Historical control data of rare events: issues, chronological patterns and their relevance for toxicological evaluations.

Historical control data (HCD) give context for a measurement by providing a biological reference frame. HCD are used in the evaluation of toxicological bioassays for quality and performance control, informal statistical false discovery rate mitigation, and to estimate the biological relevance of observed potentially adverse findings. The current commentary shortly highlights 5 points that should be considered when working with HCD of rare events: 1) HCD database (HCDB) size, 2) the issue of rare events, 3) potential chronological patterns, 4) using point estimates to summarize HCD and 5) independence from treatment bias, i.e., HCD are mostly informative for primary toxicity. It is argued to use exploratory data analysis and to apply ad hoc time windows for assessment based on an HCDB that is as large as possible to monitor for potential structure and systemic bias in the data.

The database makes the poison: How the selection of datasets in QSAR models impacts toxicant prediction of higher tier endpoints.

As the United States and the European Union continue their steady march towards the acceptance of new approach methodologies (NAMs), we need to ensure that the available tools are fit for purpose. Critics will be well-positioned to caution against NAMs acceptance and adoption if the tools turn out to be inadequate. In this paper, we focus on Quantitative Structure Activity-Relationship (QSAR) models and highlight how the training database affects quality and performance of these models. Our analysis goes to the point of asking, "are the endpoints extracted from the experimental studies in the database trustworthy, or are they false negatives/positives themselves?" We also discuss the impacts of chemistry on QSAR models, including issues with 2-D structure analyses when dealing with isomers, metabolism, and toxicokinetics. We close our analysis with a discussion of challenges associated with translational toxicology, specifically the lack of adverse outcome pathways/adverse outcome pathway networks (AOPs/AOPNs) for many higher tier endpoints. We recognize that it takes a collaborate effort to build better and higher quality QSAR models especially for higher tier toxicological endpoints. Hence, it is critical to bring toxicologists, statisticians, and machine learning specialists together to discuss and solve these challenges to get relevant predictions.

Proposals for new spray drift exposure values in orchards and vineyards for residents and bystanders.

In the EU, predicted exposure to spray drift for residents and bystanders from applications in orchards and vineyards is based on data from one study published in 1987, where one downwind distance (8 m) was considered. CropLife Europe conducted sixteen new GLP compliant studies in 4 EU countries, 8 in orchards, 8 in vineyards with early and late season applications, using adult and child mannequins located 5, 10 and 15 m downwind from the last row to measure dermal and inhalation exposures. The resulting "Bystander Resident Orchard Vineyard (BROV)" database comprises 288 observations and offers a more comprehensive option for exposure prediction. There were differences between adult and child, crop type, leaf cover and distance from the sprayer, supporting the derivation of mean, median, 75th and 95th percentile exposures for each subset. Exposures did not generally correlate with wind speed, wind direction, sprayer type, spray quality, spray concentration or amount applied. Dermal and inhalation exposure were lower in vineyards than in orchards and further analysis is required to understand why.

Proposals for new transfer coefficient (TC) values for worker re-entry activities in grape vineyards.

New transfer coefficient (TC) values were derived for vineyard workers handling treated grapevines during harvesting and crop maintenance activities. Re-entry exposure and dislodgeable foliar residue (DFR) studies were performed in Europe, covering hand harvesting, pruning/training, pruning/tying and pruning/shoot lifting. Foliar applications of fungicides (iprovalicarb, dimethomorph, dithianon, pyrimethanil and fenbuconazole) were made and 73 workers at 16 sites were monitored over one working day. Exposure was measured on inner and outer dosimeters, face/neck wipes and hand washes. In concurrent DFR studies, leaf punches were taken at each site during the time of worker re-entry. Potential exposure values correlated well with DFR values. TC values were derived for various re-entry activities for potential and actual exposure, with and without gloves. The harvesting task resulted in lower TC values than the other crop maintenance tasks. Additional TC values reflecting the use of protective gloves can be derived from the results. The TC values are much lower than current European Food Safety Authority (EFSA) default values. This project addresses a data gap identified by EFSA for specific EU TC values to permit more realistic and reliable re-entry worker exposure estimates for grapes.

Use of guinea pig data to obtain starting points for skin sensitisation risk assessment - A commentary.

The increasing drive to understand the likelihood of skin sensitisation from plant protection products (PPPs) in workers and the general public has resulted in recent initiatives to establish a quantitative risk assessment (QRA) methodology applicable to these products and their exposure scenarios. The effective evaluation of skin sensitising substances requires not only the identification of that toxicological hazard, but also determination of relative sensitising potency. Typically, this has been achieved by interpretation of local lymph node assay (LLNA) dose response data, delivering what is known as the EC3 value. This permitted regulatory division of skin sensitisers into defined potency sub-categories, but more importantly enabled derivation of a no expected sensitisation induction level (NESIL) as the point of departure for QRA. However, for many existing substances there is no LLNA data, only older guinea pig results exist. To avoid additional (in vivo) testing, an approach has been outlined to employ guinea pig data and existing regulatory guidelines on the determination of potency sub-categorisation to provide a guinea pig based NESIL. The approach adopts a conservative extrapolation from LLNA NESIL benchmarks to deliver points of departure as the basis for the type of QRA process already in successful use by other industries.

Difficulties in translating in vitro hazards of local acute irritants to relevant human risk: Learnings from Captan and Folpet.

In order to assess the regulatory value of New Approach Methodologies (NAMs), authors should provide their opinion on the physiological and exposure relevance of observed in vitro effects for correlation with predicted in vivo effects. Further, peer-reviewers should be encouraged to request such information during review. This is critical to scientifically transition to animal-free, reliable, robust and -- most importantly -- relevant regulatory toxicology and risk assessment approaches. Recently published studies using NAMs for the fungicides Captan and Folpet illustrate the difficulties and limitations of applying NAMs to adequately assess the toxicological relevance of these substances.

Understanding and overcoming the technical challenges in using in silico predictions in regulatory decisions of complex toxicological endpoints - A pesticide perspective for regulatory toxicologists with a focus on machine learning models.

There are many challenges that must be overcome before in silico toxicity predictions are ripe for regulatory decision-making. Today, mandates in the United States of America and the European Union to avoid animal usage in toxicity testing is driving the need to consider alternative technologies, including Quantitative Structure Activity Relationship (QSAR) models, and read across approaches. However, when adopting new methods, it is critical that both new approach developers as well as regulatory users understand the strengths and challenges with these new approaches. In this paper, we identify potential sources of bias in machine learning methods specific to toxicity predictions, that may impact the overall performance of in silico models. We also discuss ways to mitigate these biases. Based on our experiences, the most prevalent sources of bias include class imbalance (differing numbers of "toxic" vs "nontoxic" compounds), limited numbers of chemicals within a particular chemistry, and biases within the studies that make up the database used for model building, as well as model evaluation biases. While this is already complex for repeated dose toxicity, in reproduction and developmental toxicity a further level of complexity is introduced by the need to evaluate effects on individual animal and litter basis (e.g., a hierarchal structure). We also discuss key considerations developers and regulators need to make when they use machine learning models to predict chemical safety. Our objective is for our paper to serve as a desk reference for model developers and regulators as they evaluate machine learning models and as they make decisions using these models.

Assessing the risk of induction of skin sensitization to plant protection products: A quantitative approach.

Exposure to skin sensitizers is common and regulated in many industry sectors. For cosmetics, a risk-based approach has been implemented, focused on preventing the induction of sensitization. First, a No Expected Sensitization Induction Level (NESIL) is derived, then modified by Sensitization Assessment Factors (SAFs) to derive an Acceptable Exposure Level (AEL). The AEL is used in risk assessment, being compared with an estimated exposure dose, specific to the exposure scenario. Since in Europe there is increased concern regarding exposure towards potentially sensitizing pesticides via spray drift, we explore how existing practice can be modified to allow Quantitative Risk Assessment (QRA) of pesticides for bystanders and residents. NESIL derivation by the Local Lymph Node Assay (LLNA), the globally required in vivo assay for this endpoint, is reviewed alongside consideration of appropriate SAFs. Using a case study, the principle that the NESIL in µg/cm2 can be derived by multiplying LLNA EC3% figure by a factor of 250 is adopted. The NESIL is then reduced by an overall SAF of 25 to establish an exposure level below which there is minimal bystander and resident risk. Whilst this paper focuses on European risk assessment and management, the approach is generic and universally applicable.

Practical guidance to evaluate in vitro dermal absorption studies for pesticide registration: An industry perspective.

While there are some regulatory assessment criteria available on how to generally evaluate dermal absorption (DA) studies for risk assessment purposes, practical guidance and examples are lacking. The current manuscript highlights the challenges in interpretating data from in vitro assays and proposes holistic data-based assessment strategies from an industry perspective. Inflexible decision criteria may be inadequate for real data and may lead to irrelevant DA estimates. We recommend the use of mean values for reasonably conservative DA estimates from in vitro studies. In cases where additional conservatism is needed, e.g., due to non-robust data and acute exposure scenarios, the upper 95% confidence interval of the mean may be appropriate. It is critical to review the data for potential outliers and we provide some example cases and strategies to identify aberrant responses. Some regional regulatory authorities require the evaluation of stratum corneum (SC) residue, but here, as a very simple pro-rata approach, we propose to review whether the predicted post 24-h absorption flux exceeds the predicted elimination flux by desquamation because otherwise it is not possible for the SC residue to contribute to systemic dose. Overall, the adjustment of DA estimates due to mass balance (normalization) is not recommended.

A new conceptional model for deriving average dermal absorption estimates from studies with multiple tested concentrations for non-dietary risk assessment of pesticides.

Dermal absorption values are used to translate external dermal exposure into potential systemic exposure for non-dietary risk assessment of pesticides. While the Environmental Protection Agency of the United States of America (US EPA) derives a common dermal absorption factor for active substances covering all related products, the European Food Safety Authority (EFSA) requests specific product-based estimates for individual concentrations covering the intended use rates. The latter poses challenges, because it disconnects exposure dose from applied dose in absorption studies, which may not be suitable in scenarios where concentration is not relevant. We analyzed the EFSA dermal absorption database, collected 33 human in vitro studies from CropLife Europe (CLE) companies, where ≥3 in-use dilution concentrations were tested, and 15 dermal absorption triple pack datasets. This shows that absolute dermal absorption correlates with absolute applied dose on a decadic logarithm-scale, which is concordant with the toxicological axiom that risk is driven by exposure dose. This method is radically different from the current European approach focused on concentrations and offers new insights into the relationship of internal and external exposure doses when utilizing data from in vitro studies. A single average dermal absorption value can be simply derived from studies with multiple tested concentrations, by calculating the y-intercept of a linear model on a decadic logarithm scale while assuming a slope of 1. This simplifies risk assessment and frees resources to explore exposure refinements. It also serves as a basis to harmonize dermal absorption estimation globally for use in exposure-driven risk assessments.

Review of air concentrations of pesticides for estimating exposure to vapour in European risk assessments.

In current European non-dietary risk assessment for bystanders and residents, one of the plant protection product exposure pathways to be addressed is vapour inhalation. At present, active ingredients are grouped according to vapour pressure and assigned corresponding values. Risk assessments are driven by only two default air concentration values. Sampling is inconsistent, background data are sparse and many factors having an impact on air concentrations are not considered. Within the changing regulatory landscape over the last 20 years, criteria for volatility grouping and consequently for vapour exposure estimation have been applied heterogeneously. Here we review the background data currently used in the exposure assessment guidance to demonstrate the arbitrary nature of derived air concentration values and their inconsistent application in exposure assessment. In doing so we discuss air concentration from a risk assessment perspective and how active ingredients are grouped according to vapour pressure. We examine the database which at present forms the basis for predicting inhalation exposure to PPPs, particularly the two concentration levels driving risk assessments, and we discuss several other factors having an impact on air concentration. In conclusion, we recommend an urgent revision of the default air concentration values and assumptions applied in assessing vapour exposure.

Measured air concentrations of pesticides for the estimation of exposure to vapour in European risk assessments.

There is an identified need to revise the default air concentration values and assumptions applied in assessing vapour exposure in the risk assessment of bystanders and residents to plant protection products. To address this, we evaluated inhalation exposure via vapour using previously unpublished data from 29 field and wind tunnel studies. The database comprises 35 trials with 11 active ingredients covering a wide range of scenarios with respect to vapour pressure, crops, application rates and European regions. Of the 961 individual measurements, 634 were below the Limit of Detection (LOD), 282 were between the LOD and Limit of Quantification (LOQ) and only 45 (4.7%) were quantifiable. Ten individual non-normalized samples exceeded 0.1 µg/m³. Of the 81 first-day measurements after the application, 36 were

An OECD TG 428 study ring trial with 14C-Caffeine demonstrating repeatability and robustness of the dermal absorption in vitro method.

The dermal absorption potential of 14C-Caffeine applied as a 4 mg/mL concentration (10 µL/cm2 finite dose) was investigated in six laboratories under Good Laboratory Practice conditions using an OECD TG 428-compliant in vitro assay with flow-through cells and split-thickness human skin. Potential sources of variation were reduced by a standardized protocol, test item and skin source. Particularly, skin samples from same donors were distributed over two repeats and between labs in a non-random, stratified design. Very similar recovery was achieved in the various assay compartments between laboratories, repeats and donors, demonstrating that the assay can be robustly and reliably performed. The absorption in one laboratory was 5-fold higher than in the others. This did not clearly correlate with skin integrity parameters but might be associated with an accidental COVID-19 pandemic-related interruption in sample shipment. It is possible that other factors may affect dermal absorption variation not routinely assessed or considered in the current method. The mean receptor fluid recovery, potential absorption (recovery in receptor fluid and skin except tape strips 1 and 2) and mass balance of caffeine was 6.99%, 7.14% and 99.13%, respectively, across all and 3.87%, 3.96% and 99.00% in the subset of five laboratories.

Expanding the toxicologist's statistical toolbox: Using effect size estimation and dose-response modelling for holistic assessments instead of generic testing.

It is tempting to base (eco-)toxicological assay evaluation solely on statistical significance tests. The approach is stringent, objective and facilitates binary decisions. However, tests according to null hypothesis statistical testing (NHST) are thought experiments that rely heavily on assumptions. The generic and unreflected application of statistical tests has been called "mindless" by Gigerenzer. While statistical tests have an appropriate application domain, the present work investigates how unreflected testing may affect toxicological assessments. Dunnett multiple-comparison and Williams trend testing and their compatibility intervals are compared with dose-response-modelling in case studies, where data do not follow textbook behavior, nor behave as expected from a toxicological point of view. In such cases, toxicological assessments based only on p-values may be biased and biological evaluations based on plausibility may be prioritized. If confidence in a negative assay outcome cannot be established, further data may be needed for a robust toxicological assessment.

Metribuzin-induced non-adverse liver changes result in rodent-specific non-adverse thyroid effects via uridine 5'-diphospho-glucuronosyltransferase (UDPGT, UGT) modulation.

Metribuzin is a herbicide that inhibits photosynthesis and has been used for over 40 years. Its main target organ is the liver and to some extent the kidney in rats, dogs, and rabbits. Metribuzin shows a specific thyroxine (T4) profile in rat studies with T4 increases at low doses and T4 decreases at higher doses. Only the T4 decreases occur together with histopathological changes in the thyroid and weight changes of liver and thyroid. A set of experiments was conducted to investigate metribuzin's endocrine disruptor potential according to European guidance and regulations. The results indicate that a liver enzyme modulation, i.e. of the uridine 5'-diphospho-glucuronosyltransferase (UDPGT, UGT), is most likely responsible for both increased and decreased plasma thyroxine level and for thyroid histopathological observations. Animals with high T4 levels show low UGT activity, while animals with low T4 levels show high UGT activity. A causal relationship was inferred, since other potentially human-relevant mode of action (MOA) pathways were excluded in dedicated studies, i.e. inhibition of deiodinases (DIO), inhibition of thyroid peroxidase (TPO) or of the sodium importer system (NIS). This liver metabolism-associated MOA is considered not relevant for human hazard assessment, due to species differences in thyroid homeostasis between humans and rats and, more importantly, based on experimental data showing that metribuzin affects UGT activity in rat but not in human hepatocytes. Further, we discuss whether or not increased T4 levels in the rat, in the absence of histopathological changes, should be considered as adverse and therefore used as an appropriate hazard model for humans. Based on a weight of evidence approach, metribuzin should not be classified as an endocrine disruptor with regard to the thyroid modality.

Compounded conservatism in European re-entry worker risk assessment of pesticides.

We review the risk parameters and drivers in the current European Union (EU) worker risk assessment for pesticides, for example considering crop maintenance, crop inspection or harvesting activities, and show that the current approach is very conservative due to multiple worst-case default assumptions. As a case study, we compare generic exposure model estimates with measured worker re-entry exposure values which shows that external cumulative exposure is overpredicted by about 50-fold on average. For this exercise, data from 16 good laboratory practice (GLP)-compliant worker exposure studies in 6 crops were evaluated with a total number of 184 workers. As generic overprediction does not allow efficient risk management or realistic risk communication, we investigate how external exposure can be better predicted within the generic model, and outline options for possible improvements in the current methodology. We show that simply using averages achieves more meaningful exposure estimates, while still being conservative, with an average exposure overprediction of about 9-fold. Overall, EU risk assessment includes several numerically unaccounted "hidden safety factors", which means that workers are well protected; but simultaneously risk assessments are biased towards failing due to compounded conservatism. This should be considered for further global or regional guidance developments and performing more exposure-relevant risk assessment.

Using historical control data in bioassays for regulatory toxicology.

Historical control data (HCD) consist of pooled control group responses from bioassays. These data must be collected and are often used or reported in regulatory toxicology studies for multiple purposes: as quality assurance for the test system, to help identify toxicological effects and their effect-size relevance and to address the statistical multiple comparison problem. The current manuscript reviews the various classical and potential new approaches for using HCD. Issues in current practice are identified and recommendations for improved use and discussion are provided. Furthermore, stakeholders are invited to discuss whether it is necessary to consider uncertainty when using HCD formally and statistically in toxicological discussions and whether binary inclusion/exclusion criteria for HCD should be revised to a tiered information contribution to assessments. Overall, the critical value of HCD in toxicological bioassays is highlighted when used in a weight-of-evidence assessment.

A Nonmechanistic Parametric Modeling Approach for Benchmark Dose Estimation of Event-Time Data.

We propose benchmark dose estimation for event-time data, using a two-step approach. This approach avoids estimation of complex models and has been previously shown to give robust results for summarizing relevant parameters for risk assessment. In the first step, the probability of the event of interest to occur (in a certain time interval) is described as a function of time, resulting in an event-time model; such a model is fitted allowing an individual curve for each dose, and relevant estimates are extracted. In the second step, a dose-response model is fitted to the estimates of t50 obtained from the event-time model in the first step. Given a predefined benchmark response, the benchmark dose is then estimated from the resulting model. This novel approach is demonstrated in two examples. Our application of the time-to-event model showed a gain in power compared to the traditional analysis of end-of-study summary data.

Alternatives to statistical decision trees in regulatory (eco-)toxicological bioassays.

The goal of (eco-) toxicological testing is to experimentally establish a dose or concentration-response and to identify a threshold with a biologically relevant and probably non-random deviation from "normal". Statistical tests aid this process. Most statistical tests have distributional assumptions that need to be satisfied for reliable performance. Therefore, most statistical analyses used in (eco-)toxicological bioassays use subsequent pre- or assumption-tests to identify the most appropriate main test, so-called statistical decision trees. There are however several deficiencies with the approach, based on study design, type of tests used and subsequent statistical testing in general. When multiple comparisons are used to identify a non-random change against negative control, we propose to use robust testing, which can be generically applied without the need of decision trees. Visualization techniques and reference ranges also offer advantages over the current pre-testing approaches. We aim to promulgate the concepts in the (eco-) toxicological community and initiate a discussion for regulatory acceptance.

Scatter plotting as a simple tool to analyse relative organ to body weight in toxicological bioassays.

In toxicological bioassays, organ weight is often expressed as a ratio to body weight or another denominator to account for natural differences in animal sizes. However, the relationship of treatment-induced organ and body weight change is complicated and relative weights may accordingly confound a toxicological assessment. In addition, the statistical assessment of relative weights is challenging. The examples given in this document show that toxicological interpretation of organ weight data in relation to body weight can be vastly improved by simple bivariate scatter plotting. Conversely, plots of relative organ weight are of limited value and may lead to an incorrect interpretation of toxic effects when used in isolation. Scatter plots are useful for qualitative hazard characterization and to generate hypotheses. Bivariate summary statistics indicate effect levels and help to explore the actual correlation of organ to body weight.