The problem APODEMUS addresses is deceptively simple to state but notoriously hard to solve. European regulations require that marketed or applied plant protection products do not adversely affect animal health or the environment, and regulatory ecological risk assessments evaluate potential impacts of pesticide exposure on birds and mammals. In the lower tiers of these assessments, toxicity endpoints derived from laboratory studies with standard test species, most commonly the Norway rat, are compared against worst-case assumptions about potential field exposures. Dietary exposure is estimated depending on the type of pesticide application, whether spray or seed treatment, and the feeding guild of the species in question, from herbivore to granivore. If an unacceptable risk cannot be excluded at this stage, higher tiers come into play, and these can involve resource-intensive field studies quantifying the proportion of diet animals obtain from treated areas, the proportion of time they spend there, and the impacts of exposure on field populations. Yet such studies struggle to cover the many relevant combinations of crop and region, and environmental factors such as weather and small-scale landscape composition introduce variability that is difficult to control.

Population models have been suggested as tools to complement and extend field data in these higher-tier assessments. Simulation models can predict population-level effects from modelled life-cycle processes and individual-level exposure and effects, and spatially explicit versions are particularly relevant when exposures vary due to the spatial behaviour of animals in heterogeneous landscapes. They can estimate exposures and effects under different landscape compositions, extend field studies with additional scenarios, and estimate the effectiveness of possible mitigation strategies. Crucially, the protection goals of risk assessments, which include the survival of individuals and the long-term sustainability of populations of potentially vulnerable species in a local agricultural field and its surrounding area, can be addressed directly: exposure-induced mortality can be observed in simulations, and long-term population dynamics can be compared in simulated landscapes with and without pesticide exposure. Individual-level exposures depend on the spatial and temporal overlap of animals with treated fields and exposed off-crop areas, and toxicokinetic-toxicodynamic models can be incorporated to capture the toxicological effect dynamics of temporally variable exposures.

Building such a model, however, demands a systematic and transparent development process. The APODEMUS team followed Pop-GUIDE, the guidance for the development, use and interpretation of population modelling for ecological risk assessment, which is recommended by the European Food Safety Authority for bird and mammal population models. Pop-GUIDE requires documentation and justification of all decisions crucial for model structure, acceptability and domain of application, and it structures development into five phases. In Phase 1, the team defined the model objectives: APODEMUS is intended to capture and simulate individual- and population-level exposures to pesticides and their effects on wood mouse populations in realistic agricultural landscapes across Europe. Dietary exposure is the only exposure route considered for individual mice, with individual exposure varying according to differences in simulated behaviour and the spatio-temporal pattern of exposed diet. Individual-level exposure-effect relationships are represented explicitly, extrapolating from laboratory data with surrogate species to the wood mouse, considering both sublethal and lethal effects depending on individual weight or life stage. The objectives indicate a realistic and precise model, spatially explicit and individual-based, with outputs comparable to field data including individual body weights, population abundance and population structure over time.

Phase 2 was a comprehensive literature review reflecting the current published knowledge about wood mouse ecology. The team searched Google Scholar using predefined keyword combinations guided by Pop-GUIDE characteristics, screened results for relevance, and applied strict eligibility criteria: publications had to address relevant aspects of wood mouse ecology, present original data from European study sites, include a full description of methodology, and be written in English. After screening, 509 publications remained for full-text review, and another 153 were excluded during that review, for instance because they focused on parasites of wood mice or community studies with minor contributions of the species. A total of 341 publications were reviewed in detail, from which 1295 data items were extracted and recorded. For most model characteristics, information from at least 30 publications was identified, suggesting sufficient data to inform the model. Notably, information on wood mouse physiology, particularly ontogenesis and maturation, was scarce, and dedicated searches plus data from related species were needed to inform the dynamic energy budget sub-model.

That dynamic energy budget, or DEB, model is a distinguishing feature of APODEMUS. DEB theory describes the uptake of energy from food and its allocation to maintenance, growth, maturation or reproduction according to the organism’s needs over time, and the ‘stx’ variant relevant for placental mammals includes a dependent foetus and suckling neonate phase. Data gathered for the wood mouse included laboratory growth curves of males and females at different feeding levels, field growth curves, laboratory reproduction data, and general life history descriptors. These allowed the fitting of model parameters for male and female animals and the parameterisation of a mammalian reproduction model previously developed for the Wistar rat. The biological detail is striking: the wood mouse gestation period is around 23 days, followed by roughly 21 days of lactation; males reach sexual maturity at around seven weeks in the laboratory while females mature after eight weeks; and dominant or solitary females produce on average 3 to 4 litters of 5.6 young over 120 days, with an average interbirth interval of 26.5 days, suggesting mating occurs shortly after birth and females become pregnant during lactation. In the DEB model, mature females upregulate feeding, and the additional energy enters a reproduction buffer that goes towards the growth and maturation of dependent offspring, with two concurrent litters, one suckling and one in the womb, simulated by partitioning that buffer.

The spatial architecture of APODEMUS is equally carefully constructed. The heterogeneous agricultural landscape, comprising arable fields, pastures, woods, shrublands and hedgerows, is represented as a grid with each cell assigned to a habitat type. Wood mice use several habitats, with usage depending on habitat type, season and management, and male home range areas can reach up to 2 hectares in agricultural landscapes during the breeding season while female home ranges reach up to 0.5 hectares, meaning a single mouse regularly uses several habitat types daily. Habitat attractiveness is described by a one-dimensional index that drives the location and extent of home ranges, subsuming abiotic and biotic features such as vegetation structure, food availability and species interactions, and varying over time to reflect seasonal changes. Woody habitat serves as the baseline standard because it is constantly used throughout the year. Movement is aggregated into the representation of home range dynamics, since more abundant and reliable information exists on home ranges than on detailed movement patterns. Each simulated mouse establishes a home range and adjusts it in size and shape to fulfil its needs, with highly attractive habitat allowing smaller ranges. During the breeding season, males aim to include nearby females in their home ranges while females aim to exclude them, and if space is insufficient, mice move across the landscape in search of a suitable spot at the cost of losing dependent offspring, suppressed breeding and increased mortality. This exclusive female home range behaviour also means density dependence can emerge in the model, reflecting the negative density-dependent regulation described in wood mouse populations.

Diet deserves special attention because it represents the route of pesticide exposure addressed by the model. Wood mice consume a broad diet including seeds, fruit, leaves, roots, acorns, chestnuts, arthropods, earthworms and fungi, driven mainly by food availability rather than preference. The team assumed that simulated mice can always fulfil their daily energy needs from food available in the landscape, supported by the species’ broad diet and mobility, which means simulated wood mice cannot starve but instead shift their home ranges to cover seasonally productive habitats. Consequently, habitat use is indicative of consumed diet: the time spent during the daily activity period of a mouse in each habitat type corresponds to the diet intake from that habitat, and thus pesticide exposure is driven by an individual’s temporal pattern of habitat use. The effective exposure of an individual is determined by the total daily food consumption rate, the proportion of time spent in habitats with exposed food, and the exposure concentration in the food, with exposure levels varying temporally and spatially across the landscape. To capture individual-level effects of time-variable exposures, individuals will be equipped with a toxicokinetic-toxicodynamic sub-model linked to the DEB module, potentially using a physiologically based kinetic approach that mechanistically captures physiological and biochemical processes across organ and tissue compartments. TKTD models are specific to each compound and require calibration and testing for every pesticide addressed.

Phase 3 and Phase 4 of the development benefited significantly from stakeholder involvement, mainly achieved through a dedicated workshop hosted in Hirschberg, Germany, from 28 to 29 February 2024, bringing together regulators, small mammal experts and ecological modellers. The workshop focused on decisions for which literature information was lacking or ambiguous but important for the model objectives, and the discussions helped refine the representation of processes and identify remaining uncertainties. Beyond refining the conceptual model, the workshop improved communication and transfer of expertise between stakeholder groups, imparting learnings about species ecology, mechanistic modelling and regulatory needs to all participants. The team also identified suggestions for improving Pop-GUIDE itself, including adding diet as a standard characteristic, considering existing models explicitly in early development stages, and adapting the decision tree on generality, realism and precision, which is strongly geared toward threatened and endangered species in the United States, to the European context where common focal species in agricultural landscapes are considered.

The conceptual model now provides the blueprint for formalisation and implementation as computer code, with documentation following the ODD protocol and TRACE standards recommended by EFSA. Model evaluation will compare simulated patterns, such as seasonal population sizes and structures, to independent field observations using pattern-oriented modelling, and data from three field study types used in higher-tier risk assessments, PT studies tracking daily time spent in focal fields, acute field effect studies checking individual survival for 7 to 10 days after application, and long-term capture-mark-recapture studies, can all be used for evaluation, with virtual trapping introduced in the model to simulate capture numbers and recapture rates. Sensitivity and uncertainty analyses will identify how parameter uncertainties translate into model output uncertainty, and the model will be evaluated across different landscapes and climate regions to ensure general applicability across Europe. The team emphasises that APODEMUS was developed to provide reliable relative mean predictions of effects by comparing treatment to control outcomes rather than to explicitly address the natural variability of the ecological system. Once the implemented model, calibrated to realistic conditions and validated with field effect data, is ready, it will simulate wood mouse population responses to a wide range of environmental conditions, long-term and including multiple generations, with population-level properties such as abundance, density and population structure as the key endpoints evaluated, and exposure multiplication factors used to identify margins of safety. The model and its documentation will be made publicly available, and open access will further facilitate model evaluation and use, offering regulators a transparent tool that could transform how pesticide risks to small mammals are weighed across the diverse agricultural landscapes of Europe.

Subject of Research: Development of the APODEMUS conceptual population model for wood mice in pesticide risk assessment

Article Title: Concept for APODEMUS – a wood mouse population model for pesticide risk assessment

Article References: Singer, A., Schmolke, A., Becher, M. A., von Blanckenhagen, F., van den Brink, N., Grimm, T., Ibrahim, L., Imholt, C., Jacob, J., Jakoby, O., Laucht, S., Løvik, A. N., Martin, T., Muñoz, C. C., Preuss, T. G., & Galic, N. (2026). Concept for APODEMUS – a wood mouse population model for pesticide risk assessment. Food and Ecological Systems Modelling Journal, 7, Article e175714. https://doi.org/10.3897/fmj.7.175714

Image Credits: AI Generated

DOI: 10.3897/fmj.7.175714

Keywords: wood mouse, APODEMUS, population model, pesticide risk assessment, ecological modelling, EFSA, Pop-GUIDE, dynamic energy budget, spatially explicit model, ecotoxicology, Concept, wood