Development and Evaluation of Aquatic and Terrestrial Food Web Bioaccumulation Models for Per- and Polyfluoroalkyl Substances

Abstract

There is a need for reliable models to predict the food web bioaccumulation and assess ecological and human health risks of per- and polyfluoroalkyl substances (PFAS). This present study presents (i) the development of novel mechanistic aquatic and terrestrial food web bioaccumulation models for PFAS and (ii) an evaluation of model performance using available laboratory and field data. Model predictions of laboratory-measured bioconcentration factors and field-based bioaccumulation factors of PFAS in fish were in good agreement with observed data as measured by the mean model bias (MB), representing systematic over- or under-estimation and the standard deviation of the MB, representing general uncertainty. The models provide a mechanistic framework for evaluating the combined effect of simultaneously occurring uptake and elimination processes and indicate food web-specific magnification of PFAS, with the highest degree of biomagnification occurring in food webs composed of air-breathing wildlife. Albumin-water, structural protein–water, membrane-water distribution coefficients, and renal clearance rate are among the most important model parameters. With further development and testing, these models may be useful for future PFAS screening and risk assessment initiatives and advance bioaccumulation studies of PFAS by providing a mechanistic framework for PFAS bioaccumulation.

Short abstract

Food web bioaccumulation models for predicting PFAS in aquatic and terrestrial food webs were developed and evaluated.

Introduction

In the nearly 20 years since the first report of the widespread global distribution of perfluorooctanesulfonic acid (PFOS) in wildlife,¹ numerous studies have been conducted to assess the bioaccumulation of PFOS and other per- and polyfluoroalkyl substances (PFAS). PFAS have been detected in environmental media and biota from various regions, including the Arctic,²⁻⁵ Great Lakes,⁶⁻⁸ Baltic and Mediterranean,⁹ North Atlantic Ocean and Gulf of Mexico,¹⁰⁻¹² Antarctic¹³ and Asia.¹⁴⁻¹⁸ Several studies have been conducted to assess the occurrence and bioaccumulation of other PFAS of concern including investigations of perfluoroalkyl phosphonic and phosphinic acids (PFPAs and PFPiAs), perfluoroether carboxylic and sulfonic acids (PFECAs and PFESAs).¹⁹⁻²²

A recent review has highlighted the lack of models for assessing PFAS bioaccumulation.²³ To better understand the biological significance of local and global PFAS contamination, there is a need for models that can predict the bioaccumulation and exposure of PFAS in wildlife and humans. Existing bioaccumulation models for nonpolar lipophilic contaminants (e.g., polychlorinated biphenyls, PCBs) are routinely used for substance screening and prioritization, development of environmental quality guidelines and site-specific risk assessments.²⁴ These models rely on the chemical octanol–water partition coefficient of the neutral species (K_OW,n) and the octanol-air partition coefficient of the neutral species (K_OA) because lipophilic substances predominantly bioaccumulate in the lipids of organisms and these properties well represent the lipid partitioning of lipophilic substances.²⁵

K_OW,_n and K_OA,_n have largely been ignored in assessments of PFAS bioaccumulation. PFAS are generally oleophobic and do not readily accumulate in lipid-rich tissues, but rather are distributed in biofluids and tissues containing proteins and phospholipids (e.g., plasma, liver, kidney and brain). The majority of well-studied PFAS including PFOS and other perfluoroalkyl acids (PFAAs) are anionic, hence exchange between biota and air is negligible. This makes K_OA,n not only a virtually impossible property to measure, but also of little significance as respiration to air is essentially a nonexistent route of elimination for air-breathing organisms. Protein–water and membrane-water distribution coefficients (D_PW and D_MW) are important properties for PFAS bioaccumulation, as these compounds exhibit a relatively high affinity for albumin and transporter proteins and phospholipids.^3,26−28 Partition coefficients of PFAS in these biological phases (e.g., K_PW,i and K_MW,_i) can be effectively measured in vitro using validated methods such as equilibrium dialysis and solid-supported lipid membranes.²⁹⁻³³ In silico models for estimating chemical biopartitioning are also available.³⁴

Armitage et al.²⁷ presented a general model to predict the bioconcentration of ionizable organic compounds in fish. The model utilizes pH-dependent octanol–water distribution coefficient (D_OW) to represent partitioning in neutral lipids and D_MW to represent the phospholipid–water distribution coefficient. The model also utilizes a nonlipid organic matter-water distribution coefficient (D_NLOM-W) to characterize partitioning into organism proteins. Sun et al.³⁵ recently developed and tested a similar model for predicting bioaccumulation of PFAS in fish. A key feature of this model is the incorporation of renal elimination, which has been shown to be an important pathway for some PFAS.³⁶⁻³⁹

In the present study, we expand on these efforts by modifying existing food web bioaccumulation models,^24,25,40,41 for the purpose of simulating bioaccumulation of a wide-range of PFAS in both aquatic and terrestrial food webs. Specific objectives are to (i) modify equations and parameters of existing mechanistic food web bioaccumulation models for application to PFAS, (ii) apply the models to predict PFAS concentrations and bioaccumulation metrics in laboratory and field environments and (iii) evaluate model performance and model bias. We further utilize the models to assess food-web specific bioaccumulation behavior and evaluate the influence of various physical-chemical, biological and environmental factors. The purpose of these models is to serve as a mathematical description of some of the current understanding of the mechanisms of bioaccumulation of PFAS in biota and a tool for predicting concentrations of PFAS in organisms of aquatic and terrestrial food webs.

Theory

Food Web Bioaccumulation Models for PFAS

A mass balance approach, which has proven to be a useful mechanistic template for predicting concentrations of nonpolar lipophilic contaminants in biota,^24,25 is used to derive bioaccumulation models for PFAS in aquatic and terrestrial food webs. Figure 1 presents a conceptual diagram of the components and processes considered for water-respiring and air-breathing organisms in aquatic and terrestrial food webs. Equations 1 and 2 are used to calculate steady-state PFAS concentration in water-respiring and air-breathing organisms, respectively.

Figure 1.

Schematic illustration of a food web bioaccumulation model for predicting contaminant concentrations in aquatic and terrestrial food webs comprised of various water-respiring and air-breathing organisms.

Water-respiring organisms:

1

Air-breathing organisms:

2

C_WD,C_A,n, C_D, C_B are the chemical concentrations (mol/m³) in water (freely dissolved), air (neutral molecules only), diet and the organism, respectively. The respiratory uptake rate constant and respiratory elimination rate constant (1/d) for water-respiring organisms are represented as k_W1, k_W2, respectively. Respiratory uptake and respiratory elimination rate constants (1/d) for air-breathing organisms only account for the organism-air exchange of neutral (nonionized) PFAS, represented as k_A1,n, k_A2,n respectively. Φ_n is the fraction of neutral species of PFAS at the pH of the organism, determined by the Henderson–Hasselbalch equation. k_D, is the dietary intake rate constant. k_F, k_RENAL, k_BIL, k_MAT, k_MET are rate constants for elimination via fecal, renal, biliary, maternal transfer and metabolism, respectively. Also, for this steady-state solution, the influence of organism growth can be represented by as growth rate constant k_G, which is determined as change in body weight (W_B) and hence volume (V_B) with respect to time, represented as k_G=dV_B/(V_B · dt).

k_RENAL is governed by an estimated renal clearance rate (CL_R mL/d), which can be derived from glomerular filtration rate of the organisms (GFR, mL/d), the fraction of unbound chemical in plasma (f_u), and the degree of net reabsorption (Abs%) or net secretion (Sec%). k_MET,PRE·C_B,_PRE was included to account for the possible formation of certain PFAS due to biotransformation of precursor compounds. k_MET,PRE is the biotransformation rate constant (1/d) of the precursor and C_B,_PRE is the concentration of the precursor in the organism (mol/m³). A summary of model equations and procedures used to estimate rate constants is provided in the Supporting Information (Sections S1 and S2). Table S1 is list of all model parameters, including symbols, definitions and units.

For nonpolar lipophilic substances, k values are often derived from K_OW,n or K_OA,n.^24,25 For PFAS, this approach does not work because octanol does not adequately represent the partitioning properties of the biological media and tissues in which PFAS accumulate. However, k values for PFAS can be derived using measured or estimated distribution coefficients, accounting for both the neutral and ionic species at the relevant pH, in biological media which PFAS do appear to accumulate in.

Overall, we use six different distribution coefficients to represent equilibrium partitioning of a given PFAS in organisms and derive k values. These include distribution coefficients for albumin-water (D_ALB-W), transporter protein–water (D_TP-W), structural protein–water (D_SP-W), neutral lipid–water (D_NL-W), phospholipid–water (D_MW), and carbohydrate-water (D_CW). Protein- and lipid specific distribution coefficients are included in the models because partitioning of organic acids into different types of proteins and lipids can vary substantially. For example, partitioning of PFAS into plasma proteins (represented by D_ALB-W) is much greater than that in structural proteins (i.e., D_ALB-W ≫D_SP-W).⁴² Similarly, partitioning of anionic PFAS into phospholipids is much greater than into neutral lipids.^33,43D_TP-W is used to represent partitioning into fatty acid-binding proteins, which may play an important role in PFAS toxicokinetics.⁴³⁻⁴⁵ Distribution coefficients for carbohydrates were included to represent the diet of herbivores. The organic carbon–water distribution coefficient (D_OC) is used in the models to represent partitioning into sediments and soils. It is important to note that anionic PFAS are generally water-soluble, exhibiting both hydrophobic and oleophobic properties and hence tend to demonstrate surface active behavior and partition between media interfaces.⁴⁶ However, distribution coefficients of PFAS used in the present study generally represent equilibrium partitioning into bulk media (e.g., proteins, phospholipids). Also, these substances can aggregate and form micelles or mixed-micelles, generally at high concentrations.^47,48 Micelle formation was not considered in the present modeling study of ecosystems exhibiting relatively low PFAS concentrations.

To simplify and avoid the need for users to search for several distribution coefficients that may be hard to find or not exist, we developed correlations between the various distribution coefficients and log K_OW,n of the PFAS, which can be readily calculated (Figure 2). It is important to stress that while log K_OW,n can be a useful parameter for estimating distribution coefficients, it is the distribution coefficients, which tend to differ greatly from K_OW,n, that drive the models.

Figure 2.

Plots showing the relationship between the octanol–water partition coefficient of neutral species (Log K_OW,_n) and (A) measured protein–water distribution coefficients (D_ALB-W), membrane-water distribution coefficients (D_MW), and structural protein–water distribution coefficients (D_SP-W) and (B) gill assimilation efficiency and dietary assimilation efficiency values (E_W, E_D). Regression equations and r² values generated using Excel (Microsoft Corporation, version 16.85): Log D_ALB-W = 0.4469 × Log K_OW,n + 1.9362 (r² = 0.6459); Log D_MW = 0.9117 × Log K_OW,n – 1.1855 (r² = 0.8707); Log D_SP-W = 0.9124 × Log K_OW,n – 2.8604 (r² = 0.9312); Log D_OC = 0.613 × Log K_OW,n – 0.6567 (r² = 0.9312). The measured values of D_ALB-W, D_MW, D_SP-W include observations for C6–C11 PFAAs, ADONA, Gen X, 9Cl-PF3ONS and PFECHS. Measured values of D_OC include observations for C6–C11 PFAAs, FOSA, N-MeFOSAA and N-EtFOSAA. See Table S5 for more details. Empirical data used to generate these relationships with K_OW,n comprise measurements for a relatively small number of PFAS, thus extrapolations for PFAS with K_OW,n values outside this range is not recommended.

The distribution coefficients are used to estimate the biota-water distribution coefficient (D_BW) and biota-gut distribution coefficient (D_BG):

3

and

4

where S_B, S_W, S_F are the chemical’s sorptive capacity or “solubility” in biota, water and gut digesta, respectively. C_{B_Eq}, C_{WD_Eq}, C_{G_Eq} are equilibrium concentrations of the chemical in biota, water and gut digesta, respectively; and ν represents the volume fraction (m³/m³) of the various constituents in biota and gut digesta (assumed equivalent to excreted feces). The fraction (ν) of lipids, proteins, carbohydrates and water in the digesta/feces is determined from the composition of the diet (i.e., ν_NL,D, ν_PL,D, ν_ALB,D ν_TP,D, ν_SP,D, ν_C,D and ν_WD) and corresponding assimilation efficiencies of the various constituents (i.e., ε_NL, ε_PL, ε_ALB, ε_TP, ε_SP, ε_C, ε_W).

Figure S1 illustrates this approach for estimating PFAS solubility (i.e., sorptive capacity) in specific tissues/compartments of a fish (e.g., S_PLASMA, S_LIVER, S_MUSCLE), as well as the whole-organism (S_B). Specifically, S values for tissues and whole organisms are estimated using the estimated distribution coefficients (e.g., D_NL-W, D_MW,D_ALB-W, D_TP-W, D_CW) and volume fractions of the different constituents (e.g., ν_NL,, ν_PL,, ν_ALB-W, ν_SP, ν_C, ν_W).^49,50 Allendorf et al. recently used a similar approach, utilizing distribution coefficients and tissue composition for estimating the equilibrium distribution of PFAS in mammals.⁴³ Estimated S values for PFOS in specific tissues/compartments and whole fish are shown in Table S2. Table S3 shows estimated solubility values of PFHxS, PFOS, PFOA, PFNA, PFDA, PFUnA, PFDoA and FOSA in water, sediments, whole fish, fish muscle/fillet, plasma, eggs and liver.

Respiratory exchange in air-breathing animals only occurs for neutral molecules, as ionized species are not available to partition into the gas phase. The organism-air partition coefficient of neutral molecules (K_BA,n) is determined as

5

where S and K values are for neutral PFAS species only. Respiratory elimination in air-breathing organisms is assumed to be negligible for the majority of PFAS, which are typically anionic at environmental and biological pH. While respiratory elimination of ionized PFAS appears to have little influence on the mass balance in the organism, it has an important influence on the bioaccumulation behavior of ionizable PFAS because a key elimination route (i.e., respiration) is not accessible for these substances.

Thermodynamic Evaluation of Bioaccumulation

To provide a thermodynamically valid reference point for the occurrence of biomagnification or food web bioaccumulation of PFAS, it is useful to view bioaccumulation factors as ratios of chemical activities, recognizing that an increase in chemical activities of PFAS with trophic level indicates biomagnification, while a decrease in chemical activities of PFAS with trophic level indicates trophic dilution^51,52 The use of chemical activities is especially important for ionic substances because of their high sensitivity to environmental factors (e.g., pH, ionic strength, temperature). At most environmental concentrations, the chemical activity a (unitless) of a PFAS in biota or other environmental media can often be approximated by the chemical concentration (C, mol/m³) and solubility or sorptive capacity (S, mol/m³), i.e. a = C/S × F, where F (unitless) is the fugacity ratio.^49,50 For many PFAS, the sorptive capacity of the PFAS salt can be an appropriate reference point for the calculation of chemical activity, while the melting point of the salt can serve to determine F following Walden’s Rule. Further details of the chemical activity approach, including rationale, assumptions, benefits and limitations are provided in the Supporting Information (Section S7).

A chemical-activity-based biomagnification factor (BMF) is determined as the ratio of the PFAS activity in the predator (a_B) and its diet (a_D), i.e., a_B/a_D. A predator-diet chemical activity ratio >1 indicates the occurrence of biomagnification while a ratio <1 indicates trophic dilution. As with existing models representing nonpolar lipophilic substances, food digestion and absorption,⁵³⁻⁵⁵ are important components of the current PFAS bioaccumulation model as they determine the maximum biomagnification potential (BMF_MAX) in the organism:

6

where G_D and G_F are the feeding rate and fecal excretion rate and S_D and S_F are sorptive capacities in the diet and gut digesta, respectively. The predicted BMF may ultimately be lower than BMF_MAX, which is only approached if elimination rates via respiration, renal clearance, biliary, metabolism and reproduction (i.e., k_R2, k_RENAL, k_BIL, k_MET, k_MAT) are small.

The developed models generate a predicted activity-based trophic magnification factor (TMF), derived from the antilog of the slope (10^m) of a log–linear regression of model predicted chemical activities in the various organisms (a_B) and corresponding trophic level (TL) values, i.e, log₁₀ (a_B) = m× TL + b, where m and b are the slope and intercept, respectively.

Methods

Model Application and Parameterization

For the present study, the models were programmed in Excel spreadsheets (Microsoft Corporation, version 16.85). The models were applied to simulate PFAS bioaccumulation in (i) fish under laboratory and field conditions⁵⁶⁻⁵⁸ (ii) Lake Ontario food web^6,59 and (iii) Canadian Arctic terrestrial and marine mammalian food webs.^3,5,60 Modeling of PFAS bioaccumulation in fish included simulations of PFAS uptake and elimination behavior in laboratory-exposed rainbow trout (Oncorhynchus mykiss).^56,57 Further, we compared predicted BCFs with all available laboratory BCF data presented by Burkhard,⁵⁸ which includes data for a wide range PFAS in different fish species. Also, we compared predicted BAFs of PFAS in a hypothetical lower trophic and upper trophic fish species with BAF values reported by Burkhard,⁵⁸ which includes data for fish of varying trophic levels. Reported concentrations of PFAS in water, sediment, soil, vegetation or biota were used as input values are shown in Table S4 in the Supporting Information. Model predicted concentrations in organisms (C_B) in units of ng/g and chemical activities in organisms (a_B, unitless), as well as bioaccumulation metric values (BCFs, BMFs, BAFs, half-lives) were estimated and compared to observed values reported in the original studies. Lastly, we conducted a series of simulations of PFAS bioaccumulation via direct uptake coupled with bioformation via biotic transformation of known precursor compounds such as N-Ethyl perfluorooctane sulfonamide (N-Et-FOSA) and bis(perfluorohexyl) phosphinic acid (C6/C6 PFPiA). See the Supporting Information for more details.

Physical-chemical properties of various PFAS are shown in Table S5. Figure 2 shows relationships between K_OW,_n and measured distribution coefficients (D_NL-W, D_MW, D_ALB-W, D_SP-W and D_OC), as well as dietary assimilation efficiency (E_D) and gill assimilation efficiency (E_W) values. Measured values of D_MW,^31,33D_ALB-W,^29,30,32 and D_SP-W(42,43) and D_OC(61) have been reported for some PFAS. For other PFAS, D_MW, D_ALB-W, D_SP-W and D_OC values were estimated from the correlations of measured data with calculated K_OW,_n. E_D and E_W values were also estimated based on relationships with K_OW,_n. Organism properties and process rates are shown in Table S6. Food web structure and predator–prey relationships of the studied food webs is shown in Table S7.

Model Performance and Sensitivity Analyses

To assess model performance, we compared model predicted bioaccumulation metrics of PFAS (e.g., BCF_, BAF, BMF) or predicted concentrations of PFAS in organism (C_B) to corresponding observed values. Following previous studies,²⁴ we calculated mean model bias (MB) as the geometric mean of the ratio of predicted and observed values (see Section S6 in the Supporting Information). The MB value provides a measure of systematic overprediction (MB > 1) or underprediction of the model outcomes (MB < 1). Further, the standard deviation (SD) of the mean MB was determined to provide a measure of model performance. The SD combines uncertainty introduced by a number of experimental and model related factors, including error in the observed data used for model comparison. A sensitivity analysis was also conducted to assess the % change in model output corresponding to a 10% change (increase) of a given model parameter.

Results and Discussion

Model Performance

The model was found to reasonably predict the time-course of PFAS uptake and depuration in laboratory-exposed rainbow trout (Figure S2 and Table S8). The predicted whole-body half-life (t_1/2, d) of PFOS in rainbow trout exposed via water and diet were 12 and 8 d, respectively. The observed PFOS t_1/2 values in these experiments were 15 ± 2.0 (SE) and 13 ± 1.8 (SE) d, respectively.^56,57 Model predicted BMFs (C_B/C_D, kg/kg) were also in good agreement with observed data (Figure S3 and Table S8). The predicted whole-body BMF (C_B/C_D, kg/kg) of PFOS in rainbow trout was 0.20, close to the reported value of 0.32 ± 0.05 (Table S8). The overall mean MB for BMFs of the seven studied PFAS was 0.75 (1 SD range: 0.38–1.5), indicating an average 25% underestimation of observed values and model uncertainty equivalent to approximately a factor of 2 (i.e., 1.5/0.75), as characterized by the SD of the MB. Model predicted BCFs (C_B/C_WD, L/kg) were also in good agreement with observed BCFs in rainbow trout (Figure S3 and Table S8). The predicted BCF of PFOS in carcass, liver and blood of rainbow trout following aqueous exposure were 930, 2,390 and 1,740 L/kg, respectively, close to those reported BCFs of 1,100 ± 150, 5,400 ± 860 and 4,300 ± 570 L/kg (Table S8).

Figure 3 shows model predicted whole-body BCFs versus all observed BCFs for several PFCAs, PFSAs, F53-B and HFPO-DA.⁵⁸ Predicted whole-body BCFs ranged between 0.5 for HFPO-DA and 74,000 L/kg for PFTeA. The mean MB for the nine PFAS evaluated was 0.98 (SD range: 0.31–3.10). Predictions of BCFs in specific tissues (liver and muscle), were also in good agreement with the available BCF data. The mean MB for the BCFs of nine PFAS evaluated in liver and muscle were 0.61 (SD range: 0.11–3.22) and 0.98 (SD range: 0.13–7.40), respectively.

Figure 3.

Model predicted versus observed bioconcentration factors (BCFs, L/kg, wet wt.). Observed data are from Burkhard.⁵⁸

Model predicted BAFs in lower trophic and upper trophic fish of a hypothetical aquatic food web were compared to field-based BAFs (Table S9). Model predicted BAFs ranged between 0.03 for PFBA and 6.2 × 10⁶ for PFTeA. Model predicted BAFs of PFAS are generally consistent with field observations. For example, predicted BAFs of PFOS in lower trophic and upper trophic fish were respectively 3.27 and 3.67, which are comparable to observed values (3.55 ± 0.83).

Model predicted concentrations of PFOS in organisms of the Lake Ontario food web were in reasonable agreement with observed concentrations (Figure 4). Predicted concentrations of PFOS (ng/g wet wt.) in Lake Ontario Mysis (8.8), Diporeia (70), alewife (21) smelt (71), sculpin (87), lake trout (108) and herring gull eggs (262), were close to observed concentrations in those organisms, with an overall mean MB of 0.5 (1 SD range: 0.27–0.81), (Table S10). On average, concentrations of PFOS in the various organisms of the food web are underestimated by the model by 50%. The upper standard deviation of the mean MB is an underestimation by 19% and the lower standard of the mean of the mean MB is an underestimation by 73%. For other PFAS in this food web, the mean MB values ranged between 0.21 and 0.35. This model performance is consistent with Sun et al.,³⁵ which taken together demonstrate reasonable confidence in these models for use in risk assessments and/or regulatory purposes.

Figure 4.

Model predicted versus observed concentrations (ng/g wet wt.) of individual PFAS in organisms of (A) Lake Ontario food web, (B) Canadian Arctic marine food web, and (B) Canadian Arctic terrestrial food web. Error bars represent 95% confidence intervals of the observed data.

Model predicted concentrations (ng/g wet wt.) of PFOS in Arctic marine wildlife (e.g., eider ducks, white-winged scoters, beluga whales, ringed seals, polar bears) were close to observed concentrations, with an overall mean MB of 0.62 (1SD range: 0.32–1.2), (Figure 4 and Table S10). The results indicate that on average, the model underpredicts PFOS concentrations in Arctic marine biota by approximately 40%. Model predicted concentrations of PFOA, PFDA and PFUnA in Arctic marine biota were also generally comparable to observed values, with overall mean MB values ranging between 0.26 to 2.3. Mean MB for PFNA predictions in the Arctic marine food web was 4.1 (SD range: 1.4–12). Facilitated renal elimination of PFNA via organic anion transporter proteins in these wildlife species is a possible reason for overpredicting PFNA concentrations in this food web. In the present model, PFNA was assumed to exhibit a high degree of renal reabsorption in birds and mammals (Abs% = 99%). Spatial variation in concentrations in water and dietary sources may be another factor affecting the concentrations of PFAS in these highly mobile animals.

A comparison of model predicted and observed concentrations in the Canadian Arctic terrestrial food web is shown in Figure 4c. Field studies of PFAS in the lichen-caribou-wolf food chain show large variations in the concentrations of PFAS among lichen and other vegetation (Table S4). Geometric mean (95% CI) of PFOS concentrations in lichen and vascular plants (caribou diet) were respectively 0.013 ng/g dry wt. (95%CI: 0.002–0.025) and 0.062 ng/g dry wt. (95%CI: 0.002–0.12). Based on these input values, the predicted concentrations of PFOS in caribou liver was 0.36 ng/g (95% CI: 0.01–1.4), close to those observed in field collected liver tissue of caribou (0.67, 95% CI: 0.38–0.96). Model predicted PFOS concentrations in wolf liver (3.8 ng/g, 95% CI: 0.2–13.8) were approximately 2.7 times higher than observed concentrations (1.4 ng/g, 95% CI: 0.63–2.2), but the wide 95% CI of the predicted PFOS concentrations in wolf liver tissue (due to the wide variation in PFOS concentrations included in base input concentrations for tundra vegetation) are consistent with the observed concentrations. Predictions of other PFAS were generally consistent with observed data. With the exception of PFUnA, the overall mean MB for all PFAS predictions in caribou and wolves were 0.26 and 1.21, respectively. The model underpredicted PFUnA concentrations in caribou by over a factor of 10, which may be due to the large variation in PFUnA concentrations in terrestrial vegetation.

Sensitivity Analysis

The model parameters that have the most influence over the predicted BAF of PFOS in aquatic water-respiring organisms are related to dietary assimilation efficiency and feeding rate, gastrointestinal extraction efficiencies of proteins and lipids, (E_D, G_D ε_PL,ε_SP,,ε_ALB), gill ventilation rate and assimilation efficiency (G_W, E_W) and distribution coefficients (D_MW, D_ALB-W, D_SP-W), (Figure S4a). BAFs in aquatic organisms are also sensitive to changes in parameters related to the renal elimination (GFR, Sec%, Abs%). For air-breathing organisms, the most important model parameters affecting the predicted BMF of PFOS include those related to dietary assimilation efficiency and feeding rate, gastrointestinal extraction efficiencies of proteins and lipids (Figure S4b). The results also show model predictions of PFOS in air-breathing organisms are very sensitive to parameters related renal elimination (GFR, Sec%, Abs%).

Modeling results demonstrating the influence of body size, feeding rate and exposure conditions on PFOS bioaccumulation potential in aquatic organisms are shown in Table S11 in the Supporting Information. Also, the influence of model parameters governing renal elimination on predicted bioaccumulation metrics (BCFs, BMF, elimination half-lives) for PFOA and PFHxS in fish and mammals are shown in Table S12 in the Supporting Information. In the present study, model predictions of PFOA and PFHxS in fish are based on the assumption of net renal secretion, with %Sec values of 200% and 500%, respectively. These values correspond to k_RENAL/k_TOTAL values of 0.92 and 0.99, which is consistent with previous observations and estimates.^35,38,39 Model predicted BCFs of PFOA and PFHxS under these assumptions are 7.4 and 3.6, respectively, substantially lower than those assuming high renal reabsorption (Abs% = 99%), which results in BCFs equal to 69.7 and 135, respectively (Table S12). In the present study, PFOA and PFHxS were assumed to exhibit net renal reabsorption (Abs% = 90%) in birds and mammals. Model simulations for a 60 kg mammal indicate the BMF of PFOA and PFHxS with this level of renal reabsorption are 0.7 and 0.5, respectively. Predictions based on a high degree of renal reabsorption of PFOA and PFHxS (e.g., Abs% = 99%) results in biomagnification (BMF > 1), while predictions based on low renal reabsorption or net renal secretion results in BMFs < 1 (Table S12). PFAS renal clearance rates and elimination half-lives are highly variable between species and sex.³⁷ In the current model, estimates of Abs%, Sec% and k_RENAL for PFAS in different organisms are highly uncertain and those model predictions should be regarded with caution.

Food Web Biomagnification Behavior

Table 1 shows predicted activity-based trophic magnification factors (TMF) of legacy PFAS, other PFAS of concern, as well as several lipophilic persistent organic pollutants (POPs). With the exception of FOSA and C6/C6 PFPiA and C6 PFPA, predictions assume the substance is not metabolized. In the present study, PFOA, PFHxS, HFPO-DA, DONA and C6 PFPA are assumed to exhibit net renal secretion in fish (Sec% = 500% for PFHxS, Sec% = 200% for PFOA, HFPO-DA, DONA and C6 PFPA), but are assumed to exhibit net renal reabsorption in birds and mammals (Abs% = 90%).

Table 1. Model Predicted Trophic Magnification Factors (TMFs) of Legacy PFAS, Other PFAS of Concern, and Neutral Lipophilic Persistent Organic Pollutants.

	CAS	LogKOW,nc	LogKOA,nc	LogDNL-Wc	LogDMWc	LogDALB-Wc	LogDSP-Wc	model predicted TMF in terrestrial food web	model predicted TMF in aquatic piscivorous food web	model predicted TMF in marine avian/mammalian food web
Legacy PFAS
PFHxSa	355-46-4	5.2	7.6	1.8	4.0	4.9	1.9	0.3	0.2	0.5
PFOS	1763-23-1	6.4	8.1	2.8	4.9	4.8	3.0	5.8	1.4	3.0
PFOAa	335-67-1	5.3	7.2	2.0	3.5	4.2	2.0	0.6	0.4	0.7
PFNA	375-95-1	5.9	7.5	2.5	4.3	4.5	2.5	5.0	1.3	2.7
PFDA	335-76-2	6.5	7.8	2.9	4.8	4.9	3.1	5.5	1.4	2.9
PFUnA	2058-94-8	7.2	8.1	3.4	4.5	4.7	3.7	7.0	1.3	3.2
PFDoA	307-55-1	7.8	8.4	3.8	5.9	4.9	4.2	7.4	1.6	3.6
FOSAb	754-91-6	5.6	7.6	4.8	3.9	4.4	2.3	0.1	0.6	0.3
Other PFAS of Concern
HFPO-DAa	62037-80-3	4.2	6.4	0.3	2.4	3.4	1.0	0.3	0.2	0.4
DONAa	958445-44-8	4.9	7.2	1.7	3.1	4.2	1.6	0.4	0.4	0.6
F-53B	73606-19-6	7.0	8.4	3.9	5.7	5.1	3.6	6.7	1.7	3.7
PFTECA	329238-24-6	6.6	8.0	3.5	5.3	4.9	2.0	5.7	1.5	3.1
PFECHS	335-24-0	5.1	7.8	1.3	4.5	4.7	1.8	3.2	1.0	2.1
C6/C6 PFPiAb	68412-69-1	8.1	9.6	4.1	5.9	5.6	4.3	0.1	0.3	0.2
C6 PFPAa,b	68412-68-0	3.5	11.2	0.6	2.2	3.5	0.7	0.004	0.08	0.04
Legacy POPs
PCB153	35065-27-1	6.9	9.0	6.9	7.1	4.6	3.6	7.2	2.7	4.0
Mirex	2385-85-5	7.5	8.0	7.5	7.7	5.0	4.0	7.2	2.6	4.2
β-HCH	319-85-7	3.8	8.6	3.8	4.0	2.9	1.9	6.9	1.0	2.4
β-endosulfan	33213- 65-9	3.7	6.8	3.7	3.9	2.8	1.8	7.0	1.0	2.5

^aModel simulations assumed these PFAS exhibit net renal secretion in fish (Sec% = 500% for PFHxS and 250% for PFOA, HFPO-DA, DONA and C6 PFPA) and reduced renal absorption in birds and mammals (Abs% = 90%). For simulations of all other PFAS, a high degree of renal reabsorption was assumed for birds and mammals (Abs% = 99%), while for fish, renal reabsorption % (Abs%) was assumed to be 0% of GFR. f_u and renal secretion % (Sec%) was set to zero.

^bBiotransformation was assumed. Biotransformation half-life values (t_1/2,MET) used to derive k_MET, in the models, were set to 20 d for FOSA, 7 d for C6/C6 PFPiA and 5 d for C6 PFPA. For simulations of all other PFAS, k_MET was assumed to be zero (i.e., nonmetabolizable).

^cCalculated physical-chemical properties shown in bold font are measured values, while other values are calculated. Details of all measured and calculated properties are provided in Table S5.

Predicted activity-based TMFs of PFOS in the aquatic piscivorous, marine avian/mammalian and terrestrial mammalian food webs are 1.4, 3.0 and 6.5, respectively. PFOS is predicted to biomagnify in all three food webs, but the degree of magnification is greatest in food webs containing air-breathing wildlife. The predictions are consistent with TMFs reported in the literature, which have shown BMFs and TMFs of PFAAs in food webs containing birds and marine mammals tend to be higher than those of aquatic organisms and food webs.^3,4 For example, observed BMFs of PFOS (on a chemical activity basis) in beluga whales (beluga:cod) and polar bears (polar bear:ringed seal) of the Canadian Arctic marine food web were approximately 25 and 20, respectively (Table S10). The observed activity based BMF of PFOS in herring gulls from Lake Ontario (herring gull:smelt) is approximately 8.8. Conversely, observed activity-based BMFs of PFOS in aquatic organisms (e.g., sculpin:benthos and lake trout:smelt) from Lake Ontario are relatively low, ranging between approximately 1.3 and 1.5 (Table S10). Other field studies have similarly reported relatively low BMFs of PFOS in aquatic organisms (BMFs between 1 and 2).¹⁴

PFHxS, PFOA, GenX and ADONA are predicted to have low biomagnification potential in all three food webs (TMFs < 1), primarily due to high renal elimination of these compounds. F-53B, a chlorinated polyfluoroalkyl ether sulfonate compound, is predicted to exhibit bioaccumulation behavior similar to PFOS, with relatively high TMF values in all food webs.

The influence of physical-chemical properties on PFAS biomagnification potential is further illustrated by contour plots showing the relationship between model predicted TMFs and physical-chemical properties (K_OW,_nD_MW, D_ALB-W,K_OA,_n) of hypothetical substances (Figure S5). Separate plots are shown for neutral compounds (Figure S5a–c) and anionic compounds (Figure S5d–f) for three food webs (aquatic piscivorous, terrestrial mammalian and marine avian/mammalian). The majority of the studied PFAS are anionic (e.g., PFAAs), which are represented by the data in Figure S5d–f. The predicted TMFs of anionic PFAS increase with increasing K_OW,_n (and hence D_MW and D_ALB-W), (Figure S5d–f). For these compounds, K_OA,_n does not have any effect, due to the fact that the neutral fraction of the substances at biological pH (i.e., 6–7.5) is negligible (i.e., 99.99% anionic). For neutral substances (e.g., legacy POPs), the model predicts that chemicals with log K_OA,_n < 5 are efficiently eliminated in air-breathing animals.

The results show that PFOS and longer-chain PFAAs are predicted to generally exhibit a high degree of biomagnification (TMF > 1). These compounds exhibit relatively high K_OW,n and hence D_MW and D_ALB-W values, i.e., K_OW,n values in the range of approximately 10⁶ to 10⁸, with D_MW and D_ALB-W values in the range of approximately 10⁵ to 10⁶. Predicted TMFs of PFAS are lower in aquatic food webs compared to the terrestrial mammalian and marine avian/mammalian food webs (Figure S5d–f). This is consistent with previous observations and predictions for low K_OW-high K_OA neutral lipophilic substances (e.g., β-HCH).⁶²

F-53B is shown to exhibit a high degree of biomagnification in all food webs, due to a relatively high K_OW,n (10^7.0) and hence D_MW (10^5.7) and D_ALB-W (10^5.1). Conversely, more water-soluble, less hydrophobic PFAS such as HFPO-DA and DONA (K_OW,n,D_MW and D_ALB-W < 10⁵), are shown to exhibit negligible biomagnification in all food webs (TMF < 1), which is mainly due to increased renal elimination.

Influence of Precursor Compounds

Abiotic and/or biological formation of PFOS and other PFAAs is possible due to the presence of precursor compounds in the environment⁶³⁻⁶⁷ In vitro depletion studies have shown N-Et-FOSA can be metabolized to FOSA and FOSA metabolized to PFOS.^63,67 In vivo biotransformation of bis(perfluoroalkyl) phosphinic acids (PFPiAs) to perfluoroalkyl phosphonic acids (PFPAs) has also been reported.^68,69 Results of model simulations for these metabolizable PFAS, coupled with direct uptake of biotransformation products present in the environment, are shown in Figure S6 and Table S13. When precursor and intermediate compounds (N-Et-FOSA and FOSA) in surface water were set at 1,000 times and 500 times, respectively, below those of PFOS, the predicted BAF of PFOS in fish was 4,060 L/kg. This is only slightly higher (18% higher) than the predicted BAF of PFOS in the absence of precursors, (i.e., PFOS exposure alone, BAF = 3,450 L/kg). However, assuming equivalent initial concentrations of precursor compounds and PFOS, the predicted BAF of PFOS was approximately 160 times higher (5.6 × 10⁵ L/kg), (see Table S13). The available monitoring data indicate that the relatively more hydrophobic precursor compounds (N-Et-FOSA and PFOSA) are present at substantially lower concentrations than PFOS, and in many cases below detection limits. Modeling results indicate that when N-Et-FOSA and FOSA are present at concentrations close to those of PFOS, the influence of the precursors on tissue residue levels of the terminal end-product (PFOS) can be significant. Similar results are shown for biotransformation of C6/C6 PFPiA to C6 PFPA (Figure S6).

Implications for Risk Assessment and PFAS Screening Initiatives

Currently, there are numerous efforts worldwide to assess ecological and human health risks associated with legacy PFAS. Regulatory authorities are tasked with screening potentially several hundreds of PFAS. For example, approximately 1,200 PFAS are currently listed on the US Environmental Protection Agency (USEPA) Toxic Substances Control Act (TSCA) Inventory.⁷⁰ The models developed and tested here are promising tools that can support these important initiatives. The models are designed to predict internal whole-body and tissue-specific concentrations (C_B, ng/g) and activities (a_B, unitless) of individual PFAS and can be applied to aquatic (freshwater, marine) and terrestrial food webs. The models can also be used to estimate daily intake values (i.e., DI, mg/kg body weight/d) of various wildlife species and humans. Predictions of chemical activities (a_B, unitless) may be useful for employing a chemical activity-based risk assessment approach. This approach has been shown to be useful for assessing exposure risks of other contaminants of concern.^49,50 For substance screening purposes, the models provide insight into the influence of physical-chemical properties on PFAS bioaccumulation behavior. This information is important for bioaccumulation assessment (B) during evaluations of commercial chemicals for potential persistence, bioaccumulation potential and toxicity (PBT).

The developed models are highly dependent on appropriate values for the distribution coefficients. Specifically, D_MW, D_ALB-W, and D_SP-W are important parameters that can greatly influence model outputs. Thus, experimental and/or in silico approaches are needed to derive accurate values of these parameters. The results of the present study indicate that these distribution coefficients are highly related to K_OW,_n. Thus, calculated K_OW,_n values may provide some utility for PFAS screening initiatives.

In the developed models, the renal clearance rate (CL_R), used to derive the renal elimination rate constant (k_RENAL), is a key factor governing the bioaccumulation potential of some PFAS, especially relatively more water-soluble, less hydrophobic compounds. More research is needed to better predict renal reabsorption/secretion of PFAS, ideally with studies focused on developing quantitative-structure–activity-relationships (QSARs) with chemical hydrophobicity, membrane permeability/affinity and protein binding. Information regarding biotransformation rates of metabolizable PFAS, as well as precursor-product pathways, would also be beneficial. Lastly, further development and testing of these models will help to expand our understanding of key mechanisms and uncertainties related to PFAS bioaccumulation behavior.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.4c02134.

• Additional details related to model equations, model parameterization and application, including supplemental tables (Tables S1–S13) and supplemental figures (Figures S1–S6) (PDF)

• Electronic versions of spreadsheet models used in the present study (XLSX)

• Model updates and downloads can be found online at https://www.sfu.ca/rem/toxicology/our-models.html

The authors declare no competing financial interest.