Linking Mechanistic Effects of Pharmaceuticals and Personal Care Products to Ecologically Relevant Outcomes: A Decade of Progress

Abstract

There are insufficient toxicity data to assess the ecological risks of many pharmaceuticals and personal care products (PPCPs). While data limitations are not uncommon for contaminants of environmental concern, PPCPs are somewhat unique in that an a priori understanding of their biological activities in conjunction with measurements of molecular, biochemical, or histological responses could provide a foundation for understanding mode(s) of action and predicting potential adverse apical effects. Over the past decade significant progress has been made in the development of new approach methodologies (NAMs) to efficiently quantify these types of endpoints using computational models and pathway-based in vitro and in vivo assays. The availability of open-access knowledgebases to curate biological response (including NAM) data, and sophisticated bioinformatics tools to help interpret the information also has significantly increased. Finally, advances in development and implementation of the adverse outcome pathway framework provide the critical conceptual underpinnings needed to translate NAM data into predictions of ecologically relevant outcomes required by risk assessors and managers. The evolution and convergence of these various data streams, tools, and concepts provides the basis for a fundamental change in how ecological risks of PPCPs can be pragmatically assessed.

Importance of the Topic (Background)

In the 1990s endocrine-disrupting chemicals (EDCs) achieved notable public visibility, resulting in substantial research and regulatory activity focused on defining potential human health and ecological effects (Colborn et al. 1996; Hotchkiss et al. 2008). Concerns for EDCs in the environment arose from their ability to negatively impact reproduction and development in exposed organisms at sometimes exceedingly small doses. Some of the most potent EDCs are human pharmaceuticals (e.g., 17α-ethinyl estradiol [EE2]) and veterinary drugs (e.g., 17β-trenbolone) that are specifically designed to modify endocrine function in target species, but also can impact nontarget organisms when entering the environment through sources like wastewater treatment plants (WWTPs), agricultural run-off, etc. (Caldwell et al. 2012; Aris et al. 2014; Ankley et al. 2018). As understanding of EDCs evolved it became apparent that many pharmaceuticals and personal care products (PPCPs) intended to alter specific biological pathways similarly enter aquatic and terrestrial environments, with the potential to cause adverse effects in nontarget species (Ankley et al. 2007; Boxall et al. 2012; Arnold et al. 2014). With this realization, PPCPs became—and continue to be—widely regarded as environmental contaminants of concern, resulting in a variety of efforts focused on evaluating their potential for adverse effects (e.g., SETAC 2005; 2008; Boxall et al. 2012).

In 2011 a workshop was held in Niagara, Canada to help define needs to better assess and predict the potential ecological risks of the hundreds, perhaps thousands, of PPCPs potentially entering the environment. This workshop employed a systematic ranking exercise featuring an international group of experts to identify and prioritize research and regulatory needs for the assessment of PPCPs (Boxall et al. 2012). These needs were posed in the context of questions. The question ranked number four, from a total of more than 100 considered at the workshop was “How can ecotoxicological responses, such as histological and molecular level responses observed for PPCPs, be translated into traditional ecologically important end points, such as survival, growth, and reproduction of a species?”.

There are multiple reasons to consider molecular, biochemical, and histological endpoints (sometimes termed biomarkers) as an alternative (or complement) to whole organism responses such as survival or reproduction for assessing chemical risks (SETAC 1992). These types of mechanistic changes typically can be measured in a far shorter timeframe than the apical endpoints historically used for whole animal toxicity tests, so data collection can be much more resource efficient. In addition, because mechanistic responses generally occur earlier and at lower chemical doses than apical effects, they are more sensitive in terms of detecting the potential for adverse impacts. Finally, molecular, biochemical, and histological changes often can be linked to perturbation of specific biological pathways/processes, thereby potentially lending insights as to the nature and identity of chemical stressors in complex environmental settings.

Ecological assessments of PPCPs would greatly benefit from all the attributes associated with mechanistic effects data. For example, although availability of traditional ecotoxicity data for some PPCPs has increased in the past decade, many of these chemicals are data poor (especially personal care products), often with little or no empirical toxicity information even for those often detected in the environment (Gunnarsson et al. 2019; Pronschinske et al. 2022). Consequently, approaches to efficiently fill data gaps concerning potential for biological effects are needed. Also, most methods for assessing toxicity in the environment employ apical endpoints such as decreased survival, reduced growth, occurrence of malformations, etc., either in organisms collected from the field or in controlled lab studies with surface water, effluent, sediment, or soil samples. Since the targeted organisms almost always are exposed to complex mixtures it is very difficult to connect any observed adverse apical effects to specific chemicals or perturbed pathways.

Proposals for the use of mechanistic endpoints to assess the ecological risks of chemical contaminants are, of course, not new (SETAC 1992). But relatively little progress has been made in this area over the years in terms of practical applications. One notable factor contributing to a lack of progress involves inadequate understanding of biological activities of many environmental contaminants, which limits the identification of molecular, biochemical, or histological measurements suitable for chemicals/chemical classes of concern. One of the factors fueling optimism for use of mechanistic endpoints for assessing PPCPs is that in many instances—especially for pharmaceuticals—substantial a priori knowledge exists concerning biological pathways affected by these chemicals in target species (human, livestock, pets, etc.) obtained as part of discovery and development, as well as safety assessments. The potential to effectively leverage this information to help identify mechanistic measurements to support PPCP risk assessments for nontarget organisms was apparent to the participants at the original workshop (Boxall et al. 2012). As described in greater detail in the following section, the evolution of data collection and bioinformatic tools over the past decade has helped bring this hope ever closer to reality.

A second factor that has historically limited widespread use of mechanistic data to assess the ecological risks of chemicals involves a lack of transparent, causal linkages between observed changes in molecular, biochemical, and/or histological endpoints and impacts on survival, growth, and reproduction (SETAC 1992). Again, however, developments over the past 10 years now provide a more robust basis for achieving this critical linkage both for PPCPs and other environmental contaminants of concern, thereby enhancing acceptance of alternative types of data by regulators and risk assessors.

Understanding of the Topic (Current State-of-the-Science)

A Paradigm Shift in Regulatory Toxicology

In the few years immediately preceding the Niagara PPCP workshop, there had been a notable evolution in thinking about needs and testing approaches in regulatory toxicology. These changes were fueled in part by mandates such as the Registration, Evaluation and Authorisation and Restriction of Chemicals (REACH) program in Europe aimed at assessing the potential risks of an ever-larger chemical universe in an era of reduced resources for testing and a desire to decrease reliance on whole animal experimentation. A widely acknowledged strategic vision concerning alternative approaches to chemical safety assessment from the National Academy of Sciences (NAS) advocated greater use of novel computational methods and in vitro test systems to help define the potential for biological effects of data-poor chemicals (Krewski et al. 2010). Although the emphasis of this vision was almost exclusively on human health, the challenge of more efficiently collecting bioeffects data for large numbers of chemicals was/is equally applicable to ecological assessments (Villeneuve and Garcia-Reyero 2011; Villeneuve et al. 2019).

A wide array of powerful data collection techniques and concepts in bioinformatics developed over the past few years increasingly provide a basis for achieving the types of alternative approaches to chemical safety assessment proposed by the NAS (Krewski et al. 2010). A descriptive term capturing many of these tools has recently become common in the field of regulatory toxicology. New approach methodologies (NAMs) are techniques designed to assess the potential toxicity of chemicals that require minimal use of whole animal tests. Consequently, NAMs can encompass a variety of tools, including curated databases, computational models, and in vitro test systems (Kavlock et al. 2018). A slightly broader definition of NAMs arguably also would include short-term in vivo tests with molecular and biochemical endpoints (including ‘omics) that would serve to generate knowledge to help reduce/optimize animal testing. A common attribute of all NAMs is an emphasis on employing pathway-based data to generate insights concerning the potential biological activity of chemicals in a cost-effective, timely manner.

As approaches to generate molecular and biochemical data become more prominent in the field of toxicology so too does the need to translate this information into potential whole-organism responses meaningful to risk assessors—in the case of ecological effects, changes in survival, development. and reproduction relevant to populations. In the same timeframe as the PPCP priorities workshop was planned/held, the adverse outcome pathway (AOP) framework was proposed as a basis for making causal linkages in responses across biological levels of organization in an accessible, transparent manner (Ankley et al. 2010). An AOP depicts an interaction(s) of a chemical with a biological target (e.g., receptor, enzyme, nucleic acid, etc.) as a molecular initiating event (MIE), which can result in measurable perturbations at progressively higher biological levels of organization (key events; KEs) culminating in adverse responses in individuals and population.

The MIE and intermediate KEs in an AOP typically are the types of measurements made using different NAMs; as such, AOPs provide a basis for predicting potential negative apical effects using alternative data streams. Consequently, the AOP framework has received significant attention from different international bodies involved in chemical regulation and risk assessment (Carusi et al. 2018; FitzGerald 2020). In 2012 the Organisation for Economic Cooperation and Development (OECD) initiated a formal effort focused on development, standardization, review, and communication of AOPs, aimed at supporting their use in chemical risk assessments (OECD 2012). A critical resource supporting practical implementation of the framework is an open-access AOP-Wiki, a platform promoting collaborative AOP development and sharing of AOP knowledge (SAAOP 2021). There currently (March 2022) are more than 380 AOPs relevant to human and/or ecological effects of chemicals at different stages of development in the AOP-Wiki.

In the following section we describe the state-of-the-science concerning NAMs and AOPs suitable for supporting assessment of PPCPs in the environment, specifically in the context of addressing how molecular, biochemical, and histological data can be effectively generated and employed to predict potential ecological effects. Our examples are intended as illustrative rather than exhaustive. The relative brevity of this overview precludes critical consideration of all possible novel approaches that could be used to support PPCP evaluations.

Assessing Ecological Risks of PPCPs: Knowledgebases and Bioinformatic Tools

There are several curated, open-access knowledgebases that provide data relevant to assessing potential ecological hazards of PCPPs. For example, there are REACH dossiers that include ecological effects data for a number of PPCPs of possible interest (ECHA 2022). A resource specifically focused on potential ecological effects is the ECOTOX Knowledgebase, which has been maintained through the US Environmental Protection Agency (USEPA) lab in Duluth, Minnesota for more than 30 years (Olker et al. 2022). ECOTOX is a publicly available resource that provides curated toxicity results for ecologically relevant species from the peer-reviewed literature and EPA sources. Relevant studies are identified through application of systematic methods to conduct literature searches and review/screen references, with methodological details and toxicity results extracted into a structured database following standardized protocols and a controlled vocabulary. As of March 2022, ECOTOX included toxicity results for over 12,000 unique compounds for a wide range of aquatic and terrestrial organisms, with both apical (growth, development, reproduction, and mortality) and mechanistic endpoints (e.g., organ- and cellular-level, biochemical, and genetic effects). A recent redesign of the user interface and incorporation of common identifiers have increased the accessibility, transparency, and interoperability of toxicity records in ECOTOX, thereby enhancing searches for chemicals such as PPCPs (Olker et al. 2022).

With the interest in PPCPs in the environment, over the past 10 years there has been a substantial increase in the amount of relevant toxicity data for these chemicals in the ECOTOX Knowledgebase. Select PPCPs of concern (e.g., EE2, propiconazole, celecoxib, tramadol) have been specifically targeted for literature searches and data extraction; however, many other PPCPs also have had toxicity data added to ECOTOX. For the current paper we conducted a global analysis of the amount and types of toxicity data currently available for PPCPs in ECOTOX. A challenge for this analysis was defining the universe of chemicals classified as PPCPs, because there is currently no comprehensive listing of chemicals considered to be personal care products. To obtain a representative “snapshot” of the universe of PPCPs, a combination of informational sources was used: specifically, DrugBank on-line (Wishart et al. 2018) for pharmaceuticals (n=4,610 compounds), and for personal care products, a combination of the Combined 2000/2006 EU Cosmetics Ingredients Inventory (NORMAN Suspect List, n=2,878) and the EPA Consumer Products Suspect Screening list (Phillips et al. 2018, n=1,705). The indicated personal care products lists were obtained from the CompTox Chemicals Dashboard (USEPA 2021a). While there was some overlap of the pharmaceutical and personal care product categories across the databases, we arrived at a list of 8,547 unique chemicals.

A February 2022 ECOTOX download had toxicity information for 1,974 of the compiled PPCPs, including test results for 7,850 taxa from 23,968 publications, with highest representation of studies on terrestrial and aquatic invertebrates (38%), fish (26%), vascular (mostly terrestrial) plants (19%), and mammalian wildlife (11%). These studies include a variety of endpoints, with results for most of the 1,974 PPCPs (57%) including measurements both of apical endpoints and mechanistic responses. Data availability across the PPCPs is highly variable, with the number of publications abstracted in ECOTOX ranging from just one for 622 PPCPs to more than 200 for several other compounds (e.g., EE2, 17β-estradiol, bisphenol A, carbolic acid, nicotine, methanol, acetone, and others). Similarly, the taxa ranged from 1 to >1,000 unique species with toxicity results for a given compound, with data from only a single taxonomic group for many compounds (e.g., only fish studies currently are available for 243 PPCPs). While these ECOTOX searches and summaries provide insight as to the relative extent of testing across the universe of PPCPs and types of observed effects, it is important to recognize that these are not necessarily all-encompassing, as the PPCP lists used are representative rather than exhaustive, and ECOTOX does not include all published ecotoxicological studies (although updates are continually occurring).

There have been different efforts to assemble datasets curating what is known about PPCPs in terms of production volume, use categories, and chemical and biological properties, etc., that aid in estimating ecological exposures and the potential for biological effects (e.g., Kostich et al. 2010; Gunnarsson et al. 2019). A particularly notable effort in this regard was undertaken by contributors to the iPiE (Intelligence-led Assessment of Pharmaceuticals in the Environment) project in Europe, who developed an on-line system summarizing basic physico-chemical, pharmacokinetic, and toxicity data for a wide array of pharmaceuticals (iPiE 2021). A similar effort in the US focused on capturing and summarizing pharmacokinetic attributes of PPCPs to support ecological risk assessments (Berninger et al. 2016). The Mammalian Pharmacokinetic Prioritization for Aquatic Species Tool (MaPPFAST) features a curated database containing information for more than 1000 different PPCPs, representing 99 drug classes (Berninger et al. 2016). The data can be used for prioritization of PPCPs based on the concept of species read across, where extensive mammalian pharmacokinetic data can inform risk assessments for aquatic vertebrates for which little or no pharmacokinetic information exists. The tool uses a probabilistic scoring system across four primary pharmacokinetic endpoints (volume of distribution, clearance rate, half-life, peak plasma concentration) to establish a hazard score (1–10; low to highest hazard) for individual compounds or drug classes. The probabilistic approach allows for the estimation of missing data, and for PCPPs not in the database to be scored based on their drug class or physical-chemical properties. Access to MaPPFAST is currently available upon request from that the study’s lead author (Berninger et al. 2016).

A compelling argument supporting use of mechanistic data to assess potential hazards of PPCPs is that there often is knowledge concerning protein targets (MIEs) and biological pathways affected by chemicals of concern (Ankley et al. 2007; Gunnarsson et al. 2008). An important source of this information is DrugBank, which is an open-access database with entries for more than 500,000 drugs and drug products (Wishart et al. 2018). However, because the biological content in DrugBank is largely for humans and common mammalian models such as rodents, there is a need to translate this information into the potential for drug interactions and effects in nontarget species. Gunnarsson et al, (2008) described how this challenge could systematically be addressed through consideration of conserved protein targets relevant to PCPPs, and Verbruggen et al. (2018) described a basic tool for identifying relevant conserved targets. Expanding further on the concept, a dynamic, systematic approach for the comparative evaluation of drug targets across species was created. The Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS) tool is a web-based bioinformatics approach developed by the USEPA to predict potential susceptibility of organisms to chemicals that act on defined protein targets (LaLone et al. 2016). The open-access SeqAPASS tool incorporates protein sequence data from the extensive databases maintained through the National Center for Biotechnology Information (NCBI 2016) and transparently compares sequence data for specific proteins across, potentially, thousands of species. Consequently, if DrugBank or other resources report the protein target(s) for a given PPCP, SeqAPASS can be used to determine relative conservation of this MIE in a wide range of vertebrate, invertebrate and plant species, thereby identifying taxonomic groups most likely to be affected by a given chemical (e.g., LaLone et al. 2014). While the tools described by LaLone et al. (2014; 2016) and Verbruggen et al. (2018) represent an innovative conceptual basis for estimating hazard based on cross-species conservation of drug targets, the techniques are not intended to consider key pharmacokinetic considerations, such as PPCP metabolism, that can vary among organisms thus influencing potency and effects. Ultimately, the integration of effects-based predictive tools such as SeqAPASS with exposure/pharmacokinetic data for PPCPs captured, for example, by MaPPFAST could help address this challenge.

Assessing Ecological Risks of PPCPs: In vitro and In vivo Assays

One of the more prominent NAMs for screening biological activity of data-poor chemicals are in vitro assays, especially batteries of tests conducted using cost-effective, high-throughput (HTP) technologies. These types of assays typically measure molecular and biochemical alterations in components of specific biological pathways, which often correspond to early KEs (including MIEs) in AOPs (Fay et al. 2018). Pioneering work in the area of HTP testing was conducted through the USEPA ToxCast program, which has employed almost 2000 different assays/endpoints to determine bioactivities of several thousand chemicals. TocCast results are freely accessible (USEPA 2021b), so provide a ready source of pathway-specific bioactivity data for a range of compounds. For example, there are bioactivity measurements in the current ToxCast dataset for more than 1500 pharmaceuticals found in DrugBank.

An important attribute of ToxCast data for ecological assessments of PPCPs is the diversity of targets/pathways considered. Data from ToxCast can help confirm therapeutic targets of pharmaceuticals identified, for example, through DrugBank. However, ToxCast also can help identify targets of PPCPs not included in DrugBank, as well as uncover biological perturbations not necessarily associated with intended effects of the chemicals. Many pharmaceuticals are pleiotropic in that they can affect multiple biological pathways which may or may not be reported in databases such as DrugBank as side-effects. Consequently, ToxCast results can highlight unanticipated biological activities of PPCPs that could result in negative ecological impacts.

A challenge in employing ToxCast data to assess potential ecological hazards of PPCPs is that the majority of the HTP assays are based on mammalian systems so applicability to nonmammalian species is sometimes uncertain. However, it is possible to use bioinformatic approaches such as SeqAPASS to explore the taxonomic domain of applicability of different ToxCast assays. For example, LaLone et al. (2018) conducted extensive SeqAPASS analyses suggesting that data from a relatively high proportion of ToxCast assays focused on endocrine pathways should be applicable to most vertebrates.

Finally, HTP results can be useful for identifying taxa-specific molecular and biochemical measurements (including in vitro assays) that could support directed ecological assessments of different classes of PPCPs. For example, some pharmaceuticals elicit their therapeutic effects through activation or inhibition of a range of nuclear hormone receptors. To help screen for the potential ecological effects of chemicals that interact with the estrogen, androgen, thyroid hormone, and peroxisome proliferator-activated (ɣ) receptors, Medvedev et al. (2020) developed a multiplexed reporter assay that utilizes receptor constructs from representative mammalian, avian, reptile, amphibian, and fish species. This system was evaluated with several single chemicals by Medvedev et al. (2020), but also could be employed to screen complex environmental mixtures of contaminants (including PPCPs) that interact with these receptors with an approach analogous to studies using mammalian based HTP assays (e.g., Blackwell et al. 2019).

As genomic knowledge has become increasingly available for different species, so to have methods to rapidly assess molecular and biochemical changes in a wide range of organisms exposed to chemical stressors. While there is a desire to rely less on whole animal experimentation in the field of toxicology, there nonetheless can be important advantages associated with the use of in vivo systems for testing. For example, evaluation of intact animals allows consideration of complex interactions among multiple cell types, tissues, and/or organ systems, and accounts for pharmacokinetic variables (including metabolism) that are difficult to consider in vitro. Further, many newer in vivo approaches can rival in vitro NAMs in terms of cost-effectiveness and/or rapid data generation. For example, a consortium of North American scientists recently developed the EcoToxChip project ( Basu et al. 2019), which uses a polymerase chain-reaction array-based technology to characterize pathway responses to chemicals following in vivo exposures. One of the earliest EcoToxChip case studies involved exposure of double-crested cormorant embryos to two pharmaceuticals (EE2, fluoxetine; Crump et al. 2021) and many more case studies involving PPCPs are underway.

Significant progress also has been made recently to develop in vivo high throughput transcriptomics-based assays that would be useful for assessing potential ecological hazards of PPCPs. This research builds on evidence from the human health community that derivation of points of departure based on concentration-response modeling of global gene expression following short-term exposures may provide protective toxicity benchmarks mirroring those derived from much longer-term exposures (e.g., Thomas et al. 2007; 2019; NTP 2018; LaRocca et al. 2020). Based on promising results in mammals, investigators have begun exploring the application of these approaches in ecological hazard assessment using the model pharmaceutical EE2 (e.g., Pagé-Larivière et al. 2019; Alcaraz et al. 2021). Efforts are underway to develop both standardized assay protocols and low cost, high quality, transcriptomics platforms to support ecological high throughput transcriptomics testing as part of a tiered testing framework for hazard characterization (Thomas et al. 2019). These emerging approaches are currently being applied to characterize several PPCPs, including those frequently detected in the environment for which available ecological toxicity data are currently limited.

Assessing Ecological Risks of PPCPs: AOPs

Adverse outcome pathways can serve multiple roles in the assessment of PPCPs. The most notable of these is providing a credible and transparent basis for predicting potential adverse effects based on NAM data. For example, knowledge of the molecular target (MIE) for a given chemical (e.g., obtained from DrugBank or in vitro determination of specific bioactivities) allows the identification of relevant AOPs from the AOP-Wiki and/or the open literature. This in turn enables a broad prediction of potential apical effects of a given PPCP (or PPCP class) of concern. However, AOP information often can provide even more detailed insights as to possible hazard(s). Specifically, an important component of AOP development is definition of the biological domain of applicability of the construct (OECD 2018), thus targeting possible adverse outcomes in specific taxa. This information can help identify appropriate in vivo tests with data-poor chemicals in terms of species and endpoint selection, as well as highlight biological measurements suitable for monitoring in a field setting (e.g., Corsi et al. 2019).

In most instances prediction of apical effects based on AOP relationships will be qualitative; however, there is an increasing availability of quantitative AOPs (qAOPs) that enable quantitative forecasts of effects based on the degree of perturbation of a MIE or other early KE. An example of a qAOP for PPCP effects on fish reproduction was described by Conolly et al. (2018). In this instance, the MIE is inhibition of the steroidogenic enzyme aromatase (cytochrome P450 19), which is the target of pharmaceuticals designed to treat breast cancer and can also be affected by azole fungicides used in some personal care products (Trösken et al. 2004). Conolly et al. (2018) demonstrated how in vitro HTP measurements of aromatase activity inhibition could be used to quantitively predict decreased egg production and population size in the fathead minnow (Pimephales promelas).

Adverse outcome pathways also can inform the identification and/or development of NAMs suitable for assessing potential hazard. A relevant example of this from a human health perspective involves prediction of skin sensitization by personal care products such as soaps or cosmetics. An AOP developed for skin sensitization (AOP 40; SAAOP 2021) identifies KEs that can be captured with different in vitro assays, thereby reducing reliance on whole-animal testing for PPCP skin sensitization assessments (Schultz et al. 2016; de Avila et al. 2019). Assays/endpoints identified through AOP knowledge may feature measurement of a specific MIEs, but they could also include further downstream KEs which often are easier to quantify than the initial chemical-target interaction. For example, it can be difficult to directly measure binding to and activation of a nuclear receptor, but comparatively easy to measure the downstream consequences as changes in expression of specific mRNA products or proteins. An elegant example of this is the use of measures of vitellogenin (egg yolk protein) gene or protein induction in fish to indicate exposure to chemicals that bind to and activate the estrogen receptor (Sumpter and Jobling 1995). Identification of KEs associated with specific pathways/outcomes is useful for endpoint selection both in lab studies with individual chemicals and in field settings with organisms exposed to complex contaminant mixtures. Accordingly, it is reasonable to view AOP and assay/endpoint development as a complementary, iterative process critical to the broader use of NAMs to assess PPCPs.

Several hundred AOPs are described in the AOP-Wiki, many of which would be useful to the ecological assessment of PPCPs (Table 1; Supplementary Information Table S.1). Although the indicated AOPs are at different stages of development and review, they nonetheless can provide unexpected insights as to the possible effects of PPCPs, such as ketoconazole altering sexual behavior and subsequently reducing fecundity in birds and reptiles through inhibition of steroid synthesis (Table 1). In addition to the AOPs listed in Tables 1 and S.1, many other AOPs in the Wiki are anticipated to be applicable to ecological receptors, even if not explicitly tested in taxa of concern. For example, AOPs derived from experimentation with lab rodents (mice, rats) are likely directly applicable to terrestrial mammalian wildlife given within class conservation of many molecular pathways and processes. As previously noted, in the absence of direct empirical data concerning biological domain of applicability, bioinformatics tools such as SeqAPASS can provide reasonable predictions about potential species susceptibility based on MIE conservation. The variety of MIEs captured from our evaluation of the AOP-Wiki demonstrate that many PPCPs might impinge on targets/pathways which are conserved across species, such that unintended off-target species effects could occur. Finally, it is important to note that most AOPs provide evidence adequate to identify potential hazard but need to be combined with information regarding chemical potency, exposure, toxicokinetic properties, etc. to evaluate risk. Nonetheless, continuing to define and parse new MIEs and downstream KEs, and define taxonomic domain(s) of applicability remains important given the ever-increasing number of PCPPs found in or potentially entering the environment.

Table 1.

Select molecular initiating events (MIEs) and adverse outcome pathways (AOPs) associated with exposure to pharmaceuticals and personal care products (PPCPs) available in the AOP-Wiki database. This is an illustrative subset of those captured in Supplemental Information Table S1.

MIE	Representative PCPP(s)		Therapeutic Use(s)	Taxonomic Applicability	Key Events+	Adverse Outcome(s)	AOP ID(s)*
Androgen receptor agonism	17β-Trenbolone		Livestock growth enhancement	Fish	-reduced gonadotropin, T, and E2 -reduced vitellogenesis -impaired oocyte development	-reproduction dysfunction -decreased population trajectory	23a
Aromatase activity inhibition	Exemestane, Fadrozole, Letrozole Clotrimazole	Chemotherapeutic agents Antifungals		Fish	-reduced plasma E2 -impaired vitellogenesis -impaired oocyte development and ovary maturation	-reproductive dysfunction -decreased population trajectory	25a
					-reduced E2 synthesis -increased testis differentiation	-increased male-biased sex ratio -decreased population trajectory	346c
Cyclooxygenase activity inhibition	Celecoxib, Diclofenac sodium, Indomethacin, Ibuprofen		Nonsteroidal anti-inflammatory drugs	Birds	-decreased prostaglandin F2α plasma concentrations -renal ischemia and renal proximal tubular necrosis -increased blood K concentrations	-increased mortality -population decline	177c
CYPB7 activity inhibition	Ketoconazole		Antifungals	Birds, Reptiles	-decreased 7α-hydroxy pregnenolone synthesis in brain -decreased dopamine release -decreased locomotor activity or sexual behaviour	-decreased reproductive success -decreased population trajectory	218c 219c
Deiodinase I inhibition	Propylthiouracil		Antithyroid drugs	Fish	-decreased serum T3 -reduced swim bladder inflation -impaired swimming performance	-increased mortality -population decline	157a 158a
Glucocorticoid receptor activation	Beclomethasone dipropionate		Anti-asthmatics	Fish	-increased cripto-1 expression -inhibition of activin signaling -inhibition of fin regeneration	-reduced swimming performance -decreased population trajectory	334b
Hepatic vitamin K epoxide reductase inhibition	Warfarin		Anticoagulants	Humans, Rodents, Birds, Mammals	-uncoupling of oxidative phosphorylation -impaired osteocalcin carboxylation -clot development failure -hemorrhage, blood loss, anemia -impaired O2 delivery, nutrient delivery, and waste removal	-reduced fitness and/or mortality -acidosis, hypovolemic shock, organ dysfunction -bone deterioration -impaired recruitment -population decline	187c
Inhibition of 5-hydroxytryptamine transporter	Fluoxetine		Anti-depressants	Molluscs	-increased muscular waves and water retention in foot -increased valve movement -depletion of energy reserves -increased locomotion -increased serotonin	-increased predation -population decline	97c
Sodium channel inhibition	Amiodarone Carbamazepine, Topiramate, Phenytoin, Valproate, Zonisamide Lidocaine		Anti-arrhythmics Anticonvulsants Anaesthetics	Fish, Arthropods, Cnidaria	-decreased sodium conductance -reduced swimming speed	-reduced feeding -increased predation -reduced survival	91c
Thyroperoxidase inhibition	Methimazole, Propylthiouracil		Antithyroid drugs	Amphibians	-decreased thyroid hormone synthesis -decreased serum T4	-altered metamorphosis	175c

⁺Acronyms included in key event description defined as follows: E2 = 17β-estradiol; K = potassium; O₂ = oxygen; T = testosterone; T3 = triiodothyronine; T4 = thyroxine.

^*This table contains annotated and/or condensed overviews of associated AOPs. Detailed AOP information can be accessed by adding a corresponding AOP ID to the end of following website address: https://aopwiki.org/aops/ (SAAOP 2021)

^aAOP endorsed or approved by the Organization for Economic Co-operation and Development (OECD), Extended Advisory Group on Molecular Screening and Toxicogenomics (EAGMST), Working Party on Hazard Assessment (WPHA), and/or Working Party of the National Coordinators of the Test Guidelines Programme (WNT).

^bAOP flagged as ‘Open for citation & comment’ or ‘Open for adoption’.

^cAOP is putative and flagged as ‘Under Development: Contributions and Comments Welcome’.

Future Research Priorities (A Path Forward)

Ten years ago, at the research needs prioritization workshop in Niagara, the 4^th ranked question concerned the feasibility of using mechanistic data to predict adverse ecological effects of PCPPs (Boxall et al. 2012). This was not a new consideration in the field of ecotoxicology. More than 20 years earlier, SETAC sponsored a Pellston workshop focused on the use of biomarkers to support assessment of the ecological effects of contaminants (SETAC 1992). A major need identified at the Pellston meeting was easily measured, widely accepted molecular, biochemical, and histological endpoints capable of discerning specific chemicals (or chemical classes) of concern. A second highlighted need from the workshop was a framework(s) that could causally link changes in these types of endpoints to effects on survival, growth, and development (SETAC 1992). Over the past decade significant progress has been made with respect both to development of NAMs to assess the potential for chemicals to perturb biological systems and use of AOPs to translate resultant data into knowledge suitable for decision-making. The concepts underlying most NAMs are not new; what has changed has been the ability to efficiently collect and curate large amounts of data. Similarly, the basic underpinnings of the AOP concept have guided toxicologists for decades. What differentiates recent progress from past efforts to use pathway-based approaches in toxicology has been the focused work by many individuals and organizations to formally define AOP development, content, review, and dissemination which, in turn, has led to widespread consideration of utility in both scientific and regulatory communities (Carusi et al. 2018; FitzGerald 2020).

With these developments, the field is at an exciting juncture in terms of a practical ability to harness predictive toxicology tools for the ecological assessment of PPCPs, thereby addressing Question 4 from the Niagara workshop. Generation of data from in vitro and in vivo systems such as those described in the previous section makes it possible to identify molecular and biochemical responses for specific classes of PCPPs relevant to perturbations in biological processes of concern. Further, increasing availability of a variety of open-access interactive knowledgebases and computational tools (e.g., Table 2) directly supports predictive approaches to assessing hazards and risks both of existing and new PPCPs. In considering these developments in the context of Question 4 it is tempting to conclude that what is needed already has been achieved in terms of collecting and translating mechanistic data for PPCPs into information useful to risk assessors and managers. Unfortunately, this is not yet the case. Critical challenges remain in areas such as filling key data gaps and synthesizing existing information, including integrating various knowledgebases and computational tools in a manner that produces readily accessible, actionable insights.

Table 2.

Examples of available resources and their current interoperability that allow for automated or semi-automated collation of data for PPCP prioritization based on hazard.

Similar color shading indicates direct connections through data identifiers within each data node. Chemical connections are green; Effects connections are red, and Species connections are blue. Dashed square indicates tools that have connections among all 3 common nodes. Abbreviations: CAS – Chemical Abstracts Service Registry Number; DTXSID - Distributed Structure-Searchable Toxicity Database substance identifier; NCBI – National Center for Biotechnology Information.

Several types of new data/knowledge are needed to fully support implementation of a NAMs-based approach to assessing potential ecological impacts of PPCPs. For example, although an important objective in the use of NAMs is to reduce animal testing, there remains a critical requirement for focused whole-animal toxicity data to support practical implementation of this goal. In the case of PPCPs, apical effects data in diverse taxa are lacking for many relevant MIEs/pathways. Consequently, there is a need for strategic testing in phylogenetically diverse species with model chemicals representing different classes of PPCPs (see Ankley et al. [2005] for examples of species/test chemical selection for pharmaceuticals). Further, since PPCPs typically are designed not to be acutely toxic (i.e., many have very large acute to chronic toxic effects ratios), testing needs to focus on longer-term assays with sublethal endpoints (Ankley et al. 2005). Development of a robust taxa-chronic effects database with PPCPs that operate via different, well-defined pathways would provide a basis for validating and building regulatory confidence in PPCP-relevant NAMs, as well as informing the development of AOPs.

Similarly, additional HTP testing with PPCPs is warranted; as noted above, there are in vitro data for many PPCPs, but this represents only a fraction of the universe of this broad class of chemicals. Expanding the number/types of PPCPs with systematically collected in vitro data serves several purposes, including (a) identifying (or confirming) different bioactivities for what are often data-poor chemicals, (b) prioritizing chemicals (or chemical classes) requiring more in-depth testing or monitoring, and (c) guiding this testing in terms of selecting potentially sensitive species and endpoints.

There is a pressing need to expand current efforts synthesizing existing knowledge and new test data into AOPs that capture the diversity of MIEs/pathways impacted by PPCPs. Expanding the number of relevant AOPs is arguably the most important undertaking needed to fully address Question 4 from the Niagara workshop in terms of supporting predictive applications of NAM data. Ideally, this effort would include the development of qAOPs so that NAM data could be directly translated into adverse apical effects in a manner suitable for quantitative risk assessment (e.g., Conolly et al. 2018). As noted above, there are a variety of AOPs relevant to PPCPs, but more are required, especially in underrepresented taxonomic groups such as invertebrates or plants. The need for development of a more comprehensive assemblage of peer reviewed AOPs is not unique to PPCPs; challenges and potential solutions for achieving this goal were recently discussed by Carusi et al. (2018). For example, one impediment to the development of AOPs suitable for formal OECD review/approval is a lack of professional recognition commensurate with the significant amount of work required to produce a robust Wiki entry. A promising approach to addressing this lack of recognition currently is being implemented by different peer-reviewed scientific journals (including Environmental Toxicology and Chemistry) through the publication of “AOP Reports”, short review-type papers that summarize the content of full entries in the AOP-Wiki (e.g., Song and Villeneuve 2021).

A final need critical to supporting a predictive approach to assessing PPCPs involves continued identification and—when needed—development of knowledgebases and tools that capture and integrate existing chemical property and effects data. Ideally, these systems would be interoperable in terms of seamlessly interacting to provide information relevant to a given chemical. A summary of the PPCP-relevant knowledgebases and tools discussed in the preceding section is presented in Table 2, which also shows how the various systems do (or could) interact with one another. The CompTox Chemicals Dashboard is one example of an interoperable, integration of multiple databases and tools (USEPA 2021a). The Dashboard houses basic physico-chemistry and, when available, human health-oriented toxicity and exposure information for a large universe of chemicals (900,000 and growing). It is an open-access central hub that collates information from many tools and databases both internal and external to USEPA, making it highly interoperable and readily available for applications such as PPCP assessment. Significantly, the Dashboard currently integrates AOP-Wiki links and SeqAPASS output for, respectively, connecting chemical and/or bioactivity information to biological pathways and understanding their conservation across species. The ECOTOX Knowledgebase has been curated in a manner to expedite connections across tools based on chemicals, effects, and species. The literature review and curation process include connecting common data identifiers with coded fields that represent various elements of published study descriptions. The identifiers facilitate ECOTOX searches and enhance interoperability with other knowledgebases such as DrugBank, MaPPFAST, and the AOP-Wiki, as well as tools such as SeqAPASS (Olker et al. 2022). For example, programmers have been developing features to strategically link SeqAPASS analyses to ECOTOX output concerning potentially susceptible species (Sara Vliet, USEPA, personal communication). However, development of features supporting seamless interoperability of the data sources/tools shown in Table 2 is still evolving and needs to be a focus of additional efforts to merge information and data relevant to chemicals, chemical effects, and species in an integrated pipeline to support automated PPCP hazard identification and prioritization.

The accompanying Text Box summarizes our overall suggestions for priority research and implementation questions over the next 10 years in the context of Question 4 from Boxall et al. (2012). Although resources for conducting basic whole animal toxicology work with ecologically relevant species remain limited, this information is critical to interpreting and “anchoring” mechanistic data to apical responses. Additional work also is needed to expand the universe of PPCPs for which there are pathway-based bioactivity data from in vitro (including HTP) assay platforms. Synthesizing in vitro and in vivo data into new AOPs provides the pragmatic basis for translating NAM data into information useful for risk assessors and managers. Finally, the continued development of user-friendly interoperative knowledgebases and tools applicable to assessing the potential ecological effects of PPCPs is a critical activity.

In closing, we are optimistic that as the questions in the Text Box are addressed, in 10 years we will be presenting a “success story” for PPCPs highlighting the use of predictive approaches in 21^st century ecotoxicology. Furthermore, application of the types of concepts and tools discussed herein to the challenge of PPCPs will also provide a critical basis for productive use of these approaches in ecological assessments of other data-limited chemicals of concern.

Text Box 1.

Priority research and implementation questions for the next 10 years to support translation of histological and molecular level responses observed for PPCPs to traditional ecologically important hazards. Order of the questions does not indicate relative priority within the group of the three.

1. How can we develop and implement a systematic approach to collect in vitro bioactivity and in vivo effects data for a greater diversity of PPCPs and taxa?

2. How can the number of AOPs (including qAOPs) relevant to predicting the effects of PPCPs be expanded?

3. What is needed in terms of additional knowledgebases and computational tools to efficiently conduct hazard/risk assessments for the ecological effects of PPCPs?

Data Availability:

No new data were generated for this review manuscript.