Volume 43, Issue 3 p. 513-525
Special Series
Open Access

Cross-Species Extrapolation of Biological Data to Guide the Environmental Safety Assessment of Pharmaceuticals—The State of the Art and Future Priorities

Luigi Margiotta-Casaluci

Corresponding Author

Luigi Margiotta-Casaluci

Institute of Pharmaceutical Science, Faculty of Life Sciences & Medicine, King's College London, London, United Kingdom

Address correspondence to [email protected]

Contribution: Conceptualization, Visualization, Writing - original draft, Writing - review & editing

Search for more papers by this author
Stewart F. Owen

Stewart F. Owen

Global Sustainability, AstraZeneca, Macclesfield, Cheshire, United Kingdom

Contribution: Conceptualization, Writing - original draft, Writing - review & editing

Search for more papers by this author
Matthew J. Winter

Matthew J. Winter

Biosciences, Faculty of Health and Life Sciences, University of Exeter, Exeter, Devon, United Kingdom

Contribution: Conceptualization, Writing - original draft, Writing - review & editing

Search for more papers by this author
First published: 17 April 2023
Citations: 3

Abstract

The extrapolation of biological data across species is a key aspect of biomedical research and drug development. In this context, comparative biology considerations are applied with the goal of understanding human disease and guiding the development of effective and safe medicines. However, the widespread occurrence of pharmaceuticals in the environment and the need to assess the risk posed to wildlife have prompted a renewed interest in the extrapolation of pharmacological and toxicological data across the entire tree of life. To address this challenge, a biological “read-across” approach, based on the use of mammalian data to inform toxicity predictions in wildlife species, has been proposed as an effective way to streamline the environmental safety assessment of pharmaceuticals. Yet, how effective has this approach been, and are we any closer to being able to accurately predict environmental risk based on known human risk? We discuss the main theoretical and experimental advancements achieved in the last 10 years of research in this field. We propose that a better understanding of the functional conservation of drug targets across species and of the quantitative relationship between target modulation and adverse effects should be considered as future research priorities. This pharmacodynamic focus should be complemented with the application of higher-throughput experimental and computational approaches to accelerate the prediction of internal exposure dynamics. The translation of comparative (eco)toxicology research into real-world applications, however, relies on the (limited) availability of experts with the skill set needed to navigate the complexity of the problem; hence, we also call for synergistic multistakeholder efforts to support and strengthen comparative toxicology research and education at a global level. Environ Toxicol Chem 2024;43:513–525. © 2023 The Authors. Environmental Toxicology and Chemistry published by Wiley Periodicals LLC on behalf of SETAC.

INTRODUCTION

Patient access to effective and safe medicines relies on the successful completion of a complex four-dimensional drug discovery and development process (Wagner et al., 2018). Typically, this requires many years of research (average, 8.3 years; range, 3.6–16.6 years) and has an average cost estimated to be between $1.3 and $2.8 billion per drug (DiMasi et al., 2016; Wouters et al., 2020). This process generates large volumes of in silico, in vitro, and in vivo data across multiple mammalian species, including preclinical rodent and non-rodent species and clinical (human) data (Namdari et al., 2021). The interpretation and extrapolation of these multidimensional data drive human safety assessment and related decision-making (e.g., project closure vs. progression to clinical phases). Although the potential value of this knowledge to support the environmental risk assessment (ERA) of human pharmaceuticals (a legal requirement in Europe since 2006) has been recognized by the European Medicines Agency's guidelines (European Medicines Agency, 2006), the lack of a formalized process to guide the extrapolation of mammalian data to wildlife species has represented an important factor hampering this ambition (Winter et al., 2010).

The ERA of human pharmaceuticals is typically considered a stand-alone tiered process, and it relies on the experimental determination of adverse and non-adverse exposure concentrations using environmentally relevant model species representing three trophic levels (e.g., fish, invertebrates, algae). The assessment is conducted using standardized tests designed to detect apical effects on development, growth, reproduction, and survival (see European Medicines Agency, 2006; Organisation for Economic Co-operation and Development [OECD], 20112012a2013; US Food and Drug Administration, 1998). Considering the biological differences between mammalian species and other taxa, this experimental approach may appear like a pragmatic and effective strategy to generate empirical data, while minimizing uncertainties associated with the extrapolation of toxicity data across distant species. However, a recent large-scale analysis of ecotoxicity data available for 975 approved small-molecule drugs revealed that a complete set of regulatory compliant multispecies ecotoxicity data (e.g., across all three levels) is lacking for 88% of compounds (Gunnarsson et al., 2019). Similar conclusions were reached by Burns et al. (2018) by considering the 1912 active pharmaceutical ingredients (APIs) registered in the United Kingdom. Filling that data gap with experimental data would require decades of work (typically 2–3 years per compound) and, in the case of fish, the use of many thousands of protected animals. Specifically, Burns et al. (2018) found that only 11% of the 1912 APIs registered in the United Kingdom have ERA data. This data coverage decreases further if we consider the 332 APIs identified as priority compounds for ERA in 76 different prioritization exercises published in the scientific literature, as only 3% of those compounds have ERA data. Using this UK data coverage as a proxy, we could conclude that approximately 1700 APIs available on the market lack ecotoxicity data. The lack of data, by itself, does not automatically imply that regulatory relevant in vivo fish testing is needed. To estimate possible realistic testing requirements for the 1700 untested APIs, we can extrapolate the information already available for the 208 APIs that do have ERA data. In this case, the data collected by Gunnarsson et al. (2019) indicate that approximately 50% of those compounds triggered a fish early life stage test (OECD, 2013), 10% a bioconcentration assessment (OECD, 2012a), and approximately 5% a fish full (or reduced) life cycle test (OECD, 20112012b; US Environmental Protection Agency, 2016). Applying these proportions to the full list of 1700 APIs that lack data, this would conservatively translate into the use of >300 000 fish and would require a (likely unavailable) testing capability in contract laboratories able to accommodate >800 early life stage tests and 85 life-cycle tests. With this consideration in mind, there is an urgent need to develop more intelligent, efficient, and cost-effective approaches to prioritize compounds of concern and better predict potential adverse effects associated with the presence of human drugs in the environment.

This ambition could be realized, at least partially, by developing a battery of novel scalable and, most importantly, predictive in vitro and in silico methods (e.g., artificial intelligence [AI]–powered quantitative structure–activity relationships, in vitro organoids) for the detection of species-specific toxicity in the species of interest (e.g., fish; reviewed by Langan et al., 2024). However, even in a scenario in which this novel battery of “replacement, reduction, and refinement” (3Rs)–friendly methods were to become available, the volume of ecotoxicity data generated for human pharmaceuticals will always be much smaller than the volume of mammalian data generated for the purposes of human safety and efficacy assessment. Thus, framing ecotoxicity data within a wider data-rich cross-species knowledge base, which includes mammalian data, would bring significant advantages, facilitating data interpretation and weight-of-evidence (WoE) evaluations and increasing the overall confidence of the decision-making process for ERA. This mammalian data–driven “read-across” approach to streamlining ERA has been heralded for a number of years now (see Berninger & Brooks, 2010; Huggett et al., 2003; Rand-Weaver et al., 2013; Winter et al., 2010), yet are we any closer to accurately predicting environmental risk based on known human risk? The present review will discuss the advancements achieved in this field in the last 10+ years as well as future research priorities, with a special focus on fish and other aquatic species.

STATE OF THE ART

The interpretation of the toxicological relevance of complex biological data relies on the integration of multidimensional evidence generated along a continuum of processes that link exposure, mechanisms, and adverse effects.

Pharmacodynamics—Linking drug–target interaction to adverse effects

A key element needed to predict the potential toxicity of pharmaceuticals across species is the assessment of the evolutionary and functional conservation of drug targets. The higher the conservation between nontarget species (e.g., fish) and humans, the higher the probability of target-mediated effects (Rand-Weaver et al., 2013). In the last 10 years, this field of research has progressed at a very high pace, transitioning from the analysis of single targets (e.g., 5ɑ-reductase; Margiotta-Casaluci et al., 2013) to the large-scale evaluation of all known drug targets and the generation of publicly available informatic tools that allow user-friendly data exploration within an ERA-focused context, such as ECOdrug (Gunnarson et al., 2019; Verbruggen et al., 2018) and Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS; LaLone et al., 20132016). Importantly, these resources allow the assessment of the evolutionary conservation of drug target genes and proteins, in species of ecotoxicological relevance.

This systems-level understanding of the evolutionary conservation of drug targets has facilitated our ability to predict the hazard of pharmaceuticals in the environment (Gunnarsson et al., 2019; LaLone et al., 2013). A growing number of studies have demonstrated that mode of action–related effects can be accurately extrapolated from mammals to fish for several classes of pharmaceuticals (Table 1). These include antidepressants and other drugs targeting the central nervous system (see Brooks, 2014; Huerta et al., 2016; Margiotta-Casaluci et al., 2014; Tanoue et al., 2017; Valenti et al., 2012; Winter et al., 2021; Winter, Redfern, et al., 2008), cardiovascular drugs (Giltrow et al., 2009; Margiotta-Casaluci et al., 2019; Owen et al., 2007; Winter, Lillicrap, et al., 2008), adrenergic agonists (Weil et al., 2019), steroids and antisteroids (Margiotta-Casaluci et al., 2013; Runnalls et al., 20132015; Thorpe et al., 2003), steroidal and nonsteroidal anti-inflammatory drugs (NSAIDs; Margiotta-Casaluci et al., 2016; Marmon et al., 2021; Mehinto et al., 2010).

Table 1. Examples of experimental studies carried out using (wild-type) fish models, highlighting the potential wider translational value of mammalian data generated during the drug discovery and development process
Target system Study Species Drug MoA-relevant endpoints
Central nervous system Valenti et al. (2012) Fathead minnow (Pimephales promelas) Sertraline

  • Plasma drug concentration

  • SERT binding

  • Behavior

Central nervous system Margiotta-Casaluci et al. (2014) Fathead minnow (Pimephales promelas) Fluoxetine

  • Plasma drug concentration

  • Drug metabolism

  • Behavior

Central nervous system Huerta et al. (2016) Fathead minnow (Pimephales promelas) Oxazepam

  • Drug concentration in plasma, brain, liver, muscle

  • Behavior

Central nervous system Tanoue et al. (2017) Fathead minnow (Pimephales promelas) Tramadol

  • Drug concentration in plasma and brain

  • Drug metabolism

  • Behavior

Tanoue et al. (2019)
Central nervous system Winter, Redfern, et al. (2008) Zebrafish (Danio rerio) 25 compounds

  • Seizure-like behavior

  • Maximum tolerated concentration

Central nervous system Winter et al. (2021) Zebrafish (Danio rerio) 57 compounds

  • Whole-body drug concentration

  • Behavior

  • Brain activity

Cardiovascular system Milan et al. (2003) Zebrafish (Danio rerio) 100 compounds

  • Heart rate

Cardiovascular system Parker et al. (2014) Zebrafish (Danio rerio)

  • Adrenaline

  • Cisapride

  • Haloperidol

  • Terfenadine

  • Theophylline

  • Verapamil

  • Whole-body drug concentration

  • Maximum tolerated concentration

  • Heart rate

  • Atrium:ventriculum beat ratio

  • Stroke volume

  • Blood flow

  • Blood vessel diameter

Cardiovascular system Margiotta-Casaluci et al. (2019) Zebrafish (Danio rerio)

  • Captopril

  • Losartan

  • Propranolol

  • Maximum tolerated concentration

  • Heart rate

  • Atrium:ventriculum beat ratio

  • Stroke volume

  • Blood flow

  • Blood vessel diameter

Reproductive system Runnalls et al. (2013) Fathead minnow (Pimephales promelas)

  • Desogestrel

  • Drospirenone

  • Gestodene

  • Levonorgestrel

  • Reproductive activity

  • Egg production

  • Breeding frequency

  • SSCs

Reproductive system Margiotta-Casaluci et al. (2013) Fathead minnow (Pimephales promelas) Dutasteride

  • Reproductive activity

  • Egg production

  • Breeding frequency

  • SSCs

  • Plasma E2, T, 11-KT

  • Gonad histopathology

  • Sperm quality

Reproductive system Runnalls et al. (2015) Fathead minnow (Pimephales promelas)

  • Ethinylestradiol

  • Levonorgestrel

  • Plasma drug concentration

  • Reproductive activity

  • Egg production

  • Breeding frequency

  • Plasma E2 and 11-KT

  • SSCs

Reproductive system LaLone et al. (2012) Fathead minnow (Pimephales promelas)

  • Growth

  • Reproductive activity

  • Egg production

  • Breeding frequency

  • F1 hatching rate

  • Ex vivo gonad T and E2 production

  • SSCs

  • Gill histopathology

  • Expression of MoA-relevant genes

Respiratory system
Development
Immune system Kugathas and Sumpter (2011) Fathead minnow (Pimephales promelas)

  • 10 synthetic GCs (in vitro)

  • Beclomethasone dipropionate

  • Prednisolone

  • Target transactivation

  • Plasma glucose

  • Blood leukocyte count

Metabolism Rainbow trout (Oncorhynchus mykiss)
Immune system Kugathas et al. (2013) Fathead minnow (Pimephales promelas)

  • Becloemthasone dipropionate

  • Plasma cortisol

  • Plasma glucose

  • PEPCK and GR gene expression

  • Blood leukocyte and thrombocyte count

  • SSCs

Metabolism
HPI axis
Immune system Margiotta-Casaluci et al. (2016) Fathead minnow (Pimephales promelas) Becloemthasone dipropionate

  • Drug metabolism and plasma drug concentration

  • PEPCK and GR gene expression

  • Blood glucose

  • Leukocyte subpopulations response

  • AR gene expression

  • SSCs

Metabolism
Secondary pharmacology
Immune system Mehinto et al. (2010) Rainbow trout (Oncorhynchus mykiss) Diclofenac

  • Drug metabolism and bile drug concentration

  • ptgs1 and ptgs2 gene expression

  • Kidney, liver, intestine histopathology

Renal system
GI system
Immune system Patel et al. (2016) Fathead minnow (Pimephales promelas) Ibuprofen

  • Plasma drug concentration

  • Prostaglandin E

  • ptgs1 and ptgs2 gene expression

  • E2 = estradiol; 11-KT = 11 keto-testosterone; MoA = mode of action; SERT = serotonin transporter; SSCs = secondary sexual characteristics; T = testosterone; GC = glucocorticoid; GI = gastrointestinal; GR = glucocorticoid receptor; HPI = hypothalamic–pituitary–interrenal; AR = androgen receptor; PEPCK = phosphoenolpyruvate carboxykinase; ptgs = prostaglandin-endoperoxide synthase.

In some cases, the adverse outcome pathway (AOP) framework (Ankley et al., 2010; Conolly et al., 2017; Margiotta-Casaluci et al., 2016) and AOP-informed data-visualization strategies have been used to synthesize the evidence mentioned above (see Margiotta-Casaluci et al., 2016; Marmon et al., 2021) and have demonstrated the great potential of these data-driven approaches to support ERA and follow-up ecopharmacovigilance strategies (Gunnarsson et al., 2019; Marmon et al., 2021). Moreover, meta-analyses have been used to generate quantitative comparative evaluations of mode of action–relevant responses across species (i.e., effect magnitude and direction), highlighting the value of this approach to support the WoE assessment of toxicology data from a cross-species extrapolation perspective (Margiotta-Casaluci et al., 2019). The successes achieved with the mode of action–centered strategy discussed above holds great promise for the scaling up of this predictive approach to inform the ERA of groups of pharmaceuticals acting via common pharmacological mechanisms. For example, Marmon et al. (2021) demonstrated the value of network pharmacology methods and pharmacokinetic (PK)/pharmacodynamic (PD) data synthesis to support hazard and risk assessment of a mixture of 25 NSAIDs.

It is important to highlight that the experimental efforts explicitly aimed at characterizing the cross-species concordance of mode of action–driven effects of pharmaceuticals (for ecotoxicology applications) have been largely focused on fish, whereas a relatively limited amount of research has been carried out on invertebrates. Some notable exceptions are represented by antidepressants (see Campos et al., 2012; Fong & Ford, 2014; Rivetti et al., 2016), beta-adrenergic receptor agonists and other cardiovascular drugs (see Dzialowski et al., 2006; Stanley et al., 2006; Villegas-Navarro et al., 2003), and steroid hormone agonists (see Kaur et al., 20152016), whose effects have been investigated also in mollusks and crustaceans. Larger sets of invertebrate data are available for apical endpoints, like those quantified during regulatory ecotoxicology testing (e.g., algal growth, Daphnia magna reproduction). Analyzing this apical regulatory relevant data set, Gunnarsson et al. (2019) observed that the toxicological sensitivity of fish, Daphnia, and algae to pharmaceuticals may be comparable when the relevant drug target is conserved across these three taxa. Moreover, Coors et al. (2023) recently proposed that this scenario may even be leveraged to reduce the in vivo fish testing needed for the safety assessment of legacy pharmaceuticals, suggesting that the no-observed-effect concentrations generated with Daphnia and algae may also be sufficient for the protection of fish.

Drug uptake and PKs—The importance of internal exposure

The ERA of pharmaceuticals typically considers the concentrations of the compound of interest (either nominal or measured) in the relevant environmental matrices outside exposed living organisms (e.g., water, sediment, soil). From a practical perspective, this approach facilitates hazard and risk evaluations by providing a parameter that is easy to interpret, even without prior knowledge of the pharmacological features of the compound or the need for complex analytical chemistry. Indeed, the simple comparison of measured environmental concentrations and effect concentrations (e.g., lowest-observed-effect concentrations) determined in ecotoxicity studies has been proposed as a rapid method to prioritize chemicals, including pharmaceuticals, for ERA in the aquatic environment (Donnachie et al., 2016; Johnson et al., 2017). Despite its obvious advantages, this approach can only allow retrospective evaluations, whereas modern ecotoxicology should strive for high-precision predictions that are able to inform hazard- and risk-management strategies before the marketing of the compound, or at least when market penetration is still limited.

Considerable progress has been achieved in the measurement (Tanoue et al., 2020; Wilkinson et al., 2022) and modeling (see Kapo et al., 2016; Oldenkamp et al., 2018; Wilkinson et al., 2022) of environmental concentrations of pharmaceuticals in rivers, and dedicated databases have now been created to collate exposure data published globally (Graumnzi & Jungman, 2021). However, from a pharmacological and cross-species perspective, predictive toxicology methods would benefit greatly from understanding uptake and PK profiles of pharmaceuticals in non-target (i.e., non-human, environmentally relevant) species, because it is the concentration of chemicals inside the organism (e.g., in the blood or other tissues or more precisely at the target site) that ultimately drives toxicological risk (Huggett et al., 2003; Hutchinson et al., 2014; Margiotta-Casaluci et al., 2016; Nyman et al., 2014). The importance of this concept was highlighted by large experimental studies on synthetic glucocorticoids, in which compounds with comparable in vitro potency showed very different abilities to elicit in vivo effects, which were in line with their different uptake and PK features (LaLone et al., 2012; Margiotta-Casaluci et al., 2016).

The read-across hypothesis and the fish plasma model (Huggett et al., 2003; Rand-Weaver et al., 2013) have provided a theoretical framework to interpret the pharmacological risk of pharmaceuticals in fish, based on a comparison with pharmacologically and toxicologically relevant measured blood concentrations in humans. The theoretical and experimental utility of this approach has been demonstrated in several large-scale prioritization studies (see Fick et al., 2010; Gunnarsson et al., 2019; Marmon et al., 2021; Sumpter & Margiotta-Casaluci, 2022) and in a growing number of experimental studies performed on a wide range of classes of pharmaceuticals, including beta-blockers (Giltrow et al., 2009; Owen et al., 2009; Winter, Lillicrap, et al., 2008), beta-adrenergic receptor agonists (Weil et al., 2019), antidepressants (Margiotta-Casaluci et al., 2014; Valenti et al., 2012), anxiolytics (Huerta et al., 2016), opioids (Tanoue et al., 2017), synthetic glucocorticoids (Margiotta-Casaluci et al., 2016), and synthetic progestins and estrogens (Runnalls et al., 2015). These experimental studies have been instrumental in the validation of the fish plasma model and have helped to refine testing approaches to account for factors such as plasma protein binding, metabolic activation of prodrugs, and pH (Chang et al., 2021; Henneberger et al., 2022; Tanoue et al., 2017).

Although in vivo experimental studies are highly valuable for advancing this field, they often still require the use of protected animals; thus, there remains an urgent need to integrate all existing data and information to generate scalable modeling methods and databases that are in line with the 3Rs-driven vision of modern toxicology. In line with this, already in 2005, Reimschuessel and colleagues (from the US Food & Drug Administration) published a curated database (Phish-Pharm) that includes absorption, distribution, metabolism, and excretion (ADME) and other PK information extracted from over 700 articles (to date) on 191 aquatic species (Reimschuessel et al., 2005). This data centralization exercise was followed by the work of Berninger et al. (2016), who generated a database containing ADME data for 1070 APIs, which can be used to drive the translation of PK considerations across species and inform the development of fish-specific PK models (Brinkmann et al., 2016; Nichols et al., 1990). Although not widely used, the development of PK models in environmental toxicology has been growing slowly but steadily over the last decade, unsurprisingly, in parallel with rapidly improving analytical chemistry capabilities (see Grech et al., 2019; Larisch et al., 2017; Stadnicka et al., 2012). In a recent example, Larisch and Goss (2018) assessed which physiological parameters drive interspecies differences in fish PK and identified lipid content, ventilation rate, uptake efficiency from food, and metabolism rates as the most important factors. Because PK models have only been developed for very few fish species, Wang et al. (2022) proposed to overcome this limitation by developing a generalized physiologically based (PB) PK model for fish that could facilitate the application of PK modeling for the ERA of pharmaceuticals more widely. Beyond fish species, a similar approach was used by Baier et al. (2022) to develop a generic avian PBPK model for avian species, validated in three bird species using a set of pharmaceuticals that included antibiotic, antiparasitic, antifungal, and sedative compounds.

One of the main challenges hampering the transition toward the wider application of PK modeling in ecotoxicology is the availability of species-specific ADME data; however, in vitro and computational methods can represent a valuable 3Rs-friendly alternative that builds on the advancements achieved in the human ADME and in vitro-to-in vivo modeling fields (Davies et al., 2020; Lombardo et al., 2017). Several fish in vitro methods have been established for the characterization of drug uptake through gills (Chang et al., 2021; Stott et al., 2015) and intestine (Langan et al., 2017), hepatic metabolism (Baron et al., 2017) and clearance (Baron et al., 2017; Nichols et al., 2018).

The integration of the PK and PD considerations, as discussed above, could greatly accelerate the transition from hazard to risk assessment because it would allow us to predict whether, under a given exposure scenario, the compound of interest could achieve target-site concentrations high enough to activate a pharmacological response. Hence, the extrapolation of PK/PD across species could represent a powerful tool to generate accurate toxicity prediction, even in those cases when species-specific data availability is limited. An additional area of promise that could be leveraged into future PK/PD models is the expanding field working to better predict uptake into fish, invertebrates, and terrestrial organisms (Carter et al., 2024; Miller et al., 2019). Combining these uptake models based on machine learning and mathematical models that include ionization with in vitro systems (listed above) that can determine metabolism and excretion offers the prospect of a tiered approach to understanding the risk of these chemicals but without the need for in vivo experimentation.

PRIORITIES FOR FUTURE RESEARCH

Assessing the conservation of drug targets—From structure to function

A central hypothesis of the read-across concept for the ERA of pharmaceuticals is that the higher the evolutionary conservation of human drug targets in other species, the higher the probability that the drug will induce target-mediated effects in those species (Rand-Weaver et al., 2013). This hypothesis prompted an intense program of research that culminated in the development of the databases ECOdrug (Verbruggen et al., 2018) and SeqAPASS (LaLone et al., 2016), mentioned previously. These novel and centralized knowledge platforms have been important in accelerating the implementation of predictive approaches within a discipline that typically relies on experimental data. However, the interpretation of the biological significance of gene similarity data across species is far from being straightforward. Typically, a given biological target can play numerous functional roles in different cell types or tissues. Most of these functions may be highly similar between humans and other species, such as fish (Colbourne et al., 2022). However, others may not. For example, the gene (n3rc1) encoding the glucocorticoid receptor (GR) of many species of teleost fish displays a medium degree of similarity (~48%) to the human GR (Verbruggen et al., 2018). Despite this relatively low gene sequence homology, drug-induced GR modulation elicits similar effects in both fish and humans, including the perturbation of the immune system, reproductive function, and glucose metabolism (Hamilton et al., 2022; Margiotta-Casaluci et al., 2016). This apparent discrepancy is clarified once we consider the conservation of protein homology specifically in functional domains. Indeed, the DNA binding domain and ligand binding domain of the GR protein display, respectively, 98.4% and 86.5% similarity to the human protein (Schaaf et al., 2009). This example suggests that future emphasis should be focused on assessing drug target conservation at the level of individual gene/protein domains, rather than merely at the whole-gene/protein level (Kruger et al., 2012; Nitta et al., 2015). In this respect, precise genome editing approaches (e.g., using clustered regularly interspaced short palindromic repeats [CRISPR]/CRISPR-associated [Cas] proteins) can be useful for targeting specific sequences within a given gene with relative ease and allowing measurement of the resultant functional impact in the species of interest (Cornet et al., 2018; Kroll et al., 2021; Winter et al., 2022). The progressive knockout of different coding regions could also help to determine the functionally important parts of a given gene where this is not known (Klann et al., 2017; Wright & Sanjana, 2016), a process that has been rendered achievable through the advent of CRISPR-mediated approaches. Given the rapid growth of genetic screening and functional genomics applied to drug discovery (Gianni & Farrow, 2020), this knowledge may also be extrapolated from existing mammalian data without the need for further experimental determination. An ecotoxicological application of this approach could focus on a priority set of drug targets, such as those most likely to be frequently modulated by pharmaceuticals in the environment.

The interpretation of the functional implications of a given drug–target interaction is challenged further by a duplication event that occurred in the genome of teleost fish 350–400 million years ago (Glasauer & Neuhauss, 2014) and by interspecies differences in resultant protein function. For example, many teleost species possess two GRs (GR1 and GR2) as a result of gene duplication, and the distribution of ligand binding affinity values to the various isoforms in different species varies greatly (Hamilton et al., 2022). This, of course, has important implications for the assessment of functional conservation across species. Furthermore, the GR plays an important additional osmoregulatory function in fish (but not in humans; Bury et al., 2003; Prunet et al., 2006), which may directly affect the risk profile of GR-modulating chemicals in aquatic species. This aspect of read-across, however, can only be addressed on a case-by-case or, rather, species-by-species basis; and this requires expertise that is not widely available. This last point raises the importance of the need for appropriate training of the next generation of comparative physiologists, pharmacologists, and toxicologists to equip them with the knowledge to interpret the potential effects of chemical exposure between, often rather disparate, groups of species of relevance to ERA.

In all, these examples suggest that the study and classification of target functions across species is the next big step needed to enhance the predictive power of precision toxicology frameworks (Priority Research Question 1—Textbox 1). For example, such functional information could be used to guide the development of robust, quantitative AOPs and AOP networks (Margiotta-Casaluci et al., 2016; Spinu et al., 2020) that, in turn, may inform the development of 3Rs-driven testing strategies to identify biological biomarkers that could be used for ecopharmacovigilance. One important challenge to this vision is that the mapping of drug target functions across species would require significant resources, and in practical terms, a full mapping may be unachievable. However, a step-by-step strategy may be used to overcome this challenge. Firstly, a large volume of knowledge has already been generated over decades of comparative physiological research. We suggest that future research efforts should be focused on the extraction and classification of such information (e.g., with the help of text-mining tools supported by AI) within informatic databases designed to support evidence synthesis (Brooks et al., 2021; Lee et al., 2020; Priority Research Question 2—Textbox 1). This ambitious work has already been initiated by ongoing international activities, such as the MONARCH Initiative (https://monarchinitiative.org/), which is an integrative data and analytic platform that extracts and synthesizes all available evidence to connect phenotypes to genotypes across species (Shefchek et al., 2020).

Textbox 1..

Priority research questions to advance the application of comparative pharmacology and toxicology methods to guide the environmental safety assessment of pharmaceuticals.

  • 1)

    In the last 10 years, we have successfully characterized the evolutionary conservation of the genes coding almost all drug targets in many non-target species. Can we generate a similar knowledge map for the functional conservation of drug targets?

  • 2)

    Comparative biology knowledge is currently scattered across many different sources. What is the best way to extract that knowledge and make it usable for environmental toxicology applications?

  • 3)

    Cross-species extrapolation approaches can generate highly granular predictions at the subapical level. On the other hand, regulatory decision-making is only interested in apical adverse effects. What is the quantitative relationship between subapical adverse endpoints and apical effects across species?

  • 4)

    Can we use modeling approaches to overcome the experimental (and resource) limitations of using model laboratory organisms?

  • 5)

    Can we develop a scalable battery of new approach methodologies (NAMs) for the characterization of uptake, pharmacokinetics (PK), and pharmacodynamics (PD) of chemicals in fish species?

In addition to this data mining approach, modern genetic modification techniques coupled with advanced imaging approaches offer an exciting opportunity to shed light on the functional role of drug targets in fish and other species of environmental relevance. This approach can be applied to cells or directly in vivo, for example, using nonprotected zebrafish larvae, and is already being applied in assessing drug safety and efficacy for human health purposes. For example, we have used a combination of in vivo imaging and CRISPR/Cas9-gene editing in larval zebrafish to assess the impact of risk gene knockout on cardiovascular function (Winter et al., 2022). Using a pan-neuronal Ca2+ sensing transgenic zebrafish and light sheet microscopy, we have also profiled the effect of 57 central nervous system–active drugs/chemicals on brain activity in non-protected larval zebrafish (Winter et al., 2021). Using these types of approaches, gene–phenotype relationships can be established for virtually any desired organ/system; and, as the second study shows, the number of compounds that can be realistically assessed is typically much greater than that using traditional ecotoxicological testing paradigms. Higher-throughput and higher-content approaches such as this also offer advantages in terms of low compound requirements and higher statistical power, along with equal suitability for method standardization. Most importantly, in certain contexts, such imaging-based approaches provide greater opportunities for the identification of specific mode of action–relevant effects that can help to establish a quantitative relationship between target modulation and adverse phenotype (Priority Research Question 3—Textbox 1). For example, the ability to link alterations in drug-induced neuronal activity with specific brain regions known to be target-rich (Winter et al., 2021) suggests that a large-scale application of functional assessment and gene-editing approaches may allow highly granular mapping of gene function in fish, using 3Rs-friendly methods.

Higher-throughput ecotoxicological testing is also being aided by the availability of automation systems which were previously the preserve of pharmaceutical company in vitro drug screening facilities. These include wider application of high-content imaging systems (Green et al., 2016; Lempereur et al., 2022; Westhoff et al., 2020), microfluidics for animal manipulation and exposure (Pulak, 2016), and advanced image analysis scripts for processing the large amounts of data generated by such platforms (see Caicedo et al., 2017; Otterstrom et al., 2022). The data generated using these approaches could then be used to build and train models to predict the potential impact of functional conservation, without the need for costly animal testing at all. Of course, caution still needs to be applied with regard to the interpretation of such data in the context of their potential relevance for the ERA process. Specifically, it would be essential to consider the quantitative relationship between novel functional data (e.g., cardiovascular parameters) and the apical functions typically considered by the ERA framework (i.e., development, growth, reproduction), which is centered on the protection of population dynamics rather than on the protection of individuals (as in the case of human risk assessment; Priority Research Question 3—Textbox 1). While the use of these techniques can help us to reveal the potential for drug–target interactions in environmentally relevant species, understanding the wider ecological consequences of such effects is crucial to use these data appropriately in the refinement of ERAs. This is perhaps where more research into the integration of approaches such as functional brain imaging, alongside more directly ecologically relevant holistic markers of effect (e.g., behavior), may prove especially fruitful (Bertram et al., 2022).

LARGE-SCALE SPECIES EXTRAPOLATION OF EXPERIMENTAL DATA USING MODELING APPROACHES

One of the major sources of uncertainty during the cross-species extrapolation process is the limited availability of species-specific ADME parameters that would allow the accurate prediction of drug concentrations within target tissues (e.g., via PBPK modeling) and, in turn, the assessment of hazard and risk associated with organ-specific target engagement dynamics. This translational challenge is exacerbated by the fact that drug exposure and uptake routes for wildlife (e.g., uptake via gills and skin, ingestion via food) are generally different from the administration routes typically used in most preclinical and clinical studies (e.g., oral administration of a capsule/tablet, injection). As discussed above, recent research has led to the development of novel in vitro models to characterize drug uptake and ADME in a small number of fish species, for example, rainbow trout and zebrafish. Refining, validating, and scaling up those models should be considered a research priority. For example, high-throughput human in vitro ADME (HT-ADME) screening has become an important step in drug discovery, and the large volume of data generated with these approaches is guiding the development of AI-powered in silico tools for the prediction of ADME profiles (Davies et al., 2020; Göller et al., 2020). In line with these advancements, we foresee that medium or HT-ADME screenings could become a reality for fish species too, eliminating the need to rely on animal tests for this purpose. Furthermore, mass-spectrometry (MS)-imaging and high-sensitivity-MS applied to both whole zebrafish larvae and adult fish organs would allow the quantification, with high precision, of drug concentrations and distribution within specific tissues/organs. This, in turn, would shed light on fish-specific in vivo ADME that may be difficult to predict via in silico or in vitro methods alone (Asslan et al., 2021). Such MS-based approaches could also be used to detect changes in protein expression and more precisely reveal the distribution of a given drug target (Brunner et al., 2022; Lombard-Banek et al., 2016). These approaches have already started to be applied to ecotoxicologically relevant species groups such as daphnids and zebrafish (Perez et al., 2017; Schirmer et al., 2022), achieving relatively high levels of spatial resolution. Integrating these techniques with those discussed for the investigation of target function (e.g., in vivo imaging and gene editing) could prove incredibly powerful in identifying tissue-specific drug concentrations needed to trigger phenotypically observable target-mediated effects. This, in turn, could be used to inform the development of more predictive computational models. However, with >34 000 teleost species identified to date (of which ~50% live in freshwater; Froese & Pauly, 2022), achieving satisfactory taxonomic coverage using experimental tools is likely unrealistic. This challenge becomes even more problematic if we consider the millions of invertebrates and terrestrial species that could also be exposed to pharmaceuticals in the environment.

In parallel with the experimental approach focused on a small number of species, modeling methods should, therefore, play a major role in the extrapolation of ADME profiles across species (Davies et al., 2020; Priority Research Question 4—Textbox 1). For example, Bayesian-PBPK modeling (Krauss & Schuppert, 2016), multitask deep neural network models (Wenzel et al., 2019), and population-based ADME simulators (Wedagedera et al., 2022) are just a few examples of advanced modeling methods currently used to predict human PK profiles and estimate their variability within human populations. We foresee that comparable modeling concepts could be applied to simulate the variability of uptake and ADME data across species for ecotoxicological purposes, either from a few fish species to many or from humans to non-humans. A major obstacle to this predictive approach, however, is the limited knowledge of the key mechanistic and functional processes that determine uptake and ADME in wildlife species. Hence, any future research in this field cannot ultimately advance without an investment in fundamental biological research. This mechanistic-comparative effort should leverage the rapid advancements achieved in genome sequencing, with genome assemblies available for 3278 species distributed across 24 different phyla and 64 classes (with 684 ray-finned fish species; Hotaling et al., 2021). This vast amount of data could be used to drive the assessment of evolutionary and functional conservation of key molecular drivers of ADME processes across species. In turn, this knowledge could be integrated in existing databases (e.g., ECODrug) to guide the cross-species extrapolation of both PD and PK.

From a PD perspective, chemical-induced -omics signatures generated in a standardized manner using cell lines from a wide array of species may support comparative toxicology evaluations, including the assessment of a drug's modes of action. In the biomedical field, large-scale transcriptomic and proteomic signatures generated for thousands of genetic and chemical perturbagens (including pharmaceuticals) across many different human cell lines have unlocked novel opportunities to expand our understanding of the mode of action of marketed pharmaceuticals, supporting, for example, drug repurposing efforts (Subramanian et al., 2017). It possible to envisage that the expansion of such approaches to wildlife-derived cell lines may enable mode of action concordance studies and hazard assessment across a wide set of more relevant species groups. Recent initiatives have started to explore this avenue. For example, the EcoToxChip project is aiming at generating polymerase chain reaction arrays for wildlife (including fish, amphibians, and birds) designed to assess the effects of chemicals on 384 genes of ecotoxicological relevance (Basu et al., 2019). It is important to highlight that experimental reproducibility, inter-cell type response variability, biological coverage, phenotypic and in vivo relevance, cell metabolic competence, and data interpretation within a regulatory framework remain important challenges to overcome for the robust application of -omics signatures for chemical safety assessment within a regulatory decision-making context. Nonetheless, embedding new approach methodologies (NAMs)–based -omics approaches within a wider WoE framework may provide a valuable resource to address the challenge of predicting drug-induced effects across species.

Overcoming barriers for regulatory acceptance

Governments have evolved different chemical regulation frameworks over many years. Essentially, inherent hazard is balanced against risk to reach a point of protection for human health and the environment. Risk assessors use a wide range of information on chemicals in their evaluations, ranging from the knowledge of molecular structure to the results of complex in vivo studies. The risk assessors then inform the risk owners (e.g., governments), and chemicals are either permitted (with or without additional restrictions) or declined. This demarcation between risk assessor and risk owner is critical because it is at the heart of the barrier for regulatory acceptance. Risk assessors are scientific experts and have the appropriate skill sets to reach a conclusion on the assessment, while the risk owner is unlikely to be an individual but, rather, public and governmental authorities acting on behalf of an elected government that accept the assessor's judgment on behalf of society. In this context, key barriers to the acceptance of novel hazard- and risk-assessment methods (e.g., NAMs) concern the confidence, transparency, and domain of applicability of such methods so that the risk assessor can assure the risk owners of the safety of a chemical. That threshold for assurance is the barrier to regulatory acceptance. If an assessor has limited confidence in the data presented, then they cannot recommend acceptance to the risk owner.

In this context, cross-species extrapolation frameworks should be seen as a valuable strategy to support the overall evidence synthesis and WoE exercise underlying the safety assessment of chemicals as well as the development of ecotoxicology-relevant NAMs with high potential for regulatory acceptance. For example, comparative pharmacology and toxicology knowledge could be used to identify human in vitro assays/endpoints that can be readily extrapolated to fish without the need to develop fish-specific assays or, on the other hand, identify toxicity endpoints for which the development of fish-specific NAMs is essential (Priority Research Question 5—Textbox 1). It is important to note that the considerations discussed in the present article are specifically referred to pharmaceuticals; however, they can be adapted and applied to any chemical class, including pesticides and personal care products.

A critical action needed to bring this vision to life is the expansion and integration of databases explicitly designed to facilitate the synthesis of comparative biology data relevant for chemical risk assessment (Priority Research Question 2—Textbox 1). Novel AI-powered text mining tools could play a major role in the practical implementation of this concept. For example, the MONARCH Initiative Database (Shefchek et al., 2020) has developed a tailored data analytics pipeline to extract, annotate, and integrate a wide range of gene-to-phenotype data across species, which are presented using a user-friendly interface. Other databases have also been established with an explicit comparative vision (e.g., Comparative Toxicogenomics Database; ECOTOX Knowledgebase; Davis et al., 2022; Olker et al., 2022), while others are expanding very rapidly and effectively the integration of multiple relevant cross-species data types (e.g., ChEMBL, the US Environmental Protection Agency's CompToxChemical Dashboard). Although some expert judgment is still needed to evaluate the toxicological relevance of interspecies data and their practical implications, these examples highlight that a tremendous volume of comparative biology knowledge is already available to any user willing to implement cross-species extrapolation as part of their ERA of pharmaceuticals. Nonetheless, future research should certainly focus on increasing the effectiveness and integration of available databases with the explicit aim of supporting future chemical ERA. Another prerequisite for progress is the training of the next generation of experts with the right skill set needed to lead comparative pharmacology and toxicology research and its translation into decision-making and/or regulatory applications. Hence, we call for an active and synergistic involvement of academia, governmental agencies, regulators, and industry to support and strengthen comparative biology and toxicology education at the global level.

CONCLUSIONS

One of the most important observations that can be made concerning the last 10 years of research in the cross-species extrapolation field—applied to the ecotoxicology of pharmaceuticals—is that the tremendous advancements achieved to date have emerged thanks to the synergistic partnership between industry, government agencies, and academia. We propose that this synergy should be expanded further if we want to achieve our ambitious global protection goals for all species and not just a few. In line with our position, LaLone et al. (2021) have called for the formation of a global cross-sector collaborative consortium aimed to advance the development and implementation of cross-species extrapolation methods in regulatory toxicology. The authors have led the formation and launch of a new International Consortium to Advance Cross-Species Extrapolation in Regulation (ICACSER; https://www.setac.org/page/scixspecies). We welcome and support this initiative, and we encourage any interested reader to join the ICACSER consortium. Cross-species extrapolation is also at the heart of a new large research project, funded by the European Commission, called Precision Toxicology. This project, started in 2021, will explore the concept of “phylotoxicology,” which aims at replacing traditional animal testing with an “evolutionarily diverse model suite of organisms from multiple branches of the tree of life” (Colbourne et al., 2022). Although many research questions remain open, the progress obtained in the last 10 or so years and the rejuvenated global interest and initiatives in comparative toxicology suggest that cross-species extrapolation research will remain at the center of future ecotoxicological research. Implementing this vision in regulatory toxicology would enable us to achieve the true global protection of human and environmental health.

Acknowledgments

The present study was supported by a CRACK IT Phase 1 grant (NC/C021102/1) awarded by the UK National Centre for the Replacement, Refinement and Reduction of Animals in Research (NC3Rs) to Luigi Margiotta-Casaluci. Additional support to Luigi Margiotta-Casaluci is provided by King's College London, School of Cancer and Pharmaceutical Science. Matthew J. Winter is supported by funding from The Royal Society, the NC3Rs, the Biotechnology and Biological Sciences Research Council, AstraZeneca and the University of Exeter, Faculty of Health and Life Sciences.

    Disclaimer

    The opinions expressed herein are those of the authors only and do not necessarily reflect the opinion of the institutions to which the authors are affiliated. Stewart F. Owen is employed by and holds shares of AstraZeneca. All other authors claim no conflict of interest.

    Author Contributions Statement

    Luigi Margiotta-Casaluci: Conceptualization; Visualization; Writing—original draft; Writing—review & editing. Stewart F. Owen, Matthew J. Winter: Conceptualization; Writing—original draft; Writing—review & editing.

    Data Availability Statement

    No data were generated for the preparation of this article.