Polychlorinated biphenyl tissue-concentration thresholds for survival, growth, and reproduction in fish

Literature review

A comprehensive literature search was conducted to initiate this analysis on the effects of PCBs on fish. We focused on the endpoints of mortality, growth, and reproduction (MGR). Multiple databases were searched including Google Scholar (Google), Web of Science (Thomson Reuters), and SCOPUS (Elsevier). We excluded studies that focused on human health or other nonfish species, as well as studies that evaluated only PCB exposure and bioaccumulation. The primary literature was the preferred source of all data; however, previous reviews on the effects of PCBs on fish (Monosson 2000; Meador et al. 2002) gave critical insights into data selection and evaluation and provided a check to ensure that critical studies were included in the present meta-analysis.

Data inclusion criteria

Six criteria were developed to identify appropriate data for our meta-analysis (Figure 2), based on standards for determination of injury under the NRDAR framework (as laid out in 43 CFR Part 11.62; US Department of the Interior 2009) and the guidelines set forth by Rosenberger (1987).

The first criterion required studies to evaluate MGR effects. The MGR endpoints provide practical and well-defined data that are useful in decision making (US Environmental Protection Agency 2003). Increased mortality and decreased reproductive capacity are clearly deleterious to fish populations and are considered injurious to fish (see 43 CFR Part 11.62; US Department of the Interior 2009). Growth impairment (reduction) in fish also has a substantial impact on populations, because smaller fish are more susceptible to loss through predation, do not compete for resources as well, and can indicate reduced fitness of the population (Groh et al. 2015).

The second criterion required that the effects be unequivocally attributable to PCB exposure. For inclusion, studies must have shown that fish exposed to PCBs exhibited adverse effects relative to a control group, thus restricting our primary analysis to laboratory studies.

The third criterion required that fish be exposed to PCB mixtures, as occurs in the environment. Although all PCBs are structurally similar, the number and orientation of chlorines influence toxicity, both empirically and mechanistically (Safe et al. 1985; Safe 1990, 1994). Laboratory studies of individual congeners, although mechanistically informative, do not represent how PCBs are likely to be encountered in the environment. Moreover, data from single PCB congener studies are not readily translated from a laboratory study to a field exposure (Giesy and Kannan 1998). In addition, studies that only included the most potent dioxin-like PCBs (PCBs 77, 81, 126, and 169; Giesy and Kannan 1998) were excluded. Fish toxicity studies using only dioxin-like PCBs ignore the potential for impacts from other PCB modes of action. To facilitate the use of multiple PCB mixtures, the dose metric of total PCBs was used.

Criterion 4 required that PCB exposure concentrations be represented as quantifiable tissue residue concentrations. It is well established that PCBs bioaccumulate in tissues (Veith et al. 1983, 1979; Safe 1994; Meador et al. 2011), and the use of a tissue residue approach as an exposure metric provides the most direct connection to the observed effects (McElroy et al. 2011). This criterion required studies to measure PCBs in exposed fishes or provide sufficient data to allow tissue concentrations to be calculated. Quantification and conversion steps for estimation of PCB tissue residues are described in detail in the PCB exposure concentration selection and translations section).

Fifth, PCB-induced effects must be quantifiable for inclusion in this analysis. This criterion required that biological responses in PCB-exposed fish be measurably different from control or lower exposure organisms. Ideal data presented exposed fish effects that were statistically different from effects in control fish. This approach provided scientifically defensible determination of effects. In each case, the lowest-observed-effect concentration was selected for inclusion. Within this analysis, the effect terminology used was lowest-observed-adverse-effect residue concentration (LOAER), to more accurately reflect the type of data being collected (further described in the Data selection and translation section). The LOAER provided a definitive measure of adverse outcome from each study, that is, a single concentration for which statistically significant, PCB-related toxic effects were observed, presumably within the linear portion of the dose–response curve.

The sixth criterion required that the LOAER effect levels be paired with an exposure concentration to provide a single point for comparison across studies.

Data selection and translation

Data from studies meeting the 6 selection criteria were compiled. The primary data extracted from each study were individual LOAERs for MGR and the PCB tissue concentrations at which each of those effects occurred. Each paired set of primary data (tissue concentration +LOAER expressed as percentage effect) was referred to as a datapoint, within the context of the present review. Ancillary data were also collected for each datapoint, including PCB mixture type, fish species, and fish life history information (specifically temperature, salinity, and fish life stage). Other study data collected included route and duration of PCB exposure. All extracted data were converted to standardized units (e.g., μg/g for PCB tissue concentrations) to facilitate metadata analysis. Conversions for each type of data followed set procedures (described in the Effect data selection and translation section). Original and converted values specific to each datapoint are given in the Supplemental Data (Tables S1–S3). Abbreviations and information for each mixture are also given in Supplemental Data (SI Table PCB).

Effect data selection and translation

The MGR data were quantified using effect-specific endpoints. Mortality effects were defined by changes in survival between PCB-exposed and control groups. Studies with measurements of hatchability and larval survival were included among mortality studies. Hatchability compared the number of surviving, viable larval fish with the initial number of eggs. Growth effects were defined by change in weight or length in PCB-exposed treatments relative to control groups. Only reduced growth data (i.e., fish at LOAER significantly smaller than control fish) were included in our analysis, because decreased growth is generally accepted as an adverse effect. Growth data were excluded in cases in which size-selective mortality, (i.e., smaller fish dying first) may have influenced growth calculations (Seelye and Mac 1981; Garrido et al. 2015). For reproduction effects, only reproductive impairment in adult fish was included in the analysis under this category. Often studies on reproduction reported effects on eggs or larval fish (i.e., hatchability, larval survival; Hansen et al. 1973; Schimmel et al. 1974; Freeman and Idler 1975; Hugla and Thome 1999); those endpoints were more accurately characterized as mortality responses and were included in the mortality dataset. For adult fish, reproductive endpoints included egg production (fecundity), fertilization rate, and number of spawns.

The MGR effects at the LOAER were normalized to percentage effect relative to control responses to provide for comparisons across studies (Equation 1). The LOAER and control effects were expressed as responses that were reduced because of toxicity (e.g., % survival, body weight, number of eggs produced). This provided consistency across levels of effect, particularly for mortality endpoints, for which responses were commonly either %mortality or %survival. Any effect orientation changes are given in the Supplemental Data, Tables S1 to S3 . Furthermore, Equation 1 oriented all data so that percentage effect would increase with PCB tissue concentrations (e.g mortality, growth inhibition, and reproductive impairment). Normalization helped identify the actual impact of PCBs, beyond the natural variation in MGR. A small number of studies had quantifiable differences in mortality between exposed and control fish, although statistics were not reported. These studies were included if the percentage effect was >50%. This allowed the inclusion of more data, while reducing the chance that differences between the control and exposed groups were not real. These cases are identified in the Supplemental Data (Tables S1–S3).

(1)

PCB exposure concentration selection and translations

Concentration data at the LOAER were most commonly reported as micrograms (μg) of total PCBs/gram (g) of whole-body wet weight. The LOAER concentration data were standardized to this reporting metric using the protocols described below.

The most common translation was from dietary dose to whole-body wet weight. A 50% conversion from measured PCB dietary dose to whole-body wet weight was utilized (Equation 2). Previous work established 50% assimilation of PCBs at dietary dose concentrations <200 μg/g, with approximately 25% assimilation efficiency at greater exposure concentrations (Opperhuizen and Schrap 1988). Findings in other studies were similar (Gruger et al. 1975, 1976; Madenjian et al. 1999). Meador et al. (2002) also utilized a 50% assimilation efficiency estimate for exposure in their examination of PCB-induced effects on salmonid species (Meador et al. 2002). In Equation 2, x, the concentration of PCBs (μg) in food (g wet wt) was multiplied by the assimilation factor (0.5) to estimate y, the concentration of PCBs (μg)/g of whole-body wet weight.

(2)

A few studies reported study-specific bioconcentration factors (BCFs) instead of directly reporting PCB tissue concentrations. For example, Veith et al. calculated BCF values from PCB tissue concentrations and the aqueous (exposure) PCB concentrations, but did not actually report the tissue concentrations (Veith et al. 1979). The reported BCF and aqueous concentrations in these cases were used to determine whole-body tissue concentrations (Equation 3). Study-specific adjustments to Equation 3 are provided in the Supplemental Data (Tables S1–S3). In Equation 3, the study-reported BCF was multiplied by x, the concentration of PCBs in water, to estimate y, the concentration of PCBs (μg)/g of whole-body wet weight.

(3)

One study included in the present analysis (Black et al. 1998a) used intraperitoneal injection for dosing and required translation of injected dose to estimate whole-body wet weight tissue concentration. Evidence supported a 75% assimilation of measured injected PCB dose to total PCB wet weight (Equation 4; reviewed by Meador et al. 2002). In Equation 4, x, the concentration of PCBs (μg) injected/g body wet weight was multiplied by the assimilation factor (0.75) to estimate y, the concentration of PCBs (μg)/g of whole-body wet weight.

(4)

The PCB measurements reported as dry weight were converted to wet weight based on an assumption of 80% moisture content for fish (dry wt 20% of wet wt; Equation 5; Meador et al. 2002). Lipid normalized concentrations were translated to wet weight assuming 4.6% lipid content in fish (Equation 6). If study or species-specific lipid concentrations were reported/available, they were used.

(5)

(6)

Total PCBs were used as the metric for data standardization. Most included studies used total PCBs to report concentrations. Total PCBs were also often used in injury assessment or risk assessment cases (e.g., DeVault et al. 2001; Blazer et al. 2006). The detailed congener information required to calculate toxic equivalency factor (TEQ)-based exposure estimates were not available for most of the studies. An extensive evaluation of total PCBs versus TEQs in terms of concentration–response (see the Supplemental Data, total PCB vs TEQ) suggested that total PCBs provided the best approach for the development of PCB injury thresholds for this dataset.

Sensitivity distribution development

Probabilistic sensitivity distributions were developed separately for the MGR LOAER datasets. Probabilistic sensitivity distributions, similar to chemical toxicity distributions and species sensitivity distributions (SSDs), were applied to our current datasets to develop a probabilistic injury assessment for PCBs. This allowed us to identify the percentage of species expected to display significant mortality, growth, or reproductive impact at any given total PCB concentration. The use of LOAERs anchored these distributions to measured effect concentrations. We recognize that the included species and PCB mixtures may not represent the full range of sensitivities. This approach simply represents the best available information we have at this time, which can be updated with new information as it becomes available.

The development of the probabilistic injury distributions followed a modified SSD approach (Berninger and Brooks 2010). Tissue concentration datasets were log-transformed and tested for normality. Each PCB LOAER concentration within a dataset (n = # points within dataset) was assigned a numeric rank (i) from low to high. Ranks were then converted to a cumulative probability (percentage rank [j]) using the Weibull equation (Equation 7). The PCB tissue concentration–percentage rank pairs were used to generate a probability-log scale linear regression for the MGR datasets. These probability regressions estimated the number of species experiencing significant effects at any given concentration (previously described in depth by Solomon and Takacs 2001; Berninger and Brooks 2010; Berninger et al. 2016, 2011). The slope (m) and intercept (b) of each regression line were used to identify the percentage of species affected (centile [C]) or a tissue concentration value (x), using NORMSDIST and NORMSINV functions in Microsoft Excel® (Equations 8 and 9). Probabilistic distribution regressions were used to determine thresholds of effect at specific PCB tissue concentrations across multiple orders of magnitude (Equation 8) and to determine concentrations associated with a specified percentage of species effected (5–95%). Distributions were developed for mortality, growth, and reproduction independently.

(7)

(8)

(9)

PCB concentration: Effect regressions

The next step in the meta-analysis was the direct evaluation of the relationship between PCB tissue concentrations and effects. Log-transformed concentrations were compared with their associated percentage effects at the LOAER using linear regression. Given the range of LOAERs across studies, this produced PCB concentration–effect relationships across a wide range of potential concentrations. Each dataset (MGR) was evaluated to determine whether it fit the statistical assumptions required for linear regression, including normality and homogeneity of variance. Values with studentized-deleted residuals > 3 were identified as potential outliers, and specifically evaluated further. Regression calculations, statistics, and graphing were conducted in SigmaPlot (Ver 13.1; Systat Software).

Regression analysis was conducted on raw and condensed concentration–effect datasets. The raw data on effects were highly variable, so analysis on condensed data was conducted to reduce variability and produce regression analyses with greater predictive power, as is common practice within this type of meta-analysis (Monosson 2000; Meador et al. 2002). The LOAER tissue concentrations were divided into 5 groups (20th percentiles or quintiles), which provided better resolution and could also be applied across all MGR datasets without resulting in empty groups (null sets). The PERCENTILE function (Microsoft Excel) was used to identify 20th, 40th, 60th, and 80th percentile concentrations from the entire dataset, which were subsequently used to define quintile boundary concentrations for individual mortality, growth, or reproductive datasets for consistency (Equation 10). The central tendency of tissue LOAER concentrations within each quintile (0–20, >20–40, >40–60, >60–80, >80–100) was calculated as the geometric mean. Standard error of quintile means was calculated according to Equation 11. All central tendency values for log-transformed LOAERs were calculated as geometric means and referred to as geomean(s). Arithmetic means (averages) and standard errors (SEs) for percentage effect were calculated for each quintile. Normality of quintile values was assessed as above.

(10)

The quintile percentiles (y; 0.2, 0.4, 0.6, 0.8) identify concentrations (x) that mark the bounds of each quintile.

(11)

where the geometric mean of a quintile qG = quintile geometric mean; qX = standard deviation of the log values in that quintile; and n = number of values within the quintile.

Linear regressions along with applicability statistics were regenerated using the quintile data from each dataset (MGR). For all the regressions, the significance (p < 0.05) and the coefficient of determination (r²) were calculated by comparing percentage effect and log PCB tissue concentrations (LOAER values).

PCB concentration: Effect regression threshold calculation and prediction development

The MGR quintile regressions provided slope and intercept information that was then used to generate concentration–effect regression thresholds. Percentage effect can then be estimated at any given PCB exposure concentration (Equation 12), and PCB concentrations estimated for any given percentage effect (Equation 13). Effect estimates were calculated in a standard spreadsheet program to simplify the process and provide a standardized approach that can be applied in a variety of regulatory and decision-making scenarios. The MGR percentage effects could be estimated across order of magnitude ranges of PCB tissue exposure concentrations and PCB tissue concentrations calculated for percentage effects from 5 to 95%.

(12)

(13)

Applicability limits within our analysis were defined based on the input data. We established a lower cutoff of 0.1 μg/g for the subsequent analysis of LOAER values. The lowest LOAERs within the MGR datasets occurred in the range of 0.1 μg/g, but not below. Furthermore, background PCB concentrations, at and below the 0.1 μg/g level, were observed in control fish without reported MGR effects. The PCB concentrations below this 0.1 μg/g cutoff are not expected to elicit adverse MGR effects based on our review of the literature. The upper limit for prediction of MGR effects based on PCB tissue concentration was set at 3000 μg/g, based on the average molecular weight of PCBs and the corresponding threshold for mortality from narcosis alone. In other words, if PCB concentrations were to reach this level in tissue, all fish would be expected to die from the nonspecific narcotic mode of action. This cutoff was more theoretical than practical because the data clearly showed that mortality occurred well below 3000 μg/g. Values above or below these limits were identified within tables and figures to denote that they were beyond practical applicability of the models.

PCB concentration: Effect threshold uncertainty analysis

Uncertainty regressions were developed for the MGR datasets. These were based on the assumption that each quintile percentage effect represented the central point of a normal distribution of percentage effects. To identify this distribution, an SE was multiplied by a probability factor, analogous to the approach used to calculate confidence intervals. Dataset SE was calculated as the average of the SEs for percentage effect at each quintile. The probability factor, the degree of normal variability around each percentage effect quintile, was determined using a 2-tailed t distribution. The t values were calculated at 50, 60, 70, 80, 90, 95, and 97.5% probability (Equation 14). Dataset SEs were multiplied by the t values at each probability level to establish dataset-specific uncertainty factors. The uncertainty factor was added to (and subtracted from) the quintile average percentage effect values to generate upper (and lower) uncertainty for each probability level. These percentage effect uncertainty sets (50, 60, 70, 80, 90, 95, and 97.5%) were regressed against the geometric mean quintile tissue concentrations to generate a series of uncertainty regressions. Using the slope and intercepts from these regressions (Equations 12 and 13) provided uncertainty ranges for the concentration–effects regressions.

(14)

The probability (y), in decimal percent, and degrees of freedom (df) were used to calculate a t value.

Evaluating testing parameters across individual studies

We compared ancillary information within datasets to test our assumption that including all available data provided robust and generally applicable estimates of PCB effects in fish. Previous injury assessments have often limited data selection based on species of concern, life history characteristics (e.g., temperature and salinity), or the mixture of PCBs deemed to be present in the geographic area. This limits the amount of data available to make assessments, often to only 1 or 2 studies. Although limited data prevented meaningful multivariate analysis, 5 ancillary variables were evaluated: PCB mixture, species, life stage, salinity tolerance, and temperature preference. Analysis of the PCB mixture was of particular interest given that 13 different PCB mixtures were included in our study and it has been well established that different mixtures of PCBs have different potencies (Tillitt et al. 1991; Harris et al. 1993). Twenty different fish species were included. The other ancillary data were evaluated as binary variables: life stage (early life stage [ELS] or juvenile/adult), salinity tolerance (fresh or saltwater), and temperature preference (cold or warm water).

Average percentage effect and geomean PCB tissue concentration, along with SEs, were calculated for each subgroup of ancillary variables within the MGR datasets. Differences among subgroups within each variable were evaluated using an analysis of variance (ANOVA) with post hoc testing. Data from the binary variables (life stage, salinity, and temperature) were used to generate linear regressions for each subgroup, and a concentration–effect regression was generated for each individual PCB mixture and fish species with sufficient data (n ≥ 3).

Application of metadata analysis to evaluate cumulative effects

We developed the cumulative largest effect model (CLEM) to integrate simultaneous adverse MGR effects expected at a given exposure. The CLEM uses the most sensitive MGR effect at a given concentration. The CLEM additively combines mortality and growth given that fish exposed to PCBs experience mortality monotonically with concentration, whereas those that survive have reduced growth (Nebeker et al. 1974; DeFoe et al. 1978; Mauck et al. 1978; Bengtsson 1980; Mayer et al. 1985; Thomas and Wofford 1993; Fisher et al. 1994; Black et al. 1998a; Orn et al. 1998; Gutjahr-Gobell et al. 1999). The largest cumulative effects, either reproduction (R) or the additive mortality and growth (M+G) effect rates, were used to generate the CLEM. The CLEM prevents double counting of any single effect. Furthermore, the reproductive dataset was used independently, based on the fact that reproductive effects of PCBs on fish generally occurred at lower concentrations than effects on growth and mortality, along with the idea that reproductive effects will generally occur through different pathways than mortality and growth-related effects. Moreover, reproductive effects limit populations through decreases in natality, whereas mortality and growth are ultimately measures of loss to the juvenile/adult within a population. Independent analysis of each of these adverse effects fails to consider the influence that each of these outcomes exerts on a population simultaneously. Therefore, we developed the CLEM for the evaluation of potential effects of PCB exposures on fish populations across a wide range of exposures, as well as through multiple, co-occurring mechanisms.

The CLEM calculations were derived from reproduction and mortality + growth dataset regressions. The mortality and growth regression percentage effect responses were added at designated concentration intervals (0.01, 0.1, 1, 10, 100, and 1000 μg/g). The CLEM concentration–response regression was then based on the most sensitive endpoint, either mortality + growth or reproduction percentage effects, at each designated exposure concentration. The CLEM regression identified cumulative concentration-effects across the range of expected PCB concentrations, following procedures previously described (Equations 12 and 13).

PCB concentration–mortality effects regression

The linear regressions of percentage mortality and PCB tissue concentration had nearly identical slopes for the full dataset and for the quintiles condensed datasets (Supplemental Data, Table S8 and Figure 4). Calculated quintile average datapoints (PCB tissue concentration and percentage mortality pairs; Table 7) were used to generate the quintile regression. For example, the second quintile, Q2 (20–40th percentiles), fell between 5.6 and 23.0 μg/g and contained 6 LOAER values (Table 1, Q2: 6.0, 6.7, 14.0, 14.3, 15.0, and 17.2 μg/g), resulting in a geometric mean PCB tissue concentration of 11.3 μg/g (± 1.0 μg/g SE); the corresponding percentage mortalities (Table 1, Q2: 5.6, 13, 66, 54, 28, and 50%, respectively) resulted in a Q2 average percentage mortality of 36.1 ± 9.9%. The log PCB concentration values were used in both regressions because concentration was log-normally distributed. Both regressions were significant (p ≤ 0.05) and passed the normality (p > 0.05) and homogeneous variance (p > 0.05) tests. The mortality quintile regression improved the dataset correlation (full: r² = 0.46 vs quintile: r² = 0.79; Supplemental Data, Table S8). The expected percentage mortality was calculated across a range of PCB exposure tissue concentrations (Table 8). The quintile regression indicated that mortality would not be expected at or below PCB tissue concentrations of 0.1 μg/g. However, at a concentration of 0.5 μg/g, the quintile regression predicted a mortality of 10%. The quintile regression predicted that mortality would increase by approximately 20% with every order of magnitude increase in exposure concentration. Furthermore, expected PCB exposure concentrations were predicted at a set of given mortality rates (%; Table 9). This provided a way to identify the concentrations associated with expected magnitudes of effect, for example, 10% mortality would be expected at 0.4 μg/g and 25% mortality would be expected at 2 μg/g.

The uncertainty regressions representing a range of effect probabilities were generated (Figure 4) and allowed probability ranges to be calculated for any PCB tissue concentration (Supplemental Data, Uncertainty). For example, at a 5-μg/g PCB exposure concentration, the predicted 32.5% mortality has a 50% confidence interval (CI) ranging from 26.5 to 38.5% and a 95% CI range from 10.3 to 54.5% mortality.

During the initial regression diagnostics process, one datapoint (0.32 μg/g and 92% mortality; Hugla and Thome 1999) was identified as an outlier (studentized deleted residual value > 3). A preliminary regression including this datapoint predicted substantial mortality at PCB tissue concentrations not supported by any other studies within the dataset. For those reasons, this datapoint was excluded from further analysis and all regressions excluded the outlier.

Influence of specific PCB mixtures on mortality

The average LOAER values for PCB mixtures followed the same concentration–response trend as the quintile data (Figure 5A and Supplemental Data, Table Average PCB Mixtures). Regression analysis showed a significant, positive log-linear relationship among PCB mixture concentration–mortality pairs (regression p = 0.001; r² = 0.74), and none of the average PCB mixture values were observed to be an outlier or to be exerting undue influence on the regression. This trend suggested dose–response as an important factor among average PCB mixture datapoints (Figure 5A). However, differences in potency certainly exist among technical mixtures and individual PCB congeners, as established in the literature. Furthermore, due to a paucity of available, appropriate data, mixture-specific LOAER values would be based on 1 or 2 datapoints for most PCB mixtures. For those mixtures with multiple datapoints, individual mortality-based LOAER regressions for both A1254 (n = 13) and A1016 (n = 4) had positive dose–response relationships (Figure 5B). For A1254 there was a significant log-linear regression (p = 0.01; r² = 0.46), whereas A1016 LOAERs in fish increased with concentration; however, the relationship was not significant (p = 0.2 r² = 0.58), likely due to the small number of studies in the dataset (n = 4). The percentage effects associated with the different PCB mixtures were significant. However, no individual pairs of LOAER values for PCB mixtures tested as significantly different (Tukey's test, p > 0.05). This may be due to the wide range of mortality responses and the overall influence of PCB concentration on the percentage effect. For example, the A1242 LOAER caused 100% mortality (Nebeker et al. 1974), whereas a study with a mixture of Aroclors caused an 8.2% increase in mortality (Fisher et al. 1994). The strong trend of percentage effect increasing with concentration, observed across the mortality dataset generally, supported our inclusion of all the vetted data, which, in turn, provided better resolution across the spectrum of PCB exposure-related mortalities.

Influence of species tested on mortality

Whereas species average LOAERs varied in both PCB concentration and percentage mortality, the relationship between geomean PCB LOAER and average percentage mortality for individual species generally followed the same concentration–response trend as the quintile regression (Figure 6A) and was significant (p = 0.01, r² = 0.39). Residual analysis of that regression found no outliers. The geometric mean mortality-based LOAER values in fish spanned 4 orders of magnitude, from 0.75 μg/g for tilapia to 399 μg/g for fathead minnow (Supplemental Data, Table Average Fish Species). Many species occurred singly within the dataset. Species with multiple mortality studies generally had large SEs. The geomean mortality LOAER values for pinfish (38 ± 24 μg/g; n = 2), mummichog (9.7 ± 2.3 μg/g; n = 2), sheepshead minnow (93 ± 42 μg/g; n = 6), and fathead minnow (399 ± 116 μg/g; n = 5) had SEs ranging from 20 to 60% of the geomean mortality LOAER value. Species-specific data for both PCB LOAER values and percentage effect failed the normality test and were therefore assessed with an ANOVA of ranked data. No significant differences were observed in either geomean mortality-based PCB tissue LOAER values (p = 0.24) or average percentage mortality (p = 0.78) among the species evaluated in this dataset.

Individual regressions for fathead minnow (n = 5) and sheepshead minnow (n = 6) showed significantly positive concentration–response data (p = 0.04, r² = 0.81; and p = 0.002; r² = 0.92, respectively; Figure 6B). Slopes of the linear regressions for both were steeper and shifted to the right, relative to the quintile mortality regression, suggesting a more homogeneous response among the populations of fish tested (steeper slope) and lesser sensitivity (shifted right), compared with the all-species regression. However, these datasets were small and confounded by datapoints generated from multiple PCB mixtures (4 for fathead minnow, 2 for sheepshead minnow). Overall, the lack of significant difference suggests that each species represents an overlapping continuum of their individual sensitivities. Again, these analyses suggest the usefulness of an overall approach that is more inclusive of all available data and species.

Influence of life history variable on mortality

No significant differences were observed among averages of paired test condition variables (warm/cold water, fresh/salt water, adult/ELS). There was a great degree of overlap among PCB LOAER and percentage mortality averages for these data. Enough data were available that concentration–response regressions could be generated for each life history variable set: life stage (adult/ELS; Figure 7A), temperature (warm/cold; Figure 7B), and salinity (fresh/salt water; Figure 7C). Each of these life history variable regressions of the LOAER mortality data was significant (p < 0.05). However, the regression slopes and intercepts, and the degree of overlap, suggested that these differences in fish life history did not appear to influence the PCB tissue concentration–mortality relationship. Life history–based paired variable regressions of LOAER values for mortality did not vary substantially from the quintile regression that represented the same data from the entire dataset. Overall, this analysis suggested that when one is deriving mortality-based effects thresholds in fish, there is little support for excluding or separating data based on these life history parameters.

PCB concentration: Growth effects regression

There was minimal difference in the linear regressions between the full growth dataset and the quintile dataset (Figure 8 and Supplemental Data, Table S8). Both regressions passed the normality test (p ≥ 0.05) and the homogeneous variance test (p > 0.05). The full dataset regression approached significance (p = 0.09) but was poorly correlated (r² = 0.18), whereas the quintile dataset regression was significant (p = 0.04), with an improved correlation coefficient (r² = 0.80). The expected percentage growth was calculated across a range of PCB tissue concentrations (Table 8). As PCB tissue exposure concentrations increased, the predicted growth inhibition did not exceed 50% based on the regression. The PCB-induced growth inhibition had an upper limit, that is, only a certain degree of growth inhibition was observed before alternative pathways of toxicity were manifested in the exposed organisms, specifically mortality. This was observed in half of the studies in which both mortality and growth inhibition were monitored (DeFoe et al. 1978; Mayer et al. 1985; Thomas and Wofford 1993; Gutjahr-Gobell et al. 1999). In these studies, once mortality began to occur, individuals were removed from the population being evaluated for effects on growth. Thus the range of percentage growth reductions within individual studies of the present analysis achieved a maximum of 66% (DeFoe et al. 1978; Table 2), and the average percentage growth inhibition in the fifth quintile (Q5) was 37 % (±11%; Table 7). Furthermore, at a certain PCB exposure concentration, the percentage growth inhibition could be predicted. For example, a 25% inhibition of growth was predicted at 12 μg/g (Table 9).

The uncertainty regressions were generated for the growth dataset, representing a range of probabilities (Figure 8) and allowed probability ranges to be calculated for any PCB tissue concentration (Supplemental Data, Uncertainty). For example, at 10 μg/g, the quintile growth regression predicted 24.4% growth inhibition with a 50% CI ranging from 18.2 to 30.5%, and a 95% CI range from 1.2 to 47.5% from the uncertainty regressions.

Influence of PCB mixture on growth

A concentration–percentage effect regression of the 10 PCB mixture average datapoints (LOAER concentration–growth effect pairings) was not significant (Figure 9A; p = 0.78; nonsignificant regression line not shown). The plotted average PCB mixture datapoints relative to the quintile growth regression (Figure 9A) suggested differences. They did not follow the same relative exposure concentration–effect patterns observed in the overall growth regression, with several individual mixtures shifted away from the quintile regression line. No significant differences among mixture average LOAER concentrations (p = 0.36) or effects (p = 0.63) were observed (Kruskal–Wallis, p > 0.05). The majority of PCB mixtures were only tested once within the growth dataset, and those that were tested in multiple studies had highly variable results. The 3 PCB mixtures with multiple study datapoints (A1254, A1248, PCB-BLA) had large SEs for the LOAER concentration (Supplemental Data, Table Average PCB Mixtures). Strong concentration–response trends were observed for A1248 (with 3 datapoints, the geomean concentration was 62 ±12 μg/g) and A1254 (with 5 datapoints, the geomean was 20 ± 8.5 μg/g; Figure 9B). The regression of A1254 datapoints was similar to the quintile regression. The slope of the A1248 regression was much steeper, but all datapoints in that regression came from one species (fathead minnow), suggesting that these differences did not wholly result from differences in PCB mixtures.

Influence of species on growth dataset

Differences in species average PCB concentration LOAER values for growth inhibition across the 11 different species of fish were substantial, varying by 3 orders of magnitude, but were not statistically significant (Supplemental Data, Table Average Fish Species). Percentage growth effect also varied by species, but variation was not significant. Most species average growth concentration–response datapoints fell near the quintile regression line (Figure 10). The species that were not near the quintile regression (e.g., catfish and zebrafish) represented a single study, but were still within the 70% CI (Figure 8).

Influence of life history variable on growth dataset

No significant differences in average LOAER or percentage growth inhibition were observed for salinity (salt vs fresh) or temperature (warm vs cold). For life stage, a significant difference was observed between the LOAER growth values of ELS fish (78 ± 22 μg/g) and adult fish (12 ± 9.5 μg/g). However, the distribution of the datapoints for growth LOAERs in adult fish overlapped the range of the ELS LOAER values with the exception of one value (Figure 11A). This finding suggests that the significant differences of the average responses may be an artifact of testing approaches (e.g., species tested, PCB mixture, concentrations tested), rather than a true difference. The pattern holds for salinity and temperature variables, for which the individual LOAER values were distributed across the entire range of responses (Figures 11B and C). An analysis comparing individual variable regressions with that of the quintile regression was not conducted, because individual variable regressions were not significant.

Growth effects in context with other effects

Growth inhibition LOAER values for PCB were influenced by mortality, particularly at higher exposure concentrations. Indeed, no studies had reductions in growth >66% of the control treatments (Table 2). It was likely that PCB-induced growth impacts in fish were limited due to the co-occurrence of mortality at the same exposure concentrations. Because mortality occurred at a given exposure concentration within a study, those individuals that died were not available for assessment of other adverse outcomes, including impairment of growth. Significant growth impacts occurred in several studies at the same exposure concentrations at which significant mortality was also occurring (DeFoe et al. 1978; Mayer et al. 1985; Thomas and Wofford 1993; Gutjahr-Gobell et al. 1999). In some cases, PCB-induced reductions in growth occurred at the next lower exposure concentration, where mortality occurred but was not statistically significant (Nebeker et al. 1974; Mauck et al. 1978; Fisher et al. 1994; Black et al. 1998b; Orn et al. 1998). Growth effects caused by PCBs in fish may also be less prominent due to size-selective mortality. If, as has been suggested, smaller fish are more susceptible to PCBs and die first (Seelye and Mac 1981; Garrido et al. 2015), then the growth (or size) differences between average control and PCB-exposed fish would appear to be not as great, whereas the impact on individual fish (not measured) may be greater. Conversely, when expected mortality was low, there was a greater degree of impact on growth (Tables 1 and 2). This was the case with fathead minnows, a species observed to be among the least sensitive to PCB-induced mortality. Fathead minnows had the greatest degree of growth impairment, 50% and 66% reduction in growth compared to controls (Nebeker et al. 74; Defoe 1974; DeFoe et al. 1978).

The majority of studies reporting effects on growth also monitored and reported other effects occurring concurrently. In the evaluation of A1248 and A1260 (DeFoe et al. 1978), MGR effects were reported. A number of studies measured sublethal effects along with mortality, including physiological effects (Hansen et al. 1976; Mauck et al. 1978; Cleland et al. 1998; Shiau and Chen 1992; Thomas and Wofford 1993), reproductive and physiological effects (Black et al. 1998a; Orn et al. 1998; Gutjahr-Gobell et al. 1999), physiological and immunological responses (Leatherland and Sonstegard 1978; Mayer et al. 1985; Jorgensen et al. 2004), and behavioral responses (Fisher et al. 1994; McCarthy et al. 2003). The PCBs disrupt fish and other organisms through several different pathways and mechanisms of action (Monosson, 2000; Meador et al. 2002). Thus, it was not surprising that most studies reported simultaneous adverse effects occurring at single exposure concentrations of PCBs, including growth inhibition and mortality.