Development of a physiologically based toxicokinetic model for lead in pregnant women: The role of bone tissue in the maternal and fetal internal exposure.

Epidemiological studies have shown associations between prenatal exposure to lead (Pb) and neurodevelopmental effects in young children. Prenatal exposure is generally characterized by measuring the concentration in the umbilical cord at delivery or in the maternal blood during pregnancy. To assess internal Pb exposure during prenatal life, we developed a pregnancy physiologically based pharmacokinetic (p-PBPK) model that to simulates Pb levels in blood and target tissues in the fetus, especially during critical periods for brain development. An existing Pb PBPK model was adapted to pregnant women and fetuses. Using data from literature, both the additional maternal bone remodeling, that causes Pb release into the blood, and the Pb placental transfers were estimated by Bayesian inference. Additional maternal bone remodeling was estimated to start at 21.6 weeks. Placental transfers were estimated between 4.6 and 283 L.day-1 at delivery with high interindividual variability. Once calibrated, the p-PBPK model was used to simulate fetal exposure to Pb. Internal fetal exposure greatly varies over the pregnancy with two peaks of Pb levels in blood and brain at the end of the 1st and 3rd trimesters. Sensitivity analysis shows that the fetal blood lead levels are affected by the maternal burden of bone Pb via maternal bone remodeling and by fetal bone formation at different pregnancy stages. Coupling the p-PBPK model with an effect model such as an adverse outcome pathway could help to predict the effects on children's neurodevelopment.

A quantitative AOP of mitochondrial toxicity based on data from three cell lines.

Adverse Outcome Pathways (AOPs) are increasingly used to support the integration of in vitro data in hazard assessment for chemicals. Quantitative AOPs (qAOPs) use mathematical models to describe the relationship between key events (KEs). In this paper, data obtained in three cell lines, LHUMES, HepG2 and RPTEC/TERT1, using similar experimental protocols, was used to calibrate a qAOP of mitochondrial toxicity for two chemicals, rotenone and deguelin. The objectives were to determine whether the same qAOP could be used for the three cell types, and to test chemical-independence by cross-validation with a dataset obtained on eight other chemicals in LHUMES cells. Repeating the calibration approach for both chemicals in three cell lines highlighted various practical difficulties. Even when the same readouts of KEs are measured, the mathematical functions used to describe the key event relationships may not be the same. Cross-validation in LHUMES cells was attempted by estimating chemical-specific potency at the molecular initiating events and using the rest of the calibrated qAOP to predict downstream KEs: toxicity of azoxystrobin, carboxine, mepronil and thifluzamide was underestimated. Selection of most relevant readouts and accurate characterization of the molecular initiating event for cross-validation are critical when designing in vitro experiments targeted at calibrating qAOPs.

PBPK Modeling to Simulate the Fate of Compounds in Living Organisms.

Pharmacokinetics study the fate of xenobiotics in a living organism. Physiologically based pharmacokinetic (PBPK) models provide realistic descriptions of xenobiotics' absorption, distribution, metabolism, and excretion processes. They model the body as a set of homogeneous compartments representing organs, and their parameters refer to anatomical, physiological, biochemical, and physicochemical entities. They offer a quantitative mechanistic framework to understand and simulate the time-course of the concentration of a substance in various organs and body fluids. These models are well suited for performing extrapolations inherent to toxicology and pharmacology (e.g., between species or doses) and for integrating data obtained from various sources (e.g., in vitro or in vivo experiments, structure-activity models). In this chapter, we describe the practical development and basic use of a PBPK model from model building to model simulations, through implementation with an easily accessible free software.

Mapping blood lead levels in French children due to environmental contamination using a modeling approach.

The decrease in levels of lead in air and drinking water over the last 40 years has resulted in an overall decrease in blood lead levels (BLLs). However, there is no known safe level of lead regarding developmental effects in children. This paper maps predicted BLLs of children in France, resulting from a simulated chronic exposure in two steps, with the aim of identifying areas with environmentally overexposed populations. Probabilistic estimates of BLLs based on environmental contamination were obtained and compared to biomonitoring data. First, the contribution of various environmental exposure pathways was estimated using a multimedia exposure model: spatialized data on soil, drinking water and air contamination, together with data on food contamination and ingestion, was joined using geostatistical approaches. In a second step, a Physiologically Based Toxicokinetic (PBTK) model provided estimates of BLLs. Probabilistic estimates of BLLs were obtained by simulating uncertainty and variability of exposure levels, physiological characteristics and lead-specific parameters in the PBTK model. The median and 95th percentile of predicted BLLs in children aged 1 to 11 were compared to recent biomonitoring data obtained in France in young children (SATURNINF study): median predictions were overestimated in infants and in agreement with median observed BLLs in children aged 3 to 6. Upper bounds of predicted BLLs were protective due to uncertainties in exposure estimates. The main source of exposure appeared to be drinking water in children over 2 years old, and vegetal food and milk in children under 2 years old. Although elevated drinking water lead levels were not related to large geographical areas, the relatively fine resolution map also pinpointed geographical areas of concern due to elevated soil lead levels.

Application in a biomonitoring context of three-spined stickleback immunomarker reference ranges.

The relevance of a biomarker for biomonitoring programs was influenced both by the knowledge on biomarker natural inter-individual and site variabilities and by the sensitivity of the biomarker towards environmental perturbations. To minimize data misinterpretation, robustness reference values for biomarkers were important in biomonitoring programs. Specific three-spined stickleback, Gasterosteus aculeatus, immune reference ranges for field studies had been determined based on laboratory data and one reference station (Contentieuse river at Houdancourt). In this study, data obtained in one uncontaminated and three contaminated sites were compared to these reference ranges as a validation step before considering them for larger scale biomonitoring programs. When the field reference range were compared to data from the uncontaminated station (Béronelle), only few deviations were shown. In this way, data coming from uncontaminated station (Béronelle) was integrated in the field reference ranges to improve the evaluation of site variability. The new field reference ranges provided better discrimination of sites and spanned a larger range of fish lengths than the initial reference ranges. Furthermore, the results suggest lysosomal presence during several months and phagocytosis capacity in autumn may be the most relevant immunomarkers towards identifying contaminated sites. In the future, combining this reference value approach with active biomonitoring could facilitate the obtention of data in multiple stream conditions.

Modeling acetylcholine esterase inhibition resulting from exposure to a mixture of atrazine and chlorpyrifos using a physiologically-based kinetic model in fish.

Aquatic organisms are exposed to mixtures of chemicals that may interact. Mixtures of atrazine (ATR) and chlorpyrifos (CPF) may elicit synergic effects on the permanent inhibition of acetylcholinesterase (AChE) in certain aquatic organisms, causing severe damage. Mechanistic mathematical models of toxicokinetics and toxicodynamics (TD) may be used to better characterize and understand the interactions of these two chemicals. In this study, a previously published generic physiologically-based toxicokinetic (PBTK) model for fish was adapted to ATR and CPF. A sub-model of the kinetics of one of the main metabolites of CPF, chlorpyrifos-oxon (CPF-oxon), was included, as well as a TD model. Inhibition of two esterases, AChE and carboxylesterase, by ATR, CPF and CPF-oxon, was modeled using TD modeling of quantities of total and inactive esterases. Specific attention was given to the parameterization and calibration of the model to accurately predict the concentration and effects observed in the fish using Bayesian inference and published data from fathead minnow (Pimephales promelas), zebrafish (Danio rerio) and common carp (Cyprinus carpio L.). A PBTK-TD for mixtures was used to predict dose-response relationships for comparison with available adult fish data. Synergistic effects of a joint exposure to ATR and CPF could not be demonstrated in adult fish.

Water quality of the Meuse watershed: Assessment using a multi-biomarker approach with caged three-spined stickleback (Gasterosteus aculeatus L.).

The use of a multi-biomarker approach with three-spined sticklebacks (Gasterosteus aculeatus) through an active biomonitoring strategy appears to be a promising tool in water quality assessment. The present work proposes to assess the efficiency of these tools in the discrimination of some sites in a large scale on the Meuse basin in Europe. The study was part of an EU program which aims to assess water quality in the Meuse across the French-Belgian border. Sticklebacks were caged 21 days upstream and downstream from the wastewater treatment plants (WWTPs) of Namur (Belgium), Charleville-Mézières (France), Bouillon (Belgium) and Avesnes-sur-Helpe (France). First, the state of a variety of physiological functions was assessed using a battery of biomarkers that represented innate immunity (leucocyte mortality and distribution, phagocytosis activity, respiratory burst), antioxidant system (GPx, CAT, SOD and total GSH content), oxidative damages to the membrane lipids (TBARS), biotransformation enzymes (EROD, GST), synaptic transmission (AChE) and reproduction system (spiggin and vitellogenin concentration). The impacts of the effluents were first analysed for each biomarker using a mixed model ANOVA followed by post-hoc analyses. Secondly, the global river contamination was assessed using a principal component analysis (PCA) followed by a hierarchical agglomerative clustering (HAC). The results highlighted a small number of effects of WWTP effluents on the physiological parameters in caged sticklebacks. Despite a significant effect of the "localisation" factor (upstream/downstream) in the mixed ANOVA for several biomarkers, post-hoc analyses revealed few differences between upstream and downstream of the WWTPs. Only a significant decrease of innate immune responses was observed downstream from the WWTPs of Avesnes-sur-Helpe and Namur. Other biomarker responses were not impacted by WWTP effluents. However, the multivariate analyses (PCA and HAC) of the biomarker responses helped to clearly discriminate the different study sites from the reference but also amongst themselves. Thus, a reduction of general condition (condition index and HSI) was observed in all groups of caged sticklebacks, associated with a weaker AChE activity in comparison with the reference population. A strong oxidative stress was highlighted in fish caged in the Meuse river at Charleville-Mézières whereas sticklebacks caged in the Meuse river at Namur exhibited weaker innate immune responses than others. Conversely, sticklebacks caged in the Helpe-Majeure river at Avesnes-sur-Helpe exhibited higher immune responses. Furthermore, weak defence capacities were recorded in fish caged in the Semois river at Bouillon. This experiment was the first to propose an active biomonitoring approach using three-spined stickleback to assess such varied environments. Low mortality and encouraging results in site discrimination support the use of this tool to assess the quality of a large number of water bodies.

A two years field experiment to assess the impact of two fungicides on earthworm communities and their recovery.

Recent EFSA (European Food Safety Authority) reports highlighted that the ecological risk assessment of pesticides needed to go further by taking more into account the impacts of chemicals on biodiversity under field conditions. We assessed the effects of two commercial formulations of fungicides separately and in mixture, i.e., Cuprafor Micro® (containing 500 g kg-1 copper oxychloride) at 4 (C1, corresponding to 3.1 mg kg-1 dry soil of copper) and 40 kg ha-1 (C10), and Swing® Gold (50 g L-1 epoxiconazole EPX and 133 g L-1 dimoxystrobin DMX) at one (D1, 5.81 10-2 and 1.55 10-1 mg kg-1 dry soil of EPX and DMX, respectively) and ten times (D10) the recommended field rate, on earthworms at 1, 6, 12, 18 and 24 months after the application following the international ISO standard no. 11268-3 to determine the effects on earthworms in field situations. The D10 treatment significantly reduced the species diversity (Shannon diversity index, 54% of the control), anecic abundance (29% of the control), and total biomass (49% of the control) over the first 18 months of experiment. The Shannon diversity index also decreased in the mixture treatment (both fungicides at the recommended dose) at 1 and 6 months after the first application (68% of the control at both sampling dates), and in C10 (78% of the control) at 18 months compared with the control. Lumbricus terrestris, Aporrectodea caliginosa, Aporrectodea giardi, Aporrectodea longa, and Allolobophora chlorotica were (in decreasing order) the most sensitive species to the tested fungicides. This study not only addressed field ecotoxicological effects of fungicides at the community level and ecological recovery, but it also pinpointed some methodological weaknesses (e.g., regarding fungicide concentrations in soil and statistics) of the guideline to determine the effects on earthworms in field situations.

A generic PBTK model implemented in the MCRA platform: Predictive performance and uses in risk assessment of chemicals.

Physiologically-based toxicokinetic (PBTK) models are important tools for in vitro to in vivo or inter-species extrapolations in health risk assessment of foodborne and non-foodborne chemicals. Here we present a generic PBTK model implemented in the EuroMix toolbox, MCRA 9 and predict internal kinetics of nine chemicals (three endocrine disrupters, three liver steatosis inducers, and three developmental toxicants), in data-rich and data-poor conditions, when increasingly complex levels of parametrization are applied. At the first stage, only QSAR models were used to determine substance-specific parameters, then some parameter values were refined by estimates from substance-specific or high-throughput in vitro experiments. At the last stage, elimination or absorption parameters were calibrated based on available in vivo kinetic data. The results illustrate that parametrization plays a capital role in the output of the PBTK model, as it can change how chemicals are prioritized based on internal concentration factors. In data-poor situations, estimates can be far from observed values. In many cases of chronic exposure, the PBTK model can be summarized by an external to internal dose factor, and interspecies concentration factors can be used to perform interspecies extrapolation. We finally discuss the implementation and use of the model in the MCRA risk assessment platform.

The MCRA toolbox of models and data to support chemical mixture risk assessment.

A model and data toolbox is presented to assess risks from combined exposure to multiple chemicals using probabilistic methods. The Monte Carlo Risk Assessment (MCRA) toolbox, also known as the EuroMix toolbox, has more than 40 modules addressing all areas of risk assessment, and includes a data repository with data collected in the EuroMix project. This paper gives an introduction to the toolbox and illustrates its use with examples from the EuroMix project. The toolbox can be used for hazard identification, hazard characterisation, exposure assessment and risk characterisation. Examples for hazard identification are selection of substances relevant for a specific adverse outcome based on adverse outcome pathways and QSAR models. Examples for hazard characterisation are calculation of benchmark doses and relative potency factors with uncertainty from dose response data, and use of kinetic models to perform in vitro to in vivo extrapolation. Examples for exposure assessment are assessing cumulative exposure at external or internal level, where the latter option is needed when dietary and non-dietary routes have to be aggregated. Finally, risk characterisation is illustrated by calculation and display of the margin of exposure for single substances and for the cumulation, including uncertainties derived from exposure and hazard characterisation estimates.

Reliability evaluation of biomarker reference ranges for mesocosm and field conditions: Cellular innate immunomarkers in Gasterosteus aculeatus.

Due to their sensitivity to environmental contamination and their link with fish health status, innate immunomarkers are of great interest for environmental risk assessment studies. Nevertheless, the lack of knowledge about the effect of confounding factors can lead to data misinterpretation and false diagnostics. So, the determination of reference values was of huge interest for the integration of biomarkers in biomonitoring programs. Laboratory immunomarker reference ranges (including cellular mortality, leucocyte distribution, phagocytosis activity, respiratory burst and lysosomal presence) that consider three confounding factors (season, sex and body size) were previously developed in three-spined stickleback, Gasterosteus aculeatus, from our husbandry. Usefulness of these reference ranges in biomonitoring programs depends on how they can be transposed to various experimental levels, such as mesocosm (outdoor artificial pond) and field conditions. Immunomarkers were therefore measured every 2 months over 1 year in one mesocosm and in one site assumed to uncontaminated (Houdancourt, field). Differences between immunomarker seasonal variations in mesocosm and field fish on one side and laboratory fish on the other side were quantified: in some cases, seasonal trends were not significant or did not differ between mesocosm and laboratory conditions, but overall, models developed based on data obtained in laboratory conditions were poorly predictive of data obtained in mesocosm or field conditions. To propose valuable field reference ranges, mesocosm and field data were integrated in innate immunomarker modelling in order to strengthen the knowledge on the effect of confounding factors. As in laboratory conditions, sex was overall a confounding factor only for necrotic cell percentage and granulocyte-macrophage distribution and size was a confounding factor only for cellular mortality, leucocyte distribution and phagocytosis activity. Confounding factors explained a large proportion of immunomarker variability in particular for phagocytosis activity and lysosomal presence. Further research is needed to test the field models in a biomonitoring program to compare the sensitivity of immunomarkers to the confounding factors identified in this study and the sensitivity to various levels of pollution.

Investigating the interaction between melamine and cyanuric acid using a Physiologically-Based Toxicokinetic model in rainbow trout.

Following outbreaks of feed and food adulterations with a melamine and cyanuric acid mixture in 2007 and melamine in 2008 respectively, the kinetics and toxicodynamics of the mixture have been investigated particularly in sensitive species such as the rainbow trout. Tissue concentrations and intensity of the adverse effect, melamine-cyanurate crystal formation in kidney, were reported in similar experimental conditions. Here, a recent PBTK model for rainbow trout has been applied to model the kinetics of both single compounds based on residue levels in tissues. Both PBTK models for the single compounds were combined and a model of crystal formation for the mixture melamine-cyanuric acid was also added to predict the intensity of crystal formation under the assumptions that crystals formed either in urine or in kidney tissue. Modelling the kinetics of melamine and cyanuric acid provided a better understanding and prediction of intensity of crystal formation in case of sequential exposures with varying intensity or co-exposure. This study demonstrates, for the first time, how fish PBTK models can play a key role in the understanding and prediction of toxicokinetics and toxicodynamics of mixtures. This study also illustrates how adverse effects may potentially occur even when the compounds are not administered together as a mixture.

Modelling the effect of season, sex, and body size on the three-spined stickleback, Gasterosteus aculeatus, cellular innate immunomarkers: A proposition of laboratory reference ranges.

Innate immunomarkers reflect both environmental contamination and fish health status, providing useful information in environmental risk assessment studies. Nevertheless, the lack of knowledge about the effect of confounding factors can lead to data misinterpretation and false diagnoses. The aim of this study was to evaluate the impact of three confounding factors (season, sex and body size) on three-spined stickleback innate immunomarkers in laboratory conditions. Results shown strong seasonal variations in stickleback innate immunomarkers, with higher immune capacities in late winter-early spring and a disturbance during the spawning period in late spring-summer. Sex and body size had a season dependant effect on almost all tested immunomarkers. Reference ranges were established in laboratory-controlled conditions (i.e. laboratory reference ranges) and compared with data obtained from in vivo chemical expositions. The predictive power of the statistical model depended on the immunomarker, but the control data of the in vivo experiments, realized in same laboratory conditions, were globally well include in the laboratory reference ranges. Moreover, some statistical effects of the in vivo exposures were correlated with an augmentation of values outside the reference ranges, indicating a possible harmful effect for the organisms. As confounding factors influence is a major limit to integrate immunomarkers in biomonitoring programs, modelling their influence on studied parameter may help to better evaluated environmental contaminations.

Generic physiologically-based toxicokinetic modelling for fish: Integration of environmental factors and species variability.

One of the goals of environmental risk assessment is to protect the whole ecosystem from adverse effects resulting from exposure to chemicals. Many research efforts have aimed to improve the quantification of dose-response relationships through the integration of toxicokinetics. For this purpose, physiologically-based toxicokinetic (PBTK) models have been developed to estimate internal doses from external doses in a time-dependent manner. In this study, a generic PBTK model was developed and adapted for rainbow trout (Onchorhynchus mykiss), zebrafish (Danio rerio), fathead minnow (Pimephales promelas), and three-spined stickleback (Gasterosteus aculeatus). New mechanistic approaches were proposed for including the effects of growth and temperature in the model. Physiological parameters and their inter-individual variability were estimated based on the results of extensive literature searches or specific experimental data. The PBTK model was implemented for nine environmental contaminants (with log kow from -0.9 to 6.8) to predict whole-body concentrations and concentrations in various fish's organs. Sensitivity analyses were performed for a lipophilic and a hydrophilic compound to identify which parameters have most impact on the model's outputs. Model predictions were compared with experimental data according to dataset-specific exposure scenarios and were accurate: 50% of predictions were within a 3-fold factor for six out of nine chemicals and 75% of predictions were within a 3-fold factor for three of the most lipophilic compounds studied. Our model can be used to assess the influence of physiological and environmental factors on the toxicokinetics of chemicals and provide guidance for assessing the effect of those critical factors in environmental risk assessment.

Mixture Concentration-Response Modeling Reveals Antagonistic Effects of Estradiol and Genistein in Combination on Brain Aromatase Gene (cyp19a1b) in Zebrafish.

Comprehension of compound interactions in mixtures is of increasing interest to scientists, especially from a perspective of mixture risk assessment. However, most of conducted studies have been dedicated to the effects on gonads, while only few of them were. interested in the effects on the central nervous system which is a known target for estrogenic compounds. In the present study, the effects of estradiol (E2), a natural estrogen, and genistein (GEN), a phyto-estrogen, on the brain ER-regulated cyp19a1b gene in radial glial cells were investigated alone and in mixtures. For that, zebrafish-specific in vitro and in vivo bioassays were used. In U251-MG transactivation assays, E2 and GEN produced antagonistic effects at low mixture concentrations. In the cyp19a1b-GFP transgenic zebrafish, this antagonism was observed at all ratios and all concentrations of mixtures, confirming the in vitro effects. In the present study, we confirm (i) that our in vitro and in vivo biological models are valuable complementary tools to assess the estrogenic potency of chemicals both alone and in mixtures; (ii) the usefulness of the ray design approach combined with the concentration-addition modeling to highlight interactions between mixture components.

High-Throughput Analysis of Ovarian Cycle Disruption by Mixtures of Aromatase Inhibitors.

BACKGROUND: Combining computational toxicology with ExpoCast exposure estimates and ToxCast™ assay data gives us access to predictions of human health risks stemming from exposures to chemical mixtures. OBJECTIVES: We explored, through mathematical modeling and simulations, the size of potential effects of random mixtures of aromatase inhibitors on the dynamics of women's menstrual cycles. METHODS: We simulated random exposures to millions of potential mixtures of 86 aromatase inhibitors. A pharmacokinetic model of intake and disposition of the chemicals predicted their internal concentration as a function of time (up to 2 y). A ToxCast™ aromatase assay provided concentration-inhibition relationships for each chemical. The resulting total aromatase inhibition was input to a mathematical model of the hormonal hypothalamus-pituitary-ovarian control of ovulation in women. RESULTS: Above 10% inhibition of estradiol synthesis by aromatase inhibitors, noticeable (eventually reversible) effects on ovulation were predicted. Exposures to individual chemicals never led to such effects. In our best estimate, ∼10% of the combined exposures simulated had mild to catastrophic impacts on ovulation. A lower bound on that figure, obtained using an optimistic exposure scenario, was 0.3%. CONCLUSIONS: These results demonstrate the possibility to predict large-scale mixture effects for endocrine disrupters with a predictive toxicology approach that is suitable for high-throughput ranking and risk assessment. The size of the effects predicted is consistent with an increased risk of infertility in women from everyday exposures to our chemical environment. https://doi.org/10.1289/EHP742.

Analysis of community-level mesocosm data based on ecologically meaningful dissimilarity measures and data transformation.

The principal response curve (PRC) method is a constrained ordination method developed specifically for the analysis of community data collected in mesocosm experiments, which provides easily understood summaries and graphical representations of community response to stress. It is a redundancy analysis method and is usually performed on log-transformed abundance data. The choice of a measure of dissimilarity between samples and the choice of the data transformation significantly affect the results of multivariate analysis. Dissimilarity measures that are more ecologically meaningful than the Euclidean distance can be incorporated into the PRC using distance-based redundancy analysis. The present study investigates the ordinations produced by a small selection of dissimilarity measures: the Euclidean distance using log-transformed and Hellinger-transformed data and the Bray-Curtis dissimilarity using raw and log-transformed data. It compares 2 data sets from experiments on the effect of the anti-inflammatory drug diclofenac and the insecticide chlorpyrifos on macroinvertebrate communities. The choice of dissimilarity measure can determine the outcome of a risk assessment. For the diclofenac data set, the PRCs were different depending on the dissimilarity measure: the community no-effect concentration was lowest for the Bray-Curtis on log-transformed data and Hellinger dissimilarity measures. For chlorpyrifos, however, the PRCs were similar for all dissimilarity measures. Environ Toxicol Chem 2017;36:1667-1679. © 2016 SETAC.

Effects of bisphenol A on different trophic levels in a lotic experimental ecosystem.

Bisphenol A (BPA) is commonly used by manufacturers and can be found in many aquatic ecosystems. Data relative to BPA ecotoxicity are only available for studies in laboratory conditions on macro-invertebrates and fish. There is thus a lack of information for other trophic levels such as macrophytes. Moreover, the impacts of BPA within an ecosystem context, i.e. with populations from different trophic levels studied at long term in environmental conditions, have never been assessed. We carried out a long-term lotic mesocosm study in 20 m long channels under three exposure concentrations of BPA (nominal concentrations of 0, 1, 10 and 100 µg/L) delivered continuously for 165 days. Three trophic levels were followed: macrophytes, macro-invertebrates (with a focus on Radix balthica) and fish (Gasterosteus aculeatus). Significant effects were shown at 100 µg/L BPA on the three trophic levels. BPA had a direct impact on macrophyte community structure, direct and indirect impacts on macro-invertebrates and on fish population structure. Gonad morphology of fish was affected at 1 and 10 µg/L of BPA, respectively for female and male sticklebacks. In addition to these ecotoxicity data, our results suggest that fish are good integrators of the responses of other communities (including macro-invertebrates and macrophytes) in mesocosm systems.

Modelling Structure Activity Landscapes with Cliffs: a Kernel Regression-Based Approach.

Quantitative Structure-Activity Relationship (QSAR) models are increasingly used in hazard and risk assessment. Even when models with linear relationships between activity and a small number of descriptors are built and validated regarding predictivity and statistical assumptions, similar structures can exhibit large differences in activity known as similarity paradoxes or activity cliffs. In order to reduce the impact that similarity paradoxes can have on predictions we have devised a statistical method based on Nadaraya-Watson kernel regression. According to our method, activity cliffs filter out contributions of neighbouring chemicals especially along the cliff axis. Our method decreases density-based certainty in particular for chemicals with strong prediction errors and the implementation of Structure-Activity Landscape Index (SALI) curves shows that our method improves the prediction of activity cliff ranks. We also provide useful indications on the density-based applicability domain and the reliability of individual predictions.

A Kernel-Based Method for Assessing Uncertainty on Individual QSAR Predictions.

The assessment of uncertainty attached to individual predictions is now a priority for sound decision-making in risk assessment. QSAR predictive uncertainty is affected by a variety of factors related to the quality of the training set data, the adopted statistical models, and the distance between the query chemical and the training set. We developed a method to quantify uncertainty associated with individual linear QSAR predictions that integrates both model and experimental error uncertainty and that defines an applicability domain based on the density of training set data. Our method is based on chemical spaces defined by latent variables identified by Partial Least Squares (PLS) regressions. The method provides a kernel regression estimate of the activity of interest as well as a measure of predictive uncertainty based on a mathematical estimation of the domain of applicability and on local propagation of uncertainty associated with training set data.