Structure-activity relationships in toxicology and ecotoxicology: An assessment

High-throughput prediction of oral acute toxicity in Rat and Mouse of over 100,000 polychlorinated persistent organic pollutants (PC-POPs) by interpretable data fusion-driven machine learning global models

This study utilized available oral acute toxicity data in Rat and Mouse for polychlorinated persistent organic pollutants (PC-POPs) to construct data fusion-driven machine learning (ML) global models. Based on atom-centered fragments (ACFs), the collected high-throughput data overcame the applicability limitations, enabling accurate toxicity prediction for a wide range of PC-POPs series compounds using only single models. The data variances in the Rat training and test sets were 1.52 and 1.34, respectively, while for the Mouse, the values were 1.48 and 1.36, respectively. Genetic algorithm (GA) was used to build multiple linear regression (MLR) models and pre-screen descriptors, addressing the "black-box" problem prevalent in ML and enhancing model interpretability. The best ML models for Rat and Mouse achieved approximately 90 % prediction reliability for over 100,000 true untested compounds. Ultimately, a warning list of highly toxic compounds for eight categories of polychlorinated atom-centered fragments (PCACFs) was generated based on the prediction results. The analysis of descriptors revealed that dioxin analogs generally exhibited higher toxicity, because the heteroatoms and ring systems increased structural complexity and formed larger conjugated systems, contributing to greater oral acute toxicity. The present study provides valuable insights for guiding the subsequent in vivo tests, environmental risk assessment and the improvement of global governance system of pollutants.

Glutamic Acid Uptake Inhibition Assay in Cultured Hep G2 Cells as an Alternative Method for Evaluating Potential In Vivo Eye Irritation

Glutamic acid (GA) content was measured in cultured Hep G2 cells, after treatment of the cells with test compounds. The results with 37 chemicals were compared with their respective rabbit eye irritation data, of which 17 were determined according to the OECD test, and the other 20 in range-finding studies. The chemicals were mainly organic solvents (alcohols, esters, amines, acids and others). The xenobiotics were applied to the cells for 4 hours at 5 different concentrations. The cells were then incubated for 15 minutes with tritiated GA. GA uptake inhibition was measured by liquid scintillation counting, and the results were expressed as the GI50 value, which is the concentration of test compound required to induce a 50% reduction in GA uptake. A linear correlation coefficient r = 0.66 was found between the log GI50 and the mean corneal opacity scores. This value is comparable to that obtained in total protein and uridine uptake inhibition studies. However, r = 0.81 was found when the log GI50 was compared with range-finding scores, indicating that a closer relationship exists between cytotoxicity and the latter.

Genotoxicity response of Vicia faba seedlings to cadmium in soils as characterized by direct soil exposure and micronucleus test

To overcome the drawbacks of the Vicia faba root tip micronucleus test in soil using the solution extract method, we conducted a potting experiment by direct soil exposure. Cadmium was spiked into 3 typical soils (brown soil, red soil, and black soil) to simulate environmental concentrations (0.625, 1.25, 2.5, 5, and 10 mg kg^-1). Multiple Vicia faba tissues (primary root tips, secondary root tips, and leaf tips) were sampled, and mitotic index (MI), chromosome aberration frequency (CA), and micronucleus frequency (MN) were used as endpoints after a seedling period of 5 days. The results showed a response between Cd concentrations and multiple sampling tissues of Vicia faba, and the secondary root tips responded to Cd stress the most, followed by primary root tips and leaf tips. Soil physicochemical properties (e.g., pH, total phosphorus, total organic carbon, etc.) influenced the genotoxicity of Cd, and pH was the dominant factor, which resulted in the genetic toxicity response of Cd in soils in the order: red soil > brown soil > black soil. The lowest observable effect concentration (LOEC) of Cd was 1.25 mg kg^-1 for both brown soil and red soil and 2.5 mg kg^-1 for black soil. In view of this, we suggested that soil properties should be considered in evaluating genotoxicity risk of Cd in soil, especially with soil pH range, and the secondary root tips should be taken as suitable test tissues in the MN test due to its more sensible response feature to Cd stress in soil.

QSPR of the bioconcentration factors of non-ionic organic compounds in fish using extended topochemical atom (ETA) indices

Bioconcentration refers to the absorption or uptake of a chemical from the media to an organism's tissues leading to greater concentration in tissues than that in the surrounding environment. Considering the importance of bioconcentration from the viewpoint of ecological safety assessment, a QSPR study was conducted based upon log BCF of 122 non-ionic organic compounds in fish using the recently introduced extended topochemical atom (ETA) indices. In deriving the models, principal component factor analysis (FA) followed by multiple linear regression (MLR), stepwise regression, partial least squares (PLS) and principal component regression analysis (PCRA) were applied as statistical tools. This was repeated with non-ETA (topological and physicochemical) descriptors and a combination set including both the ETA and non-ETA descriptors. The ETA indices suggested negative contributions of functionalities of nitro, amino and hydroxy substructures and positive contributions of branching, volume and functionality of chloro substituents. Again, the predictive ability of the developed models was compared with the previously reported models. Finally the validation of all the QSAR models was discussed based on random division, sorted log BCF data and K-means clusters for the factor scores of the original variable (ETA) matrix without the response property values. The results suggest that ETA parameters are sufficiently rich in chemical information to encode the structural features contributing to the bioconcentration of the non-ionic organic compounds in fish and thus these merit further assessment to explore their potential in QSAR/QSPR/QSTR modelling.

Modeling the bioconcentration factors and bioaccumulation factors of polychlorinated biphenyls with posetic quantitative super-structure/activity relationships (QSSAR)

During bioconcentration, chemical pollutants from water are absorbed by aquatic animals via the skin or a respiratory surface, while the entry routes of chemicals during bioaccumulation are both directly from the environment (skin or a respiratory surface) and indirectly from food. The bioconcentration factor (BCF) and the bioaccumulation factor (BAF) for a particular chemical compound are defined as the ratio of the concentration of a chemical inside an organism to the concentration in the surrounding environment. Because the experimental determination of BAF and BCF is time-consuming and expensive, it is efficacious to develop models to provide reliable activity predictions for a large number of chemical compounds. Polychlorinated biphenyls (PCBs) released from industrial activities are persistent pollutants of the environment that produce widespread contamination of water and soil. PCBs can bioaccumulate in the food chain, constituting a potential source of exposure for the general population. To predict the bioconcentration and bioaccumulation factors for PCBs we make use of the biphenyl substitution-reaction network for the sequential substitution of H-atoms by Cl-atoms. Each PCB structure then occurs as a node of this reaction network, which is some sort of super-structure, turning out mathematically to be a partially ordered set (poset). Rather than dealing with the molecular structure via ordinary QSAR we use only this poset, making different quantitative super-structure/activity relationships (QSSAR). Thence we developed cluster expansion and splinoid QSSARs for PCB bioconcentration and bioaccumulation factors. The predictive ability of the BAF and BCF models generated for 20 data sets (representing different conditions and fish species) was evaluated with the leave-one-out cross-validation, which shows that the splinoid QSSAR (r between 0.903 and 0.935) are better than models computed with the cluster expansion (r between 0.745 and 0.887). The splinoid QSSAR models for BAF and BCF yield predictions for the missing PCBs in the investigated data sets.

A fragment constant QSAR model for evaluating the EC50 values of organic chemicals to Daphnia magna

The quantitative relationship between the median effective concentration (EC50) of organic chemicals to Daphnia magna and the number of molecular fragments was investigated based on experimental EC50 values for 217 chemicals derived from the literature. A fragment constant model was developed based on a multivariate linear regression between the number of fragments and the logarithmically transformed reciprocal values of EC50. Functional correction factors were introduced into the model. The model was verified using an independent set of randomly selected data. The mean residual of the final model was 0.4 log-units. The robustness of the model was discussed based on the results of three jackknife tests.

The application of predictive models in the environmental risk assessment of ECONOR

Environmental risk assessment of products requires information on the physico-chemical properties, persistence and ecotoxicity of the product, its constituents and possible metabolic and degradation products. Experimental investigations are usually required to generate this information and consequently risk assessment can be costly and time consuming. One possible approach to minimising the amount of experimental testing is to supplement experimental data with data predicted using models such as quantitative structure-activity relationships (QSARs). Using these models, information can be generated based primarily on the knowledge of the chemical structure of the substance(s) under investigation. In this study predictive models were used to assess the environmental risk of the veterinary medicine, ECONOR which contains the active ingredient valnemulin. Available experimental data on the properties, degradability and ecotoxicity of valnemulin was supplemented with predicted data. Where possible, experimental data was used to validate the predicted approaches and this indicated that the predictions were accurate. Information on usage, properties and degradability was input to fate models to predict environmental concentrations (PECs) of valnemulin in soil, pore water and groundwater. Comparison of PECs with experimental and predicted ecotoxicity data for valnemulin indicated that that even under 'worst case' scenarios the environmental risk posed by valnemulin was low.

SAR models for estimating the percutaneous absorption of polynuclear aromatic hydrocarbons

A structure-activity relationship (SAR) of the in vitro percutaneous absorption of polycyclic aromatic hydrocarbons (PAH) is described. The data set consisted of 60 three to seven ring PAH. Over 50 numeric descriptors were generated from the modeled molecular structures. Computer aided methods were used to evaluate descriptors and develop linear expressions relating the percent of dermally applied PAH dose absorbed through skin (PADA) to PAH structure. Three regression models with one and two variables were developed. The log octanol/water partition coefficient (log P) was the most important variable in determining percutaneous absorption. An inverse relationship between log P and the skin penetration properties of the PAH was observed. Nearly 40 of 60 PAH tested had PADA-values within a factor of two of benzo[a]pyrene (BaP); well over 50 of 60 had PADA-values within a factor of three. The results lend support to the use of isotopically labeled BaP as a surrogate for measuring the dermal flux (in vivo and in vitro) and estimating the dermal bioavailability of PAH from complex mineral oil and coal-tar derived mixtures.

Assessment of QSARS for Predicting Fate and Effects of Chemicals in the Environment: An International European Project

Abstract In 1993, an international project on QSAR has been started with funding from the Commission of the European Union. The first part of the project is focused on preparing an overview of existing models for the prediction of environmental parameters such as bioconcentration, sorption, degradation and ecotoxicity. Emphasis will be given to defining the limitations of the models. Since all models, including QSARs, have their limitations, it is important that these limitations are known in case QSARs are actually used and applied within the risk assessment context. The second part of the project is directed towards experimental research on new developments with emphasis on the use of multivariate techniques and quantum chemical properties. In this short paper, a general outline of the project will be given, as well as some first results. Results of experimental work within this project will be published in the proceedings of the 6th International Workshop on QSAR in Environmental Sciences and will appear in this same journal.

Model-based description of distribution of chemicals in biosystems for the continuous dose

The distribution of xenobiotics in a morphologically compartmentalised biosystem is described for the continuous dose of readily metabolised chemicals. The time hierarchy of individual processes the compounds undergo in the biosystem (transport-considered as uni-directional due to metabolism, protein binding, and metabolism) is used to simplify the model description. The explicit expressions for the steady-state concentrations (ci) in individual compartments as a function of transport rate constants and metabolism rate constants are derived. Further substantial simplification can be achieved if the phases of the test biosystem can be considered as approximately identical. Under the conditions ci can be used to calculate the explicit expressions for the average concentrations in all aqueous phases and in all membranes. The model parameters (the membrane/water partition coefficient, the water-to-membrane transport rate constant) are substituted by the reference (usually 1-octanol/water) partition coefficient using plausible mutual relationships. The dependences of the steady-state concentrations in individual compartments as well as the average quantities in all aqueous phases and in all membranes on hydrophobicity and on the metabolism rate constant are analysed in detail. The presented explicit expressions for the steady-state concentrations create a firm basis for: (1) derivation of model-based QSAR equations for toxicity caused by individual mechanisms of action and for further quantities of environmental interest like bioconcentration factors and total metabolism rates as well as for (2) scaling and comparison of QSAR equations among individual organisms.

Modelling of human acute toxicity from physicochemical properties and non-vertebrate acute toxicity of the 38 organic chemicals of the MEIC priority list by PLS regression and neural network

Linear and non-linear modelling of human acute toxicity (as human lethal concentrations; HLCs) of the 38 organic chemicals from the 50 priority compounds of the Multicentre Evaluation of In Vitro Cytotoxicity (MEIC) programme was investigated. The models obtained were derived either from a set of 23 physicochemical properties of the compounds or from their acute toxicities to five aquatic non-vertebrates together with the physicochemical properties. For the linear type, modelling was performed using a partial least square projection to latent structures (PLS) regression method; for the non-linear models, both PLS regression and neural network were utilized. A neural network using a combination of backpropagation and cascade-correlation algorithms was applied in this study. The results generally reveal a slightly better predictive performance of the models obtained from PLS regression than those obtained from neural networks. However, the model composed of physicochemical properties (PC-model) from the trained neural network using a back propagation algorithm with pruning technique proved superior to that trained with a combination of backpropagation and cascade-correlation algorithms after leave-one-out cross-validation. The predictive power of the PC-models, whether linear or non-linear, was comparable with that of the corresponding models consisting of both structural descriptors and the ecotoxicological tests (ECOPC-models), except for the battery (ECOPC-model) from the neural network. The composition of the 'best' PLS and neural network models points to the importance of the combination of physicochemical properties reflecting lipophilicity, size, volume, intermolecular binding forces and electronic properties of the molecule. All the aquatic non-vertebrate tests are shown to be essential in explaining human acute toxicity. However, the degree of contribution differed, with the crustacean (Artemia salina) and the bacterial (Microtox) bioassays being more important to the linear and non-linear PLS models, whereas the crustacean (Artemia salina and Streptocephalus proboscideus) tests, and the rotifer (Brachionus calyciflorus) assay were important to the neural network models. The organochlorine (lindane) and bipyridinium (paraquat) pesticides were common outliers in all the models. Moreover, the latter two compounds and the organophosphate (malathion) pesticide were also common outliers in all ECOPC-models. Other types of pesticides, however, fit the models. The predicted HLCs of a number of non-pesticides, including some chlorinated compounds, also deviated from the observed HLCs by more than one order of magnitude.

Human acute toxicity prediction of the first 50 MEIC chemicals by a battery of ecotoxicological tests and physicochemical properties

Five acute bioassays consisting of three cyst-based tests (with Artemia salina, Streptocephalus proboscideus and Brachionus calyciflorus), the Daphnia magna test and the bacterial luminescence inhibition test (Photobacterium phosphoreum) are used to determine the acute toxicity of the 50 priority chemicals of the Multicentre Evaluation of In Vitro Cytotoxicity (MEIC) programme. These tests and five physiocochemical properties (n-octanol-water partition coefficient, molecular weight, melting point, boiling point and density) are evaluated either singly or in combination to predict human acute toxicity. Acute toxicity in human is expressed both as oral lethal doses (HLD) and as lethal concentrations (HLC) derived from clinical cases. A comparison has also been made between the individual tests and the conventional rodent tests, as well as between rodent tests and the batteries resulting from partial least squares (PLS), with regard to their predictive power for acute toxicity in humans. Results from univariate regression show that the predictive potential of bioassays (both ecotoxicological and rodent tests) is generally superior to that of individual physicochemical properties for HLD. For HLC prediction, however, no consistent trend could be discerned that indicated whether bioassays are better estimators than physicochemical parameters. Generally, the batteries resulting from PLS regression seem to be more predictive than rodent tests or any of the individual tests. Prediction of HLD appears to be dependent on the phylogeny of the test species: cructaceans, for example, appear to be more important components in the test battery than rotifers and bacteria. For HLC prediction, one anostracan and one cladoceran crustacean are considered to be important. When considering both ecotoxicological tests and physicochemical properties, the battery based on the molecular weight and the cladoceran crustacean predicts HLC substantially better than any other combination.

QSAR models for predicting the acute toxicity of selected organic chemicals with diverse structures to aquatic non-vertebrates and humans

The linear and non-linear relationships of acute toxicity (as determined on five aquatic non-vertebrates and humans) to molecular structure have been investigated on 38 structurally-diverse chemicals. The compounds selected are the organic chemicals from the 50 priority chemicals prescribed by the Multicentre Evaluation of In Vitro Cytotoxicity (MEIC) programme. The models used for the evaluations are the best combination of physico-chemical properties that could be obtained so far for each organism, using the partial least squares projection to latent structures (PLS) regression method and backpropagated neural networks (BPN). Non-linear models, whether derived from PLS regression or backpropagated neural networks, appear to be better than linear models for describing the relationship between acute toxicity and molecular structure. BPN models, in turn, outperform non-linear models obtained from PLS regression. The predictive power of BPN models for the crustacean test species are better than the model for humans (based on human lethal concentration). The physico-chemical properties found to be important to predict both human acute toxicity and the toxicity to aquatic non-vertebrates are the n-octanol water partition coefficient (Pow) and heat of formation (HF). Aside from the two former properties, the contribution of parameters that reflect size and electronic properties of the molecule to the model is also high, but the type of physico-chemical properties differs from one model to another. In all of the best BPN models, some of the principal component analysis (PCA) scores of the 13C-NMR spectrum, with electron withdrawing/accepting capacity (LUMO, HOMO and IP) are molecular size/volume (VDW or MS1) parameters are relevant. The chemical deviating from the QSAR models include non-pesticides as well as some of the pesticides tested. The latter type of chemical fits in a number of the QSAR models. Outliers for one species may be different from those of other test organisms.

Electronic properties and free radical production by nitrofuran compounds

Substitution of nifurtimox tetrahydrothiazine moiety by triazol-4-yl, benzimidazol-l-yl, pyrazol-l-yl or related aromatic nitrogen heterocycles determines changes in the quantum chemistry descriptors of the molecule, namely, (a) greater negative LUMO energy; (b) lesser electron density on specific atoms, especially on the nitro group atoms, and (c) modification of individual net atomic charges at relevant atoms. These variations correlate with the greater capability of nifurtimox analogues for redox-cycling and oxygen radical production, after one-electron reduction by ascorbate or reduced flavoenzymes. Variation of the nitrofurans electronic structure can also explain the greater activity of nifurtimox analogues as inhibitors of glutathione reductase and Trypanosoma cruzi growth, although other factors, such as molecular hydrophobicity and connectivity may contribute to the latter inhibition.

Analogy Models for Prediction of Human Toxicity

Estimating the toxicity to humans of chemicals by testing on human subjects is not considered to be ethically acceptable, and toxicity testing on laboratory animals is also questionable. Therefore, there is a need for alternative methods that will give estimates of various aspects of human toxicity. Batteries of in vitro tests, together with physicochemical and toxicokinetic data, analysed by efficient data analytical methods, may enable analogy models to be constructed that can predict human toxicity. It may be possible to model non-specific toxicity relating to lipophilicity, or basal cytotoxicity, for a series of diverse compounds with large variation in chemical structure and physicochemical properties. However, local models for a series of similar compounds are generally expected to be more accurate, as well as being capable of modelling more-specific interactions. Analogy models for the prediction of human toxicity are discussed and exemplified with physicochemical and cytotoxicity data from the first ten chemicals in the multicenter evaluation of in vitro cytotoxicity (MEIC) project.

Survey of the QSAR and in vitro approaches for developing non-animal methods to supersede the in vivo LD50 test

Quantitative structure-activity relationship (QSAR) studies and in vitro studies in which correlations with LD50 have been sought are reviewed. QSAR methods have shown some success in relating LD50 to certain physicochemical properties of the compound, particularly lipophilicity, but have been less successful in correlating LD50 with electronic properties of molecules (related to reactivity) or structural variables. It is concluded that insufficient evidence is available to determine whether QSAR methods can be of general use in predicting the acute toxicity (LD50) of chemicals, and that until further work is undertaken to develop QSARs for a much wider range of homologous series of compounds, this situation is unlikely to be resolved. New chemical descriptors that are more directly relevant to the mechanism of toxic action of the chemical should be identified. Cytotoxicity in vitro is poorly correlated with LD50, but good correlations have been obtained between toxicity in vivo and in vitro, using systems in which the toxic endpoint reflects the probable mechanism(s) of acute toxicity of the test chemical (e.g. the assessment of neurotoxins using neural cell systems). Therefore, it seems that the successful application of in vitro methods requires a better understanding of the mechanisms of acute toxicity in vivo and the development of mammalian cell culture systems that can model more closely the metabolic fate of the chemicals in vivo.

Estimation of mutagenic/carcinogenic potential of environmental contaminants by ion-molecule reactions and tandem mass spectrometry

The ability to produce and detect products of model DNA/carcinogen ion-molecule reactions is demonstrated in the ion source and the collision cell of a triple quadrupole tandem mass spectrometer. Reaction between adenine and benzoyl chloride in the ion source is shown to produce the DNA adduct benzoyl adenine. The daughter ion mass spectrum of the reaction product is compared to that of the synthesized standard. Mass chromatograms of the reaction between mass-selected pyridine ions and various analytes eluting from a GC column into the collision cell are demonstrated and illustrate the ability to detect only the GC eluates that react with pyridine. This technique could provide a rapid and sensitive method for screening complex environmental samples for carcinogens, as well as for estimating the relative mutagenic/carcinogenic potential of environmental contaminants.

MEIC--a new international multicenter project to evaluate the relevance to human toxicity of in vitro cytotoxicity tests

A new international project to evaluate the relevance for human systemic and local toxicity of in vitro tests of general toxicity of chemicals has been organized by the Scandinavian Society of Cell Toxicology under the title Multicenter Evaluation of In Vitro Cytotoxicity (MEIC). The basic assumptions underlying the project, as well as the practical goals and the design of the program are outlined. The list of the first 50 reference chemicals is presented. The chemicals are an otherwise unbiased selection of compounds with known human acutely lethal dosage and blood concentrations, including LD50-values in the rat or mouse. Most agents also have other data on human toxicity and toxicokinetics, including more extensive animal toxicity data. International laboratories already using or developing in vitro tests of various partial aspects of general toxicity are invited to test the substances, the results of which will be evaluated by us. The predictivity of the in vitro results for both partial and gross human toxicity data will be determined with combined use of univariate regression analysis and soft multivariate modeling. The predictivity of the in vitro results will be compared with the predictivity of conventional animal tests for the same chemicals. Finally, batteries of tests with optimal prediction power for various types of human toxicity will be selected. The need for and possible uses of such batteries are discussed.

Nitrofuran inhibition of yeast and rat tissue glutathione reductases. Structure-activity relationships

Nitrofuran derivatives bearing unsaturated five- or six-membered nitrogen heterocycles or related substituents were more effective inhibitors of yeast and rat tissue glutathione reductases than those bearing other groups, such as nifurtimox, nitrofurazone and 5-nitro-2-furoic acid. The inhibitory action proved independent of electron withdrawal from the reduced enzyme, as a consequence of redoxcycling of the nitro group. Uncompetitive kinetics was obtained with nitrofurantoin and nifurtimox. Most of the assayed nitrofurans inhibited the yeast enzyme Coenzyme A glutathione disulfide reductase activity, though less than oxidized glutathione reduction. The transhydrogenase activity was not inhibited to a significant degree. Benznidazole (a 2-nitroimidazole derivative), 2-nitroimidazole, 5-nitroindole and chloramphenicol did not inhibit glutathione reductase. Under the same experimental conditions, liver glutathione peroxidase was not affected by the nitro compounds.

Evaluation of the In Vitro Uridine Uptake Inhibition Assay in Comparison with the In Vivo Eye Irritation Test as Prescribed by the EEC

In vivo eye irritation tests were carried out on 17 substances (mainly solvents) according to the experimental protocols laid down in the European Community legislation on dangerous substances and the OECD Guidelines. Erythema, chemosis, corneal opacity and iritis were observed and evaluated according to the interpretation rules laid down in EEC legislation.

The uridine uptake inhibition assay was carried out on Human Hep G2 cells according to the method of Shopsis & Sathe (1), modified by Dierickx & Martens (2). Inhibition of uridine uptake was expressed as the UI50 (Hep G2 cells, mM).

Comparison of the results of our two studies revealed only a poor relationship between the UI50 values and the mean scores obtained for corneal opacity (Spearman rank correlation; r=0.57, p<0.02). Consideration of just the monoalcohols, together with the ketones and the formamides, gave a good linear correlation between the UI50 (Hep G2 cells) and the <<Se>> (r=0.93). If a UI50 value of 50mM is accepted as a limit for classification of irritancy for the eye, no false positives or false negatives are obtained.

The UI50 can only be used within strictly defined chemical classes, as a first screening method for detecting possible irritants.