Integrating Higher Tier Studies with Mechanistic Models in Bee Risk Assessment

Recent advancements in mechanistic models have the potential to significantly enhance the preci-sion and reliability of environmental risk assessments for bees. In this paper, we explore the inte-gration of data from higher tier studies with models to evaluate the potential improvement in the accuracy of risk assessment outcomes. Over the past decades, various global bee risk assessment paradigms have been developed, providing exposure estimates and using laboratory-derived endpoints such as acute LD50 and chronic NOED values for screening/tier I risk assessment. These paradigms calculate exposure tox-icity ratios (ETR) or risk quotients (RQ) and use different trigger values to identify cases of low risk to bees or those requiring further refinement or mitigation. The revised EFSA bee guidance pub-lished in 2023 takes a step further by proposing a predicted colony-level effect calculation, which can be compared to a numerical protection goal. BEEHAVE is a mechanistic model that simulates real-world conditions and has been validated using field and semi-field data, has a module called BEEHAVEecotox, which includes exposure and effect elements for honey bees. Like the tier I risk assessment scenarios, the model utilises information on product use, exposure, with laboratory-derived toxicity data. We selected three well known plant protection active substances to act as case studies; the herbi-cide glyphosate, and two insecticides thiacloprid and dimethoate, the latter which is used as a posi-tive control for effects on bees in laboratory and higher tier tests. They were selected as expecting to be of low, medium and high risk to bees in a tier I risk assessment, respectively. Each was sub-jected to a tier I risk assessment according to four different paradigms (SANCO 2002, BeeRex, EFSA 2013, and EFSA 2023). Where low risk could not be concluded at tier I the risk assessment was per-formed by simulating tunnel test conditions in BEEHAVEecotox using analogous input parameters for toxicity and exposure. The ability of BEEHAVEecotox to simulate tunnel test conditions was validated by comparing to empirical data from studies conducted with thiacloprid and dimethoate at the same application rates as employed in the tier I risk assessment. For example, an application of the insecticide dimethoate at 400 g/ha which is a typical use rate as a toxic reference treatment in a semi-field tunnel test, predicted a 100% effect on colonies (tier I risk assessment, EFSA 2023) whereas BEEHAVEecotox simulations predicted a 57% mean maximum effect. These values overestimated the tunnel test data mean maximum effect of 46%, demonstrating the conservative nature and protectiveness of the model simulation. Overall, mechanistic models such as BEEHAVEecotox offer the potential to refine risk assessments beyond that of simplistic exposure tox-icity ratios and thresholds, translating individual-level assessments to colony-level effects consider-ing time and colony processes.