Framework for traits-based assessment in ecotoxicology

Abstract

A key challenge in ecotoxicology is to assess the potential risks of chemicals to the wide range of species in the environment on the basis of laboratory toxicity data derived from a limited number of species. These species are then assumed to be suitable surrogates for a wider class of related taxa. For example, Daphnia spp. are used as the indicator species for freshwater aquatic invertebrates. Extrapolation from these datasets to natural communities poses a challenge because the extent to which test species are representative of their various taxonomic groups is often largely unknown, and different taxonomic groups and chemicals are variously represented in the available datasets. Moreover, it has been recognized that physiological and ecological factors can each be powerful determinants of vulnerability to chemical stress, thus differentially influencing toxicant effects at the population and community level. Recently it was proposed that detailed study of species traits might eventually permit better understanding, and thus prediction, of the potential for adverse effects of chemicals to a wider range of organisms than those amenable for study in the laboratory. This line of inquiry stems in part from the ecology literature, in which species traits are being used for improved understanding of how communities are constructed, as well as how communities might respond to, and recover from, disturbance (see other articles in this issue). In the present work, we develop a framework for the application of traits-based assessment. The framework is based on the population vulnerability conceptual model of Van Straalen in which vulnerability is determined by traits that can be grouped into 3 major categories, i.e., external exposure, intrinsic sensitivity, and population sustainability. Within each of these major categories, we evaluate specific traits as well as how they could contribute to the assessment of the potential effects of a toxicant on an organism. We then develop an example considering bioavailability to explore how traits could be used mechanistically to estimate vulnerability. A preliminary inventory of traits for use in ecotoxicology is included; this also identifies the availability of data to quantify those traits, in addition to an indication of the strength of linkage between the trait and the affected process. Finally, we propose a way forward for the further development of traits-based approaches in ecotoxicology. Integr Environ Assess Manag 2011;7:172–186. © 2011 SETAC

INTRODUCTION

Ecological risk assessment (ERA) of chemicals aims to determine the probability and extent of an adverse effect occurring in an ecological system, with the ultimate goal of protecting the long-term viability of populations, communities, and ecosystems. Because it is not feasible to test all species and chemical combinations, a major challenge for ERA is to extrapolate the population vulnerability sensu Van Straalen (1994) (see below) of different species based on the available exposure and effects data. Generally, extrapolation between species or populations is based on taxonomy; i.e., it is assumed that a given laboratory model species is representative of a broader faunal group. However, it has been recognized that ecological and physiological factors are also important in determining vulnerability to chemical stress, and that they play a key role in influencing effects at the population and community level. It was recently proposed that incorporating more information about how a given species' traits might contribute to its susceptibility may eventually permit better prediction of the potential for adverse effects to a broader range of species (Baird et al. 2008). This does not imply replacement of taxonomy-based approaches but rather building on them and improving existing knowledge and methods using a different perspective. This is of particular importance because ecotoxicological data are generally collected on the basis of taxonomy (Culp et al. this issue2010). Subsequent analysis using traits may help elaborate these data. In the near future, molecular techniques using next-generation sequencing, such as DNA barcoding (Hajibabaei et al. 2007), will facilitate the combination of species taxonomy with traits.

We propose a framework for incorporating species traits in ecotoxicology and risk assessment. We define a trait as a phenotypic or ecological character of an organism, generally measured at the individual level, but often applied as the mean state of a species. The ideas for this framework were established at the SETAC Traits-Based Ecological Risk Assessment (TERA) workshop held in Burlington, Ontario, Canada, in September 2009.

To structure the presented framework, we used the vulnerability conceptual model of Van Straalen (1994), in which the ecotoxicological effects of a chemical toxicant on a population are described in 3 categories (Figure 1):

• 1.
  External exposure: the extent to which organisms are exposed to the toxicant
• 2.
  Intrinsic sensitivity: the potential of the toxicant to affect an organism's ability to survive, develop, and reproduce; includes 2 subcategories: toxicokinetics (bioaccumulation, distribution, and transformation) and toxicodynamics (target site considerations and compensation mechanisms)
• 3.
  Population sustainability: the potential for a population to recover from any toxic effect; includes 2 subcategories: demography and recolonization

Figure 1. Categories for use of traits in ecotoxicology, after Van Straalen (1994). The categories external exposure, intrinsic sensitivity, and population sustainability mechanistically contribute to the vulnerability of a population (species). Each category can be broken down into a series of processes and subprocesses. To each category or process, groups of traits can be related.

For each of these categories, we discuss the potential application of traits-based approaches in the sections below. This is not meant to be an exhaustive review but rather an attempt to exemplify where traits could fit within the existing ERA paradigms. We evaluate a potential mechanistic approach for linking traits with vulnerability and provide potential options for the future development of the field.

EXTERNAL EXPOSURE

External exposure is the first step, or category, in the sequence that determines whether an organism will be affected by a toxicant in the environment. Clearly, in the absence of exposure, there can be no direct effect, whereas increasing levels of exposure are likely to lead to potentially growing levels of impact. After entering the environment, the distribution of the toxicant in various environmental matrices will be determined by a variety of fate processes resulting in a range of environmental concentrations in the various potential habitats in the ecosystem. This can be predicted using the properties of the toxicant and landscape characteristics by means of fate models. The traits of a species will then determine the potential likelihood and magnitude of external exposure of an organism, which will subsequently lead to an internal dose. Therefore, traits can be used to evaluate whether a species is likely to come into contact with the toxicant by describing where, when, and at what life stage exposures may occur in a contaminated habitat. Traits such as food or habitat preference are clearly the 2 key factors in determining likely exposure routes by uptake through either food or the ambient matrix (e.g., water, soil, sediment, air). In the following discussion we elaborate further on how various traits may influence external exposure.

Habitat choice

Habitat choice of a species is a major determinant of whether a species will come into contact with a toxicant. For example, soil exposure will be less relevant for species that live exclusively on vegetation or aerial species because of limited contact with the contaminated soil. Furthermore, the habitat choice of a species may also vary at different stages of the life cycle, and this needs to be accounted for when evaluating potential risks. Habitat choice has been used in a number of case studies to explain differences in exposure, for example in evaluating the potential risks of Cd for the little owl Athene noctua in a river floodplain (Kooistra et al. 2005) and Hg poisoning in the loon Gavia immer (Nacci et al. 2005). Furthermore, the duration and intensity of direct contact with a toxicant may vary with the lifespan and home range of an organism. The longer an organism lives, the greater the duration of exposure that can be expected, assuming exposure persists. An animal's home range size can have a major influence on the exposure to a toxicant if the toxicant is distributed heterogeneously throughout the habitat, because exposure varies widely from none to significant levels, but this may be of less importance for homogenously distributed toxicants for which exposure is ubiquitous and uniform. The habitats may also vary in characteristics that influence toxicant bioavailability; for example, pH and organic matter content are important variables that determine metal accumulation (Van den Brink, Lammertsma, et al. 2010). Habitat preference on a smaller scale, e.g., preference for hedgerows with lower pH versus preference for open fields with higher pH, may therefore be an important factor determining metal exposure.

Food choice

Food choice or preference can determine exposure to toxicants. As some substances accumulate in the food chain, such as Se in aquatic ecosystems and PCBs in aquatic and terrestrial ecosystems (Maul et al. 2006), predators can be exposed to higher concentrations of a toxicant, even with limited exposure through habitat. As with the trait habitat choice, food choice may vary with life stage of a species as well. If food is the main exposure route, the daily food intake determines the extent of exposure to a toxicant. This can add complications because many traits will be correlated; for example, daily food intake is related to the metabolic rate of a species and both are correlated with body mass (Nagy 2005). Other life-cycle traits, such as hibernation, aestivation, or migration, may alter external exposure through a seasonal change in habitat, which could either reduce or increase external exposure. For example, some species of mammals, birds, fish, and insects exhibit migratory behavior that results in long-distance journeys, in order to feed or reproduce in a specific habitat that may influence the amount and frequency of external exposure.

A range of other behavioral traits can also affect the external exposure of a species. Active avoidance of contaminated media or food will reduce external exposure, particularly in the face of a spatially heterogeneous contamination, and the organism can move to less contaminated areas. Avoidance behavior can be triggered by internal receptors within the context of chemotaxis and the resulting behavioral avoidance leads to decreased contact with the toxicant and therefore to decreased external exposure. Avoidance behavior has been shown for invertebrates avoiding contaminated sediment (De Lange et al. 2006), soil (Natal-da-Luz et al. 2008), or water (Schulz and Liess 2001); for fish, such as zebrafish avoiding Cu and acid mining drainage (Moreira-Santos et al. 2008); and for mammals and birds through avoidance of contaminated food (Linder and Richmond 1990).

INTRINSIC SENSITIVITY

Traits can contribute significantly to intrinsic sensitivity differences among species. These traits, with the exception of size, are not usually considered by ecologists, but they can be of significant toxicological importance. We shall summarize a range of traits that contribute to the acquisition of contaminants and the abilities of organisms to cope with chemical exposure.

Toxicants generally elicit direct effects by interacting with a certain biological compartment in the organism. This varies depending on the toxicant, leading to physiological consequences in the organism, potentially resulting in toxicity. The intrinsic sensitivity of an organism to a toxicant partly depends on the concentration of toxicant that is able to reach the target site (Meador et al. 2008). Even with external exposure, differences in traits may mean that in some species the toxicant is unable to reach the target site, whereas in others it may do so easily. Thus, internal concentrations at the target site or whole body concentrations (as a surrogate of target site concentration) may provide a better basis for linking traits to intrinsic sensitivity than external concentrations alone.

The toxicological processes of bioaccumulation, biotransformation, and the internal distribution of toxicants between organs and tissues determine the internal concentration at the target site. This has also been called the biologically relevant dose (Paustenbach 2000). Hence any trait that affects these processes may also help explain the sensitivity of the organism. Mechanistically, intrinsic sensitivity of an organism to a chemical stressor is a result of processes that can be grouped into toxicokinetics and toxicodynamics. Toxicokinetics describe the time course of uptake, distribution, biotransformation, and elimination of the toxicant, i.e., all processes that together determine the concentration of the toxicant at the target site. Toxicodynamics describe the time course of the actual toxic action at the target site, its physiological consequences, and how suborganism-level effects manifest themselves in organism-level consequences. Toxicokinetic–toxicodynamic effect models (TKTD models) were recently reviewed by (Ashauer et al. 2006; Ashauer and Brown 2008) and are valuable because they provide a methodology for linking traits to the processes of toxicity.

Toxicokinetics

Bioaccumulation

Whether considering organic or inorganic contaminants, the literature is rich with examples of sympatric species varying (sometimes by orders of magnitude) in contaminant body burdens. Intuitively, it makes sense that such interspecific differences in bioaccumulation are a function of species traits. Organisms accumulate contaminants directly from their ambient media (e.g., water, air, soil, sediments) and from their diets; morphological and physiological traits will be fundamental drivers of these bioaccumulation processes. For example, in the case of metals, biodynamic modeling approaches (Luoma and Rainbow 2005) have demonstrated that species-specific physiological traits related to ionoregulation and digestive processes drive bioaccumulation differences among taxa. Similarly, accumulation of organic contaminants has been shown to be related to organism size and lipid content (Hendriks et al. 2001), of which the latter is discussed in more detail below.

For organic contaminants in particular, body size and related surface area-to-volume relationships can exert a profound influence on the bioconcentration of contaminants. Several investigators have reported that bioconcentration is inversely proportional to the volume or weight of different species (Preuss et al. 2008). Baird and Van den Brink (2007) included dry mass as 1 of 5 characteristics that could together describe approximately 79% of the interspecific variability in sensitivity to several different compounds. Although biomass is easy to measure, it is possible that the surface area-to-volume relationships may be more powerful predictors of bioconcentration rate differences between species. However, surface area-to-volume relationships are significantly more difficult to measure than biomass, particularly for small invertebrates with complex 3-dimensional gill surfaces. In addition to size and/or body mass differences among species, the nature of body surfaces may be radically different among species. Crustaceans, for example, tend to have calcium-rich integuments; by contrast, insects may be soft and membranous or heavily armored with chitin. The means by which different integument types or biological barriers affect the diffusion rates of organic contaminants remain still poorly understood (Boudou et al. 1991), although methods do exist. Further progress through genomic and proteomic approaches will likely be made in this area in the near future.

Respiratory strategies (e.g., having surfaces such as gills) can also be important determinants of bioaccumulation. Indeed, respiratory strategy was also among the 4 key explanatory traits used by Baird and Van den Brink (2007), as described above. Buchwalter et al. (2002) found that water-breathing (dissolved oxygen–breathing) insects tend to be more permeable to water and have faster chlorpyrifos uptake rates than do comparably sized air-breathing species. Across all species in that study, water permeability was strongly correlated with chlorpyrifos uptake rates. In both respiratory strategies, body size still exerted first-order control of overall permeability. However, the use of morphological traits as predictors of physiological processes should be treated with some caution. More research is needed to identify the importance of morphological traits in relation to bioaccumulation and to develop quantitative relationships between traits and their affected processes. For example, the presence or absence of gills may be less informative than the relative permeable surface area of a water-breathing species. Intersegmental membranes in some insect species may serve as gaseous exchange surfaces and can be similarly permeable to external gills. Therefore, other characteristics, such as the degree of sclerotization, may not be good approximates for integument permeability, as some species with membranous integuments can be quite impermeable, despite common perceptions to the contrary (Buchwalter et al. 2002).

In contrast to traits that focus on the interface of the organism and its environment, the internal composition of species can vary considerably as well. Perhaps the best understood trait in relation to the bioaccumulation of organic compounds is lipid content (Hendriks et al. 2005). We make a distinction between the overall quantity of lipid (percentage lipid) and the qualities of various lipid pools at the organism level. Although the initial critical body residue concept did not explicitly consider the role of lipids in toxicity (McCarty and Mackay 1993), Di Toro et al. (2000) later identified the membrane lipid fraction (polar lipids) in the organism as the generic site of toxic action for contaminants eliciting baseline toxicity such as organic compounds, uncouplers, and inhibitors of photosynthesis or ATP synthesis (Escher and Hermens 2002; Hendriks et al. 2005).

Storage lipids, in contrast, may act as a transient sink for hydrophobic organic contaminants in organisms. The amount and composition of storage lipids within a given organism undergo dramatic seasonal fluctuations as compared with polar lipids (Naesje et al. 2006) and depend on food quality and quantity (Goulden and Place 1993), density dependence (Cleuvers et al. 1997), and life stage (Bychek and Gushchina 1999). The composition and distribution of lipids within an organism may modify intrinsic toxicity; therefore, insights into lipid dynamics and the relationship to contaminant partitioning could provide a stronger basis for understanding toxicity of tissue residues and predicting effects. Because hydrophobic organic chemicals preferentially partition into lipids, it has become common practice to normalize bioaccumulation data to the lipid content of the sample (Escher and Hermens 2002). Furthermore, the partitioning of hydrophobic contaminants in tissues follows predictable patterns. For instance, fugacity-based approaches are used increasingly to understand and predict bioaccumulation differences among species. Species' lipid content and body size are the most important traits included in quantitative bioaccumulation models (Arnot and Gobas 2004; Hendriks et al. 2005). Nevertheless, the role and influence of lipid should be interpreted with caution, particularly when comparing different studies, given the uncertainties associated with lipid measurement. For example, a variety of analytical methods using different solvent combinations and ratios generated differences in lipid concentrations for the same sample (Smedes 1999; Manirakiza et al. 2001).

Depending on the contaminant and the species, diet may be a primary exposure pathway and may therefore alter toxicity (Fisher and Hook 2002). For terrestrial vertebrates, diet is assumed to be the major source of exposure for many contaminants. Diet can also be the primary route of exposure for many metals to aquatic invertebrates (Martin et al. 2007), although it is generally accepted that dietary uptake of organic chemicals is less important than uptake from ambient media (Gomes et al. 2004). In those cases in which direct uptake from water by animals appears relatively unimportant (e.g., Se), food web dynamics may drive bioaccumulation differences among species (Stewart et al. 2004; Conley et al. 2009). In this example, both food choice and digestive processes drive bioaccumulation differences among species.

Differences in the dietary assimilation of metals appear to be profoundly different among species and diets. Dietary assimilation efficiencies of Cd in predatory stoneflies were found to range from 85% to 90% (Martin et al. 2007), whereas they could be as low as 5% in zebrafish (Danio rerio Hamilton) fed daphnids (Liu et al. 2002) and as low as 3% in silversides (Menidia sp.) fed copepods (Reinfelder and Fisher 1994). Dietary uptake models for organic contaminants in fish have been reviewed by Barber (2008), where assimilation efficiencies ranged from 40% to 100%. Assimilation efficiencies of chemicals must be distinguished from assimilation efficiencies of food components such as lipids. The extraction of lipids from food during the gut passage in fish, for example, increases the fugacity of organic contaminants in the gut, which in turn leads to further transfer of organic contaminants from the gut into the fish (Gobas et al. 1999). This provides the mechanistic explanation for biomagnification, i.e., an increase of fugacitiy in food chains (Kelly et al. 2004). Assimilation efficiencies of organic contaminants in several taxa have been reviewed by Hendriks (2001), who also established a quantitative relationship to organism body weight. Dietary uptake of toxicants (metals or organics) depends on the food choice, ingestion rate, and assimilation efficiency, as well as on the concentration within the food source, which is in turn triggered by the traits of the food source itself. Despite a growing body of literature associated with the dietary assimilation of metals in aquatic organisms, this remains a relatively difficult trait to measure with precision, because assimilation can vary with diet type, ration, and concentration.

Bioaccumulation of contaminants is a function of both uptake (directly from surrounding media and diet) and loss. Mechanistically, much less is known about the traits that drive loss rate differences among species for a given contaminant, but numerous studies have measured profound differences among species. The rapid elimination of metals by the caddis fly Hydropsyche (Cain et al. 2006) may help explain the observed metal tolerance of this genus. Furthermore, in the case of Cd in aquatic insects, it appears that the elimination capacity of species is not arbitrary, but seems to cluster phylogenetically (Buchwalter et al. 2008). Thus, more comparative studies in different faunal groups may eventually be a means of predicting the ability of a taxon to eliminate toxicants based on phylogenetic considerations. The relationship between elimination rates of organic compounds and body size has been quantified for several taxa (Hendriks et al. 2001; Kooijman et al. 2004).

Another important process that can shift elimination rates of organic compounds is biotransformation (see below). Further traits triggering elimination of a compound include ventilation rate, fecal egestion, growth dilution, and reproduction (maternal transfer) and the influence of these traits on elimination are reviewed and discussed in detail elsewhere (Barron 1990; Mackay and Fraser 2000) but will be addressed briefly in the following section.

Several other traits might be expected to play roles in differential contaminant bioaccumulation but have received significantly less attention. For example, species with high growth rates may be aided in limiting bioaccumulation relative to slower-growing species via the process of growth dilution. Slow growth rates may contribute to the relatively high bioaccumulation tendency of some freshwater mussels, for example, but high water filtration rates may be a more important trait in this regard. The ability to close the operculum in mollusks might be advantageous for avoiding exposure in the short term but is unlikely to be beneficial in chronic exposure scenarios (Cope et al. 2008). Other processes may also result in reduction of contaminant burdens in tissues. For example, the transfer of some contaminants such as Se to eggs is known to be important to fish (Coyle et al. 1993), daphnids (Lam and Wang 2006), and insects (Conley et al. 2009). For these types of contaminants, traits such as fecundity, which are known to vary tremendously among taxa, may play a modifying role. However, this trait is likely to be more important in determining population recovery and resilience (see below). Other contaminants, such as Cd and Mn, may be lost during the molting process, as was shown for a shrimp species (Keteles and Fleeger 2001) and a mayfly (Cid et al. 2010), and molting frequency can also vary widely across species.

Internal distribution

Once a toxicant is taken up into an organism, its internal distribution is driven by partitioning, either through passive diffusion or active biological transport, which emphasizes the relevance of body size as a trait. Partitioning depends on the physicochemical characteristics of the toxicant, and for organic compounds partition may be determined by the fugacity capacities of the compartments (tissues, organs) involved (Mackay 2004). Consequently, the traits lipid content, lipid distribution, and lipid composition (e.g., storage lipid, membrane lipids) will affect the partitioning processes of organic chemical stressors and, by doing so, alter the temporary concentration at the target site.

Aside from body size and lipid characteristics, the way in which basic “Bauplan” affects the distribution of contaminants is poorly understood. For example, even the fundamentals of how different types of circulatory systems, presence or absence of barriers separating the site of action from the rest of the organism, or characteristics of specific organs affect the distribution of contaminants are unknown. Once further relevant traits and their relationship to internal distribution are identified, these may be linked to pharmacokinetic models, such as the physiologically-based pharmaco-kinetic (PBPK) and adsorption distribution metabolism excretion (ADME) models (Barber 2008) or dynamic energy budget models (DEB) (Kooijman and Bedaux 1996), which are well developed for some model systems but are rarely applied to nonmodel species because of their high data demand. In contrast, processes of active biological transport may be very suitable to incorporate into these types of models. For example, processes like the sequestration of metals into cell-specific granules, e.g., cuprosomes for Cu (Joosse and Verhoef 1987) or transport of xenobiotics across membranes, which may also be inhibited by other organic compounds (Epel et al. 2008), can influence the distribution of toxicants significantly but yet are difficult to combine with available modeling approaches.

Biotransformation

Biotransformation is defined as the enzymatic conversion of a toxicant to a structurally different product with altered chemical and toxicological properties. This process is one of the major confounding factors in the prediction of toxicokinetics and toxicodynamics of organic chemicals. Birds, mammals, fish, and many aquatic invertebrates are able to metabolize a range of organic toxicants extensively (Stegeman and Kloepper-Sams 1987; Boon et al. 1997; Livingstone 1998), although this ability appears to be species-specific (Chambers and Carr 1995). Biotransformation includes direct chemical changes in the structure of the parent compound (phase I reactions) together with the conjugation of the parent compound with hydrophilic groups (phase II reactions) to facilitate excretion. Various enzymes and/or proteins are involved in these biotransformation pathways, e.g., cytochrome P450, mixed-function oxidase, metallothioneins (see also below), and glutathione-S-transferase. The presence and translation rates of all these enzymes reflect biotransformation potential and can differ even between closely related species (Rust et al. 2004). Biotransformation can lead to the formation of compounds that can be either more or less toxic than the parent, depending on the compound or enzyme combination (Perkins and Schlenk 2000). Both metabolites and conjugates have been shown to persist in several invertebrates and fish (Preuss et al. 2008), although their contribution to toxicity has yet to be determined. Biotransformation may dominate toxicokinetics; if biotransformation is faster than efflux, toxicokinetics is no longer a simple partitioning process between 2 phases (Preuss et al. 2008). Hence any trait that determines biotransformation is important for explaining the variability in intrinsic sensitivity between species and could therefore be described in general as biotransformation potential (see below). Biotransformation rates are also related to the substance's structural properties (e.g., structure, functional groups present). For contaminants that can be biotransformed, failure to address biotransformation can lead to either over- or underestimation of toxicological risk depending on the toxicity of the transformation product. Applying a traits-based approach to this topic could be done in a very detailed way, measuring the presence and amount of all possible enzymes within all species. However, this approach would come up with thousands of new traits defined by the type and subtype of an enzyme. Because this approach would likely take an unreasonable amount of time, other approaches will be necessary.

One possible alternative might be to use the biotransformation potential of a species as a trait. For fish it was demonstrated that in vitro tests can be used to determine the biotransformation potential of a species–compound combination (Fitzsimmons et al. 2007; Dyer et al. 2009). Nevertheless, confounding factors for the use biotransformation rate as traits are many, especially because biotransformation rates, just as sensitivity, strongly depend on the species–compound combination and the physicochemical conditions (Fitzsimmons et al. 2007). Before biotransformation can be used within a traits-based approach, further research in this area is needed, as well as data-mining algorithms to fill the gaps in mechanistic knowledge and to consolidate this information.

Toxicodynamics

Target sites

As described above, the effect caused by a toxicant is mediated by its mode of action (MOA) and concentration at the target site (Escher and Hermens 2002). Target sites can be polar lipids in membranes, receptors, transporters, ion channels, or enzymes, as well as DNA or intracellular systems that steer cell division (European Centre for Ecotoxicology and Toxicology of Chemicals. [ECETOC] 2007). Significant conceptual progress has been made in recent years with the establishment and classification of toxicants based on their target sites, sometimes in the absence of complete mechanisms of action (Verhaar et al. 1992, 1996; Escher and Hermens 2002; ECETOC 2007). As an example of how traits may be applied to a specific site of action, the estrogen receptor is discussed in Box 1.

Box 1. The estrogen receptor as a site of actionThe relevant toxicants for the estrogen receptor are xenoestrogens, a class of chemicals that act on the estrogen hormone system. This class of chemicals produces mostly sublethal effects at low concentrations, i.e., below baseline toxicity. Once in the tissue of an organism, the potency of these organic chemicals is directly related to their affinity, activity, and reversibility (see main body of text for explanation) at the receptor site. Ligand binding leads to activation or deactivation of the receptor, producing agonistic or antagonistic response within the biological system (ECETOC 2007). Estrogenic potency of substances can be measured with various in vitro assays, from estrogen receptor (ER) binding assay to reporter gene and proliferation assays. These assays differ in the ER used (e.g., Erα, ERβ), the organism the receptor is derived from (e.g., human, rat, fish) and the cellular surrounding (e.g., yeast cells, human, rat, or fish cell lines or primary cell culture). Binding to the ER does not necessarily provide information on the estrogenic potency of the substance as the MOA may be partially estrogenic or antiestrogenic. Estrogenic potency varies between receptor subtypes (Katzenellenbogen et al. 2000) and the organism from which the receptor is derived (Segner et al. 2003). Another important factor influencing the activity of the receptor–ligand complex is the cellular surrounding (Katzenellenbogen et al. 2000). Although xenoestrogens lead to very specific effects in vertebrates (which express the estrogen receptor), they act mostly as baseline toxicants in arthropods (which lack the estrogen receptor). However, the presence or absence of the estrogen receptor is often unknown at the species level, and, as for other sites of action, has only been investigated in a few model species.

ECETOC (2007) has reviewed the concepts and availability of data on MOAs across biota in the context of developing intelligent testing strategies for toxicants (ECETOC 2007). In principle, as soon as a biologically active molecule has reached its site of action, which in the case of agrochemicals is often fairly well defined, their interaction will initiate an effect chain, which may eventually lead to damage of tissue or system function. This toxicological effect may originate from either antagonistic or agonistic behavior of the toxicant (ECETOC 2007).

Despite the highly complex nature of toxicant–target site interactions, it may be possible to identify traits that can serve as reasonable descriptors. The mode of action approach, which links chemical types with certain target sites, provides a good start for this, although knowledge is currently restricted to a few model organisms at relatively high levels of taxonomic resolution (kingdom, phyla, and subphyla). The presence or absence of the site of action can be considered as a trait for an individual organism that will influence its sensitivity to a particular toxicant. This may be relatively easy to define if the toxicant is a pesticide or another chemical with a well-known mode of action. For a xenoestrogen or a acetylcholinesterase inhibitor, the trait modality would be presence or absence of estrogen receptor or acetylcholinesterase enzyme, respectively. Although this detailed knowledge about sites of action is only available for a few model organisms, for the short term it might be useful to assume that closely related species share these because a range of studies suggest that drug target sites are evolutionarily well conserved (Gunnarsson et al. 2008).

Intrinsic sensitivity will also be affected by the binding affinity of the toxicant to the site of action attributable to differences in steric constitutions and its reversibility (i.e., reconstitution of receptors), rate of de novo synthesis of the site of action (upregulation of gene expression), activity of enzymes, and speed of aging of the receptor. It is unclear to what extent the sites of toxic action vary in their constitution on species level; not many comparative studies have evaluated this, let alone the other complex processes described above. In the future these issues might be addressed in comparative studies using molecular techniques including genomics and proteomics.

Compensation

Numerous compensatory abilities have been described that help organisms overcome chemical-induced stress and stabilize the homeostasis of the cell. Knowledge about stress compensation in organisms relates primarily to the suborganism level; little information is available at the organ or whole-body level. At the suborganism level, mechanisms range from 1) sequestration or metabolization (covered in the Biotransformation section) of the offending agent, through 2) complex cellular machinery to combat the secondary oxidative or toxic challenge faced by the organism, to 3) repairing or regenerating damaged macromolecules, and finally 4) controlled cell death (apoptosis), which can also be considered a mechanism of compensation.

We shall briefly discuss how compensatory or coping mechanisms may fit into a traits context. Korsloot et al. (2004) describe existing stress response systems and their specificity, which is used here to give an overview of the existing knowledge base (Table 1). The cellular stress response system as a whole is a very complex network of different single specific response systems, each consisting of a complex network of interacting enzymes and messengers, leading to different gene expression profiles. Some single stress-response systems can be linked to various types of stressors, such as metallothioneins (MTs) or antioxidant capabilities for metals, although these stress systems are also induced by other stressors. The induction of MTs or metallothionein-like proteins (MTLPs) protects cellular machinery from offending metals by complexation of metals, which was shown to vary considerably across species in the extent to which metals are associated with MTs or MTLPs (Buchwalter 2008). Similarly, some faunal groups are known to store metals in extracellular granules, which are considered physiologically inert. Thus, the extent to which species can sequester their metal body burdens can be considered a highly relevant trait in determining their susceptibility. The oxidative stress response systems of animals can play a major role in ameliorating chemically induced oxidative stress. Because these antioxidant systems are highly conserved, a relatively small suite of tools is necessary to make meaningful measurements across a wide variety of taxa. Although the MTs, MTLPs, and oxidative stress responses are not entirely specific, other stress response systems, such as the basal signal transduction and the general stress protein system, are even less specific for certain stressors and therefore more difficult to apply in TERA. Similarly to the biotransformation potential described above, the ability to repair macromolecules damaged by a toxicant is an important factor that can affect the sensitivity of an organism (Marquis et al. 2009). Techniques such as the COMET assay may be a means of better understanding the susceptibility of different faunal groups to DNA damage or perhaps to compare their repair capacity. Other avoidance mechanisms such as dormancy exist, whereby the stress response leads to a change in physiology and/or behavior (Joosse and Verhoef 1987). Although a great deal of fundamental knowledge about compensatory mechanisms exists for certain model species, their variability is not well known because they are studied chiefly in model organisms. Again, comparative toxicology research programs would be needed to take what is known in these areas and apply it across different faunal groups. At this point, however, our understanding of truly toxicological traits is extremely limited.

Table 1. Stress response systems and their position in the sequence of stress compensationa

Stress response system	Involved molecules, enzymes systems, or processes	Specificity for stressors
(A) = primary; (B) = secondary; (C) = repair; (D) = apoptosis. cAMP = cyclic adenosine monophosphate; PKA = protein kinase A; MAPK = mitogen-activated protein-kinase; HSP = heat shock proteins; HSF = heat-shock (transcription) factors; ROS = reactive oxygen species; SOD = superoxide dismutase; GSH = glutathione; CAT = catalase; MTs = metallothioneins; AHH = aryl hydrocarbon hydroxylase; AhR = aryl hydrocarbon receptor; INA = ice nucleating agents; AFP = antifreeze proteins; ATP = adenosine triphosphate; ETC = electron transport chain. a Adapted from Korsloot et al. 2004.
Basal signal transduction systems (B)	Second messenger such as cAMP, Ca^2+/calmodulin, PKC, MAPK-cascade	Unspecific, all stressors
General stress protein system (B)	Heat-shock proteins (hsp70 and hsp90 families), ubiquitin, HSF	Protein denaturing stressors, quite unspecific
Oxidative stress response system (A)	ROS, SOD, GSH, CAT, and many more	Unspecific
Metallothionein system (A)	MTs	Metal stressors, very specific
Mixed function oxygenase (MFO) system (A)	Cytochrome P450, AHH, AhR, hemoproteins, several oxidases	Quite specific for organic pollutants, in insects
Cellular response system (development of tolerance), general	HSP, MT, MFO, INA, AFP	Quite specific, also natural stressors, organic pollutants
Cell aging/death response (D)	Oxidative stress responses, MAPK, ATP, ETC, Ca^2+	All stressors, especially oxidative stressors
DNA repair (C)	DNA polymerases	Genotoxic and mutagenic stressors

POPULATION SUSTAINABILITY

As outlined above, a population's vulnerability to a toxicant not only includes its intrinsic sensitivity to the toxicant but is also influenced by population-level and community-level processes (Van Straalen 1994). ERA generally aims to preserve the long-term viability of populations (Ratte 1996; Forbes and Calow 1999), which we consider to be population sustainability (Figure 1). To that end, higher-tier approaches have been developed that include semi-field or field studies at the population and community level, such as aquatic and terrestrial model ecosystem studies and nontarget arthropod and earthworm field studies. The advantage of these approaches is that they include more realistic exposure to the toxicant and many of the ecological processes that may influence the responses of a population or community to toxicant stress. This methodology has limitations, however, in that clearly not all circumstances or scenarios can be covered by such experiments (Grimm et al. 2009). Consequently, approaches that would allow extrapolation to other scenarios would be a useful development. To meet this need, ecological models have been developed to predict effects on populations and communities (Grimm et al. 2009), including simple population models and others considering food webs or metapopulation approaches. The input parameters for such models are fundamentally based on the life history of the organism (Stark et al. 2004; Rowe 2008) along with the inherent sensitivity of the various life stages, if applicable (Preuss et al. 2009). For the purpose of this discussion, we have categorized these life history attributes into demographic and recolonization traits.

Demography

Demographic traits are those that influence the population growth rate (or intrinsic rate of increase) and ultimately drive population densities and age distributions (and these are also key for recovery considerations; see below). Key demographic traits include life span and survival, generation time (the interval between reproductive events), voltinism (the number of reproductive events per y), and the number of offspring (or clutch size) per reproductive event. The combination of these traits to age-specific birth, growth, and death rates determines population growth rate (Sibly and Hone 2002); the former depends on inter- and intraspecific competition for resources, predators, parasites, and so forth, which in turn may depend directly or indirectly on population density. The importance of such traits for population growth rates and the influence of toxicants on them have been demonstrated in a large number of studies. In most population-level experiments, toxicant effects have been measured at relatively low population densities using life-table response experiments (Levin et al. 1996) and generally these have shown that population growth rate declines with increasing toxicant concentration, largely due to decreases in the survival, fecundity, and/or a delay in time to maturity of individuals. However, some studies have shown that toxicant effects may differ at high densities (Forbes et al. 2001) as a result of buffering effects when resources were limited, presumably because toxicant-induced mortality resulted in more resources for the survivors (e.g., Hooper et al. 2003). In other studies, however, toxicant effects were exacerbated when resources were scarce; for instance, populations of the rotifer Brachionus calyciflorus were 7 times more sensitive to pentachlorophenol when tested under limited rations (Cecchine and Snell 1999). These considerations have important consequences for risk assessments at the population level, because of differences between the density of test and field populations. Consequently, understanding the demographic traits of a species will be an important factor in developing a mechanistic understanding of effects and the population level, and allowing the development of models to predict effects over a range of population and environmental conditions.

Recolonization

Recolonization of a habitat can be described by several traits. First, a species must be able to reach a new habitat. This is determined by aspects such as dispersal capacity and dispersal mode. Once in the new habitat, a population then needs to be established. This involves mode of reproduction and other demographic traits described in the previous section.

Dispersal capacity describes the ability of a species to disperse to a new area. Populations of species with a low dispersal capacity have a higher risk of local extinction due to the presence of contaminants than species with a high dispersal capacity, which can easily move to colonize new habitats. Species with a high dispersal capacity will tend to have a longer dispersal distance. The distance traveled is partly related to body size; for example, active dispersal of aquatic invertebrates might occur on a smaller scale than dispersal of birds or mammals. Nonetheless, some invertebrates can travel impressive distances, such as the monarch butterfly (Altizer et al. 2000). Also, the species distribution in an area affects the potential for recolonization. Species with a patchy distribution have a smaller ability for colonization to new areas or spreading in their biotope than other species with a dense distribution. This means that suitable patches in their biotope will be empty for longer, reducing their recolonization potential. Similarly, territorial behavior limits a species' ability to move freely in available space, thereby reducing its recolonization potential. A species that does not display this behavior, on the contrary, can settle down wherever it is able to because it is not threatened by conspecifics and is able to colonize new habitats or places in its biotope more easily. Territorial behavior can occur at all life stages or only during breeding. Whether a species can actually settle in a new habitat at a certain point in time is also determined by its trophic level. Recolonization of an empty habitat follows certain arrival rules. In general, early colonizers create a more favorable habitat for subsequent arrivers (Lake et al. 2007), so lower trophic levels need to arrive first before higher trophic levels (predators) can be sustained.

Dispersal mode can be active or passive. Active dispersal is mostly used by vertebrate species but also by some invertebrates, such as butterflies. Passive dispersal is used primarily by invertebrate species, by means of water flow (hydrochory), air (anemochory) or attached to other animals (zoochory). Dispersal mode can change with different life stages. Active dispersal capacity is generally higher in adults than juveniles. Passive dispersal is sometimes best developed for resistant egg stages, such as zooplankton resting eggs (ephippia) attached to waterfowl (known as zoochorous dispersal) or earthworm cocoons transported by water (known as hydrochorous dispersal). The dispersal mode is related to the locomotion type of a species, that is, the way it moves in its environment. For example, aquatic invertebrates can be crawlers or active swimmers, with the latter offering more potential for active dispersal. Drift is a passive mode of dispersal for aquatic invertebrates. Downstream drift from upstream or tributary areas is by far the most frequently cited mechanism for recolonization within lentic systems (Wallace 1990). Higher drift can occur due to disturbance, chemical exposure, parasites, or pesticide application (Schulz and Liess 2001).

Mode of reproduction has an influence on the speed of recolonization of a new habitat. For parthenogenetic reproducing species, a single individual is enough to start a new population. For species that reproduce sexually, at least 1 male and 1 female need to be present to start a new population. A combination of a resistant stage with good dispersal capacity (ephippia, cocoon) and parthenogenetic reproduction (Daphnia, earthworm species) gives a species an advantage. In general, species that have a short generation time and produce large clutches (typical r-strategists) will be able to establish a new population quickly.

TOWARD A MECHANISTIC LINK BETWEEN TRAITS AND VULNERABILITY

Population vulnerability is the aggregated endpoint of external exposure, intrinsic sensitivity, and population sustainability (Figure 1). To address population vulnerability quantitatively, it can be broken down into a series of semimechanistic models that could predict how various traits in these categories contribute to population vulnerability. Although we do not attempt in the present study to match each of the conceptual categories in Figure 1 with corresponding mechanistic models, it is evident that a set of mechanistic models could be used to describe the main aspects of population vulnerability (Pastorok et al. 2003; Grimm et al. 2009). The advantage of these semimechanistic models is 2-fold: they divide population vulnerability into several interlinked processes and so facilitate the identification of traits that affect these processes, and they consist of equations or algorithms that quantify the effect of chemicals on organisms, populations, or communities based on process related parameters.

We illustrate a framework (Figure 2) for linking species traits quantitatively with aspects of population vulnerability described as a series of mechanistic effect model components. Figure 2 uses a simple bioaccumulation model, a process leading to intrinsic sensitivity, as an example to illustrate this link. The model equation describes the change in internal toxicant concentration (C_int) as a sum of uptake from water (uptake rate constant k_in, toxicant concentration in water C_w), and food (food intake rate constant k_f, toxicant concentration in food C_f), and assimilation efficiency (AE) as well as elimination processes (elimination rate constant k_out). Previously quantified relationships between traits and model parameters are represented in the table on top in Figure 2, using the traits lipid and size in combination with the model parameters uptake rate constant (k_in) and assimilation efficiency (AE) as illustrative examples. Parameters of mechanistic effect models are then viewed as functions of a series of species traits. Each model parameter can depend on multiple traits, and each trait can modify multiple parameters. The crucial step lies in the establishment of quantitative relationships between model parameters and traits on basis of fundamental mechanistic knowledge and empirical data, which classically involves 1) a priori hypothesis about trait–parameter relation (mechanistic link), 2) verification of the hypotheses leading to establishing the mechanistic link, and 3) empirical quantification of the mechanistic link (trait–parameter correlation). Once these relationships or links between traits and model parameters are established, the traits of a species with unknown sensitivity or vulnerability can be used to generate a semimechanistic estimate of those effects. Theoretically, the population vulnerability of a virtual species can be constructed from quantitative relationships between species traits and mechanistic effect models (lower part of Figure 2), where the virtual species stands for any of the many untested species. Similarly, these predicted effects on a species can subsequently be compared with measured data to test the established trait–parameter relationships and hence validate these traits-based models. Such studies to test and evaluate the predictive power of the traits-based approach to ecotoxicology are a prerequisite for use in a regulatory context. For bioaccumulation, chosen here, excellent examples of quantitative trait–parameter relationships exist already. For example, Hendriks et al. (2001) related toxicokinetic rate constants quantitatively to weight, lipid content, and trophic level of organisms. Model parameters for this aspect of population vulnerability can then be generated for untested species, provided that the relevant traits are known. The outlined approach for extrapolation of toxicant effects to untested species based on their trait composition implies that traits may be more suitable for this purpose when quantified on a numerical scale. The use of categorical trait descriptions results in larger uncertainty for extrapolation. Hence, wherever possible, traits quantified on a numerical scale are preferred.

Figure 2. Mechanistic link between species traits and population vulnerability using mechanistic models. The framework is illustrated using a bioaccumulation model as example (see text for details). Table at top represents hypothesized quantitative relationships between model parameters and traits. These relationships are further propagated into the models, as model parameters are viewed as functions of trait suites. Traits that modify parameter values of the model are indicated with arrows. In this way, effects on species with unknown sensitivity, but a known trait composition, so-called “virtual species,” can be predicted by a recombination of quantitative trait information and linking this to mechanistic effect model parameters.

The framework for a mechanistic link between traits and population vulnerability can be used with all kinds of mechanistic models and at different levels of biological organization. The traits described in the previous sections and in Table 2 outline the possibilities but do not claim to be a comprehensive collection of relevant traits. Another example of well-established quantitative links between traits and aspects of population vulnerability are population models. Simulation of population dynamics using individual-based models, for example, is based explicitly on species traits. These might be life cycle traits to predict population dynamics (Preuss et al. 2009) or a mix of life cycle and ecological traits to predict recovery potential (Van Den Brink et al. 2007). Currently, many activities are undertaken to promote the use of mechanistic models within environmental risk assessment of chemicals (Grimm et al. 2009). It can be expected that approaches linking traits and models will receive increasing attention in the near future, especially in view of the potential for interspecies extrapolation when linked to traits.

Table 2. Availabilitya and linkageb of traits to processes that affect population vulnerability in ecotoxicology

Vulnerability factor	Affected process	Trait	Availability	Linkage
Assignment of the categories varies greatly depending on the taxa. This analysis is designed to give a broad overview based on the experience of the authors as consensus and preliminary judgment on fish, aquatic invertebrates, aquatic plants, birds, and mammalian wildlife. a Availability: data are already in existence or can be easily acquired. Categories: low = x (minimal data, limited taxa); medium = xx (good review available, some taxa); high = xxx (database, many taxa). b Linkage: the link between the trait and affected process. Categories: x = unknown or proxy (plausible but not proven); xx = hypothesized (some evidence for some taxa); xxx = established (relationships available for several taxa).
External exposure	Likelihood and magnitude of exposure	Habitat choice	xxx	xx
		Home range	xx	x
		Food choice	xx	xxx
	Magnitude of exposure	Life span	xx	xx
		Migration	xxx	xx
		Ingestion	x	xx
		Active avoidance	x	xxx
	Likelihood of exposure	Hibernation	xx	xx
Intrinsic sensitivity	Bioconcentration	Mode of respiration	xx	xx
		Size of organism	xxx	xxx
		Surface area: volume	x	xx
		Integument permeability	x	xx
	Bioaccumulation	Assimilation efficiency	x	xxx
		Type of lipid	x	xx
		Lipid content	xx	xx
		Toxicant elimination ability	x	xxx
		Egestion rate	x	xx
		Transporters in membranes	x	xx
		Somatic growth rate (dilution)	xx	xxx
		Reproduction (transfer to offspring)	x	x
		Molting loss	x	x
	Internal distribution	Type, amount, distribution of lipid	x	xxx
		Target site distribution	x	x
		Circulatory system	xxx	x
		Sequestration	x	xxx
	Biotransformation	Biotransformation potential or conjugation	x	xxx
	Site of action	Presence of target	x	xxx
		Binding affinity	x	xx
	Compensation mechanisms	Antioxidant capabilities	x	x
		Heat-shock proteins	x	x
		DNA repair mechanism	xx	x
		Metallothioneins	x	xxx
		Mixed-function oxygenase	x	xxx
Demography	Intrinsic recovery rate	Generation time	xxx	xxx
		Life span	xx	xxx
		Voltinism	xxx	xxx
		Clutch size	xx	xxx
		Survival to reproduction	xx	xxx
Recolonization	Dispersal	Dispersal mode	xx	x
		Locomotion	xxx	x
		Drift	xx	xx
		Distance traveled	xx	x
		Territorial behavior	xxx	x
		Anadromy	xxx	x
		Trophic level	xx	x
		Resistance stages	xxx	xx
		Reproduction mode	xxx	xx

KNOWLEDGE STATUS FOR USE OF TRAITS IN ERA

Now that we have explored how and which traits can be mapped onto an established ecotoxicological framework for assessing population vulnerability (Figure 1), we are able to build further on this concept. Table 2 evaluates the availability of trait data and strength of the trait linkage to ecotoxicological processes (i.e., how well the trait and the affected process could be mechanistically linked). The list of traits presented is not exhaustive, but our intention is to provide a preliminary inventory (and we welcome future elaboration) that can be linked to the toxicological processes leading to the vulnerability of a population. From this analysis, it appears that traits linked to external exposure are mainly ecological traits such as life history, habitat, and food type-related traits. Traits that map to intrinsic sensitivity are primarily morphological and physiological. Strong linkages between processes and traits were identified for bioaccumulation and biotransformation. Data availability for bioaccumulation is good, but in most cases biotransformation data are very limited (although they have a strong mechanistic link). Internal distribution is strongly linked to sensitivity, but available knowledge (probably due to practical considerations) is limited and focused on larger organisms (e.g., mammals). Information concerning target site traits and traits describing compensatory mechanisms is lacking, even though there are some strong mechanistic linkages. Population sustainability is also influenced by a range of ecological traits with a strong focus on life history traits. Traits and population sustainability were found to have strong links between them, and in many cases large amounts of data are available.

CURRENT CHALLENGES AND THE WAY FORWARD

The use of traits in ecotoxicology offers great potential for the future development of ERA. Nevertheless we recognize that some significant challenges will have to be overcome for traits-based ERA to fulfill its potential. The broader challenges that confront the field, for instance on how to combine taxonomy-based data with traits, are also addressed in other contributions to this issue (Baird et al. this issue2010; Culp et al. this issue2010; Van den Brink et al. this issue2010). In the present work, we highlight those challenges that are specific for the field of ecotoxicology. To succeed in developing the mechanistic approaches described above, it will be important to further identify and elaborate the relevant traits (suites) for use as model parameters. An adjacent challenge will be establishing means of translating fundamental ecological knowledge on traits into a form in which it is suitable to be used within the modeling framework. For example, although a good deal of fundamental knowledge about target site and compound interactions for certain model species is often available, the suitability of such model species for making generalizations about other taxa remains unknown. It would be possible for TERA to bridge this gap, if meaningful and predictive traits for target sites' presence or absence could be formulated based on practical experimental observations and herewith establish a mechanistic link between mode of action and target sites. Similarly, developing the mechanistic approach would also call for a better integration and combination of the different effect models that can be used to estimate population vulnerability. The existing bioaccumulation, TKTD models, PBPK or ADME models and population models cited above offer a good starting point for this initiative. Modeling will also help us to prioritize which traits to quantify for a large number of species. For example, sensitivity analysis can pinpoint those traits in a given model or submodel of population vulnerability that have the greatest influence on the predicted endpoints.

The linkage, or lack of linkage, of traits through phylogeny is also an important consideration and an area for future research. Traits are not purely independent entities, so before we can arbitrarily assign organisms into groupings based on shared traits, we need to better understand their linkages to each other. For example, the need to make contact with atmospheric air results in the tendency for many air breathing aquatic insect taxa to have well developed swimming abilities as evidenced in Hemiptera. Thus, well developed swimming capacity and air breathing often co-occur in species. Similarly, certain predation styles lend themselves to higher crawling mobility, as seen in predatory stoneflies. Little is known about trait correlation with respect to contaminant susceptibility. Poff et al. (2006) use the term “trait syndromes” to describe traits that tend to cooccur and observed that different phylogenetic groups tended to occupy different “morphospace.” The tendency for closely related species to have similar traits is pervasive (Blomberg et al. 2003), and yet there are also many examples of related species being significantly different from each other with respect to a given trait. For aquatic organisms, we currently lack a fundamental understanding as to which traits are more likely to follow phylogeny and which are likely to be more labile. Poff et al. (2006) provide an excellent summary of trait lability in aquatic insects for a suite of traits associated with life history, mobility, morphology, and ecology. They further suggest that the most phylogenetically labile traits are the most informative, because they are most responsive to local selection. Buchwalter et al. (2008) provide a different perspective, by examining the extent to which physiological traits follow phylogeny. In this example, physiological traits that contribute to Cd sensitivity were compared across several insect species and found to be tightly linked to phylogeny. From an experimentalist's perspective, the consideration of phylogeny is attractive because it raises the possibility of predicting or extrapolating trait states on the basis of phylogeny and having a rational framework for testing these predictions. In contrast, mechanistic and empirical relationships between traits and processes can exist; the role of phylogeny remains unclear, as is the case, for example, for size-related traits. Therefore, a major area needed to explore the use of traits for ecological assessments is to better understand the linkages among traits both within an individual taxon and across taxa. In this area, and also for the characterization of many physiological and morphological traits, rapid progress will be made in the coming years with techniques such as next-generation sequencing and bioinformatics approaches as many genomes will gradually become available. A more detailed discussion of how this can contribute to TERA is discussed in Baird et al. (this issue2010).

For each of the relevant fields or categories that we have identified in the framework, it could be useful to establish think tanks or organized communities to elaborate further the ideas developed here and also to explore which approaches could be applied from other disciplines. For example, are there aspects of resistance management of pesticides that could assist our learning with developing our understanding of compensation mechanisms in nontarget organisms? However, “speaking the same language” is often a challenge in a multidisciplinary science like ecotoxicology, and the same will certainly be the case for TERA. We therefore see it as critical that the traits community comes together to develop an ontology that can be used to allow effective and efficient information exchange (Baird et al. 2008).

How then should we move forward with TERA? We have shown how traits could be used in ecotoxicology to enhance ERA. A number of key activities are needed to build on the ideas presented here. First, it is clear that traits-based approaches rely on the availability of good quality data for the trait in question. These data need to be stored and organized in a systematic way, so that data sharing and exchange is simplified. To this end, we recommend the development of traits ontologies and databases; further discussion of options for these is discussed in a related paper (Baird et al. this issue2010). Despite the challenges that remain, we believe that traits-based approaches have the potential to enable ecotoxicologists to develop a more mechanistic approach to understanding the reasons for differences in the vulnerability of populations to toxicants. Ultimately, this enhanced understanding should allow improvement in our ability to extrapolate between species, providing more effective risk assessment methodologies.

EDITOR'S NOTE

This is 1 of 5 papers reporting on the results of a SETAC technical workshop entitled “ Traits-based Ecological Risk Assessment (TERA): Realising the potential of ecoinformatics approaches in ecotoxicology,” held 7-11 September 2010 in Canada Centre for Inland Waters, Burlington, Ontario, Canada to evaluate the potential of traits-based ecological risk assessment among experts of different fields of biomonitoring and environmental risk assessment.

Acknowledgements

The authors thank all participants of the TERA workshop for feedback, discussion, and participation. Input from Ghent University was facilitated via the fund BOF09/24J/092.