidean distance-based method. In the future, the data gap for other chemicals within the application domain on their chronic toxicity for D. magna and P. subcapitata could be filled using the models developed here.     

A baseline inhalation toxicity model for narcosis in mammals

This paper presents the results of an analysis of the rodent inhalation literature and the development of a quantitative structure–activity relationships (QSAR) model for 4-hour LC50 as baseline toxicity to complement the baseline toxicity model for aquatic animals. We used the same literature review criteria developed for the ECOTOX database which selects only primary references with explicit experimental methods to form a high-quality database. Our literature review focused on the primary references reporting a 4-hour exposure for a single species of rodent in which the chemical had been clearly tested as a vapour and for which the exposure concentrations were not ambiguous. An expert system was used to remove reactive chemicals, receptor-mediated toxicants, and any test that produced symptoms inconsistent with non-polar narcosis. The QSAR model derived for narcosis in rodents was log LC50 = 0.69 × log VP + 1.54 which had an r ² of 0.91, which is significantly better than the baseline toxicity model for aquatic animals. This simple model suggests that there is no intrinsic barrier to estimating baseline toxicity for in vivo endpoints in mammalian or terrestrial toxicology.     

Structure-Toxicity Relationships for Aliphatic Compounds Encompassing a Variety of Mechanisms of Toxic Action to Vibrio fischeri

Abstract

QSARs based upon the logarithm of the octanol-water partition coefficient, logP, and energy of the lowest unoccupied molecular orbital, E_LUMO were developed to model the toxicity of aliphatic compounds to the marine bacterium Vibrio fischeri. Statistically robust, hydrophobic-dependent QSARs were found for chloroalcohols and haloacetonitriles. Modelling of the toxicity of the haloesters and the diones required the use of terms to describe both hydrophobicity and electrophilicity. The differences in intercepts, slopes, and fit of these models suggest different electrophilic mechanisms occur between classes, as well as within the diones and haloesters. In order to model globally the toxicity of aliphatic compounds to V. fischeri, all the data determined in this study were combined with those determined previously for alkanones, alkanals, and alkenals. A highly predictive two-parameter QSAR [pT₁₅ = 0.760(log P) ?0.625(E _LUMO) ?0.466; n = 63, s = 0.462, r ² = 0.846, F = 171, Pr > F = 0.0001] was developed for the combined data that models across classes and is independent of mechanisms of action. The toxicity of these compounds to V. fischeri compares well to the toxicity (50% population growth inhibition) to the ciliate Tetrahymena pyriformis (r ² = 0.850).     

Comparison of in silico models for prediction of Daphnia magna acute toxicity

Eight in silico modelling packages were evaluated and compared for the prediction of Daphnia magna acute toxicity from the viewpoint of the European legislation on chemicals, REACH. We tested the following models: Discovery Studio (DS) TOPKAT, ACD/Tox Suite, ADMET Predictor?, ECOSAR (Ecological Structure Activity Relationships), TerraQSAR?, T.E.S.T. (Toxicity Estimation Software Tool) and two models implemented in VEGA on 480 industrial compounds for 48-h median lethal concentrations (LC₅₀) to D. magna, matching them with experimental values. The quality of the estimates was compared using a standard statistical review and an additional classification approach in which the hazard predictions were grouped using well-defined regulatory criteria. The regression parameters, correlation coefficient being the most influential, showed that four models (ADMET Predictor?, DS TOPKAT, TerraQSAR? and VEGA DEMETRA) had similar reliability. These performed better than the others, but the coefficient of determination was still low (r² around 0.6), considering that at least half the predicted compounds were inside the training sets. Additionally, we grouped the results in four defined toxicity classes. TerraQSAR? gave 60% of correct classifications, followed by DS TOPKAT, ADMET Predictor? and VEGA DEMETRA, with 56%, 54% and 48%, respectively. These results highlight the challenges associated with developing reliable and easily applied acceptability criteria for the regulatory use of QSAR models to D. magna acute toxicity.