Environ. Sci. Technol. 2009, 43, 2858–2864
A Combined Food Web Toxicokinetic and Species Bioenergetic Model for Predicting Seasonal PCB Elimination by Yellow Perch (Perca flavescens) K E N G . D R O U I L L A R D , * ,† GORDON PATERSON,‡ AND G. DOUGLAS HAFFNER† Great Lakes Institute for Environmental Research, University of Windsor, 401 Sunset Avenue, Windsor, ON, Canada, N9B 3P4, and Environmental and Resource Studies Department, Trent University, 1600 West Bank Drive Peterborough, ON, Canada, K9J 7B8
Received September 16, 2008. Revised manuscript received February 10, 2009. Accepted February 24, 2009.
A commonly used toxicokinetic model was coupled to a bioenergetic submodel optimized for yellow perch and an empirical growth submodel to predict daily PCB elimination flux in three size classes of fish under seasonally variable temperatures. Across seasons, the bioenergetic model predicted highly variable gill ventilation and fecal egestion rates which varied by 74.2-111.2 fold and 35-65 fold, respectively, over the annual cycle. The empirical growth model accounted for seasonal trends in overwintering lipid losses evident for all fish size classes and growth, evident only for the small fish size class, during warm water periods. The toxicokinetic model described seasonal trends in congener specific PCB mass balance of fish, but tended to overestimate PCB elimination for less hydrophobic congeners when the recommended gill transfer efficiency term (EW) of 0.54 was used. Downward adjustment of EW to an average value of 0.14 produced the strongest model fit for several low KOW PCBs but had less effect on model performance for mid- to high KOW congeners. The toxicokinetic model was less sensitive to parameters involved in fecal elimination of PCBs when applied to low and midKOW PCBs. This study demonstrates the importance of seasonal trends in metabolic rate, growth, and overwintering weight loss as factors that modify PCB toxicokinetics in temperate fish.
Introduction A number of organic chemical bioaccumulation models exist which provide algorithms for estimating toxicokinetic parameters such as pollutant uptake and elimination rate constants for a range of aquatic species (1-3). These models and associated algorithms have formed the basis for advanced food web bioaccumulation models that simultaneously solve steady state chemical bioaccumulation in multiple organisms inhabiting specific environments (4-6). Such food web models have become useful regulatory tools for facilitating * Corresponding author e-mail:
[email protected]. † Great Lakes Institute for Environmental Research, University of Windsor. ‡ EnvironmentalandResourceStudiesDepartment,TrentUniversity. 2858
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 8, 2009
environmental management decisions for persistent organic pollutants (POPs) and for the categorization of persistent, bioaccumulative, toxic (PBT) substances. Often, food web bioaccumulation models are adapted for predicting chemical exposures in species where no empirical toxicokinetic data are available. Empirical studies documenting contaminant toxicokinetics are limited to a small number of species and such research is usually performed under controlled laboratory conditions. Indeed, most of these models have adopted and progressed toward the development of generalized predictive algorithms rather than using species specific bioenergetic and/or physiological approaches (6). Validation of these models is usually performed by comparing model predictions against field observations. Such exercises are typically performed at the ecosystem scale and test model functioning as a whole (4, 5, 7), but do not allow evaluation of individual parameters within the model code. Despite the extensive use of food web bioaccumulation models, there have been few attempts to independently evaluate algorithms for key toxicokinetic parameters included in such models beyond training data sets used as part of the algorithm generation process (6). Recently, empirical data for congener specific PCB elimination rates were collected for three size classes of yellow perch (Perca flavescens) housed under ambient temperature conditions (8, 9). These data provided the opportunity to rigorously test toxicokinetic model algorithms from an established food web bioaccumulation model (6) to predict daily elimination flux over different body size ranges of fish, across different environmental temperature conditions, and under different conditions of growth and seasonal weight loss. We used a species optimized bioenergetics model developed and calibrated for yellow perch to predict fish gill ventilation and fecal egestion rates (10, 11) and an empirically derived growth model to predict daily changes in body weight and proximate body composition to supply inputs required by the toxicokinetic model to estimate daily chemical elimination flux. Our first objective was to evaluate the ability of the toxicokinetic submodel to describe time-dependent PCB depuration and concentration trends in yellow perch over the seasonal cycle. Additionally, this study investigated the mass transfer efficiency terms describing chemical flux across gill surfaces and between the animal and its feces to determine whether further calibration of these parameters could improve model performance.
Methods The empirical data set used for model evaluation was derived from a PCB depuration study conducted at ambient temperatures in mesocosm enclosures. Details of the experimental protocol, analytical procedures, and empirical measurements of PCB elimination rate coefficients are provided in Paterson et al. (8, 9) and are also provided in the Supporting Information. Seasonal temperature data, size specific growth rates, and temporal trends in whole body lipid and lean dry protein contents were used in conjunction with a modified version of the Wisconsin bioenergetics model (10, 11) to calculate gill ventilation rates and fecal production rates for each size class of fish. The bioenergetic model output and growth model output was then used as inputs to the toxicokinetic model to predict daily elimination flux of PCBs. Bioenergetic Model. A modified version of the Wisconsin bioenergetics model calibrated for yellow perch (10) was adapted for fish in each size class (Table S1). Food consumption rates (C; kJ · g-1 · d-1) were calculated according to: 10.1021/es802567p CCC: $40.75
2009 American Chemical Society
Published on Web 03/17/2009
C)
(GL + GP) + (SMR · A) + SDA + U EC
(1)
where GL is growth of somatic lipid (kJ · g-1 · d-1), GP is growth of somatic protein (kJ · g-1 · d-1; measured as lean dry weight), SMR is standard metabolic rate (kJ · g-1 · d-1), A is an activity multiplier term (unitless), SDA is the specific dynamic action (kJ · g-1 · d-1), U is energy lost to excretion (kJ · g-1 · d-1), and EC is the energy assimilation efficiency from food (unitless). Growth, in the model, is an empirically derived term and separately considers growth or weight loss of lipid and protein within the organism. Protein is assumed to represent the lean dry weight portion of the animal and is similarly defined as nonlipid organic matter (NLOM) in Arnot and Gobas (6). Lipid growth carries an energetic value of 39.3 kJ · g-1 and protein growth carries an energetic value of 18.0 kJ · g-1. Weight loss, either of lipid or protein, simply reverses the sign of GL or GP. Gonadal growth was not included in the model owing to a lack of parameterization of gonad production in laboratory fish. An improved standard metabolic rate (SMR) algorithm for yellow perch was reported by Enders et al. (11) as follows: SMR ) 0.01 · W B0.72 · e(0.19 · T ) ·
( )
0.024 · DO2 WB
(3)
U ) 0.0253 · T 0.58 · e-0.299 · (SMR · A + |GL | + |GP |) (4) In eq 4, the terms GL or GP were excluded from the calculation in a given time step if they were positive but included as the absolute positive value of the term when they were negative. This approach assumes that energy excretion is not associated with tissue growth, but occurs when tissues are catabolized. The diet energy assimilation efficiency (EC) in eq 1 is often set as a constant value. However the toxicokinetic submodel is sensitive to diet properties and digestive efficiencies for different dietary components. Thus, EC was estimated for the diet according to the recommended algorithm associated with the toxicokinetic model (6): 39.3 · XL,D · EL + 18.0 · XP,D · EP EDFood
C EDFood
(6)
The fecal egestion rate (QEX; g feces · g-1 · d-1) is required as an input to the toxicokinetic submodel. QEX is calculated as: QEX ) QC · (1 - (XL,D · EL + XP,D · EP))
(7)
The second output of the bioenergetics model required for whole body elimination rate estimates is the gill ventilation rate (QV; mL · g-1 · d-1). Total ventilation energy requirements are similar to eq 1, except that GL and GP are excluded from the expression. In other words, tissue growth is stored as potential energy with an assumed negligible anabolic ventilation cost while tissue catabolism contributes to total metabolic rate but without excess contribution to ventilation activity. Ventilation energy requirements are converted to a water flow rate according to: QV )
In eq 3, the absolute positive value of lipid or protein growth is applied. This presumes that growth and weight loss contribute to equivalent energy costs associated with tissue catabolism or anabolic processes. The energy excretion term is given by:
EC )
QC )
(2)
where WB is the body weight (g wet weight), T is the water temperature (°C), and DO2 is the oxycalorific coefficient for converting oxygen respired to energy of fish (14.30 kJ · g-1 O2; ref 12). For the aquaculture studies, the activity multiplier (A) is established at 1 (13). The terms SDA and U were estimated using the Wisconsin model algorithms (10). These terms are typically estimated as multiplier terms applied against consumption. Since consumption is being solved for in this model, SDA and U are calculated as proportional to the combined costs of (GL + GP + SMR · A). The equations for each term included in the bioenergetics submodel are as follows: SDA ) 0.172 · (SMR · A + |GL | + |GP |)
(kJ · g-1) of food. Arnot and Gobas (6) recommended values for EL (0.92) and EP (0.60) were applied. The solved consumption rate from eq 1 is converted into a mass flow (QC; g food · g-1 · d-1) by considering the energy density (EDFood; kJ · g-1) of consumed food;
(SMR · A) + SDA + U DO2 · CO2 · EO2
(8)
where CO2 is the concentration of oxygen dissolved in water (g O2 · mL-1) and EO2 is the oxygen extraction efficiency (0.60 unitless; ref 14) across the gills. Fish tanks were continuously aerated and oxygen concentrations were assumed to be saturated and estimated by: CO2 ) 14.45 - 0.413 · T + 5.56 × 10-3 · T2
(9)
where T is the water temperature (°C; (12)). Toxicokinetic Model. Toxicokinetic parameter estimation methods were obtained from Arnot and Gobas (6). For the present study, the focus was to describe contaminant elimination dynamics and therefore only parameters associated with chemical elimination are described (Table S1). Under depuration conditions, the change in chemical concentration of the animal is governed by:
( )
∆WB ∆CB CB - (k2 + kEX + kM) · CB )∆t WB
(10)
where CB is the chemical concentration (ng · g-1 wet weight) in the animal, WB is the body weight (g wet weight), and the terms k2, kEX, and kM refer to mass elimination rate constants (d-1) for chemical depuration across the gills, to feces, and as a result of metabolic biotransformation, respectively. Although change in weight with time is typically incorporated with other elimination terms as a growth dilution rate constant, it is expressed in eq 10 as a separate term to allow for both biodilution or bioamplification and can be either negative, as in the case of growth, or positive, as in the case of weight loss. Body weight (WB) is estimated at each time step as the summation of water (WH2O), lipid (WL), and lean dry weights (WP): WB ) WH2O + WL + WP
(5)
where XL,D and XP,D refer to the mass fraction of lipid and protein in the diet (unitless), EL and EP are assimilation efficiencies of dietary lipid and protein (unitless), the coefficients 39.3 and 18.0 refer to the energy density (kJ · g-1) of lipid and protein, and EDFood is the total energy density
(11)
Changes in body weight and proximate composition are empirically derived terms specific to the population of animals being modeled. Metabolic biotransformation (kM) was assumed to be negligible compared to diffusion based depuration processes as is commonly assumed when applying the bioaccumulation model to polychlorinated biphenyls in fish (2-5). VOL. 43, NO. 8, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2859
The gill elimination rate constant is calculated as: k2 )
EW · QV KBW
(12)
where EW is the chemical exchange efficiency across the gills, QV is the gill ventilation rate (mL · g-1 · d-1) from eq 8, and KBW is the biota/water partition coefficient (unitless). The biota/ water partition coefficient is estimated as: KBW ) XL,B · KOW + XP,B · 0.05 · KOW
(13)
where KOW is octanol/water partition coefficient of the chemical (unitless), XL,B is the fraction of lipid in the animal, and XP,B is the fraction of lean dry protein in the animal (unitless). The lean dry protein partitioning capacity was established by Debruyn et al. (15) to be 5% of the lipid partitioning capacity. The recommended algorithm for estimating EW is given as:
(
EW ) 1.85 +
155 KOW
)
-1
(14)
Over the KOW range of PCB congeners tested in the present study, eq 14 predicts EW to remain stable at a value of 0.54. Thus, a constant EW of 0.54 was applied for each congener in initial model simulations. Fecal elimination rate constants (kEX) were estimated by: kEX )
EEX · QEX KEX,B
(15)
where EEX is the PCB organism/fecal exchange efficiency, QEX is derived from eq 7, and KEX,B is the feces/biota partition coefficient (unitless). The algorithm for estimating EEX is given by: EEX )
1 3 × 10-7 · KOW + 2.0
(16)
The KEX,B is estimated from the relative proportions of lipid and lean dry protein in the animal (XL,B and XL,P, respectively) and its feces (XL,EX and XP,EX, respectively) and is given by: KEX,B )
XL,EX + 0.05 · XP,EX XL,B + 0.05 · XP,B
(17)
Model Evaluation and Calibration. The model was solved by finite difference using a spreadsheet software under a daily time step starting from day 0 to day 365. Empirical functions were generated from the experimental data to describe water temperature, growth, and weight loss of lipids and lean dry weight (proteins) for each size class at daily iterations. The model was initiated using mean day 0 chemical mass for each congener and size class of fish. Model performance was evaluated by computing the geometric mean model bias for individual PCBs. The geometric mean model bias was calculated as the ratio of predicted chemical mass to measured chemical mass for each fish and taking the geometric mean of ratios across the sampling points. Day 0 fish were not included in the geometric mean calculation because the model was initialized with day 0 data. Only PCB congeners which demonstrated significant mass elimination over the study, i.e., had significantly lower chemical mass in day 365 fish compared to day 0 fish, were included in the model evaluation. As per objectives, model calibration was restricted to parameters in the toxicokinetic submodel described in eqs 10-17. The toxicokinetics submodel uses the output from the two biological submodels, the bioenergetic submodel 2860
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 8, 2009
FIGURE 1. Top: body weight trends for small (9), medium (O), and large (*) size classes of yellow perch over time. Bottom: lipid weight trends in each size. Lines represent empirical model growth curves predicted for each parameter and animal size category. (QV and QEX) and the empirical growth submodel (daily changes in body weight, lipid, and lean dry protein) to make predictions on daily PCB losses by fish. Among the parameters described in the toxicokinetic submodel, growth (eq 11), KBW (eq 13), and KEX,B (eq 17) are time dependent functions related to daily changes in proximate body composition (lipids and protein) and total body weight. Since these parameters are directly related to the growth submodel, they were not considered for calibration. Similarly, the parameters QV and QEX were excluded from calibration as these reflect time dependent bioenergetic model outputs. This left two remaining parameters available for calibration, EW and EEX. Model simulations were initially run using the recommended values for EW (0.54) and EEX (eq 16) to estimate mass elimination and PCB concentration trends. Average model bias was examined across the PCBs to examine for trends in bias with chemical KOW. Model calibration was performed by adjustment of EW and/or EEX terms and evaluating model performance improvements. Adjustments for EW were applied globally across all congeners until the best fit was established. For EEX, the constant in eq 16 (i.e., value 2.0) was adjusted until satisfactory fit was achieved.
Results and Discussion Growth and Body Composition. Mean body weights at the beginning of the study were 10.1 ( 2.2, 45.9 ( 13.4, and 86.7 ( 9.7 g for small, medium, and large fish, respectively. Body weights of individual fish through time are presented in Figure 1. The small fish were the only group to exhibit growth with final body weights significantly (P < 0.05; ANOVA) higher at the end of the study. No significant (P > 0.05; ANOVA) growth occurred for medium or large fish. Most of the growth of small fish occurred during the first 90 d when water temperatures approached the optimal growth temperature
for yellow perch (23 °C). A simple linear growth curve was established for the small fish as follows: during the first 90 d, fish body weights were predicted by WB ) 0.069 · day + 9.43
(18)
After day 90, fish weights were increased by a daily increment of 0.14 g · day-1. The above model explained 31.5% of the variation in body weight and performed better than linear or exponential growth models. For medium and large classes, the initial mean body weight for each group was used without growth. Empirical model predictions of daily fish body weights as a function of time are provided in Figure 1 (top). Changes in temperature over the seasonal cycle had a strong effect on lipid content of fish (Figure 1, bottom). Winter lipid losses were described in refs 8 and 9 and have been demonstrated in field populations of yellow perch (16). There were highly significant (P < 0.001 for all regressions) positive relationships between % lipids and water temperature. An analysis of covariance (ANCOVA) indicated nonsignificant (P > 0.6) temperature/lipid interactions across size. Following adjustment for temperature, % lipids were significantly higher (P < 0.001; ANCOVA) for the small fish and no differences (P > 0.1; ANCOVA) occurred between medium vs large sizes. Separate % lipid versus temperature relationships were derived for the small and combined medium + large sizes. Linear regression analysis yielded the following equations describing % lipid changes with temperature (T; °C): Small Size: %Lipid ) 0.23 ( 0.03 · T + 5.7 ( 0.6 R2 ) 0.44; n ) 77; p < 0.0001 (19) Med/Large Sizes: %Lipid ) 0.22 ( 0.03 · T + 3.8 ( 0.6 R2 ) 0.39; n ) 76; p < 0.0001 (20) Figure 1 summarizes the fit of model predicted body and lipid weights against time for each size class. Lipid mass predictions were significantly correlated (P < 0.05; ANOVA) with empirical measurements for each size. However, owing to error propagation in body weight and % lipid relationships with time, the empirical lipid model explained only 10, 12.8, and 18.4% of the variation in whole body lipid mass for small, medium, and large fish, respectively. Protein content of yellow perch was determined as the difference between dry weight and lipid weight. There was a strong correlation between lipids and % moisture described by: %Moisture ) -0.68 ( 0.03 · %Lipid + 76.9 ( 0.33 (21) This relation was combined with the lipid and body weight predictions to estimate daily protein weight changes with time. Bioenergetic Model Output. The bioenergetics model provided output of energy costs was consistent with those reported for yellow perch in other studies. Bajer et al. (15) reported model simulation results using the Wisconsin and Karas and Thoresson models parameterized for 20 g yellow perch at 21 °C and fed a 1% daily ration. The total daily energy costs for the above simulations ranged from 0.094 to 0.110 kJ · g-1 · d-1. This compared favorably with the present simulation results for day 347, where 19.2 g yellow perch at 21 °C were predicted to have a total daily respiratory cost of 0.106 kJ · g-1 · d-1. Bioenergetic model estimated gill ventilation and consumption rates were highly influenced by water temperatures and adhered to general rules of allometry with the smallest fish exhibiting the highest weight specific bioenergetic rates (Figure S1). For gill ventilation rates, the predicted minima occurred on days 201 (small fish) and 210 (medium and large
fish) coincident with the lowest water temperatures (2.4 °C). For small, medium, and large sizes, gill ventilation rates changed by 74.2, 111.2, and 110.6 fold, respectively, over the study. The bioenergetics model predicted a complete cessation of food consumption for the medium fish during days 136-193 when temperatures were between 2.7 and 9.1 °C (Figure S1). Both small and large fish were predicted to continue feeding at this time, although at minimal levels. These predictions are consistent with aquaculture notes where minimal feeding activity occurred at water temperatures e10 °C (10). During this period, standard metabolism was low and predicted to be satisfied by lipid catabolism. Secondary alterations in feeding rates were predicted due to growth (small fish) and/or changes in lipid + protein content with time. For each size class, predicted feeding rates ranged from 0.05 to 3.26, 0 to 1.71, and 0.04 to 1.42% body weight per day for small, medium, and large fish, respectively. Fecal egestion rates are directly proportional to feeding rate (eq 7) and were calculated to be 36.5% of food consumption. Toxicokinetic Model Output. For each size class of fish, PCB elimination rates were negatively related to chemical KOW (8, 9). In general, empirical observations of PCB elimination rates were slower than measured in laboratory studies that used constant temperatures and chemical halflives were shown to be seasonally dependent (8, 9). In the present study, the toxicokinetic model was evaluated only for PCBs that demonstrated significant mass elimination during the study. For the small, medium, and large sizes, there were 11 (log KOW range 5.2-5.5), 31 (log KOW range 5.2-6.7), and 42 (log KOW range 5.2-6.9) congeners, respectively, demonstrating significant chemical loss. The large fish experienced significant losses for a greater number of PCBs even though the bioenergetic model indicated that these fish exhibited lower gill ventilation and fecal egestion rates. This is due to individuals from the large size experiencing the greatest overwintering lipid losses during the year. These seasonal lipid losses would have produced higher fugacity gradients between the animal and water which elevates diffusive flux across respiratory surfaces and to feces (17, 18). In contrast, the small size class had higher lipids and experienced growth which decreases the fugacity gradient between the animal and its environment (19). The model fit to the empirical data was initially evaluated using the recommended values for EW (0.54) and predictive algorithm for EEX. Model predictions of PCB whole body burdens and lipid normalized chemical concentrations through time were positively correlated with empirical data. Lipid normalized PCB concentration trends in each size class are provided for PCB 19 and 28/31 in Figure 2. For PCB 19, the model greatly underestimated chemical concentrations and chemical mass for each size class of fish. For PCB 28/31, the model provided a closer fit to the data than evident for PCB 19, but still overestimated chemical elimination. As chemical KOW increased, the model performance improved and satisfactory fits to the data were evident for compounds having log KOWs exceeding 6. Model performance across different PCBs was established by measuring geometric mean model bias as described in data analysis. A geometric mean model bias of 1 indicates perfect fit between model predictions and observations. Figure 3 summarizes model bias for individual PCBs as a function of chemical KOW. For the small size class, model bias was substantially lower than unity (0.001 to 0.35) for all chemicals. For the medium and large size classes, model bias showed a distinct KOW trend, such that model bias of under prediction was evident for low KOW chemicals, but approached the expected value of unity for mid- to high KOW congeners. To establish which elimination parameters were contributing to excess PCB elimination for low KOW chemicals, VOL. 43, NO. 8, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
2861
FIGURE 3. Model bias across PCB congeners demonstrating significant mass elimination over the study. Bias was estimated as the geometric mean of predicted/observed ratios of total chemical mass in fish across sampling time points, excluding day 0 values. Top graphic presents model bias from simulations using the recommended gill transfer efficiency (EW) value of 0.54 for small (9), medium (O), and large (*) size class fish. Bottom graphic represents model bias for simulations using size-optimized EW′s of 0.11, 0.15, and 0.16 for small (9), medium (O), and large (*) sized fish.
FIGURE 2. Observed and predicted lipid corrected PCB concentrations through time for PCBs 19 (O) and 28/31 (2). Solid lines are toxicokinetic model predictions using the recommended gill transfer efficiency (EW) value. Dashed lines are toxicokinetic model predictions using size-optimized EW values. the relative magnitude of model predicted gill (k2) and fecal elimination (kEX) coefficients were examined. Gill elimination coefficients are inversely proportional to chemical KOW, while fecal elimination is weakly scaled to KOW via the KOW dependence of EEX described in eq 16. For the eleven chemicals undergoing significant mass elimination in the small size class, the model indicated that >95% of predicted chemical losses occurred via gills. Similarly, for the significantly eliminated chemicals in the medium and large size classes, gill elimination contributed >80% of total chemical elimination. Even though the relative proportions of gill losses decreased as a function of KOW, the model always predicted gill elimination to dominate whole body excretion of PCBs. This suggests that model predictions of depuration were most sensitive to the parameters incorporated into eq 12. In addition to the above considerations, parameter perturbation trials were performed where the EEX value was set to 0 and EW held constant at the recommended value. This effectively excludes fecal elimination from the model, yet for low KOW PCBs, the model was still found to substantially overestimate 2862
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 8, 2009
chemical elimination. Thus, changes to EEX provided little improvement to the model fit, and for chemicals with log KOW values less than 6, the model was relatively insensitive to this parameter. Only downward adjustments of EW were successful in calibrating the model to match observations. Model calibration was therefore performed by adjusting EW (substituting the same EW value for all chemicals) until the KOW dependence of geometric mean model bias was removed. These calibrations were performed while holding EEX at the recommended values. For the small size class, the optimized EW was 0.11 which produced an overall model bias of 1.7 ( 0.1. For the medium and large size classes, the optimized EW values were 0.15 (overall model bias ) 1.5 ( 0.3) and 0.16 (overall model bias ) 0.99 ( 0.02, respectively). Figure 2 also summarizes model predictions of lipid normalized chemical concentrations for PCBs 19 and 28/31 when the calibrated EW values were used. For PCB 19, the superiority of the fit of model predictions using the calibrated EW value is evident. Similar observations of model performance were evident for other PCB congeners with log KOW values less than 5.8. Geometric mean model bias using the calibrated Ew terms are also presented in Figure 3. The optimized EW values established in this study (averaging 0.14) were lower than those reported elsewhere. Norstrom et al. (12) proposed an EW of 0.75 for their model of yellow perch PCB bioaccumulation. Morrison et al. (5) applied the same EW used by Norstrom. The recommended EW of 0.54 from eq 14 was derived by Gobas and Mackay (20), fitted to empirical measurements of EW for rainbow trout (14), and used in several food web bioaccumulation simulations by Gobas and coauthors (3, 4, 6, 7). Direct measurements
of Ew for hydrophobic chemicals have been made for a limited number of fish species using respirometer-metabolism chambers. Chemical EW values between 0.45 and as high as 0.80 have been reported using such approaches in different species of salmonids and in channel catfish (14, 21-24). The present study did not measure gill ventilation rates of yellow perch, thus over predictions of the magnitude of QV by the bioenergetic model would have resulted in calibrated EW values being lower than the true value. However, estimates of yellow perch metabolic rate and respiration rate derived from the present simulations were consistent with other bioenergetic modeling studies for this species as well as respirometry measurements of oxygen consumption by yellow perch (11, 13). Overestimates of the biota/water partition coefficient (KBW) derived via eq 13 might contribute to calibrated underestimates of EW. Equation 13 uses the expression provided by Debruyn et al. (15) which accounts for partitioning capacity contributed by lipid and nonlipid organic material. This equation provides slightly higher estimates (∼14%) of KBW compared to classic algorithms that assume only lipids contribute to tissue partitioning capacity (3, 20). However, the differences in KBW between the updated and classic algorithms for KBW is too small to explain the low calibrated value of EW. Similarly, experimental bioconcentration factors (BCFs) for laboratory fish exposed to PCBs via contaminated water suggests good fits between steady state BCFs and equilibrium partitioning predictions, particularly for lower KOW PCBs (25). Alternatively, failure of assumptions related to the gill exchange model may explain the discrepancy between calibrated PCB EW values and empirical measurements. The simple flow-limited model employed by eq 12 assumes that chemical flux across the gills is proportional to water flow, i.e., ventilation-limited diffusion. The present model does not allow for variation in EW as a function of gill ventillation, although Gobas and Mackay (20) recognized that EW could vary as a proportion of QV. McKim and Goeden (21) tested this by subjecting respirometer-metabolism chamber held brook trout to waters containing different oxygen concentrations to stimulate higher ventilation in animals over a gradient of hypoxia. Although the authors observed higher endrin uptake rates in fish with increasing ventilation, the increase in endrin uptake was not commensurate with increases in QV. Fish exposed to hypoxic waters (30% saturation) exhibited a 4.3 fold increase in gill ventilation rate but only a 2.2 fold increase in endrin uptake rate (21). The EW decreased from 0.80 ( 0.11 under oxygen saturation to 0.47 ( 0.11 under hypoxia (21). Similar observations of hypoxia induced changes in [14C] decanol EW across the gills of rainbow trout were reported by Schmieder and Weber (26). Noting discrepancies between flow limited uptake of chemical exchange across gills of guppies and rainbow trout, Sijm et al. (27) suggested that chemical uptake rates become less dependent on QV when QV exceeds 0.6 L · g-1 · d-1. Above this threshold, the authors hypothesized that bottlenecks related to desorption kinetics of chemical from blood proteins to the aqueous phase constrain EW such that EW will decrease with further increases in gill ventilation. In the present study, model predicted QV values commonly exceeded the hypothesized threshold of 0.6 L · g-1 · d-1 (Figure S1). For large, medium, and small size classes, gill ventilation rates in excess of 0.6 L · g-1 · d-1 occurred for more than 117, 128, and 143 d, respectively. The latter periods corresponded to the highest metabolic activity of fish and also reflected the period where most of the chemical elimination took place. The calibrated EW values were also observed to be the lowest for the small size class which had both higher overall QV values and a greater number of days over which QV exceeded 0.6 L · g-1 · d-1.
Additional simulations were established to test the threshold rule described above. In these simulations, EW was set to the recommended value of 0.54 when the daily QV was 0.6 L · g-1 · d-1 or lower. When QV exceeded the threshold, EW was adjusted in proportion to the increase in ventilation rate above the threshold according to: EW ) 0.54 ·
0.6 QV
(22)
The modification provided by eq 22 was found to still overestimate PCB elimination for low KOW congeners, although to a lesser degree than that provided by the original model formulation. Optimization of the modified model as described above indicated that model bias across PCB congeners was removed, and similar model performance realized as the calibrated EW described previously, when eq 22 was modified by substituting the value of 0.54 with 0.31, 0.32, or 0.30 for the small, medium, and large fish, respectively. Unfortunately, due to the nature of temperature changes during cooling and warming periods and timing of fish sampling, we were unable to separately calibrate EW under different conditions of gill ventilation. Performing similar model calibrations for fish depurated at constant temperature treatments would be useful to verify the above threshold QV predictions. Combining Sijm’s QV threshold and the yellow perch bioenergetic model enables the prediction that yellow perch EW will approach a value of 0.3 at water temperatures
