Mixture Math: Deciding What to Add in a Cumulative Risk Assessment

Abstract

Component-based approaches for cumulative risk assessment provide an important tool for informing public health policy. While current quantitative cumulative risk assessments focus narrowly on pesticides that share a mechanism of action, growing scientific evidence supports expansion of their application to encompass stressors that target a common disease. Case studies have demonstrated dose additive effects of chemicals with different mechanisms of action on liver steatosis, craniofacial malformations, and male reproductive tract developmental disruption. Evidence also suggests that nonchemical stressors such as noise or psychosocial stress can modify effects of chemicals. Focused research attention is required before nonchemical stressors can routinely be included in quantitative cumulative risk assessments.

Introduction

Large-scale biomonitoring efforts such as the National Health and Nutrition Examination Survey (NHANES) and innovative data analysis approaches have increased our understanding of the complex array of combined exposures and mixtures to which people are commonly exposed [1–3]. Furthermore, it is well-established that vulnerable populations are disproportionately affected by higher chemical burdens and myriad nonchemical stressors (e.g., poverty, social inequality) [4,5], and that certain life stages (e.g., early development, pregnancy, elderly) are more sensitive to disruption from environmental exposures [6,7]. Developing and evaluating methods for combining disparate chemical and nonchemical stressors in a cumulative risk assessment framework is critical for prioritizing interventions and protecting public health.

Evaluating human health risk from exposure to mixtures can center on whole mixtures (e.g., diesel exhaust, tobacco smoke) or component-based approaches that use data from individual chemicals and additivity models to estimate mixture effects [8–10]. Limitations to the application of whole mixture approaches include a lack of toxicity data on many whole mixtures and deficiencies in methods to determine sufficient similarity of a tested mixture to exposure-relevant variations of that mixture (i.e., read-across methods for complex mixtures are not well-established in a risk assessment context). These limitations combined with the relative abundance of single chemical data, have led to a reliance on component-based cumulative risk assessment approaches. Any approach for estimating cumulative risk based on the identified components in a mixture requires the analyst to make assumptions. One important assumption is that chemicals will act jointly according to a specified model of additivity. A significant body of research over many decades has been dedicated to identifying a default strategy for calculating additive effects from chemical mixtures [11,12]. Dose addition has emerged as a reasonable default assumption for chemicals within a common group [8]. Examples of the application of dose addition to chemical groups include toxic equivalency factors applied to dioxin-like chemicals [13] and relative potency factors for mixtures of polycyclic aromatic hydrocarbons [14]. More recently, relative potency factors have also been proposed for use with per- and polyfluoroalkyl substance mixtures [15]. In each case, potency factors are used to convert concentrations or doses of class members to reference chemical (e.g., benzo[a]pyrene in the case of polycyclic aromatic hydrocarbons) equivalents, which are then added to get a total mixture dose for risk estimation. Despite significant support for the use of dose addition as a default model, the critical issue remains of which chemicals belong in a common assessment group.

A second common assumption in cumulative risk assessment is that chemicals present in a mixture will not display less-than-additive (antagonistic) or, more importantly for public health protection, greater-than-additive (synergistic) effects. Debate on the public health significance of these potential interactions remains. Relatively few examples of synergy were identified in one review that focused on low dose mixtures [16]. A more recent review of the literature found additional examples of synergy and confirmed or identified some chemical classes with synergistic potential, including pesticide combinations (e.g., pyrethroids and azoles) and endocrine disrupting chemicals combined with certain metals [17]. Both reviews noted important deficiencies in the existing literature on detecting deviations from additivity and provided recommendations for mixtures studies. The authors recommended inclusion of lower doses (i.e., around points of departure) and sufficient doses to accurately define dose-response relationships, use of rigorous statistical methods for comparing predicted to observed data, and evaluation of higher order mixtures [16,17].

A critical mass of mixtures research has elucidated best practices for conducting informative studies on combined exposures [18] and has brought clarity to the key questions in the field of component-based mixtures risk assessment. These key questions boil down to which chemicals should be included in cumulative risk assessments and can we anticipate greater-than-additive or less-than-additive interactions based on knowledge of individual chemical mechanisms of action. Toward the goal of answering these questions, we will explore the utility of Adverse Outcome Pathways as a potential framework for developing and testing hypotheses about joint action. Finally, we will touch upon moving beyond chemical mixtures to include nonchemical stressors in cumulative risk evaluations. Answering these critical questions is necessary for advancing mixtures toxicology and moving toward a more public health protective framework for cumulative risk assessment [19].

Which chemicals should be included in a cumulative risk assessment?

Many factors can influence the decision of which chemicals to include in a cumulative risk assessment. Regulatory mandates, such as those in the Food Quality Protection Act of 1996 (FQPA) and Superfund legislation, dictate circumstances that require consideration of mixtures. For example, the FQPA requires EPA to consider the cumulative effects of pesticides that share a mechanism of action. Establishing a common mechanism group is a data intensive process [20] as evidenced by the limited number of classes with quantitative cumulative risk assessments (i.e., organophosphates, N-methyl carbamates, chloracetanilides, triazines, and pyrethroids). The 2016 EPA Pesticide Cumulative Risk Assessment: Framework for Screening Analysis Purpose provided a more flexible and less resource-intensive screening-level alternative to the 2002 EPA Guidance on Cumulative Risk Assessment of Pesticide Chemicals That Have a Common Mechanism of Toxicity [21]. Recently, researchers applied the 2016 guidance for establishing a candidate common mechanism group using a weight of evidence approach to evaluate available data from nine dinitroaniline pesticides (Figure 1, top panel) [22]. The dinitroaniline case study evaluated structural similarity, physicochemical properties, in vitro bioactivity, and in vivo data, but did not find a consistent pattern of similarity among the chemicals and concluded that they should not be included in a cumulative risk assessment [22].

Figure 1.

Comparison of approaches for cumulative risk assessment based on chemicals that share a mechanism of action (top panel) or based on chemical and nonchemical stressors that target a common disease (bottom panel). PM = particulate matter. See an example of the common mechanism deliberation in reference [17] and the disease-based pathway in reference [33].

Disease-centered grouping for cumulative risk assessment offers a promising alternative to the common mechanism grouping approach described above (Figure 1, bottom panel). The underlying premise of disease-centered grouping is that structurally diverse chemicals with disparate mechanisms of action can target pathways that converge on an endpoint or disease and contribute to dose additive toxicity. A common feature of work in this area is the use of Adverse Outcome Pathway (AOP) networks to organize available data and develop hypotheses about the joint action of chemicals [23,24]. In this use case, a comprehensive and quantitative AOP for each relevant pathway is not required. Rather, the AOP network provides a useful map of the intersecting pathways that could indicate potential additivity or interactions among chemicals that perturb those pathways.

The EuroMix project, funded through the European Union’s Horizon 2020 program, developed a series of case studies using AOP networks to evaluate mixtures of chemicals that impact reproductive function, craniofacial development, and liver steatosis [25]. A special issue of Food and Chemical Toxicology includes numerous studies from the EuroMix project as well as a brief editorial overview [26]. For the liver steatosis case study, researchers developed an AOP network with multiple molecular initiating events (MIEs) including activation of various receptors (e.g., peroxisome proliferator-activated receptor, pregnane X receptor, aryl hydrocarbon receptor) that converged on the key event of liver triglyceride accumulation and progressed to the adverse outcome of liver steatosis [27]. Pesticides operating through similar (imazalil and thiacloprid) and different (chlothianidin) MIEs were evaluated alone and as binary and tertiary combinations in multiple in vitro assays corresponding to different events in the AOP network with results demonstrating that a dose addition model provided a good fit to the data [27]. In vitro toxicogenomic studies in human HepaRG liver cells with binary mixtures of chemicals exhibiting varying levels of similar action demonstrated that dose addition provided a good model for most combinations with a tendency for synergy noted only at the highest concentrations of some combinations [28]. Next, they tested an equimolar mixture of 8 steatotic chemicals (amiodarone, benzoic acid, cyproconazole, flusilazole, imazalil, prochloraz, propiconazole and tebuconazole) and measured triglyceride accumulation in HepaRG cells [29]. Here, they compared different additivity models (e.g., dose addition and independent action) to observed mixture responses and found some indications of greater-than-additive interactions [29]. Finally, researchers evaluated individual chemicals and binary mixtures of imazazil, thiacloprid, and clothianidin in a 28-day Wistar rat study to compare in vitro to in vivo results. They found that the observed liver weight increase [30] and transcriptional changes indicating xenobiotic metabolism and nuclear receptor activation [31] generally conformed to a model of dose addition. In the craniofacial development case study, chemicals displaying notably different mechanisms of action were evaluated in an embryonic stem cell model [32] and a developmental zebrafish model [33] and binary mixtures were found to conform to an assumption of dose additivity. The final case study evaluated the effects of in utero exposure to endocrine disrupting chemicals with different mechanisms of action (estrogenic dienestrol, and antiandrogenic chemicals linuron and flutamide) on male reproductive tract development [34,35]. In these studies, flutamide appeared to be the driver of observed effects, while linuron and dienestrol did not meaningfully contribute. Other researchers have found that mixtures containing chemicals that disrupt androgen signaling via distinct mechanisms (e.g., pesticides that bind to and antagonize the androgen receptor and phthalates that decrease testosterone production) conform to a model of dose addition [36]. Based on the AOP network for male reproductive tract development and extensive data from mixtures studies, Kortenkamp concluded that, in order to be health protective, a cumulative risk assessment aimed at reproductive toxicant phthalates should include androgen antagonists, chemicals that disrupt steroid synthesis, InsL3 production, and prostaglandin signaling, as well as aryl hydrocarbon receptor agonists [37]. A similar framework for evaluating mixtures of chemicals that act on different targets involved in cancer development has been proposed to stimulate additional research [38].

Currently, there is scientific support for grouping chemicals that share common disease targets through different mechanisms of action for the examples that have been presented. Additionally, this work provides a blueprint for mixtures studies to investigate the effects of chemicals identified in AOP network analyses for different diseases (e.g., cardiovascular disease, neurodevelopmental disruption, metabolic disease). Finally, large-scale chemical screening efforts (e.g., Tox21, ToxCast) offer a wealth of useful data for prioritization of chemicals for the type of mixtures work outlined above. Assays linked to key events within an AOP network could be used to identify novel candidate chemicals that are likely to perturb relevant pathways [39]. Although more complex systems (e.g., zebrafish, organoid) can be used to evaluate chemical combinations, careful consideration should be given to the integrity of target pathways within the model system. In effect, is the complexity of the model system adequate for reflecting the potential toxicokinetic and toxicodynamic interactions among chemicals that perturb different pathways?

Adding nonchemical stressors to the mix

Nonchemical stressors (physical, biological and psychosocial) can contribute to disease development in much the same way as chemical stressors and should be considered for inclusion in the disease-centered cumulative risk assessment paradigm. A recent survey of the literature on nonchemical stressors in children’s environments found that certain nonchemical stressors (e.g., economics, educational attainment, exposure to violence) can adversely influence health and well-being [40]. However, they found a dearth of studies evaluating combined effects of chemical and nonchemical stressors. There is evidence that exposure to nonchemical stress can exacerbate adverse health effects associated with chemical exposures from epidemiological studies that consider both chemical and nonchemical stressors (e.g., air pollution and psychosocial stress) [41]. Epidemiological studies on combined chemical and nonchemical stressors are critical for understanding real-world exposures and prioritizing combinations for further study, while toxicological studies could provide an opportunity to deconvolute complex exposure scenarios and aid in quantifying effects and elucidating mechanisms of action of nonchemical stressors.

Incorporation of human-relevant animal models of psychosocial stress (e.g., social defeat stress [42], maternal deprivation to mimic early life adversity [43]) hold promise for increasing our understanding of behavioral and physiological variability in stress response and facilitate studies on combinations of chemical and nonchemical stressors. Examples of recent toxicological studies exploring chemical and nonchemical stressors include evaluation of cardiovascular effects from combinations of intermittent noise and ozone [44] and measurement of behavioral effects in offspring following exposure to prenatal stress and lead [45]. In both cases, findings involved nuanced interpretation (e.g., sex-specific response and differences over time, respectively) and highlight the challenges in quantifying the effects of nonchemical stressors. While beneficial factors (e.g., access to greenspace, exercise, healthy diets) clearly play a role in modifying risk [46], more work is required to develop practical approaches for incorporating both positive and negative nonchemical factors into cumulative risk evaluations. Current work in developing health impact assessments that incorporate multiple chemical (e.g., air pollution) and nonchemical (noise, heat, physical activity) stressors should inform future efforts [47]. Modeling approaches that attempt to provide a more complete picture of stressor interactions offer useful tools. For example, a systems dynamics model, designed to present relationships among interacting factors, was proposed for depicting the chemical and nonchemical stressors that impact neurodevelopmental risk in children [48].

Conclusions

Component-based approaches for cumulative risk assessment continue to offer pragmatic solutions for estimating risk from exposure to mixtures despite some well-known limitations. A critical question is which stressors should be included in cumulative risk assessment in order to protect public health. The current body of evidence on joint effects of chemicals that have a common disease target suggests that a dose addition model provides a reasonable default approach for evaluating risk, regardless of whether the chemicals have similar or different mechanisms of action. While grouping chemicals according to a common mechanism of action is dictated by legislation for pesticides, more flexible grouping approaches are supported by scientific evidence. Furthermore, cumulative risk assessments that encompass diverse stressors that target a common disease have the potential to be more health protective by accounting for a more comprehensive set of exposures. Nonchemical stressors can act on similar pathways as chemical stressors and can also impact disease development. Therefore, nonchemical stressors should be included in disease-centered cumulative risk analyses. Additional studies that quantify the joint action of chemical and nonchemical stressors will help to advance cumulative risk assessment practice. Finally, increased scientific attention to comparing component-based predictive approaches to data generated from whole, real-world mixtures is recommended.

Highlights:

• People are exposed to numerous diverse stressors throughout their lifetimes.

• Current cumulative risk evaluations focus on chemicals with shared mechanisms.

• Support is growing for a disease-centered approach.

• More work is needed in expanding cumulative risk to include nonchemical stressors.