SUMMARY

The AURORA project (Actionable eUropean ROadmap for early-life health Risk Assessment of micro- and nano- plastics) is a Horizon 2020 initiative focused on understanding the impact of micro- and nano- plastics (MNPs) on human health during pregnancy and early life.

A primary goal of AURORA Task 5.5 (Work Package 5) is to deliver an actionable framework and roadmap for risk assessment by integrating the scientific findings from the research objectives of other project Work Packages. Additionally, it is necessary to gather results from CUSP (European Research Cluster to Understand the Health Effects of Micro- and Nano- plastics) partner projects, external research initiatives and the scientific literature.

Human exposure to MNPs is unavoidable, yet reliable empirical evidence regarding health effects, particularly during early life, is broadly lacking. For vulnerable early-life stages, a precautionary framework is initially essential. Focussing on exposure-led risk assessment is a pragmatic first-step, applying the logic that risk is most effectively managed by minimising exposure.

The proposals presented here implement a hazard banding and probabilistic exposure modelling framework, with the flexibility to iteratively improve as more robust data emerge. The models are designed so that all inputs are anchored to citable empirical data. The vision is that, as the underlying database develops, comprehensive exposure scenarios can be developed and risks quantitatively assessed with full transparency.

To support the longer-term transition from qualitative hazard banding to data-driven risk characterisation, a companion Hazard Characterisation Database has been designed and built — capturing particle identity, physicochemical properties, chemical load, and exposure matrix data in a standardised, organised format.

These elements form the basis of a Risk Assessment framework that supports the vital work of the AURORA project. It is actionable, extensible, auditable, and future-facing. Crucially, the framework indentifies a set of standard data requirements for MNP researchers and regulators, while already offering clear pointers to exposure mitigation - and therefore risk-reducing - strategies to protect vulnerable human early-life stages.

Interactive AURORA 5.5 Roadmap

Follow the implementation of each phase, scenario, and planned development in detail.

OPEN →

BACKGROUND

Current human health risk assessment (HHRA) for micro- and nano-plastics (MNPs) is supported by six primary frameworks, each at varying stages of implementation (Koelmans et al., 2026). The Southern California Coastal Water Research Project (SCCWRP) framework is the most operationally advanced and is in active use for establishing California's drinking water screening levels. MICROPLASTIC LAB utilises a fully probabilistic, Toxicologically Relevant Metric (TRM)-based approach that is well-aligned with international standards but lacks official regulatory adoption for human health. POLYRISK employs a modular, mechanism-based structure focused on inhalation, while MOMENTUM serves as a case-specific prototype for indoor air risks. PLASTICHEAL offers the PlasticRiskCat tool for qualitative hazard prioritisation, but remains largely at the developmental stage.

At present, while the classical Risk = Hazard x Exposure paradigm is considered appropriate and remains the foundation for MNP health risk assessments, it is more applicable to traditional chemical risk assessment which relies on mass-based concentrations. MNPs present significant technical difficulties due to their extreme diversity in size, shape, polymer type, and aging state. Adaptation of assessment approaches is needed to account for the unique complexities of particles.

Key limitations to be addressed include:

• Data Gaps: A full quantitative assessment is currently hindered by a lack of standardised exposure data and dose-response relationships.
• Complex Hazards: MNPs present a multiplicity of hazards related to physical particle toxicity, chemical hazards from leachates/additives, adsorbed environmental contaminants ("Trojan Horse") effect including mixture and synergistic toxicity, and microbiological hazards from pathogens colonising their surfaces.
• Analytical Limits: Identification of the smallest nanoparticles (≤100 nm) in biological samples remains technically challenging. Recently, questions have been raised over the accuracy and potential for false positives inherent in some analytical techniques (Thomas et al., 2026).

To make this paradigm operational for MNPs, recent reviews (Brachner et al., 2020; Noventa et al., 2021; Vogel et al., 2024; Christopher et al., 2024; Koelmans et al., 2026) have identified the need for more comprehensive strategies, including standardisation of metrics, elucidation of particle size distributions, linking external intakes to internal dose and kinetics, integrated hazard assessment approaches, and the adoption of tiered, modular architectures.

We have developed a framework that comprehensively addresses these requirements. It compliments the outputs of CUSP partner projects, while tailoring risk assessments to the AURORA-specific circumstances of the foetus, neonates, infants and toddlers.

AURORA has so far provided a conceptual roadmap specifically for early-life vulnerability (Christopher et al., 2024). This roadmap has now been elaborated into a suite of operational tools: four probabilistic exposure models covering the key early-life stages (foetal, neonatal, infant, toddler), each with detailed methodological justifications and empirically anchored input parameters; a structured Hazard Characterisation Database capturing particle identity, physicochemical properties, chemical load, and exposure matrix data in a standardised format; and an integration pathway linking the exposure and hazard domains for particle-specific risk characterisation.

EXPOSURE ASSESSMENT OF EARLY LIFE-STAGES

As a first step, we have developed 'proof-of-concept' probabilistic models designed to estimate potential human exposure to micro- and nano- plastics (MNPs), with a specific focus on sensitive early-life stages. The aim is to demonstrate the applicability of recommended best-practice approaches to exposure modelling. Note that the models currently do not represent the totality of daily exposures, but rather a subset of pathways unique to each life stage. However, due to their flexibility and modularity, they can be adapted as new requirements and data are identified.

The models use probabilistic (Monte Carlo) modelling, an established method in risk assessment (Wardani et al., 2024). Instead of producing single "worst-case" estimates, the models run 10,000 simulations for each scenario, drawing from statistical distributions for each input parameter (such as water intake or dust concentration). This can generate a more realistic range of potential exposure outcomes based on empirical data while allowing user modification, thus providing a better understanding of the likely MNP exposures under defined conditions.

Results are presented as a histogram showing the distribution of calculated doses, with key metrics including mean and 95th percentile highlighted. This exposure level is then combined with a user-defined hazard score (based on polymer type and chemical additives) in a simple 3x3 risk matrix to provide a final "Low," "Medium," or "High" risk characterisation.

As proof-of-concept, four distinct life-stage models are presented, with each targeting unique exposure pathways:

The Maternal/Foetal model estimates the potential for maternal MNP exposure to transfer to the foetus. Two modelling options are provided: a traditional conservative "forward dosimetry" approach, and a "reverse dosimetry with plausibility filter" that rejects biologically impossible scenarios based on empirical data from human placentas.

The Neonate / Early Infancy (0–6 months) model covers both the newborn period (0–28 days) and the early breast/formula-feeding phase of infancy, focusing on the high-exposure pathway of MNP release from polypropylene baby bottles during formula preparation, with cumulative ageing effects from repeated use.

The Infant (6-12 months) model simulates exposure for a crawling child, combining inhalation of resuspended dust with ingestion from hand-to-mouth contact.

The Toddler (1-3 years) model assesses parallel exposures from direct ingestion of dust, along with chemical leaching from the mouthing of plastic toys. Currently there are no validated quantitative data on MNP shedding from mouthed plastic toys, however the model serves as a methodological placeholder for MNP leaching which can be adapted when validated data become available.

Detailed methodological justifications for each scenario's calculation approach, parameter choices, and the supporting evidence underpinning each parameter are provided within each model's section below. These justifications cover the rationale for the chosen equations, distribution types, and citation bases that drive the probabilistic exposure estimates.

While reviewing the models, it is important to understand their current limitations and areas marked for future improvement:

• Particle Size Distributions & Dose Metric: The models assume a single dominant particle size category per scenario rather than a full particle size distribution, and use particle count as the dose metric. Particles are treated as homogeneous units — a 100 nm and a 10 µm particle each count as "1", which ignores orders-of-magnitude differences in mass, surface area, and potential toxicity. Polymer density is also not accounted for (e.g., a Polyvinyl Chloride (PVC) particle has greater mass than an identically-sized PE particle). Results should therefore be interpreted as a proxy for exposure rather than a definitive toxicological dose. Future versions will aim to incorporate a multi-metric reporting strategy (particle number, surface area, chemical load, and mass — in that order of priority) as described in the Hazard Characterisation section. A particle size distribution framework (lognormal, power-law, binned) has been designed within the hazard characterisation database to enable this transition, but has not yet been integrated into the exposure models.
• Normalisation Metric Uncertainty: The models present results normalised to body weight (particles/kg-bw/day) as standard practice in chemical risk assessment. However, the applicability of BW-normalisation for particle toxicity is debated — harm may relate more to surface area interaction with organ linings or to the absolute flux of particles crossing biological barriers. To address this, the models include a "Normalisation Metric" selector offering both options for scientific flexibility.
• Reverse Dosimetry: This approach, used in the Maternal/Foetal model, anchors outputs to real-world placental burden data (current default ~4,000 particles at birth). Rejecting any simulation that exceeds this limit—this is a theoretically sound mass-balance technique but has not been formally validated against independent datasets.
• Kinetics: Gut translocation and placental transfer fractions are modelled using broad distributions because direct human data for MNP uptake is sparse.
• Dermal (skin) exposure pathways are not yet modelled. Young children have extensive skin contact with MNPs via contaminated surfaces, bathwater and handled objects — but no dermal absorption or contact-transfer pathway is currently included in any scenario.
• Breast milk as an exposure route for neonates is not yet modelled. The Neonate scenario currently only captures formula feeding via polypropylene bottles. Maternal offloading of MNPs through breast milk represents a significant gap in the 0–6 month exposure assessment.

Calculation justifications are documented within each model’s reference pages. The modular design makes it straightforward to update model parameters as new empirical data becomes available. Input parameters are to be based on the best available evidence, with iterative improvements.

Discussion is needed on expansion to capture all relevant exposure parameters. Plausible exposure routes are presented below.

Modelling Particle Kinetics

All four exposure models share a common set of kinetic parameters that account for how particle size and shape influence movement across biological barriers — from the point of intake (gut or lungs) through to systemic circulation and, where relevant, transfer to the foetus. These parameters are defined for five standard size categories spanning the nanoscale to coarse microplastic range, with morphology-dependent modifiers applied for fibres and fragments. Life-stage-specific physiological multipliers further adjust these baselines to reflect the developing anatomy and physiology of early-life stages. The following sections define these parameter sets and explain how they translate external exposure estimates into internal doses.

From External Exposure to Internal Dose — The Kinetic Pipeline

The kinetic parameters defined in this section are not abstract — they describe the sequential transformations that convert the external dose (the concentration of MNPs in air, water, or food, multiplied by intake rate) into the internal dose (particles reaching systemic circulation or target tissues). Understanding this pipeline is essential for interpreting the scenario model equations. The general sequence is:

1. External (Intake) Dose: The model first calculates the total particles encountered per day — for example, particles in ingested water, in inhaled air, or in dust transferred via hand-to-mouth contact. This is the gross environmental load and represents the upper bound of possible exposure.
2. Respiratory Tract Deposition: Not all inhaled particles are retained. The ICRP-based get_lung_deposition_fractions() function (defined per life stage and particle size) splits the inhaled load into three compartments:
   • Alveolar deposition — deep lung fraction, available for translocation to systemic circulation.
   • Conducting airways (bronchiolar + bronchial) deposition — cleared via the mucociliary escalator to the pharynx and swallowed, thereby contributing to the ingested dose.
   • Oropharyngeal / extrathoracic deposition — directly swallowed without reaching the lungs, also contributing to the ingested dose.

   Localised effects at deposition site not modelled kinetically: Particles deposited in the respiratory tract or retained in the gut lumen — particularly fibres and those in the nanoscale range — may cause localised biological effects (oxidative stress, inflammatory signalling, cellular damage) without ever translocating into systemic circulation. These local effects are not modelled by the kinetic pipeline (which tracks translocation only), but they are accounted for in the hazard scoring framework through the physical particle hazard domain (particle size, morphology, surface chemistry) and polymer-specific toxicity scores, where the potential for localised tissue responses is factored into the overall hazard characterisation.

3. Gut Barrier Translocation: Of the total ingested load — comprising direct ingestion plus particles swallowed via mucociliary clearance (MCC) from the respiratory tract — only a fraction crosses the gut epithelium into the portal or systemic circulation. This fraction is the product of three factors:
   • Size-dependent baseline translocation fraction (from the Reference Translocation Parameters table below).
   • Morphology modifier (sphere = 1.0×, fibre = 0.5×, fragment = 0.9×).
   • Life-stage gut permeability multiplier (neonate = 2.0×, infant = 1.5×, toddler = 1.2×, adult/pregnancy = 1.0×).
4. Systemic Circulation and Distribution: Particles that cross the gut barrier or translocate from the lungs enter the bloodstream. From here, they distribute to organs, are trapped in filtering tissues (liver, spleen, placenta), or are cleared via the reticuloendothelial system. For the early-life models, the key distribution steps are:
   • Placental trapping — a fraction of the maternal systemic dose is retained in the placental tissue (modelled as ~5% default, see table below).
   • Placental transfer to the foetus — a further fraction passes through to the foetal compartment (modelled as ~8.7% for 50 nm PS, van Boxel et al., 2025).
   • Pulmonary translocation — particles deposited in the alveoli may translocate directly into the bloodstream, bypassing the gut entirely. This is a separate entry route to systemic circulation.
   • Dermal penetration — not yet modelled. Skin contact with MNP-contaminated dust, textiles, and bathwater represents a plausible exposure pathway (particularly through compromised skin such as nappy rash or eczema), but no dermal absorption or contact-transfer pathway is currently included in any scenario. This is flagged as a development gap in the Introduction's list of limitations.
5. Internal (Target Tissue) Dose: The end result of the exposure model. At present, this internal dose estimate is combined with a semi-quantitative hazard banding score (based on polymer type, particle characteristics, and chemical load) in a simple 3×3 risk matrix to produce a 'Low / Medium / High' risk characterisation — rather than being formally compared to a toxicological reference value. This is the appropriate approach given the current lack of established health-based limits for MNPs. The exposure models therefore provide the dose side of the risk equation, while the hazard characterisation database provides the tools needed to eventually transition to a more quantitative, metric-aware risk assessment as data become available.

Size Categories and Morphology Modifiers

Five discrete size bins span the nanoscale to coarse microplastic range. These bins define the baseline translocation and deposition fractions — larger particles have lower translocation potential regardless of shape. The morphology modifiers (sphere, fibre, fragment) are then applied multiplicatively on top of these size baselines to account for the altered kinetic behaviour of non-spherical particles:

• Spheres/beads: 1.0× (reference morphology — most empirical data uses spheres)
• Fibres: 0.3× for pulmonary translocation (Sturm 2012 — fibre geometry impedes alveolar translocation); 0.5× for gut translocation (limited empirical data, fragment ratio used as conservative default)
• Fragments: 0.8× for pulmonary translocation (Ni et al., 2026 — irregular shape reduces aerodynamic diameter alignment); 0.9× for gut translocation (near-spherical behaviour)

Reference Translocation Parameters

Citations are annotated as Direct (empirically measured in the relevant species and route) or Derived (extrapolated from surrogate data, animal models, or mechanistic first principles):

Life-Stage Physiological Multipliers

Life-stage modifiers account for the developing physiology and assumed vulnerability of early-life stages, with healthy adult baseline as 1.0×:

These multipliers are semi-quantitative estimates based on known physiological differences between life stages. Neonates and infants have a high surface-area-to-body-weight ratio (SA), which enhances the gradient for pulmonary particle translocation and is reflected in the elevated multipliers for that column. Alveolar macrophages (AM) are the lung's resident immune cells responsible for clearing deposited particles; neonates have fewer alveolar macrophages than adults, resulting in reduced clearance capacity. The gut permeability multipliers reflect the progressive maturation of intestinal tight junctions from the immature neonatal barrier through to near-adult integrity in toddlers. These values are intended as indicative modifiers for risk assessment and should be refined as empirical data become available. Supporting evidence and baseline values for each parameter category are detailed in the Reference Translocation Parameters table directly above.

HAZARD CHARACTERISATION DURING EARLY LIFE STAGES

The models currently implement the proposal of Christopher et al. (2024), applying a semi-quantitative banding approach that uses a points-based system to prioritise particles based on polymer, particle-specific, and chemical dimensions. This is a scientifically appropriate methodology given the current lack of detailed toxicological data needed to establish formal reference values (e.g., Tolerable Daily Intakes) for most micro- and nano- plastics.

By necessity, current MNP risk assessments must be exposure-led. Considering the vulnerability of early life stages and in the absence of firm quantitative hazard data, priority is given to identification of key points of exposure. A qualitative / semi-quantitative assessment approach can already point to appropriate and pragmatic risk mitigation and management measures, thereby reducing risk. Over time, more precise quantification of hazard and exposure parameters should be possible.

To enable such improvements, a structured Hazard Characterisation Database has been designed and built, capturing particle identity, physicochemical properties, chemical load, and exposure matrix data in a standardised format. Key domains include:

• Particle Characterisation: Polymer type, dimensions (nm precision), morphology, crystallinity, specific surface area, and density — enabling both particle-count and mass-based metrics.
• Surface & Interfacial Chemistry: Zeta potential (surface charge), hydrophobicity, and refractive index — forming the basis for predicting how particles agglomerate in biological fluids (Two-State Classification approach).
• Chemical Load (Trojan Horse): Both intentional additives (phthalates, bisphenols, flame retardants) and adsorbed environmental contaminants (POPs, heavy metals, PFAS) linked by CAS number.
• Exposure Matrices: Standard reference mixtures (e.g. European housedust, bottled water) that combine multiple particle types with fractional abundances and continuous particle size distributions — enabling realistic exposure inputs for the Monte Carlo models.

The database is currently populated with demonstration data. Its flexible structure allows new particle types, mixtures, and composition data to be added as empirical evidence accumulates, providing the infrastructure needed to support the transition from qualitative hazard banding toward data-driven, particle-specific risk characterisation. The database therefore sets a standard for characterisation data requirements. See the Hazard Characterisation section for full details.

What remains aspirational (future development): Several important enhancements are planned but not yet implemented. These depend on the availability of further validated empirical data and partner input:

• Continuous particle size distributions for hazard scoring: As measurement and analytical techniques improve, continuous statistical distributions could replace the current discrete scoring classes to better capture the continuum from micro- to nano- sized particles (Koelmans et al., 2026).
• Mechanism-based grouping: Application of read-across, Adverse Outcome Pathways (AOPs), and Integrated Approaches to Testing and Assessment (IATA) as proposed by the POLYRISK project (Vogel et al., 2024; Koelmans et al., 2026). This is under review but not yet operationalised within the current framework.
• Reference materials: Utilisation of environmentally realistic reference materials (Brachner et al., 2020) to enhance the scientific rigour of hazard scoring.
• Microbiological hazards: Broadening scoring criteria to include microbiological hazards and pathogens as a distinct risk category (Noventa et al., 2021).
• PBPK modelling: The models currently incorporate aspects of particle kinetics (gut translocation, inhalation deposition, placental transfer) at varying levels of detail, but a full Physiologically Based Pharmacokinetic (PBPK) model — incorporating size-stratified kinetics, protein corona effects, and quantification of biological barriers such as the blood-brain barrier — remains a medium-term objective.
• Formal dose metrics: The transition from single-metric (particle-count) reporting to the proposed multi-metric approach (particle number, surface area, chemical load, and mass) is structurally supported by the database but has not yet been implemented in the exposure models.

Dose metric considerations: A recurring cross-cutting question is which dose metric best represents biologically relevant exposure. The current models default to particle number (particles/day), the most widely reported metric in both environmental sampling and experimental toxicology. However, mass, surface area, and chemical load each capture different dimensions of potential harm, and none is sufficient on its own. The Hazard Characterisation database stores all the data needed for a multi-metric approach (particle dimensions, morphology, density, surface area, and chemical loading), and a multi-metric reporting strategy has been proposed, but the implementation of surface area and chemical load as derived metrics awaits further development (see the Hazard Characterisation section for the full discussion).

Kinetic modelling: All kinetic parameters — gut translocation, lung deposition fractions (ICRP model), mucociliary clearance, placental trapping, and foetal transfer — are defined in the Modelling Particle Kinetics section above, with size-dependent baseline values, morphology modifiers, and life-stage physiological multipliers. That section provides the complete set of reference translocation parameters with their empirical evidence levels. Future developments will aim to include direct lung-to-bloodstream transport (pulmonary translocation), dermal absorption, breast milk offloading, and a size-stratified placental transfer function — the database structure already supports expanding to include these processes.