Developing Prototypes of a Modernized Approach to Assess Crop Protection Chemical Safety

In 2019, the US EPA Administrator issued a directive directing the agency away from reliance on vertebrate tests by 2035, whilst maintaining high quality human health and environmental risk assessments.  There is no accepted approach to achieve this. The decade long duration of the crop protection (CP) chemical R&D process therefore requires both the invention, and application, of a modernized approach to those CP chemical projects entering corporate research portfolios by the mid-2020s.  Consequently, we conducted problem formulation discussions with regulatory agency scientists which created the problem statement: "Develop, demonstrate, and implement a modern scientifically sound and robust strategy that applies appropriate and flexible exposure and effects characterization without chemical specific vertebrate tests to reliably address risk, uncertainties, and deficiencies in data and its interpretation with equivalent confidence as do the currently accepted test guidelines and meet the regulatory needs of the agencies".  The solution must provide the knowledge needed to confidently conclude human health and environmental protective risk assessments. Exploring this led to a conceptual model involving the creation, and parallel submission of a new approach without reliance on chemical specific vertebrate tests.  Assessment in parallel to a traditional package will determine whether it supports some, or all, of the necessary risk management actions.  Analysis of any deficiencies will provide valuable feedback to focus development of tools or approaches for subsequent iterations. When found to provide sufficient information, it will form the technical foundation of stakeholder engagement to explore acceptance of a new approach to CP chemical risk assessment.

mation gaps, and (v) engage and communicate with stakeholders.Similar activities are being developed around the world, e.g., in Canada (Bhuller et al., 2021), the EU (Escher et al., 2022), and the UK 2 .In addition to any new science that may need to be incorporated, these roadmaps also share elements with the US EPA NAM workplan, particularly concerning the importance of establishing confidence, demonstrating application, and engagement with stakeholders.Indeed, opportunities for collaboration and discussion of case studies have been created between regulatory agencies (Kavlock et al., 2018) or exist through the IATA program at OECD (OECD, 2020).

The problem formulation
The overall goal of this work was to develop a new approach for knowledge generation that meets the regulatory needs whilst substantially reducing, and ultimately eliminating, reliance on currently required vertebrate animal toxicology testing.It is also critical to meet the regulatory and societal need to register safe uses of new crop protection AIs.By selecting potential case studies from our existing late research portfolio, we can conduct and evaluate parallel case studies that will act as an architype to explore this goal.Using the available data tools, frameworks, and NAMs that are available now will permit the definition and refinement of the uncertainties on exposure and hazard to determine whether they provide sufficient precision to make consistent, appropriate risk management decisions.Doing this should enable confidencebuilding across stakeholders in the use of new approaches to perform the risk assessment of new AI uses.Further, it will identify what gaps, if any, exist in the available toolkit and thereby identify future research needs.
However, considering the above, it was clear that there was no obvious way to select which of the many tools and frameworks would best be utilized to achieve the goals of the Administrator's directive within the originally proposed timeline.Therefore, it was necessary to rapidly understand how one could transform our approach to address the implications of this US EPA directive.To accomplish this, we used published problem formulation approaches that lay out the components of a problem formulation effort, including planning and scoping, and developing the problem statement to be addressed (NRC, 2009;Sauve-Ciencewicki et al., 2019).
The initial planning and scoping of the problem addressed the issues around evaluating a new chemical, scientific drivers, and opportunities with changing technology for both human safety and environmental risk assessments.As new AIs for crop protection are rarely used or registered only in a single country, these discussions were held with multiple regulatory agency scientists, in separate sessions, to gain a breadth of ideas and concerns to be organized into several separate Syngenta-agency problem statements.The focus of the problem formulation discussions was mainly on cal active ingredient (AI) research programs begin by identifying biologically active chemical leads.These are then expanded into thousands of candidate molecules that are optimized over several years to design and select a few tens of molecules for further consideration.This optimization process uses high-throughput efficacy screening studies with panels of assays and computer models to eliminate those that have features known to drive toxicity, undesirable environmental fate properties, or have inadequate effect on the target pests or plant diseases.Subsequently, over the next 1-3 years, these molecules are further narrowed to just a few through low-throughput testing including whole animal studies and initial field efficacy studies.When a single molecule is finally selected, the remaining development phase can typically take an additional 8 years to generate the requisite regulatory efficacy and safety information.It is this final phase that utilizes the traditional US EPA and OECD1 test guideline studies (US EPA, 2023a) in scope for elimination by the US EPA's directive.However, there is no known and accepted approach to broadly meet this need without using the traditional guideline vertebrate studies.

Availability of potentially useful new approaches
Despite this clear gap in regulatory science practice, considerable scientific progress has been made by the combined activities of the academic, regulatory, and industrial scientific communities.This has resulted in the development of many potential new approaches at a wide variety of levels of technical and regulatory readiness.Some, such as in silico modelling, read-across and grouping approaches, and some simple in vitro or ex vivo tissue culture systems, are already used.Others, such as high-throughput and highcontent data sources including multiple omics-based test systems; complex ex vivo and in vitro assays such as microphysiological organ-on-a-chip systems; in vitro stress and bioactivity assay panels; in-vitro-to-in-vivo extrapolation approaches; and machine learning tools, are areas of exploration.Any of these could, in theory, be applied to meet the needs of the directive, but it is unclear how to make effective use of this modern science for regulatory decisionmaking on a new agrochemical.

Regulatory agencies and approaches to evolution in regulatory science
Nevertheless, it is now the right time to explore this evolution in regulatory science.The US EPA has been working toward this goal for almost 20 years through its computational toxicology research program and has most recently published a revised plan for moving away from the use of whole animal studies in research and for chemical registration decisions (Kavlock et al., 2003;US EPA, 2021).The US EPA NAM workplan has defined five objectives to achieve this goal of reduction in animal testing whilst ensuring the Agency decisions are protective of human health and the environment: (i) evaluate regulatory flexibility, (ii) develop suitable baselines and metrics, (iii) establish scientific confidence and demonstrate application, (iv) develop NAMs to address infor-To register a pesticide, an agency needs to address the sufficiency of data for making a risk management decision (see, e.g., US EPA, 2013).Consequently, agencies that focus on risk assessment as the driver for their risk management determinations will need knowledge or data to enable a quantitative assessment of exposure and hazard to derive conclusions on the risk to humans and environmental species.Table 1 lists an example of these risk assessment needs.

Study/data requirements
Most of the demand to use vertebrate animal models is related to hazard identification and characterization to inform the human health risk assessment needs for multiple potential types of hazards, susceptible life-stages, and routes and durations of exposure (Fig. 1A).However, whilst the intent of ecotoxicological testing is also to generate data that address key knowledge needs for a risk assessment, this risk assessment need is generally aimed at protecting populations in an environmental compartment, rather than individuals.The requirements to protect endangered species at the individual level is a notable exception to this.In many instances, an iterative process is required, and data may be collected in tiers of increasing complexity and costs (US EPA, 1992).Ideally, this process will reflect a tiered ecological risk assessment approach, in which acute ecotoxicity tests are conducted to identify potentially susceptible taxa, followed by more intensive ecotoxicity testing for those species with the greatest risk.While such an approach has historical acceptance (Suter, 2008), many chronic ecotoxicity tests remain compulsory for pesticide registration, including those involving vertebrate animals (40 CFR Part 158) 1 .Consequently, several vertebrate and non-vertebrate studies currently performed are used to identify potentially susceptible taxa and characterize hazard for these species (Fig. 1B).In addition to toxicity and ecotoxicity testing, vertebrate studies are used to understand adsorption, distribution, metabolism and excretion (ADME) (usually rats), aquatic bioconcentration (fish), and food residue levels (poultry, ruminants).Broadly similar lists of mandated study data requirements can be drawn up for other agencies across the world, the strengths and weaknesses of traditional toxicology and ecotoxicology data and how they are used to inform risk assessments based on vertebrate animal studies.In addition, they clarified the diverse agency viewpoints, concerns, drivers, and practical requirements in transitioning to using NAMs for agrochemicals.The separate problem formulation discussions were ultimately merged into a single overarching problem statement encompassing the totality of viewpoints.
The issues brought up across the various discussions included the use of too many animals that did not ultimately contribute to decision-making, uncertainties in the relevance of endpoints, challenges in integrating NAMs into the testing paradigm, confidencebuilding among regulatory agency scientists and with the public, the post-registration expansion of uses which alter exposure scenarios, scientific consensus on applicability of NAMs, inadequate use of prior knowledge, and a lack of focus on the risk assessment and risk management needs.One essential component was that to meet the goal for agrochemical risk assessments, we would need to cover both human and environmental safety assessments and so consider all testing across vertebrate taxa.These discussion points led to the creation of an overarching problem statement defining the needs.
Problem statement: Develop, demonstrate, and implement a modern scientifically sound and robust strategy that applies appropriate and flexible exposure and effects characterization without chemical specific vertebrate tests to reliably address risk, uncertainties, and deficiencies in data and its interpretation with equivalent confidence as do the currently accepted test guidelines and meet the regulatory needs of the agencies.

Exploring this problem
With the establishment of an agreed upon problem statement, the problem was further explored to identify the breadth of questions and issues that would need to be addressed to create a conceptual model of the project (Sauve-Ciencewicki et al., 2019).ing studies while maintaining high confidence in safety and risk decisions (Craig et al., 2019).As a second example of a regulatory opportunity that could enable NAMs, the Australian regulatory approach for pesticide evaluation has an inherent flexibility.In this case there is no specific list of test guideline studies required by law, but rather a set of risk decisions that need to be determined.
The law permits the Australian Pesticides and Veterinary Medicines Authority (APVMA) to issue guidelines that describe how APVMA produces its determination, and therefore creates the potential for rapid adoption of NAM-based approaches via updating of its data guidelines.

Development and evolution of new test methods
There has also been considerable research activity in the academic and industrial development of in silico and in vitro experimental NAMs over the last 40 years.Despite all this research effort, in general these approaches have not yet delivered the long hopedfor guideline in silico or in vitro test methods that are broadly accepted as replacements for all vertebrate studies.Originally, this effort focused on creating in vitro test methods or test batteries that can replace existing guidelines studies, rather than focusing on the actual risk assessment knowledge need.The hope was that a simple substitution of a regulatory test method based on animal experimentation by an equivalent non-animal method would be achievable.Therefore, the existing animal study outcomes were used as the prediction goal to validate assay performance.Apart from a few notable exceptions (see below), this approach has not yet succeeded.Later, responding to the U.S. National Academy of Sciences 2007 report Toxicity Testing in the 21 st Century: A Vision and a Strategy, there was a switch to a focus on mechanistic considerations and testing of pathways of toxicity using a battery of test methods.Recently, there has been a shift back to re-exploring in vitro models of target organs using advanced cell cultures such as human organotypic systems (Nitsche et al., 2022).However, there has been little investment in these technologies to address ecotoxicological risks.Furthermore, when regulatory acceptance of NAMs is clearly understood, they are used by industry (e.g., direct contact acute toxicity, study waivers based on weight of evidence assessments).What is perhaps less visible to the wider scientific community is that the development and application of NAMs by industry to support new AI invention is commonplace to aid chemical design and selection decisions.

Historic experience of agrochemical R&D
Through decades of experience gained from studies on both research and registered agrochemicals, combined with the historic and continuing efforts in investigative toxicology, a detailed mechanistic understanding has been developed, across many pesticidal MOA and chemical classes, of the actual on-and off-target molecular initiating events (MIEs) and the associated adverse outcome pathways (AOPs) that agrochemicals frequently impact.Consequently, it is often possible to draw conclusions on one molecule based on other molecules in a group.For pesticides, these groups may be established by MOA (e.g., Herbicide, Fungicide and Insecticide Resistance Action Committee (HRAC, FRAC generally for similar purposes (see, e.g., the comparison in Chen et al., 2023).
For ecological risk assessment, it is important to consider these needs in the context of specific environmental compartments.Several test guidelines are available to address aquatic ecotoxicity in multiple taxa, including plants, invertebrates, and fish.Because these species inhabit a common environmental compartment with similar exposure, their effects data could be considered together to inform the risk assessment.For example, if acute toxicity screens indicate that plants are the most sensitive taxa, then protecting aquatic plants would be protective of invertebrates and fish.Such an approach is particularly useful for certain pesticides that are known to target specific organisms through a known mode of action (MOA).Similar data-driven logical arguments can be made for suggesting that it is unnecessary to conduct additional testing with vertebrates if they will not be the primary driver of risk management decisions.

Regulatory knowledge needs for risk assessment and study data requirements are not synonyms
Whilst exploring the problem, it became critical to differentiate between the mandated guideline study data requirements (that are now usually conducted to inform risk assessments), and the actual knowledge needed to support decisions made by a regulatory agency.The results of the toxicity studies using vertebrate animals (e.g., a rat chronic no observed adverse effect level (NOAEL)) are not the knowledge needed, although they have historically provided useful information to help meet that actual need.The knowledge needed is in fact the conclusion on the risk (e.g., a chronic dietary risk assessment).As mentioned above, the agencies and all stakeholders must have confidence that the data used to support those conclusions is fit and suitable for that purpose.With sufficient flexibility in permitted approaches to meet this need, other lines of evidence could be appropriate for use in a new scientifically defensible approach to pesticide risk assessment.
A critical insight from exploring the problem was that by focusing on meeting the requisite risk assessment needs, one might draw on multiple lines of evidence to generate the necessary information to reach risk-based conclusions without the need for chemical-specific vertebrate guideline studies, thereby addressing the problem statement.

Flexibility of practice
One pre-requisite for using a new approach is that the framework of legal requirements and administrative interpretative guidance permits and enables its actual use and does not inadvertently result in a de facto technological "lock-in" to the existing paradigm when potentially superior approaches are available.There are several ways to achieve this in the different regulatory systems around the world.For example, the US EPA Office of Pesticide Programs published a document that identified guiding principles for data requirements and included statements regarding the US EPA's flexibility in determining data needs, alternative approaches to answering the risk questions, and opportunities to waive test guideline studies (US EPA, 2013).The outcome of this flexibility was illustrated in a recent publication that quantified the value of waiv-NAMs to also address regulatory endpoints (Ramanarayanan et al., 2022;OECD, 2022) or through risk assessment-based, weight of evidence approaches to study waivers (Hilton et al., 2022).

Development of new conceptual methods
This collective research effort has shown that it is currently not possible, nor reasonable, to expect a substitution of all in vivo endpoint data with in vitro tests that provide the same types of hazard information as are being generated now using guideline animal studies.Therefore, new conceptual thinking will be needed to incorporate modern scientific approaches and maintain safety, or we must wait for some unknown future time when our ability to model all relevant biology without animal studies develops sufficiently to meet this challenge (Wolf et al., 2022;Bhuller et al., 2021).
This need for new thinking has been recognized and explored, resulting in numerous approaches having been proposed as possible ways forward.For example, a HESI (Health and Environmental Sciences Institute) project proposed a framework able to incorporate new approaches in toxicology and exposure assessment until sufficient information to make a risk assessment decision was known (Pastoor et al., 2014;Embry et al., 2014).The US EPA Computational Toxicology program proposed, and exemplified, an approach to derive health protective points of departure for use in risk assessments, using a tiered approach based on in vitro assays with in vitro-to-in vivo extrapolation (Thomas et al., 2019;Paul Friedman et al., 2020).Similar approaches have been proposed for use in cosmetic ingredient risk assessments to address the challenge posed by vertebrate studies being banned in the EU on cosmetic ingredients since 2013 (Dent et al., 2021;Middleton et al., 2022;Alexander-White et al., 2022).Also, Wolf et al. (2020) proposed a combination of the best available science including using new methods and existing data and highlighted the need for iterative industry and regulatory engagement to ensure that sufficient information to make a risk assessment decision was generated.In general, these approaches rely on an iterative integration and generation of multiple lines of evidence to refine uncertainties in exposure and hazard data to provide sufficient certainty in a risk assessment.

Developing the conceptual model
Mindful of the problem exploration described above, we then developed an overall conceptual model for our project that should answer the problem statement (Fig. 2).The approach to creating a solution to our problem statement is to work with willing agencies and make a parallel regulatory submission to evaluate all the human and environmental safety and risk concerns using both the traditional animal toxicology data and a package of information based on new approaches.This will enable a direct comparison of the assessments from both the NAM effort that addresses the problem statement and a traditional dossier that includes all the requisite test guideline studies.In this way we will be able to identify the technical, financial, administrative, and/or legal challenges that will need to be solved.It will create real-world exemplars for and IRAC) classification), structural similarity, or other means of establishing similarity, such as clustering of bioactivity profiles.Since the two initial pieces of information available for an AI in the research pipeline are its chemical structure and pesticidal MOA, it is frequently possible to predict which species are most likely sensitive for ecotoxicology assessments and which hazards are likely to occur in the toxicology studies performed for human safety assessments.
An historic focus on avoiding certain undesirable toxic hazards has motivated the creation of complete catalogues of all possible MIEs that may drive certain toxic endpoints and the use of in silico model / in vitro assay combinations to explore them (e.g., the "toxome" for developmental and reproductive toxicities produced by bioinformatic analysis to produce a comprehensive DARTable genome, Janowska-Sejda et al., 2022).This mechanistic understanding provides a direction to plan toxicity testing at early stages of a research project, where chemical (quantitative) structure-activity relationship and MOA information are used routinely to predict toxicity in early stages and prioritize chemicals for selection or deselection on this basis.Later, a combination of this information with a chemical-specific understanding of ADME, integrated using biologically-based mathematical models, results in quantitative AOP models that also inform compound selection.Furthermore, the application of transcriptomics and metabolomics measurements is now routine in early shortterm in vivo studies to provide further mechanistic insights and estimates of chronic no-effect levels.
This combination of quantitative AOP understanding with broad surveys of in vivo "bioactivity" provided by the omics methods aids chemical selection and progression decisions from research into full development.However, these modern data generated for progression decision-making are currently not required by, nor are they routinely used in, regulatory submissions that instead rely on a study paradigm that has evolved from an approach first described in the middle of the 20 th century (Lehman et al., 1949).Such studies may, however, be useful additions to permit informed waiving based on protective omics points of departure (Johnson et al., 2022).In addition, as these non-guideline research-phase studies are not in scope for the US EPA workplan describing vertebrate testing reduction targets, they may have some utility in reducing the need to conduct some animal studies and building confidence in the transition to vertebrate testing-free risk assessments.
There has been some progress in adoption of NAMs for screening and prioritization (Judson et al., 2015(Judson et al., , 2018;;Kleinstreuer et al., 2017) but much less so for application to support regulatory decisions and specifically for use within regulatory risk assessment.There are some notable exceptions where replacement has indeed succeeded.One such case is for direct contact acute toxicities driven by simple chemical mechanisms, which resulted in the skin sensitization approaches adopted into a regulatory guideline (OECD, 2023).Also, as regulatory practice continues to evolve, some NAMs are starting to be used as supporting information to aid the understanding of potential mechanisms driving toxicity that might occur and so are used in addition to, rather than as a replacement of, guideline studies.More recently, some in the agrochemical industry have also started to adapt the use of some groups, e.g., 62% of herbicides, 42% of fungicides, and 36% of insecticides are covered by the top 5 MOA groups of each (HRAC, FRAC and IRAC classification data accessed November 2022).Therefore, these data-rich MOA groups and uses are desirable to act as exemplars of our approach.If successful, this will enable the adaptation of the approach into an efficient general method for any MOA group, i.e., an IATA which is likely to be reused in multiple future examples.Also, as it is likely that the multiple lines of evidence generated to support the initial parallel submission will contain redundant information, each line of evidence (or combinations) could be tested for its sufficiency to meet agency need.In this way our case studies will also inform on the approaches required to support smaller or new MOA groups, where the same wealth of comparative data is not available.
Additionally, we need to consider how to ensure the best use of available regulatory agency resources for their evaluations of the parallel submission.Therefore, there needs to be alignment between the proposed possible agricultural uses on crops in a particular country, or perhaps a requirement for an import tolerance for pesticide crop residues in traded commodities, and the research projects that might form suitable case studies.wider stakeholder discussions to help build confidence in the approach.We acknowledge that it may not be possible to gain regulatory approval without animal data the first time.Therefore, we may identify new research goals that need to be solved before we can successfully deliver this integrated new approach for registration.Also, based on the historic precedent of the evolution of the current paradigm, confidence will be built on repeated successful practice of the new approach.Therefore, there are likely to be multiple iterations to ensure we can generalize the methods to cover all possible AIs and MOA groups.

Choice of exemplar
The choice of projects to use as case studies to develop the new approach is driven by the intersection of technical, regulatory, and commercial considerations.At its core the technical solution will require an assessment of the weight of all available relevant evidence.Fortunately, there is a wealth of historic toxicology and ecotoxicology data available for pesticides that provides valuable information for these comparative assessments.Due to commercial resistance management and chemical tractability considerations, AIs are highly skewed towards a few privileged MOA

Fig. 2: Conceptual framework of our proposed Product Safety Regulatory Science Research (PARTNER) Project to develop and build confidence in a new modernized approach to evaluate agrochemical risk
A conceptual model of the PARTNER project that when completed will address our problem statement: "Develop and implement a modern scientifically sound and robust strategy that applies appropriate and flexible exposure and effects characterization without chemical specific vertebrate tests to reliably address risk, uncertainties, and deficiencies in data and its interpretation with equivalent confidence as do the currently accepted test guidelines and meet the regulatory needs of the agencies."The parallel evaluation of traditional and NAMbased regulatory submissions will, depending on the outcomes, be used to build confidence in the approach or to find gaps that need to be addressed to meet the risk assessment needs of the regulatory agencies.
and environmental fate parameters may also still be under development at this stage.However, pre-and early post-emergent foliar herbicide uses are well understood, with many historical examples that can inform the exposure assessment.While the final GAP and AI-specific information are still being developed, the existing approaches to estimate exposure based primarily on the anticipated and well-understood use scenarios for other similar products can be utilized.The surrogate residue and environmental fate information for suitable existing products may be derived from publicly available risk assessments such as EPA's risk assessment documents, the Joint FAO/WHO Meeting on Pesticide Residues (JMPR) documents, or maximum residue level (MRL) records.The exposure values derived through these approaches can be used to make credible first-tier predictive risk assessments and drive either decision-making or further refinement of the uncertainties of exposure or hazard (Pastoor et al., 2014).

Parallel alternative data package
At the core of this new approach is the parallel creation of an alternative data package that will be suitable to inform the risk assessment needs of the agencies.This alternative will then be assessed and compared to a traditional data package to determine whether it also supports appropriate risk management actions for human and environmental safety.For our initial case studies, we will maximize the reuse of existing information by performing a comparative assessment of our new AI.Our safety hypothesis is that, when sufficient structural, bioactivity, and toxicological effect similarity is established in a "data-rich" MOA group using multiple diverse lines of evidence as shown in Figure 2, then the hazard endpoints for both human health and environment protective risk assessments of a new pesticide in that group can be defined based on the existing published regulatory conclusions.
In our new ACCase herbicide case study, these existing data are derived from the 19 HRAC Class A herbicides of the aryloxyphenoxypropionate (FOP), cyclohexanedione (DIM) and phenylpyrazomline (DEN) structural classes, and from the 4 IRAC Group 23 insecticides of the tetronic and tetramic acid (TA) derivatives class.Our safety hypothesis necessitates the ready use of the regulatory agencies' existing conclusions on these chemicals.Therefore, our first step was to identify relevant endpoints used in the risk assessments on all these chemicals.Although the utility of making such safety data accessible, usable, and integrable has been highlighted previously (e.g., Lanzoni et al., 2019), these data are not yet completely available in a suitable format or publicly accessible database.However, this information is available in unstructured formats in the published documents describing regulatory decisions.Additionally, for aquatic ecotoxicity data there has been an effort to curate relevant endpoints into the Envirotox database (Connors et al., 2019).Therefore, we created a workflow illustrated in Figure 3 to access these requisite data.For human and environmental risk assessments we use these data in slightly different ways, as introduced below and described further in upcoming publications.
Finally, as we need to develop a parallel submission to generate confidence in decisions made using the new approach, there must be a reasonable expectation of a sufficient commercial case to continue the investment in the data generation using the traditional approach and registration in those countries willing to perform parallel evaluations.
In our case, these constraints result in preferences for projects that have potential uses in corn, soybean, and/or cereals and a preference for weed or disease control over insect control solutions.Our first example project that met these criteria was selected to be a novel acetyl-CoA carboxylase (ACCase) inhibitor herbicide with potential uses in soybean weed control.

Exposure assessment
Exposure assessments for pesticides are conducted using standard assessment methods and tools based on Good Agricultural Practices (GAP), product-specific residue and metabolism data, and environmental fate information for the new AI.
Human dietary exposure assessments are based on standard consumption rates for different sub-populations and empirical residue concentration in food items.The consumption data is generally derived from consumer surveys and the residue concentration in food items is generally derived from regulatory residue studies.Often in screening-level assessments, the maximum residue level (MRL) for food commodities is used, which may be refined in higher-tier assessments using data from field residue studies or food monitoring programs.Regulatory dietary assessments are conducted using established and well-tested tools such as the Dietary Exposure Evaluation Model (DEEM-FCID; US EPA, 2023b) or International Estimated Dietary Intake (IEDI) tool3 .In North America, estimated drinking water concentrations (EDWC) are calculated for all cropping scenarios using the standard assessment tool Pesticides in Water Calculator (PWC; US EPA, 2023b) that uses scenario-and AI-specific parameters.The aggregate dietary (food and drinking water) exposure is then estimated using DEEM-FCID.In other regions, drinking water estimates are not aggregated, and regulatory determinations are based on cut-off criteria (e.g., in the EU the predicted drinking water concentration must be below 0.1 µg/L).The occupational exposures are proportional to the amount of AI handled by the pesticide handlers.This can be estimated based on crop use scenarios and the proposed wellunderstood application methods for these uses combined with the anticipated use rates.
Environmental exposures for ecological risk assessment will be informed by AI-specific environmental fate and metabolism data, which can parameterize existing models such as the PWC and the Terrestrial Residue Exposure model (T-REX; US EPA, 2023b).
At an early stage of product development, there is a need to establish the weight of evidence to support the use of alternate risk assessment methods.Although the final GAP for the new AI may not be available, a range of possible use rates and application times is known.The AI-specific metabolism, crop residue, et al., 2000), and tools such as the Ecotoxicity Calculator can facilitate the generation of risk curves (Dreier et al., 2021) that can provide risk estimates for multiple use scenarios.

Submission and registration
There are currently numerous steps between the development of the registration data package and the final registration of a new AI and associated pesticide products.So, in addition to the need for a robust and fit-for-purpose technical solution that meets agency need for decision-making, there may also need to be an evolution in various regulatory and administrative aspects that enable that decision-making to use a new approach.
The US EPA processes can be used as an illustrative example, whilst acknowledging that other agencies have different administrative processes and will therefore need to develop additional solutions to meet their specific needs.The US EPA has established guidance that allows acceptance of alternative approaches or sufficient information in lieu of a guideline study, enabling a study to be waived.Currently, these waivers have been developed on a per study basis.However, there are approximately 40 vertebrate guideline studies that are required, or conditionally required.Depending on the expected use and exposure scenarios, our comparative assessment-based approach would likely reuse the same information for each study waiver.To avoid this repetition introducing undesirable inefficiencies, our proposed dossier to fulfill the needs of an US EPA registration should consist of all study waivers in one document with citations to the journal articles or specific data held by the agency to facilitate linking the studies used in the readacross to support the technical argument.
Also, in the United States there are Pesticide Registration Improvement Extension Act (PRIA) requirements associated with a conditional ruling on pre-application of both waivers and protocols.The front-end and 90-day review process could remain the

Human health risk assessment
The human health risk assessment will be formed from a comparative assessment from across the registered ACCases class.After compiling these data, a human health risk assessment will be performed without any chemical-specific animal studies utilizing the RISK21 framework (Embry et al., 2014).The range of hazard will be defined as the range of endpoints for existing chemicals within this "data-rich" MOA class, with the expectation that the endpoints of the new AI will lie within these bounds based on multiple alternative lines of evidence justifying the read-across.Exposure estimates for the new AI will be defined as described above and compared to the hazard endpoints to complete this predictive human health risk assessment.

Ecotoxicology risk assessment
As with human safety, the distribution of endpoints from a comparator group will provide an estimate of hazard for a new member of a pesticide class.For ecotoxicology these endpoints can be fitted to a chemical toxicity distribution (CTD), which is similar to a species sensitivity distribution (Dreier et al., 2015).A conservative estimate can be drawn from this distribution (e.g., an HC5 value), establishing an ecological threshold of toxicological concern for the class, (ecoTTC, Belanger et al., 2015).Such an approach is common for substances that commonly lack toxicity data, such as industrial chemicals (Williams et al., 2011), cosmetics (Kroes et al., 2007), and food additives (Kroes and Kozianowski, 2002), but this approach has also been used to estimate hazard for data-rich chemicals, such as pesticides (Rizzi et al., 2021).Although HC5 values will be used as a surrogate endpoint for deterministic risk assessment, probabilistic risk assessments can also be performed.An effect distribution, represented as a CTD, can be integrated with an exposure distribution in a risk curve to characterize risk.Such an approach is relevant for pesticides (Solomon

Conclusions and next steps
In response to the US EPA Administrator's directive on moving the agency away from reliance on vertebrate tests whilst maintaining high-quality risk assessments, a problem statement was created.Exploring this problem led to the development of a conceptual model for future activities that, when finished, will solve the problem.The solution to the problem will be to provide the knowledge needed for a regulatory agency to confidently complete human health and environmental protective risk assessments.The conceptual model involves the creation and parallel submission of a new approach that will provide sufficient confidence to perform these risk assessments without reliance on some, or all, of the chemical-specific vertebrate tests.This alternative will be assessed in parallel to a traditional data package to determine whether it also supports some, or all, of the risk management actions necessary for human and environmental safety.If it does, then it can serve as an exemplar providing the technical basis of future stakeholder engagement to catalyze a transition to the new approach.If it fails, in part or entirely, then an analysis of the deficiencies will provide valuable feedback to focus development of suitable tools and additional approaches that will be required for subsequent success.same, which would fulfill a well-established role of pre-consultation meetings to address questions prior to submission.Similar process steps may exist in other regulatory agencies that will have to be adapted to fit the proposed new approach.
We anticipate that when using a comparative assessment, the same information will need to be reformatted to fit the different regulatory agency's needs.This format can be reused to make submission with each additional new AI from the same MOA class.Therefore, it will be useful to make handling of comparative assessments efficient and publicly available to meet multiple agencies' needs and yet scalable to subsequent new AIs.Careful creation, curation, and storage of appropriately modeled structured content that comprises the comparative assessment part of each MOA group's IATA should enable these data to be used and reused in multiple submissions.Mechanisms to ensure this content is updated as new information is generated in the transition period, yet remains accessible and reusable for multiple potential registrants of new Ais, will have to be developed and should be an active discussion topic during wider stakeholder engagement.

Stakeholder engagement
This NAM-based submission will not only need to be acceptable to multiple national agencies but also to all stakeholders including the general public due to the multiple technical, regulatory, policy, social, and political implications of any alteration in how food and environmental safety is ensured.Various diverse approaches will need to be used to achieve this.Initially, wider regulatory agency feedback beyond the parallel submissions with the pioneering agencies will require a suitable vehicle to explore the case studies and draw appropriate lessons; to transition effectively this will ideally move beyond one-to-one conversations.One possible method would be the adaptation of the MOA group-based comparative assessment approach into OECD IATA case studies.Also, supranational regulatory agencies such as WHO/JMPR have often requested NAM-based approaches.A discussion of these case studies at JMPR could enable wider regulatory visibility and discussion on their suitability for agrochemical assessment.To ensure consistency between registrants, other companies developing new AIs will also need to be engaged and encouraged to explore and adopt these approaches.The recently initiated Transforming the Assessment of Agrochemicals HESI project (Wolf et al., 2022) might be a useful venue to further develop the approach outlined herein.Moving beyond considerations of technical and regulatory suitability, the adoption of the approach is likely to require some changes in policy.Therefore, there will need to be wider participation to gain public confidence and trust and ensure appropriate support from policy-makers.Submission of both a conventional vertebrate animal test guideline dossier and NAM-based dossier will assist in establishing both technical confidence in the approach and gaining support and acceptance from the public.An important outcome of this effort will be a detailed evaluation of the numbers and species of animals avoided while establishing safe uses, protecting human health and the environment, and improving sustainable agricultural productivity.

Tab. 1 :Fig
Fig.1: The studies used to provide information for risk assessment

Fig
Fig. 3: Data processing needed to support a new modernized approach