Application of the benchmark dose-response modelling approach for risk characterization of chemicals

Abstract

Toxicology is the discipline that investigates the possible adverse effects of chemical exposure on human, animal and environmental health. Chemical risk assessment is the process that aims to identify potentially hazardous substances and describes the probability of adverse outcomes associated with their exposure. Biological changes and adverse effects do not occur after a threshold level is surpassed, but gradually and following a sequence of linked events. Traditionally, the no-observed-adverse-effect-level (NOAEL) approach has been used to detect the highest dose at which no adverse effect was observed. However, the NOAEL approach has methodological limitations and disadvantages that have resulted in it being increasingly replaced by the scientifically more advanced benchmark dose (BMD) approach. The BMD-modelling approach is a flexible method that takes all uncertainty and variability in the data into account, providing better estimates of doses leading to the potential adverse effects. Nonetheless, there are a number of knowledge gaps that need to be addressed and a lack of consensus persists regarding certain methodological aspects of this modelling strategy.

The overall aim of this thesis was to contribute to the BMD field and expand the knowledge base by applying this approach to the areas of risk assessment and pharmaceutical development, addressing some identified challenges and discussing potential improvements. In particular, this thesis covers three topics that are interconnected, namely the choice of the Critical Effect Size (CES) (study I and III to VI), the analysis of multiple endpoints (study II to VI) and the assessment of chemical mixtures (study I, V and VI). These topics were applied to data from studies on chemicals, namely per- and polyfluoroalkyl substances (PFAS) (study I and VI), polychlorinated biphenyls (PCBs) (study IV and V) and a candidate drug in pharmaceutical development (study II and III) and the pesticide norflurazon (study VI).

Study I combined human and animal data in order to derive the probabilistic risk for a 10% decrease in total triiodothyronine (T3) hormone levels, depending on residency time. The human data consisted of perfluorooctanesulfonic acid (PFOS) and perfluorohexanesulfonic acid (PFHxS) serum levels from the resident population in Ronneby, a Swedish village that was highly exposed to PFAS through contaminated drinking water. The animal data originated from a 6-month subchronic study in monkeys, exposed to PFOS once a day. This integrated probabilistic risk assessment (IPRA) analysis demonstrated that longer exposure periods were associated with a larger proportion of the population at risk, ranging from 2.1% (90% C.I. 0.4% – 13.1%) to 3.5% (90% C.I. 0.7% – 21.8%) for residents exposed to PFOS and PFHxS for at least 1 or 29 years, respectively. This risk was mostly distributed among women, and exposure duration was thegreatest source of uncertainty (60.8%). It was concluded that IPRA is an advantageous method to calculate the risk for adverse effects, in comparison to the deterministic Margin of Exposure aproach (MoE).

Study II analyzed data from three subsequential safety assessment studies performed in rats to investigate the potential toxicity of an anti-oncogenic candidate drug in pharmaceutical development. The partial least squares (PLS) modelling approach was used to detect associations between clinical signs observed during the study, a 5% body weight decrease and pathological findings noted after study termination. Piloerection, eyes half shut and slightly decreased motor activity were the signs that were most strongly associated with the pathological findings, and the models accurately predicted the injuries observed in the thymus, testes, epididymides and bone marrow. The findings indicate that an evaluation of clinical signs as an integrated toxicity evaluation has potential 3R (Replacement, Reduction and Refinement of animal use) gains, especially in terms of Refinement of animal studies. The study suggests that the PLS-modelling approach can be employed to predict pathological changes, monitor animal welfare and support the decision-making process during pre-clinical safety and toxicity assessment studies.

Study III analyzed the same data as study II, but applied the BMD-modelling approach instead, with a different objective, namely to describe potential relationships between the dose and the findings made in the 63 examined endpoints. The endpoints modelled included biochemistry and hematology endpoints, body weight changes, organ pathology findings and clinical observations. The resulting BMDs and BMDLs were compared to the study NOAEL (or LOAEL) and were often lower than the estimates of the NOAEL approach. A 5% change was also compared to the findings based on an adversity threshold derived from the observed and endpoint-specific magnitude of change. Additionally, the BMD-modelling was also considered to have a strong focus on the Refinement of animal studies. In summary, it was shown that modelling multiple endpoints is desirable, providing a more complete overview of the potential toxicity of a candidate drug and improving the pharmaceutical development process.

Study IV assessed the potential toxicity of PCB-156 (2,3,3′,4,4′,5-hexachlorobiphenyl) following a 90-day study in rats exposed daily through their diet. Dose-dependent toxicological effects were described, including body and organ changes but also in the assessed retinoid system endpoints. Retinoid disruption and effects in the organs of rats were demonstrated employing the BMD dose-response modelling approach, revealing that the apolar liver retinoid concentrations were the most sensitive endpoint. The retinoid system was shown to be sensitive to PCB-156 exposure, and it was suggested that its endpoints should be more often considered for chemical risk assessment purposes.

Study V employed the BMD method to calculate relative potency factors (RPFs) for seven PCBs (PCB-28, 77, 105, 118, 128, 153 and 156) and one PCB-mixture. PCB-126 was used as an index chemical, and the eight 90-day regulatory toxicity studies for the individual congeners were performed under the same conditions (the PCB-mixture study was 28 days long). The liver apolar retinoids levels and concentration, and the remaining endpoints examined, estimated greater RPFs than those calculated by the World Health Organization (WHO) in 2006 (Van den Berg et al., 2006), being suggestive of a hazard underestimation. In fact, the potency factors estimated in this study, based on the ethoxyresorufin-O-deethylase (EROD) enzymatic activity (a historically used endpoint to calculate RPFs), were the lowest in comparison to other endpoints for which RPFs were calculated. In summary, RPFs were useful to describe the potential toxicity of structurally similar compounds, expressed in units equivalent to the index chemical, and the retinoid system proved once again to be susceptible to changes following low-dose PCB exposure.

Study VI focused on the choice of CES, a matter of debate when applying the BMD method to continuous data. Currently, there is no internationally harmonized approach to choosing the CES, and five strategies were examined: the EFSA default value of 5% or 10%, the US EPA 1 SD approach, an endpoint-specific CES based on historical data, the General Theory of Effect Size (GTES) and expert judgment. All examined strategies featured advantages and limitations, and the different choices of CES led to distinct reference values when applied to five case-studies, analyzing PFAS, PCB-156 and a pesticide (norflurazon) data. Although some of these strategies delivered similar CES values, it was not always the case, and reliance on a single method to choose the CES is not recommendable. It was concluded that expert judgment is irreplaceable and that the decision-making process performed by risk assessors and managers regarding the likely threshold of adversity should be supported by BMD analysis of the data comparing different CES. This could lead to a better overview of the data package and understanding of the doses leading to different magnitudes of effects, which would lead to better motivation of the choices and decisions made.

In conclusion, this thesis demonstrates that the BMD method is a flexible modelling approach to assess the potential effects of several classes of substances, such as PFAS, PCBs and candidate drugs. Possible applications in the chemical risk assessment and pharmaceutical development areas were demonstrated. Additionally, it was shown that the BMD approach has a strong 3R potential and extracts a considerable amount of information from the data. The BMD approach is in chemical risk assessment to stay, and much like a Swiss army knife, it is a useful and multi-purpose tool that will support you in the derivation of reference values of superior quality.