Selection of chemicals and model organisms
The model chemicals were selected to present a few key differences with respect to environmental partitioning and bioaccumulation behavior. In terms of hydrophobic partitioning to organic matter and lipid reservoirs, pyrene and PCB-153 represent a range of moderate to highly hydrophobic chemicals with log KOW values of 5.2 and 6.9, respectively. Black carbon (BC), formed through the incomplete combustion of fossil fuels, biofuel, and biomass, and emitted in both anthropogenic and naturally occurring soot, has a strong binding capacity for organic chemicals. As for bioavailability, the selected chemicals differ in their relative association with BC in that the planar aromatic structure of pyrene results in stronger association with environmental BC compared to lipid than is the case for the more bulky PCB-153 molecule (Koelmans et al. 2006). The 3rd difference relates to persistence, i.e., degradation and susceptibility to biotransformation within the animal tissues. Biotransformation of PCB-153 is found to be negligible in our test organism groups (Gobas et al. 1989; Drouillard et al. 2001; Goerke and Weber 2001; Paterson et al. 2007). In contrast, biotransformation of pyrene and other polycyclic aromatic hydrocarbons (PAHs) have been reported to be at higher rates than PCB-153 in polychaetes (Kane-Driscoll et al. 1996; Selck et al. 2003b; Jorgensen et al. 2008), fish (Namdari and Law 1996) and birds (Ronis and Walker 1989) but is unlikely to occur in mayflies (Landrum and Poore 1989).
We selected 4 groups of organisms representing different habitats: 3 aquatic (infaunal, epifaunal and/or pelagic, and pelagic) and 1 terrestrial. The aquatic organisms included deposit-feeding polychaetes, mayfly larvae (Hexagenia sp.), and fish (Perca flavescens, yellow perch). Aside from habitat, these organisms also differ with regard to biotransformation capability of organic contaminants (see above). Whereas the aquatic invertebrates have a relatively uniform diet of ingested sediment and/or seston, the yellow perch generally represents aquatic vertebrates that possess a complex diet. The aquatic species are exposed to chemicals in water and via ingested food and/or sediment. The terrestrial group includes a top avian predator represented by the little owl (Athene noctua), which is exposed to organic compounds through food web accumulation and complex diet components that include both vertebrates and invertebrates. In addition, chemical exposures via respired air may also contribute to bioaccumulation in terrestrial species (little owl). The 4 selected organism groups differ in routes of uptake (terrestrial versus aquatic, vertebrate versus invertebrate) and their ability to biotransform organic compounds (PCB-153 versus pyrene for owl, polychaete, and fish).
Probabilistic scenario studies were performed to quantify propagation of variance in the traditional bioaccumulation metrics. Using a modeling approach, our goal was to prescribe values for different chemical/physical factors (e.g., chemical KOW, chemical partitioning to proximate components of the animal, its diet, and sediment), physiological factors (e.g., organism respiratory ventilation rate, potential for biotransformation, feeding rate, digestive physiology, and so forth), environmental conditions (e.g., contaminant concentration in food items and respired media) and ecological factors (e.g., foraging range, dietary proportions, and so forth) and their associated measures of variation to develop a range of bioaccumulation scenarios encountered in nature. Indeed, modeling approaches have provided a framework for assessing and better understanding the sources of variation in bioaccumulation among various pollutants and taxonomic species (Thomann and Connolly 1984; Gobas and Mackay 1987; Luoma et al. 1992; Wang and Fisher 1999; Arnot and Gobas 2004; Luoma and Rainbow 2005; Hauck et al. 2007; Moermond et al. 2007; Wang and Rainbow 2008; Ciavatta et al. 2009; De Laender et al. 2010).
Model simulations for the 2 chemicals (PCB-153 and pyrene) and the 4 organism groups were performed using a general 1-compartment bioaccumulation model designed for organic compounds. Further details about the bioaccumulation model and its predictive algorithms can be found in Arnot and Gobas (2004). The Arnot and Gobas (2004) model was selected because of the generality of its framework, which allowed ready adaptation to the different animal species and chemicals being considered for simulation. This model has a long history, originating in its general framework from Thomann and Connolly (1984) and undergoing several iterations with advances in parameter information and predictive algorithms (e.g., Clark et al. 1990; Gobas 1993; Morrison et al. 1997). The general equation for the model is given as:
Where Corg (ng · g−1 wet wt), Cr (ng · mL−1) and Cfood (ng · g−1 wet wt) refer to chemical concentrations in the animal, respired media and ingested food, respectively. The rate coefficients are denoted as k values and for simplicity in model, simulations were assumed to follow first-order kinetics. The rate coefficients are individually defined as follows: kv is the uptake rate coefficient from respired media (mL · g−1 wet wt · d−1), kfood is the uptake rate coefficient from ingested food (g food · g−1 wet wt · d−1), k2 is the elimination rate coefficient (d−1) across respiration surfaces, kex is the fecal elimination rate coefficient (d−1), km is the metabolic biotransformation rate coefficient (d−1) and kg is the growth dilution coefficient (d−1) (Table 1). Other processes such as dermal absorption and elimination, molting losses, or reproductive losses (maternal deposition to eggs) were not included in model simulations because of lack of species-specific data and/or confidence about the degree of uncertainty associated with these parameters. Species- and chemical-specific rate coefficients were estimated on the basis of equations summarized in Supplemental Data 2.
For some organism simulations, multiple diet items (up to 3 diet items) were included in the model and/or separate contributions of respired sediment porewater and overlying water to chemical uptake. The expanded model, solved for steady state, is thus given as:
Where the terms p(o,w), p(p,w) refer to the proportion of overlying water respired and proportion of porewater respired by the animal such that p(o,w) + p(p,w) = 1. This term was substituted to a single air ventilation and air concentration term for the little owl. Qv express the ventilation rate, flow of water across gills and integument (mL/g body wt/d). Similarly for organisms having multiple diets, the proportion of a given diet to the total food ingestion is provided by p(i). Each diet has its own chemical assimilation efficiency term (Efood(i)) and the same value was assigned to the fecal egestion efficiency term (Eex(i)) for the equivalent diet item. The term Ew refers to the chemical exchange efficiency across the gills and/or integument. The organism–water partition coefficient (Korg,w), or bioconcentration factor, was estimated according to:
Where pw(org), plip(org) and pNLOM(org) are the proportions of water, lipid, and nonlipid organic matter in the animal, respectively. KOW is the n-octanol–water partition coefficient and φNLOM is the partition capacity of nonlipid organic matter in the organism relative to octanol (Debruyn and Gobas 2006).
For fecal elimination, each dietary food item contributes to a diet-specific fecal production rate based on the digestibility of proximate components in the diet as follows:
Where AEw, AElip, and AENLOM represent diet digestibility (unitless) of water, lipid, and nonlipid organic matter components, respectively, in the ingested food item. Similarly, the proximate composition of feces produced from each digested diet type results in a diet-specific organism–feces partition coefficient (Korg,ex) described below in Equation 5:
Additional modifications were performed for simulations with benthic organisms (mayfly larvae and polychaetes) to consider the effect of chemical distribution between individual sediment components: labile organic matter (LOM), black carbon (BC), inorganic matter (IM) and porewater (PW). Any chemical associated with porewater was considered bioavailable via respiratory surfaces as well as from the small amount of porewater ingested as part of sediment feeding activity. Chemicals associated with LOM, BC, and IM were treated as separate diet components with different chemical assimilation and digestibility terms. Among the latter components, only LOM was considered to be partially digested and assimilated such that it provided nutritional value to the animal. It was further assumed that animals did not engage in selective feeding of sediment components. Chemical partitioning to BC is nonlinearly related to porewater concentration and often estimated using a Freundlich equation (Koelmans et al. 2006). Assuming the chemical distribution among sediment components is in equilibrium, individual sediment component concentrations were estimated by mass balance and the difference in their relative partitioning capacities according to:
Where pLOM, pIM, and pBC represent the proportions of labile organic matter, inorganic matter (pIM and φIM), and BC, respectively. The term φIM is the partition capacity of inorganic matter in sediments relative to octanol. The term KBC is the Freundlich constant and nBC the Freundlich coefficient (Accardi-Dey and Gschwend 2002; Koelmans et al. 2006). In the uncertainty analysis for mayflies and polychaetes, uncertainty in bulk sediment concentration (i.e., LOM, BC, IM, PW) was covered by varying Cw(pw) (see discussion in Supplemental Data 2).
We used equations from 3 common bioaccumulation endpoints: BAF (L/kg wet wt; Equation 7), BMF (kg wet wt/kg wet wt; Equation 8) and BSAF (kg l.w./kg org. w, Equation 9) to assess the contribution of the various factors to the observed differences between laboratory and field measures of bioaccumulation.
In Equations 7 and 8, chemical concentrations in the organism (Corg), water (Cw), and food (Cfood) are expressed on a wet weight basis, although lipid-normalized BAF and BMFs can be calculated by dividing the BAF by the proportion of lipid (plipid(org)) in the organism or by multiplying the BMF by the ratio of the proporption of lipid in the food (plipid(food)) over plipid(org).
A 3rd commonly used bioaccumulation endpoint is the BSAF, which by convention is always reported on a lipid (organism concentration) and organic carbon (sediment concentration; Csed) normalized basis. The BSAF can be related to rate constants via the following expression:
Where pOC(sed) is the proportion of organic carbon in bulk sediments, pfood is the proportion of diet composed of nondetrital material and psed is the proportion of diet consisting of ingested bulk sediments.
Additional information on modeling BC sediment water partitioning uncertainty, modified equations for benthic feeding species are documented in Supplemental Data 2 for mayflies and polychaetes. Supplemental Data 2 further provides documentation of model parameters, parameter values, and literature sources for selected values and ranges for each of the organism–chemical simulation trials.
For each of the parameters, we examined the uncertainty in 2 steps. First, a literature search was performed to identify parameters considered as main drivers of variation. The results and motivations for this selection are provided as Supplemental Data 1. In the second step, model uncertainty and sensitivity was quantified using Monte Carlo simulations by assigning probability distributions to the most influential parameters as identified in step 1 (Supplemental Data 2). Among the different models, between 29 and 38 model inputs and parameters of the total 56 had variability associated with them. An overview of motivation, parameters and assumed parameter distributions used in probabilistic model simulations is provided in Supplemental Data 2. Monte Carlo simulations used 10 000 iterations for each organism–chemical simulation trial using Crystal Ball software (Goldman 2002) interfaced with a Microsoft Excel spreadsheet. Overall model uncertainty was evaluated by examining the distribution (mean, range, or standard deviation) and percentiles (1%, 5%, 25%, 50%, 75%, 95%, and 99%) of model output trials across simulation iterations. Model sensitivity analysis was performed to determine which parameters contribute the greatest degree of variation in model output. Although some model parameters are likely to be correlated to one another, i.e., they covary, they were treated as independent when running the model uncertainty and sensitivity analysis. Examples of potentially covarying model parameters include respiratory ventilation rates and animal feeding rates which are both scaled to animal metabolic rate (Drouillard et al. 2009), potential interactions between chemical assimilation efficiencies across respiratory surfaces and air or water ventilation rate (Drouillard et al. 2009) or chemical assimilation efficiencies from food with animal feeding rate (Drouillard and Norstrom 2003). These potential limitations of the model uncertainty analysis are expected to inflate variation in model output relative to more complex models that explicitly consider covarying model parameters as part of the probabilistic assessment. However, such influences are only likely to become important if multiple covarying parameters are found to be strong contributors to overall model variation. All results in the text are presented as mean ± standard deviation (SD).
Model validation using empirical data
Model simulation output was compared to relevant field data in order to gauge the degree to which predicted variation in simulation output matched empirically measured field variation for selected bioaccumulation metrics. Empirical data used for comparisons included published data as well as unpublished raw data from the authors or reported summary statistics of bioaccumulation metrics. Applicable empirical data sets could not be found for all organism–chemical combinations. The model trials from Monte Carlo simulations and empirical data were used to generate box and whisker plots to present 1st, 5th, 25th, 50th (median and mean), 75th, 95th, and 99th percentiles of each metric for comparisons.
For mayflies and yellow perch, empirical field data were obtained from the Great Lakes Institute for Environmental Research (GLIER), University of Windsor, Canada (Drouillard 2010; Kashian et al. 2010). This data included samples of organism concentrations or benthic invertebrate BSAFs collected from the Detroit River, Michigan, USA, and were considered consistent with model simulations for these organism–chemical trials because the simulations used Detroit River water (river-wide averages from 1998–2008) and sediment (1998 river-wide survey) (Drouillard et al. 2006, 2010). The yellow perch data consisted of 24 fish samples (total lengths ranging from 15–26 cm) collected from 4 locations in the Detroit River from 2000 to 2003. The data included lipid contents and PCB-153 concentrations in skinless fillet samples. Matched data were not available on yellow perch diets to compute field BMFs in this species and therefore comparisons were made between empirical data and model output of steady state fish concentration estimates. Yellow perch fillet sample data were converted to whole-body residue estimates by lipid normalizing the PCB-153 concentration in fillet samples and then multiplying by the assumed whole-body lipid fraction of 0.065 (Drouillard et al. 2009). The mayfly data consisted of a GLIER data set of 12 BSAFs calculated from pooled samples of Hexagenia spp. and matched sediments obtained from 9 sampling locations in the Detroit River collected during July 2008 (Drouillard 2010).
The polychaete data were obtained from Nesto et al. (2010) and Ruus et al. (2005). Nesto and colleagues reported 8 matched field BSAF estimates for PCB-153 and pyrene in the polychaete species Perinereis rullieri from 2 sampling stations in the lagoon of Venice, Porto Marghera, Italy. In the above case, BSAF metrics were reported at the 2 stations during 4 different sampling months. Ruus et al. (2005) reported laboratory bioassay-derived BSAFs (pyrene, n = 6 estimates; PCB-153, n = 3 estimates) in the polychaete species Nereis diversicolor exposed to 2 harbor sediments from Kristiansand Harbor, Norway, following 28-d laboratory exposures. Because of the limited sample numbers for PCB-153 in the laboratory bioassay data set, box and whisker plots were only generated for pyrene using both field and bioassay BSAF data.
For the little owl, an appropriate data set on BMFs in adults of this species could not be located. A surrogate data set on avian PCB BMFs for guillemots (Uria aalge) was used for comparison purposes. This is a marine species, mainly feeding on small fish and marine pelagic invertebrates (Mehlum 2001). The difference in ecology between these 2 species may hamper the comparison. The main reason for choosing the little owl as a model species is that relatively much is known about its ecology in Northwestern Europe, enabling the compilation of a case study for modeling purposes, e.g., detailed information on diet composition, prey concentrations, and so forth. For the guillemot, much empirical information is available on the derivation of the BMFs from a specific study by Lundstedt-Enkel et al. (2005), making it an excellent study to compare the modeling results, whereas key parameters for the model are relatively less well known. Despite the differences between the species, comparison of their specific accumulation is feasible, because the most important route of uptake for both species is through the diet. Hence, this comparison is based on the same process, namely, food web accumulation, which in our opinion justifies the comparison between the 2 species even though they may seem to be very different. The guillemot BMF data was obtained from Lundstedt-Enkel et al. (2005), based on measured concentrations of PCB-153 in 29 guillemot samples and 72 herring (Clupea harengus) samples. These authors used a statistical bootstrapping technique to compute randomly sampled BMF ratios from individual guillemot and prey samples, to generate a data set of 50 000 BMF combinations on which to report summary statistics. From this data set, Lundstedt-Enkel et al. (2005) reported a mean guillemot BMF of 31.2 ± 21.4, geometric mean of 24.7 and BMF range of 3.04 to 158. These summary statistics were used in conjunction with a new set of Monte Carlo simulations, assuming a lognormal distribution, to generate a surrogate empirical data set on which to compute quartiles for box and whisker plots and to compare with the little owl model simulation output.