Key challenges and developments in wildlife ecological risk assessment: Problem formulation

Abstract

Problem formulation (PF) is a critical initial step in planning risk assessments for chemical exposures to wildlife, used either explicitly or implicitly in various jurisdictions to include registration of new pesticides, evaluation of new and existing chemicals released to the environment, and characterization of impact when chemical releases have occurred. Despite improvements in our understanding of the environment, ecology, and biological sciences, few risk assessments have used this information to enhance their value and predictive capabilities. In addition to advances in organism-level mechanisms and methods, there have been substantive developments that focus on population- and systems-level processes. Although most of the advances have been recognized as being state-of-the-science for two decades or more, there is scant evidence that they have been incorporated into wildlife risk assessment or risk assessment in general. In this article, we identify opportunities to consider elevating the relevance of wildlife risk assessments by focusing on elements of the PF stage of risk assessment, especially in the construction of conceptual models and selection of assessment endpoints that target population- and system-level endpoints. Doing so will remain consistent with four established steps of existing guidance: (1) establish clear protection goals early in the process; (2) consider how data collection using new methods will affect decisions, given all possibilities, and develop a decision plan a priori; (3) engage all relevant stakeholders in creating a robust, holistic conceptual model that incorporates plausible stressors that could affect the targets defined in the protection goals; and (4) embrace the need for iteration throughout the PF steps (recognizing that multiple passes may be required before agreeing on a feasible plan for the rest of the risk assessment).

INTRODUCTION

Wildlife risk assessments are conducted to predict impacts on populations of wildlife as a result of being exposed to environmental stressors. Many regulations focus on anthropogenic chemicals (either synthetic or manipulated naturally occurring substances). Increasingly, biotic and physical stressors are being considered in wildlife risk assessments, for example, in response to zoonotic diseases, invasive species, or changes in habitat quality resulting from land use practices and climate change (Gust et al., 2018; McFarland et al., 2012). For this article, we define wildlife as air-breathing vertebrates and primarily restrict this to animals residing mostly in terrestrial settings. Because of many cultural factors, criteria used in decision-making for wildlife are different from those for other organisms or in other environments. For example, migratory birds are managed on a continental, even global basis. In addition, risks to large charismatic mammals are often managed differently than are those for small mammals. Wildlife species can receive special protections in parks and wildlife refuges. Management of populations of waterfowl, ungulates, and other mammals (e.g., mink, raccoon, squirrel, beaver, etc.) considers harvesting from hunting activities. Not only is take considered, but they should also be safe for human consumption, adding human health, welfare, and cultural components to the decision-making.

Risk assessment for wildlife may entail several layers of decision-making. For example, the spatial extent of a risk assessment may cross local, regional, state, tribal, or international boundaries and be subject to various treaties. A federal, state, or provincial government entity might manage the land as an owner, trustee, or a regulator. The primary goal of a risk assessment is to inform decision-making as determined by applicable regulations, community interests, or other interests.

Problem formulation (PF) is a critical initial step in planning risk assessments for wildlife. This, from a science-based perspective, should be done in the context of the relative impact of other stressors that wildlife populations typically experience (e.g., limiting food resources, weather and/or climate impacts, habitat loss, predation pressures, zoonoses and other transmissible diseases, parasites, mating habits, etc.). The value of the risk assessment depends on how well it meets the goals and specifications defined through the PF process and how it facilitates subsequent decision-making. Properly scoping the problem is critical to success in any assessment of risk (McPartland et al., 2022; Sauve-Ciencewicki et al., 2019). Exposures to and effects on wildlife species depend on the manner of exposure to the stressor, for example, the environmental chemical release, chemical/physical properties of chemicals, biological uptake and bioaccumulation potential, and the life history of the species of concern, and the complexity of biotic and physical interactions affecting the ecological system.

A particular challenge of PF in risk assessment is to integrate the specific rules and values applied to terrestrial organisms and their habitats (as defined by the regulatory or other drivers for the risk assessment) into physical, biological, ecological, and possibly evolutionary reality. Since the 1990s, our view and knowledge of ecological structures in general, and terrestrial systems and organisms specifically, have changed dramatically. The hierarchical patch dynamics paradigm first presented by Wu and Loucks (1995) captures many of these features. These include:

• Ecological structures are complex, meaning that there are numerous interactions and that these interactions store information regarding the events in the etiology of the system. Some of this information exists in the genetics of the organisms, life history and phenotypic plasticity, spatial interactions, and in specifics of the population dynamics and structure of the landscape.

• These systems are dynamic.

• The interactions can be governed by processes that are best described as combinations of deterministic and stochastic processes.

• Numerous hierarchies have interactions that are dynamic, with connections that vary in time and space.

• Spatial interactions within a landscape are key to the dynamics and the impacts of disturbances.

Risk assessment and decision-making tools for wildlife require a paradigm, methods, and data that describe these systems. The tools to accomplish this integration exist but are yet to be adequately incorporated into PF, and thus remain largely absent from wildlife ecological risk assessment (ERA) processes.

Like all ERAs, wildlife ERAs are often driven by regulations that address one of three basic management decision needs which can overlap: (1) prospective—assessing the incremental risk of a particular future anthropogenic stressor (e.g., new chemicals and/or products) in the environment, in which case all other stressor effects only set a baseline condition unless there are potential interactions between the future anthropogenic stressors and possibly others; (2) retrospective or following forensic efforts to identify likely causal effect of legacy stressors—informing decisions about how to mitigate the threats to wildlife populations; and 3) resource driven—assessing current and future threats from environmental chemical releases or other stressors that might affect wildlife as valued resources. However, the terms “prospective risk” and “retrospective risk” were offered initially to distinguish between cases involving legacy contamination and new products used for industrial or agricultural purposes. At least one author contends that all risk assessments are intended to be predictive, that is, they are forecasting a future event, and therefore we should dispense with the use of the constructs as they are superfluous and redundant (see Fairbrother et al., 1997). Ultimately, the wildlife ERA PF must respond to the identified management need but should also be cognisant of and incorporate “bigger picture” ecological context (to the extent possible) to recognize, account for, and to the extent possible, minimize unexpected ecological and human welfare impacts.

Although some regulations that apply to wildlife ERA have been updated recently (e.g., EFSA Risk Assessment Guidance for Birds and Mammals), the methods used to evaluate risks to wildlife species have not changed markedly in decades. Meanwhile, there have been substantive advances in computational capacity, systems ecology, and analytical techniques over the past two decades that provide ways to improve the relevance and accuracy of wildlife risk assessments. These include processing big data as a way to identify response patterns that can induce interactions among co-occurring stressors. Modeling capabilities and computational processing capacity have expanded on several fronts that, if anticipated during PF, would surely alter the type and quantity of data gathered. Bayesian networks can be used to infer causality that was not achievable heretofore. Systems ecology, coupled with computational capabilities, allows deeper understanding of food web interactions that are vital to characterizing indirect effects, which are critically important when ecosystem services are explicitly explored as assessment endpoints. The advancement of eDNA and miRNA characterization has greatly improved the ability to evaluate ecosystem structure and function. Finally, these advances need not be treated as stand-alone techniques, but rather can be mixed and matched to yield robust assessments. In PF, it is sufficient to be aware that valuable data can be assembled using these emerging methods and approaches. Additional discussion of some of these methods appears in companion papers (e.g., Bean et al., 2022; Rattner et al., forthcoming; Morrissey et al., forthcoming).

Much of what appears in risk assessments continues to derive from organismal-level endpoints. The work on adverse outcome pathways (AOPs) focuses on understanding cascades of mechanisms of action at suborganismal and organismal levels and theoretically explores linkages to population dynamics. However, absent empirical, project-specific measurements of populations, what is measured often remains disconnected with the stated protection goals. That is not to discount the value of AOP analysis, but rather to recognize the inherent limitations of the method and inform strategic transition of useful organism-centric AOP output toward incorporation into higher scale ecological assessment methods and/or models.

Challenges posed by having complex mixtures (i.e., several chemicals) and multiple stressors (i.e., biological, chemical, and physical parameters) have been recognized from the outset of risk assessment. A general rule of modeling that redounds to risk assessment is to reduce the problem to its simplest level. For regulatory purposes that has meant consideration of a single chemical and a single organism. However, when applied to real-world conditions, this level of simplification often fails. And it fails in both directions, that is, it can overestimate risk in some situations and underestimate risk in others. This is because in the real world, system dynamics expressed through multiple nested feedback loops reveal consequential indirect effects. Such responses can only be anticipated by examining system interactions that occur at higher levels than organisms. Therefore, the risk assessment focus must be tailored in the PF stage so that it aligns with the magnitude of the decisions to be made and the inherent complexity of the social–ecological system in question.

Although spatially explicit models that account for complex behaviors and spatial movements were published more than a dozen years ago, they have not received widespread acceptance. Spatially explicit models seek to incorporate life histories of target species and consider both direct and indirect effects (e.g., migration and recruitment patterns, habitat use changes initiated from seasonal niche patterns, and heterogeneous distribution of stressors including chemicals) to better inform wildlife ERA (Hope et al., 2011; Johnson et al., 2009; Pistocchi, 2008; Wickwire et al., 2011; also see Morrissey et al., forthcoming). Evaluating opportunities to incorporate scientific advancements into wildlife ERA should be an ongoing effort, and the present effort represents a necessary attempt to advance the process with a lens focused on the PF stage of wildlife ERA.

The PF provides the opportunity to focus efforts, gain stakeholder input, and plan a path forward for risk assessment and risk-management decision-making. The PF stage of a risk assessment is analogous to a blueprint for constructing a building. Some blueprints are simple and readily transferable from location to location, such as the basic structure of a pole barn. The function is relatively simple, reusable, and readily transferred to a new location. In contrast, a high-rise office building necessarily requires more detailed blueprint development with exquisite detail that reflects specific site conditions. In such a case, blueprint development will need to occur de novo. Extrapolating this blueprint-development analogy to wildlife ERA, we can see that PF can be straightforward and readily standardized in simple, commonly encountered, and well-validated applications. However, there is a danger in applying one-size-fits- all PF approaches to complex wildlife ERA scenarios. Just as a pole barn blueprint is insufficient to produce a functioning office building, overly simplistic PF processes are not capable of incorporating the necessary details and tackling the inherent complexities of higher order ecological scenarios and the ultimate decisions that affect wildlife and stakeholders involved when applied to complex wildlife ERAs. Risk assessors and risk managers often implement a tiered or phased approach to refine an initial PF blueprint (USEPA, 1997, 1998). As depicted in Figure 1, an iterative process is applied to allow for increasingly complex analysis or testing of hypotheses that serve as a stepwise approach to reduce uncertainty and focus the risk assessment. Emerging science and tools should be considered in selecting measurement endpoints that reflect the agreed complexity of the conceptual model.

FIGURE 1.

Problem formulation (PF) development includes efforts to (1) frame the problem, (2) explore the problem to develop a conceptual site model, and (3) map the approach to how the wildlife ecological risk assessment (ERA) should be addressed. Each of these three components of the PF may be revisited (i.e., refined) as more knowledge is gained and from stakeholder input

Expanding on the analogy of the PF as a blueprint for construction, Figure 2 illustrates the iterative PF framework. This framework begins with considerations that frame the problem, leading to a decision to proceed to a wildlife ERA or to determine that a qualitative decision and/or solution is available or appropriate (i.e., exit the ERA process). When the decision is made to proceed with the wildlife ERA, the PF efforts focus on considerations that explore the problem and develop the conceptual model, leading to the decision to perform a qualitative screening or a quantitative, detailed wildlife ERA. Regardless of whether the path forward is screening or detailed, the PF must map the approach that will be used. Specifically, this stage of the iterative PF identifies the exposure and toxicological approaches that will address the conceptual model, including clearly defining what constitutes an unacceptable level of impact considering an acceptable level of uncertainty, define the analysis plan and strategy, and identify the uncertainties that will exist even after the approach is implemented. At this point of the iterative PF development, if stakeholders agree, then the wildlife ERA and data collection can begin. If stakeholders do not agree, then refinement of the problem is appropriate. Once the wildlife ERA begins, several key questions can lead to the conclusion of the ERA or these can lead to further refinement of the problem or the approach, as illustrated in Figure 2.

FIGURE 2.

The iterative problem formulation framework encourages decision-making to (1) frame the problem, (2) explore the problem to develop a conceptual site model, and (3) map the approach to how the wildlife ecological risk assessment (ERA) should be addressed. At each decision point, the concepts may be refined to achieve stakeholder agreement before moving forward toward a solution. ^aRe-explore the problem and conceptual model or remap the approach, as appropriate. ^bOn some occasions (albeit rare), it may be appropriate to reframe the problem

Terrestrial and aquatic ecological systems are often intricately connected and intertwined (cf, Munns, 2006; USEPA, 2004). Our definition of wildlife in this article derives largely from our understanding of the perceived needs to enhance approaches to terrestrial wildlife ERA per se (van den Brink et al., 2022). Advances in our understanding of functional attributes of aquatic ecosystems (e.g., importance of submerged aquatic vegetation in providing structural and functional nurseries for young fish) can inform terrestrial ones. Because all ecological systems are open and connected, there can be substantive communication between terrestrial and aquatic systems through the exchange of nutrients (predator/prey) and contaminants. For example, the life history of salmon links marine, freshwater, and terrestrial communities. Salmon acquire most of their biomass in oceanic waters. There, they also accumulate contaminants such as PCBs, some legacy pesticides, and mercury. When the salmon return to freshwaters to spawn, they become prey to eagles, bears, and other wildlife. Salmon carcasses are food for insects, spiders, birds, and bats (Nakano & Murakami, 2001; Richardson et al., 2010). Dispersal of these predators and scavengers from streamside into uplands can result in deposition of nutrients and contaminants far into terrestrial habitats (Bilby et al., 2003; Morrissey et al., 2011). We therefore draw at times from our understanding of aquatic ecosystems to inform terrestrial wildlife ERA during PF, yet recognize differences in the types of data collected from both field and controlled laboratory investigations.

COMMON CHALLENGES IN WILDLIFE ERA PF AND DEVELOPMENTS OF POTENTIAL VALUE

The practice of risk assessment differs across jurisdictions, classes of chemicals, and types of additional stressors, as do the types of decisions that are required. For example, at legacy-contaminated sites, PF requires focused attention on the setting, the context, the extent and complexity of contaminant mixtures present, and the protection goals as articulated by affected stakeholders. For other types of risk assessment, such as for pesticide registration, in addition to focusing on only a single contaminant, there are agreed a priori protocols to follow—in other words, the equivalent of PF has been negotiated by government and industry in advance of any application.

Extensive guidance and policies exist for conducting wildlife ERA that are application- and jurisdiction-specific (see van den Brink et al., 2022). In addition, reports of new information and methods continue to emerge, filling data gaps and providing novel methods, approaches, and techniques that can be applied to wildlife ERA to increase accuracy and ecological relevance and link more directly to decision-making. If these improved methods and data are to be integrated into wildlife ERAs, they must be introduced in the PF so that the appropriate data can be acquired, in the correct locations and at the correct times, or to facilitate the identification and integration of the models that most appropriately relate to the risk question/decision being addressed. Some approaches that are not themselves “new science” can be applied in a way that is markedly different from current practice, and thus offer improvements to how wildlife ERA is conducted. Multiple challenges to defining and implementing PF for wildlife ERA (and ERA in general) were identified, along with potential developments to address these challenges. These are described below.

Challenge 1—Clearly identifying linkage between risk conclusions and risk-management goals

Decisions made by risk managers are implemented to achieve risk-management goals (USEPA, 1998). The characteristics of these goals define the scope, focus, and conduct of the risk assessment. Example characteristics include, but are not limited to, spatial extent (site-specific vs. national or international in scale); the specific stressors considered (i.e., one vs. many); nature of biological receptor (single species vs. multiple species/taxonomic groups); level of biological organization (individual, population, community, ecosystem); nature of decision (site remediation vs. permissible environmental releases); and so forth. Although risk-management goals are often generalized statements (USEPA, 1998), it is essential to establish clear definitions and linkages between the characteristics underlying the goals and decision criteria that will lead to management actions. It is important to also recognize detrimental impact for some management actions.

Protection goals must be balanced with understanding the uncertainties and associated variation in predicting risk. Initially, assessments are designed to minimize the chance of underestimating the potential for impacts. Ultimately, the risk assessment should characterize risks in a manner that allows managers to understand the uncertainties and still be confident in meeting their protection goals. Clearly identifying the protection management goals ensures they are linked to possible management actions, while acknowledging any constraints.

Development—Consider ecosystem services when defining management goals.

Management goals should not only consider ecological effects on wildlife, but also how wildlife risk influences the welfare and well-being of people. This is the concept of ecosystem services (Munns et al., 2015; cf, Selck et al., 2016). We encourage consideration of stakeholder values and concerns when formulating wildlife ERAs. Too often in the past, wildlife ERAs have been planned through the narrow lens of regulatory mandates, with limited regard for those things about which people care (Munns & Rae, 2015). The PF and conceptual model development, and ultimately environmental decisions, would benefit from an expanded view from strict regulatory requirements to consider ecosystem services more explicitly.

Development—Comprehensive conceptual models.

Ecosystems are inherently complex, with many interactions and with many unknowns that may be community specific, such that it is necessary to use simplifying assumptions. However, it is important to identify and, if possible, quantify how factors beyond the chemical(s) under investigation may be influencing wildlife risks, and therefore could influence management decisions. This is initially captured by first developing a comprehensive conceptual model that incorporates chemical exposure and the receptors within a multistressor environment that are the focus of the question (i.e., problem), but that also includes other factors that may influence management decisions. This comprehensive conceptual model allows the inclusion of molecular biology, AOPs, systems and population modeling, multiple stressors, community ecology, use of field observations, potential causal links, and evaluation of management alternatives. Development of this conceptual model would benefit from consultation with a multidisciplinary team of experts to better understand how these analyses can contribute to understanding the decision to be made, as well as their associated levels of uncertainty.

Challenge 2—Collecting enough of the right data to answer the question

A key focus of the wildlife ERA PF is to define the “correct” amount of the “right” data to conduct the necessary analyses to answer the question under investigation with an acceptable level of uncertainty. This includes sufficient data of sufficient quality to quantify the causal link between exposure to chemicals and potential for adverse effects to support the management decision. Regardless of the wildlife receptor of interest, wildlife ERAs require the use of multiple assumptions to provide the means for a decision. Screening approaches that rely extensively on conservative assumptions are typically employed in the initial stages. These approaches serve to focus the assessment especially when time and money are limited and should be viewed as one-tailed tests. Some emerging new approach methods (NAMs) require a sound understanding of their reliability to get consistent results before application. They also require an understanding of false-positive and false-negative results (Type I and Type II errors, respectively). By definition, effective screening assays require a demonstrated high false-positive rate and a relatively low false-negative rate to be useful.

Because of the conservative assumptions, confidence should be high that risks are absent or trivial for exposures to chemicals that pass the screen. However, failing the screen typically does not indicate unacceptable risks; rather it simply suggests that we lack sufficient data to make a risk-based decision. Determination of a confident risk-based decision often requires collection of additional data or information. Further, assumptions and management goals may differ based on spatial and temporal scales. Migratory birds may experience exposures in a pulsed acute or transient manner, and their exposome broadens to include wintering, breeding, and transient migratory stopover exposures.

Development—Establishing data quality objectives, quality assurance and quality control protocols, quantitative model documentation in a detailed analysis plan that considers new science and analysis techniques.

The data quality objectives (DQO) process (USEPA, 1994, 2006) represents a tool that can be used to clearly identify the linkages between the question to be addressed, the decision to be made, and the amount and quality of data needed. The DQO process should also consider all possible outcomes from data collections and decide a priori what decisions could be made given those ranges of outcomes. This is also an opportunity to identify the uncertainties that will exist at the end of a wildlife ERA so stakeholders can agree if those uncertainties will be acceptable for decision-making. Frameworks like AOPs may also play a role in identifying data needs that may support the decision (i.e., providing evidence of interspecies extrapolation and possibly provide empirical evidence to justify the magnitude of assessment or uncertainty factors, etc.). In particular, if the goal is to parse out the effects of a single chemical in a mixture and multiple-stressor environment, frameworks such as AOPs can define the data that are necessary to detect the chemical signature across several levels of biological organization. Endpoints such as growth or reproduction are not sensitive enough to detect the toxicity pathway perturbed by a single chemical in a mixture (also the case if a single chemical has multiple modes of action). To the extent possible, data acquisition should consider evidence to support a determination of causation (see Pearl, 2019).

Additional data can focus on information that provides more exact exposure estimates to better characterize spatial extent of the substance relating to exposure, toxicity to the species of interest, or ecological criteria that may be relevant. Assessments beyond screening assessments should consider data collection beyond chemical concentrations in abiotic media. For example, uncertainty in exposure modeling could be reduced by measuring chemical concentrations in wildlife dietary items or by measuring bio-accessibility of metals in soil or diet items. Regardless, a plan is needed to first identify these assumptions, rank them based on their relative influence on the risk estimate, and develop a plan to further collect information to refine these criteria. Before any additional information is collected, the required precision and accuracy of the data should be determined, how the data are to be collected should be described, and the range of decisions that could be made given the possible range of outcomes should be considered. This process is widely known as the DQO process (USEPA, 2006). In addition, if models are to be developed and applied, there should be quantitative model documentation (e.g., transparent and comprehensive model evaluation [TRACE—https://cream-itn.eu/trace]).

Developments provided as part of the following challenges provide more specific examples of advances in science that must be considered as part of any data analysis plan.

Challenge 3—Understanding population-level effects

Management of risks to wildlife is typically suggested at the population level (Barnthouse et al., 2008; USEPA, 1999, 2016). Toxicity data intended for mammalian wildlife are typically extensive and include health parameters consistent with those used to assess human health (e.g., clinical chemistries, histopathology, indicators of immune suppression, etc.). Extrapolation of effects from lower to higher levels of organization is often required to support management decisions (Bean et al., 2022). Often, risks are calculated based on the assumption that organismal-level effects (e.g., reproduction, growth, and survival) are directly equivalent to population-level risks (Rohr et al., 2016; USEPA, 1997, 2016); however, there are few field data that support this assumption. Further, chemical impacts on endpoints such as growth and reproduction are only discernible in a field setting if the chemical has a single mode of action and is occurring in isolation of other stressors. However, most chemicals have multiple modes of action and often occur in mixtures, and so energetic trade-offs between growth and reproduction and survival are difficult to quantify unless the toxicity pathway and physiological mode of action are identified (Ananthasubramaniam et al., 2015; Murphy et al., 2018). Rattner et al. (forthcoming) suggest behavioral endpoints may be of greater ecological importance.

Our understanding of metapopulation dynamics has improved for many species and useful models have been developed (Barnthouse et al., 2008; Hanski, 1994; Raimondo et al., 2021; Wiens, 1997).

Sample et al. (2000) tested the relative importance of mortality and reproduction using individual-based population models for generalized r- and K-selected bird species exposed to a contaminant that affected reproduction and survival. They found that, to prevent population reductions of 20%, adult survival should not be reduced more than 5% for either the r- or K-species, and age-zero survival and fecundity should not be reduced more than 5% for the r-species. These simulations suggest that basing population-level risk-management decisions on individual-level end-points may underestimate population responses. This issue has been supported by analyses by Rowe (2008) and Kamo and Hayshi (2011).

Development—Improved understanding of the implications of organismal effects on populations.

Ideally, long-term monitoring data for target populations would be available that clearly reflect the combined effects of all stressors that could be relevant in understanding the relative impact of the introduced chemical stressor. However, in reality, even when such long-term data exist (e.g., Barclay et al., 2011; Conner et al., 2013; Getz et al., 2001; Krebs et al., 2019), the results often highlight the variability of populations and the challenges of understanding the key drivers behind the observed dynamics. Consequently, considering the amount of time and resources required to collect these datasets, it is unlikely that this type of information will be widely used directly in many wildlife ERAs.

As an alternative to direct population monitoring, population modeling provides a means to explore the potential implications of exposure to contaminants and other stressors on target populations (when known). Comparative assessments of the implications of contaminant exposure to populations for species with different life-history characteristics have demonstrated that the one-size-fits-all approach often used for setting acceptable effect levels is flawed (e.g., Gleason & Nacci, 2001; Sample et al., 2000; Stark et al., 2004). Simply stated, the susceptibility to similar levels of reproductive impairment would likely be quite different for species that exhibit high neonatal predation rates relative to those that do not (e.g., mouse vs. moose populations).

Use of population models in ERA is not new (e.g., Barnthouse et al., 2008; Pastorok et al., 2001). Although they are more common in certain applications (e.g., pesticide risk assessment; Etterson et al., 2021), they continue to be underutilized more broadly in ERA because of a range of challenges (Raimondo et al., 2021). However, research efforts continue to advance these tools (e.g., Ananthasubramaniam et al., 2015; Martin et al., 2013; Murphy et al., 2018, and many others) and improve guidance on their use to support environmental management decisions (Raimondo et al., 2021; Schuwirth et al., 2019).

Given the effort required to use these tools, their direct use in wildlife ERA is predominant in cases with limited contaminants and receptors (e.g., one each). This could change if researchers explored the susceptibility of wildlife species to contaminant-induced effects based on their life-history characteristics and considered the integration of toxicity modes of action into a whole organism response that may influence population-level effects. Application of a consistent approach and consideration of other stressors could provide a tool to help set acceptable effect levels for wildlife species.

Development—Interspecies extrapolation of organism-level effects to population-level analyses.

As discussed, the biological level of organization of protection is generally considered the population or metapopulation (Barnthouse et al., 2008; USEPA, 1999, 2016). However, maintaining metapopulation densities requires an understanding of current stressors and factors that limit them (e.g., available nesting sites, community predator compositions, availability and quality of food resources). Each metapopulation may have a different set of stressors that function to regulate them, and rarely are they known a priori. It is also important to understand the influence of immigration and emigration rates particularly if field monitoring efforts are considered. This must occur in addition to understanding the exposure pathways. Integration of the additional stressor of exposure to chemicals must be considered in a manner relative to other metapopulation-regulating mechanisms. This may require the use of assumptions to understand the relative influence of chemical exposure.

Additionally, the influence of the toxic effect should be considered in an ecological context, however, with caution (see Porter et al., 1984). Small reductions in Northern Bob-white egg production in a controlled laboratory situation where photoperiod is constant may mean little in a field situation where nest egg predation is high. Ecological theory suggests that energy is limiting (e.g., food) and that somatic maintenance could conceivably divert limited energy from reproduction to repair of the toxic insult, resulting in lower reproductive output (Nisbet et al., 2000). Increases in lethargy from toxicant exposure in the laboratory could translate to reduction in predator vigilance during breeding or a reduction in mating success (Fairbrother et al., 1988). Therefore, consideration of toxic endpoints relative to the ecology of the species and the communities in which they reside is important.

Improving our understanding of seasonal ecology and how individuals shift in how they experience the environment can improve our estimates for exposure. Additionally, considerations of exposure relative to season and contact in time with receptors (e.g., pesticide application) are also relevant. Use of spatially explicit models that consider movements of individuals that are governed by habitat preferences and life-history attributes (e.g., nest provisioning requirements) can improve exposure estimates for prospective (e.g., pesticide use) and retrospective risk assessments (e.g., contaminated sites; Morrissey et al., forthcoming). Seasonal differences in niche partitioning, diet shifts, and social interactions of wildlife species can affect exposure where acute (e.g., short-term herding or flocking and roosting behavior), subchronic and chronic exposures should all be considered. This aspect is considerably complicated for evaluating chemicals that bioaccumulate in species that are long-range migrants and may experience exposures from distant locations.

Often responses to stressors, particularly those that affect growth and reproduction, will be modified by life-history constraints for particular species, and so life-history models could be included to predict effects of contaminants across different species or between populations of the same species.

Challenge 4—Improving exposure estimation

Wildlife experience their environment (and therefore are exposed to contaminants therein) in a multidimensional context. Understanding exposure of wildlife to anthropogenic chemicals in the environment is intricately tied to a sound understanding of species-specific life-history attributes. Their longevity and mobility mean that exposure can vary over time and space. Attributes of the environment, food, habitat, contaminants, and the interactions among these attributes are also variable. Whereas screening-level assessments employ simplifying assumptions to conservatively represent this complexity, improving accuracy in exposure estimation for higher-tier assessments requires understanding and integrating this complexity. Higher tiers of assessment allow the incorporation of more site-specific and species-specific information.

There are significant gaps, however, in our understanding of many attributes that feed into estimating exposure to wildlife. These include species-specific soil and sediment ingestion rates, bioavailability and/or bioaccessibility of contaminants in food and abiotic media, food and water ingestion rates, seasonal niche shifts in food choices and preferences, and changes in habitat preferences. Fate and transport of chemicals through food webs must be considered. Additionally, dermal exposures are rarely considered for lack of information (e.g., dust-bathing birds). Many others are discussed in Morrissey et al. (forthcoming).

If these improved methods and data are to be integrated into wildlife ERAs, they must be introduced in the PF so that the correct data can be acquired in the correct locations and at the correct times, or to facilitate the identification and integration of the models that most appropriately relate to the risk question or decision being addressed.

Development—Improvements in spatially explicit/habitat-focused models and ecological understanding.

Wildlife experience the environment in ways consistent with life-history attributes that are guided by habitat requirements (Morrissey et al., forthcoming). These can vary seasonally, where species can become gregarious only during months of the year (e.g., red-winged blackbirds, white-tailed deer). The same species can shift in diet preferences based on seasonal availability of food and energetic demands. Diet shifts and life-history attributes can affect indirect soil consumption, which can add significantly to exposure. Newer, individual-based models have been developed that can account for seasonal differences in diet shifts, habitat preferences, and other life-history attributes. These models can display variation in metapopulation exposures (Johnson et al., 2007), but may not add value for small mammals with limited movement (Johnson et al., 2020); further work is needed on additional species with relatively larger home ranges (see Morrissey et al., forthcoming). These issues must also be considered in the PF exercise before any such model is used.

These models typically provide individual daily estimates of exposure for a metapopulation based on species-specific criteria and seasonality. Exposure probability is governed by habitat suitability and other life-history constraints. These models can also facilitate “what if” questions where virtual remediation can occur and users can model changes in exposure based on these spatial remediation choices.

Development—Improvements in understanding bioavailability/bioaccessibility in various matrices.

Chemical physical properties (e.g., water solubility, fat solubility, affinity to organic carbon, etc.) can inform where in the environment exposures could occur through fate and transport properties. Additionally, parameters are required to estimate exposure through food, water, and inadvertently through soil and sediment (Morrissey et al., forthcoming). Improvements in understanding physiological differences between species may help in improving exposure estimates, particularly for oral exposures. Our understanding of gut physiologies has improved for factors that influence systemic absorption such as gastrointestinal pH, gut retention times, and differences between monogastric and those with alternative physiologies (e.g., ruminants, hindgut fermenting species). The development of technologies to quantify exposure through the collection of biosamples (e.g., plasma, hair, feather, or other tissue-based criteria) adds significantly greater precision when the value of such measurements can be validated and verified.

Development—Improvements in understanding species-specific life-history attributes and influences on exposures.

As mentioned, many species vary in their seasonal, spatial, and habitat-specific food preferences. Much has been learned recently regarding considering the whole exposure or exposome approach (Morrissey et al., forthcoming) where life histories are tied to multiple exposure events to include various substances (i.e., mixtures). Information is continually developed that better informs species-specific parameters important in calculating and estimating exposure concentrations. Consideration of these approaches and incorporation of these traditional and nontraditional approaches is required during PF.

Challenge 5—Incomplete use of toxicity information to ascertain the threshold for adverse effects

Defining the level of protection and/or prediction is critical to understanding the potential for adverse effects and a necessary step in scoping the problem. The paradigm used in defining safe versus unsafe levels of exposure for wildlife typically involves using animal surrogates in controlled laboratory experiments where statistical differences between treatments have been established as a principal criterion for establishing thresholds for adverse effects. Here, no adverse observed effect levels and lowest observed adverse effects levels (NOAELs and LOAELs, respectively) were used for screening and higher tiered ERAs. These methods failed to provide thresholds and neglected much of the dose–response information important in understanding the magnitude of response should exposure exceed a toxicity reference value (TRV). This process was modeled after historical human health risk assessment, though methods to characterize the dose–response functions for both human health and for wildlife have improved.

Development—Use of benchmark dose and similar models to capture complete dose–response curves for toxicity data.

Current protocols are insufficiently designed to capture the threshold for adverse effects. Protocols used for new chemicals and pesticides require the use of three treatments and one control (e.g., OECD, 2018; USEPA, 1996). New dose–response models allow capturing the variation and providing optimal fit of dose–response data to determine the threshold for adverse effects (e.g., Bayesian benchmark dose models; Filipsson et al., 2003; Shao & Shapiro, 2018); however, these models provide more accurate fits when additional treatments are used (Wignall et al., 2014). New protocols are being suggested that integrate additional treatments but use fewer animals (see Bean et al., 2022). These improvements allow for a more robust fit of dose–response curves and calculation of the threshold for adverse response within user-defined levels of confidence.

Development—Use of AOPs to inform extrapolations of toxicity data.

Relatively few wildlife laboratory animal models are available to investigate dose–response thresholds in a controlled laboratory environment. Therefore, extrapolation of those responses to other species is critical. The use of shared biological pathways between laboratory and wildlife species is vital to interspecies extrapolation (Madden et al., 2014). The development of AOPs can provide empirical evidence to support extrapolation of effects between species (LaLone et al., 2016), when differences in kinetics are considered, and so should be incorporated into the PF so that appropriate data are collected during the study (Perkins et al., 2019). Therefore, if an AOP in a rodent laboratory model can be shown to be plausible in white-tailed deer, this would constitute evidence to support the scaled use of this threshold without the use of uncertainty factors. Accurate scaling would ultimately involve the use of physiologically based toxicokinetic models; however, we recognize that these are often labor intensive and data prohibitive.

Additionally, assessment of the toxicity of mixtures in the context of multiple-stressor scenarios (or just a single chemical with multiple modes of action) would also benefit from approaches such as the AOP framework (Rattner et al., forthcoming). This allows evaluating toxicity patterns across multiple levels of biological organization, thereby identifying a toxicity “signature” that could be attributed to a particular mixture and/or identify the toxic agents in a complex mixture for retrospective assessments. Establishing PF with an AOP structure in mind, with key events measured at several levels of biological organizations, would allow these mixture/multiple-stressor patterns to be detected. Accordingly, next steps for prospective and retrospective risk assessment would include developing integrative models that can scale levels of biological organization such as Bayesian networks, or dynamic energy budget models (e.g., Murphy et al., 2018). These data and modeling frameworks need to be built into the PF so that appropriate data are collected with data quality assurances. Methods for establishing data quality requirements for AOP framework development have been published (Collier et al., 2016) and can be readily applied to the PF phase of WERA.

Finally, initial screening for species susceptibility in the PF could make use of the Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS) tool (https://www.epa.gov/chemical-research/sequence-alignment-predict-across-species-susceptibility) that compares protein sequences across a wide variety of taxa and determines if the species of interest will have the appropriate molecular receptor for the chemical of interest (LaLone et al., 2016).

Development—Use of wildlife-specific cell lines and tools.

Using new data from wildlife species-specific cell lines coupled with the development of AOP analysis with laboratory toxicity information may provide the means in a weight-of-evidence approach to support data integration for species extrapolation (Madden et al., 2014; Rattner et al., forthcoming). The use of new, noninvasive field methods (environmental DNA, feather, fecal samples, etc.) where appropriate, provides additional information in a data integration approach to refine toxicity thresholds (i.e., toxicity reference values) by comparing tissue effect levels with modeled exposure estimates of adverse effect. The collection of data from these tools should be considered in the PF phase to include DQOs.

Challenge 6—Developing computational models that factor in indirect effects and multiple stressors to estimate exposure and effects

Wildlife are exposed to multiple chemicals, other stressors, and may be affected via indirect pathways, such as habitat degradation or prey scarcity. One example is the large-scale aerial application of two pesticides, diflubenzuron (DFB) and Bacillus thuringiensis kurstaki (Btk), for the control of the sponge moth (Lymantria dispar), an introduced, invasive forest pest responsible for tree defoliation. Both pesticides have virtually no direct vertebrate toxicity. However, application of these pesticides resulted in reduced abundance and diversity of Lepidoptera (Sample et al., 1993a, 1996; Whitmore et al., 1993), resulting in dietary shifts in songbirds, away from more nutritious and easily digested Lepidoptera larvae, to less digestible insect taxa (e.g., Homoptera, Diptera, and Coleoptera; Sample et al., 1993b). Reduced fat reserves, postulated to be caused by reduced food abundance, increased effort to obtain food, and/or reduced food quality, were observed in songbirds (Whitmore et al., 1993); similar effects have been observed for Caprimulgids (Wagner et al., 2021). The PF must be sufficiently robust to accommodate evaluation of direct and indirect effects in addition to multiple-stressor effects.

Development—Consideration of chemical-related effects relative to other stressors.

As emphasized by Burton (2017a,b) and Suter (2017), a focus on chemicals alone is not appropriate. Although their dialogue was related to aquatic habitats, it also applies to terrestrial habitats. During PF, the conceptual model would benefit from including factors that have significant potential to influence wildlife, including their habitat, so that data can be collected to address potential causal links between stressors and wildlife. This includes the potential influences of climate change (Landis et al., 2013). Burton and Suter cite several frameworks and guidance that exist for doing this; however, it is rarely done. Although most examples are for aquatic habitats, these frameworks and guidance (e.g., USEPA, 2000) can be applied to assist in the identification of nonchemical stressors and describe how these stressors influence wildlife. Although it may not be possible to quantitatively evaluate all stressors, indirect effects on wildlife and causal links should be identified, and their potential influence on the risk decision described.

Development—Dynamic energy budget models.

The theory of natural selection rests on the assumption that populations are limited in resources (Martin et al., 2013). Adaptation and repair of toxic injury can conceivably result in diversion of resources devoted to reproduction to somatic maintenance. Models have been developed based on these assumptions; some of which have supporting field observations as empirical evidence (Baas et al., 2018; Murphy et al., 2018; Nisbet et al., 2000).

Development—Bayesian networks and relative risk model.

The developments of the past 25 years in computation, geographical information systems, modeling, ecology, genomics, molecular biology, and data science can be used and linked to improve the fundamental goal of decision-making. These tools can be used to build a conceptual framework that supports computations to estimate risk and evaluate different management options.

Several modeling approaches have been applied to ERA that link the PF with subsequent analyses and can span several levels of biological organization. Among these approaches are Bayesian networks that can integrate inputs from individual-based models, nested matrix models, and dynamic energy budgets. In the late 2000s, the relative risk model (RRM) methodology began applying Bayesian networks as the computational framework. Since then, there have been a number of studies incorporating invasive species, disease, a contaminated site, pesticides, and incorporating human health (see the review by Landis, 2021).

Kaikkonen et al. (2021) present a critical review of the extensive literature using Bayesian networks in environmental risk assessment and risk management. It is the definitive comparative review of the state of the art in the early 2020s. It highlights that Bayesian networks have been used extensively to estimate fire risk and mitigation, for resource management (such as large ungulates), and for aquatic environments. Few examples are available for terrestrial wildlife (e.g., Carriger & Barron, 2020; Marcot & Penman, 2019). A Bayesian network performs the risk calculation and can include an AOP. Interactions between the nodes are described by conditional probability tables.

In parallel with increased use of Bayesian networks has been the development of the RRM, which uses a series of networks, rankings, and scoring to estimate the relative risk of multiple endpoints in different parts of the study area owing to multiple stressors (Landis & Wiegers, 1997, 2007; Wiegers et al., 1998).

The key to the structure of the RRM is a graph that incorporates sources of stressors, the description of the stressors, the location/habitat node that represents the spatial interactions of the stressors, organisms, and end-points, then effect and finally impact. The structure of the model is formally known as a directed acyclic graph. Starting in the 2010s, these models were built using Bayesian networks with the nodes representing the sources, stressors, locations, effects and impacts, and the links between them described by the lines of influence. Ayre and Landis (2012) demonstrated that the Bayesian networks provided a better description of the risk and how management alternatives can be evaluated. Johns et al. (2017) evaluated risks to the belted kingfisher and Carolina wren from exposure to mercury.

Typically, the development of the conceptual model and the Bayesian network is an iterative process. It is also possible to add management alternatives, to evaluate their influence on risk, and to evaluate causal links. The utility node will compute how well each management option works in various combinations as judged by cost, social values, and meeting the regulation. These alternatives can be considered in PF so that the right questions are asked and the correct data are assembled to inform potential decisions.

The building of a conceptual model is an early step in building the appropriate calculation. The calculation of risk can confirm the adequacy of the exposure determination or point to where additional research is needed. The conceptual model may have to be altered if it becomes apparent that new endpoints are to be considered or if other latent variables are driving the output risk determination. In risk communication, diagrams can be used to explain the process and calculate risk in real time before an audience to demonstrate how small changes in some variables dramatically alter risk outputs in a sensitivity analytical framework.

Challenge 7—Characterization of uncertainty to adequately inform decision-making

No risk assessment has complete, comprehensive, and exact information. Consideration of uncertainties starts in the PF when developing the analysis plan. However, use of assumptions is nearly always required. Addressing the uncertainty associated with the use of these assumptions, however, is often neglected and left to a few qualitative statements at the back of the report. A range of methods from quantitative to qualitative has been used. Tables have been used where the direction and relative magnitude of influence using an assumption is characterized, but these estimates should be used to warn about the accuracy of the risk estimate that, in turn, may be used to influence the outcome or decision.

Development—Use of new tools and methods to fill data gaps.

Bridging the relative influence of the potential for adverse effects across various biological levels of organization (e.g., cell, organ, individual, population, community) is an area of PF that often introduces various levels of uncertainty. There is also inherent uncertainty in each level of biological organization outside estimates that data can provide. However, there have been some advancements to better characterize variation and better estimate potential uncertainty in extrapolation where clear data do not exist.

The development of NAMs is an area of new research that provides biologists with greater tools to address uncertainty associated with extrapolation of effects between species (Bean et al., 2022). The use of data-derived uncertainty factors (DDUF) similar to those used for human health purposes can provide a technical basis for extrapolation, but are not without considerable effort (USEPA, 2014). Bayesian distributions are useful in refining assessment or uncertainty factors for human health and may also be valuable in interspecies extrapolation where data are not available to develop DDUFs (Simon et al., 2016). Given that the composition for each community is dynamic and unique, inferences on how stressors may interact to affect species assemblages is often uncertain.

Development—Improving characterization of uncertainty.

Use of assumptions should be explicit and highlighted in discussions regarding uncertainty when data are lacking. When data are lacking (e.g., to describe exposure parameters, food ingestion rates, etc.) ranges of criteria can be used where distributions can form the basis for probabilistic analyses or used in a semiquantitative way to show best and worst case conditions and most probable outcomes (Burmaster & Anderson, 1994; Hammonds et al., 1994; Regan et al., 2003). When resources are lacking, qualitative estimates regarding the direction and relative magnitude (e.g., high, medium, or low) of uncertainty can be provided in tables to aid in interpretation for each parameter used in the risk assessment where assumptions were used.

Challenge 8—Working within constrained regulatory frameworks

In the original ERA guidance, the steps, collectively known as PF, became established as the initial block in the process (NASEM, 2016; USEPA, 1992, 1998). However, the name does not convey the vital role of this segment to the process of calculating risk and its critical nature in decision-making. The initial steps in the risk assessment process require a series of specifications for each risk assessment tool to facilitate decision-making for a social group that includes multiple decision-makers and influential stakeholders. In many instances, some of the specifications in regulations were set decades ago or by legal precedent and are no longer consistent with the current state of the science. A clear example is the chemical-by-chemical evaluation and regulation of toxic substances. For pesticides, the chemicals exist as one component of a formulation, which is then added to an environment already containing thousands of other chemicals. Other chemicals, changes in soil quality, landscape alterations, and climate are also variable and can affect the criteria used for management. In this context, the risk analysis is separated from its environmental context.

Although some legislation recognizes the need for flexibility in defining the problem and development of the conceptual model (e.g., various habitats and community composition for contaminated sites), some are very prescriptive on what data will be used to understand exposure and toxicity and how those data will be interpreted (e.g., pesticide registration). Here, advances in science cannot be used without changes to guidance or policy even when such information advances the purpose of the legislation. However, there is value in improving our understanding of toxicity to target and nontarget species as well as developing models to better inform ecological relationships to improve our predictions regarding indirect effects. Even with restrictive policies, some new scientific advances can be used in a weight-of-evidence approach.

Development—Recognizing regulatory policies do change, present new/novel approaches along with standard approaches.

Good policies are built on good technical foundations. Improvements suggested herein provide more accuracy and possibly greater precision; however, they are often data and resource consuming. The use of traditional screening methods (e.g., use of hazard quotients) is likely appropriate; however, the use of novel methods provides the promise of greater accuracy where decisions are based on complex scenarios and thus difficult. These methods can be part of a phased approach to decide if further study is needed. Screening methods (e.g., HQs) should not be used to make remedial or conditions-of-use decisions when conditions of hazard (not risk) are suggested.

Mandated testing designs for pesticides present an opportunity for improvement (see Bean et al., 2022). Incorporating more treatments while not increasing the number of animals can provide more robust modeled threshold responses, thus moving away from problematic issues associated with NOAELs and LOAELs. Flexibility in the use of novel toxicology methods observed in some jurisdictions allows for a more complete understanding of exposure and effects for human health extrapolations (Lilienblum et al., 2008). These methods can drastically improve our ability to extrapolate effect data between species, improve our predictive capabilities in estimating exposure, and better understand transitional elements crossing biological levels of organization.

SUMMARY AND RECOMMENDATIONS FOR IMPROVING WILDLIFE ERA PF

Understanding the influence of exposure to anthropogenic chemicals to terrestrial wildlife species either before environmental release or thereafter requires an explicit conceptualization of the problem and the decisions that could be feasibly made as a result. In many cases, implicit (e.g., pesticide registration) or explicit (contaminated sites) PF plans have focused predominantly on exposure and toxicity interactions. Although these processes have value in screening (and possibly risk communication) approaches, advances in science and new methods can provide more accuracy and add confidence to risk projections when they are conceptualized in robust, holistic PF.

The iterative PF stage is critical to and a central component of the risk assessment process (Figures 1 and 2). With effective stakeholder engagement to identify outcome assessment needs related to utility analyses (Carriger et al., 2015), PF should focus on the intersection of problem detection, problem solving, establishing causal relationships, and management. Mapping the approach involves characterizing the exposome and the properties of the environment from source to external distribution to uptake and incorporation into biological tissues. These processes are dictated by the physical, chemical, and biological state of the environment that could involve the presence of mixtures and multiple stressors or be influenced by the life history of the species. Internal distribution is further influenced by toxicokinetics (i.e., absorption, distribution, metabolism, and excretion processes), which are species specific. Wherever possible, quantitative relationships (either through probabilistic or mechanistic/statistical models should be developed or used; Morrissey et al., forthcoming). Problem solving and establishing causal relationships involves using as much biological information as possible from several levels of biological organization that can inform ecological impacts. This should make use of population and community ecology and could employ the AOP knowledge base read across extrapolations, probabilistic frameworks, and cross-species extrapolations (Rattner et al., forthcoming). Problem solving and management are also constrained in an appropriate regulatory framework. Here we provide some examples and recommendations where these methods and such data must be included from the beginning of the risk assessment process (i.e., PF).

The following points are provided as a summary:

• Establish clear and specific protection goals early in the process; include explicit definitions of terms (e.g., what is meant by the “population”), and consider that goals may vary depending on attribute (specific species, ecological processes, etc.).

• Consider how data collection, including the use of new methods, will affect decisions, given all levels of uncertainty and all possible outcomes, and develop a decision plan a priori.

• Engage all relevant affected stakeholders in creating a robust, holistic conceptual model that incorporates plausible stressors that could affect the targets defined in the protection goals.

• Embrace the need for iteration throughout the PF steps (recognize that multiple passes may be required before agreeing on a feasible path for the rest of the risk assessment).

DATA AVAILABILITY STATEMENT

This article involves the input from all authors where consensus was reached on all topics. This article contains perspectives based on previous published work; therefore, the article contains no analysis of new data.