Final Report Summary - MODNANOTOX (Modelling nanoparticle toxicity: principles, methods, novel approaches)
The physicochemical properties of nano-sized materials are distinct from the properties of equivalent bulk substances and are also often unpredictable. As the use of nanomaterials increases, so must the research into any potential adverse effects on the environment or health. ModNanoTox addressed these research needs by developing a number of well-documented and technically advanced models describing the behaviour of engineered nanoparticles in organisms and the environment. ModNanoTox was a small FP7 funded project, which started in November 2011 and ended in October 2013 (2 years), and involved EU and US teams. In Europe: University of Birmingham, EMPA, EAWAG, Roskilde University and the SME In Silico Toxicology GMBH. In the US: University of Nebraska (full partner), University of South Carolina (informally) and Rice University (MoU).
ModNanoTox focused on the development of computational models, to complement and support research on and regulation of the environmental and human implications of exposure to engineered nanoparticles. In recent years, a number of laboratory-based nanotoxicology projects have been funded, notably under the FP7 programme, which are recently completed or currently in progress and which have generated major nanosafety-relevant datasets. A significant research effort is also under way in the US. ModNanoTox aimed to evaluate and data-mine the best available datasets from these sources, and fit them into new models. ModNanoTox also developed a number of well-documented, integrated and technically advanced models describing the behaviour of engineered nanoparticles in an environmental or biological context to comprehensively address the following key hypotheses:
1. Toxicity of nanoparticles is the result of physicochemical properties and this has been documented reliably in completed/ongoing studies. Properties found to be relevant include composition, size, surface area, shape, structure & surface modifications (WP1,2).
2. Nanoparticle reactivity can be modelled computationally and can be linked to toxicity (WP1).
3. Toxic responses from cell culture studies and whole organisms can be correlated and rationalised and can be translated into tools useful for model development (WP2).
4. Bioaccumulation into cells or whole organisms can be characterized and modelled using biodynamic principles (i.e. by characterizing uptake rate constants from food and water as well as loss rate constants) (WP3).
5. Toxic responses from cell culture studies and whole organisms can be modelled reliably by QSAR type approaches (WP4).
6. Environmental exposure concentrations can be assessed reliably and incorporated in appropriate models (WP5).
7. Mechanistic effect models can be developed by extrapolation from ecological and (eco)toxicological observations and can be built into risk assessment models (WP6).
The project resulted in 4 publications to date (several others to be submiteed), and has fed data into several ongoing EU FP7 projects including NanoTransKinetics, NanoMILE, MARINA and NanoVALID and newly started projects eNanoMapper, SUN, and FutureNanoNeeds.
Project Context and Objectives:
ModNanoTox was a € 1M EU-US modelling project entitled “Modelling nanoparticle toxicity: principles, methods and novel approaches” with 6 full partners (5 EU and 1 US) and a duration of 24 months. It was one of the first of the modelling cluster of projects, funded in parallel with EU FP7 NanoTransKinetics (coordinated by UCD) and a US EPA STAR-N1 project, ENPI (Coordinated by Rice University). The project partners were:
1. University of Birmingham (UoB) UK, Coordinator
2. Roskilde Universitet (RU) Denmark
3. Eidgenoessische Anstalk fur Wasserversorgung Abwasserreinigung und Gewässerschutz (EAWAG) Switzerland
4. Eidgenoessische Materialpruefungs-und Forschungsanstalt (EMPA), Switzerland
5. In Silico Toxicology Gmbh (IST) Switzerland
6. University of Nebraska, Lincoln (UNL) US
The goals of ModNanoTox were to develop a number of well-documented and technically advanced models describing the behaviour of engineered nanoparticles in organisms and the environment. The background to these models was a thoroughly documented database, constructed based on:
(1) An advanced evaluation of physicochemical properties of nanoparticles and in silico modelling of their reactivity; and
(2) Assessment of the characterisation methodologies as well as toxicity protocols used to develop biological responses in toxicological studies.
The whole dataset was evaluated for internal consistency and then compared with other relevant sets. The evaluation stage was followed by development of toxicity models based at the individual organism level, using statistical and mechanistic models, in parallel with models predicting environmental fate. The toxicity and fate models were integrated in mechanistic models to predict the long term risks of engineered nanoparticles for populations under realistic environmental conditions. The risk assessment framework was developed using results from the project’s new models.
The overarching objective of ModNanoTox was to assimilate data from major EU and US funded projects on nanotoxicity (in many of which the ModNanoTox consortium has had direct links) and from this to generate models of the relationships between engineered nanoparticle (ENP) properties and toxicity. The project comprised 6 research workpackages, each of which addressed different aspects of model development with an overall environmental focus. The work packages were:
WP1 - Physicochemical properties assessment
WP2 - Data evaluation
WP3 - Bioaccumulation modelling
WP4 - QSAR models
WP5 - Exposure & Fate modelling
WP6 - Population models and risk assessment
The specific objectives of the ModNanoTox project are summarised as follows:
The overall objective of WP1 was to generate a computational fundamental understanding of nanoparticle reactivity, in order to inform other modelling activities within the project, particularly QSARs (WP3). Having reviewed and evaluated physicochemical descriptors (size, shape, phase, concentration, composition, surface modification, method of synthesis) and studied the effect of size and structure on nanoparticle reactivity the work was completed with a set of simulations of nanoparticle solubility as a function of size and structure.
The overall objective of WP2 was to generate a database that would support model development within and beyond the duration of the project and contribute to the nanotoxicology community needs for a reliable compilation of all relevant and up to date data. Having initiated the evaluation, classification and recording scheme for ENP toxicity data (data mining), and generated an initial database to support work in workpackages 3, 4 and 6, and having evaluated existing environmental exposure data for ENPs, in order to prioritise data for modelling in workpackage 5, the final objective of WP2 was to complete the database with studies up to the point of project end. A further important objective was to produce an analysis of data quality, address data gaps and identify challenges and next steps.
The overall objective of WP3 was to develop toxicokinetic/toxicodynamic models suitable for use in risk assessment. Having evaluated the data quality and availability from WP2 for use in toxicokinetic modeling and toxicodynamic analysis, and having made a selection of variables suitable for describing bioaccumulation as a toxicokinetic and mortality as a toxicodynamic model system, the final objective of WP3 was to assess the sensitivity of the models to different parameters identified across different organisms/cells/organs and exposure conditions to obtain generally applicable information, which could further guide environmental risk assessment for nanoparticles.
The overall objective of WP4 was to develop QSAR modeling of nanomaterial toxicity as far as current data availability allows. Having redeveloped the Lazar QSAR framework for nanotoxicological data, but realized current limitations in data availability for the model, the final objective of WP4 was to work on improving the accessibility and sustainability of the WP2 database by convert it into a format compatible with emerging standards (ISA-TAB NANO).
The main objective of WP5 was to advance and improve environmental exposure assessment models. Having reviewed and evaluated available models, work then progressed into the development of a model which combined nanomaterial flows into the environment with an environmental fate model, both connected on a local scale using geographic information on river flow and wastewater treatment effluents.
The main objective of WP6 was to model effects on ecosystems and then combine these with models from other workpackages to generate a risk assessment framework. Having produced a set of realistic worst-case scenarios for exposure of metal nanoparticles in freshwater, marine, pelagic and sedimentary environments and an assessment of species groups most likely to be at risk from engineered nanoparticles, WP6 also produced a population level risk model and a conceptual framework for risk assessment.
ModNanoTox resulted in 3 published articles (with several more in advanced drafts process and 20 technical deliverables, two of which are publically available via the projects website (www.birmingham.ac.uk/generic/modnanotox/), with the others available on request to the coordinator (email: firstname.lastname@example.org).
WP1 Physico-chemical properties assessment
Atomistic modelling has shown that NP properties such as size, surface roughness, fractal dimension etc. can all impact on transport (& uptake and likely toxicity) mechanisms and that interaction of the surfaces with water can affect the NP surface reactivity. Analysis of the 1st hydration layer of the water molecules around Ag NPs and the radial distribution functions facilitated prediction of water molecule residence times at different surfaces and how these were affect by NP size, shape and structure. The size (number of Ag atoms) of the AgNP affects the shape of the particle, meaning that the size of the particle can affect the proportion of different surface configurations, which is important considering that some surfaces are more reactive than others purely as a result of their geometric configuration. The modelling was also able to estimate how attractive differently sized AgNPs are to water molecules. The residence time for water molecules in the 1st hydration layer around the AgNPs at different surface positions showed that certain sizes / morphologies are more attractive to water than others, and so they are potentially more liable to undergo dissolution – this is an area for future research.
Additionally, the process of Ag NP agglomeration was studied: Ag NPs in a simulated water box were monitored in terms of the changes in aggregate structure over time, focussing on the effect on different physicochemical parameters, for example, molecular roughness. The analysis software utilised in ModNanoTox was not designed to cope with such large systems and thus significant re-writing of the software was required; modelling approaches were therefore being pushed and expanded in terms of the software capabilities. However, now the code is re-written, it can be applied to other questions quickly and easily, and as such, the code can be shared upon request. The work on NP modelling aimed to address the lack of significant overlap currently between theory and experiment, and make the overlap more comprehensive. Three of the recommendations from the Intelligent Testing Strategy report support the importance of modelling of NP physico-chemical parameters:
- “F.6.1 The use of extrapolation and read-across between: materials of similar/related characteristics; species; human and environmental; as well as in vivo and in vitro should be more exploited.
- F.6.2 The impact of exposure route/pathways on PC characteristics (e.g. surface properties, protein corona etc.) and, therefore, the impact on toxicity, needs investigation.
- F.6.3 There is a need for better understanding of how NM interact with their environment, as factors such as dispersion, aggregations and agglomeration can influence NM toxicity”.
Two publications are in final stages of drafting on this work. Data regarding atomistic modelling of the effect of NP size and shape on ligand stabilisation, using for example citrate, is being developed further within NanoTransKinetics and will feed also into the overall corona work there.
WP2 Data evaluation
The final form of the ModNanoTox database (DB) contains 99 studies dealing with the toxicity of nanoparticles (NPs) in aquatic organisms. Of the 99 studies included in the database 1% were published in 2007, 4% in 2008, 15% in 2009, 15% in 2010, 20% in 2011, 31% in 2012 and 14% in the first 10 months of 2013 (note that this includes data only to end October 2013 as the project ended then). This reflects the fact that ecotoxicity studies have traditionally lagged behind toxicity studies, and this was also observed for nanoecotoxicity studies compared to nanotoxicity studies: although the first nanotoxicolology papers were published in 1990s, the first nanoecotoxicology papers came out in 2005/2006. It is clear that the pace of research and publication in this arena (ecotoxicity focusing specifically on aquatic species and silver nanomaterials) is increasing rapidly, so a key goal now is to support researchers to ensure that the data they are generating is of sufficient quality/type for use in modelling studies to progress understanding rapidly and effectively.
Information extracted from each paper was grouped under 4 categories selected to describe the papers: Study Details, Particle Details, Assay Details and Study outcomes. Note that these were determined by the ModNanoTox consortium for the specific purpose of data quality assurance for subsequent modelling studies relating to NP interactions with the environment. One of the project deliverables (D2.4) contains the full list of publications included in the final database, as well as the list of publications assessed but not included in the database, along with the justification of their exclusion on the basis of insufficient characterisation.
During the last few months of the project, efforts were made to align the ModNanoTox database with the format of ISATab. The ISAconfigurator (a tool provided by the ISATab support community) creates experiment-specific configuration files in XML format that can be used to build spreadsheets in the ISAcreator. The ModNanoTox spreadsheet is separated into four tables, as explained above: Study Details, Particle Details, Assay Details and Study Outcomes. For each table we created a configuration file which maps the information from the ModNanoTox spreadsheet to the related ISA-TAB category/form and its fields and requests a data type or an ontology term (e.g. a species or a defined unit). Up to date ISAconfigurator files for the ModNanoTox database can be found at the public online repository and an example is included in ModNanoTox deliverable D3.3.
This ISAcreator tool creates and edits ISA-TAB files. Configuration files designed for particular assays help to map data from existing spreadsheets into ISA-TAB and can be reused with little modification for similar assays. ISAcreator provides a mapping tool to import spreadsheet files to deal with legacy data.
Challenges identified in terms of the conversion include the fact that ModNanoTox database has no common units as it is extracted from literature data and this is dependent on whatever units the studies reported. Significant additional manual effort would be required to convert all studies to common units, and the risk of introducing errors would be high. Additionally, ISATab uses a hierarchical ontology that our database cannot capture - we would have to move towards a database format with relational representation which would also require additional effort. Challenges exist on the ISATab side also as for the conversion, since the nano aspects are not yet included in the tools, and the fact that ISATab supports only single tables as input and the tables should be complete and the entries per column should have the a specific format (e.g. boolean, sting, integer, …).
Quality Assurance criteria / classification of degree of NP characterisation
The central purpose of the ModNanoTox database was to facilitate the modelling work packages to access good quality data with which to develop and validate their models. To achieve this ModNanoTox partners developed a set of criteria with which to rank the extent and appropriateness of nanoparticle characterisation in each study, leading to a quality score for each study. Thus, each study that passed the initial QC (based on their source (industry or lab-synthesized) and the extent of in-house characterisation (none or some) with possible scores between 0 and 2; Combined scores of 0 meant exclusion from the database) was then scored for the completeness of particle characterization – note that this is specific for each particle chemistry, as not all descriptors are relevant for all particles. Studies were evaluated against a matrix of physico-chemical parameters as determined by specific methodologies and characterisation under relevant exposure conditions and over relevant timescales, giving a characterisation score for each, which is included as a field in the database. This score gives the database user an indication of the confidence they should have in the study from a characterisation perspective. The criteria per particle are a ‘most informed guess’ of what parameters will influence endpoints of interest: for silver NPs dissolution is obviously a key parameter, while it is not relevant for titania. Crystal structure is likely a relevant parameter for titania NPs, as it may be present as one of three structural configurations (polymorphs), but is not relevant for silver NPs, with one only possible structure.
The QC of the biological assays and end-points was based on basic scientific principles – the studies were too variable to standardize. Typical parameters recorded regarding the biological assays included species used, gender / life-stage, maintenance and preparation, media conditions (pH, ionic strength, temperature), exposure route, exposure duration, endpoints measured, endpoint method and controls included.
Future additions to the current characterisation matrix could / should include details of the coating characterisation (binding affinity, degree of coverage, displacement by other biomolecules in situ etc.) as well information regarding the ageing and/or transformation(s) in situ in a time resolved manner over timescales matching the exposure duration.
As the scientific community matures, and understanding of the importance of nanomaterials characterisation reaches the wider community, and is debated in the open literature and with journal editors, the degree of in situ characterisation reported in publications (i.e. characterisation in the relevant exposure medium) appears to be improving. Using the data in the ModNanoTox database, some analysis was performed to assess whether this is indeed the case: In order to test whether a significant correlation exists between advancing publication years and the characterization parameters, Two-Factor ANOVA without replication was performed. Results showed that when all parameters were taken into account there is significant difference in scoring change with time but that even more significance was calculated when comparison was made in terms of the different individual characterisation parameters (e.g. size, size distribution, shape/morphology, zeta potential, etc.).
Agglomeration is increasingly used to characterize NPs year on year, though initially (2008 studies) it was not assessed at all. The fact that agglomeration is correlated with the media in which the NPs are dispersed and that year-on-year the reporting of aggregation/agglomeration increased led to strong positive scoring correlation with time (rAP=0.804 and rAS=0.941) and significant variance with time (pA=8.56*10-11) which suggests that more and more studies follow a protocol of characterization both in situ and during the exposure time of the biotargets to NPs. Since non-linear correlation was found to be significant at the 0.01 while linear at the 0.05 level, exponential fitting was performed on the calculated data that provided highly significant values of R2=0.992 and reduced χ2=0.005.
ModNanoTox has demonstrated that in situ characterisation of NMs in the exposure media and over the time-course of experiments is improving with time, as greater awareness builds in the community of the need for detailed characterisation, it is clear that there is still some way to go to generate data of sufficient robustness for modelling. Note that these studies related to NMs exposed in aqueous media, which for the OECD model systems typically contains salts only, making them much simpler from a characterisation perspective than river or sea waters which contained dissolved organic matter etc. NM characterisation in soil and sludge present further challenges, and are likely showing less improvement over time than aquatic ecotoxicity studies.
WP3 Bioaccumulation modelling
The primary aim of this WP was to utilise the ModNanoTox database of studies assessing impacts of Ag NPs on aquatic organisms to develop a statistical assessment to determine the influence of NP characteristics such as size, coating and concentration on NP bioaccumulation (as a toxicokinetic variable) and toxicity (as a toxicodynamic variable), ultimately in order to develop toxicokinetic/toxicodynamic models suitable for use in risk assessment. For the species considered, the team were also interested to understand whether feeding during exposure plays a role in modulating the bioaccumulation or toxicity.
For bioaccumulation, feeding and NP characteristics including surface coating appear to be most important, and for toxicity it is not yet clear if feeding is a significant factor, but hydrodynamic and nominal NP size and coating are drivers of toxicity. Non-significant factors are also assessed as it is also important to demonstrate those factors that don’t influence bioaccumulation / toxicity in order to prevent unnecessary replication of studies in the future.
The most significant challenge for the bioaccummulation modelling resulted from the sparcity of the available dataset: following identification of data sets suitable for toxicokinetic (TK) and/or toxicodynamic (TD) modelling only 11 studies initially (M12 of the ModNanoTox project) were identified covering both waterborne and dietborne exposure to Ag NPs. Narrowing this down to only waterborne exposure of the organisms resulted in data for 14 different species from 20 studies in the database which resulted in just 143 data points in total, spread across these 14 species. Unsurprisingly, given its status as an OECD standard test organism, Daphnia was the species used in the most studies (4) and resulted in 57 data points in total, while the second most studied species was fish gills where internal concentration has been studied, but with only 11 datapoints. Within this also there were variations in the coating on the Ag NPs resulting in sub-groups of particles based on coatings – steric (PEG, PVP), electrostatic (carbonated, citrates, HA), and uncoated. This is a very clear limitation in the development of models and needs to be addressed as a matter of priority. Clearly there was virtually no redundancy in the dataset and no studies were replicated. Here also, it is clear that temporal data, and several datapoints for each measured endpoint would improve the amount of data available for the models dramatically without increasing the number of new studies needed.
In an effort to increase the number of data points in each category an effort was also made to group species according to broader criteria, e.g. Crustaceans & molluscs, Organs (gills, liver and intestines), fish cells and microalgae. Sub-groups of particles were also made based on coatings – steric (PEG, PVP), electrostatic (carbonated, citrates, HA), and uncoated. Media composition was also assessed as this can affect NP size and surface charge, thus a sub-grouping of these was made based on ionic strength (high, low, added biopolymers to disperse), while some studies only use dechlorinated tap-water.
The quantification of bioaccumulation and determination of a bioaccumulation factor for waterborne versus foodborne exposure illustrated that there was an approximately linear relationship for both such that internal concentration increases at same rate as the external rate. Comparing water versus foodborne exposure, a higher bioaccumulation rate was observed via waterborne exposure, possibly as a result of food-avoidance in organisms exposed via foodborne route as the food quality drops when spiked with NPs. The key finding was that the organism is highly important: higher bioaccumulation was observed in cells versus microalgae versus Mollusca etc. This was a much more significant impact than that of the particle characteristics, where only the hydrodynamic diameter affected bioaccumulation, but its impact was low.
The most important conclusions from this work were that organism-related characteristics are more important than particle properties to explain bioaccumulation (bioconcentration) and mortality. A slow elimination of AgNPs was observed for some species and much higher for others, as well as higher uptake from waterborne versus dietborne exposure, although ingestion of NPs is an important route of uptake.
WP4 QSAR models
The aim of WP4 was to develop QSAR modelling of nanomaterial toxicity using the lazar framework. Lazar (lazy structure–activity relationships) is a modular framework for predictive toxicology, whose approach is similar to the read across procedure in toxicological risk assessment in that lazar creates local QSAR (quantitative structure–activity relationship) models for each compound to be predicted. Model developers can choose between a large variety of algorithms for descriptor calculation and selection, chemical similarity indices, and model building. Within ModNanoTox, the open access OpenTox framework (www.opentox.org) for QSAR modelling was extended to include facilities for nanomaterials, which are much more complex that chemicals. Within ModNanoTox, the open access OpenTox framework (www.opentox.org) for QSAR modelling was extended to include facilities for nanomaterials, which are much more complex that chemicals. Work carried out as part of ModNanoTox included the regression of algorithms, which were extended to use quantitative descriptors; in addition, some chemical libraries were integrated to include descriptors, although most still relevant only to small molecules still. Ongoing challenges include the ability to compute NP properties, and this will continue to be addressed via projects such as eNanoMapper. More information on the lazar framework can be found at the In Silico toxicology webpage: http://www.in-silico.de/).
In terms of utilizing the ModNanoTox database as the basis for lazar modelling, similar challenges were encountered as those for bioaccumulation modelling, but here focused around the NP physico-chemical properties: Size, surface area/ charge (zeta potential), shape/ modification, dose, dissolution data are available for various studies in the database, but in most selections of studies, only a subset of the total dataset is complete which leads to missing values. The missing properties are hard to compare due to different methods of measurement and test environments. Additionally, the variety of properties is very sparse for modelling purposes. This is consistent with the fact that a single dataset (NPs in pancreatic cells) has been used to validate multiple different QSAR modelling approaches, and that papers are published without full statistical robustness (i.e. treating the replicates as separate datapoints) at present as a result of lack of datasets. Further evidence for this critical lack of data is apparent from the report on Intelligent Testing Strategies recommendation F.5.2: “There is an urgent need to conduct and verify structure activity relationships using an array of PC properties and hazard endpoints for different NM, so more work at all levels (characterisation, exposure, hazard and risk data) is required to generate and validate such systems”.
Thus, at present the lack of substantial datasets in the field of NPs is presently the limiting factor for lazar QSAR modelling. Even if the models are valid, they will have an extremely narrow applicability domain, which makes them effectively unusable for any realistic application at present. However, as more and more datasets become available from the ongoing EU FP7 projects (and elsewhere) the lazar approach should eventually become applicable to nanotoxicity QSAR modelling. Full details are reported in Deliverables D4.1 and D4.2.
WP5 Exposure & Fate modelling
The main objective of WP5 was to advance and improve environmental exposure assessment models and support other WPs with data on likely environmental concentrations of nanomaterials. An important outcome from this study was the evaluation of existing qualitative and quantitative models and an assessment of how far these approaches can be utilised to understand environmental exposure, as well as the outlook for future modelling and experimental efforts. A detailed assessment and comparison of literature data regarding concentrations of NPs in the environment (focussing on surface water, waste water and sludge) was undertaken, which provided a comparison of measured, modelled and measured & modelled data regarding NP presence in surface waters versus in wastewaters. In all cases, an analysis of the assumptions used in the reported studies, as well as the challenges associated with background NPs and NP dissolution, was provided.
Mechanistic fate modelling of NMs in the environment was also performed (in collaboration with ETH Environmental Fate model). This approach attempted to combine two models, i.e. the model of flows of NMs during production, use and disposal into the environment and some environmental concentrations, with a model looking at agglomeration and heteroagglomeration to natural colloids etc. The team selected as a case study the Glut river in Switzerland, as they had access to detailed information regarding the number of treatment plants, the amount of wastewater output into the river, etc. and thus were looking at NMs coming into the river from households allowing a focus on impact from consumer products containing NPs. Using the model, the team looked at different fate scenarios and found that the most important reaction is the heteroagglomeration of the engineered NPs with natural particles (colloids) in the river water. The study modelled free NPs in water, NPs in water attached to larger particles, and NPs-attached to larger particles in sediment. NPs were found adsorbed to suspended solids and in sediments. If the attachment efficiency was high then NP fate will be primarily influenced by sedimentation, whereas the opposite is likely if attachment efficiency is low. The key learning from this approach is that it is easy to couple the two models as they have compatible ways to calculate the parameters. This was the first time to include mechanistic models of NP concentrations in natural waters. The most significant problem or data gap is that currently it is only possible to do scenario modelling, as literature is missing the alpha parameter for heteroagglomeration, which must be determined experimentally.
Two publications have been submitted bases on this, one of which was highlighted as an editor’s choice. An important outcome from ModNanoTox (which MARINA is taking forward) is that with current exposure & fate modelling methods cannot yet do any validation of the PEC (predicted environmental concentrations) values obtained, and cannot say yet whether measured concentrations are relevant.
WP6 Population models and risk assessment
The two key aims for WP6 were to model effects on ecosystems and then combine these with models from other workpackages to generate a risk assessment framework. Focussing on Environmental Risk Assessment (ERA), current scenarios are based on predicted (PEC) or measured (MEC) environmental concentration, and effect measures divided by a safety factor (PNEC). Bioaccumulation potential (B) and persistence (P) are ‘simple’ cut off criteria.
Mechanistic modelling offers an integrated approach that builds on and complements existing ERA in order to enable:
• Testing of effects under relevant exposure scenarios
• Incorporation of bioaccumulation (toxicokinetics) & chemical persistence data
• Relation of toxicokinetics to individual effects leading to prediction of effects on populations.
Experimentally, it is possible to measure effects at individual or below individual level, but what is required is the protection of populations. However, it is not possible to measure effects at the population level with the required resolution. The current approach is determination of predicted environmental concentrations and lab experiments to get a NOEC value, which is divided by the PNEC value. Biopersistence criteria tell us about persistence / accumulation which are used to determine a cut-off criteria. Thus, the current scheme does represent everything from release to effect. ModNanoTox is proposing a more effective modelling approach that enables determination of the worst case scenario – i.e. effects on populations. The proposal contains all the exiting elements, such as the NOEC/PNEC, but now with spatial and temporal resolution, and including exposure scenarios, and integrating toxicokinetics/toxicodynamics into the population models. In an ideal world where we have the necessary data, it then becomes possible to test population effects under the different exposure scenarios. This is perceived as useful in risk assessment of pesticides, which are recognized as high risk chemicals, but has much broader scope also. This approach also provides the information that comes from the standard approach, and can calculate all the “standard” parameters: PEC, “p” for PBT, “b” for PBT etc. so nothing is lost and much is gained using this approach.
Another important aspect for consideration is which species to test. While this might seems like a small step, in reality for risk assessment there will be a huge impact. ModNanoTox suggests that in order to have a realistic worst-case scenario there is also a need for qualitative assessment of biological traits that can increase exposure. If organisms are not exposed, toxicity is not relevant, i.e. new approach: that incorporates a better integration of exposure & effect measures during the testing planning phase, in order to address those populations most at risk and assess only those. The types of questions that would inform this would include: Where in the environment do we find the highest NP concentrations? Which species are most affected by these concentrations?
An example of the type of qualitative approach could be as follows: Is the NP particle bound, agglomerated or present in surface water, etc.? Depending on how it is found in the different compartments, different species will be affected. E.g. if the NP is primarily located in sediment, then sediment dwellers that feed in the sediment are the most at risk group. Others in close proximity but that do not feed in the sediment less affected. Even if only small amounts left in the water column, some species are efficient at extracting this – for example filter feeders. These organisms will have high exposure. Organisms that do not filter will have low exposure. This is a practical approach, focussing on what is appropriate to test and forms the basis for a simplified risk assessment framework based on a “decision & testing” principle linked to likely environmental behaviour.
The next step is then to add on bioaccumulation (multiple timepoints) & long term testing of effects on survival, reproduction and somatic growth (note timepoints for both should be aligned). These can all be included in a population model (e.g. Dynamic Energy Budget (DEB) approach), which directly relates observed effects to predicted exposures. The approach can be tailored to either dynamic (spatial and/or temporal) exposure scenarios or static worst case PECs. Integrating all the information into the population model can be used to assessment of population level risks. A clear advantage of the approach is that once developed it is not time-consuming to test different exposure scenarios, and potentially also different species (although this requires some re-coding).
Interestingly, the approach developed within ModNanoTox addresses several of the recommendations from the Intelligent Testing Strategy report whose recommendations for risk assessment include:
- “F.3.9 The most relevant species should be identified for specific scenarios (realistic exposure) and guidelines/standard protocols recommended.
- F.3.11 Protocols must be identified that can be applied at all stages of the NM life-cycle and should be carried out at life-cycle stages that best represent realistic exposure scenarios.”
ModNanoTox developed a model for organism starvation using DEB and assessed how starvation varies with the size of animals. The sub-models involved in IBM are feeding rate; functional response; assimilation rate (animals ingest depending on body length, or availability in environment); and assimilation efficiency from good in their gut. Once energy is obtained it can be spent on growth or reproduction. Within WP6, ModNanoTox partners modelled growth and reproduction versus nutrient availability for Daphnia magna initially and assessed how the animals deal with starvation and how this is distributed across the population. The same parameterisation was applied also for Capitella teleta with the result that the outcome at population level exhibits a boom-bust life cycle. Modelling a sediment dilution experiment, whereupon population behaviour in sediments with different carbon contents were assessed, demonstrated that the model predicts well. Sensitivity analysis of the parameters included in the model produced a qualitative rank in terms of importance: In both species parameters related to feeding and survival were most importance – this means then that when applying the approach to other species, there are the parameters to be most careful with.
Two publications are in final stages of preparation from this work, one on the population models and the other on risk assessment.
Integrating and Cross-cutting issues
Quality and reproducibility issues for studies related to aquatic toxicity of NPs
While this is not unique to the field of nanosafety research or nanoecotoxicology, there is a surprising lack of overlap in terms of the studies reported to date in aquatic toxicity of Ag NMs (which is a relatively well-studied subset of all NM ecotox studies). Thus, of 10 papers extracted from the database relating to daphnia magna ecotoxicity, the studies reported address four different life stages (6 on neonates, 1 on 1-day old, 3 on adults of 7 days and 1 on adults of 10ays), Ag particles with 8 different surfaces / capping agents (uncapped, citrate, organic, PCP, coffee, EDTA, ASAP and carbonate) and the particles sizes stated range from 10nm through to <150nm with a cluster around 30±10 nm. Clearly identifying any significant trends from data with this degree of scatter (or lack of between-study replication) becomes extremely challenging, or indeed impossible. Indeed the ModNanoTox attempts at Lazar modelling were restricted for just this reason – there were too many gaps in the dataset for significant training sets to be identified, and particularly for any predictive modelling to be performed. Further analysis of the efforts made and the challenges identified are given in ModNanoTox Deliverable D4.2.
Part of this lack of replication of studies may be related to the lack of high quality and affordable sets of test materials with systematically varied properties such that individual projects / groups are forced to make their own particles leading to ongoing issues of studies not being comparable and part due to the difficulty to secure funding for replication studies, although clearly this has become a topic of very significant debate over recent months, with articles in the Economist (17th and 19th October 2013) on scientific data reproducibility and an initiative by the National Institutes of Health in the US to fund data replication studies, and a series of articles in Nature on the topic following a report in March 2013 that researchers at Amgen pharmaceutical company could replicate only about 10% of clinical studies published in the literature. This has sparked a lot of reaction from scientists, especially in the clinical area, where it is claimed that the experiments are so technical that it can take up to a year of hard graft to master the approaches sufficiently to reproduce existing data and that reproducibility comes with actually seeing how an experiment is performed (by visiting a lab or via a video protocol), as unlike reading, visualization eliminates the errors of interpretation or misinterpretation.
Proposed revision to current risk assessment approach to integrate predicted exposure levels with predictions of likely ecological effects
Combining all the issues described and discussed in the preceding sections, ModNanoTox partners have identified a number of changes to OECD protocols for aquatic and soil testing, that are relatively easily implementable and will greatly enhance the quality of nanosafety data for modelling purposes.
The first and most important is the generation of data at multiple timepoints during the course of exposure, and not just at the start and end. The reasons for this are multiple, but the primary one is to ensure more data for modelling from each study as more data means more robust models. Additionally, temporal information can also allow correlation of onset rates with (for example) uptake rate, and can also allow transient effects, that organisms can recover from, to be observed that would otherwise be missed.
A second key recommendation relates to feeding and depuration steps in experimental design. Lack of feeding is considered to be biologically not relevant, and can mask effects such as enhanced uptake in food or secondary toxicity result from biding of key nutrients. The OECD guideline for acute immobilisation (OECD-2023) specifically states that daphnia should not be fed during the 48 hours of exposure, so likely this is the source of many of the studies in the literature not including feeding as part of their assays. Although for classical chemicals which partition according to their octanol-water coefficient, and are metabolised, digested and excreted via well-known and predicable pathways, this is not an issue, for NMs which are actively transported into cells via receptor mediated pathways, and whose bioaccumulation, biokinetics and biodistribution are not yet understood, some simple alterations to the OECD study design would facilitate the generation of significantly more useful data for modelling.
Additionally, for determination of bioaccumulation it is essential to understand how much of the applied NP dose has been taken up by the organism and actually internalised. Studies that do not include a depuration step include NPs remaining in the gut as part of the internalised dose, leading to overrepresentation of uptake (in a similar manner that NPs stuck to cell membranes can lead to erroneous cellular uptake quantification. Note that OECD test guidelines do not include information on depuration, since the endpoints being assessed (acute mobility impairment (OECD-202) and reproduction (OECD-211) do not consider bioaccumulation or internal dose. Inclusion of these steps are relatively easily accommodated in existing protocols, and would greatly enhance the relevance of the data for modelling purposes, and the predictive capacity of resulting models.
As part of the ongoing dissemination and exploitation activities from the ModNanoTox project, the following list of exploitable outcomes was identified, and linked with the stakeholder groups for which they are deemed to be of most relevance:
1. Database (Excel and ISATab formats) of studies relating to Ag NP toxicity to aquatic organisms (99 studies from 2008-2013): eNanoMapper (FP7), NSC WG4 (databases), ISATab community, EU-US CoR Databases & ontology.
2. Quality Assurance criteria / classification for degree of NP characterisation in literature studies: NSC WGs 1 (materials), 2 (hazard), 3 (exposure) and 4 (databases), EU-US CoR Databases & ontology, Scientific community.
3. Adaptions to OECD protocols for aquatic assessment to include temporal information and facilitate determination of internal dose: Scientific community, QualityNano (FP7) for Inter-laboratory comparison, OECD and CEN, Regulatory organisations, MARINA, NanoValid, NanoMILE, FutureNanoNeeds & NanoREG.
4. 4 Modelling approaches and outcomes (including what didn’t work & why) on fate & exposure, bioaccumulation, QSAR models and population models: NSC WG6 (Modelling), EU-US CoRs on modelling, Modelling cluster of projects.
5. Atomistic modelling data and outputs regarding effect of size on particle stability, and NP dissolution: EU FP7 NanoTransKinetics & FutureNanoNeeds, NSC WG1 (Materials) and 6 (Modelling), NP producers / SMEs, Regulatory agencies.
6. Quality and reproducibility issues for studies related to aquatic toxicity of NPs: Scientific community, QualityNano (FP7), MARINA, NanoValid & NanoREG.
7. Proposed revision to current risk assessment approach to integrate predicted exposure levels with predictions of likely ecological effects: Risk assessors / regulators including ECHA and EMA, Policy makers, Industry, including pesticides, NSC WG5 (Risk assessment), NanoREG.
These outputs cover a diverse range of activities, and address a number of different stakeholders. Many of the identified stakeholders are accessible via the NanoSafety Cluster (NSC), including the NSC Working groups (WGs) and the EU-US Communities of Research (CoRs) as well as specific projects mentioned (e.g. NanoTransKinetics, QualityNano, MARINA, NanoValid, NanoMILE, eNanoMapper, FutureNanoNeeds, SUN and NanoREG). In addition, for each of the listed projects, specific plans are already in place / currently being implemented to ensure direct sharing of knowledge and incorporation of ModNanoTox findings into the ongoing work of these projects. These plans are summarised below.
This is the sister project to ModNanoTox, but has a 36 month duration (compared to ModNanoTox’s 24 months) and this is continuing with its activities (specifically around the nanoparticle corona and how this determines uptake by cells and biological barriers). Via the joint project meetings and bilateral collaborations with Giancarlo Francesze (University of Barcelona, partner in NanoTransKinetics), ModNanoTox has fed its data regarding particle stability models at the atomistic level from WP1, which they are now scaling-up to a more coarse-grained modelling. Additionally, UoB and University of Barcelona are collaborating on modelling of aggregation of small particles (highly transportable biologically or NP fragments) and how they interact within a confined space in order to understand the effect of stabilising and the effect of shape on both the ligand (orientation, adsorption) and NP surface properties and where/how they agglomerate. The initial focus is on citrate as the capping agent/ligand which is a commonly used stabiliser for metal nanoparticles, and particularly relevant to Ag nanoparticles, but it is also being used as a model of other biomolecules with similar functional groups / similar steric arrangements to allow development of predictive models.
QualityNano is the FP7 research infrastructure for nanosafety assessment whose primary goal is to improve and enhance the quality of nanosafety assessment, via provision of access to state of the art facilities, training in protocols and best practice, and facilitation of inter-laboratory comparisons and/or round robins (RRs). To date the RRs have focused on NP physicochemical characterisation and in vitro toxicity testing, and nothing has been done on either in vivo studies or ecotoxicity. This presents a unique opportunity for onward exploitation of the ModNanoTox revisions to OECD-211 for daphnia reproduction studies, which is being presented to QualityNano as a proposal for RR evaluation within WP2. The revised OECD-211 test guidelines include additional temporal data (multiple time points from 0-21 days), depuration and sample collection for ICP-MS to assess NM uptake. Performance of such as an ILC would be a very important step forward for the field, and something that UoB and the ModNanoTox partners would be willing to champion via QualityNano. A proposal to this effect is being submitted to QualityNano in early 2014 as a matter of priority, and reflects a key stage in the final ModNanoTox exploitation plan.
Another key area where QualityNano can support the onward dissemination / exploitation of ModNanoTox’s outputs, specifically those related to inclusion of mechanistic and population modelling into risk assessment, will be presented at the regulatory stakeholder training / dissemination session planned for the final QualityNano conference in 2015.
One of the two FP7 projects focussed on development of reference materials and methods, MARINA aims to develop and implement and intelligent testing strategy and a MARINA will a strategy for Risk Management integrating the advances in Human Risk Assessment, Environmental Risk Assessment, Management of Accidental Risk and Risk reduction strategies. ModNanoTox can, and is (via project partners EMPA and UoB) feed into this process. Specifically, the ModNanoTox modelling data from WP5 regarding NP environmental exposure models and the ModNanoTox data on bioaccumulation models (WP3) will be fed into MARINA’s WP7 on Environmental Fate/behaviour assessment. Additionally, ModNanoTox’s data on population & mechanistic modelling and the recommendations regarding simple additions to the current risk assessment framework will be fed into MARINA (WP13 – Environmental Risk Assessment).
Focussed on development of reference methods for hazard identification, risk assessment and LCA of engineered nanomaterials, NanoValid shares one common partner with ModNanoTox (UoB) and action has been taken to ensure direct transfer of knowledge and project outcomes, especially in terms of bioaccumulation modelling and particle stability modelling. Protocols and data have been shared and indeed NanoValid is also a route for RR comparison assessment of protocols as needed.
The ModNanoTox coordinator is also coordinating NanoMILE and thus there has already been significant flow of information from ModNanoTox into NanoMILE, specifically in terms of the database approaches being shared with the knowledge management partner for NanoMILE (Biomax Informatics AG) as well as with the QSAR modelling partner (NovaMechanics). Additionally, the deliverables relating to NP descriptors for modelling (D2.1 and D2.2 relating to carbon-based and metal-based materials, respectively) and the database deliverable (D2.4) have been fed into NanoMILE as a solid basis for onward development of nanoSARs.
The modified OECD-211 protocol for Daphnia reproduction is also being shared with NanoMILE, and NanoMILE partners have indicated willingness to participate in an inter-laboratory study of the revised protocol, facilitated via QualityNano.
In order to ensure maximum uptake of the ModNanoTox database beyond the lifetime of the ModNanoTox project, it was agreed to share the database deliverables with the EU FP7 funded project on database and ontologies for nanosafety, eNanoMapper, which started in late 2013. An agreement will be put in place between the ModNanoTox coordinator and eNanoMapper in order to ensure that outcomes building on, or utilising, the ModNanoTox database acknowledge the very significant effort dedicated to the development of the database by the ModNanoTox team. Up to date ISAconfigurator files for the ModNanoTox database can be found at the ISATab public online repository.
Moving beyond so-called legacy NPs, such as the OECD test materials, FutureNanoNeeds will assess how characterisation, hazard and exposure approaches used in existing regulatory frameworks need to be adapted to account for non-spherical and multi-component NMs. ModNanoTox’s findings from modelling NP behaviour in water, and the link between transportability and surface roughness & fractal dimensions may be important in terms of non-spherical particles. The simulations in WP1 found that for Ag NPs, surface roughness depends on particle shape, which in turn is influenced by particle size. This could have important implications for complex particle structures (e.g. nanoflowers), and indeed will also affect the nature and conformation of bound biomolecules (corona). ModNanoTox are feeding this data forward into FutureNanoNeeds also, and offering molecular modelling as a potential route to predict effects from geometrically complex NPs.
UoB’s role in FutureNanoNeeds involves assessment of NP impacts in daphnia connected to exposure route and determining metabolomics impacts and corona evolution in parallel to the NP localisation (based on stable-isotope analysis). The ModNanoTox modifications to the mobilisation and reproduction assays will also be shared with FutureNanoNeeds and as far as possible temporal and bioaccumulation aspects will be considered.
NANoREG is developing a common European approach to the regulatory testing via a concerted effort to harness ongoing research activities at the member state level. It aims to provide legislators with a set of tools for risk assessment and decision making instruments for the short to medium term, by gathering data and performing pilot risk assessment, including exposure monitoring and control, for a selected number of nanomaterials used in products; as well as developing new testing strategies adapted to a high number of nanomaterials where many factors can affect their environmental and health impact. ModNanoTox’s approaches to mechanistic and population modelling are important additions to the regulatory tool box, and will be fed into NanoREG via the various member states (UK, Denmark, and Switzerland) as well as by the individual partners.
The Sustainable nanotechnologies project aims to evaluate environmental, health and safety (EHS) risks along the lifecycle of manufactured nanomaterials and incorporate the results into tools and guidelines for sustainable manufacturing. ModNanoTox data regarding exposure & fate modelling is being developed further in SUNM via EMPA and the former EMPA postdoctoral fellow funded from ModNanoTox (Dr. Fadri Gottschalk) who has subsequently started a small SME on risk assessment (ETSS) and is now also a partner in SUN alongside EMPA. The next steps for the model (to be developed within SUN include the need to incorporate a much better description of NP dissolution and the role of NOM and other binding parameters. The model could also add speciation and concentration of NMs e.g. based on stable isotopes to look at dissolution in real water at very low concentrations (data which could come from NanoMILE).
This snapshot illustrates the projects where ModNanoTox partners have identified clear linkages and pathways for impact, but the consortium’s hope is that this document will provide signposts for other projects regarding the data available within ModNanoTox, and the coordinator is open to sharing the non-public deliverable reports upon request, and subject to appropriate acknowledgement of ModNanoTox efforts and data-sharing in any resulting outputs.
Summary of key ModNanoTox outputs mapped to NSC WGs
This section summarises the key outcomes from ModNanoTox of relevance for each of the WGs, and points interested readers towards the more detailed information, including relevant ModNanoTox deliverable numbers should more detailed information be required.
WG1: Materials: Data on NP stability as function of size, crystal face, ligand etc, NP characterisation – minimal data for aquatic toxicity based on PCA analysis, NP characterisation – gap analysis and recommendations - D1.1 – D1.4 D2.4 and relevant future publications.
WG2: Hazard: Recommendations regarding adjustments to OECD protocols (e.g. temporal data)
Results from bioaccumulation models, Results from population models, Recommendations regarding data replication / reproducibility, Recommendations regarding modelling approaches e.g. Lazar approaches - D6.1 D6.2 D3.1-D3.3 D6.3 D2.4 D4.1-4.3 and relevant future publications.
WG3: Exposure: Data from exposure modelling in surface water, waste water and sludge, Mechanistic fate model for NMs in the environment combining flows of NMs during production & use with model looking at agglomeration and heteroagglomeration to natural colloids, Gap analysis in terms of data availability (surface binding, heteroagglomeration) - D5.1 D5.2 D2.4 and relevant published and future papers.
WG4: Databases: Database of 99 aquatic studies in excel & ISATab formats, QC and NM characterisation classification, Data gaps & resultant challenges for QSAR - D2.3 D2.4 D3.3 and relevant future publications.
WG5: Risk assessment: Recommendation for inclusion of mechanistic and population modelling in environmental risk assessment, Adaptions to OECD protocols for environmental end-points to allow temporal collection and facilitate bioaccumulation modelling - D6.3 and relevant future publications.
WG6: Modelling: Challenges in modelling limited data sets, Advances to Lazar framework for NPs (Extended regression algorithms to use quantitative descriptors; included some chemical libraries to integrate descriptors), Approaches for population models, Approaches for bioaccumulation modellingApproaches for NM modelling, Approaches for environmental fate modelling - D4.1 D4.2 D4.3 D6.3 D1.3 D1.4 D5.1 D5.2 and relevant published and future papers.
In terms of dissemination to regulatory agencies and industry, the present document will serve as a first signposting, as well as the targeted publications from WP6, and WP2 will aim to present at the final QualityNano conference in early 2015 specifically in the regulatory session. Other dissemination activities are not planned at present (as the project is now ended) but consortium partners will continue to cite the project outcomes in their ongoing communication activities, and the numerous other projects which ModNanoTox has fed into (as described above) will also help to ensure that ModNanoTox’s outputs will have enduring impact beyond the project.
List of Websites:
Public web site: http://www.birmingham.ac.uk/generic/modnanotox/index.aspx
Coordinator's contact details: Professor Eugenia Valsami-Jones, University of Birmingham, E.email@example.com personal page: http://www.birmingham.ac.uk/staff/profiles/gees/valsami-jones-eva.aspx